I get the figure of speech. It’s more about being extremely delighted in reference to the broader political and international context - things that are slightly more important than the empirical accuracy of a gambling website.
I've edited it to "I couldn’t be more delighted by their progress," but under slight protest. At every moment, millions of people are dying of war and preventable disease, but you still ought to be able to say you're delighted by specific things sometimes without adding "...aside from the broader political and international context, of course".
No . . . it's my fortieth birthday! But despite the specter of impending mortality I'm still able to spare a little happiness for Polymarket's victory : )
Please explain how critically important today is while neither your nor my life changes at all, beyond having to deal with a few more hysterical whiners than usual for a few weeks
Please explain how critically important the day that we discovered a dinosaur killer asteroid will hit Earth in 2028 is, while neither your nor my life changes at all, beyond having to deal with a few more hysterical whiners than usual for a few weeks.
I was hoping Harris would win, but Trump winning isn't the end of the world. There are upsides:
- the baseless lawfare against Elon Musk in retaliation for allowing rightists on Twitter will stop instead of intensifying, which could lead to more advances from Musk's companies
- while Trump's somewhat volatile, this can be an asset in international relations; he managed to drive a wedge between Beijing and Pyongyang when no-one else could, he's willing to drive re-armament of the West, and there are potential outcomes of WWIII where I'd be worried Harris might balk at doing the needful, so while I'm slightly more worried about him in the chair than Harris (largely due to his age) it's not drastically worse.
I'm very curious to understand the world model that makes you think that Trump - who consistently said that the US shouldn't help Ukraine or Taiwan, would be more willing to engage in a war with China over Taiwan.
(Also the wedge he alledgedly drove between China and North Korea is completely non-existant. NK is just as protected by China today as it ever was, and Trump's diplomatic openings toward NK didn't change anything to the NK nuclear program.)
(As a non-American who mostly cares about how the US election will impact my own country and our allies, I think that international relations is the main field were Harris was completely superior to Trump in all regards)
>I'm very curious to understand the world model that makes you think that Trump - who consistently said that the US shouldn't help Ukraine or Taiwan, would be more willing to engage in a war with China over Taiwan.
I don't particularly think that*. I do think that if he did engage in such a war, and there was a small nuclear exchange ending with Beijing run out of nukes but refusing to surrender, Trump (or Vance) would be certain to call the bluff and methodically burn Chinese cities until the PRC either surrendered or suffered total state failure; I am less certain that Harris has the sheer bloodymindedness necessary to do that, which both in the event gives them a chance to either build more nukes to kill Westerners, or lick their wounds to *try again* in 10 years, and also back-chains to make the PRC leaders less hesitant to pull the trigger in the first place (though this is counterbalanced by the greater chance of coming in at all).
As I said, I *am* slightly more worried about him in the chair than Harris, and like you was hoping for a Harris win on this alone, but there *are* complexities there.
*I think the disparity in chance of coming in is not all that big, in the end, because Trump wants to look "tough on China" (and "tough" in general) and that ties his hands if the PRC does something outrageous like attack Taiwan. But, you are correct, the disparity is in Biden/Harris' favour because Trump is less predictable.
> I do think that if he did engage in such a war, and there was a small nuclear exchange ending with Beijing run out of nukes but refusing to surrender, Trump (or Vance) would be certain to call the bluff and methodically burn Chinese cities until the PRC either surrendered or suffered total state failure
China has perhaps 500 nukes, the US has 5k. However, nuclear war is not a pissing contest where the US will comfortably win by these numbers alone.
A few megatons dropped on the Bay Area would likely be devastating to the US economy, and I don't think that you can be confident that US missile defense will avoid that outcome. Perhaps half of their nukes will hit US cities. The fact that they run out of nukes sooner than the US might not be all that critical, if two shooters manage to hit each other a few times, the fact that one of them has a larger clip and can hit his opponent more often does not prevent him from also bleeding to death.
Furthermore, while there are solid game theoretical reasons to pre-commit to retaliating against nuclear strikes on your cities in kind, I think there are limits to this kind of vengeance. If you launch your own ICBMs after detecting the enemy launches, you can still claim that your goal is to save your deterrence capabilities before your ICBMs are destroyed. You can also claim that one of your goals is to prevent the enemy from launching further nuclear attacks.
The morality of such a counterattack might be debatable, but it looks very different if you delay your response and nuke back at leisure, conditional on how the diplomacy is going.
It is the difference between returning fire when being shot at (which would generally be self-defense) and walking over to some incapacitated enemy and shooting them in the head (which is generally considered murder).
Still furthermore, the US has nuke parity with Russia. If it spends half of its nukes glassing China without getting hit too badly, Russia will be the leading nuclear power in a world where nuclear war has been normalized.
Lastly, I am highly doubtful that the mere threat of nuclear annihilation will force anyone to surrender. For Japan, it was more like the last straw as their conventional war did not go very well either. Do not expect the CCP to roll over unless you offer them a favorable deal. Likewise, do not expect total state failure. The CCP rose to power in what was initially a protracted war. You can diminish their state capabilities greatly, but the CCP will likely still rule over what remains of their population. Take Gaza: the IDF has bombing the shit out of Hamas for a year now, and they are still not defeated.
For the limited effects of strategic air power, see also:
As others have said Trump speaks in poetry, not in prose.
Or to put it differently, he should be taken seriously, but not literally.
You can view these as apologia, and make snarky comments. Or you can view them as ways to understand the man and the policies he's likely to implement.
Think of him as Nixon. You don't have to like the man, or his policies. But you're the idiot, not him, if you would rather mock, misquote, and make jokes than attempt to understand what's going on.
And at the end of the day, Nixon actually went to China -- regardless of what the Pauline Kaels of the world thought of him.
I don't understand how anyone thinks Harris was superior to Trump on foreign policy. What has she done? Given charge of border security, she was slow to visit the US-Mexico border, and ended up doing almost nothing about it. https://theweek.com/in-depth/1023057/kamala-harris-vice-presidential-track-record One may consider this immigration policy instead of foreign policy, but I don't see how one may consider immigration policy NOT to impact foreign policy.
The most experience I can find of her foreign relations seems to be having visited 21 countries and met with more than 150 world leaders. I cannot find any reference to anything she has DONE with those leaders. It does show she hasn't really impressed them, as foreign leadership goes through other people to contact the US, such as secretary of state or national security advisor. https://www.cnn.com/2024/09/08/politics/kamala-harris-foreign-diplomacy/index.html
My impression is that she has been hidden away so that she can do as little damage as possible, while showing how modern-thinking the US is by having a woman as a vice-president.
In contrast, Trump seems decisive in foreign affairs, though time will tell how good his policies prove to be. He initiated serious trade issues with China (including effectively a trade war), moved the American embassy from Tel Aviv to Jerusalem. and started (though it ended up falling through) talks with North Korea.
The largest problem I have with Trump's foreign policy is his having pulled out unilaterally from the Iran nuclear deal. I do believe it was a bad deal, but the US had already agreed to it under a previous administration, and this tarnishes any future foreign deals since any future president could undo them.
The claim isn't based on Harris's track record since she doesn't have much of one one way or the other. She's mostly assumed to be a generic Democrat.
Trump seems a lot softer on China now than he did during his first term, flip flopping on the Tiktok ban and sounding very hesitant to support Taiwan against Chinese aggression.
Meanwhile he's been consistently soft on Russia going back way further than that, including the incident he was originally impeached for where he held up Congressionally mandated aid to Ukraine for his personal political advantage.
And a large chunk of the senior foreign policy leaders of his first administration came out against him, including Secretaries of State and Defense and at least one Chairman of the Joint Chiefs. That alone says pretty bad things about the quality of his leadership.
>while showing how modern-thinking the US is by having a woman as a vice-president
<mildSnark>
Suggestion to the Democratic party for 2028: How about asking Kemi Badencoch if she'd like to run a larger country? Yeah, there is that phrase about "natural-born citizen", but I'm sure that can be worked around... :-)
The wise man would say that the stochasticism built into the US electoral system, the fact that any deal might be revoked 4 years later, means you damn well better make a *win-win* deal, and *stick to it*, because legalistic arguments about how you are technically allowed to do x or y because the treaty says so, are not going to work for you...
You do realize that there were many peace treaties signed in the Middle East under Trump, don't you? Also Trump didn't send billions in aid the terrorist nation of Iran like Obama or implicitly support Iran's nuclear program like Biden.
If you don't think Iran is a terrorist state, then no need to discuss any further.
Have you heard with full CONTEXT when Trump said he the US shouldn't help Taiwan? Ukraine I am sure he has said. What he will actually do, well, do you like to gamble :-)
Trump has consistently said that Ukraine is not worth WW3. Do you disagree?
Beyond that is pretty much projection of what people want to project.
Does this mean he'd give Putin a blank check? Well I guess, if you assume he's a Russian agent, then ...
Does this mean he can see a situation where there's no winning strategy (pull out of Vietnam in 1967, not in 1975)? Possibly.
Does this mean he thinks that giving Ukraine permission to use their fighter jets and cruise missiles, to force Putin to negotiate? Also possibly. I full expect Trump's character is such that he would love to both try negotiating a deal with Putin AND having Ukraine unleash thunder on Putin if the deal is not accepted.
Mainly what this means is exactly what the words say -- that Trump is not going to risk nuclear war on the US for the sake of Ukraine (or Taiwan for that matter, and probably not for any country in Europe given the way they have behaved over the past 20 years or so).
If you think that's a scandal, that the US SHOULD be willing to destroy civilization for the sake of virtue signaling, let me ask why you haven't already left the US, given the death camps and genocides that are supposedly surely coming? If not, then maybe you're a little too invested in playing word games with zero consequences, and your opinion about something that actually matters deserves to be utterly ignored?
I disagree in the sense that giving up Ukraine is a signal that one can in fact annex another country. We give up Ukraine to avoid WW3, and Russia decides Georgia has insulted them and ought to be invaded and annexed. That's OK; they're a lot smaller country than Ukraine. Same thing in Belarus, but they're landlocked, so less important. Then Poland, Moldova, and Romania, because they're a natural progression, like Manifest Destiny. Then Germany, since it's long overdue, after all.
Russia invaded Ukraine, you recall, because they were about to join NATO. The REASON for the war is the reason to defend them, and though WW3 would not be a desirable outcome, if one is not willing to risk it then Russia will have license to do as they want. Mutually assured destruction PREVENTS world wars, unless one side decides it must not enter such a war, which means the other side wins.
I'm especially enjoying the turning on Merrick Garland, the former Martyr of Justice who had been so egregiously cheated of his rightful spot on the Supreme Court.
I enjoy even more that the "but he is a convicted felon, 34 FELONIES" stuff went nowhere.
This was supposed to be damning, but it also came from the set of people who in other instances demanded we should scrap terms like "felon" because they were de-humanising:
" People with criminal justice histories are referred to in an array of dehumanizing labels, such as “inmates,” “criminals,” “prisoners,” “convicts,” “delinquents,” “felons,” and “offenders.”
...Dehumanizing labels stereotype and marginalize people rather than support them while they rebuild their lives. Individuals with justice system involvement are not defined by their conviction history. The words we use to reference people should reflect their full identities, and acknowledge their capacity to change and grow.
Be mindful of how you speak about The Fortune Society’s mission, the people we serve, our dedicated staff, and board members. We encourage you to use humanizing language — your example will inspire others.
Person or individual with justice system involvement; Person or individual impacted by the justice system; Person or individual affected by the justice system"
Throughout history and across the world, dehumanizing language has facilitated the systemic, inhumane treatment of groups of people. This is certainly the case for people impacted by the U.S. criminal legal and immigration systems, and that’s why it’s so important to use language that actively asserts humanity. Many people and organizations are moving away from using terms that objectify and make people’s involvement with these systems the defining feature of their identities. But many others—politicians, media outlets, and more—still use harmful and outdated language like “convict,” “inmate,” “felon,” “prisoner,” and “illegal immigrant.”
There are better alternatives — alternatives that center a person’s humanity first and foremost. These include “person who was convicted of a crime,” “person who is incarcerated,” “person convicted of a felony,” and “person seeking lawful status.” These words and phrases matter. Choosing people-first language is a step toward asserting the dignity of those entangled in these dehumanizing systems.
...Calling a person who was convicted of a crime a “criminal,” “felon,” or “offender” defines them only by a past act and does not account for their full humanity or leave space for growth. These words also promote dangerous stereotypes and stoke fear, which stigmatize people who have been convicted of crimes and make it harder for them to thrive.
“Until this carceral state and the people of this country begin to understand the power of the words that seek to dehumanize the incarcerated or justice-impacted people, there will never be a real and substantive conversation about criminal justice reform,” said Wright. “Our humanity is maintained and respected by not referring to us in those impersonal and definitive terms, but by acknowledging our intrinsic value as human and not by defining us by the worst day or act in our lives.”
So if someone vents about how Trump is a felon, kindly remind them gently that such terms are discriminatory and dehumanising and they should refer to him instead as "a person with justice system involvement" 😁
If they say "how can a convicted criminal be worthy of, or fit for, the presidency?", remember:
"Individuals with justice system involvement are not defined by their conviction history. The words we use to reference people should reflect their full identities, and acknowledge their capacity to change and grow." Don't uphold the carceral state or define people by the worst day or act in their lives!
None of those restrictions on language apply to straight white males. Trump is obviously that. So you can call him a felon, fascist, mis-gender him, dead-name him, anything you want.
And I'm not being a sarcastic asshole.
This is the actual stated truth. The less like a straight white male you are, the more you can claim people be careful their language around you.
This makes me nothing but sad. The most dangerous thing about Trump, to my mind, is that he dehumanizes people every chance he gets. Dehumanization is the first step toward quite a few historical evils.
I don't like these arguments from hypocrisy, because the person who's making them is being equally hypocritical.
IE I interpret you as saying "The Democrats believe Trump should be stigmatized for being a felon, but they don't believe random robbers should be stigmatized as felons, therefore they're hypocritical".
But you're also implicitly saying "I believe random robbers should be stigmatized as felons, but Trump shouldn't be stigmatized as a felon", which is also hypocritical.
I’m very worried about an increase in lawfare now that the people who endorse it are getting into power! (Especially the guy who turned my porn site x.com into a set of political ads.)
Don't know what you're talking about, there's plenty of good porn there still... Though, people are getting more serious about moving to Bluesky now that Musk has been, uh, more emboldened.
My main delight is that it's a blow against a) legacy media and b) cancel culture.
If you pay attention to events closely, you'll see that most news outlets (e.g., top sites on Google when you search a news story), are grossly distorted on any vaguely culture war topic, and distorted in favour of the radical left. To me, the clearest insurance was the coverage of Depp v. Heard. It was really disgusting to see people defend an abuser (Heard) like that, having actually watched 100% of the court footage.
Similarly I've had horrible experiences of cancel culture.
You saw it in how the Democrats ran the election. One of their core strategies was slandering their opponents and calling them bigots, sexists, extremists, fascists, racists, etc... E.g., they characterize JD Vance as an extremist despite him almost not getting the VP nomination due to being too democrat leaning (see the leaked JD Vance dossier). I'm sick of the hypocritical gaslighting, so to me, the election result feels like a breath of fresh air.
What is sad is that so many have been sold on the propaganda that Trump is Hitler. I've had democrats friends talking about how "they want to kill the guy." Fyi, voicing death wishes about a president is a federal crime.
It is insane and sad. They spout how they want to literally kill their political opponent, while complaining that the republicans are the divisive party?!
This was one of several sufficient reasons for me to vote for Trump.
I cannot image Scott Alexander, who I used to have tremendous respect for and still respect, somehow thinks there is a balance of power in this country.
Outside of elections this country is run by Democrats and their allies. I guess Scott can't see how incredibly stilted political speech in this country when most (though not all) of the allowed speech is what he agrees with.
The closeness of North Korea and China is vastly exaggerated. People act as if North Korea is a Chinese proxy. Whereas NK had a far closer relationship with the Soviet Union.
No, I don't think North Korea is a Chinese proxy. It *is* a Chinese ally, and the PRC has covered for them time and again since 1991. The *only* time the PRC backed off a kerfuffle, TTBOMK, is when Trump played the madman card.
Hey, you score 10 out of 10 for ineffectiveness and vacuity! Your comment is absolutely guaranteed to lead to no change at all in OP's view of the situation, and to increase their difficulty understanding the point of view of people who are less distressed than they are or in fact not at all. It is completely free of content, just a big fart of meanspiritedness.
How can your feedback possibly be valuable? Do you think OP hasn't noticed that more than half the country just voted against his views? Or that he discounts all those millions of your-views-suck votes because he's been waiting to get a read from Martin Blank, mean teen extraordinaire?
Deiseach did a good job of providing the same feedback, but with decent manners, mention of several issues that are now likely to be decided in a way she thinks is fair, and a bit of rueful acknowledgment that she couldn't help enjoying watching the other side absorbing just how badly it had lost.
Setting aside what counts as an intelligent and civilized response, let's think about effectiveness in convincing OP he might actually be wrong about some things. Deiseach may conceivably have given him a glimpse of the interior of an intelligent person who is not amoral but has arrived at conclusions very different from his. You, on the other hand, have probably strenthened OP's view that the opposing side is made up of meanspirited tinyminds who can't even argue an issue well. Fuck you too, buddy.
Banned for one week. The main reason it's only one week is that it's election week and everybody is crazy now and I assume they'll calm down soon enough.
Have you ever posted on wireheading? (I am in the middle of Ringworld Engineers) Is it feasible? Is it any better than the best available drug? What do you think of it philosophically?
Nobody REALLY knows how good wireheading is, because it's never been really developed. OTOH, there were rats that though it was better than sex or eating, even when they were starving.
On the *other* other hand, when I'm waked up from a sound sleep, and want to go back to sleep, offers of sex or food would not tempt me at bit. And yet sleeping is not an ecstatic experience. It's mostly a non-experience. But when you are tired and get in bed lying there feels great, and sleeping looks, in your anticipation, like a wonderful thing. It's actually kind of a ripoff. You think sleeping is going to be 8 hours of wonderfulness. In reality, giving in to the craving to fall asleep is wonderful, but only lasts 5 mins. or so. So I'm not sure we can know that the wireheaded rats experienced hours of something that is far pleasanter than sex or eating. Maybe it was sort of a ripoff for them -- a powerful illusion, at the moment of choosing wireheading, that they were about to experience something wonderful.
Heh. Sometimes I think even orgasms are sort of a rip-off, because they're so brief. You try to hang on to the physical ecstasy but it shrinks down to nothing in a few seconds.
Short version: I would be sad if everybody was turned into blissed-out lumps of jelly, but it seems like there are probably ways to do it that are better than that.
The Deep Utopia post is good, and you could write more on that and probably will. I'm a Christian, and it ties in nicely with the hard question of "What does God want us to do, besides not breaking moral law and besides giving Him due credit via prayer and worship?" Aquinas said "To be happy," and the Westminster Confession says "To glorify God," but neither of those is an answer till it's fleshed out (which Aquinas does a bit). Whether one is religious or not, it's clear humans are built in such a way as to help figure out an answer, and that the answer is going to be different vocations for different people. Actually, one good subquestion is what the best life is if you're a quadraplegic of normal IQ and we can't use technology to hook up your nerves and give you back your arms and legs or add to your IQ. That's a good question because it's simpler than for most people.
My understanding of climate change (stay with me) is that the more authoritative estimates of costs per degree of warming are about 1% of GDP
Meanwhile it seems that this is being revised upwards according to catastrophe bias. One new study cited news headlines and public pronouncements of prominent climate scientists as the source of their priors, which did not match the actual costs literature. Their study's lowest cost estimate was 6x higher than the previous highest one, and I get the impression that no insiders really challenged them and many said approving things, while their study generated a powerful round of news headlines
Here we have an information ecosystem that appears to be giving a popular estimate of costs 10x higher than the literature, and the popular perception is now altering the literature
The implications of this across the century in terms of miscalibrated policy costs must be in the range of $50 trillion dollars. If prediction and forecast markets can gain the prestige to replace popular headlines as information touchstones then this is an event super worth celebrating
1) an extremely precise definition of what exactly you’re assessing (which has no reason to stay the same, and the devil is in the details),
2) a model of the overall economy,
3) another precise model about what kind of climate impacts to expect.
And more importantly, there’s the fact that the result is an abstract figure with a mostly academic interest – making 1) close to what we actually care about is a difficult task.
All in all, I think these points can easily explain a 10x difference.
And if there’s a prediction market for the costs of climate change, I expect these points to heavily affect the price in as well.
I agree it's a difficult task to satisfactorily set up a prediction market on what the consensus academic position on global costs per degree of warming is and what it will be in x years
However, compared with the existing information ecosystem I consider it to be an easy task to radically improve on it. For instance in "Sea-level rise and its possible impacts given a ‘beyond 4°C world’ in the twenty-first century" by Nicholls et al, there is a reference figure of the exposed population in the high case, if all flood control infrastructure construction is halted until 2100
This figure is 187 million, and is how the news headlines describe the study results. However, in the low case with maximum adaptation costing I believe .02% of global GDP, the population displaced in 2100 is 5000-10,000, and this more closely resembles the median estimate than the headlines. Prediction markets have the potential to publicly elicit this kind of information to a much greater degree vs the existing system of incentives
>My understanding of climate change (stay with me) is that the more authoritative estimates of costs per degree of warming are about 1% of GDP
I hope you are abridging here, because that relation can be linear for only a very short range. Or are you saying a temperature rise of 50°C would incur a cost of 50% of GDP?
Yeah, the IAM model estimates are fairly linear out to about 4 degrees and look quite troubling by about 10 degrees. Luckily the IPCC projections for high emissions temperature levels tend to level out at about *6-7 degrees
This does assume that decarbonization is sure to happen by 2100 so you could argue that there could be a worst case emissions scenario not covered. The high emissions scenarios do include global coal increasing by 6x at that point though
edit: 6-7 degrees rather than 4 degrees with the 6x coal scenario
How much it costs is going to depend on where you live, and for some folks the costs are going to be intolerable. Your argument seems to assume that the costs are evenly distributed over the world population in some "fair" manner. This clearly is not going to happen. It's already killing people now and then. (Death spikes during heat waves, unexpectedly strong storms killing people, etc.)
Definitely one weakness with global cost estimates is that it doesn't account for marginal costs. This is something politicians are accounting for and the IPCC constantly mentions
There is a problem with that. They don't account for the marginal cost of climate policy. Just as Sub-Saharan Africa will have high costs from warming they will, by my lights, have higher costs than that from being blocked from developing with coal and gas
My general take is that we live in a world where the large majority of weather related deaths are from temperature, and a large multiple of those are from cold temperatures, including in Africa. Economic costs are not being balanced correctly against avoiding warming costs
Hmmm, this seems a bit like criticizing people (or systems) when they turn out to be right here... If the election had been close but happened to land on Trump then I would agree with you, but this election was sufficiently lopsided that I think giving a call of pure 50% is very slightly embarrassing, and I'm very curious to hear from pollsters where they think things went wrong.
Theo's 'Neighbor' theory is just one, and your right that with only one data point it's a bit ridiculous to say that he was definitely right, but I also think it's very important to note that Theo didn't just armchair reason his way to success. He tested his theory by commissioning a private poll with his own methodology, and it probably cost ~$100k, that's a huge structural advantage real money prediction markets have over reputation based ones. They offer profit incentives for spending money to gather *new* evidence, instead of just optimally summarizing existing evidence.
Great analysis. What i am left wondering at the end of it: you are correct that prediction markets predicting the correct result does not entail that their model is right. is there any kind of empirical testing or a certain way prediction markets could perform that would cause you to conclude they are efficient? Or is this just a question that is immune to empirical testing bc we’ll never know the actual probability curve for the election?
Before the election, Silver emphasized that 50% was a mean outcome, but not a modal outcome. Given that polling errors would likely be correlated, he noted that a landslide win would be fairly likely, with an over 40% chance that a single candidate would take all 7 swing states. See e.g. https://x.com/NateSilver538/status/1848060370268938538.
He also gave it around a 25% chance that Trump would win the popular vote. And it looks like he's doing to win it by a decent margin.
Not knocking Nate Silver at all here, but being able to predict that if there is a large polling error it would be correlated doesn't mean it's not unfortunate that there was a large polling error.
Using the posts framing, I would say that Nates model would put this result at a little less than 20% probably, or somewhere between 2 and 3 coin flips coming up heads (between 2-3 bits of evidence). Not a ton of evidence, but enough to shift my credence towards polymarket, especially since my prior against it was way less than 90% (though I did think before the election it was off)
Yeah, I think Nate’s model was right based on public information, but my gut tells me that based on this result and the herding that a lot of pollsters never published pro-Trump polls or were systematically overestimating Harris’s likely voters, making assumptions that her turnout would be similar to Biden’s in 2020. Without better polling firms, we will continue to see surprising election results.
I think this is right, but one thing that makes it hard to systematically resolve this is that the social landscape is always shifting, and pollsters are always trying to resolve the problems of earlier polls. There's always polling error election after election, but the direction isn't consistent, otherwise pollsters would correct for it. If there were methodological information which allowed private pollsters to construct more accurate polls, the public pollsters would have access to it too.
Desertopa said "There's always polling error election after election, but the direction isn't consistent, otherwise pollsters would correct for it."
You may be being sincere here but if you look at the RealClearPolitics average going back to 2004, the error is almost always, 5 out 6 times, in favor of Democrats:
2004 Bush R +1.5 +2.4 => 1.1
2008 Obama D +7.6 +7.3 => 0.3
2012 Obama D +0.7 +3.9 => -3.2
2016 Clinton D +3.2 +2.1 => 1.1
2020 Biden D +7.2 +4.5 => 2.8
2024 Harris D +0.1 -3.3 => 3.4
Yes, small sample size applies, but remember these are averages of polls for each year (at least 9 polls in the average each year) AND these polls were informed by many other polls a week or two before. (It is surprising how organizations polls in early to mid October, but NOT the week before the election.)
So the idea the polls will self-correct sound a lot like the argument that "of course the legacy media is NOT biased, if they were people wouldn't trust them and would not listen to them."
Well people don't trust them (see the latest Gallup polls on this this) and they have a lot less less power and influence than they did in 1980 but they continue to be highly biased.
TWO
Desertopa said "If there were methodological information which allowed private pollsters to construct more accurate polls, the public pollsters would have access to it too."
It's not access to special methodology that allows private polls to be more accurate. I believe they are more accurate for two reasons:
1) Money - polls are expensive. A candidate or party is willing to spend the enormous amount of money to get the best possible information
2) Focus - public polls are usually meant to drive readers / viewers to a news organization and so want as many crosstabs as possible and just need to get to close to the results. Private polls are going to be more focused so they can get more data points and spend more time of getting exactly the right information.
If we include midterm elections though, we don't see this sort of systematic skew. Are polling averages systematically biased towards Democrats in presidental elections, but not in midterm elections? Maybe, but it's not at all clear that we should expect that to be the case.
As far as the accuracy of private versus public polls, in statistics, getting an adequate sample size for high precision is much, much easier than getting a sufficiently representative sample for high accuracy. Public polls aren't inhibited in their ability to get enough data points to be sufficiently accurate, the real issue is whether their sample sets actually reflect the overall voting population, which is a matter of methodology.
Maybe money and focus can produce accurate polls, but candidates and parties do not produce accurate polls. Their internal polls are much worse than public polls. You are trying to explain something false.
Yes, I think the "coin" analogy is extremely misleading. We didn't get a single binary output - we got the fact that the country overwhelmingly preferred Trump, and if that doesn't result in _some_ kind of signal that should be measurable by prediction markets, then honestly, what use are they?
If the markets predict a 50/50 chance that Kasparov will win a chess game against Carlsen, then the game starts and it turns out that Kasparov's been dead for 3 years and "forfeits" ... we wouldn't say "shucks, guess the markets lost this coin flip". Forecasters should start to ask questions about how they can maybe notice the next time a player is dead and factor that into their predictions.
Last figures I saw showed him with a little under 51% of the vote and Harris with about 47.5%.
He won a bunch of swing states ... by something like 51%-49% in each case.
In what world is any of that an overwhelming preference?
He did better than Harris, for sure. The electoral-college system, which can turn a bunch of near-50:50 results into a landslide either way, will deliver him a nice big count of EVs, for sure. But what about this suggests that the US electorate had so strong a preference for Trump that pollsters and pundits who said "it looks really close" are as wrong as a chess prognosticator who didn't notice that one player is dead?
Hmm, I think you and the other replies here are making a basic logic mistake. Prediction markets are not trying to estimate the % of the country that prefers a candidate. They are trying to estimate the % chance that candidate will _win the election_. The two values both have a % in them, but please don't get them confused.
So, I admit that I made the same mistake when I said America "overwhelmingly preferred Trump", which is more a statement about the former than the latter. What I should have said is that "Trump was overwhelmingly likely to win this election". Knowing what we know now, that is a correct statement. After all, Trump didn't win "a bunch" of swing states - he won ALL of them, because his 51-47 "coin" was "flipped" millions of times for each one. The election didn't hinge on which way the wind was blowing on Tuesday. Carlsen's victory against his dead opponent didn't hinge on whether he got enough sleep the night before.
Since this was a historically strong and certain election result, you have two possibilities:
- Prediction markets should been able to predict the result ahead of time with at least a little confidence.
- Prediction markets can't distinguish a country's 51-47 preference, and are thus useless for predicting the outcome of US elections.
I think we'd prefer the second wasn't true, but at the moment it's not looking good.
I have no idea what gives you the impression that I don't understand the difference between "x% votes for X" and "x% chance that X wins".
The claim I was commenting on was your statement that "the country overwhelmingly preferred Trump". He only got ~51% of the popular vote! If he'd had a couple of percent less he wouldn't have won the Electoral College either! That isn't what an overwhelming preference looks like.
Anyway, you've now retracted that, so fair enough. But I still don't buy that "Trump was overwhelmingly likely to win this election". I mean, I guess it depends on what that means. He _did_ win the election (unless there was some huge fraud that no one has caught a glimpse of, I guess, but I don't see any reason to believe that and mention it only because worrying about election fraud on the basis of no evidence is kinda Trump's brand), and probably for almost all elections it's true that if you could rewind the world by 24 hours and run the election again you'd almost always get the same result. But I don't see that that's a useful sense of "overwhelmingly likely to win".
The relevant question is: _given the available information_, was it overwhelmingly likely that Trump would win? I don't see why we should believe that. It was, in fact, pretty close. All the polling indicated that it would likely be very close. Polls have systematic errors and they're really hard to predict. If you think it was "a historically strong and certain election result" in some sense that implies "someone competent would have called it confidently in advance", I think you're under the influence of hindsight bias or something of the kind.
I _do_ think that prediction markets can't reliably predict a 3.5-point electoral preference, at least not if what you want them to do is to say "90% that X wins" in 90% of cases where X wins by 3.5 points or something like that. So far as I know, nothing can do that. I don't know of any reason to think anything should be able to do it.
(Whereas chess prognosticators _are_ able to predict that Magnus Carlsen will win a chess game against a corpse. It's a silly analogy.)
"Hmm, I think you and the other replies here are making a basic logic mistake. Prediction markets are not trying to estimate the % of the country that prefers a candidate. They are trying to estimate the % chance that candidate will _win the election_."
Then why are you talking about how the American people "overwhelmingly" prefer Trump?
The alleged overwhelmingness of Trump-preference, has little to do with the expected probability that a candidate will win an election; it's the *ratio* of actual Trump-preference to error margin in our measure of Trump-preference that matters. If 50.01% of the population of each swing state is committed to voting for Trump, and 49.99% for Harris, and the Orbital Brain Scanners allow us to predict this with +/-0.001% accuracy, then we will be justifiably very confident that Trump will win, If the actual preference is 60/40 in Trump's favor, but people so thoroughly and bafflingly lie to pollsters that the polls all have +/-20% error bars, then the smart money will be "meh, leaning Trump but still pretty much a toss-up".
And Trump winning with 60% of the vote, in that hypothetical, doesn't make the smart money wrong. Trump winning with 51% of the vote, *definitely* doesn't.
Erm, yes, did you read the end of the comment you replied to? It is entirely conceivable that our inconveniently complex world is such that prediction markets can't distinguish voter preferences well enough to say anything useful.
In this world, prediction markets for election results might as well not exist. To repeat, this election was a very strong result (only really beaten by Obama's and Clinton's landslide victories, which, as I recall, didn't require superforecasters to see coming). If you claim that it's STILL not strong enough for prediction markets (what you call the "smart money") to move away from 50-50, then I ask again: what use are they?
>Last figures I saw showed him with a little under 51% of the vote and Harris with about 47.5%.
>He won a bunch of swing states ... by something like 51%-49% in each case.
>In what world is any of that an overwhelming preference?
Agreed! What I wish I had would be a two-dimensional plot of voters' views on something like:
Would like Harris to be POTUS: -10 ... 10
Would like Trump to be POTUS: -10 ... 10
showing the density of the distribution throughout this space
( Personally, I'd put Harris at -10 and Trump at -9, marginally lesser evil. )
The most important thing for anyone to remember is that, whatever one's view, there are a _lot_ of people who disagree, and (for Americans), these are your countrymen. You need to live with them.
Just for reference, in 2020 Biden got 51.3% of the popular vote compared to Trump’s 46.8% and no one was saying the country overwhelmingly preferred Biden. That would’ve been silly. 51% isn’t an overwhelming anything. Biden won by 7 million votes. So far, Trump is leading by 4.6 million votes. Biden also won more electoral votes in ‘20 than Trump will have won in ‘24.
EDIT: This last claim I made, about Biden receiving more electoral votes in ‘20 than Trump would in ‘24, now appears to be false. It now appears Trump will secure 312 electoral votes to Biden’s 306.
Yes. I’m wrong. You’re right. Some people probably said Biden’s narrow victory indicated that “the country overwhelmingly preferred” Biden, though I can’t find instances of that wording. Maybe you can. I just don’t think whoever you found who made that same claim could reasonably substantiate it. And given that Trump won by an even narrower margin, I don’t think the claim can be substantiated in his case either.
What do you mean? We got almost exactly what Nate Silver predicted. Off from the polls by about 2 points in every state. That seems to vindicate Nate Silver more than the people who said it would be an easy win for one side.
I thought it was going to be Harris, my basic reasoning was: all the polls are herding, this is probably because they're terrified of another "Hillary 2016: 90% chance" moment, and so they're disproportionately throwing out Democrat leaning outliers.
I noticed that the vibes were good on the republican side but I figured the argument there was: polls are about equal, but they always underestimate Trump so he's actually winning in a landslide. I figured the polling companies were now taking that into account (taking it into account too much, even, as above).
Can I ask what kind of things you saw which made it obvious to you?
It's the early vote. Compare a state like Florida or Texas on targetsmart to 2020. They had equal suburban and rural votes but were missing MILLIONS of urban votes (I'm not exaggerating) compared to 2020. It's when the early vote started coming in mid october betting sites across the board started to shift. This quixotic lone bettor theory scott posits is a tad dubious and I think there is a simpler explanation. Nevada is a prime example of early voter data (which makes up 80% of total vote) being borne out directly in the final results as most basic analysis would have lead one to conclude. (Ralston the main figure on early vote data from Nevada is super partisan and jumped through hoops to explain how kamala could "still" win. But even he was pessimistic in his framing. He also analyzed without the broader context of early voter data from other states which brought everything more into frame).
Yeah, it's those missing urban votes I'd like to see explained. Why didn't they turn out? I am being slightly mean here, but she proposed legal weed and free money so black men could legally deal drugs without risk of prison, why didn't that work?
Did she just not click with black/Asian/Latino-Hispanic voters? Why not? Is her weakness her background, that her father was not US black but Jamaican black, and her mother Indian, so she's not 'authentically' black by the US context and not Indian 'enough' for those voters? Obama had the same problem, but he was married to Michelle, née Robinson, whose father was a precinct captain in Chicago and so those contacts aided his career. Harris is married to a white Jewish lawyer in San Francisco and doesn't have that family network behind her.
I believe it was mainly covid and mail ballots( and only secondly the hispanic vote shift). people locked down at home easy to reach by political ads, have less to do, not out having fun. take NJ for example. Everyone was mailed a mail ballot and they won huge. This time they weren't and they had to request it and it barely stayed blue. they have a huge swath of low propensity urban voters and were able to juice the turnout in this demo far beyond normal levels due to covid and using the emergency provisions provided by covid to make last minute changes to local election law and procedure. They will say it was only to counteract the reduced turnout forecasted on election day from covid concerns, but In reality I don't think anyone really doubts in many cases they used their expanded powers to get the most bang for the buck possible. A lot of these measure were rolled back in 2024 such as in NJ. Trump mentioned this exact line of reasoning on Rogan btw as an example of how they "cheated".
Are you saying that if we made voting easier and more convenient the DEM party would keep their advantage? How's making voting easy and convenient not a desirable property of a democratic society?
Ideally you want informed people to vote and people who essentially vote randomly or based on irrelevant information to not vote. If identity and party politics weren't relevant to voting, I think everyone would agree with this. Since identity and party are relevant, and ease of voting typically skews towards Democrats, the calculus is more complicated.
some ppl say a basic filter would be helpful like having to wait in line to vote on eday, or having id , or passing a basic civics test , or filing tax retunrs, or anything to show a minumum level of investmnet to the society you going to determine and influence with your vote. i think everyone is happy with the opportunity to vote being universal and free, but the level of ease is another question. it becomes political quickly because of how these measures would benefit one side or the other. the but as far as the theory of a functioning healthy democracy i think its an open question .
I think it might not be mail-in ballots. I *think* California sends everyone a mail ballot (I got one this year without requesting it, though I'm not sure that's not just because I requested it last time). But California went from +29% Dem last time to only +18% Dem this time. Obviously it's still really blue, but an 11 point shift is huge.
just so u know the vote is only half in in california and they count the mail ballot last which are more blue so the shift isnt going to be that big in the end but still a shift. dont we explain that by mexicans and asians shifting? those were very real changes that happened this election in the electorate, but volume of vote was still also down in the most densly populated areas for other reasons.
Well, an 11 pt shift is only 5.5 pct of voters changing their vote. So not necessarily that huge.
Note also that the Senate election shows a much less dramatic shift. The last several elections in CA since 2012 had the R candidate get 37.5, 43.8, 45.8, 39, and 42 this year. (I skipped 2016, when two Democrats made the final).
I think that Trump's magic as a politician is just making the entire news cycle so damn toxic that most people just tune out. From exit polls, racism was 4% of people's main concern, while the economy was 40%. Trump distracting everyone with culture wars crap, and media taking the bait every time, made most normal people tune out of politics. And since Dem voters skew younger and less conscientious, lower turnout hurts Dems a lot more. Plus Dems were the incumbent after COVID, so they were already perceived to be worse on the economy.
Sorry, but I think the reason people didn't vote for Kamala seems to be pretty simple, though only in retrospect. The voters were presented with two worldviews:
Trump is a dangerous, convicted felon who will amass power and never leave office, just as he did in 2020.
VS
Trump will solve the economic problems of this country and actually manage illegal immigration, just as he did in his prior term in office.
More voters merely believed the second instead of the first.
She attended HBU and was a member of a *very* popular and social - across generations and geography - sorority there. That sorority was very much over the moon about her becoming VP, and many of them wore pink and green to vote in 2020.
How much influence black women have over the vote of black men, I do not know.
Well, as someone who "didn't vote for Harris, just against Trump", I think she wasn't at all inspiring as a candidate. And political promises are easy to make, but which one's will be carried out?
I also think there's an inherent bias against women as leaders in the US, but Harris wouldn't have inspired me no matter what race or gender she had. And distasteful as I find it, many are inspired by Trump.
Well, he does seem to claim that, but the studies he claims to rely upon is behind a sealed door. I won't subscribe to see if there's a pig in that poke.
I also thought it was obvious Trump was winning. But nothing directly to do with polls. It was instead over the non-endorsements of billionaires with newspapers. I was just trusting that they knew more about the odds.
An endorsement in a newspaper, is costly reputation wise, and should be used when it might get gains for the newspaper in the administration.
Endorsements are also more valuable when you do them less often.
The editorial boards of those two newspapers are almost never going to endorse a Republican candidate.
So it makes sense to hold off on endorsements for losing democratic candidates.
Interesting. I saw that as basic game theory. If they didn't endorse and Kamala won, they knew they wouldn't face retribution. But if they endorsed Kamal and Trump won, they would.
Trump already punished Bezos in 2018 by killing a DOD contract. He tried to do the same thing to CNN in 2017 by blocking the AT&T Time Warner deal.
Exactly. Authoritarianism doesn’t happen all at once. It creeps in by degrading institutions, and one of the first things to go is independence of oligarchs because it’s bad for business.
If Harris had won then I expect there would be retribution against billionaires who endorsed Trump. (In fact we've already seen signs of this for Elon.)
I think we likely would have seen retribution against Elon, in the sense that Elon has been using his ownership of a massive social media platform to influence the electoral landscape, including in ways that are dubiously legal.
This is a case where tightening (or just enforcing) restrictions on billionaires influencing the electoral process would be serving the will of her constituents. We can't examine the counterfactual where Harris won to see how she would have treated various billionaires who supported Trump, but I suspect that most who simply offered him their endorsement would not have been treated in the same category as Elon.
Regardless, as GreatSwordsmith points out, something's plausibly wrong when Silver et. al. say that Trump is only a slight favorite and then sweeps every swing state and the popular vote. Oh yeah, they also underestimated him in 2020 and 2016! Perhaps while everyone was voting for the Democrat, their "neighbor" really was voting Trump! Further, didn't other prediction markets follow Polymarket too?
I know embarrassingly little about finance and betting so I'm having to look into that more to see what you mean. What's your theory for them underestimating Trump three times in a row? I'm not a Silver doubter - the data shows he is very good. Perhaps there's just a Lizardman's Constant level of people who are paranoid about their anonymous Trump poll answer becoming public, getting mocked, and then getting pelted with coconuts?
A coin coming up heads three times in a row just isn't very unlikely. If Silver (and the pollsters) is honestly trying to predict the election as best as he can, it's not at all unlikely that the error might point the same way three times.
It's not (only) the final result prediction that's underestimated Trump, it's the individual state polls -- which were outside of their error estimates a lot more than three times.
This is exactly what you would expect if "error estimate" is being defined as just the expected variance from random sampling given sample size (which we can unambiguously calculate), but not any systematic bias (which we mostly can't).
OK, people like Nate Silver can kind of barely sort of quantify systematic bias by comparing polling errors across multiple polling houses, states, and elections, but not with great precision. And the pollsters themselves, I be;ieve just quote the random-sampling error estimates, which will underpredict the real error if there's any bias at all.
Yeah, that's no lie *exactly* what I was thinking when I got your reply notification, but it's not crazy to think that there's a pattern at this point at the margins either. Yes the sample size for a Trump effect isn't large, so it's difficult to judge.
When the market with money on the line goes Trump much harder than the ones that don't, that further makes me wonder. Yes Scott could be right that Theo was just replacement level smart and rich and lucky, but Trump being such a unique politician makes the idea of less random coin flip noise plausible as well. I used to dismiss the shy Trump voter thing until this happened.
Well, hindsight is always correct, but Trump did have the advantage over Biden, then Biden dropped out and the replacement by Harris gave what we can now call a temporary advantage to the Democrats.
The mistake was assuming this would last, that abortion was such a big deal every single woman in America would vote for Harris (women can be pro-life, too, you know), and that the mantra of "Literal Hitler! Fascist dictatorship!" hadn't worn itself out over eight years of repetition and would work to scare people to the voting booths.
In reality, it looks like Trump maintained his advantage and Harris just wasn't present enough to catch people's interest. I know it's been said that "She does too have policies, look at her website" but those weren't put up until very late in the day, and mostly the campaign seems to have been on vibes - vote for First Female Ever! First Asian Ever! First Black Female Ever!
Trump is Trump. Like him or loathe him, everyone has a pretty good notion of what he's like and what his policies may or may not be. Harris was "fun, joy, brat summer, coconuts!" but when you looked deeper, there was no 'there' there.
The stunts with McDonalds and the garbage truck may have been stupid campaign stunts, but they were stupid fun. Harris did the "oh hey Tim I love Doritos, toss me a bag" appearance at a gas station, which was also a stupid campaign stunt, but it wasn't fun. She did a "drinking beer with Gretchen Whitmer" appearance also, I think? But also not fun.
And the garbage truck was an immediate response to what Biden said. Harris' campaign didn't seem to have that level of being able to respond quickly and seize the moment.
Shoulda gone on Joe Rogan, Kamala! Trump did and Vance did, and now Trump is back in the Oval Office and Vance is taking over your old office.
> The mistake was assuming this would last, that abortion was such a big deal every single woman in America would vote for Harris women can be pro-life, too, you know)
I don't think it's that, exactly. Instead, I think many voters didn't let pro-choice views dictate their presidential vote.
Florida is a good example here. The state had an abortion-rights referendum on the ballot, which received 57/43 for the pro-choice side (the proposition failed to get the 60% required to pass, though). The state's presidential vote, however, we 56/43 for Trump. That means that at least 13% of Florida's voters cast a Trump/pro-choice ticket.
If I generalize that lesson, it seems to be that a large fraction of voters don't care about what traditional leading media figures think they should care about, even if their position on "issues" in isolation would be conventional.
I am thinking of a female relative by marriage, who is quite dear to me. She is the perfect Dem voter (though curiously, she had a job at a federal agency, and was only happy there and happy with her boss during the 8 GWB years).
She views politics as an unfolding tragedy. Kamala and Doritos could have been plenty fun and that would not have moved her at all. Sure, she didn't need to be amused; her vote was secure. But I think the audience most receptive to Dem messaging tends to be pretty lugubrious about politics, and is not looking for levity. Certainly the young people inject a lot of drama into all their internet comments. They always sing from the same sad song sheet, and their idea of entertainment is making SNL *less* funny for political reasons.
The Simpsons continue to be brilliant with their flag and abortion division of the electorate.
Re: arbitrage: real-money prediction markets can't stably have very-different numbers for the same question. If they do, it becomes profitable to e.g. buy "Trump" at 50c on one exchange and buy "Harris" at 40c on another, getting a dollar for 90c risk-free (since one of the two will always pay out) and forcing the numbers closer together again.
Play-money markets don't always work this way, because the play-money is not fungible and debatably desirable.
I saw PredictIt was the only site on ElectionBettingOdds not favoring Trump on Monday night, so I looked at that (since I'd put in $20 there before Kalshi won their court case and then forgot about it), and saw there did seem to be duplicated markets there that should have been arbitraged but their limits on new entrants prevented that. But I noticed that the question for whether there would be a female President priced "No" at 49 cents, while Harris & Trump were both at 53/54 cents in their named next President market, so I figured that was the closest thing to arb and bought 40 shares. That difficulty in arbitraging should lower the status of PredictIt relative to sites like Polymarket.
>> Perhaps while everyone was voting for the Democrat, their "neighbor" really was voting Trump!
I'm... confused about this.
Nate Silver et al. know that polls have statistical bias, and they correct for them. If a poll underestimated Trump in 2016 (as many did) and 2020, surely NS measured the bias and corrected for it this time around. So... what's going on here?
I'm still confused how this method is supposed to work.
I have no clue who any of my neighbours might vote for. I'm sure there's some loudmouths who can't help advertising their political allegiances to their neighbours, but none of them lives around me.
It's unfortunately much more complicated then that.
If your poll in 2016 should have been Trump +5 and your poll in 2020 should have been Trump +1 what adjustment should you apply to your 2024 poll? Trump -3 because that's the naïve progression? Trump +3 because that's the average even if we have no reason to suspect the average will apply? +5 because he's not the incumbent? +1 because that represents people now having experienced a term of Trump?
Not to mention if you're a polling aggregator each of the polls you're aggregating has done independent attempts to correct for previous results. It should be noted though that 2020 and 2024 are pretty inline with historical polling, 2016 was the outlier in how far off the polls were.
This is the big issue here. The polling averages always reflect pollsters best efforts to correct for previous results. Attempting to correct for bias in the polling aggregate is applying a correction on top of the correction the aggregators are already trying to apply. In general, the direction of error hasn't been consistent over time. With enough sample size for elections with Trump specifically, polling aggregators would correct for that too, but we haven't had data to assume one.
I'm wondering here if it was the early vote effect. People in comments below are talking about the early votes, and I wonder if there was an expectation that this and mail-in votes would swing for Harris.
The (ahem) disagreement over the 2020 result was in part that Trump was (or looked to be) in the lead at the close of one day's counting, then the next morning suddenly it was Biden, and overnight all these mail-in ballots turned up to be counted?
So I'm now wondering if the expectation was "Harris is polling just a small gap ahead of Trump, but when the mail-in votes come in, those will be Democrat votes" and the assumption was that even if Trump led in early counts, once the rest of the ballots were counted Harris would pull ahead, and hence the polls were adjusted to take this into consideration?
What seems to be different this time, if I can trust the online sources hanging around in shady alleyways going "psst, wanna buy a post-election analysis?", is that the Republicans learned the lesson and made efforts to have people use early voting and mail-in ballots themselves, and this made the difference - it cut away the advantage from the last election, as well as being much tighter over watching counts, being present when returns were being made, and so on.
> However, the baseline assumption of the Silver Bulletin model is that while the polls could be wrong again — and in fact, they probably will be wrong to some degree — it’s extremely hard to predict the direction of the error. Empirically, there’s basically no correlation in polling error from one cycle to the next one.
But looking at the link in that quote (https://fivethirtyeight.com/features/2022-election-polling-accuracy/, table "Polling bias is pretty unpredictable from election to election") there was a pretty consistent bias in favor of democrats since 2013. Not fully, but calling it "basically no correlation" seems wrong to me. Maybe it was so in the past, but something having changed since Trump seems not implausible.
Problem is working out how much to compensate. And that'll vary by circumstances. You can't just add a blanket 2% if trump was underestimated by 2% or whatever
Silver's model generates a prior based on fundamentals (economy, incumbency, etc) and then adds polls to it, and the polls are weighted on their past reliability. But you still have the garbage in garbage out problem, if the polls were herding towards 50/50 (as he suggested https://www.natesilver.net/p/theres-more-herding-in-swing-state) then you can't tell what way to weight it to cancel out.
The state level errors are correlated, so you would expect the swing states to mostly go together. And Nate Silver's model actually had less probability on a PV/EC gap than the markets
The fact that Trump won by a lot doesn't really affect the correctness of the odds beforehand. When I'm about to bet on the outcome of a coinflip, it doesn't really matter whether the coin has been already flipped and is hidden from me or the flip is yet to happen, and it also doesn't matter whether it's "heads all the way or just 60% heads".
Correct odds are always relative to the amount of knowledge at the time. Give me two people at random and ask me who is more intelligent, I'll say it's 50-50 and that's a correct bet. If afterwards turns out that one of them beat the other on 9/10 of the tests we used, that doesn't make my previous odds incorrect.
Agreed, I understand that stuff about probability theory. But Trump's uniqueness as a political figure makes me think it's plausible that the shy Trump voter theory really might be right! We’ll never know because a larger sample size is impossible for clear reasons.
Exactly! This is the most convincing argument from me (apart from the early voting data). Timur Kuran (author of the famous book on preference falsification) also makes this argument: https://x.com/timurkuran/status/1850200912062566873
I'm very surprised by the popular vote, to be honest. Can anyone inform me about that? Either every single rural person turned out to vote this time, or the voters in the big population centres didn't. If it were the case of 2016 where Hillary was expected to win in a landslide, I'd understand the voters staying home because they thought a blue victory was a sure thing, but the polling did consistently show it as only a slim margin for Harris over Trump.
So it looks like the Harris campaign managed to not do their jobs properly (one of which seems to have been the emphasis on appealing to women voters over trying to appeal to men voters, and when they did try to appeal to guys, it was too flat-footed to work).
"I eat carburetors for breakfast"? My friend, you look like you eat the entire truck (I can say that and it's not fat-shaming because I am a Person of Chonkitude myself). Granted, I only know Irish rural background myself, but if these guys are genuine American country folks, I'm a New York sophisticate in a size zero dress.
It seems that many latin americans decided to switch sides from Democrats to Republicans in this elections. Including in otherwise Democratic strongholds like NYC. https://www.bbc.com/news/articles/cze3yr77j9wo That drove up the popular vote. Trump also gained among working class african americans (although they still predominantly vote Democrat).
This makes a lot of sense to me and I've wondered why the usually christian (albeit catholic) and socially more conservative latinos vote for Democrats.
Now it seems there is no longer a big difference in political opinions between the working class with northern European ancestry and southern European ancestry (or mestizos) as both switched to Republicans (only the latter did so a few decades later).
By the way the advert is horrible and seems extremely ingenuine. And interestingly this is exactly what I thought about yesterday when I saw some images from campaigns of Harris and Trump. Trump is kitchy and obnoxious ... but Harris just seems fake and Trump does not. Of course he is really extremely fake in what he says. But it does not feel so cringeworthy as when I see Harris and Obama poiting fingers at each other in a "go girl" pose like they are 16 year old schoolgirls. I just don't believe them that this is really who they are (and if they were, it would perhaps be even worse). I believe Trump his behaviour ... It also shows he is a self-centered narcissist who talks bullshit but he is not pretending to be someone else. I would not vote for him if I were American but I would be very hesitant to vote for someone as fake as the current Democratic leadership.
Part of it, I think, is that with early Democratic Party you did have a strong strain of Catholicism from the Irish and Italians, so there was a common cultural background there to appeal to Latinos, as well as the reputation for being for the working man and the blue-collar workers.
As the Democrats slowly dropped that in favour of social liberalisation and chasing the college-educated vote, there were only the traditional links holding people to the party - we vote for the Democrats because our parents/grandparents voted for them, and they're the pro-immigrant party.
Catholicism is now probably, for the newer generations, a matter of 'cultural Catholic' or 'brought up Catholic' than deep faith, and for the first generation, again speaking as a cradle Catholic, there's a lot of folk piety but not much real understanding of the theology, as well as falling away to being influenced by the secular society around you (see all the polls done on contraception, divorce, abortion, etc.) The same way Pelosi and Biden can claim to be devout Catholics and still support abortion (sorry, is it 'reproductive justice' or 'healthcare rights' is the favoured term today? I can't keep up with how it switches).
So the cleavage point probably is "they're not representing the little guy like me" and Trump at least makes the mouth-sounds about bringing back good blue-collar jobs and immigrants have to follow the rules to get into America. Immigrant families who did follow the rules to get to America and become citizens don't like queue-jumpers any more than the rest of us do.
I think the Democratic party has generally taken the Hispanic vote for granted, on the presumption that the Hispanic vote should be convinced that the Republican party hates Latinos. It's a widespread opinion among the Democratic voter base that Republican animosity against illegal immigrants is a dogwhistle for hatred of Hispanics in general, but apparently not a particularly widespread opinion among Hispanic voters as a whole.
I think Trump is a bad candidate well beyond the usual scale of bad candidates, but the Democratic party had fallen pretty hard for the idea that the minority groups it's traditionally counted as constituents will vote based on identity politics rather than policy or cultural alignment, and while I don't think the average voter today actually does pay much attention to policy, a lot of minority groups are significantly more in cultural alignment with the Republican than Democratic party.
I think many people were disgusted by their choices for President, and it happened to favor Trump, as he had more support; people would turn out to vote FOR Trump, but if they were against Trump, it looks like they stayed home.
I base this on Trump winning about the same (maybe slightly fewer) votes in 2024 as he won in 2020, but Harris coming up short of Biden's 2020 by about 14 million votes. So lots of people didn't think Harris was worth voting for, but people still turned out to "protect" against a Harris presidency.
Low turnout in democratic safe states seems to explain a lot of it. Which is entirely sensible of you are efficiently allocating your get out the vote and persuasion efforts to swing states
I think Harris was pretty bad as a candidate and the fundamentals were against her, but she performed slightly better in the swing states, which is evidence her campaign was doing effective things.
Trump will win the popular vote, but don't use the current numbers to asses the margin. There's still millions of votes to be counted in NY and California.
One of the best criticisms I've seen of that ridiculous ad is that its entire purpose is to appeal to men, but it does not talk about a single issue men care about. "Be brave enough to vote for OUR interests" isn't how you win over a demographic.
> Either every single rural person turned out to vote this time, or the voters in the big population centres didn't.
Not even that, it appears that the voters in big population centres increasingly moved towards Trump. He didn't win New York City, but it went from 80-20 to 70-30, with the new votes coming from poorer neighbourhoods in Queens and the Bronx.
>So it looks like the Harris campaign managed to not do their jobs properly
The opposite seems to be the case, since her drop-off from 2020 in swing states (ie, where the campaign was operating) was substantially less than her drop-off overall.
When Silver would say that Trump/Harris were just a slight favorite, he was talking about the median outcome. Even a model that is extremely close at the median outcome (Trump vs. Harris winning) can have a wide outcome distribution, and I am certain that Nate Silver would admit ex ante that this was a strong possiblity (in fact Trump sweeping all swing states was the modal outcome).
“Agreed, I understand that stuff about probability theory. But Trump's uniqueness as a political figure makes me think it's plausible that the shy Trump voter theory really might be right! We’ll never know because a larger sample size is impossible for clear reasons.”
Also, the fact that real money markets seemed to think this make this plausible.
He won the swing states by tiny margins. Given that polling methodology has a 6 point error margin, it would be very hard to predict elections that over and over get decided by a couple of percentage points. We're basically flying blindly in that scenario.
Nate Silver, if you were paying attention, said that the most likely outcome of this election was Donald Trump sweeping every swing state and the popular vote, and the second most likely outcome was Kamala Harris sweeping every swing state and the popular vote, and then a bunch of lesser probabilities after that.
I don't think there's anything wrong with Nate's reasoning or forecasting. I think the problem is with your misunderstanding of what a close election looks like. Because this is exactly what a close election looks like - the winner getting their safe states and most or all of the swing states, but *only* those states and only 51% of the popular vote.
Whether the winner absolutely sweeps the swing states, or whether some minor variation leaves one or two of them in the loser's camp, is lost in the noise. But if you care about that bit of noise, the Silver got it right albeit with very low confidence.
It's genuinely baffling that someone could look at a one-off event and say "Forecaster A said there was a 40% chance of rain, B said there was a 60% chance of rain, then it rained. So forecaster A was totally wrong."
You just can't use a one time result to assess the accuracy of a prediction market. Any *single* observation is consistent with a probabilistic estimate, you need to assess a range of predictions or a large number of observations.
But seriously, doesn't that make them more effective for their stated purpose? Technically the profit is an inefficiency in the system as it makes use of arbitrage.
Acting well takes a good deal of intelligence, for you must take on another role based on your understand of who that character is, and yet still maintain who you are as a separate identity. And switch between these identities on demand.
Right, but once you select for 2 percent of the population, you limit your ability to pick good-looking people as well, which movies always require.
I'm not saying smart people can't be good-looking; I'd actually expect the correlation to be weakly positive on account of decreased mutational load. But if you pick *only* from the top 2 percent of the population in IQ, the cutest person is going to be less cute than if you pick from the general population.
"limiting access to >130 IQ people always makes it more effective"
If your point is that you won't find enough people that look the part and also have >130 IQ, then that is reasonable, and I agree, for there is a lot of competition for intelligence across many fields. But if you could get them, then movies with such IQ actors should still sell well, for they should "be" the characters in addition to looking good while doing it.
> you limit your ability to pick good-looking people as well, which movies always require
This seems to be a Hollywood idiosyncrasy. Movies/TV shows in the UK don't particularly focus on casting lookers more so than ordinary looking people who can act really well.
Couldn't you make the case that, a day before resolution, when interest rates etc stop mattering, any prediction markets trades not at 0 or 1 are mispriced? Isn't the whole point here that, because there is a truth of the matter, no intermediate value could possibly be the true value? It's not clear to me that markets, in general, have "accurate prices", which I why I don't simultaneously trade in all possible markets for all possible things. There is only "worth my money". Or am I missing something about the intent of the post?
I don't think so. For example, if I'm going to flip a coin, any bet other than 50% is "mispriced", because we have enough evidence about the process to know that it ought to point to 50%.
I agree 50% is the calibrated understanding of a coin flip, and I'm not frequentist myself. My comment is mostly about how these financial instruments are abstractions that strive to be related to probability, but aren't probabilities. The market structure is precisely why they might be valuable. But the market behavior is what we interact with, not the probability we are imagining. My initial response was, it is a breach of etiquette to *say* a market is mispriced, the appropriate etiquette is to *trade* that a market is mispriced. But if you acknowledge the trades aren't worth the hassle, then, what's left? The probability you infer from the market is not a product of that market. I don't mean to be pedantic here, it just seems the complaint isn't about mispricing, it's about real money markets not doing the secondary, aspirational thing you want -- letting you generate probabilities -- which, as you point out, can be achieved without any price mechanism at all.
I'm trying to make the case that "miscalibrated", the term related to predictions, is much better, I suppose. Perhaps I *am* being pedantic, but "mispriced" might not be a word it's possible to use when buyers and sellers agree on a trade.
Looking forward to it. The post basically says "if you had x prior, then you should update to y posterior after the given amount of evidence" in the context of metaculus and polymarket, and to me the math checks out. Curious how you are going to show that it's wrong.
I'm not Alexej, but Scott's basic mistake, as others have pointed out, is acting as though the only information we got was that Trump won. We also learned that he won by a substantial margin, which is evidence that the probability of his losing was substantially less than .5, which is consistent with Polymarket but not with various of its competitors.
This is a good point, although I would hope Alexej would have more in order to conclude Scott is "completely wrong about almost every single thing", rather than just this one mistake, although it is a pretty significant one.
I'm not sure how you would quantify the amount Trump won by into Bayesian information though.
Totally valid concern to raise, however as I wrote in another comment, I don't think it pans out. Do you have an actual model for why (and how) the winning margin matters as evidence? I can think of a bunch of examples where it clearly doesn't.
Presumably if a candidate won the election by a single vote, or ten votes or 0.0001% of the votes, you would acknowledge that election as having been basically a coin toss.
By what percentage of the vote (nationwide or averaged over swing states) does a candidate have to win for you to say "the probability of that candidate winning was 55%"? What about 60% or 70%? At what popular vote margin do we retroactively assess the candidate as having had a 90 or 95% chance of winning?
I am genuinely curious as to what sort of math you might have for this. But without math, it's meaningless.
To add to that, even if a candidate has exactly one more vote than the other, then they have an exact 100% chance of winning (I know about electors, but there's always a vote that wins the election, let's go with that one). It's the rule, more votes win the election, there's no randomness in it, the ballott counters don't do random coin tosses.
So why do we have odds? They are determined in terms of the knowledge we have at the time. So any discussion that doesn't involve some concept of prior knowledge is moot.
thanks I like Scott but I feel like he was missing the ball on this one. It's like I told him to fetch but instead of bringing back the ball he brought back a kids old baseball glove. No! Stop! Don't chew on that! Go get the ball! I hope u help him find it.
One of the interesting things about tonal languages is that, on top of the complexity of tones, you will sometimes get regional dialects that just make up their own tones for common words, so you have to throw everything you know out the window when visiting a new town. In some cases, tonal patterns just feel inverted. The upwards 2nd tone is transposed with the falling 4th tone, and vice versa, across several common words. You can get really confused just by crossing a river. (河, hē, hé, hě, hè, in Hefei, Beijing, Wuhan, and Ürümqi, respectively).
Ok I was bluffing when I said "tonal languages." In fact, almost all languages are tonal, even English. We use them for grammar and emphasis instead of semantics. And we can flip our tones too. This will have hilarious or irritating results, depending on your perspective. In one interview, the Pythons recalled the somewhat symphonic nature of their arguments behind the scenes. When the Welsh Terry Jones was upset, he'd lilt up at the end of every sentence and get very high pitched. John Cleese, with his boarding-school RP, would slow his speech and go down at the end of every sentence. So you would get this cacophony of accelerating piccolo and descending stair step baritone when they were fighting over some sketch. As a result each side would very poorly judge when the other was getting riled up.
So it goes with linguistic registers. We don't really think too much about them, we don't teach them or talk about them, so they can get all mixed up and backwards.
I've noticed this happening with a different sort of register in English, if I can call it one: how we convey certainty.
Most of the people I work with use nuance to show conviction. They want to show off that they've really studied something, that they see both sides. They do so by hedging, steelmanning other positions, showing deference to other views (which, after all, might change their view on the margins, but pose no serious threat of reversal!).
MAGA vibes completely flummox this crowd, because that crowd doesn't play by those rules. Its leading voices often deploy hyperbole to demonstrate the same thing. "I've really studied this" becomes "I know it's true, I'm 100% right, and everybody who doesn't agree with every one of my points is completely wrong, (give or take an insult about their intelligence)."
This isn't a criticism. I'm just Attenboroughing here. I'm noticing two groups, who otherwise somehow navigate all the same cultural spaces just fine, yet somehow came up with two completely opposite and incompatible ways to demonstrate that they really believe what they say.
It is as though we are all trapped in some elaborate farce.
I first caught on to this when Trump was on some AM Christian radio show years ago. The host said, with genuine sympathy, "Hey you caught some flak for saying Obama was literally the founder of ISIS. But what I hear you saying there is that his policies ended up leading to a situation where..." At this point Trump cut him off and said "No. I meant Obama literally founded ISIS."
It took me a long time to realize what the hell was happening in that clip. I think I finally get it? Kinda? It honestly still sounds absurd to me, I have to strain to hear in the wavelength where it makes any kind of sense.
So I say MAGA opened my eyes (ears?) to this. But after noticing it a while, I no longer see it as political. I know rights who speak nuance, and lefts who exaggerate for emphasis (oof, or so I hope anyway).
It brings to mind the conflict vs. mistake dichotomy, but it's really not that. It is related though, because team emphasis could easily get mis-labeled as conflict theorists. They're just using an inverted register to say "I actually believe my position, I'm not just bullshitting you." And when you raise probing questions to clarify their main points, find limits, ask for some nuance or both-sides-ism, it gets interpreted as, "why don't you adopt a more bullshitting posture," or in other words, "nah I think you're just bullshitting." So they double down and turn up the emphasis another degree. They shout louder to get through whatever earplugs you seem to be wearing.
This creates an unstable feedback loop. Discussions disintegrate shortly after, both sides leave questioning the other's sanity or seriousness.
If this becomes a separate post instead of just a reply, I thought of this when I read Alexej.Gerstmaier's 7 Nov comment:
"This post is completely wrong about almost every single thing. Will write my own Substack article to refute it"
Which is perhaps the most copypastable comment I have ever read on the internet.
But also... ok hear me out... maybe just a great textbook on how to talk like you actually mean what you say.
If I were Scott or EY I'd have to give this dichotomy a catchy name I guess. Emphasis talkers and nuance whisperers? Eh, idk, feels a bit silly.
If anybody can think of a better name, say so and I'll dub you Thingnamer.
Very insightful comment! I once talked to an HR lady about the statement "Dogs are larger than cats". She insisted I was wrong while I called her autistic
Back in the early 2000s, I'd follow Hollywood Stock Exchange, a non-money prediction market for movie revenue, to get an idea which new film to review this week. (Editors don't like it when you review what turns out to be a dud that nobody is talking about instead of the new bombshell.)
Hollywood Box Office, which was owned the Wall Street firm Cantor Fitzgerald (which was devastated by 9/11 due to their offices being in the World Trade Center), was usefully accurate, in part because it encouraged insider trading. HSX encouraged people whose jobs gave them access to watch movies early (or see the daily rushes even earlier) to wager pretend money. Insider trading is a good thing when it comes to upcoming movies because it's good that more people hear about movies worth watching and, especially, more people hear what movies to avoid wasting their money and time watching.
Cantor Fitzgerald wanted to turn HSX into a real money futures trading exchange. But I never heard how insider trading was supposed to be handled with real money on the line. Would the SEC be in charge of prosecuting pool guys who overheard their clients raving about a movie they are working on? It seemed pointless.
The laws for insider trading don't apply by default to futures. (Unless the future is actually related to a security, for which insider trading laws do apply).
Box-office futures are a bit of a strange story in financial products. It was controversial whether they should be allowed to exist given the legal framework for commodities. They were approved in 2010 by the CFTC by a 3-2 vote, only to be banned one month later by Dodd-Frank.
The question is whether insider trading is a good thing or a bad thing. When you can't make money off of insider trading, as in the turn-of-the-century Hollywood Stock Exchange, it's clearly a (small scale) good thing because it reveals insider knowledge, which I found useful in choosing which upcoming movies to review.
But if you can cheat other people out of real money, it's more of a question.
My realization when I became a weekly movie reviewer in 2001 was that some movies were obviously pretty good and that other movies were obviously pretty bad, which came as a surprise to me. For example, to compare early 21st century sword and sandal movies, Ridley Scott's Gladiator is clearly pretty good and Oliver Stone's Alexander is clearly pretty bad.
But that was hard to predict ahead of time, except among insiders who'd seen the movies or at least footage from them.
It's less a matter of subtle aesthetic taste on the part of the critic than that making a good movie is hard, and even successful directors like Scott or Stone misfire at times. So some movies perceptibly hit on all 8 cylinders, while others obviously don't.
So, to my surprise, much of reviewing movies is pointing out that this movie works or that movie doesn't work. I thought it was going to be high level Pauline Kael vs. Andrew Sarris aesthetic argument, but a lot of going to previews of new releases is just reporting "This is awesome" or "This is awful."
Generally speaking I agree with your comments here but the example given is not apt because Oliver Stone’s past performance is certainly an indicator of his future results
Realistically speaking, and setting review sites aside that seem to believe 90% of movies are good, Stone has never made a single movie worth the ticket price
Yeah, the debate about insider trading policy is interesting, because there are definitely tradeoffs with allowing or restricting it. There's variance among governments and markets about how to approach insider trading. AFAIK, in the US, only "securities" have the insider trading restrictions. Securities are generally the things you'd think of as investment products: stocks, bonds, etc. Other financial products, like commodities and derivatives thereof, have much less (if any) restriction on insider trading. Indeed, with commodities, insider trading is pretty much the whole reason for the existence of the markets: producers and consumers want to be able to hedge their risk via buying/selling futures.
Relevant to the topic at hand, prediction markets wouldn't be considered securities, so insider trading restrictions wouldn't apply by default. In some cases that's a good thing--you want the markets to aggregate information even from insiders. But in other cases, it could lead to perverse incentives or disorderly markets (e.g. if a person can both trade the contract and influence the outcome). I'm not up to speed on how prediction markets currently handle insider trading; although I expect it's a solvable problem, the right solution might not be obvious or straightforward.
> All of these factors gave me a 90% prior that Metaculus was better calibrated than Polymarket on the elections; now that Trump’s won, I update to 88%.
Your earlier point about not updating too much on a single coin flip should, I think, also make you update less on those factors. Effectively we can think of metaculus as having flipped five heads so far (more or less), so you should start out believing in metaculus but only by like 75% or so, not 90%.
I think the studies of why Metaculus is better includes many less-than-perfectly correlated results from each election, plus the evidence from the (many different) questions on my contest, plus the common sense evidence from Theo being one guy making the bet.
hm. I think you get ~2 bits per election for ~3 elections (the results aren't perfectly correlated but they are mostly correlated, and I don't think metaculus won literally every time? I'm ), maybe another 1-2 bits more from your contest (a lot more information but also it should be more discounted since the main advantages of election betting markets, like high liquidity and money involved, are missing). I don't know how much we should discount the one guy thing by (it seems more like inside view than outside view?). So call it about 8-10 bits overall. 10 bits gives us about 86% of Metaculus being better, but if we assume Metaculus lost one of those bets (and I think that's a fair assumption - the linked first sigma post does mention some races they did better on) that goes down to about 80%. Getting more precise than this requires a lot more research than I'm willing to do right now, but to a first approximation I think the 2024 election should drop us from about 80% to about 75% (which is what I get if I add in the 2024 prediction gap) that Metaculus is better.
Were you aware that Theo commissioned private polls, with his own methodology and at his own expense?
1. This means he's literally bringing information into the market that was unavailable to anyone else (and nobody would probably procure if there weren't a chance of recouping the investment; ie. non-money markets, poll aggregators)
2. If I say that this fact means that *my* prior is 90% that the money markets are giving a *more* accurate prediction than the other forecasters, so now my prior is 92% -- does that make me a good Bayesian?
3. The actual results indicate that the polls systemically underestimated Trump (again) -- if all the polls on Nov 3 happened to have landed on the actual state results +/- a random 3%, would you really think that the race was a coin toss? For reference, that would be Trump at +2 in GA & PA, +1 in WI & MI, +4 in NV/AZ, and +13 in Florida -- 60% seems a bit *low* on that map, to me.
Why a rapidly half-assed stealth edit where you simply mention the private neighbor polls and go with “I’m just not impressed by private polling”?
Because a smart whale with non-public knowledge incentivized to pay to gain such knowledge and use it to affect market prices is kinda the perfect goddamn argument for why real money could outperform play money or non-market mechanisms.
Do we have any details about what those polls were exactly, what their scope was, their methodology, etc? If he asked 1,000 people using a poor methodology and not weighting for the location of the responders, I’m sure he’d get a “Trump solid win!” conclusion but that wouldn’t count as useful new information.
If I said “Theo was correct because he consulted an astrologist” that would technically be “new information” but you’d still challenge me. Same thing here - “Theo paid for polls!” alone doesn’t update me much without knowing the details OR seeing his track record winning money using such polling.
This is the key question imo. People often like to ignore the methodology of polls because who wants to spend their free time poring through the boring details? Yet, modern polling methodology contains quite a number of free parameters whose unjustified setting could easily show whatever result the poller wants to show. I also did not go over the details of the poll in question, but to me it does not sound at all unlikely that a smart generalist could outperform the old guard who simply just wants to do their job and not rock the boat that much with too suprising results..
The crux of our disagreement is the probability you assign to "a smart generalist could outperform the old guard" in the area of voter polling. I assign that <1% probability without seeing his exact methodology OR a long track record OR numerous insanely good bets in the same election. You assign it... a higher value? 50%?
This post doesn't mention that Theo secretly commissioned his own polling - it just references him "working off of really smart reasoning." Not only does that render the particulars of the post somewhat misleading, it also affects the underlying argument, somewhat. The argument is that a single very rich trader isn't that much smarter that many poorer traders who won't bet as much as him. But as this case demonstrates, the single rich trader has the resources and incentive to create non-public knowledge to leverage for his bets. So while real-money markets may still be less accurate than play-money ones due to the ability to be skewed by individuals, the ability for rich individuals to not only reason like anyone else, but create their own proprietary information mitigates the disadvantage of real-money markets.
He edited the post after I commented (but added a new typo there, in his haste), but I don't find his edit compelling. He now claims that "I don’t find the private polls very interesting either - the existence of private pollsters implies this happens often, and we shouldn’t expect these private polls to be massively better than the public ones."
But first of all, the article states that he commissioned a major pollster who he wasn't allowed to name, per his agreement with them. So this was almost certainly not some "private pollster," but rather a well known public pollster. As such, there's little reason to assume that this happens often. As far as we know, the business model of the major pollsters isn't dependent on conducting secret polls for individuals - why assume that something like this is then routine? The original post strongly implied that he hadn't read the article he was linking and this point keeps the question open of whether he has yet read the short article.
More significantly, even if Scott is right that secret polls aren't "massively better" than the public ones, they don't have to be! Polling averages are useful and even if the secret polls weren't better than the public ones, they still give a trader alpha that can allow him to beat the market, which is what Theo did.
The issue I raised therefore still stands. While it may be true that real-money markets are more volatile, since a single very rich person can afford to take a large gamble that poorer people can't effectively bet against if the market isn't liquid enough, the rich people have the ability and incentive to create their own information that smaller traders can't. Their info doesn't have to be massively better to allow them to make money off of trading.
As far as the editing, the original version of Scott's post was:
"—But didn’t Theo give a great explanation of his strategy to the Wall Street Journal, which proves he was working off of really smart reasoning?
Yes, but there were dozens of people who could give equally-plausible arguments for their positions before the election. These were divided half-and-half into intelligent-sounding pro-Kamala arguments and intelligent-sounding pro-Trump arguments, and Theo was a completely replacement-level example of the intelligent-sounding pro-Trump arguments. We should think of him as an example of an intelligent person with a good argument who got lucky, unlike the many other intelligent people with good arguments who didn’t.
I’m 90% bullshitting here but I believe the business model of many polling companies is, if not “secret polls for private individuals”, low-profile polls for private clients; I don’t see how those companies could be in business between elections otherwise. Market research and such.
>So this was almost certainly not some "private pollster," but rather a well known public pollster.
Fine, but given that the public polls had lots of correlated non-sampling error, why would a private pollster do better? They're doing the exact same thing as the public polls, so you would expect them to have the same nonresponse bias.
First of all, I pointed out in my comment that additional polls, even if no better than other polls, still create a more useful polling average that gave Theo an advantage that he capitalized on.
There's reason to think his private polls were particularly useful, though. First, the public pollsters were evidently herding, that is, their consistent 50/50 results were unlikely even were the actual electoral split 50/50, demonstrating that they were putting their thumbs on the scales. The reason for herding is presumably to avoid embarrassment if the results end up being very different. But that issue is mostly obviated when commissioning a private poll on the condition that the results remain private.
Second, Theo suspected that traditional polling wasn't capturing Trump support and that "neighbor polling" in which you ask whom people think their neighbors would vote for would do a better job. Whether or not he was right about polling methodology, since his polling method differed from the norm, he had particularly useful information.
Do we have any details about what those polls were exactly, what their scope was, their methodology, etc? If he asked 1,000 people using a poor methodology and not weighting for the location of the responders, I’m sure he’d get a “Trump solid win!” conclusion but that wouldn’t count as useful new information.
As the linked article mentions, Theo commissioned "neighbor polls" in which people are asked how they think their neighbors will vote, instead of how they'll vote, themselves. Theo thought that this would better capture Trump support that people may be less inclined to reveal to pollsters. The details of the poll remain private per his agreement with the pollster.
Why does it matter what form the "really smart reasoning" takes?
It's one guy, acknowledged as really smart, claiming that he can beat the market but with no prior track record and only a vague description of his methodology. That shouldn't move anyone's priors very much, and the details of the cleverness matter not at all to the reassessment of the priors.
"But long after Biden won the last election, PredictIt said there was a 9% chance *he* would be the next President; some commenters suggested that maybe he would #StopTheSteal and win a fair recount (aside from the inherent implausibility of this, some of the specific scenarios bettors placed money on required him to win California, where his campaign hadn’t even asked for a recount)."
based on reading this a couple times I think "he" here refers to Trump?
Ik also very confused about this. I guess the $2 billion refers to the total trading volume, which also counts back-and-forth trading? The order book certainly didn't have $2 billion in it at any time.
> Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%. Now should we say that the “true” probability of Trump’s win was 90%?
No, because Theo's goal wasn't to drive up the odds, it was to take advantage of a mispricing. While he was confident enough to buy Trump at 60%, it's unlikely that he would have been willing to do so at 90% odds.
Unless Theo thought the true odds were >90%, in which case buying Trump at $0.90 would still be positive EV from his perspective. To the extent that he was money-constrained, we don't know how high he might have pushed the price with more available capital.
We actually have data here, and it was the economy, maco. Mostly inflation; immigration was a distant second.
Abortion and preserving democracy were the big choices among people voting for Kamala, who after all almost won.
While I support a national healthcare system, including sex changes/gender-confirmation surgery where appropriate (some people are genuinely better off that way), at least for citizens... I have to admit the line on 'transgender operations for illegal immigrants in prison' was a pretty effective slogan, as it ties together anger over cultural changes, anger over illegal immigration, and anger over crime.
Thank you for the link! But I don’t see how the poll completely disproves maco’s claim, though: in the “issues you care about”, NBC did not ask about “cultural” topics such as, say, wokism or trans rights – was it an open-ended question?
Trannies think the world revolve around them. It doesn't. Literally almost no one cares. Yes it's always good political tactic to say the other party is aligning with the freakshows ( this is what many trans people appear as to the average uneducated voter im sorry) whether it is the satanists or the pot smokers or what have you. This strategy works. But if it wasn't transtrenders it would just be whatever other subversive unhinged movement infected the youth. It is not the trans themselves that are the unique issue it is what they represent to the voter. They are a symbol of degeneracy. This always exists in society and the left has been guilty of openly feeding into it as of late.
Who just wrote a long answer focusing on that topic when I was asking about culture war in general? Called them a “symbol of moral degeneracy” (as opposed to many possible formulations highlighting their irrelevance, because a symbol is by definition significant)?
Maybe we try not to care, but it's put in our faces so much , especially if you are on the Internet much it feels as if "the trans are out In force" , so then we find ourselves dragged by its sheer oversaturation to dicuss it. This is how eric Weinstein and co. Found themselves expounding on trans issues for hours on podcasts. They were dragged in by the gravitational force once they entered the online sphere simply due to the outsized presence of trans individuals. Who are mostly AGP by the way. Trans Androphiles sleep with normal boys and dont spend their time commandeering every Internet message board known to man.
Spent an hour yesterday talking to a trans man about the election. He thinks the Twitter LGBT scene is ridiculous, because people seem to think their subgroup is the only group that matters. He hates wokeness. He was furious at radical lefties who voted for 3rd party candidates in swing states out of purism. Main concern about having Trump in office is that he has illegal immigrant family members in the US, and is worried about them. So how about reminding yourself that trans people are like other peoples -- some are childish and narcissistic, some are not. Most groups on Twitter look like self-important hysterical assholes. Twitter is designed to attract people and turn them into self-important hysterical assholes.
It was quite a choice, wasn't it? One side favored transgender surgeries for illegal immigrants in prison, while the other side would prefer to just castrate them.
You forgot the most important part of this. ...Why should taxpayers have to pay for this? In a country that doesn't even have a fully public health-care system like what Canada and Great Britain have. It really was an absolutely awful policy idea put forward by Harris, and not just because it was bad for her politically. It's because it's totally unwarranted given the current status of the American health care system. Why the heck should a non-citizen wanting expensive gender surgery take precedence over a citizen who needs heart surgery or cancer treatment in order to stay alive?
Trans are failing at the primary objective: be pretty girls! In Asia they are everywhere but they blend in and you barely bat an eye. Everyone would be far more willing to accept them if they succeeded in their primary mission. Instead they are blaming us because they did not live up to their name! If you are trans I shouldn't be able to spot u looking like a bigfoot from 100yards! Less drama and tears on twitter and more effort on transitioning please !
tim dillon show is a mainstream popular podcast that interviewed the vice president last week. he goes on rogan all the time. hes a popular GAY comedian. its funny a bit he did about a family saving up to buy their trans son a pussy for christmas on his podcast. what is wrong with sharing that. the news story about the kid dying was real and very tragic.
But honestly, it's nice to get some confirmation on the actual reason everyone hates trans people. It's fine though, deal with them as you wish. Nobody deserves happiness.
You know, I actually really like raunch, and I'm ok with hearing female genitals called pussies or cunts, and use those words myself both ias swears and to refer matter-of-factly to genitals. I'm a woman, by the way.But this joke offends me, so was trying to figure out why. And I did.
It's 2 things: The first is that it mixes sex and parents It makes me think about parents checking out their daughter's genital secretions. It's not that I disapprove of being transgressive in that particular way,it's that I just find it creepy and unpleasant, a real buzz-kill for my laughter.
The second is that it hinges on the idea that women should be ever-ready. As you probably know, even natural-born women do not have mucus running down their legs. They only secrete the lubricant when they are turned on. If you want frequently wet pussy in your life may I suggest that you work on getting way way WAY hotter than you are, and greatly improve your social skills. Until then carry a pocket pussy and a great big bottle of lube. If the bottle leaks, you too can leave trails of slime.
For some M to F trans people, being pretty may not be their primary objective. They may just want generic femaleness, or to ditch their male genitals, which have always felt to them like a grotesque lump that should not be there. And they may be attracted to women before transitioning and after, and prefer looking dykey. Also, transitioning is expensive and time-consuming, and most people, trans or not, are not in a position to travel to another country for surgery.
You aren't coming right out and complaining that it's jarring to see a big broad person with female attire and hairstyle, but I sort of think that's the root of your complaint. If so, hey, I get it. I find it jarring too. I find sexual ambiguity or change of gender jarring. I'm sure nature wired us all to distinguish between male and female, and of course the culture has too. It's natural to feel some recoil or confusion when confronted with someone who has characteristics of both genders in their appearance or behavior.
But I don't think it's reasonable to expect transgender people to spare you the experience of being jarred. What really works for weakening that reaction is just desensitization. Log enough time with trans people and their transness will just be a lot less salient, and you'll process them the way you do non-trans people: is the person an asset or a liability at work, do you like their sense of humor, do you share any interests, what do you agree or disagre with them about, etc.
Reading trans Twitter does not count as logging time with trans people. Most groups on Twitter are caricatures of themselves, and act like self-absorbed, hysterical assholes.
"Trump won because of my pet issue" might be true, but also said by everyone, including all the wrong people.
I think that giving sex changes to prisoners was dumb, probably cost her some thousands of votes, and was overall a symptom of her poor political skills. But people really hate inflation and being gaslit about it.
on twitter why does the profile say "I'm a soup puppy kitten puddle slurpy mess goo bunny" and it looks like a boy that estrogenized his whole body to look like a preteen. I come across this profile aesthetic a lot flying the trans flag. I'm sympathetic to exploring gender to an extent but this is the kind of thing throwing some people off &_&
That didn't help, but I might have let it slide if it had been in a thread about transgender issues or contained some interesting insight. Instead it was in a thread about prediction markets, and distracted from the topic with a one-sentence assertion of a very controversial opinion.
I've said before that I use some set of criteria like "true, kind, necessary". As far as I can tell this statement wasn't true (or at least the poster provided no evidence for it). The slur didn't help with kindness. And it was basically unrelated to what this thread was about. 0/3.
Was it really mispriced? Early vote data showed about equal number of suburban and rural voter in Florida and Texas but missing MILLIONS of urban vote. it was inconceivable for them to make up this deficit on e-day and they didn't (both carried by 14 points flying In the face of any polls). Anyone looking at early voter data with an open mind seems to be much more confident about their predictions in certain swing sates like AZ and NV (that both had solid early voter data, nv being the most complete). The markets had these at -300+ at points because they were using early vote data.
For myself I bet 14k on Nevada on a sportsbetting site I use with very little fears even though I'm generally very risk averse and that was the largest bet by far i ever placed .
Stil waiting for the finalized results for the 12k pay out uwu
You can’t know the party affiliation for certain, but you do know where they’re voting and that tracks with party affiliation pretty well. High early votes in rural counties (which are pretty much always lopsidedly Republican) with lower early votes in urban areas (lopsidedly Democrat) and you can make a good guess as to how enthusiastic the early voters are by party.
> Yes, but this system is meant to work in a world where amounts of money are at least somewhat even.
No! The system is meant to work in a world where the amount of money flowing through prediction markets is *a significant fraction of the economy.* The whole idea is if you can systematically make true predictions you can pull alpha out of the prediction market. You get a very *uneven* distribution of money, and that *correctly* drives your prediction strength up, because the market recognizes that you're a reliable source of accuracy. Even amounts of money would ruin prediction markets by turning them into democracy. The real problem is that being able to produce value for society has no reliable connection to predictive accuracy of bets. - Though, even then I have to say that being able to produce value via a banking job probably has a much higher connection to predictive accuracy than most other jobs, so maybe even in that regard the system worked as designed here.
I think 2% is a bit generous, we've seen that prediction markets have a rightwing bias so this is perfectly in line with expectations, I'll be updating 0.2%.
As to why prediction markets have a persistent bias when markets are supposed to correct for them: prediction markets don't just predict the future, they also *affect* the future e.g. if prediction markets predict that a certain policy will have a positive effect they're more likely to be implemented. This gives market participants an incentive to be more optimistic towards policies that if implemented would benefit them e.g. if Musk bets a million dollars to shift a policy (that would net him a billion dollars) towards a positive perception, then even if he loses that million dollars on the prediction market he more than makes up for it with the billion dollars he gains by having that policy implemented.
This brings us to the demographics of prediction markets, which is mostly rich white guys, people more likely to be rightwing. If you're a poor black woman who's inside a poverty trap living paycheck to paycheck you can't participate in the prediction market even if you want to. So even if you know that the policy that Musk wants to implement is just going to enrich rich white guys at the expense of the poor (which you might not know since living paycheck to paycheck doesn't give you a lot of time for policy analysis, but that's a whole different problem with prediction markets) you can't even correct it.
I never claimed it was Elon that was just an example, why would you even think I believed that given that I didn't talk about the whale, did talk about a fictitious policy proposal, and that the post I'm commenting on literally talks on and on about Theo.
It's not that people with vested interests can't make money by predicting correctly, it's just that they can make even more money by predicting incorrectly, which means conditional prediction markets as a whole will, on average, be skewed in the interests of the wealthy.
I didn't claim you did claim it was Elon. I'm claiming that the facts of the case don't fit your hypothetical. As far as we know Theo derived no particular benefit from the outcome other than his bets paying off. They CAN'T "make even more money by predicting incorrectly", because they lose all the money on incorrect bets. Bet incorrectly and someone like Theo can figure it out and bet against you to take that money. Nobody on earth is rich enough to buy the pot from everyone else on earth willing to take their money.
I wasn't talking about the case/Theo, I didn't even mention them. And people CAN make even more money by predicting incorrectly if that causes a policy to be implemented which makes you more money, like I explained in my comment. It's not necessary for everyone to be on the same page to skew the result e.g. the scenario in the post.
Even assuming perfect information (which as I pointed out is only true in the ancap's imagination and not in reality) this still isn't true since one group could outspend another group e.g. the scenario in the post
"Until early October, Metaculus (the top forecaster) was consistently +4% bluer than Polymarket (the top market)."
This is incorrect. I have a graph: https://photos.app.goo.gl/Tuab9CoLaL96cZpS9. The data points are only every few weeks, but it doesn't look remotely like a consistent +4 to Polymarket.
It follows that "If not for Theo, there’s no reason to think Polymarket would ever have shifted from its usual regime" is also wrong. There is no "usual regime". Theo's trading pattern is murky, but the accounts identified by the WSJ were not active in the week before the election. My own sense is that while Theo was responsible for the price spiking to 67c on 30 October, he didn't contbribute much to the price on 4 November. There were other people making big bets, and perhaps the argument goes through if you substitute "half a dozen whales" for "Theo", but the narrative where this was all about one French guy is too simplistic.
I remain of the view I expressed in the hidden thread, which I'll copy here so other people can see it:
There's more information here than whether the outcome occurred or not. If Trump had won by a very narrow margin, the result would be the same, but we would consider that a 50/50 prediction looked good. In the events which occurred, we are therefore bound to consider that it looks less good.
Sometimes, you predict something has a 10% chance of occurring, and it occurs, but the means by which it occurs was sufficiently peculiar that in retrospect you are content the prediction was sound. This is the opposite situation. It doesn't look like something that just happened to fall on one side of a knife-edge.
Theres robust early voter data for recent elections with the ability to make fine grained comparisons. If anyone looked at a state like Fl or TX the differences made it almost literally impossible for any compensatory eday vote to bring the results level with 2020 . In the swing state there were more nuance but this is the main factor that made the betting market more red than the poll aggregate. The early vote data portended underwhelming result with certain demos.
The post awards 10 percent probability to „Theo knows better than the rest of the world“, 90 percent to „Theo doesn‘t know better“, and then updates from 10-90 to 12-88 based on the election result. Scenarios such as this one, real markets incorporating early voter data better, are given 0 percent.
With the election coming into close view, too much dumb, mainstream-poll-following money moved into the markets for the smart money to keep it in check . . .
(Meant to be a joke originally, but who knows. My own first speculation was a different scenario from Buckwheatloaf's, based on the fact that we know Nate Silver's model swung slightly towards Trump temporarily, which obviously couldn't have been caused by Theo: perhaps real-money markets interpreted the slightly shifting polls as the end of the Harris honeymoon effect and momentum for Trump, therefore experiencing a bigger swing than Silver's model, which does not consider momentum, as Gergő Tisza documented to me in the Mantic-Monday post's comment section.)
This poll instantly became a hot topic for a lot of poll watchers online Many conservatives immediately mocked it, finding it unbelievable. But perhaps it gave some renewed hope to Harris supporters, perhaps leading to some deciding to take the plunge on Polymarket.
maybe mostly emotions due to the selzer poll and the perceived possibility that all the previous analysis (regarding early voter data or other data) was somehow wrong. ppl just decided to hedge their bets due to mixed signals. in retrospect obviously just a wave of fear. i dont think there was any big manipulation in the markets, more of a genuine moment of confusion and doubt. luckily for me i got really good odds on nevada due to that. (i felt the fear, but i managed to stay strong in my convictions).
I'm slightly skeptical large bet on poly could we set the odds because Isn't there a lot of sportsbook with high(er) volume? It is easier by magnitudes to send ur BTC into sportsbools and place ur bet in an instant than try to use poly market "buy shares" (convoluted and confusing to anyone habituated to sportsbetting). Wouldnt the people who set the odds across sportsbooks be placing large bets in the millions reverberate to poly.
I don't know if I'd ever blanket compare two prediction markets in that way. The users have different cultures and will bet differently on different issues, and each issue has its own considerations. I think a lot of people were trying to arbitrage between polls and Polymarket, which ended up being a weakness... But also, if Polymarket hadn't been that close to 50/50, perhaps the incentive wouldn't have been there for a billionaire to commission his own, more accurate polling. I'm conflicted on whether this is a win or not for prediction markets, and given the results of that polling was hidden from the pubic, I'm conflicted on whether all of this is good for society. Ideally pollsters would just do better.
Easy for me to say, sure, but it's consistent enough that I was able to make money on the market in question based on the prediction that A) pollsters will continue to underestimate conservative sentiment, and B) lots of people are arbitraging between polls and prediction markets. I don't know what you'd call that. Reverse arbitrage?
You could make money by both active betting like what you did and risk-free arbitrage at the same time. You'll get larger returns doing no arbitrage, but you'll do that by accepting more risk.
I said it in the other thread, but betting markets are nothing but another data point. That data point only holds water if it reaches a critical mass and even then is just another data point.
Americans have proven three times in recent memory that we absolutely suck at predicting elections. Why disparage more data points? Polling was incredibly bad at predictive power this go round (landslide + popular vote), anecdotes in my county were spot on, betting markets were accurate.
Anonymous controlled spaces like reddit were very inaccurate, anonymous slightly less controlled spaces like X were more accurate.
All of this negated by Socrates: the beginning of wisdom is knowing you know nothing.
I'm somewhat confused on what you think the 'normal operations of the market' are? A big player confidently buying up a short-term contract seems like a very normal operation!
prediction markets are meant to weigh the wisdom of the crowd by each participants confidence, and if Theo was confident for the right reason, his vote should be outweighed! The argument that it was difficult for some to participate on Polymarket seems to also apply to why Theo wasn't buying on any other markets - clearly there were some frictions applying in the other direction too
"mid October" is when early voter data began to trickle in. To correlate the shift in the markets to a lone bettor before that seems like an oversight =.=
The sportsbet site also shifted at this time not just poly. It certainly is insulated from barrier to entry but I thought it still shifted with the other books not against it.
I'm not sure how big a factor this is, but if your currency is worth more in outcome A then outcome B, then real-money prediction markets make A seem relatively more likely.
Polymarkets is, what, USDC? If we assume that's worth the same as USD, the S&P 500 went up after the election. If traders expected that to happen (I didn't), then a Trump victory being "overpriced" is what we'd expect, though I'm not sure how much overpriced - looks like the index is up about 3% but idk offhand how to turn that into prices on prediction markets. (Supposing the market thinks there's an X% probability of a Trump victory, and value of USD in a Trump victory versus a Harris victory is Y% higher, we can probably get the market prices - but we know neither X nor Y, and surely we can't figure out both of them from the market prices. Maybe we can get plausible bounds?)
And betfair and smarkets are in GBP, where it looks like the FTSE all-share went up the day after but then back down, and I dunno how to interpret that.
At any rate, this is one theoretical reason to expect play-money markets to be more accurate.
Price is an indicator of how much value the buyers and sellers place on a thing.
If something is offered for sale, and the offer is taken, by definition the thing is not "mis-priced." Value is subjective and differs from person to person, which is what makes exchanges desirable.
The value of "Trump wins" shares on a prediction market was in the buyers' and sellers' expectations, not in how the bet resolved. Post-transaction regrets, absent force or fraud, are irrelevant to price as a marker for value.
My issue is that there is a serious disanalogy to the coin case. In that case, we're trying to adjust our credences to match some chancy disposition of a coin. There is, in that case, a "true" probability, and we want our credences to reflect that. In the election case, it's a mistake to think of the election as truly random binomial distribution with p = 0.5. Now that Trump won, I believe with something like 99%+ probability that if we reset time to the second the polls opened, Trump would win again, and again. On the p = 0.5 theory, if we reset time, he'd win half the time. No, we learnt the actual, non random state of the electorate. In such cases, it seems to me that higher credences (i.e., polymarket's higher odds) are very good for their theory of the case.
Analogy: You have two expert paleohistorian friends. One is 50% confident in asteroids-killed-the-dinosaurs theory and one is 90% confident. You can't tell who to trust, but the 50% gut has a general better track record. Ultimately, you come down (due to your ignorance) with a 50% credence. If the asteroid theory is subsequently conclusively demonstrated, I would *massively* defer (with no other information, and with no expertise of my own) to the 90% guy's general understanding of that whole situation.
Totally agree that a sample size of 1 (N=1) does not cause us to update much a prior probability that was around 1.
Even if Theo was consistently beating the markets by making lots of bets and winning in the long run, he still could have lost on Trump by random chance. (I sometimes do a bit where I argue that DARPA should derisk scientific research funding by diversifying grant money across multiple research projects in the same basic area; DARPA does, in fact, do this). You have to split your money across a lot of smaller bets each of which has > 0 expected utility.
There’s not really any objective sense in which the shares can be mispriced, because there’s no “real price” to compare to. To the extent that there is, the “real price” was 100 cents, since that’s what they ended up being worth.
I think you mean something like “the best knowable guess for Trump’s victory chances was 50%”, but “knowable” is doing a lot of work there; an omniscient being would’ve probably put Trump’s victory chances at almost 100%.
I don’t think you can score particular probability estimates; you have to score over many events to get something meaningful.
It seems odd to pick a case where you ended up being wrong and say “no actually ex ante I was right here”. There’s not really a definition of “right” that can support this, and you just end up in a tangle of excuses. The point of being “right” is to make correct predictions, but this prediction was wrong.
Exactly. There are only correct prices - and this correct probability estimates for outcomes they haven’t happened yet - if there are correct values.
I think there are correct values, which I get makes me a minority here, but it’d worth nothing that “there are no true values” appears to be a phenomenon only in the 20th century and (in my opinion) the cause of so much of the chaos in the world today.
Just because there are correct values doesn't mean we know, or can know, those correct values. A particle has both a position and momentum at any given point in time, but you'll never know both, and arguably will never know either depending on how rigidly we define a "point".
> the cause of so much of the chaos in the world today
>I think there are correct values, which I get makes me a minority here
It might also be the case that there are correct values, but, in most realistic situations, it is unrealistic to expect to be able to calculate them.
Look at what happened in the lab leak v. zoonosis debate. For a _retrospective_ analysis, on a _past_ event of the virus coming from a lab or from a wet market, with both presenters doing their best to present all of the evidence they had, there was still a factor of 50 difference in the odds calculated by the judges (even excluding the presenters).
> I think you mean something like “the best knowable guess for Trump’s victory chances was 50%”, but “knowable” is doing a lot of work there; an omniscient being would’ve probably put Trump’s victory chances at almost 100%.
I think you're leaving out "as of October". I think it's perfectly reasonable for someone to make the argument that an omniscient being might put Trump's chances, relative to October, at below 100%, and if we argue that "no, the omniscient being would know the future" then you'd also be trapped into arguing "this same omniscient being would, in October, place the odds of yesterday's Powerball drawing being '12 17 37 58 62 [4]' at 100%, ergo the oddsmakers were wrong in saying it was a 1 in 292.2 million longshot", which is just silly.
> It seems odd to pick a case where you ended up being wrong and say “no actually ex ante I was right here”. There’s not really a definition of “right” that can support this, and you just end up in a tangle of excuses.
Of course there's a definition, and it's trivially simple if you actually understand the actual arguments actually being made in actuality.
If you say "the odds of a standard Vegas roulette wheel landing on Red are 50-50", and I say "no, I believe the odds are actually 47.37-52.63", and we go to Caesar's Palace and spin a wheel, and it comes up Red, I was still right.
> The point of being “right” is to make correct predictions, but this prediction was wrong.
Then by your definition, the insane man on the Strip who told both of us that the wheels at Caesar's Palace have a 100% chance of landing on Red because that's the Lizardman's favorite color was right. He was not right. His (implied) prediction came true, but he was not right. That's not how right works.
Is it worth some discussion about what it means for an election market to be accurately priced, and relatedly about where the sources of randomness are?
Particularly when contrasted with a sports match for instance. If we hypothetically say that a football match is perfectly priced just before it starts, then that is the best we can do obviously, by definition, and most of the randomness is yet to ‘take place’, in the goings on of the actual match. If we repeated the game 100,000 times we would see results that agree in frequency with our pre-match ‘true price’. In this sense, the result of a single occurrence gives very little information about the price.
What would it mean for an election to be perfectly priced on the morning of the election? By this time, it would be reasonable to assume that most people have made up their mind on which way they intend to vote. As such, this source of randomness has been largely removed. In a certain sense, aren’t the voting numbers already largely ‘decided’ by the morning of the election? And most of the randomness left comes from our inability to know and sample this perfectly. But in a hypothetical world, the one in which the ‘true price’ lives, we are aware of everyone’s voting intentions in the morning. Shouldn’t this be taken into consideration when we talk about accuracy of the prediction markets? What do you mean by the statement that you now think there is an 88% chance Meticulus was more accurate?
If we repeated the day of the 5th November 100,000 times, wouldn’t it be very likely that Trump wins a very large proportion of these elections. In this sense, doesn’t a single occurrence give much more information about the price than a single occurrence of the football match. In it, something is revealed to us that was in a certain sense already true (what the state of play was at the start of the day).
Therefore, couldn’t we argue that the true price on the morning of the election of Trump winning is much closer to 100%? (Of course there is still a certain amount of randomness from those undecided, people actually making it to the polling station etc). Therefore, how can we argue that a market estimating, on the morning, the likelihood of a Trump win at 50% is more accurate than one that had it at 60%?
"At exactly midnight on Tuesday, I pressed the “go” button for the final time on our election model this year. I knew it was going to be close. I felt like I was spinning a roulette wheel. (Appropriate, I guess, in a year when I published a book about gambling.) We’d decided ahead of time to run 80,000 simulations instead of our usual 40K.
And after 80,000 simulations, Kamala Harris won the Electoral College in … 40,012 of them, or 50.015 percent. The remaining 39,988 were split between Trump (39,718) and no majority — a 269-269 tie — which practically speaking would probably be resolved for Trump in the U.S. House."
Maybe he should have run the 100,000 after all! Reality - how rude of it not to line up neatly with the beautiful mathematical models!
I'm not mocking Nate Silver here, by the way; he's been one of the best at covering this entire election and trying to parse out what polls are really saying, what bias is there, what weighting should be given, etc. And he did say that in the end, the result comes down to people going out and voting, and not what the polls predict.
But Nate's model is built on polls. There's fundamental uncertainty baked into it.
If the world woke up on Tuesday, November 5th 2024 100,000 consecutive times, lacking any 4th-wall-breaking awareness of their predicament, I think it's a safe assumption that far more than 50,000ish of those Tuesdays would end up with Trump winning Wisconsin.
Nate's model is based on very limited information. If we hopped in a time machine to 12:00 AM November 5th 2024 80,000 times, we would watch Trump win ~80,000 times in a row because in reality voters had already (and would already, based on the information they were fated to consume that day) decided their votes, and the votes would lead to a Trump win every time. The true value of a Trump win contract was ~100c and a Kamala win contract was ~0c, and in that sense everybody mispriced it heavily down, but Polymarket less so.
A little unfair. You can't tell from one guy making one big bet and winning if he's right or just lucky, so saying that the market was overpriced is the correct view. That's not mere cope.
I don't have the shining faith in prediction markets that others do, but at least this way they're getting a trial run and seeing how things can go wrong, before anyone tries using them as "let's set government policy that will affect the real world running of the country by the prediction market! what could possibly go wrong?"
His priors don't seem to be based on anything, trying to prove anything from predictit is silly, they have a betting limit per trader (which implies that whales are better for market accuracy...)
Market was right, Scott was wrong. he retroactively sets unrealistic priors so he doesn't have to update. My prior that my uncle is a fish is 1.00
It's so frustrating, I learnt about prediction markets from him, long brilliantly written essays how they work, why they work etc. and then in their biggest test, they were right! but because he assumed that it should be more like 50/50 we can now throw everything out the window
Scott says what he bases his priors on (previous competitions), so its clearly not true his priors are based on nothing.
Furthermore, using calibration city for this is silly as it does not use the same set of questions for comparison. Manifold has thousands of "daily coinflip" and similar markets, it is very easy to reach perfect calibration on a coinflip (just predict 50%), while metaculus only has "interesting" questions (ie. A random user can't just make some easy to correctly predict question).
Personally, I use both Metaculus and Manifold and because I have a consistent profit on Manifold, I feel qualified to say that Metaculus is better.
Metaculus also has 1/10 of the number of bets analyzed. Compared to others. No wonder it is “less calibrated”.
I don’t know if Metaculus is accurate or not. My gut feeling is that the money markets are more accurate. But I do think polymarket in October was wrong.
Frankly, setting government policy by prediction markets is a dumb idea; they're only one tool among many. Of course IRL deep-pocketed interests would try to manipulate it.
I think the biggest real upside of futarchy is you can make (in the appropriate company) jokes about how there's no real argument intersex people are any better at government than anyone else.
('Futa' is also short for 'futanari', a genre of hentai involving people with external genitalia belonging to both sexes, often in anatomically impossible ways.)
My well of English has been permanently defiled, to quote "His Last Bow", by early exposure when I did get on to the innerwebs to the less savoury parts, hence every time I see "Futarchy" my mind automatically goes "Isn't that the Japanese porn thing about?"
Really do need to come up with a different name for it!
"(T)he amount of money someone puts into Polymarket is exactly proportional to their trustworthiness. This is a fair approximation for the Polymarket algorithm to use; it’s the assumption that drives prediction markets"
I think this is the problem right here. Prediction markets (all markets, really) are not working off *trustworthiness*, they're working off *confidence*. Theo was very confident that his bet was correct so he put a ton of money on it.
People can be very confident that their prediction/bet is right, and put a ton of money on it, and lose their shirts. We know this from ordinary gambling and betting. Dressing up prediction markets as "one of the single best sources of truth" is what is blinding rationalists here.
Bob has a lot of confidence in his bets (let's call them that, if we're putting real money with them, instead of predictions) so he puts a lot of money on them. All the market can tell from that is that Bob is very confident. For the market to judge Bob as *trustworthy*, it (and we) would have to follow the pattern of his bets over time to see if Bob is generally right (better than chance) if he bets on "X will happen, Y will not happen".
If Bob gets a lot right at better than mere chance levels, then we can say Bob is trustworthy if he thinks Larry the Cucumber will win the presidency in 2028. If Bob makes one big bet, wins big, and peaces out, we have no idea if he's smart, or just lucky.
Theo seems to have had a process at work for his confidence that people were underestimating Trump's chances and overestimating Harris' chances, more than just going on a gut feeling.
But the problem again is that "oh but won't the smart money correct for the one big bet?" is that the people *do* think they're smart, so they *overthink*. "Hmm - this guy must be an idiot, why is he punting so hard on Trump? All the polls say it's way too close! Unless... does he know something I don't? Has he got inside info or a really stunning algorithm here?"
And since people on prediction markets tend not to be the average punter, they do tend to assume "other people on this market are running Bayes on all their decisions" and so they follow the money. If Theo bets big, maybe I should be betting big too. Even if it's contra the run of the polls, and maybe *especially* if it's contra the run of the polls, because prediction markets are for smart people and Trump supporters are not smart people, all the polling done about who is the most educated shows that! So he can't be a MAGA-head, so he must know something!
Human nature is a wonderful thing.
As for trustworthy smart people making predictions, Ann Selzer seems to have blown up her own reputation with her Iowa poll. Nate Silver was diplomatic about it, because plainly she is a big name as an expert whose polls are highly accurate for her area of expertise (how elections go in Iowa), but even he had to go "Nah, this is nuts":
"Releasing this poll took an incredible amount of guts because — let me state this as carefully as I can — if you had to play the odds, this time Selzer will probably be wrong. Harris’s chances of winning Iowa nearly doubled in our model from 9 percent to 17 percent tonight, which isn’t nothing. Polymarket shows a similar trend, moving from 6 percent to 18 percent after the survey. But that still places Harris’s odds at around 5:1 against.
The poll has a reasonable sample size: 808 likely voters. Still, the margin of error for the difference separating the candidates in a poll of that size is ±6.6 points. That means in theory, in 95 out of 100 cases, the “real” number should be somewhere between Trump +3.4 and Harris +9.6 if Selzer had surveyed every single Iowa voter instead of just an 808-person sample.
The theory, however, doesn’t capture all possible sources of polling error — especially the fact that the overwhelming majority of people who pollsters attempt to contact never complete the survey. And we ought to be good Bayesians here: even if Harris has a very good night on Tuesday, winning a state you lost last time by 8 points is a big ask."
"A new Des Moines Register/Mediacom Iowa Poll shows Vice President Harris leading former President Trump 47% to 44% among likely voters just days before a high-stakes election that appears deadlocked in key battleground states.
The results follow a September Iowa Poll that showed Trump with a 4-point lead over Harris and a June Iowa Poll showing him with an 18-point lead over Democratic President Joe Biden, who was the presumed Democratic nominee at the time.
“It’s hard for anybody to say they saw this coming,” said pollster J. Ann Selzer, president of Selzer & Co. “She has clearly leaped into a leading position.”
"Former Republican President Donald Trump has once again cruised to victory in Iowa, tallying his third presidential win in the Hawkeye State — this time over Vice President Kamala Harris.
In the end, it wasn't close.
The Associated Press called the race about 9:40 p.m. Tuesday. According to partial, unofficial results released by the Iowa Secretary of State late Wednesday morning, Trump led Harris 56% to 43%."
In conclusion, nobody knows nuthin', I wish I had gone with my gut and punted on Trump, and stick with the established bookies if you want to win big.
I think Theo knew something. I hadn't heard he commissioned a private poll until recently, even though people had been talking about the French Trump whale earlier.
So what would people do to invest for the second Trump term? I'm thinking buy a house, since the tariffs (and deportations) are likely to cause inflation, and stocks over bonds, but I'm sure people here have some more thoughtful ideas. Energy? Finance? Small-caps?
A good bet if you think tariffs will be broad-spectrum would be to buy stock in companies that produce things in the US, as long as they also buy their raw materials/parts in the US. Their costs won't increase, so their pricing should get better relative to the competition, increasing sales and thus profits.
I have no particular companies in mind, though. We import a lot of stuff, and commerce is complicated.
Elections are not stochastic experiments. Market prices are not probabilities. The analogy of "voting with dollars" is more apt. If Joe's Exxon has gas at $3 on the corner of Fifth and Main, but Sally's Mobile has gas at $1.50 on 30th and Main, is anyone prepared to make a comment about the probability of finding gas at $2 at the corner of 2nd and Main?
Market prices give us information about what other people think the value of some good is. Mr. Market, as Benjamin Graham called it, has a mind of its own. Mr. Market combines the dollar votes of people into bids and asks. If someone comes along and says "the market price is wrong," my question is "compared to what."
Every day any number of people buy and sell stocks on the world's exchanges with a belief (or is it just a bet) that the price of some company is "too high" or "to low." Who knows how much of the action is "smart money" vs "dumb money"? Mr. Market doesn't know and doesn't care. Every dollar is just a vote. How are prediction markets any different? The more dollars voted, the greater weight the vote carries. That's the top, bottom and both sides of Mr. Market.
The Efficient Market Hypothesis proposes that market prices reflect all the information available and do so really quickly. Are prediction markets efficient? "Compared to what" is the only question that makes senses to me.
Scott, your coin flipped 3,000 times, and came up heads at least 2,000 times. It’s clearly not 50:50.
A good test for whether it was a fluke is looking at county-wide differences from the 2020 vote. If it truly was 50:50, we should have seen noise. But instead we saw a sea of red, because almost whole country shifted rightward.
I think that’s good evidence that the whale understood reality better than the play money markets, and the true probability of Trump victory was higher than 60%.
True, but are they *that* highly correlated? I guess the key question is: "in what percent of Nate Silver's simulations were the county-wide differences as skewed as the actual result?"
I feel like you're answering the wrong question. It is not "given the actual votes on Election Day, what are the odds of Trump winning?" Those are obviously 100% (minus whatever negligible amount you'd like). The question is, given the public and semi-public information available prior to November 5th (i.e. polls, analysis of the fundamentals, the method Theo was using, etc.), what are the odds of Trump winning? In other words, if you can imagine an equally likely outcome that looked the same beforehand but shifted the other way, then the amount and uniformity of Trump's win in practice does not really matter.
Another minor quibble is that since the Democrats' Electoral College advantage diminished by (I believe) about 3 points between 2020 and 2024, and the Ds won in 2020, a relatively uniform rightward shift is not incompatible with odds of 50-50.
I agree with your overall point about real-money markets doing weird things because arbitrage is harder.
FWIW I bet about $1500 and made about $300 as of this writing on Predictit. Last time I bet about $1000 and made about $100. I couldn't really use Kelly because I wasn't sure what my actual chances of winning were--I was making small bets and moving sides accordingly as news came in (same as everyone else on that site, I imagine), so stopping to record it would have cut into my winnings--I was arbitraging between 'Kamala next president' and 'next president is a woman', for example, which doesn't stay different long and often gets negated by the difference between buying and selling prices (I believe real traders call this the bid-ask spread?). I'm sure one of you folks could write code to store all movements or something.
I suspect the money on Kalshi's much smarter and I am not going to do well so I am not planning on a side hustle as a trader--something like 98% of people lose money on that. (Also, as you may have guessed, I have something of a problem with excess risk aversion.) But I thought people might find the experience interesting.
I couldn't bet on who specifically would be the next President on PredictIt because they weren't taking more bets there, so I had to bet on whether the next President would be a woman (which was priced multiple cents lower than Kamala).
Wow, I remember that from the last election. I think I was already in it at the time.
If I were starting out I'd probably go with Kalshi. I had $1000 in Predictit already (sunk cost) and the number of markets was small enough I could monitor them all. I also figured there was enough dumb money I could still have a small edge; Kalshi was going to be full of smart rationalist people with software they coded to calculate the optimal price based on AP feeds or something.
"Suppose you have a coin. You think there's a 90% chance it's fair and a 10% chance it’s biased 60/40 heads. Then you flip the coin and comes up heads. What should your new probability be? You would solve this with Bayes’ Theorem; the answer is 88% chance it’s fair, 12% chance it’s biased."
This is flawed, because Trump did not just win, he won by A LOT.
Prediction markets didn't give the odds of Trump winning by a lot as opposed to just winning, but we can infer the odds they would probably have given, from the odds they give for Trump just winning.
Imagine there's a guy who represent Polymarket and another who represent the others. If we had asked them what are the odds of Trump winning by this much (again, as opposed to just winning), we would get answers that differ greatly, because of how bell curves are shaped.
Not a small difference like 50% as opposed to 60%.
Therefore I think the outcome vindicates Polymarket by a lot more than 2%.
There were definitely sports betting markets for it, I remember watching that particular one change over the course of the night. I don't know where it started before counting began, but it went from a reasonably long shot to a dead certainty as more votes were counted.
I think this post illustrates exactly why the whole prediction market idea disgusts me: like most of capitalism, it exists in a world without any connection to morality and is only considered within itself and the milieu of money, regardless of the world actions that it is reflecting. Bleh.
I know you (Scott) love it and are way into the idea of future prediction, but honestly every time I read one of your posts about betting on prediction, it somehow turns my stomach.
But what do I know, I'm just a penniless old guy who seems to have nothing in common with people who have and play with money, more and more as the years go by. I guess your world is entirely different.
For example, I read this: "After Republican candidate Donald Trump won the U.S. Presidency, the price of Bitcoin saw a sharp upward move that saw it hit a new all-time high at $76,482 as market optimism over the potential introductions of a friendlier regulatory environment in the U.S." and I realize that the "friendlier regulatory environments" are exactly this same playground, and also exactly what I believe are responsible for huge wealth disparities in the world, leading to scarcity, fear and ultimately to autocratic governments. Yuck.
Wealth disparities are caused by the existence of huge amounts of wealth. If no wealth existed, there couldn't be any disparity. The poor in a wealthy country like the US are far wealthier than the poor of the Third World, which is why people all over the world want to move to the US. Rather than scarcity, there is increasing abundance (hence the prevalence of obesity, which is now being fixed by semaglutides).
When I sleep I also ignore the entire context of human society. Is this also immoral? This question probably sounds like a lame gotcha to you, but it is absolutely honest: I have no idea what is the meaning of your previous comment, and I would guess a lot of of other people are also confused, so feel free to elaborate for more clarity.
I'll try. What I mean is that playing with money regardless of the context (in this case betting on election outcomes that will have global affect) separates the monetary aspect of it from the greater sphere of the subject, which can only insulate the players from being effected by the good or bad that comes from it. In this particular example, I see it as the same as, for example, investing in companies that have detrimental environmental effect upon the world but being happy about making money, disregarding that these companies and indeed the entire stock market have an adverse effect on wealth disparity overall. Or perhaps like betting on dog racing, disregarding the well-being of the dogs themselves.
I'm both jet-lagged and extremely concerned about global repercussions to the *situation* that these bets are riding on, so perhaps emotional, but seriously: my reaction to reading the post was nausea.
I find that those who play with money almost always disregard the very fact that their "play" is made available to them by a constant inequality in wealth among people, and most people ignore the context in which that wealth exists in the first place.
If that doesn't explain why I posted this, I'm not sure I can.
Seeming example of this: https://x.com/RichardHanania/status/1842940892048703513. It's easy to gesture at issues without even roughly quantifying how those issues override the benefits. While it's conceivable that betting markets affect outcomes negatively since perhaps by introducing valuable information, someone could make a decision on its basis, that's ultimately just an argument against information! Sure, a campaign could adjust its practices in light of prediction market behavior, but the same goes for punditry, inasmuch as the punditry contains a useful signal.
Similarly, the follow up comments note that capitalism has led to undesirable outcomes like income inequality. Sure! You can't have much inequality, when everyone is a subsistence farmer. Everything is bad for someone (vaccines are bad for business, if you're a coroner). And it's trivial to string words together alluding to the downsides of anything, without making a meaningful argument about that thing actually being bad.
Hi, I think you'll enjoy my YouTube channel. I won't share it here so as not to spam but I get exactly how you feel. Clicking my avatar will get you to my substack to my youtube. Check out especially those videos that are more than a month old. I engaged in a serious experiment and the results are such that I'm undergoing a radical transformation and my future videos or whatever are likely/hopefully to be far less moral polemics with practical intent. The experiment didn't result in the vindication of goodness and truth that I preferred it would but I am very happy to have done it so thoroughly that I can now rule it out. I have something much better and more fun in mind.
You may be able to learn things from my experiment too If you do I would be pleased to hear from you.
Frankly, I'm not certain that all economics isn't a zero-sum game. Profit implies loss. Maybe there is no such thing as "honest" profit in any situation, someone or something loses. For example, in extraction-based capitalism, they don't even count the damage to the environment as a "cost", hence it's all "honest profit." Similarly much of globalization has meant more diesel ships at sea, worse working environments *somewhere*, etc etc, for that "honest" profit.
"Frankly, I'm not certain that all economics isn't a zero-sum game. Profit implies loss."
Consider the amount of wealth in the world 50, 100, 200, and 1000 years ago, and the amount of wealth today. Where are you thinking all that wealth came from?
as above: "Trade offs and externalities are a thing, but guess what, damaging the environment predates modern capitalism."
edit in an afterthought: where do I think all that wealth came from? off the back of people and places that had no say in the matter and got nothing in return. It didn't come from nowhere.
>where do I think all that wealth came from? off the back of people and places that had no say in the matter and got nothing in return.
That is _spectacularly_ wrong. Almost all of the increased wealth from the last few hundred years came from technological innovation, using the _same_ ores and land and worker-hours to fulfill more human wants and needs.
I don't understand. "Same ores and land"? They aren't being recycled? The increases in extraction and harvesting of environmental resources, especially in the last few hundred years, are now causing environmental and climatic catastrophes, causing more human suffering at the trade-off of more monetary wealth. There may be more people who have more money, but there are increasingly more people in general, which means more suffering despite the apparent wealth of "things". You're saying that wage workers now are better off now than slaves, sure, but having an iphone doesn't make a person "better off", just a different level of slavery. (Ok, that's hyperbolic, but you understand.)
I mean, I know Hans Rosling's work, but his work itself is paradoxical—he tracks the decrease in bad and increase in good, but ignores increases in bad. Just saying that tech innovation brings up the overall level of human wealth ignores the environmental and psychological costs.
Profit does not imply loss. The most basic form of profit is, a farmer with a tiny bit of land and a kernel of corn plants the kernel, nurtures it for a season, and harvests several ears of corn. He has made a profit. There has been no loss.
If you insist on money changing hands before it counts as "profit", the farmer sells half the corn for three obols. He has made a monetary profit of three obols. He has made a profit of three obols. And someone else has bought dinner. There has been no loss. At some point the farmer will spend his obols on a spool of yarn. Now the yarn-seller has made a profit, and the farmer has bought the makings of a new pair of socks. There has still been no loss. Only the gain of some corn, some yarn, and whatever the corn-buyer made and sold to get the money to buy the corn. The world, and everyone in it, has become richer.
It is possible for one person to profit from the loss of another. It is not *necessary* that he do so, and by blinding yourself to all other forms of profit you miss what the rest of us are trying to do here. Which is, for the most part, trying to find a way to buy accurate predictions of important future events. That's profit, and it requires no loss.
This is an example of what I was initially posting about, what you are describing is only the monetary value of the things involved in an entire system. The farmer's time and effort (and I don't mean time=money, but time as experiential, an element of their life—and not labor as work=money, ie wages, but labor as effort that might be applied to other usages. Maybe the farmer really was only happy when they danced, but couldn't due to the needs of raising the crop), plus of course the water, the land, the entirety of the environment involved that was being changed to a crop bearing strata.
There are enormous amounts of things "lost", things that are never counted in the economy of money. The economies of the "useless" (for example the arts) make huge impacts on the qualities of life. That is the context I am speaking of here.
I think this is why I wrote that I "am not convinced" that economics—as a gestalt— isn't a zero sum game, resulting in equilibrium like any other physical process. Discounting all variables that aren't able to be quantified to dollars *at the moment of transaction* ignores a great deal of the overall system. It's well known that extraction-based capitalism ignores the externals, like the environmental costs, but they come back to haunt us in time. Beyond what I mention above regarding the farmer, there is the use of land over time: if the farmer plants corn over and over, the land ceases being arable due to the monoculture sucking out nutrients. I feel like the entire game of money ignores so much else.
->This is also why I flinch at these posts on prediction markets. "Will somebody use a nuclear device in 2025? Yay! I win some money!"
If the "loss" you're concerned about is that we can't all spend all our time dancing or whatever and having food, clothing, shelter, etc, miraculously appear for us, then I don't think that's a useful framing. You can't lose what you never had, and we never had those things. Well, not since we were kicked out of the Garden of Eden, I suppose, but it's hard to blame that one on the profit-seeking capitalists.
I don't really bet on or use prediction markets, but I still find them useful. I mostly comment and read forums in fairly Democratically-aligned spaces, and I saw lots of information last week that was hard to gauge in terms of its accuracy. People were spending hours dissecting early vote results, the Seltzer +3 poll, etc etc. Some people were predicting a Harris landslide. See here [1] for a typical example of this. Whenever I saw this, rather than trying to engage with a complex statistical argument, I looked at 538 and PredictIt, and found that the odds were still about 50/50. An enormous amount of election coverage is essentially bullshit that doesn't matter or isn't predictive. This is why I find it useful: as a summarization of what's happening. I think it's important to separate the message and the messenger here.
Agreed! I, also, am just an onlooker towards the markets, but I find it very useful when someone is claiming "X is inevitable" or "Y is impossible", to look at the markets and be able to say: Well, this market says the odds of X are 40:60 and that market says 55:45, so people who have skin in the game are betting reasonably close to 50:50. As you said,
>An enormous amount of election coverage is essentially bullshit
I'm not sure I agree. If the election indeed would have been close (as in 2020), there would indeed have been no way to predict it over chance. But this was a shellacking. There were signs that this was coming, and there was presumably a way to discern those signs from the Kamala noise.
I'm not sure a biased coin is the right model for an election. The main characteristic of a (biased or non-biased) coin toss is that the physics of its flight path are so complicated to compute that even a human with a computer cannot predict its outcome.
But surely, an election result is in principle more predictable than that? With an expensive effort you could send people into Iowa, to get to know the people there and understand their values and beliefs and predict successfully how they will vote.
I'm not saying Théo's efforts were on this level, but I don't know, one is at least on shaky ground when criticizing the man who commissioned his private polls and who turned out to be right. I don't find it so unthinkable that the incentive of winning $30 million made him do a better at getting good polls than anyone else. (Hmm, though what about Wall Street? Wouldn't they also have a large monetary incentive to find out who will win?)
Also, I fully agree that Polymarket didn't have *enough* sophisticated traders to really trust their odds, but given that the most sophisticated trader made it go towards Trump (the winning outcome), it could well be that with more traders of higher sophistication, the market would have been even more skewed to the winning outcome, because, again, I would argue that an election outcome is something that is knowable, as opposed to the outcome of a coin flip.
This is a general way to model how much information you do and don't have - with a fiar coin you can't (barring an implausible level of physics modeling) reduce entropy below one bit of uncertainty, but with an election good polling and modeling often can - but still not to zero (no one predicts every election exactly right, there's always unknown factors).
"But surely, an election result is in principle more predictable than that? With an expensive effort you could send people into Iowa, to get to know the people there and understand their values and beliefs and predict successfully how they will vote."
This strikes me as about as impractical as solving the Navier-Stokes equations and setting up an elaborate diagnostic suite of flow-visualization and -quantization instruments to determine which way the coin will flip. In principle, at least in the classical limit, the coin-flip is absolutely deterministic. In practice, it involves forces that we can only assess collectively and statistically.
The same is true of elections, with any plausible level of election-forecasting resources.
I know Scott says the other big money markets were arbitraged which is why they were closer to Polymarket than PredictIt was. However they didn't seem to be arbitraged very well. At times after the French trader Theo entered, there were extended periods where there was at least a 4% difference between Polymarket and Betfair. How is that big of a spread possible? Commissions can't be that high can they?
Bertie has a 2% commission on winnings, Polymarket doesn’t have any, but there are costs in moving money in and out of crypto. There is also currency risk and opportunity costs of locking the money up in bets, increasing the further before the election you place the bets.
I believe the markets on Polymarket did correct on election day, completely removing the Theo effect. The popular vote market was trading at 73-74c Kamala the afternoon of election day, and the president market was trading at 39-42c.
I will say that Polymarket has always had a Trump bias, so many traders on Poly were expecting a Trump surge near election day. On November 2, the election market was trading at around 45-55, then on Monday and Tuesday the Trump surge happened (Theo barely traded on Monday).
An election isn't like a coin flip. It's not a binary yes/no. The fact that Trump didn't just win but swept all the swing states should increase your trust in Polymarket a lot more than if it ended up being a close win for Trump.
I don’t think this is right even though it’s intuitive. swing state sweep was among the highest probability outcomes for the Silver model, because it expects correlated polling error like we saw. Very impressive sub-prediction! Overall probability was still 50-50.
Isn't it obvious in hindsight that such an error was more likely to occur in favor of Trump than Harris? To me, the Polymarket betters seem smarter than Silver for having predicted when he didn't.
AFAIK he didn't give odds for a swing state sweep, but the single most likely outcome according to his model was Trump 312 - Harris 226 (which matches the NYT's current projection for the final result).
The night before, it had about a 28% chance of popular vote victory and 20% chance of every swing state going Trump. Both of these came true. However, these numbers are basically the same as Nate Silver's modeling.
I think this analysis ignores the "wisdom" part of the wisdom of crowds.
When you looked at something like Nate Silver or the RCP polling average or 538 you would come to a number around 50%. But those numbers were only based on polls.
Trump had beaten his poll numbers twice already, Silver and others assured us that this was a random coin flip and could just as easily go the other way for Harris. But there is no real reason to believe that. The data on polling errors is quite short and decidedly mixed and does not control for the fact that usually the candidate changed
I think introspection tells you the errors with Trump would not be completely fixed. When I model they type of person that works for a polling firm I don' think of a person who makes big adjustments to methodology but instead makes gradual adjustments. This type of data was outside of the "numbers".
My view is that you had to start with the Silver numbers as a baseline and shade them towards Trump. I was 60-40 Trump for most of the time leading up to the election. Polymarket over shot that a couple of times and I would have bought Harris but would up right at that number on election day.
A lot of this analysis falls apart when we look at Kalshi, which is US-regulated, open to entities like trading firms, and has done nearly $1B of election market volume in the last month or so. Kalshi's presidential market orderbook in particular has been consistently deep and traded entirely by US individuals *and entities* with deep pockets and high confidence (SIG being perhaps the most widely-known market maker on the site). As much as I like the idea that these are niche hobby markets for wealthy nerds, it's not really the case anymore. We're really entering a new age of prediction market accuracy.
Interesting reading this, just a completely different mindset and view of the world to my own.
My strong default would be the real money markets would be far better as in every other area of life, putting money on the line forces people to be more rational, and those that aren’t don’t generally last very long (poker, hedge funds etc). So very surprising to see research showing non money markets doing better. My best guess is this might be when the money prediction markets were very small and illiquid, but incredibly hard to see something outperforming markets on the US election, with probably a few hundred $m bet in total.
In the medium term if it is true that the non-money markets outperform, imagine this will disappear as people bet on the money markets using the information from the non-money ones.
Regarding hedge funds - I guess they aren’t called hedge funds but know a few people who work for companies that make money by using data and analytics to place bets on sports and politics.
Finally take issue a bit with the coin flip metaphor (although agree shouldn’t update much from one event). With a coin flip (in theory) you can’t figure out before the toss anything that will help you predict it. In contrast the morning of an election (particularly one with a decent margin like this one) the result is pretty knowable if you had the brains and maybe a couple of hundred million to figure it out (I would guess you could call it the right way maybe 80-90% of the time), with the major sources of unpredictability being weather, last minute political surprises and some variations in turnout that might be hard/impossible to predict in advance. No one actually does this but some people get closer than others (private polling, new data approaches, inside information etc). So the fact that the money prediction markets got closer to the ‘real’ odds of maybe 85-15 does count for something.
But do agree that this is only one event so doesn’t count for much, and also only applies to an election where there is a decent margin of victory. In contrast in 2000 or 2016 the ‘real’ odds would have been much closer to 50-50, as they were so tight.
"My strong default would be the real money markets would be far better as in every other area of life, putting money on the line forces people to be more rational, and those that aren’t don’t generally last very long"
OK, so what's the average trading lifespan of the average prediction-market player? Polymarket is only four years old; has there even been time for the market to flush out the irrational players?
And if you think that's inevitable, check out any casino in Las Vegas. Lots of people betting lots of money, most of them losing, very very few of them truly expert card-counters or poker sharks who can realisticaly expect +EV, Has betting real money "forced" them to be rational?
Arguably they are being rational, in that they are enjoying themselves and paying for entertainment that one enjoys is rational. But once you acknowledge that people aren't always trying to Maximize Moar Moneyz, that the point of making money is that you can then *spend* it on other things that they value, then you can't trust a mechanism based on the postulate that everybody is trying to rationalize their expected monetary return. You can't even trust "if they don't rationally maximize their expected return they'll go broke and purge themselves from the sample", because almost all of them will make almost all of their money elsewhere and can afford to lose indefinitely in this market if they're having fun with it or otherwise securing some valuable-to-them reward.
I think you're seriously underestimating the amount of smart money that was available to correct any obvious mispricing. There are billions of non-Americans in the world, and many of us follow your politics (arguably too intently). We're much less likely to have relevant non-public information, but we have access to basically all of the public information. And we wouldn't have had to use Polymarket; if we bet enough to seriously move any real-money market, arbitragers would have showed up and done their thing.
>I have never heard of a hedge fund betting on a prediction market and my guess is that it would either not be legal or require too much compliance paperwork to be worth it. I hope this changes!
Apparently the main thing is just that they have better options. Even if they think they have uniquely good information on the result they can just invest/short stocks that they expect to go up or down as a result. And the downside risk is much lower
Polymarket in particular has a lot more third party risk, and systemic risk around crypto, than most funds would be comfortable with.
> Polymarket in particular has a lot more third party risk
Blockchains eliminate third-party risk. They're trustless: the smart contract always does what it says it will do. There is no third party to trust. (Of course there could be a bug in a contract, or something, but Polymarket has been around for a long time and there's a lot of free money for someone if they can find a bug, and they haven't so far.)
I was using the term slightly loosely. Yes on paper it eliminates third party risk. In practise looking at it from the perspective of a fund you have the questions of: How likely are people to continue to want to buy the token I get out at the end of it. How much do I trust that the person who wrote the contract did so competently and non-maliciously. How much do I trust that my belief that it is difficult or impossible to hack the contract is true? What legal protections if any do I have? And if I like the answers to these can I convince my boss and risk management team?
> How likely are people to continue to want to buy the token I get out at the end of it.
They use USDC, which is a regulated stablecoin collateralized by treasuries and similar which are held by a traditional custodian, so that's not a major worry.
Not sure I'd call Circle a traditional custodian since almost all of what they do is crypto related. USDC briefly lost its peg to the dollar when silicon valley bank collapsed, if I was a fund I'd consider that a problematic precedent.
That's not me saying that the risk is super high. But as a fund you would want to compare it to the other options you have for using your capital. If you can invest in more traditional implements that have the same correlation with the election result, why bother.
Who decides which outcome wins on Polymarket? Often it's this weird Rube Goldberg machine called UMA. A very clever system for reaching decentralized consensus about events, but also it's definitely weird and extra risk.
That may well be the biggest risk, but they haven't done anything clearly wrong yet... resolution of some of these questions can be gray areas, is the worst I've seen.
Yeah mechanisms are like that are fine when it's mostly a hobbyist thing. But if if anyone with serious resources is involved and has an incentive to cheat you have a problem.
For a uma you just need to buy up enough of the tokens that you can swing the result. Which is potentially easier on a contested issue like the election.
Hedge funds come in all shapes and sizes. I think that there's room for small hedge funds (a couple of dudes with a seven or eight figure chunk of money) to clean up in this field.
If I believed Scott's core claim here that fake-money prediction markets are systematically more accurate than real-money prediction markets then I'd be quitting my job right now and starting up such a hedge fund because trading the diff between real money markets and metaculus would be ridiculously easy.
The question is not whether it's possible to make money, it's how much money you can make at what risk level relative to what else you can do with your capital. It doesn't matter if there are $1 bills on the ground if I picking them up you lose the opportunity to pick up $100 bills. And given the current rates of growth of the stock market etc competition is high.
I don't know about specifically the play money markets, but plenty of people have made money in the past by taking 538 or other public models results and betting that. That's why the market tends to converge towards them outside odd situations
As a heavy prediction market user, I think this argument is directionally correct. Theo's market impact created a substantially wider gap between real-money and fake-money markets that never really closed.
This runs contrary to most people's intuitions about how markets work, but I think the important thing to understand is that 5% of edge on a one-off event (i.e. the additional "arb" that was created from Theo's impact) is just not that much. Furthermore, although this was a breakout year for prediction markets, there isn't actually *that* much money hunting for opportunities here. It would have taken tens of millions of dollars to reverse the additional spread that Theo created, and while there were some traders willing to do it, the total bankroll willing to correct Theo's impact was just not on the order of Theo's own risk appetite.
The one caveat I would add is that I think we should distinguish the claim that "Polymarket's odds on election day were wrong" from the claim that "Metaculus/etc's odds on election day were right". I think it's very plausible that real-money markets were correct prior to Theo's arrival, or at the very least that the true probability was somewhere in the 4-5% gap between real- and fake-money markets at that time. But given that this gap widened by another 5% or more, *and it did so purely as the result of one trader's impact*, I think it's very reasonable to claim that Polymarket was mispriced on election night in a literal sense.
Did PredictIt take a loss because there was more money on Trump than Harris? It seems like this wouldn’t be an issue if prediction markets were more like the stock market (finite amount of action available) than a bookie.
1. the best polls said the election was 50-50. Small random factors can move it either way.
2. the reality was that virtually every demographic, and in every swing state, Trump overperformed his last election. Small random factors made no difference at all. In theory, maybe all these decisions were made by voters on election day and pollsters couldn't pick that up ---but that seems highly unlikely. The clearest story is that the pollsters missed exactly the kind of thing they should have easily picked up. If some demographic groups were moving more toward Trump and others against him - that can be tricky to resolve. But when almost every group did that, how can pollsters miss it?
3. Before the election I wondered why more people didn't bet against the French whale, given where the polls were. But now, of course, the question is why more people didnt bet with him. I don't mean in the trivial sense (Trumps probability of winning is now 100). I mean, in a more efficient betting market I would expect such a broad based move to generate levels much higher than 60% prior to the election.
Yeah, how (why) did the pollsters miss it. It's not a single coin toss. It's like in every state they got it wrong. And the money markets got it right!
I don't know why you'd still refer to it as "mispricing", when betting against "Theo" would have lost money and moved the market away from what turned out to be the correct result. And perhaps I'm just ignorant of how many people are commissioning private polls, but that doesn't seem "replacement level" to me. How many other Polymarket bettors were doing that? It also seems incorrect to say that Polymarket absent Theo is "normal market operations" while with him it isn't. The market includes all the participants, including whales.
To steelman the original version of his post (which didn't realize that Theo had his own polls), though, one could claim that "normal market operations" have enough liquidity to prevent individuals from significantly affecting the markets, as a whole. They therefore more effectively approximate the aggregate opinion. While a few large players moving markets against the consensus of the rest of the market is possible and is consistent with normal market operations, in practice, markets typically more closely resemble the situation in play-money markets in which single individuals don't move the markets as much.
This would therefore not be a fundamental distinction between prediction markets and other markets, but instead just a feature of the liquidity of these particular markets due to the factors that Scott lists later in the piece which limit participation in Polymarket.
> I don't know why you'd still refer to it as "mispricing", when betting against "Theo" would have lost money and moved the market away from what turned out to be the correct result.
I beleive the idea is that there is some correct ex ante assessment of the probabilities given avalible information (public and private in this case), and that even though Theo happened to have won, the market still may have been mispriced. If someone offers you 60-40 odds on a fair coin to land heads, the market is mispriced — even if the coin does in fact land heads, meaning the bet loses you money and moves the market away from the outcome that happened in this case.
Someone says "I bet that coin will come up heads because it's a double-headed coin!", bets enough money that lots of other people say "he must know something!", such that the local prediction market gives 90% odds of the coin coming up heads.
The coin comes up heads. But on examination, it was a perfectly ordinary coin. Was the prediction market "mispriced"?
You can define "mispriced" and "properly priced" to make it so, but if you do you'll have defined a market that can't be determined to be "properly priced" until after the fact and can't be trusted to make predictions because it sometimes gives you 90/10 odds on a fair coin toss.
Scott is using the terms in a way that points to potentially useful results.
There is no coin that was examined to have 1 vs 2 heads after the fact. There were multiple polls and disagreements about which ones were reliable and what kind of bias existed in them. Someone already replied to me about the nature of uncertainty in this case, so I'll link to that https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76203508#comment-76266715 "Theo" really did go out and collect info that others didn't possess.
I think some of your arguments imply that you shouldnt listen to prediction markets at all. If you think prediction markets only work when they weigh people approximately equally, you should just ask people and weight their opinion approximately equally. If youre going to discount market movements without legible justification, you should just believe whatever is legibly justified to you. So, I dont think theres a reason to trust this prediction market less then usual. The track record matters, and arguments like the above may be reasons to trust them less *in general*, but I think arguments to discount a particular market based on how its going are almost always wrong.
You seem to believe too much that the consensus of the opinions of intelligent people must be correct, and not enough that the data is the way you find out what's correct. We know there are big problems with the polls: shy Trump voters have happened three times in a row now, and we've observed similar things going back to at least the exit polls on Obama's first election (when exit polls stopped being good predictors). We know that the excessive clustering of polls is statistically impossible, so the pollsters must be fudging them. The Polymarket Whale wasn't just a guy with a different argument that should be counted equally to other arguments, and he didn't just do a "private poll:" he commissioned a "neighbor poll" to correct for the shy Trump voter problem. You don't seem to believe fully in markets, when you say that everyone's "votes" should count equally, not based on how much they bet. The Whale had a history of successful financial betting, high conviction, and better evidence than most people, so his vote *should* count more. I do believe all those things make markets work better than majority wins. Your pre-election article struck me as the worst of yours I've read; hearing you call the Whale a just a degenerate gambler was a low point. I've also seen you have the same blind spot before (heavily weighting the opinions of people rather than the direct data): when you missed Biden's cognitive decline, while others just watched the video over the years. I get the impression that some political emotionalism may have clouded your judgment recently as well. Another point you didn't make in this article: volume *matters* in market efficiency... and Polymarket had enough volume to be efficient this time. You point out problems with PredictIt, without mentioning that it has low volume limits. Also, hedge funds and big American banks' foreign subsidiaries *did* bet on Polymarket this time, according to Haseeb Qureshi (he is an investor in Polymarket, among other things). That 2022 data needs to be be updated, now that we finally have some volume to make markets efficient. You mention "the personal opinion of one whale," but that *is* part of the "normal operations of the market." People with better judgment, like the Whale, that look at evidence, not theories, that got richer over time in the business of betting, should have more weight in order for markets to be efficient... that's essential to how markets work, part of what makes markets accurate, and I believe markets are superior to intellect. The smart money/whales take money from the stupid money over time, making markets more accurate (another example: the play at a new poker room improves in quality over time as winners take money from losers, as the latter mostly get better or stop playing). People have always had theories (intellect), since before the scientific method; I'm glad that now we have the scientific method and markets, where being wrong has a cost (unlike areas like governance and academia).
Do we have any details about what those polls were exactly, what their scope was, their methodology, etc? If he asked 1,000 people using a poor methodology and not weighting for the location of the responders, I’m sure he’d get a “Trump solid win!” conclusion but that wouldn’t count as useful new information.
If I said “Theo was correct because he consulted an astrologist” that would technically be “new information” but you’d still challenge me. Same thing here - “Theo paid for polls!” alone doesn’t update me much without knowing the details OR seeing his track record winning money using such polling.
It's about the type of polls: neighbor polls versus conventional ones.
An excerpt from the WSJ article, based on their conversations with Théo:
Polls failed to account for the “shy Trump voter effect,” Théo said. Either Trump backers were reluctant to tell pollsters that they supported the former president, or they didn’t want to participate in polls, Théo wrote. To solve this problem, Théo argued that pollsters should use what are known as neighbor polls that ask respondents which candidates they expect their neighbors to support.... Théo cited a handful of publicly released polls conducted in September using the neighbor method alongside the traditional method. These polls showed Harris’s support was several percentage points lower when respondents were asked who their neighbors would vote for, compared with the result that came from directly asking which candidate they supported.... To Théo, this was evidence that pollsters were—once again—underestimating Trump’s support.... At the time that Théo made those wagers, bettors on Polymarket were assessing the chances of a Trump popular-vote victory at less than 40%.... In an email, he told the Journal that he had commissioned his own surveys to measure the neighbor effect, using a major pollster whom he declined to name.... Théo declined to share those surveys, saying his agreement with the pollster required him to keep the results private.
You seem to believe too much in the opinion of one intelligent person who has a clever idea which tickles your fancy and who was right once, compared to the consensus of all the many other intelligent people with vast quantities of data gathered the boring old-fashioned way.
Which, yes, is known to be biased, but biased in a way that can be studied and understood and to some imperfect extent compensated for by intelligent people who devote themselves to the task. Are you under the impression that Polymarket Theo's clever idea produces infallibly error-free data? Do you even understand what the sources of error and bias are in that case? Does Theo, even?
No, I believe in markets. And Theo wouldn't have had enough money or conviction to move markets that much if he'd only been "right once." (See the example above about poker rooms).
Before this election, people acted like someone who disagreed with the consensus must be wrong (or even illegally manipulating the market)... that was an emotional reaction, not a rational one: what people wanted to believe. There was a lot of emotion in this stuff, which makes for inefficient markets = free money to other people. Good gamblers take advantage of other people's emotions (or are influenced less by emotions themselves). It might be the single most important thing in playing poker, for example.
Those "intelligent people who devote themselves to the task" *are* Theo, and others like him. The rest are amateurs that he takes money from. It reminds me of the old joke about the economics and finance professors walking down the sidewalk when they see a $20 bill on the ground; the finance professor goes to pick it up and the economics professor says "what are you doing? if there were a $20 dollar bill there it would already have been picked up!" Theo's the guy doing the work of picking it up, so that most people never see a $20 bill on the sidewalk. The market will always be closer to reality than the ivory tower, where people don't pay for being wrong. Theo doesn't have to be infallible; markets just have to be better than every other solution (and they are).
I mentioned a couple of the sources of bias above, at least one of which Theo used.
Well, many people made money betting on when Kamala's odds shifted from 34% on October 29, to 44% on November 3. Presumably some of them were rationalists, given I saw some rationalists on twitter talking about the mispricing.
That four day shift implies one of the following:
- Polymarket's odds can be mispriced/fluctuating for reasons not related to the market topic (e.g. a whale investing)
- Trump's chances got worse in those four days for some reason.
- Someone learned new information that changed Trump's odds during that time
While perhaps all these contributed, I find the first option most significant, because I don't see the specific facts that would make 2 or 3 more likely to be significant.
If you'd have asked me before the election, I would have said I thought Trump was favored 55-45.
As would I have, and as a result I lost money on Kamala making what I thought was a +EV investment just like the rest of the "smart money". However, I think this article is much more about the election day odds than the pricing movement in the preceding week.
"But on the specific literal level, Polymarket was mispriced last Monday."
Yes, not being facetious, but on the specific, literal level it should have been 100% Trump. I defend probabilistic reasoning under uncertainty, but the election has to be interpreted as a deterministic event, and one that could theoretically be accurately determined beforehand.
Some people, for probably a combination of good and bad reasons, mediated by good or bad epistemics, used more accurate proxies and measurements to assess this, while others (including myself) failed to do so, believing incorrectly that it was a "toss-up". Metaculus and poll aggregators just had the wrong combination of proxies and measurements.
I agree we have to be Bayesian about people being right for good or bad reasons, and we may never know if there was anyone genuinely well-calibrated. But it does seem that the "toss-up" people were wrong in some fundamental way that they wouldn't have been for, say, a sports prediction.
In some sense, almost everything that happens is deterministic. It's pretty clear that most events we care about are not influenced by actual quantum randomness. With this is mind, predictions that aren't basically certain reflect our ignorance, inability to model the world well, and lack of infinite computing power.
> But it does seem that the "toss-up" people were wrong in some fundamental way
Given the above, everyone is wrong some fundamental way. The only way we can evaluate prediction sources empirically is by measuring calibration (i.e. consistency with themselves) and comparing different prediction sources (e.g. using some proper scoring rule). Even then, it's hard to evaluate a prediction with few other similar predictions, for example one for a presidential race.
So, back to the point, I think Scott is right that while empirics are important in evaluating predictions, we should also value theory heavily in valuing prediction sources. If someone used astrology to predict 13 elections in a row, I wouldn't particularly value that as prediction source, because it has no theoretical backing that would lead me to believe its likely to be right in the future.
Conversely, Nate Silver's method of averaging the results of a bunch of sources asking people who they are voting for, adding uncertainty based on past polling errors, adjusting for bias, etc. etc. seems like a pretty reasonable way to predict an election to me. When people say it makes bad predictions as a method, my question is "in comparison to what". It seems rather pointless to say its bad without recommending an alternative method, because *calibrated predictions can only be compared in relation to each other*. Though I would probably trust prediction aggregators over Nate alone.
To be more specific about polymarket, the level of volatility doesn't make sense to me based on what happened this election. And, for example, on the eve of the election 538 had "Trump winning popular vote" at 29% and polymarket at only 26% (though polymarket did have higher odds two weeks beforehand). So perhaps prediction markets overreact to news or are overly certain?
I'm not sure if this directly responds to your comment, but these are some thoughts it spurred.
Good comment. To add a thought to your thought about how to evaluate predictors not merely based on their results, but also whether their methodology is using astrology or more like Nate's, I'll mention that top forecasters often say they are a better judge of their peer's forecasting ability than their peer's forecasting track record. I agree with that, and think it's mainly because you can only get so much information from a small sample of a few hundred or few thousand forecasts (track record). Examining how those forecasters make forecasts provides a lot more information, and one skilled forecaster can learn a lot from that about a fellow forecaster's strength as a forecaster.
"But it does seem that the "toss-up" people were wrong in some fundamental way that they wouldn't have been for, say, a sports prediction."
Yes, and the difference is that there is more "irreducible uncertainty" (Philip Tetlock's term) in a sporting game than in a Presidential election.
"Yes, not being facetious, but on the specific, literal level it should have been 100% Trump."
The world is not necessarily deterministic--there is quantum randomness--but as for whether this would ever cause Harris to win >0% of elections if we re-ran a simulation of the universe a million times starting one week before the election is unclear to me.
I support using a notion of "true probability" that is meaningful in forecasting by never quite being 100% or 0%. This is achieved by adding random noise to the hypothetical simulated re-runs of the universe. We re-run the soccer game or re-run the election, but tweak the universe in some random way at some point in time before the game or before the election. So for example, a week before the election we make someone's car tire get punctured and go flat when it hits a rock on their drive to work, rather than just holding together as happened in reality. Or we cause one of the soccer players to sleep in for 5 minutes later or set their alarm for 5 minutes later than they otherwise would have a week before the game. These tweaks would undoubtedly lead to big changes in the long-term, and meaningfully affect the frequency of wins in the million universe simulation of the election and soccer game.
In this sense we might say that the "true probability" of a team winning a soccer game is low, say 60%, even if that team in fact went on to win. We know this because we believe the two teams were about equal in strength and random-tweaking of the simulations would actually change who wins a significant amount of the time.
With elections though, if you add the random noise only shortly before the election, say a week-or-so, then that's probably no enough to make Harris win a significant percentage of the time, given the margins Trump actually won by. As such, I think that the "true probability" (in the sense that I mean it in which the random noise is added a week before the election) that Trump won is likely >90%, but not quite 100%. If we ran a billion simulations, Harris would likely win some of them. (Or if you disagree with that, imagine that the random noise happened a few months before the election. At some point, you'd start giving Harris nontrivial chance of winning some of the elections in some of the simulations of the universe.)
Having said all this, the reality is that the best subjective forecast we can make on a given question is almost always more uncertainty than the "true probability" given what Philip Tetlock calls "irreducible uncertainty" in the world.
So perhaps an efficient betting market for the soccer game with a true probability of 60% would have a market price of 51%. And similarly there is probably no rational way that the betting market for Trump could have been priced at >90% (that would require irrational actors, betting on unjustified true beliefs; irreducible uncertainty in the world would prevent them from being able to know that Trump was that likely to win).
Could it be the case that the most informed best forecast one could have made on Trump winning given the irreducible uncertainty in the world was less than 60%? Could it have been less than 50%? I think both are possible, but very unlikely.
Help an average Joe understand why prediction markets are a good thing.
No participant in a market is passive. Everyone is an actor. If a person bets money on a certain outcome, they will naturally do what they can to influence things in a direction that's favorable to them. They will do this unconsciously if not consciously.
There are significant consequences to mistakenly regarding prediction markets like passive polls. It opens up the public mind to undue influence when a whale bets a certain way. Anyone who understands market dynamics knows about the momentum of sentiment when things start to go a certain way. Yes, there are opportunities for arbitrage, but history tells of how speculation can bubble to a harmful degree.
Market dynamics are one thing when based on some underlying value. But with prediction markets, we're betting on the future. We're betting on mass phenomena. We're often betting on something we are both a part of, and a part of perceiving.
Aren't we just letting Moloch having even more control or influence over us? More power and sway for money to influence and shape mass social behavior?
Scott wrote a FAQ on prediction markets that you can easily find on this very website.
My understanding is that the prediction markets enthusiasts of this website believe that markets are indeed the solution to most assessment problems.
It’s not unsound: an ordinary stock price reflect the expectation of (suitably discounted) future earnings. So betting on the stock market is already a prediction problem.
The basic idea is that people want more money. Therefore, by putting money at stake, you incentivize people to spot mistakes and correct them.
Because of this incentive, this should imply that a sufficiently large prediction market gives you an actual prediction, even without anyone giving you any odds.
It is supposed to be accurate because anyone believing otherwise can (expect to) make money simply by betting.
More generally, believing markets to be the solution to just about every problem is not particularly rare here. I’ve come to suspect that it may be an American thing. After all, they literally spend their whole lives betting that the market will go up (so that investing their economies into pension funds to fund their retirements makes sense). Apparently, on the long term, it hasn’t failed them yet.
(What would be convincing alternatives?)
This line of argument concedes that markets do fail sometimes, but only in the predictable ways studied by classical economists for a while, with well-known remedies.
I am personally skeptical and suspect that there’s a lot of motivated reasoning going on in there. That freer-markets are a lot easier to support when you’re affluent, things are going well for you, and you win off of it.
(Which, as an old Scott reminds us, is *not* the majority situation. Sure, we’re better off than subsistance peasants, but that’s not too relevant.)
Prediction markets are good because they aggregate the most accurate information, thus helping us to predict things. Attempting to influence voters is fair game in politics, but I'm unaware of prediction market speculators like "Theo" doing that because that wouldn't be the most efficient way for him to use his money to make more money. Remember that the Harris campaign had something like 3 times the funds of the Trump campaign and still lost.
> But with prediction markets, we're betting on the future.
That describes a LOT of speculative markets, such as the stock market. Someone could short a stock and try to sabotage the corporation, but that basically never happens.
Does Polymarket show the bets people are making? I might assess the prediction differently if I knew that one big bet moved it that much. This seems like a good partial solution at least.
Another point in your favor is the very large post-election move in financial markets, which suggests that those massive markets were very much unsure of a Trump victory.
I'd only expect the size of the move to be ~20% larger if everyone is updating from 50% confident than if everyone is updating from 60% confident, so I don't think the vague magnitude of the move proves much.
I don't really understand this paragraph: "Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%. Now should we say that the “true” probability of Trump’s win was 90%? Imagine he spent half his money on a yacht just before the election and only had enough money left to move the markets up to 56%. Now should we say that the “true” probability of Trump’s win was 56%? Why should the true probability of Trump’s win depend on whether a French guy bought a yacht or not?"
Surely the true probability of Trump winning should have been 100%, since he did in fact win, and the fact that the markets were trading at less than that means there's a lot more accuracy out there to be gained?
See my comment that clarifies this. I support a notion of "true probability" that is meaningful by never quite being 100% or 0%, but also reflects the fact that the best subjective forecast we can make on a given question is almost always more uncertainty than the "true probability" given what Philip Tetlock calls "irreducible uncertainty" in the world.
Three other comments on the reasonableness of a larger move given a hypothetical richer Theo:
a) Echoing other commenters, if the available information, even with additional polling, is insufficient to push the expected probability up past e.g. 70%, then Theo++ will stop buying when the price reaches that point
b) weak point: A richer Theo++, if self-made, is presumably at least marginally smarter than the real world Theo (or luckier? Teela Brown gene :-) ? ) and their opinion should count more heavily
c) stronger point: real Theo commissioned an additional poll to gather additional real information. Theo++ has a larger budget, and could commission more polls, or larger polls. At the very least, the 1/sqrt(N) random error should go down.
I remember Nate writing a model deep dive that listed this outcome (Trump sweeps all 7 swing states) as the modal outcome, out of the 128 possible outcomes for the 7 swing states. Don’t remember the exact probability he assigned to this outcome but I think it was around 7-10% of all results. The next most common individual outcome was Harris sweeping all 7 swing states, I think that was around 5%
Great post! This would be a very good reading to discuss in a PhD finance class, because it is so good for thinking about market efficiency. I took a 6-student class with Fischer Black at MIT in 1979 based on 30 to 40 questions like "Why do corporations pay dividends?" (and longer ones), and I can see him putting a betting market question and your post on the syllabus. He would stand at a lectern in his 3-piece suit (no other prof wore them, or even suits usually) listening and writing little notes, and then maybe give his opinion.
One part of the discussion would be why you can't do arbitrage between betting markets-- place hedging bets that reduce the risk to zero. Probably the answer is transaction costs plus that if you place bets big enough to make it worthwhile to set up the apparatus, you move the markets too much because they're thin.
I think transactions costs are probably a big part of the story. The other part is probably slight differences in resolution criteria that make it difficult to price those differences accurately.
I buy some of this. But I think it's interesting that there wasn't a multimillionaire in Canada, the UK, Japan, etc who decided to take the other side of this bet, given the pricing. Although I'm sure some did the arbitrage trade (I wanted to, but didn't have time to set up my wallets).
But I suppose also from a statistical analysis perspective, it's not crazy to throw out an outlier if that's having a large effect on the variable you're measuring. Since it's clear the biggest whale really did have an outsized effect on price, I think it's reasonable to question if Polymarket was really efficient. Had it been open to US persons, I think we could trust it more.
It seems like part of the problem is that some people bet their identity rather than betting as insurance + arbitrage. Are there any plans to educate people on how to think better about betting? Otherwise lowering the access barrier may not result in better results, if people bet their identity rather than their beliefs.
"Yes, but there were dozens of people who could give equally-plausible arguments for their positions before the election. These were divided half-and-half into intelligent-sounding pro-Kamala arguments and intelligent-sounding pro-Trump arguments"
Nate Silver disagrees:
"To the extent there was anything that resembled actual insight, it was that it was much easier to come up with a list of reasons that Trump would win — inflation, immigration, Joe Biden wanting to be president until he was 86, the illiberal backlash around the world — than to make the same list for Kamala Harris. I tried and could maybe come up with 6 or 7 good points for her, but not 24. So that may have weighed on my “mental model” of the race."
"Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%. Now should we say that the “true” probability of Trump’s win was 90%?"
This assumes that Theo would have continued to buy at that high a price. It's possible that he thought probability of Trump win was, say, 70%, in which case he would have bid it up to that but no higher.
> But long after Biden won the last election, PredictIt said there was a 9% chance Trump would be the next President.
This is an understatement. If I recall correctly, the Trump vs Biden election market was at 17% Trump in late Nov/early Dec. 2020, and the add all the "Trump win" results in the "What will be the electoral college margin market" it was at 23%. I know because I played some, uh, Minecraft during that time period.
It seems like all these prediction models were fundamentally flawed. Does anyone really think if we simulated Election Day 100 times, Kamala would not have won 40 times?
If we were sampling from the ensemble of possible universes which produced polling data, market prices, etc, that exactly matched what we saw right before the election, then I believe Kamala would most likely have won somewhere between 40 and 50 times.
Several people here seem to be arguing that we live in what we now know is a single deterministic universe(*) where it was ~100% certain that Donald Trump would win the election, and that forecasts should be judged correct by how confident they were that Trump would win. This may be true, but it is irrelevant for predictive purposes. We may live in a deterministic universe where the outcome of the election is preordained, but we don't know *which* such universe we live in. There are many possible deterministic universes in which the polling results were exactly as we saw and Kamala Harris won the election, and many such universes in which Donald Trump won.
"Prediction", is the art of figuring out the odds that we live in a particular subset of universes, and its useful accuracy is in how closely the predicted odds match the fraction of possible deterministic universes in which the stated outcome occurred. And, given the available data and analysis at the time, the best answer was something like "slightly better than 50% that we live in one of the universes where Donald Trump is destined to win, and about 20% one where he sweeps all the swing states".
* Or nigh-deterministic if we assume e.g. that people's decisions are fixed on election day but a large number of freak buttefly-effect accidents preventing some voters from going to the polls might have had a miniscule chance of changing the outcome.
>If we were sampling from the ensemble of possible universes which produced polling data, market prices, etc, that exactly matched what we saw right before the election, then I believe Kamala would most likely have won somewhere between 40 and 50 times. [out of 100, presumably]
Reminds me of a statistical mechanics class from half a century ago, making a distinction between a macrostate, where we know the pressure, temperature, volume, and maybe a handful of other measurements of a sample of a gas, and microstates, where we know the individual quantum states (including translational states in a box...) of each individual molecule in the gas. Good old microstate ensembles! :-)
The presupposition that these polls were modeling anything approximately our universe should have been dispelled midway through election night. The Iowa Ann Selzer poll was 16 points off. There was no 40-50% chance of Kamala winning whereas if we simulated people going to vote that day 100 times 40-50 we would have had the opposite results.
The polling data was systematically flawed and only made to appear a close election through nonrepresentative sampling and arbitrary weighting. I am not saying we live in a deterministic universe. I'm saying that there were a million red flags that must be ignored to claim these represent the voting population. It is a prime example of a garbage-in-garbage-out models.
One of the best function of prediction markets is that - like any good price signal - it cuts through the motivated reasoning and lies and gives ordinary people access to reality. On election night, even as late as 10pm the TV networks hadn't called states like GA and NC that everyone paying real attention knew were over, because of the percentages on Polymarket. Somebody watching CNN would've had to read tone and body language and have an amount of political knowledge/insight that we may take for granted here but that probably less than 5% of the American people possess. But somebody just checking Polymarket regularly all night knew this thing was all over but the shouting by 10pm.
And it performs that service very well even if it's off by 10%, which is meaningful to bettors but to normies who just want to know "is this thing the news media is yapping to me about something I should take seriously?" doesn't need to be that precise. Just as a consumer in a small town doesn't need to know the price of pork as accurately as a grocer does, but still benefits from having a device in his phone that shows him what the walmart in the next town over is charging. I think this is a great tool for average people to cut through the manipulation aimed at them.
Watching Steve Kornacki on MSNBC you were hearing, dozens of times, about counties in GA and NC, "Harris is not performing at Biden 2020 levels and the remaining vote out is likely tilted R." It was clear by, I think, about 9ET where it was going. There are detailed guidelines for "calling a state" but there was more than body language available. That's the kind of info the Polymarket people were presumably working from.
I largely agree, but I want to defend the mainstream media a little bit. Even putting aside any political bias/preferences, the mainstream media has two strong reasons to call states very late:
1. Ratings. Ratings probably spike briefly right after a critical state is called, but ratings will likely decline if/when people think the election is basically over (and for people on the east coast, they can finally go to bed).
2. Nothing is more embarrassing for the media on an election night than calling a critical state and then having to reverse it not long after.
So basically, even if we had a magically unbiased mainstream media who truly didn't care which party/candidate won, they'd have two good reasons to drag out the calling of states. Thankfully, Polymarket does not have the same concerns. And so yes, for those who don't want to stay up all-night or try to read between the lines of what news anchors are saying, they can just look at Polymarket and do some simple math.
Why was it so important for people to know that GA and NC have been called by 9pm vs 10pm? Other than pure political entertainment value, why does it matter?
This also ignores the possibility of Polymarket calling a state too early in a more complicated situation, such as what happened in 2020 when Trump was leading initially but then Biden overtook him as the night went on in certain states.
<i>we’re asking whether to trust the normal operations of the prediction market vs. the personal opinion of one whale</i>
That's an interesting definition of "normal operations of a market". Having large traders is part of the normal operation of any money market. It is also one of the reasons why these markets are relatively efficient. Large traders are (sometimes) better informed because for them there are stronger economic incentives to be better informed. This has certainly played a role here (Theo commissioned his own private polls, if that's not adding information to the market I do not know what is.) So that's what a normal operation of a market looks like.
At to the claimed better accuracy of fake money markets vs. real money markets, I am not sure it is measured correctly. Real money markets binary event prices do not equal implied probabilities for low and high implied probability events, for some pretty simple reasons (interest rates, credit risks, frictions.) So when one takes these prices and discovers that they do not match realised probabilities, a part of mismatch is due to the incorrect price to probability mapping.
>Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%
This is very, very wrong. Theo didn't move Polymarket to 90% because he didn't think Trump had a 90% chance of winning; he felt ~60% was a fair price and set limit orders to floor the market, slowly accumulating volume. Prediction markets are not binary, you are not supposed to just decide what you think will happen and go all in. Instead, you invest if you think the probability of something happening is higher than the probability expressed in the market. Theo did set the price, yes, but to the probability he thought was correct at whatever volume would fill it, not to an arbitrary probability based on a 100% certainty of a Trump win * his bankroll.
This massively under-rates Theo. It wasn't just his opinion or "a completely replacement-level example of the intelligent-sounding pro-Trump arguments"
From an interview he gave later:
> As Théo celebrated the returns on Election Night, he disclosed another piece of the analysis behind his successful wager. In an email, he told the Journal that he had commissioned his own surveys to measure the neighbor effect, using a major pollster whom he declined to name. The results, he wrote, “were mind blowing to the favor of Trump!”
He didn't just think the polls were wrong, he literally sought disconfirming information by funding an independent poll using what he believed was a better methodology. Then, he bet on the private information he received from that poll.
I speculate that the ~$40MM was probably not just his money, he probably got some external funding as well. Almost certainly he got the results and his interpretation cross-checked by multiple other experts. Even someone with a few hundred million of net worth doesn't just throw down $40MM on a whim.
This is literally exactly how prediction markets are supposed to work. You have a very informed bettor whose private information is incorporated into the probabilities.
Right. If George Soros in his trading prime had made this bet instead of some dude named "Theo", we would believe that Soros had moved the market to the more correct price, because Soros had a phenomenal track record for trading as a whale who moves markets. Yet Theo supposedly was a great financial trader himself. His lack of prior fame makes all the difference somehow. Soros would have been right, but "Theo" got lucky.
It could be that Theo is well-known (at least in France) under his real name, but trading on a crypto platform with pseudonymous accounts obscures that.
(1) Calibration city says Metaculus is less well-calibrated than other markers and tends to be biased towards 50% https://calibration.city/calibration
(2) On election night, Metaculus was useless since it does not move quickly in response to new information. Manifold, Kalshi, and Polymarket all converged on >90% Trump while TV pundits on my other screen were all talking about "multiple paths to victory" etc etc.
So from a purely popcorn-munching, I-know-the-result-sooner-than-my-friends perspective, you've gotta admit the real money markets are the better option!
Scott, I don’t know if you’ve been reading Rajiv Sethi’s Substack, “Imperfect Information,” but a lot of what you’ve been saying lately regarding potential prediction market liabilities has been explored there. The main difference is Sethi’s often talking in reference to the relative utility of models vs. prediction markets. From his most recent post:
“Markets have some significant advantages over models—they can take on board novel sources of information, can deal with historically unprecedented situations, and can respond very rapidly to evolving events. But they also have potential shortcomings as forecasting mechanisms. They can come to be dominated by a few large traders whose views may be idiosyncratic, they are subject to overreaction and correction, and when they reference events whose objective likelihood of occurrence depends on subjective assessments of this probability, they create incentives for manipulation.”
I think Sethi’s claim is not that they definitely were, but that “they can come to be.” I take it you don’t agree with him that prediction markets are at risk of this.
I think it depends on the market. Polymarket is open to people everywhere but the US, and there was a lot of attention on this election, so even "Theo" only provided a small portion of the total. PredictIt was less accurate because it's more restricted, which is precisely what Sethi said could potentially make it more accurate by not having such influence from large traders:
It should be noted that I also mused along Sethi's lines earlier (but as sort of a Devil's Advocate for an argument), and afterward noted that the actual outcome falsified that theory in that case: https://x.com/TeaGeeGeePea/status/1854870049359630447
If these markets get big enough, the primary use for many users will be hedging election outcome risk. In fact, I think it would be the main benefit of these markets maturing. For example, investors in solar companies could have bet on Trump to win to protect against stock losses.
I posted one comment about Theo being under-rated, but I think the description of "rich techie with $1M net worth should reasonably bet around $2k" is also incorrect.
If you have extreme confidence in your bet - or even better, if you have some edge - the right move is not to bet $2k or even $166k. The right move is **get more money**.
If you had an amazing but capital-intensive startup idea, you wouldn't say "hey let's see how far I can take this on $2k", you would go talk to a bunch of VCs and raise a few million dollars. Similarly, if you have actual edge in the betting markets and there's sufficient market liquidity, the play is to go and raise external money from investors or set up some kind of capital pooling with other, wealthier people. There are entire industries dedicated to this kind of thing: hedge funds, private equity, proprietary trading firms, etc.
If you have actual, real edge and not "Kamala wins is just like, your opinion, man" then you should be able to raise reasonable amounts of money with a few weeks to months of lead time (depending on your personal network).
If you don't have real edge, then it's totally appropriate to bet $2k money and book it as entertainment rather than investing.
Given how well-versed this crowd is on this topic, I hesitate to bring up this old paper, but I think it provides some theoretical arguments about the limitations of prediction markets, even without whales, that are almost never engaged directly in such discussions:
"The concluding Section 4 observes that the information about beliefs contained in equations (1) and (2) is fragile. If budgets are not statistically independent of beliefs, prices in prediction markets are determined by the joint distribution of expectations and budgets rather than by the distribution of expectations alone. If traders are risk averse, price reflects their risk preferences as well as their expectations and budgets.
For these and other reasons, I suggest that direct measurement of expectations may be preferable to inference on expectations from prices in prediction markets."
I strongly disagree with the claim that Polymarket's price that Trump would win was mispriced in the direction of being too high on e.g. November 4th.
I explained why in a series of comments I wrote on the Facebook post of a friend who shared Scott's post. Copying them here (sorry for not polishing them up, but I think they're quite comprehensible despite being written sloppily):
----
Scott says several things I disagree with in this post and it would take a full post to adequately explain why I disagree with all of it.
Listing out some issues:
(1) He claims that the markets were "mispriced" without explaining what he means by this.
(2) Later he references the concept of a "'true' probability", which he does not define.
(1b) By "mispriced" does he mean that the market price does not match the "'true' probability"? Or does he mean that the market price does not reflect the most-informed credence that humans can come up with (a different probability)?
(1c) His "mispriced" claim seems to be that the markets should have been priced closer to Metaculus's ~50% than their actual ~60% Trump price. (1b continued) Whether he thinks this because he thinks Metaculus was accurately forecasting the "'true' probability" or because he thinks Metaculus' forecast was the best forecast one could make given the information available to the world before the election is unclear.
(1d) In either case, I claim that Scott is wrong to claim that the markets ought to have been priced below 60%.
And now (because it's easier), to just say some related beliefs I hold without responding directly to Scott's post:
I believe that looking back on how the election unfolded provides us with evidence that the "true probability" of Trump winning was >90%.
Note that I've been trying to come up with a single coherence concept of "true probability" for years that works for what I mean in forecasting conversations like this one while at the same time not contradicting our knowledge of physics and how it is roughly deterministic (or random with quantum mechanics).
I don't have a single definition, so what I will say is that I mean something like "if we were to reset the Universe to the state it was in one week before the election and had some random noise (like a rock that almost punctures someone's tire in real history actually causing them to get a flat, or vice versa) to make sure that they did not all play out exactly the same (due to determinism, if quantum randomness doesn't actually make a difference), then, in what percentage of simulations would Trump win the election? "True probability" is probably a misnomer given this meaning, but this captures our meaning, similar to how we want "true probability" to mean something such that "the true probability that the D6 will land on 6 when I toss it right now is 1/6th", even though if we had enough information about how the die was oriented in your hand and how your brain worked we could predict the exact way in which you would toss it and how it would tumble through the air and hit the table and land with high probability (since quantum randomness probably doesn't affect this, unless it does with cognition somehow, but you get might point), meaning that 1/6th probably isn't *actually* the "true probability" and the true probability is probably exactly or almost exactly 1 or 0.
Note that had PA been decisive and had Trump won by 1 vote or 10,000 votes in PA, I think that the random noise injected into the simulations a week prior to the election would be enough to make it so that Harris would have won in >10% of elections in the simulated worlds that started a week before the election. But the further the election is from being close, the less of a difference that the random noise makes, and with what actually happened I think it's reasonable to predict that Trump would win in >90% of the simulated-world-elections (which I'm calling "true probability" even though my definition is arbitrary and "true probably" shouldn't have an arbitrary definition.)
All this said, might there a good reason why forecasts or market prices were only ~50-60%, even if I'm right that the "true probability" was >90% that Trump won?
Yes.
There is (what Philip Tetlock calls) "irreducible uncertainty" in the world. For forecasters trying to predict how a D6 will land, this may result in them making forecasts close to 1/6th that it lands on 6, even though they have video footage of the die flying through the air. This is because their uncertainty in the measurements introduces too much noise, and if they're slightly wrong about how the die will strike the table then that will completely change how it will land. So even though someone with a perfect knowledge of physics and the state of the world when they make their forecast midflight could more accurately predict how the die will land, the forecasters knowledge is limited and they cannot possibly know better. There is inherent uncertainty that they can't get rid of given the information they have access to.
Was 50% or 60% the best forecast one could make given the irreducible uncertainty, and assuming I'm right that the "true probability" was >90%?
Not necessarily.
On November 4th I wrote:
"My 2/10 low information inside view judgment is that Trump is about 65% likely to win PA and the election. My all-things-considered view is basically 50%.
However, notably, after about 10 hours of thinking about who will win in the last week, I don't know if I actually trust Nate and prediction markets to be doing a good job. I suspect that there may be well-informed people in the world who *know* that the markets are wrong and have justified "true" beliefs that one candidate is >65% likely to win. Such people presumably have a lot of money on the line, but not enough to more [sic] [move] the market prices far from 50%." (https://www.facebook.com/.../pfbid02kHsKK9oWUKGo3bvhs9wV9...)
In other words, I thought before the election and I think now that people more informed than me (I said I was only 2/10 informed) might be justified in making their all-things-considered believe >65% in favor of a particular candidate (unlike me, who meta-updated downward from my inside view toward believing the markets and better informed forecasters; not my "50%" in the quote was not meant to say that I believed the forecasters/Nate more than the markets, just that I didn't trust my inside view judgment of 65% Trump due to the markets and forecast aggregations assigning lower probabilities and me knowing I was relatively uninformed.)
Why do I think that? In short, because forecasters aren't close to being at their peak strength, and political prediction markets are nowhere near as efficient as e.g. stock prices. As Scott said, there probably weren't any big professional trading firms betting on Polymarket, let alone many like in stock markets. It just wouldn't be surprising if a very-informed person or group was able to correctly identifying that the prevailing wisdom of November 4th about how likely Trump was going to win was wrong.
So, expanding on "not necessarily" above: Could Trump-wins have been priced too highly on Polymarket despite Trump having a "true probability" of >90% of winning?
Yes, but (1) Scott and I and all of us reading this don't know if that's the case.
And (2) us learning now that Trump won and had a >90% "true probability" of winning is strong evidence that the hypothetical very informed forecasters I mentioned (whom may have actually existed, as I said on Nov 4th I thought was ~50% likely) were people who were at >65% Trump rather than people who were at <35% Trump. And thus it's more likely that ~60% market prices were too low rather than too high, given the best forecasts that could be made given irreducible uncertainty.
Scott also says a lot of stuff re the French whale on Polymarket, some of which I'd critique, but I don't think it's as important as the fundamental mistake Scott is making (claiming that Polymarket's price on Trump winning was too high).
If you want to criticize your opponents' reasoning for being incorrect, at least wait for them to actually be factually mistaken
It feels very strange for him to be complaining about a probability estimate that was directionally correct compared to the one scott proposes
If lottery tickets are actually negative sum, then you will get lots of opportunities to criticize people who buy them and then lose money
It feels like Scott is currently in the position of somebody who has been pro-lottery his entire life, and then suddenly posted his very first criticism of the lottery the day after winning the jackpot
Well actually I'd push back on this. I think there is a true thing that Scott could have said in his post that is not far from what he actually said.
If we consider a different question where there is more irreducible uncertainty in the than there is for a Presidential Election, such as a soccer game or the role of a die (see this comment-reply of mine above to someone else: https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76065426), such that learning the result that actually happened only causes us to update our estimate of the "true probability" that the event was going to happen to, say, 60%, rather than >90%, then I could see a post very similar to Scott's post being reasonable.
Suppose that Polymarket put the result of the soccer game at 70% that Team A wins, and Metaculus puts it at 55%. Then Team A wins. Scott could then write a post saying "Hey people, just because Team A won doesn't mean that Polymarket's 70% was better than Metaculus' 55% forecast. I still think Polymarket's 70% was price too high and that Metaculus' forecast was better."
This is related to what Tetlock calls the "wrong side of maybe" fallacy. (Hey, I've been thinking about this for 8+ years: https://imgur.com/gallery/wrong-side-of-maybe-fallacy-uG3rgGU ). Just because a forecast was on the wrong-side-of-maybe doesn't mean it was wrong or bad. It's evidence that it was bad, but not decisive evidence, and there are cases where it is not actually bad.
More generally it's also true that just because a forecast is less-close-to-the-result-that-actually-happened doesn't decisively mean that it's worse than a forecast that was closer to what actually happened. Again, it's generally *evidence* that it's worse, but that's not definitively the case, and there are exceptions. E.g. People widely regard the Dilbert guy Scott Adams as being overconfident when he stated 99% Trump in 2016.
If Polymarket had been at 99% Trump for a week before the election, yet the result had turned out exactly as it had (same number of votes everywhere, Nate and Metaculus thinking ~50%, etc), then saying that Polymarket had Trump winning priced too highly would be reasonable.
I think you can only conclude that with a larger sample size of predictions. If outcomes that Polymarket assigns a 60% probability to actually happen 60% of the time, then Polymarket is well-calibrated.
"If outcomes that Polymarket assigns a 60% probability to actually happen 60% of the time, then Polymarket is well-calibrated."
Yes, that is the definition of "well-calibrated".
However, "well-calibrated" does not mean that the Trump actually wins 60% of the time, nor does it mean that 60% is the best forecast a highly-informed, rational forecaster can make on the question.
"I think you can only conclude that with a larger sample size of predictions."
I'm not sure that I know what you are referring to by "that". Is it my last statement? That: "If Polymarket had been at 99% Trump for a week before the election [...] then saying that Polymarket had Trump winning priced too highly would be reasonable."
If so, let me clarify that it'd be reasonable because we'd know that market actors, like Scott Adams, didn't have the actual knowledge to know that Trump would win 99+% of the time, even if it was actually the case that he would win that much. Market actors have epistemic uncertainty that would prevent them from knowing that. So we'd know that the market price was irrational, even if it happened to correspond to the true frequency of Trump winning.
"If the market is well-calibrated, then if it assigns 60%, I conclude that the available evidence is consistent with a 60% probability."
It sounds like your claim is that the available evidence is consistent with the aleatoric probability that Trump would win was 60%. I wrote a concise argument against that just now in this comment[1], which I'll copy here:
"The January 2024 Less Wrong post "Will quantum randomness affect the 2028 election?" makes clear that uncertainty about election outcomes is almost entirely epistemic uncertainty, not aleatoric uncertainty. Thus, whichever outcome happens is very strong evidence of what the true aleatoric probability of that outcome happening was. Specifically, it is very strong evidence that the outcome that happened was very likely to happen. Therefore, a day before the 2024 election Trump was very likely to win (>99%). Therefore, Polymarket's ~60% price was priced too low, not too high. https://www.lesswrong.com/posts/HT8SZAykDWwwqPrZn/will-quantum-randomness-affect-the-2028-election "
Here is an explanation of what "aleatoric uncertainty" is, from this source[2]:
"So aleatoric uncertainty is, if I tell you I’m going to flip a coin, you’re roughly 50-50 on whether the coin will come up heads. But you have almost no epistemic uncertainty, because you’re quite confident that your distribution is correct. And there isn’t a way to reduce it further with a practical amount of knowledge. If you knew the whole state of the world, you could get it to zero, but you don’t. So in that situation, you have all aleatoric uncertainty and no epistemic uncertainty. So in that case, you don’t expect to learn more by getting more data about the problem. And in the epistemic case, I flip the coin, it lands, I close my hand and now I’m going to reveal the data point, what the coin is. And so I’m still 50-50, but I have no aleatoric uncertainty and all epistemic uncertainty. I just don’t know the answer, but there is definitely an answer to be learned."
(disclamer: didn't read Scott's article, got here via your comment on FB)
(1b) does he mean that the market price does not reflect the most-informed credence that humans can come up with?
Surely yes? (basis: it's always seemed to me that Scott and I conceive of probability the same way, i.e. if I have already flipped a coin but no one has looked at it, we should say the probability of heads is 50.5% or so, notwithstanding that the "real" probability is 100% or 0%). Or perhaps something further in this direction, like: almost nobody had better information than Nate Silver, and if a few people did, they probably didn't have the resources to move the market that much. (by analogy: one man has seen the coin, but not shared the results with anyone, and isn't rich enough to move the market on it)
The best argument in favor of "markets priced correctly" that comes to my mind is "wisdom of crowds": if enough low-information people in swing states were betting, that might aggregate enough information to set the price correctly even if none of the individual betters had enough information to do so.
"Surely yes? (basis: it's always seemed to me that Scott and I conceive of probability the same way, i.e. if I have already flipped a coin but no one has looked at it, we should say the probability of heads is 50.5% or so, notwithstanding that the "real" probability is 100% or 0%)"
I agree with this. (Note that "real" probability would more appropriately be called "aleatoric" probability)
"Or perhaps something further in this direction, like: almost nobody had better information than Nate Silver, and if a few people did, they probably didn't have the resources to move the market that much."
I think you're assuming unjustifiably that Nate's forecast was near the best quality possible. Even if he was very informed, he may not have applied the information he had most appropriately to form a forecast. (Two forecasters given access to the same information about a forecasting question often have different credences, due to different background knowledge from their life, different methodologies, etc.)
"and if a few people did, they probably didn't have the resources to move the market that much. (by analogy: one man has seen the coin, but not shared the results with anyone, and isn't rich enough to move the market on it)"
The French trader moving the market toward Trump may have been less informed and irrational relative to Nate, and may have effectively been betting on Trump randomly and just got lucky. We can stipulate that that was the case and I claim that still wouldn't show that the "correct market price" should have been less than 60% rather than greater.
I'll make a stronger claim: Even if *all* the Metaculus forecasters were better informed forecasters who were stronger / higher skilled at forecasting in general than *all* the Polymarket traders, that still wouldn't show that the "correct market price" ought to have been closer to the Metaculus aggregation than what the Polymarket market price actually was.
Why? Because the Metaculus traders, despite being higher skilled and more informed forecasters than the Polymarket traders were likely still not making the best estimates possible, despite inevitable epistemic uncertainty due to irreducible uncertainty in the world that would prevent them from ever being able to be as confident in the outcome as they would be if they had zero epistemic uncertainty and their credences matched the aleatoric uncertainty perfectly.
So the fact that the Polymarket price may have been largely what it was due to low-information traders is actually pretty irrelevant to the question of whether "correct market price" ought to have been what the Metaculus aggregation was or something else.
We don't know precisely how much the best possible forecast could have reduced epistemic uncertainty, but I don't think we have any reason to believe that reality tricks people into doing worse than chance on this question when they try to become very informed. On the contrary, this is the sort of normal forecasting question where we'd expect it to be possible to be able to get some evidence in the right direction. This implies that the "correct market price" should be >50%.
But how much more than 50%? I don't know precisely, but I'd be surprised if one couldn't get beyond >60%. As I wrote on November 4th, I suspected that there actually were highly informed people who were justifiably above >65% in favor of a particular candidate, though I didn't know which candidate. In other words, before the election I believed that it was possible to reduce epistemic uncertainty by enough to get over 15% of the way away from 50%. And that was before I had the benefit of hindsight of knowing that aleatoric uncertainty was >99%, and thought the election could have been close such that aleatoric uncertainty would have been <90% in one direction or another.
The fact that Nate and Metaculus was only at 50% does not at all show that it wasn't possible to reduce epistemic uncertainty enough to reach a credence of >65% (including both epistemic uncertainty and aleatoric uncertainty).
So I think the "correct market price", i.e. the price that the market should have been at if it were reflective of the credences of forecasters who had reduced their epistemic uncertainty as much as possible given inevitable uncertainty in the world, was probably about 65-90%, not 60%, and definitely not 50%. This 65%-90% is still much more uncertainty than a zero epistemic uncertainty that matches the true aleatoric uncertainty (what you called the "real probability"), but it's not so uncertain as to say that humans can't reduce any epistemic uncertainty they have when it comes to election results.
Here's a question I haven't seen much discussion of:
Should the 2024 election results make one update, to however small a degree, on whether the 2020 election was honest and accurate? Not merely the fact that Trump won, but more micro-data, like numbers of registered voters participating in particular districts, etc. Is there now new reason to hold that some of the 2020 results look less credible?
Looking at county by county data https://www.nytimes.com/interactive/2024/11/05/us/elections/results-president.html what strikes me is the uniformity of the shift. If there was an election-turning level fraud in 2020 then we'd probably expect to see the swing being particularly concentrated in particular states and counties, but as far as I can figure out we don't.
This doesn't disprove all 2020 fraud but it places limits on it.
I like Yudkowsky's idea of not critiquing people's behavior on the occasion they turned out to be right. How about following Theo's wallet for a bit longer to wait until his method fails?
'I have never heard of a hedge fund betting on a prediction market....' Well, I can tell you from personal experience that SIG was very involved in one of the busiest markets. Maybe update your priors on that one. :)
Your argument assumes that all we learned on election day was that Trump won. What we actually learned was that he won by a substantial margin, took all the battleground states, lost NY state by 12% instead of 20%, took Texas with an even larger margin than he lost NY by. That's additional information that you are ignoring.
If Silver was right, the probability of Trump winning was about .5 but the probability of his winning by this much was considerably less than that. If Polymarket was right, his winning by this much was not surprising, since it is in a world where his expected margin is significantly more than zero that the probability it will be negative is only .4.
A fair point, but I don't think that is true of other forecasters and I don't see how you can deduce it from the non-money markets. How much Scott underestimated the change in estimated reliability from this event is a much more complicated problem than I am prepared to deal with. My point was that he left out an important part of the information. I think Silver's odds on a popular majority were about 25%, for what that is worth — I don't know what the Polymarket odds were.
It shows Trump at 24% on October 1, 34% on October 21, 36% on November 1, 27% on the afternoon of November 5, over 99% by the early morning of November 6.
I correctly called this election with a much simpler methodology: I just read whatever the smart libtards were predicting and adjusted their takes by how biased I think they are. Scott, pretty smart, so I'll read his stuff, but since he has a slight liberal bias I'll give him a liberal mind virus score of 4/10. Nate Silver, also smart, but his liberal mind virus score reaches 5/10. etc. This gave me a result pretty close to reality.
The difficult part, of course, was giving the LMV score to each pundit. I went with ordering them around from least to most libtard and just going +1 +1 +1.
'All the other real-money markets rose close to Polymarket’s level because of arbitrage'
1) This is not how the arbitrage works, in fact you have it backwards here. Because PolyM was the biggest market, arbitrageurs would be *selling* Trump there heavily and buying Trump elsewhere, spread out among the other markets. They would be pushing Trump odds down below 60%. Not raising the others as high! Whether the fair price 'should' have been 57% is partially unknowable.
2) The WSJ article wrote that French Whale had put in $30m into Trump wins POTUS, but even raise that to $60m. The handle on PolyM was over $3.6 *Billion*. The Whale was a Minnow in the grand scheme of things.
3) You fundamentally misunderstand the nature of +EV and Kelly betting, it does not matter that this opportunity was a 'one-off' event [which it's not, there's always elections and other betting] you should have bet Half-Kelly [to account for less than perfect info] and not a random amount 97% smaller, if you were interested in maximizing your +EV.
The 'this is only happening once, so I can't bet what I should' is a well-known fallacy. Each horse race on each track every day is a unique event, if you have a 10% edge on GreyMareCryptoHODL but bet far less than you should because she is only racing at Aqueduct, one time, ever, versus these 9 other horses than you are doing it totally and completely wrong.
An easy example to show this is a coin flip where the true odds of Heads are 99.999% and Tails is 0.001%. I offer you the choice to bet up to $1m on heads, but only once. If you're only going to bet $5k or $10k, you fundamentally misunderstand the nature of +ev gambling, odds, and risk of ruin.
You state that "If Theo hadn’t bet on Polymarket, it would still have been at ~54%." How are you calculating this price?
I assume you are not factoring in that others buy into the market based on the price. That is if he had not bet, then the prices would have been different and others may have invested.
The amount of bet is not meaningful unless the money bet is literally life changing.
Equally, comparing Metaculus to Nate Silver is comparing different tribe's voodoo witch doctors.
What is spectacularly clear in this election is that the anti-Trump biases in particular, and anti-conservative biases in general, in many of the polls are spectacular.
For example: Trump won the popular vote by 2%. Ipsos, Morning Consult, Forbes and Marist - the top "left" polls, predicted a Harris popular vote win between 2% to 4%. These incompetent morons' own fine print say the margin of error is 3% - so every single one of them was wrong and well outside their own margin of error.
These idiots' polls swung the overall averages to +0.1% Harris as of the evening before election day.
Is it any wonder the prediction markets were screwed? They were relying on accuracy from these incompetent outfits.
This gets worse if you drill down. In the pre-election ACX publication - I noted that no candidate has won only 1 or 2 of the 3 Rust Belt swing states since 1988. I take zero credit for this - it was Rich Baris of People's Pundit who pointed this out.
Yet from November 3 to the end of November 5 - Michigan and Wisconsin had Harris favored by over 1.5 to 1 vs. Trump (i.e. Trump under 40% chance to win). Why? Because the polls said so - heavily skewed by the above polls. This despite Pennsylvania and all the other Swing States showing Trump in the lead DESPITE the garbage pollsters.
Nor is this a one election phenomenon.
In my same post - I noted the same day election poll results then (i.e. 2024) vs. the same day in 2020 and 2016 - and linked directly to the RCP page showing this information. I then contrasted the actual results vs. what the polls showed in the previous elections.
It is now the 3rd Presidential election in a row that the RCP average has been skewed by the above garbage polls to favor the Democrat candidate by a huge amount (i.e. way above margin of error) - and so anyone relying on them (most people) would act accordingly.
GIGO
As for Nate Silver: I am looking forward to see if his one foray into putting his money where his mouth is - which he lost spectacularly - will result in him actually paying:
Something I think consistently gets left out when looking at the efficacy of prediction markets is market manipulation. It could be beneficial for someone to make a huge bet for no rational reason at all, but simply because it benefits them to move the markets.
For example, someone with a lot of money who wants Trump to win might make a large bet just to move the markets, to make things look hopeless in a way that affects turnout.
Or, someone might engage in a classic "pump and dump" scheme. Invest heavily in one candidate (for no reason at all) to get it to look like that person is surely winning, let other traders jump on the bandwagon, and then sell before the results are in.
"Efficient market hypothesis" folks tend to accidentally assume that markets are infinitely deep, and that's just not the case. Even real markets are not immune to this. Large Wall Street funds might have incentive to push markets around occasionally, even large markets like the S&P! If you can afford to push the price a bit to influence some exotic options contract you're in, and the cost to push the market is less than the profit you're going to make, why not?
It's enough of a problem that there are rules around market manipulation, and those rules get broken all the time.
It feels like rationalists think there is some infinitely deep pocket out there, that has perfect information, and is ready and willing to correct any mistake the market is making. That's simply not the case.
There must be two sides to every trade. Either insider trading is rampant, in which case you need a lot of fools betting against the insiders, or insider trading is rare or impossible, in which case you just have a bunch of people trying their best at guessing an unknown outcome.
Agreed. While there is a lot of money in the election betting markets - a Presidential campaign in which $2.6 billion was spent makes say, $30 million used to skew numbers is little more than a single small state ad spend.
There were lots of fools betting. The biggest bettor appears not to be someone trying to manipulate an outcome, but merely a French speculator who saw what he concluded was a mispricing he could make lots of money on.
Have we discussed the bias created by one's own desires yet?
Scott wanted Harris to win, and he thinks the process which gave Harris higher odds was more accurate. I wanted Trump to win, and I think the process which gave Trump higher odds was more accurate.
The urge to believe the set of numbers that tells you the story you want to believe is strong. I don't know how to correct for it.
I largely agree for similar reasons, but I would go further: I'm skeptical that either probability is all that useful a signal for an event that only happens once, unless the odds get pretty extreme. Suppose we have two scenarios:
* It will rain on wedding day
* It will be sunny on wedding day
Whether it's a 50-50 chance or 60-40 chance, you have to plan for both. It's not until it gets to 90-10 or higher that maybe you can think about not planning much for the unlikely scenario, if it doesn't matter much if you're wrong. And the more severe the consequences, the more you have to consider rare scenarios. A 10% chance of death is pretty darned high. Better write your will.
In situations that are similar to an roughly-even election, numeric odds are not much more useful than saying "it could go either way" for planning. (This doesn't apply for gambling or other situations where you can make repeated bets.)
Perhaps I'm biased by working as a programmer. We write code to handle rare events without doing any formal probability calculations. What would be the point? Even if you estimated the odds correctly, such estimates are fragile.
As an individual, I have a hard time getting excited by the difference between 50:50 odds and 60:40 odds for this last election.
Now, admittedly, I have _relatively_ little at stake here - and the part that is at stake (e.g. whether there is a nuclear exchange) would not be well-constrained by even perfectly knowing the results of the election in advance.
Now, there are public policy questions where the difference between 50:50 and 60:40 for e.g. the odds for another pandemic in N years matters.
For me as an individual, even for fairly dire events, the odds have to reach something like 90:10 before I feel comfortable disregarding the low probability side.
To phrase it as a question: _Other_ than for the question of whether to place a bet, are there _any_ current prediction markets where anyone reading this would take some personal action differently depending on whether the market odds were 50:50 instead of 60:40?
You are interested in the question: Is Polymarket more reliable (or some variant of that).
But a different question is: why was a blowout predicted by polls (and poll-adjacent, like Nate Silver) as 50%?
The link between the two is Theo, and not in the sense that he guessed lucky, but in the sense that he seemed to realize that the polling industry was as morally and competently bankrupt as the news industry, and put together a somewhat more functional alternative. Polling is a wasteland because of
- push polls and similar nonsense, which mean that anyone who isn't terminally lonely or braindead hits "end call" the moment they realize the caller is doing political polling
- our betters running our institutions have made many people scared, even in the context of talking to a stranger, of voicing opinions that might get them cancelled.
Theo worked out a scheme for dealing with the second of these, if not the first, the "neighbor" method, and got the results he got.
The obvious takeaway is that some sort of alternative to current polling is required, in the sense that X is an alternative to the pathologies of the traditional news media.
Markets may play some of this role, but plenty of people who are not comfortable with the specifics of how a market works, or just don't care enough, are happy to answer questions (IF the questions are honest, a big if -- as I said, push polls, or even the GSS which is clearly a push poll even if the people in charge of it are so philosophically ignorant and clueless about the rest of the country that they can't see it; and if they don't feel there's some sort of blowback possible).
I'm not sure where this goes. Maybe X-Polls, if in 4 years X is considered trusted enough by a wide enough range of people? Maybe polls as short games on ad-supported platforms? But I think the current industry is kaput, they lost their last shreds of credibility this election. Even those, like Nate Silver, who probably aren't deliberately biased, just did not do enough to cope with the social environment in the country - in spite of the fact that this environment was clear to even a French outsider...
Say I have a coin that I suspect of being biased, so I make a market to figure it out.
An anonymous whale spends millions of dollars performing a detailed scan of the coin, builds computer model of the coin, and flips the simulated coin 1000 times. Then he bets heavily that the coin is biased, which substantially moves the market in that direction.
In this scenario, nobody would argue that the market is now mispriced. Right? The whale introduced data into the market. The whale’s model is incorporated into the wisdom of crowds that is the market. Another way of saying this is that the whale *knows that the coin is biased* and he is informing us of this fact, up to the limit of the amount of money he’s comfortable betting.
Is it possible that a different whale comes along and bets an equal or greater amount in the opposite direction? Yes, but it’s *less likely* because the coin is in fact biased and proper study and analysis of the coin should indicate that. An expensive study that says the coin is fair (or biased in the other direction, or whatever makes sense in this metaphor) would imply that someone spent a lot of money and made a huge mistake in their analysis. Can this happen? Sure. Mistakes, mismeasurements, and misjudgments are always possible. But you have to think that in a reasonable-ish world, the results of analyses should correlate with the actual state of the actual coin.
The thing about elections is that the outcome is predictable. We call such things “stochastic” as a shorthand way of indicating that we can’t practically know all the details that would be required to get the right answer. But Omega would simply figure out the answer, but building a detailed simulation of the entire Earth and everyone on it, and running it. A really expensive study commissioned by a French whale is not quite at that level but it’s somewhere on the spectrum between Omega and “some guy’s opinion.”
What bothers me is how whales are getting access to crypto. Is this what those guys are doing with the LLM that speaks whale?
How did the whale validate the code that he used to model the coin flip?
If the answer is "I dunno, but the idea of modeling the coin flip is clever and neat and he clearly put a lot of work into it and is confident of the result", then Oh Lord No. That's the formula that, in other contexts, gives us the Replication Crisis. If that's the formula that will set the price in a prediction market, then that market's predictions will at best be *slightly* better than random chance.
Typically you would do a sensitivity analysis to understand the impact of your assumptions on the outcomes. This is how modelers across all domains understand their model uncertainty, and their measurement uncertainty.
Perhaps he would conduct the same procedure on a coin known to be fair and verify that the simulation approach calculates that this coin is in fact fair.
Since this is a metaphor for the actual polling approach that took place, the truth is that we don’t know what he did to validate or analyze the model, all we know is that he put up the money and then won.
I think you sort of have to have a prior that is something like “don’t assume maximum uniform incompetence at all scales at all times.” It’s possible that the computer modeling experts simulating the coin flip forgot to convert from Imperial to SI and got a flagrantly stupid result, which nobody bothered to check for some reason. But there’s a kind of conceptual convexity to a lot of science/engineering tasks that means if you make a significant error in one part of a model that will show up as a weird result or inconsistency elsewhere in the model, and then you are forced to track it down. Perhaps this convexity doesn’t exist in soft science domains. It would be “hard to miss” such a mistake in the coin flip simulation example. You would notice problems and inconsistencies, unless you were just very incompetent, and if you’re very incompetent, why do you have tens of millions of dollars to bet on this question? Is wealth uncorrelated or negatively correlated with competence? Probably not, because if it were so, prediction markets wouldn’t work at all.
Finally, I think maybe your comment here and Scott’s OP indicate a mindset that is perhaps excessively outside-view. If you were sitting at a bar talking to a friend and they casually described the “neighbor polling” method and then told you that it suggested a massive advantage for Trump, you would probably not reply with the sort of Cartesian skepticism that I’m seeing in these comments — you would probably have said, “Oh word?” and adjusted your prior.
Tbh is seems to me that the consistent generalized for me of your argument would imply that prediction markets cannot possibly work as methods of information aggregation because people are sometimes mistaken. I’m sure that’s not what you mean or what you think. But if you have an epistemic carveout for “whales,” i.e. “bettors who have resources and confidence,” then on net you’ll just end up throwing away Bayes points.
"I think you sort of have to have a prior that is something like “don’t assume maximum uniform incompetence at all scales at all times.” It’s possible that the computer modeling experts simulating the coin flip forgot to convert from Imperial to SI and got a flagrantly stupid result, which nobody bothered to check for some reason. "
Have you ever done any serious computer modeling? Because I have. I am, in fact, taking a break from a fairly substantial modeling project to wander through the ACX garden right now. And I think I'm on pretty solid ground in expecting a *lot* of incompetence in the field, albeit not uniformly distributed.
Oh, and the errors that matter aren't usually silly mistakes like "Oops we forgot to convert to SI", but "Hey, the (n,7Li) reaction cross section is really tiny under these conditions, we don't need to waste clock cycles tracking it, do we?" Serious modeling, or polling, is filled with the opportunity to make mistakes like that. And I don't count it as incompetence when people make those mistakes. But it all too frequently crosses over into incompetence when people place high confidence in an untested model and e.g. neglect to fully evacuate the test area around Bikini. Betting forty million dollars on a single run of your untested model would be a lesser degree of incompetence, to be sure.
"If you were sitting at a bar talking to a friend and they casually described the “neighbor polling” method and then told you that it suggested a massive advantage for Trump, you would probably not reply with the sort of Cartesian skepticism that I’m seeing in these comments — you would probably have said, “Oh word?” and adjusted your prior."
I can assure you that while I would in such a situation politely say "Oh, that's really clever!", in roughly the same way I'd say that to a Las Vegas gambler explaining to me that he has a System that will let him Beat the House. I might ask about the details of validation and testing, but if those aren't forthcoming I will be changing the subject and the only prior I will be adjusting is the one about his being an overconfident fool.
That doesn't, and shouldn't, change if I then see him place a single bet on Red at the roulette wheel and win.
You have activated my trap card. I am a computer modeling specialist with a PhD in programming computer models of various house al systems in Fortran and a 15 year career in applied simulation.
Unfortunately though you didn’t say much here that I disagree with so I don’t have too much of a rebuttal.
I’ve seen a lot of shoddy modeling in my day. The Imperial to SI remark was referencing the Mars mission that was lost due to exactly that error. My intent wasn’t to say that it’s impossible to do modeling badly, and I can see how it came across that way.
On the other hand, I’ve also seen a lot of incredibly good modeling work that was disregarded because the answer output by the model disagreed with the institutional wisdom of the organization that commissioned the model. I think I have personally seen probably ten billion dollars wasted in my own field due to people not taking seriously the conclusions of simulation, not for any principled reason but because “simulation is untrustworthy.” And anybody can always endlessly nitpick the assumptions of a given model, because even career modelers like me will admit that all models are wrong. You do the best you can with the tools and time at your disposal and if the client or stakeholder wants an excuse to disregard the work, they will always find it.
This is why I am somewhat sensitive to the blanket skepticism that some lone nerd might be able to beat the house. Lone nerds beat the house all the time. We live in a world dominated by lone nerds who made insane bets and won. It is the main way of getting ahead in this world.
Final remark here is that the original conversation wasn’t really about modeling so much as data gathering. The French Whale performed a new kind of survey, basically. So metaphorically this is less like making a computer simulation of a coin flip and perhaps more like using a specific tool that simply measures the imbalance of the coin’s center of mass, in a circumstance where nobody else has done that sort of analysis. Viewed this way the French Whale is very directly and straightforwardly adding data to the wisdom of the market. It seems that if this character had bet $1000 instead of tens of millions we wouldn’t be having this conversation at all, and that fact seems to indicate that the claim is something more like “he was overconfident” (not necessarily true if he is very rich) or “this bet distorted the odds too much” (he was correct; he won the bet, so his bet *correctly updated* the odds in this case.)
I like this piece, but it overlooks one crucial fact -- how much you should update your probabilities ought to depending on the scale of victory. Election outcome isn't binary -- win or loss. To take an illustrative extreme example: suppose your prior is 90% that probability is 50-50 and 10% that it's 60-40. Then, election results come out, and Trump has swept 48 states. Your prior ought to be updated with greater magnitude than a closer victory.
...Why are you acting like he didn't tell everyone exactly what he would do? There's nothing stopping him anymore. No need to be concerned with being re-elected (legitimately), immune to consequence, full control over every branch of government, and most importantly, he has the consent of the public. This is what the people want. Would you deny them catharsis?
Prosecuting political opposition, bombing Mexican cartels, ending the Ukraine war by ending support to Ukraine, a bunch of tariffs, using military to stop protests and crime without consent of the states, deporting as many illegal immigrants as possible (and legal immigrants if they're undesirables), performing an ideological purge of federal agencies, and a bunch of other stuff I'm probably forgetting.
You can do a lot in four years when you actually have ambition.
I doubt this will happen on a large scale. His track record on this one is already weak(muh swamp, muh lock her up). But he will try to get a bunch of people that particularly annoy him fired(he did this in the past).
>bombing Mexican cartels
This sounds really based. I doubt he will do this though. Or, if he does, it will be probably a performative bombing with consent from the Mexican government.(Military operation decisions in his first administration always involved a lot of talk with generals, and they were mostly successful in dissuading him from doing truly unhinged stuff).
>ending support for Ukraine
Possible. But he sends mixed signals on the Ukraine issue all the time. According to him he will simultaneously save Ukraine by being the toughest president against Putin in history, save everybody by signing the best deal in the history of deals, and kill NATO because the European countries don't give enough money to the alliance.
>bunch of tariffs
It can happen, it would suck. But the GOP may dissuade him, it is not clear that the rest of the party is on board with his populist economic stuff.
>using military to stop protests and crime without consent of the states
This sounds based. I hope he does this. Him being a pussy during the George Floyd riots and refusing to send the military probably hurt his image. Also, the US military is pretty well trained compared to your average cop, they will not go around slaughtering random protesters, they will probably end the chaos pretty efficiently.
>deporting as many illegal immigrants as possible
Probably will happen. It is the second campaign he runs promising to deport everybody, if he fails to deliver AGAIN, I can see even hardcore Trumpers turning on him.
>performing an ideological purge of federal agencies
Elon will probably champion the idea of firing a bunch of people, this seems to be the idea behind the DOGE thing(holy cringe name). Trump will probably just want to fill a bunch of positions with his most loyal buddies(like he did the first time). We'll see who wins. At the end of the first month we either have a Milei style purge, or Dana White as Secretary of State(maybe both?).
Of course, there is the fact that Trump is really unpredictable, so he can very well find very creative ways to do shitty stuff that we will never see coming and can very well be worse than all the stuff you're expecting. But, I'll maintain that most of the stuff you're predicting won't happen.
I cannot see this. Hardcore Trumpers will find some reason to support anything he does, just as hardcore Democrats will support the candidate under any circumstances.
"the idea behind the DOGE thing"
The DOGE will do what a DOGE does, when in the presence of a Donald doing diplomacy, that is.
...You're free to gloat all you want. I just hope you're prepared to deal with the consequences of your actions. Just because you're on the winning side doesn't mean you won't be forced to make sacrifices.
It was more of a grim joke. I actually don't think he'll be good for the country.
I have given up on Ukraine and would be OK with bombing Mexican cartels, though, especially if he gets the OK from Mexico. I think even the Mexicans (probably *especially* the Mexicans) are sick of them.
>Just because you're on the winning side doesn't mean you won't be forced to make sacrifices.
nit: I gritted my teeth and pushed the button for Trump, but I don't think of that as being on the winning side. I think of him as being _marginally_ less bad than Harris (assuming that her administration would have more-or-less been a continuation of the Biden/Harris administration), largely because of Harris's anti-free-speech statements and corresponding Biden/Harris administration actions. But I don't gloat to have a liar in the White House, one who suggested looking into injecting disinfectant, with the implications of that for Trump's common sense.
He watched a lot of TV in his first term. He is not energetic, not smart, not well organized, not practical and not good at finding people who are. He's good at picking people who'll do what he wants, but not at figuring out who actually knows how to manage this and that huge, complicated novel task.
He is very energetic, specially for someone his age. But he doesn't seem to like boring policy stuff, he saves his energy for rallies, speeches and golf(the stuff he actually enjoys).
Politicians never keep their campaign promises and Trump never keeps ANY promises. No doubt among the many things you said he would do there are some he never had any intention of doing and there are some he genuinely wants to do. That's why OP's question would actually be fun to speculate about.
There were lots of things he said he'd do. He couldn't do all of them even if he met no legal resistance or political pushback. Four years is nowhere near enough time. Deporting 15 million people, for instance, is a huge job. The logistical problems alone are huge.
“You shouldn’t update much on dramatic events!” That depends very much on the event. A baseball game, no. A boxing match, absolutely yes. Just because the odds are 50% for each does not mean that the predictions are equivalent. And an election is a lot more like a boxing match.
In bayes terms, a fair coin flip has priors of 100% 50/50. A boxing match has priors of 50% 90/10 and 50% 10/90.
It’s a serious conceptual error to equate the two, and very common.
Baseball has notoriously small edges for a great team. This years World Series winners, the dodgers, only won 60% of their 162 regular season games. If a baseball team wins a matchup it doesn’t tell you very much whether they will win the next matchup.
In boxing, like chess, the better boxer wins very often. Unlike chess, bouts are so rare that it is difficult to assess how two boxers compare unless they actually fight. So when two boxers fight for the first time, odds will be relatively even, but their next match the odds will be heavily favoring the one who won previously.
A) predict it is not liquid. Liquidity is important for accuracy
B) we don’t know if the person who lost 5 million was hedging. My gut says they’re hedged.
C) the whale is known and he literally ran his own polls before the election which is why he was so sure. This is exactly why non money prediction markets are garbage. It’s just a bunch of people full of hot air like Nate silver speculating.
Nate Sliver, I believe, lost a $100K bet on the election (FL Trump margin). Not sure if he put more money elsewhere, but suggesting he is just full of hot air is incorrect.
Nate Silver is following the polls, if his data is 'off' for some reason, what do you want him to do? (shade them in the direction he would like?) (I guess he needs to find better data!)
Rothschild, David M. and Sethi, Rajiv, Trading Strategies and Market Microstructure: Evidence from a Prediction Market (November 22, 2015). The Journal of Prediction Markets
The paper looked at Intrade's trade history for the 2012 election, finding that a single whale (codenamed “Trader B”) spent >$300K on election day betting on Romney, which propped up his winning odds from 20% to 30%. (Yeah, markets had less liquidity back then.)
Romney lost by a wide margin. Part of that loss was to arbitrageurs from other venues like Betfair.
The paper puts forth a few hypotheses:
“This leaves three possibilities: (i) the trader was convinced that Romney was underpriced throughout the period and was expressing a price view, (ii) he was hedging an exposure held elsewhere, or (iii) he was attempting to distort prices in the market for some purpose.”
You talk about adjusting priors based on “Theo” doing (i), but suppose if his goals included (ii) and/or (iii), is the same adjustment warranted? A banker should have large exposures in other financial assets, in addition to the yachts.
"prediction markets work because we expect nobody to have vastly more money than anyone else, giving everyone a fair chance to compare their opinions"
Given the actual inequality in the world, why wouldn't this cast prediction markets into serious doubt? It sounds like they are premised on something that is simply false.
It's a bit strange that people writing about "real money markets" ignore every sportsbook around the world. Sportsbooks took millions and millions of bets, more than enough to offset one French whale, and landed at 60%+ for Trump on the morning of the election.
Even if you ignore them, Robinhood alone created an election derivatives market that had hundreds of millions of dollars in volume and started the day with Trump at 63c. There are more real money markets, Horatio, than are dreamt of in your Manifest conference.
I suspect a lot of the guys who do this stuff aren't big into sports. (Myself included.) Stereotypes are that way for a reason. :)
BTW, Googling 'dating in the second person' gives you articles on polyamory. Which I know you are an advocate of, but is not the subject of your blog. :)
I've enjoyed your past stuff, also BTW, but have aged out of the market and it's no longer applicable. But the 'dating as trading' metaphor is interesting; I spent way too much time trying to figure out what archetype to try to turn myself into. Had I read your article 20 or even 30 years ago, things might have been different.
The January 2024 Less Wrong post "Will quantum randomness affect the 2028 election?" makes clear that uncertainty about election outcomes is almost entirely epistemic uncertainty, not aleatoric uncertainty. Thus, whichever outcome happens is very strong evidence of what the true aleatoric probability of that outcome happening was. Specifically, it is very strong evidence that the outcome that happened was very likely to happen. Therefore, a day before the 2024 election Trump was very likely to win (>99%). Therefore, Polymarket's ~60% price was priced too low, not too high. https://www.lesswrong.com/posts/HT8SZAykDWwwqPrZn/will-quantum-randomness-affect-the-2028-election
I agree that election outcome uncertainty is epistemic, but that doesn’t mean that we should judge the market result by whether it aligns more closely with “objective” probability, whatever that even means! Heck, if you believe the universe is deterministic, then Trump winning was always 100%. But that bears no relevance whatsoever for whether betting for Trump on polymarket at 60% odds was the right bet. That is always going to be fundamentally a question about epistemic probability. Saying otherwise strikes me as very confused (plus it goes against the Bayesian framework that Scott adopts here).
So I see the points in the article about the initial perturbations in chaotic systems having huge effects as being irrelevant, because our epistemic data doesn’t come close to establishing the initial conditions with any sort of precision to allow us to predict those outcomes.
I think we should interpret questions like “were the true probabilities 50/50 or 60/40?” to be about the ratios of possible nearby universes where Trump or Harris won. In short, we have limited epistemic data about physical reality, and this data constrains the way our physical universe can be. But our data is also massively underdetermined, in the sense that there are a lot of ways the actual universe can turn out to be configured in a way that is still entirely consistent with our epistemic data. Questions about correct probabilities are thus just questions about the configuration space of nearby possible universes where all the epistemic facts are held fixed.
To say that the “true” odds were 50/50 is thus just to say that “holding all the epistemic facts before the election fixed, the ratio of possible ways to configure the universe where either Trump or Harris won comes out 50/50. There are just as many possible configuration spaces where Trump won as where Harris won, given the epistemic data”.
Of course there are questions about how to sort the configuration space, and I don’t have easy answers here, but I think in the main that’s broadly what people are gesturing at when they talk about these kinds of probabilities.
"That is always going to be fundamentally a question about epistemic probability. Saying otherwise strikes me as very confused (plus it goes against the Bayesian framework that Scott adopts here)."
Yeah, I agree with you, sorry for not making that clear--my initial comment was trying too hard to be concise and I forgot to clarify that I agree that "what the market price ought to be" (what Scott Alexander was debating) is not equal to the aleatoric uncertainty, but rather is more uncertain than the aleatoric uncertainty due to inevitable epistemic uncertainty that will remain.
It wouldn't be useful to say that market prices ought to exactly correspond to aleatoric probabilities, because that would be impossible with irrational traders given that real world traders are inevitably going to have some epistemic uncertainty. So I agree with you on the goal posts here.
"Heck, if you believe the universe is deterministic, then Trump winning was always 100%. But that bears no relevance whatsoever for whether betting for Trump on polymarket at 60% odds was the right bet."
Well it does bear relevance, because it's evidence of what the best forecast possible (given inevitable irreducible uncertainty in the world) was.
If aleatoric uncertainty was >99% Trump (as I claim the actual election outcome shows us is true), then the best possible forecast (reflecting both epistemic uncertainty and aleatoric uncertainty combined) was very likely to be >50% (since we have no reason to believe with >50% confidence that the universe fools people into doing worse than chance on this question when they become informed) and probably something like ~65%-90% (as I wrote in another comment elsewhere here).
It's still going to be <90% rather than >99% because of the irreducible uncertainty in the world, causing some remaining epistemic uncertainty even for the highest skilled most informed forecaster possible.
But saying that it's only 60% or less seems implausible to me, and we certainly can't establish that it's that low just by pointing out that Polymarket had a big low-information pro-Trump trader, or that Metaculus was at ~50% or <60% and Metaculus is pretty good.
The reality is that even people on Metaculus who we think were relatively informed, or even people like Nate Silver, were likely far from the limits of what is epistemically possible in terms of being informed and rational-at-making-forecasts-given-the-information available.
"I think we should interpret questions like “were the true probabilities 50/50 or 60/40?” to be about the ratios of possible nearby universes where Trump or Harris won."
Yes, I agree, if by "true probability" you mean "aleatoric probability". And note that before the election I had some weight, perhaps ~5-30% on the aleatoric probability being 10-90% Trump. If the election had wound up very close, e.g. PA being decisive and Trump winning by 1,000 votes, I think we'd be right to conclude then that the true aleatoric probability was not >99% Trump, but rather something like 10-90% Trump. (A good example is the 2000 election: The aleatoric probability the day before the election that Gore would win was probably 10%-90%, not <1%.)
I think that in forecasting contexts we can just ignore the hard-to-figure out quantum perturbation stuff (I don't know much about quantum mechanics) by using thought experiments that introduce noise to simulations. The way I was thinking about this is:
Suppose we re-run the universe a million times starting a week before the 2024 Presidential election, however we ensure that even if the universe is deterministic that we don't get the same outcome 100% of the time, by introducing small amounts of noise. For example, we cause a person to sleep-in 5 minutes later than they actually did in reality. Or we cause someone to get a flat tire on their way to work. Or we cause a tire that did get punctured to not get punctured instead. Etc.
Due to chaos / the butterfly effect, I think these sorts of interventions would be extremely likely to change the amount of votes that each candidate receives. In close elections, these interventions would even change who won some of the time. But they wouldn't change it so much that Harris would win almost half of the elections. I still think that in 2024 Trump would have won >>90%, probably <99%, and maybe even >99.9%. But not 100%. Less than 100%. (And if you think 100% still, then let's chance the thought experiment so that the simulations start a month before the election rather than a week, etc, to reduce the aleatoric uncertainty by enough that even in a million simulations with different bits of noise injected into each one, the same candidate wouldn't win all one million of them.)
"But our data is also massively underdetermined, in the sense that there are a lot of ways the actual universe can turn out to be configured in a way that is still entirely consistent with our epistemic data. Questions about correct probabilities are thus just questions about the configuration space of nearby possible universes where all the epistemic facts are held fixed."
I think we should distinguish between two questions: (1) What forecast should a forecaster make given the information that they have, and (2) What is the best possible forecast a forecaster can make given irreducible uncertainty in the world?
Re (1), a low-information forecaster should probably have had a credence of ~40-60% Trump, and would have been luckily-overconfident if they had had a credence of >60% Trump. We can say that this person's forecast was irrational if it was >60% given that they had sufficiently low information. They had no way of knowing the odds were >60%.
We can say this about the low-information forecaster *even if* it is true that the most informed forecaster possible could obtain a rational forecast of, say, 70% Trump. As I said previously in this comment, I think the answer to (2) is much more likely to be ~70% than <60%.
In other words, I think that the most confident the most informed person could have been in favor of Trump without being irrationally overconfident was probably ~70% (or ~65%-90% as I said before).
This does not mean that the French whale was not overconfident. This does not mean that Metaculus forecasters were necessarily underconfident--they could have just been ignorant, low-information forecasters assigning probabilities rationally given their high epistemic uncertainty that they didn't reduce as far as was physically possible.
"Questions about correct probabilities are thus just questions about the configuration space of nearby possible universes where all the epistemic facts are held fixed."
"To say that the “true” odds were 50/50 is thus just to say that “holding all the epistemic facts before the election fixed, the ratio of possible ways to configure the universe where either Trump or Harris won comes out 50/50. There are just as many possible configuration spaces where Trump won as where Harris won, given the epistemic data”."
So if by "correct" here and "true" odds you meant "rational, given the information I actually have", I agree with you.
But I don't think that's what Scott is saying in his post when he claims that the odds should have been and Polymarket was thus over-priced.
What I think Scott means by "true" when he says (your) "the “true” odds were 50/50" is "what the market price ideally should be, i.e. what it would be if it were determined by maximally-informed rational traders who reduced their epistemic uncertainty as far as possible, up to the limit of irreducible uncertainty."
And I don't think you can establish that without either showing that the person(s) you're talking about who were at 50/50 were maximally informed. And I don't think they were. I think maximally informed people would have been at ~65-90%.
>Suppose we re-run the universe a million times starting a week before the 2024 Presidential election, however we ensure that even if the universe is deterministic that we don't get the same outcome 100% of the time, by introducing small amounts of noise. For example, we cause a person to sleep-in 5 minutes later than they actually did in reality. Or we cause someone to get a flat tire on their way to work. Or we cause a tire that did get punctured to not get punctured instead. Etc.
>Due to chaos / the butterfly effect, I think these sorts of interventions would be extremely likely to change the amount of votes that each candidate receives. In close elections, these interventions would even change who won some of the time. But they wouldn't change it so much that Harris would win almost half of the elections.
Agreed re noise and chaotic dynamics. And, even setting aside quantum mechanics, the effects of thermal noise in a classical universe do very similar things...
>And I don't think you can establish that without either showing that the person(s) you're talking about who were at 50/50 were maximally informed.
There is one sense in which "maximally informed" is ill-defined. The number of people one could poll depends on the budget one has for it. To pick an even more extreme case, given an infinite budget, one could pay voters to tell their opinion (not to _change_ it, this isn't vote buying), and pay to establish a credible case that one will not disclose their answers and will not take any action for or against them based on their answers. In the limit, one polls every single eligible voter.
Now, this still will miss the voter that gets the flat tire, or the voter who changes their mind between the poll and the final vote.
But ... still, this is possible in the real world (no omniscient omegas involved). So, how close to this does one have to get to count as maximally informed?
"So, how close to this does one have to get to count as maximally informed?"
Good question, and I agree that my "maximally informed" was ill-defined. What I meant was that a "maximally informed" person is a person who reduces their epistemic certainty to near-zero and is only left with aleatoric uncertainty.
How close to zero epistemic uncertainty it possible to get?
In one sense, your person with an infinite budget who polls everyone will have extremely little epistemic uncertainty, and will only fail to predict who fails to get to the polls on election day do to a surprise wrench in their plans, etc. (Note that I think this hypothetical person would have been able to get to 90+% Trump.)
And in another sense, your hypothetical person is more informed than it was actually possible for anyone to be, since no one actually had the budget to go interview every eligible voter in a trustworthy way to get them to disclose who they planned to vote for and how likely they were to actually end up voting in a way for that person to reveal that information honestly.
So I think what I meant by "maximally informed" was actually a lower bar than your hypothetical. I meant basically 'as in formed as possible' in a practical way, rather than according to the laws of physics, or some higher standard.
Perhaps another way to say this is to point out that whoever you think was the most informed in reality (Nate Silver and his team?) weren't tackling this problem with a $100M budget, and probably weren't even tackling it with a $10M budget, but if they were, they would have been better informed. (They obviously weren't tackling it with the near-infinite budget that'd be required to gain the trust of every eligible voter and get them to honestly share with you the knowledge they have about who they plan to vote for, but that's too high a standard). Spending a $100M budget on hiring a team of forecasters and support to make a massive effort to forecast the 2024 election would have at least been realistically plausible. E.g. Someone who cared to make that happen could do that for the 2028 election. There'd be no guarantee that the project would make a good forecast just because it has a large budget, but it'd be much better equipped to become more informed and figure out the truth of who is more likely to win in an election that would not wind up being that close.
Had the 2024 election wound up like the 2020 election, even a team with a $100M budget might have struggled to determine the likely winner. Though perhaps that's because even in retrospect determining who was most likely to win the 2000 election is hard. But given that the 2024 election wasn't very close, I expect that very informed people (e.g. with a $100M budget and strong forecasting skills in general, would have been able to figure out that Trump was >60% likely to win, and probably >70% likely to win.
Presumably an extremely informed group that settled on e.g. 75% Trump before the 2020 election wouldn't have been literally maximally informed, but they'd be close enough for my purposes. And mainly I just want to say that I think it was theoretically possible for a group to become informed enough to come to a 75% forecast for Trump like that without being irrationally overconfident.
>Presumably an extremely informed group that settled on e.g. 75% Trump before the 2020 election wouldn't have been literally maximally informed, but they'd be close enough for my purposes. And mainly I just want to say that I think it was theoretically possible for a group to become informed enough to come to a 75% forecast for Trump like that without being irrationally overconfident.
That seems reasonable.
I think there are at least two other plausible candidates for a "maximally informed" bettor.
The higher bar one would be someone/some group that had access to all of the data from all of the polls that had actually been made. This is arguable, since some of the polls (such as Theo's) were completely private, and some (most?) of the polls that disclosed their aggregate results weren't disclosing their full raw data. So, these considerations make this a somewhat unreasonable standard (though still below my hypothetical of a pollster with an infinite budget :-) ).
The lower bar would be whichever person/group had the "most" poll data actually available to them. One of the snags in this bar is that the measures of the amount of poll data aren't quite comparable. The set of polled voters available to one group are unlikely to be a strict superset of the set available to some other group. If one group polled 20,000 voters, and another polled only 15,000 but did a better job of matching their polled voters to the likely-to-actually-vote population, who has "more" data?
Reasonable thoughts. Note though that poll data is far from the only kind of data that can inform a forecaster about the probability of Trump winning.
E.g. Forecasters could have learned about the probability of Trump winning by watching the Presidential debate, by watching Trump's interview on the Jake Paul podcast or the Joe Rogan podcast or Harris interviews, by seeing what kind of political discussions people are having on Twitter, by seeing what kind of reasons people give for supporting or opposing Trump, etc, by seeing how these things compare to what Biden looked like during his 2020 election campaign, by looking at voter registration numbers by party in different states compared to the 2016 and 2020 election, by observing whether your known Republican neighbor bothers to put up a Trump sign this year or not, by reading political philosophy and finding out whay Jason Brennan means by "hobits" in a political context, etc. All of these things provide information.
Your "That seems reasonable" reply suggests to me that you agree with me that Scott's claim that Polymarket had Trump-wins priced too highly is wrong. Is that the case?
Other prediction markets did move in sync with Polymarket, the distance was usually 2-4%, the biggest gap I found was Preidctit vs polymarket oct 29th, about 7%... and it was quickly corrected.
I didn't arbitrage myself, for the same reasons many people didn't. Too much risk (of markets going under), too much fiscal hassle (banks and tax authorities tend to look at your really funny when really large sums leave and enter accounts and these are events I'd rather avoid because e.g. increasing risk of going up a KYC tier or getting thoroughly audited is just not wroth it), too little gain (fees eat at profit), and too much hassle (you have to keep buying/selling every day, very few orders are at market price)
This is not a *failure* of prediction markets as much as a combination of a few failures of scale, distribution and communication style that will be smoothed out:
1. Different ways markets turn books into probabilities
2. Fees
3. Limits on bets
4. Limits on betting legally from certain jurisdictions
5. Trust by people with money in the platform
6. Financial hassle
5 and 6 looks like "Yes you could make a 3-4% return a month by arbitraging x vs y but you need a multinational structure, some pretty on-top lawyers and accountant, and should still account for a large risk of one of these orgs going bankrupt or rogue. And at that point many other arbitrages are equally tempting and (even though this seems surprising), there isn't enough leverage in the world to arbitrage all markets.
I suspect that many people made loads of money by arbitraging and I expect this will be less of an issue as time goes on and more money floods the markets & track records get established & insurance offerings pop up.
Not sure if this is the thread to discuss predictions generally, but I just checked Allan Lichtman's particular 2024 prediction (https://hdsr.mitpress.mit.edu/pub/kcnosyqm/release/1), and it must be maddening to Lichtman that he could have predicted a Trump win if he had just utilized his own (flawed) model rigorously and objectively! Lichtman says that there were four flipped keys against the Dems, but it seems obvious to me that Policy key and Foreign/Military Success key should have been flipped as well: Biden didn't achieve a New Deal level policy change (insofar as I've understood this is *not* supposed to be an easy key to flip), and Ukraine is not a Foreign/Military Success, as the Ukraine/West coalition has not, uhh, actually won the war.
Yes. His 13 Keys model is still very good. He just allowed too much personal political bias in how he applied it this time. But then, that is an edge that a purely mathematical model would have. I think it's easier to put personal bias aside when you're just dealing with poll data and/or prediction market data, and not having to judge if a candidate is charismatic/hero-like or had a major policy achievement.
I think 3 other keys are at least in doubt: 1. Contest (sitting President did have to be forced off the ticket, even if then party unity resulted)
2. Scandal (same events, really: a sitting President becoming senile, this being covered up, and then resolved behind closed doors)
3. Challenger charisma (DJT fails part of the definition because he doesn’t appeal across the board, but he smashes the other part which is being a once-in-a-generation transformational figure)
ISTM in all these cases there’s some typical-minding going on: AL perhaps doesn’t see those things as I have described them, so he doesn’t notice other people do see them that way.
Good points. A sitting President having to be forced off the ticket strongly suggests *some* sort of scandal, like you said. In this case, the scandal was the covering up of Biden's declining mental faculties.
Neighbor polling. Sounds interesting. Do you know of any good links... I find a lot of trivial stuff about the whale Theo.
Oh, and I think Scott missed the point of this election. Money markets will rule, because some people will go to all sorts of lengths if money is involved.
I overlaid the main events on the probability chart and you can see that Trump’s odds never went above 52% after Harris got the nomination … that is until Musk started paying swing-state voters to register and Trump’s odds then rose from 49% to 64% (right before the MSG rally which knocked them back down to 54%, which was the opposite of the actual Latino result)
Which argues that prediction markets (currently) lack some of the features and players common in other markets, stalling their adoption.
The Whitaker and Mazlish is a long article, difficult to summarize, but does invite one to wonder if one could catalyze adoption rates by playing with the right levers.
It inspires this very Levine-ish line:
"“Prediction markets are not a natural savings device” seems obviously true, and a good reason to think that prediction markets will never compete with stock markets for size and liquidity and efficiency. (However: “Design a product for retail investors who want to use prediction markets as a savings device” strikes me as a very good exercise for the reader, and I expect that within two years I will be writing about a buy-and-hold prediction-markets exchange-traded fund.)"
Caveat, Levine very handwavily endorses the default narrative on Théo. But his article also brings a lot of whimsy, which if not grating, ends up fungible with accuracy to some exchange rate.
He closes by noting:
"Oh meanwhile, amusingly, Polymarket might be illegal in France, where Théo is apparently located, oops..."
I recommend the whole article though. It's paywalled, but his column is free if you just sign up for his mailing list like it's 1999.
I get the feeling that Scott's punishing us with this dry spell for electing Trump.
I didn't vote and had no plans to vote ( https://youtu.be/vJriNA37xng ) but had I have known it would effect our intrepid host I wouldn't have been so cavalier about TheMostImportantElectionInHistory™.
More seriously, Scott, if you're actually hurting over this I hope your legendary resiliency asserts itself and you bounce back with vigor and joy.
Is there a possibility that the mispricing was primarily due to the rare occurrence of a US election, and can isn't it obvious then to keep expecting more mispricing and arbitrages due to the relative infrequency of these events?
You “couldn’t be more delighted”?
It's a figure of speech, I assume wireheading would work.
I get the figure of speech. It’s more about being extremely delighted in reference to the broader political and international context - things that are slightly more important than the empirical accuracy of a gambling website.
I've edited it to "I couldn’t be more delighted by their progress," but under slight protest. At every moment, millions of people are dying of war and preventable disease, but you still ought to be able to say you're delighted by specific things sometimes without adding "...aside from the broader political and international context, of course".
This isn’t any old day…
No . . . it's my fortieth birthday! But despite the specter of impending mortality I'm still able to spare a little happiness for Polymarket's victory : )
Happy birthday!
Happy birthday!
Happy birthday! <3
Happy Birthday!
Happy birthday, Scott!
Happy Birthday!
Happy birthday!
I write this 44 minutes after you posted it, which is numerologically significant to me more than you, but HAPPY BIRTHDAY. :D
Have a Happy Birthday buddy!
Happy birthday!
Happy Birthday!
Happy birthday 🥳🥳🥳🥳🥳🥳
Many happy returns! (And I hope you get a good night's sleep.)
Happy Birthday!
Happy birthday!
Happy Birthday!
Please explain how critically important today is while neither your nor my life changes at all, beyond having to deal with a few more hysterical whiners than usual for a few weeks
Please explain how critically important the day that we discovered a dinosaur killer asteroid will hit Earth in 2028 is, while neither your nor my life changes at all, beyond having to deal with a few more hysterical whiners than usual for a few weeks.
This is such a dumb argument.
A delightful birthday to you!
This reminds me of this prozd sketch from youtube https://youtu.be/_AnDFT2l_h0
I was hoping Harris would win, but Trump winning isn't the end of the world. There are upsides:
- the baseless lawfare against Elon Musk in retaliation for allowing rightists on Twitter will stop instead of intensifying, which could lead to more advances from Musk's companies
- while Trump's somewhat volatile, this can be an asset in international relations; he managed to drive a wedge between Beijing and Pyongyang when no-one else could, he's willing to drive re-armament of the West, and there are potential outcomes of WWIII where I'd be worried Harris might balk at doing the needful, so while I'm slightly more worried about him in the chair than Harris (largely due to his age) it's not drastically worse.
I'm very curious to understand the world model that makes you think that Trump - who consistently said that the US shouldn't help Ukraine or Taiwan, would be more willing to engage in a war with China over Taiwan.
(Also the wedge he alledgedly drove between China and North Korea is completely non-existant. NK is just as protected by China today as it ever was, and Trump's diplomatic openings toward NK didn't change anything to the NK nuclear program.)
(As a non-American who mostly cares about how the US election will impact my own country and our allies, I think that international relations is the main field were Harris was completely superior to Trump in all regards)
>I'm very curious to understand the world model that makes you think that Trump - who consistently said that the US shouldn't help Ukraine or Taiwan, would be more willing to engage in a war with China over Taiwan.
I don't particularly think that*. I do think that if he did engage in such a war, and there was a small nuclear exchange ending with Beijing run out of nukes but refusing to surrender, Trump (or Vance) would be certain to call the bluff and methodically burn Chinese cities until the PRC either surrendered or suffered total state failure; I am less certain that Harris has the sheer bloodymindedness necessary to do that, which both in the event gives them a chance to either build more nukes to kill Westerners, or lick their wounds to *try again* in 10 years, and also back-chains to make the PRC leaders less hesitant to pull the trigger in the first place (though this is counterbalanced by the greater chance of coming in at all).
As I said, I *am* slightly more worried about him in the chair than Harris, and like you was hoping for a Harris win on this alone, but there *are* complexities there.
*I think the disparity in chance of coming in is not all that big, in the end, because Trump wants to look "tough on China" (and "tough" in general) and that ties his hands if the PRC does something outrageous like attack Taiwan. But, you are correct, the disparity is in Biden/Harris' favour because Trump is less predictable.
Americans say the most insanely bloodthirsty things and then wonder why the rest of the world fears them
I hope to live to see this evil empire (and I don't mean China) fall
"I am less certain that Harris has the sheer bloodymindedness necessary to do that"
FWIW the prediction markets disagree with you: https://www.astralcodexten.com/p/mantic-monday-judgment-day
"They could have picked better questions (I’m not sure why “Trump in power beyond 2028” needs to be conditional), but some of these are interesting:
China more likely to invade Taiwan under Trump (25%) than Harris (17%), and Harris is more likely to fight back (75%) than Trump (54%).
Russian prospects in Ukraine better under Trump (75%) than Harris (40%)
Iranian nukes more likely under Trump (49.5%) than Harris (45%)"
> I do think that if he did engage in such a war, and there was a small nuclear exchange ending with Beijing run out of nukes but refusing to surrender, Trump (or Vance) would be certain to call the bluff and methodically burn Chinese cities until the PRC either surrendered or suffered total state failure
China has perhaps 500 nukes, the US has 5k. However, nuclear war is not a pissing contest where the US will comfortably win by these numbers alone.
A few megatons dropped on the Bay Area would likely be devastating to the US economy, and I don't think that you can be confident that US missile defense will avoid that outcome. Perhaps half of their nukes will hit US cities. The fact that they run out of nukes sooner than the US might not be all that critical, if two shooters manage to hit each other a few times, the fact that one of them has a larger clip and can hit his opponent more often does not prevent him from also bleeding to death.
Furthermore, while there are solid game theoretical reasons to pre-commit to retaliating against nuclear strikes on your cities in kind, I think there are limits to this kind of vengeance. If you launch your own ICBMs after detecting the enemy launches, you can still claim that your goal is to save your deterrence capabilities before your ICBMs are destroyed. You can also claim that one of your goals is to prevent the enemy from launching further nuclear attacks.
The morality of such a counterattack might be debatable, but it looks very different if you delay your response and nuke back at leisure, conditional on how the diplomacy is going.
It is the difference between returning fire when being shot at (which would generally be self-defense) and walking over to some incapacitated enemy and shooting them in the head (which is generally considered murder).
Still furthermore, the US has nuke parity with Russia. If it spends half of its nukes glassing China without getting hit too badly, Russia will be the leading nuclear power in a world where nuclear war has been normalized.
Lastly, I am highly doubtful that the mere threat of nuclear annihilation will force anyone to surrender. For Japan, it was more like the last straw as their conventional war did not go very well either. Do not expect the CCP to roll over unless you offer them a favorable deal. Likewise, do not expect total state failure. The CCP rose to power in what was initially a protracted war. You can diminish their state capabilities greatly, but the CCP will likely still rule over what remains of their population. Take Gaza: the IDF has bombing the shit out of Hamas for a year now, and they are still not defeated.
For the limited effects of strategic air power, see also:
https://acoup.blog/2022/10/21/collections-strategic-airpower-101/
Concentrating on stuff that Trump says is an exercise in futility. He is often contradicting himself.
In the previous term, his administration's policy was only tangentially related to his utterances.
As others have said Trump speaks in poetry, not in prose.
Or to put it differently, he should be taken seriously, but not literally.
You can view these as apologia, and make snarky comments. Or you can view them as ways to understand the man and the policies he's likely to implement.
Think of him as Nixon. You don't have to like the man, or his policies. But you're the idiot, not him, if you would rather mock, misquote, and make jokes than attempt to understand what's going on.
And at the end of the day, Nixon actually went to China -- regardless of what the Pauline Kaels of the world thought of him.
I don't understand how anyone thinks Harris was superior to Trump on foreign policy. What has she done? Given charge of border security, she was slow to visit the US-Mexico border, and ended up doing almost nothing about it. https://theweek.com/in-depth/1023057/kamala-harris-vice-presidential-track-record One may consider this immigration policy instead of foreign policy, but I don't see how one may consider immigration policy NOT to impact foreign policy.
The most experience I can find of her foreign relations seems to be having visited 21 countries and met with more than 150 world leaders. I cannot find any reference to anything she has DONE with those leaders. It does show she hasn't really impressed them, as foreign leadership goes through other people to contact the US, such as secretary of state or national security advisor. https://www.cnn.com/2024/09/08/politics/kamala-harris-foreign-diplomacy/index.html
My impression is that she has been hidden away so that she can do as little damage as possible, while showing how modern-thinking the US is by having a woman as a vice-president.
In contrast, Trump seems decisive in foreign affairs, though time will tell how good his policies prove to be. He initiated serious trade issues with China (including effectively a trade war), moved the American embassy from Tel Aviv to Jerusalem. and started (though it ended up falling through) talks with North Korea.
The largest problem I have with Trump's foreign policy is his having pulled out unilaterally from the Iran nuclear deal. I do believe it was a bad deal, but the US had already agreed to it under a previous administration, and this tarnishes any future foreign deals since any future president could undo them.
> Given charge of border security,
She wasn't. She was in charge or improving economy in countries where immigrants come from, to stop them from *needing* to immigrate
Source: https://www.youtube.com/watch?v=HyzGkEV3p2g
The claim isn't based on Harris's track record since she doesn't have much of one one way or the other. She's mostly assumed to be a generic Democrat.
Trump seems a lot softer on China now than he did during his first term, flip flopping on the Tiktok ban and sounding very hesitant to support Taiwan against Chinese aggression.
Meanwhile he's been consistently soft on Russia going back way further than that, including the incident he was originally impeached for where he held up Congressionally mandated aid to Ukraine for his personal political advantage.
And a large chunk of the senior foreign policy leaders of his first administration came out against him, including Secretaries of State and Defense and at least one Chairman of the Joint Chiefs. That alone says pretty bad things about the quality of his leadership.
>while showing how modern-thinking the US is by having a woman as a vice-president
<mildSnark>
Suggestion to the Democratic party for 2028: How about asking Kemi Badencoch if she'd like to run a larger country? Yeah, there is that phrase about "natural-born citizen", but I'm sure that can be worked around... :-)
</mildSnark>
The wise man would say that the stochasticism built into the US electoral system, the fact that any deal might be revoked 4 years later, means you damn well better make a *win-win* deal, and *stick to it*, because legalistic arguments about how you are technically allowed to do x or y because the treaty says so, are not going to work for you...
You do realize that there were many peace treaties signed in the Middle East under Trump, don't you? Also Trump didn't send billions in aid the terrorist nation of Iran like Obama or implicitly support Iran's nuclear program like Biden.
If you don't think Iran is a terrorist state, then no need to discuss any further.
Have you heard with full CONTEXT when Trump said he the US shouldn't help Taiwan? Ukraine I am sure he has said. What he will actually do, well, do you like to gamble :-)
But Taiwan? If so, please send a link.
Trump has consistently said that Ukraine is not worth WW3. Do you disagree?
Beyond that is pretty much projection of what people want to project.
Does this mean he'd give Putin a blank check? Well I guess, if you assume he's a Russian agent, then ...
Does this mean he can see a situation where there's no winning strategy (pull out of Vietnam in 1967, not in 1975)? Possibly.
Does this mean he thinks that giving Ukraine permission to use their fighter jets and cruise missiles, to force Putin to negotiate? Also possibly. I full expect Trump's character is such that he would love to both try negotiating a deal with Putin AND having Ukraine unleash thunder on Putin if the deal is not accepted.
Mainly what this means is exactly what the words say -- that Trump is not going to risk nuclear war on the US for the sake of Ukraine (or Taiwan for that matter, and probably not for any country in Europe given the way they have behaved over the past 20 years or so).
If you think that's a scandal, that the US SHOULD be willing to destroy civilization for the sake of virtue signaling, let me ask why you haven't already left the US, given the death camps and genocides that are supposedly surely coming? If not, then maybe you're a little too invested in playing word games with zero consequences, and your opinion about something that actually matters deserves to be utterly ignored?
"Ukraine is not worth WW3"
I disagree in the sense that giving up Ukraine is a signal that one can in fact annex another country. We give up Ukraine to avoid WW3, and Russia decides Georgia has insulted them and ought to be invaded and annexed. That's OK; they're a lot smaller country than Ukraine. Same thing in Belarus, but they're landlocked, so less important. Then Poland, Moldova, and Romania, because they're a natural progression, like Manifest Destiny. Then Germany, since it's long overdue, after all.
Russia invaded Ukraine, you recall, because they were about to join NATO. The REASON for the war is the reason to defend them, and though WW3 would not be a desirable outcome, if one is not willing to risk it then Russia will have license to do as they want. Mutually assured destruction PREVENTS world wars, unless one side decides it must not enter such a war, which means the other side wins.
ive also learned to use the word lawfare this past week xD
I am enjoying, and yes it is Schadenfreude, the news that the cases against Trump will now have to be dropped since he's President-Elect.
Politico has a typically calm and even-handed article on it:
https://www.politico.com/news/magazine/2024/11/07/trump-legal-failures-blame-column-00187945
I'm especially enjoying the turning on Merrick Garland, the former Martyr of Justice who had been so egregiously cheated of his rightful spot on the Supreme Court.
I enjoy even more that the "but he is a convicted felon, 34 FELONIES" stuff went nowhere.
This was supposed to be damning, but it also came from the set of people who in other instances demanded we should scrap terms like "felon" because they were de-humanising:
https://fortunesociety.org/wordsmatter/
" People with criminal justice histories are referred to in an array of dehumanizing labels, such as “inmates,” “criminals,” “prisoners,” “convicts,” “delinquents,” “felons,” and “offenders.”
...Dehumanizing labels stereotype and marginalize people rather than support them while they rebuild their lives. Individuals with justice system involvement are not defined by their conviction history. The words we use to reference people should reflect their full identities, and acknowledge their capacity to change and grow.
Be mindful of how you speak about The Fortune Society’s mission, the people we serve, our dedicated staff, and board members. We encourage you to use humanizing language — your example will inspire others.
WORDS TO AVOID
Offender, Inmate, Felon, Criminal, Convict, Prisoner, Delinquent
PHRASES TO USE INSTEAD
Person or individual with justice system involvement; Person or individual impacted by the justice system; Person or individual affected by the justice system"
https://www.vera.org/news/words-matter-dont-call-people-felons-convicts-or-inmates
Throughout history and across the world, dehumanizing language has facilitated the systemic, inhumane treatment of groups of people. This is certainly the case for people impacted by the U.S. criminal legal and immigration systems, and that’s why it’s so important to use language that actively asserts humanity. Many people and organizations are moving away from using terms that objectify and make people’s involvement with these systems the defining feature of their identities. But many others—politicians, media outlets, and more—still use harmful and outdated language like “convict,” “inmate,” “felon,” “prisoner,” and “illegal immigrant.”
There are better alternatives — alternatives that center a person’s humanity first and foremost. These include “person who was convicted of a crime,” “person who is incarcerated,” “person convicted of a felony,” and “person seeking lawful status.” These words and phrases matter. Choosing people-first language is a step toward asserting the dignity of those entangled in these dehumanizing systems.
...Calling a person who was convicted of a crime a “criminal,” “felon,” or “offender” defines them only by a past act and does not account for their full humanity or leave space for growth. These words also promote dangerous stereotypes and stoke fear, which stigmatize people who have been convicted of crimes and make it harder for them to thrive.
“Until this carceral state and the people of this country begin to understand the power of the words that seek to dehumanize the incarcerated or justice-impacted people, there will never be a real and substantive conversation about criminal justice reform,” said Wright. “Our humanity is maintained and respected by not referring to us in those impersonal and definitive terms, but by acknowledging our intrinsic value as human and not by defining us by the worst day or act in our lives.”
So if someone vents about how Trump is a felon, kindly remind them gently that such terms are discriminatory and dehumanising and they should refer to him instead as "a person with justice system involvement" 😁
If they say "how can a convicted criminal be worthy of, or fit for, the presidency?", remember:
"Individuals with justice system involvement are not defined by their conviction history. The words we use to reference people should reflect their full identities, and acknowledge their capacity to change and grow." Don't uphold the carceral state or define people by the worst day or act in their lives!
Great stuff on The Fortune Society. Much appreciated!
You must have missed the memo.
None of those restrictions on language apply to straight white males. Trump is obviously that. So you can call him a felon, fascist, mis-gender him, dead-name him, anything you want.
And I'm not being a sarcastic asshole.
This is the actual stated truth. The less like a straight white male you are, the more you can claim people be careful their language around you.
This makes me nothing but sad. The most dangerous thing about Trump, to my mind, is that he dehumanizes people every chance he gets. Dehumanization is the first step toward quite a few historical evils.
How do they feel about "bloody-handed murderer"? Specificity is useful!
I don't like these arguments from hypocrisy, because the person who's making them is being equally hypocritical.
IE I interpret you as saying "The Democrats believe Trump should be stigmatized for being a felon, but they don't believe random robbers should be stigmatized as felons, therefore they're hypocritical".
But you're also implicitly saying "I believe random robbers should be stigmatized as felons, but Trump shouldn't be stigmatized as a felon", which is also hypocritical.
I’m very worried about an increase in lawfare now that the people who endorse it are getting into power! (Especially the guy who turned my porn site x.com into a set of political ads.)
I thought that was about a game where Martians attack and invade Earth, and we defend against them.
Don't know what you're talking about, there's plenty of good porn there still... Though, people are getting more serious about moving to Bluesky now that Musk has been, uh, more emboldened.
My main delight is that it's a blow against a) legacy media and b) cancel culture.
If you pay attention to events closely, you'll see that most news outlets (e.g., top sites on Google when you search a news story), are grossly distorted on any vaguely culture war topic, and distorted in favour of the radical left. To me, the clearest insurance was the coverage of Depp v. Heard. It was really disgusting to see people defend an abuser (Heard) like that, having actually watched 100% of the court footage.
Similarly I've had horrible experiences of cancel culture.
You saw it in how the Democrats ran the election. One of their core strategies was slandering their opponents and calling them bigots, sexists, extremists, fascists, racists, etc... E.g., they characterize JD Vance as an extremist despite him almost not getting the VP nomination due to being too democrat leaning (see the leaked JD Vance dossier). I'm sick of the hypocritical gaslighting, so to me, the election result feels like a breath of fresh air.
What is sad is that so many have been sold on the propaganda that Trump is Hitler. I've had democrats friends talking about how "they want to kill the guy." Fyi, voicing death wishes about a president is a federal crime.
It is insane and sad. They spout how they want to literally kill their political opponent, while complaining that the republicans are the divisive party?!
Fyi, voicing death wishes about a president isn't a federal crime (https://en.wikipedia.org/wiki/First_Amendment_to_the_United_States_Constitution#Freedom_of_speech_and_of_the_press, https://en.wikipedia.org/wiki/Threatening_the_president_of_the_United_States#Watts_v._United_States, etc.).
Agreed. It is some sort of mass delusion. That word, I don't think it means what you think it means.
The only silver lining is this behavior is giving me greater insight into human beings and our history.
This was one of several sufficient reasons for me to vote for Trump.
I cannot image Scott Alexander, who I used to have tremendous respect for and still respect, somehow thinks there is a balance of power in this country.
Outside of elections this country is run by Democrats and their allies. I guess Scott can't see how incredibly stilted political speech in this country when most (though not all) of the allowed speech is what he agrees with.
The closeness of North Korea and China is vastly exaggerated. People act as if North Korea is a Chinese proxy. Whereas NK had a far closer relationship with the Soviet Union.
No, I don't think North Korea is a Chinese proxy. It *is* a Chinese ally, and the PRC has covered for them time and again since 1991. The *only* time the PRC backed off a kerfuffle, TTBOMK, is when Trump played the madman card.
It's one thing to be in mourning because your preferred candidate lost; it's insufferable to expect everyone else to join you.
Hey, you score 10 out of 10 for ineffectiveness and vacuity! Your comment is absolutely guaranteed to lead to no change at all in OP's view of the situation, and to increase their difficulty understanding the point of view of people who are less distressed than they are or in fact not at all. It is completely free of content, just a big fart of meanspiritedness.
On the contrary I think I provided some valuable feedback to him about what contempt some others hold his views.
Fuck this dude and people like him trying to police everyone else's lives and feelings.
How can your feedback possibly be valuable? Do you think OP hasn't noticed that more than half the country just voted against his views? Or that he discounts all those millions of your-views-suck votes because he's been waiting to get a read from Martin Blank, mean teen extraordinaire?
Deiseach did a good job of providing the same feedback, but with decent manners, mention of several issues that are now likely to be decided in a way she thinks is fair, and a bit of rueful acknowledgment that she couldn't help enjoying watching the other side absorbing just how badly it had lost.
Setting aside what counts as an intelligent and civilized response, let's think about effectiveness in convincing OP he might actually be wrong about some things. Deiseach may conceivably have given him a glimpse of the interior of an intelligent person who is not amoral but has arrived at conclusions very different from his. You, on the other hand, have probably strenthened OP's view that the opposing side is made up of meanspirited tinyminds who can't even argue an issue well. Fuck you too, buddy.
Banned for one week. The main reason it's only one week is that it's election week and everybody is crazy now and I assume they'll calm down soon enough.
Have you ever posted on wireheading? (I am in the middle of Ringworld Engineers) Is it feasible? Is it any better than the best available drug? What do you think of it philosophically?
Nobody REALLY knows how good wireheading is, because it's never been really developed. OTOH, there were rats that though it was better than sex or eating, even when they were starving.
On the *other* other hand, when I'm waked up from a sound sleep, and want to go back to sleep, offers of sex or food would not tempt me at bit. And yet sleeping is not an ecstatic experience. It's mostly a non-experience. But when you are tired and get in bed lying there feels great, and sleeping looks, in your anticipation, like a wonderful thing. It's actually kind of a ripoff. You think sleeping is going to be 8 hours of wonderfulness. In reality, giving in to the craving to fall asleep is wonderful, but only lasts 5 mins. or so. So I'm not sure we can know that the wireheaded rats experienced hours of something that is far pleasanter than sex or eating. Maybe it was sort of a ripoff for them -- a powerful illusion, at the moment of choosing wireheading, that they were about to experience something wonderful.
What a crypto-buddhist comment.
Heh. Sometimes I think even orgasms are sort of a rip-off, because they're so brief. You try to hang on to the physical ecstasy but it shrinks down to nothing in a few seconds.
It's been tried a couple times in humans, summarized here https://en.wikipedia.org/wiki/Brain_stimulation_reward#History
Yes, but "tried a couple of times" is not similar to "really developed". It could probably be optimized in several different ways.
Not in any really specific way, but you can see some thoughts that I mostly agree with on https://www.astralcodexten.com/p/book-review-deep-utopia .
Short version: I would be sad if everybody was turned into blissed-out lumps of jelly, but it seems like there are probably ways to do it that are better than that.
The Deep Utopia post is good, and you could write more on that and probably will. I'm a Christian, and it ties in nicely with the hard question of "What does God want us to do, besides not breaking moral law and besides giving Him due credit via prayer and worship?" Aquinas said "To be happy," and the Westminster Confession says "To glorify God," but neither of those is an answer till it's fleshed out (which Aquinas does a bit). Whether one is religious or not, it's clear humans are built in such a way as to help figure out an answer, and that the answer is going to be different vocations for different people. Actually, one good subquestion is what the best life is if you're a quadraplegic of normal IQ and we can't use technology to hook up your nerves and give you back your arms and legs or add to your IQ. That's a good question because it's simpler than for most people.
My understanding of climate change (stay with me) is that the more authoritative estimates of costs per degree of warming are about 1% of GDP
Meanwhile it seems that this is being revised upwards according to catastrophe bias. One new study cited news headlines and public pronouncements of prominent climate scientists as the source of their priors, which did not match the actual costs literature. Their study's lowest cost estimate was 6x higher than the previous highest one, and I get the impression that no insiders really challenged them and many said approving things, while their study generated a powerful round of news headlines
Here we have an information ecosystem that appears to be giving a popular estimate of costs 10x higher than the literature, and the popular perception is now altering the literature
The implications of this across the century in terms of miscalibrated policy costs must be in the range of $50 trillion dollars. If prediction and forecast markets can gain the prestige to replace popular headlines as information touchstones then this is an event super worth celebrating
I don’t think it’s as clear-cut as that?
For such a computation, you need
1) an extremely precise definition of what exactly you’re assessing (which has no reason to stay the same, and the devil is in the details),
2) a model of the overall economy,
3) another precise model about what kind of climate impacts to expect.
And more importantly, there’s the fact that the result is an abstract figure with a mostly academic interest – making 1) close to what we actually care about is a difficult task.
All in all, I think these points can easily explain a 10x difference.
And if there’s a prediction market for the costs of climate change, I expect these points to heavily affect the price in as well.
I agree it's a difficult task to satisfactorily set up a prediction market on what the consensus academic position on global costs per degree of warming is and what it will be in x years
However, compared with the existing information ecosystem I consider it to be an easy task to radically improve on it. For instance in "Sea-level rise and its possible impacts given a ‘beyond 4°C world’ in the twenty-first century" by Nicholls et al, there is a reference figure of the exposed population in the high case, if all flood control infrastructure construction is halted until 2100
This figure is 187 million, and is how the news headlines describe the study results. However, in the low case with maximum adaptation costing I believe .02% of global GDP, the population displaced in 2100 is 5000-10,000, and this more closely resembles the median estimate than the headlines. Prediction markets have the potential to publicly elicit this kind of information to a much greater degree vs the existing system of incentives
>My understanding of climate change (stay with me) is that the more authoritative estimates of costs per degree of warming are about 1% of GDP
I hope you are abridging here, because that relation can be linear for only a very short range. Or are you saying a temperature rise of 50°C would incur a cost of 50% of GDP?
Yeah, the IAM model estimates are fairly linear out to about 4 degrees and look quite troubling by about 10 degrees. Luckily the IPCC projections for high emissions temperature levels tend to level out at about *6-7 degrees
This does assume that decarbonization is sure to happen by 2100 so you could argue that there could be a worst case emissions scenario not covered. The high emissions scenarios do include global coal increasing by 6x at that point though
edit: 6-7 degrees rather than 4 degrees with the 6x coal scenario
How much it costs is going to depend on where you live, and for some folks the costs are going to be intolerable. Your argument seems to assume that the costs are evenly distributed over the world population in some "fair" manner. This clearly is not going to happen. It's already killing people now and then. (Death spikes during heat waves, unexpectedly strong storms killing people, etc.)
Definitely one weakness with global cost estimates is that it doesn't account for marginal costs. This is something politicians are accounting for and the IPCC constantly mentions
There is a problem with that. They don't account for the marginal cost of climate policy. Just as Sub-Saharan Africa will have high costs from warming they will, by my lights, have higher costs than that from being blocked from developing with coal and gas
My general take is that we live in a world where the large majority of weather related deaths are from temperature, and a large multiple of those are from cold temperatures, including in Africa. Economic costs are not being balanced correctly against avoiding warming costs
An error: " If you previously thought (like I did) that there was a 90% chance that Polymarket was more accurate, you should update down to 88%."
You mean Metaculus.
Thanks, fixed.
Hmmm, this seems a bit like criticizing people (or systems) when they turn out to be right here... If the election had been close but happened to land on Trump then I would agree with you, but this election was sufficiently lopsided that I think giving a call of pure 50% is very slightly embarrassing, and I'm very curious to hear from pollsters where they think things went wrong.
Theo's 'Neighbor' theory is just one, and your right that with only one data point it's a bit ridiculous to say that he was definitely right, but I also think it's very important to note that Theo didn't just armchair reason his way to success. He tested his theory by commissioning a private poll with his own methodology, and it probably cost ~$100k, that's a huge structural advantage real money prediction markets have over reputation based ones. They offer profit incentives for spending money to gather *new* evidence, instead of just optimally summarizing existing evidence.
I agree that more people should criticize others who turn out to be right, and that this is an under-provided social good. See https://www.astralcodexten.com/p/against-learning-from-dramatic-events (I edited this into the post a second ago to hopefully emphasize this point).
Great analysis. What i am left wondering at the end of it: you are correct that prediction markets predicting the correct result does not entail that their model is right. is there any kind of empirical testing or a certain way prediction markets could perform that would cause you to conclude they are efficient? Or is this just a question that is immune to empirical testing bc we’ll never know the actual probability curve for the election?
It's philosophically debatable (like, *really* debatable) whether there even is a probability curve for the election.
Before the election, Silver emphasized that 50% was a mean outcome, but not a modal outcome. Given that polling errors would likely be correlated, he noted that a landslide win would be fairly likely, with an over 40% chance that a single candidate would take all 7 swing states. See e.g. https://x.com/NateSilver538/status/1848060370268938538.
He also gave it around a 25% chance that Trump would win the popular vote. And it looks like he's doing to win it by a decent margin.
Not knocking Nate Silver at all here, but being able to predict that if there is a large polling error it would be correlated doesn't mean it's not unfortunate that there was a large polling error.
Using the posts framing, I would say that Nates model would put this result at a little less than 20% probably, or somewhere between 2 and 3 coin flips coming up heads (between 2-3 bits of evidence). Not a ton of evidence, but enough to shift my credence towards polymarket, especially since my prior against it was way less than 90% (though I did think before the election it was off)
Yeah, I think Nate’s model was right based on public information, but my gut tells me that based on this result and the herding that a lot of pollsters never published pro-Trump polls or were systematically overestimating Harris’s likely voters, making assumptions that her turnout would be similar to Biden’s in 2020. Without better polling firms, we will continue to see surprising election results.
I think this is right, but one thing that makes it hard to systematically resolve this is that the social landscape is always shifting, and pollsters are always trying to resolve the problems of earlier polls. There's always polling error election after election, but the direction isn't consistent, otherwise pollsters would correct for it. If there were methodological information which allowed private pollsters to construct more accurate polls, the public pollsters would have access to it too.
Two items:
ONE
Desertopa said "There's always polling error election after election, but the direction isn't consistent, otherwise pollsters would correct for it."
You may be being sincere here but if you look at the RealClearPolitics average going back to 2004, the error is almost always, 5 out 6 times, in favor of Democrats:
2004 Bush R +1.5 +2.4 => 1.1
2008 Obama D +7.6 +7.3 => 0.3
2012 Obama D +0.7 +3.9 => -3.2
2016 Clinton D +3.2 +2.1 => 1.1
2020 Biden D +7.2 +4.5 => 2.8
2024 Harris D +0.1 -3.3 => 3.4
Yes, small sample size applies, but remember these are averages of polls for each year (at least 9 polls in the average each year) AND these polls were informed by many other polls a week or two before. (It is surprising how organizations polls in early to mid October, but NOT the week before the election.)
So the idea the polls will self-correct sound a lot like the argument that "of course the legacy media is NOT biased, if they were people wouldn't trust them and would not listen to them."
Well people don't trust them (see the latest Gallup polls on this this) and they have a lot less less power and influence than they did in 1980 but they continue to be highly biased.
TWO
Desertopa said "If there were methodological information which allowed private pollsters to construct more accurate polls, the public pollsters would have access to it too."
It's not access to special methodology that allows private polls to be more accurate. I believe they are more accurate for two reasons:
1) Money - polls are expensive. A candidate or party is willing to spend the enormous amount of money to get the best possible information
2) Focus - public polls are usually meant to drive readers / viewers to a news organization and so want as many crosstabs as possible and just need to get to close to the results. Private polls are going to be more focused so they can get more data points and spend more time of getting exactly the right information.
If we include midterm elections though, we don't see this sort of systematic skew. Are polling averages systematically biased towards Democrats in presidental elections, but not in midterm elections? Maybe, but it's not at all clear that we should expect that to be the case.
As far as the accuracy of private versus public polls, in statistics, getting an adequate sample size for high precision is much, much easier than getting a sufficiently representative sample for high accuracy. Public polls aren't inhibited in their ability to get enough data points to be sufficiently accurate, the real issue is whether their sample sets actually reflect the overall voting population, which is a matter of methodology.
Maybe money and focus can produce accurate polls, but candidates and parties do not produce accurate polls. Their internal polls are much worse than public polls. You are trying to explain something false.
Yes, I think the "coin" analogy is extremely misleading. We didn't get a single binary output - we got the fact that the country overwhelmingly preferred Trump, and if that doesn't result in _some_ kind of signal that should be measurable by prediction markets, then honestly, what use are they?
If the markets predict a 50/50 chance that Kasparov will win a chess game against Carlsen, then the game starts and it turns out that Kasparov's been dead for 3 years and "forfeits" ... we wouldn't say "shucks, guess the markets lost this coin flip". Forecasters should start to ask questions about how they can maybe notice the next time a player is dead and factor that into their predictions.
"the country overwhelmingly preferred Trump"
Last figures I saw showed him with a little under 51% of the vote and Harris with about 47.5%.
He won a bunch of swing states ... by something like 51%-49% in each case.
In what world is any of that an overwhelming preference?
He did better than Harris, for sure. The electoral-college system, which can turn a bunch of near-50:50 results into a landslide either way, will deliver him a nice big count of EVs, for sure. But what about this suggests that the US electorate had so strong a preference for Trump that pollsters and pundits who said "it looks really close" are as wrong as a chess prognosticator who didn't notice that one player is dead?
Hmm, I think you and the other replies here are making a basic logic mistake. Prediction markets are not trying to estimate the % of the country that prefers a candidate. They are trying to estimate the % chance that candidate will _win the election_. The two values both have a % in them, but please don't get them confused.
So, I admit that I made the same mistake when I said America "overwhelmingly preferred Trump", which is more a statement about the former than the latter. What I should have said is that "Trump was overwhelmingly likely to win this election". Knowing what we know now, that is a correct statement. After all, Trump didn't win "a bunch" of swing states - he won ALL of them, because his 51-47 "coin" was "flipped" millions of times for each one. The election didn't hinge on which way the wind was blowing on Tuesday. Carlsen's victory against his dead opponent didn't hinge on whether he got enough sleep the night before.
Since this was a historically strong and certain election result, you have two possibilities:
- Prediction markets should been able to predict the result ahead of time with at least a little confidence.
- Prediction markets can't distinguish a country's 51-47 preference, and are thus useless for predicting the outcome of US elections.
I think we'd prefer the second wasn't true, but at the moment it's not looking good.
I have no idea what gives you the impression that I don't understand the difference between "x% votes for X" and "x% chance that X wins".
The claim I was commenting on was your statement that "the country overwhelmingly preferred Trump". He only got ~51% of the popular vote! If he'd had a couple of percent less he wouldn't have won the Electoral College either! That isn't what an overwhelming preference looks like.
Anyway, you've now retracted that, so fair enough. But I still don't buy that "Trump was overwhelmingly likely to win this election". I mean, I guess it depends on what that means. He _did_ win the election (unless there was some huge fraud that no one has caught a glimpse of, I guess, but I don't see any reason to believe that and mention it only because worrying about election fraud on the basis of no evidence is kinda Trump's brand), and probably for almost all elections it's true that if you could rewind the world by 24 hours and run the election again you'd almost always get the same result. But I don't see that that's a useful sense of "overwhelmingly likely to win".
The relevant question is: _given the available information_, was it overwhelmingly likely that Trump would win? I don't see why we should believe that. It was, in fact, pretty close. All the polling indicated that it would likely be very close. Polls have systematic errors and they're really hard to predict. If you think it was "a historically strong and certain election result" in some sense that implies "someone competent would have called it confidently in advance", I think you're under the influence of hindsight bias or something of the kind.
I _do_ think that prediction markets can't reliably predict a 3.5-point electoral preference, at least not if what you want them to do is to say "90% that X wins" in 90% of cases where X wins by 3.5 points or something like that. So far as I know, nothing can do that. I don't know of any reason to think anything should be able to do it.
(Whereas chess prognosticators _are_ able to predict that Magnus Carlsen will win a chess game against a corpse. It's a silly analogy.)
"Hmm, I think you and the other replies here are making a basic logic mistake. Prediction markets are not trying to estimate the % of the country that prefers a candidate. They are trying to estimate the % chance that candidate will _win the election_."
Then why are you talking about how the American people "overwhelmingly" prefer Trump?
The alleged overwhelmingness of Trump-preference, has little to do with the expected probability that a candidate will win an election; it's the *ratio* of actual Trump-preference to error margin in our measure of Trump-preference that matters. If 50.01% of the population of each swing state is committed to voting for Trump, and 49.99% for Harris, and the Orbital Brain Scanners allow us to predict this with +/-0.001% accuracy, then we will be justifiably very confident that Trump will win, If the actual preference is 60/40 in Trump's favor, but people so thoroughly and bafflingly lie to pollsters that the polls all have +/-20% error bars, then the smart money will be "meh, leaning Trump but still pretty much a toss-up".
And Trump winning with 60% of the vote, in that hypothetical, doesn't make the smart money wrong. Trump winning with 51% of the vote, *definitely* doesn't.
Erm, yes, did you read the end of the comment you replied to? It is entirely conceivable that our inconveniently complex world is such that prediction markets can't distinguish voter preferences well enough to say anything useful.
In this world, prediction markets for election results might as well not exist. To repeat, this election was a very strong result (only really beaten by Obama's and Clinton's landslide victories, which, as I recall, didn't require superforecasters to see coming). If you claim that it's STILL not strong enough for prediction markets (what you call the "smart money") to move away from 50-50, then I ask again: what use are they?
These are closely correlate because state-level outcomes are not independent but highly correlated.
>Last figures I saw showed him with a little under 51% of the vote and Harris with about 47.5%.
>He won a bunch of swing states ... by something like 51%-49% in each case.
>In what world is any of that an overwhelming preference?
Agreed! What I wish I had would be a two-dimensional plot of voters' views on something like:
Would like Harris to be POTUS: -10 ... 10
Would like Trump to be POTUS: -10 ... 10
showing the density of the distribution throughout this space
( Personally, I'd put Harris at -10 and Trump at -9, marginally lesser evil. )
The most important thing for anyone to remember is that, whatever one's view, there are a _lot_ of people who disagree, and (for Americans), these are your countrymen. You need to live with them.
Just for reference, in 2020 Biden got 51.3% of the popular vote compared to Trump’s 46.8% and no one was saying the country overwhelmingly preferred Biden. That would’ve been silly. 51% isn’t an overwhelming anything. Biden won by 7 million votes. So far, Trump is leading by 4.6 million votes. Biden also won more electoral votes in ‘20 than Trump will have won in ‘24.
EDIT: This last claim I made, about Biden receiving more electoral votes in ‘20 than Trump would in ‘24, now appears to be false. It now appears Trump will secure 312 electoral votes to Biden’s 306.
Some people were in fact saying that.
They were wrong too of course.
Yes. I’m wrong. You’re right. Some people probably said Biden’s narrow victory indicated that “the country overwhelmingly preferred” Biden, though I can’t find instances of that wording. Maybe you can. I just don’t think whoever you found who made that same claim could reasonably substantiate it. And given that Trump won by an even narrower margin, I don’t think the claim can be substantiated in his case either.
What do you mean? We got almost exactly what Nate Silver predicted. Off from the polls by about 2 points in every state. That seems to vindicate Nate Silver more than the people who said it would be an easy win for one side.
I made 10k off the election. It was really obvious that Trump would win, this was literally free money
I thought it was going to be Harris, my basic reasoning was: all the polls are herding, this is probably because they're terrified of another "Hillary 2016: 90% chance" moment, and so they're disproportionately throwing out Democrat leaning outliers.
I noticed that the vibes were good on the republican side but I figured the argument there was: polls are about equal, but they always underestimate Trump so he's actually winning in a landslide. I figured the polling companies were now taking that into account (taking it into account too much, even, as above).
Can I ask what kind of things you saw which made it obvious to you?
That was my guess too, especially as the non-herding polls leaned Harris.
It's the early vote. Compare a state like Florida or Texas on targetsmart to 2020. They had equal suburban and rural votes but were missing MILLIONS of urban votes (I'm not exaggerating) compared to 2020. It's when the early vote started coming in mid october betting sites across the board started to shift. This quixotic lone bettor theory scott posits is a tad dubious and I think there is a simpler explanation. Nevada is a prime example of early voter data (which makes up 80% of total vote) being borne out directly in the final results as most basic analysis would have lead one to conclude. (Ralston the main figure on early vote data from Nevada is super partisan and jumped through hoops to explain how kamala could "still" win. But even he was pessimistic in his framing. He also analyzed without the broader context of early voter data from other states which brought everything more into frame).
Yes, the early vote was the strongest signal. But there were other strong signals as well. Will write a substack article about this soon
Yeah, it's those missing urban votes I'd like to see explained. Why didn't they turn out? I am being slightly mean here, but she proposed legal weed and free money so black men could legally deal drugs without risk of prison, why didn't that work?
Did she just not click with black/Asian/Latino-Hispanic voters? Why not? Is her weakness her background, that her father was not US black but Jamaican black, and her mother Indian, so she's not 'authentically' black by the US context and not Indian 'enough' for those voters? Obama had the same problem, but he was married to Michelle, née Robinson, whose father was a precinct captain in Chicago and so those contacts aided his career. Harris is married to a white Jewish lawyer in San Francisco and doesn't have that family network behind her.
I believe it was mainly covid and mail ballots( and only secondly the hispanic vote shift). people locked down at home easy to reach by political ads, have less to do, not out having fun. take NJ for example. Everyone was mailed a mail ballot and they won huge. This time they weren't and they had to request it and it barely stayed blue. they have a huge swath of low propensity urban voters and were able to juice the turnout in this demo far beyond normal levels due to covid and using the emergency provisions provided by covid to make last minute changes to local election law and procedure. They will say it was only to counteract the reduced turnout forecasted on election day from covid concerns, but In reality I don't think anyone really doubts in many cases they used their expanded powers to get the most bang for the buck possible. A lot of these measure were rolled back in 2024 such as in NJ. Trump mentioned this exact line of reasoning on Rogan btw as an example of how they "cheated".
Are you saying that if we made voting easier and more convenient the DEM party would keep their advantage? How's making voting easy and convenient not a desirable property of a democratic society?
Ideally you want informed people to vote and people who essentially vote randomly or based on irrelevant information to not vote. If identity and party politics weren't relevant to voting, I think everyone would agree with this. Since identity and party are relevant, and ease of voting typically skews towards Democrats, the calculus is more complicated.
some ppl say a basic filter would be helpful like having to wait in line to vote on eday, or having id , or passing a basic civics test , or filing tax retunrs, or anything to show a minumum level of investmnet to the society you going to determine and influence with your vote. i think everyone is happy with the opportunity to vote being universal and free, but the level of ease is another question. it becomes political quickly because of how these measures would benefit one side or the other. the but as far as the theory of a functioning healthy democracy i think its an open question .
I think it might not be mail-in ballots. I *think* California sends everyone a mail ballot (I got one this year without requesting it, though I'm not sure that's not just because I requested it last time). But California went from +29% Dem last time to only +18% Dem this time. Obviously it's still really blue, but an 11 point shift is huge.
the famous scott replied to me :0
just so u know the vote is only half in in california and they count the mail ballot last which are more blue so the shift isnt going to be that big in the end but still a shift. dont we explain that by mexicans and asians shifting? those were very real changes that happened this election in the electorate, but volume of vote was still also down in the most densly populated areas for other reasons.
Well, an 11 pt shift is only 5.5 pct of voters changing their vote. So not necessarily that huge.
Note also that the Senate election shows a much less dramatic shift. The last several elections in CA since 2012 had the R candidate get 37.5, 43.8, 45.8, 39, and 42 this year. (I skipped 2016, when two Democrats made the final).
I think that Trump's magic as a politician is just making the entire news cycle so damn toxic that most people just tune out. From exit polls, racism was 4% of people's main concern, while the economy was 40%. Trump distracting everyone with culture wars crap, and media taking the bait every time, made most normal people tune out of politics. And since Dem voters skew younger and less conscientious, lower turnout hurts Dems a lot more. Plus Dems were the incumbent after COVID, so they were already perceived to be worse on the economy.
Sorry, but I think the reason people didn't vote for Kamala seems to be pretty simple, though only in retrospect. The voters were presented with two worldviews:
Trump is a dangerous, convicted felon who will amass power and never leave office, just as he did in 2020.
VS
Trump will solve the economic problems of this country and actually manage illegal immigration, just as he did in his prior term in office.
More voters merely believed the second instead of the first.
She attended HBU and was a member of a *very* popular and social - across generations and geography - sorority there. That sorority was very much over the moon about her becoming VP, and many of them wore pink and green to vote in 2020.
How much influence black women have over the vote of black men, I do not know.
Well, as someone who "didn't vote for Harris, just against Trump", I think she wasn't at all inspiring as a candidate. And political promises are easy to make, but which one's will be carried out?
I also think there's an inherent bias against women as leaders in the US, but Harris wouldn't have inspired me no matter what race or gender she had. And distasteful as I find it, many are inspired by Trump.
Voters don't seem to penalize candidates for being female and may reward them for it: https://www.cremieux.xyz/p/voters-dont-penalize-candidates-for.
Well, he does seem to claim that, but the studies he claims to rely upon is behind a sealed door. I won't subscribe to see if there's a pig in that poke.
I also thought it was obvious Trump was winning. But nothing directly to do with polls. It was instead over the non-endorsements of billionaires with newspapers. I was just trusting that they knew more about the odds.
An endorsement in a newspaper, is costly reputation wise, and should be used when it might get gains for the newspaper in the administration.
Endorsements are also more valuable when you do them less often.
The editorial boards of those two newspapers are almost never going to endorse a Republican candidate.
So it makes sense to hold off on endorsements for losing democratic candidates.
Interesting. I saw that as basic game theory. If they didn't endorse and Kamala won, they knew they wouldn't face retribution. But if they endorsed Kamal and Trump won, they would.
Trump already punished Bezos in 2018 by killing a DOD contract. He tried to do the same thing to CNN in 2017 by blocking the AT&T Time Warner deal.
And that's why Trump is especially toxic to a high functioning society.
Exactly. Authoritarianism doesn’t happen all at once. It creeps in by degrading institutions, and one of the first things to go is independence of oligarchs because it’s bad for business.
If Harris had won then I expect there would be retribution against billionaires who endorsed Trump. (In fact we've already seen signs of this for Elon.)
I think we likely would have seen retribution against Elon, in the sense that Elon has been using his ownership of a massive social media platform to influence the electoral landscape, including in ways that are dubiously legal.
https://www.youtube.com/watch?v=waPngGP7Awk
This is a case where tightening (or just enforcing) restrictions on billionaires influencing the electoral process would be serving the will of her constituents. We can't examine the counterfactual where Harris won to see how she would have treated various billionaires who supported Trump, but I suspect that most who simply offered him their endorsement would not have been treated in the same category as Elon.
Now lets consult the other 49% of universes...
My argument will revolve around why we lived in an 80%/20% universe
I bet on Trump as an emotional hedge. I hoped Kamala would win, but at least I got a "sorry" payout
Another usecase for prediction markets!
Regardless, as GreatSwordsmith points out, something's plausibly wrong when Silver et. al. say that Trump is only a slight favorite and then sweeps every swing state and the popular vote. Oh yeah, they also underestimated him in 2020 and 2016! Perhaps while everyone was voting for the Democrat, their "neighbor" really was voting Trump! Further, didn't other prediction markets follow Polymarket too?
The other prediction markets were moved by people trying to arbitrage Polymarket. PredictIt, which was harder to do the arb on, moved less.
I know embarrassingly little about finance and betting so I'm having to look into that more to see what you mean. What's your theory for them underestimating Trump three times in a row? I'm not a Silver doubter - the data shows he is very good. Perhaps there's just a Lizardman's Constant level of people who are paranoid about their anonymous Trump poll answer becoming public, getting mocked, and then getting pelted with coconuts?
A coin coming up heads three times in a row just isn't very unlikely. If Silver (and the pollsters) is honestly trying to predict the election as best as he can, it's not at all unlikely that the error might point the same way three times.
It's not (only) the final result prediction that's underestimated Trump, it's the individual state polls -- which were outside of their error estimates a lot more than three times.
Interesting, do you have a source for that exact claim?
https://www.cnn.com/election/2024/results/president?election-data-id=2024-PG&election-painting-mode=projection-with-lead&filter-key-races=false&filter-flipped=false&filter-remaining=false
https://www.natesilver.net/p/nate-silver-2024-president-election-polls-model
Which states were out of their margin of error and how often does that typically occur in presidential elections?
This is exactly what you would expect if "error estimate" is being defined as just the expected variance from random sampling given sample size (which we can unambiguously calculate), but not any systematic bias (which we mostly can't).
OK, people like Nate Silver can kind of barely sort of quantify systematic bias by comparing polling errors across multiple polling houses, states, and elections, but not with great precision. And the pollsters themselves, I be;ieve just quote the random-sampling error estimates, which will underpredict the real error if there's any bias at all.
Yeah, that's no lie *exactly* what I was thinking when I got your reply notification, but it's not crazy to think that there's a pattern at this point at the margins either. Yes the sample size for a Trump effect isn't large, so it's difficult to judge.
When the market with money on the line goes Trump much harder than the ones that don't, that further makes me wonder. Yes Scott could be right that Theo was just replacement level smart and rich and lucky, but Trump being such a unique politician makes the idea of less random coin flip noise plausible as well. I used to dismiss the shy Trump voter thing until this happened.
Well, hindsight is always correct, but Trump did have the advantage over Biden, then Biden dropped out and the replacement by Harris gave what we can now call a temporary advantage to the Democrats.
The mistake was assuming this would last, that abortion was such a big deal every single woman in America would vote for Harris (women can be pro-life, too, you know), and that the mantra of "Literal Hitler! Fascist dictatorship!" hadn't worn itself out over eight years of repetition and would work to scare people to the voting booths.
In reality, it looks like Trump maintained his advantage and Harris just wasn't present enough to catch people's interest. I know it's been said that "She does too have policies, look at her website" but those weren't put up until very late in the day, and mostly the campaign seems to have been on vibes - vote for First Female Ever! First Asian Ever! First Black Female Ever!
Trump is Trump. Like him or loathe him, everyone has a pretty good notion of what he's like and what his policies may or may not be. Harris was "fun, joy, brat summer, coconuts!" but when you looked deeper, there was no 'there' there.
The stunts with McDonalds and the garbage truck may have been stupid campaign stunts, but they were stupid fun. Harris did the "oh hey Tim I love Doritos, toss me a bag" appearance at a gas station, which was also a stupid campaign stunt, but it wasn't fun. She did a "drinking beer with Gretchen Whitmer" appearance also, I think? But also not fun.
And the garbage truck was an immediate response to what Biden said. Harris' campaign didn't seem to have that level of being able to respond quickly and seize the moment.
Shoulda gone on Joe Rogan, Kamala! Trump did and Vance did, and now Trump is back in the Oval Office and Vance is taking over your old office.
> The mistake was assuming this would last, that abortion was such a big deal every single woman in America would vote for Harris women can be pro-life, too, you know)
I don't think it's that, exactly. Instead, I think many voters didn't let pro-choice views dictate their presidential vote.
Florida is a good example here. The state had an abortion-rights referendum on the ballot, which received 57/43 for the pro-choice side (the proposition failed to get the 60% required to pass, though). The state's presidential vote, however, we 56/43 for Trump. That means that at least 13% of Florida's voters cast a Trump/pro-choice ticket.
If I generalize that lesson, it seems to be that a large fraction of voters don't care about what traditional leading media figures think they should care about, even if their position on "issues" in isolation would be conventional.
I am thinking of a female relative by marriage, who is quite dear to me. She is the perfect Dem voter (though curiously, she had a job at a federal agency, and was only happy there and happy with her boss during the 8 GWB years).
She views politics as an unfolding tragedy. Kamala and Doritos could have been plenty fun and that would not have moved her at all. Sure, she didn't need to be amused; her vote was secure. But I think the audience most receptive to Dem messaging tends to be pretty lugubrious about politics, and is not looking for levity. Certainly the young people inject a lot of drama into all their internet comments. They always sing from the same sad song sheet, and their idea of entertainment is making SNL *less* funny for political reasons.
The Simpsons continue to be brilliant with their flag and abortion division of the electorate.
Re: arbitrage: real-money prediction markets can't stably have very-different numbers for the same question. If they do, it becomes profitable to e.g. buy "Trump" at 50c on one exchange and buy "Harris" at 40c on another, getting a dollar for 90c risk-free (since one of the two will always pay out) and forcing the numbers closer together again.
Play-money markets don't always work this way, because the play-money is not fungible and debatably desirable.
I saw PredictIt was the only site on ElectionBettingOdds not favoring Trump on Monday night, so I looked at that (since I'd put in $20 there before Kalshi won their court case and then forgot about it), and saw there did seem to be duplicated markets there that should have been arbitraged but their limits on new entrants prevented that. But I noticed that the question for whether there would be a female President priced "No" at 49 cents, while Harris & Trump were both at 53/54 cents in their named next President market, so I figured that was the closest thing to arb and bought 40 shares. That difficulty in arbitraging should lower the status of PredictIt relative to sites like Polymarket.
>> Perhaps while everyone was voting for the Democrat, their "neighbor" really was voting Trump!
I'm... confused about this.
Nate Silver et al. know that polls have statistical bias, and they correct for them. If a poll underestimated Trump in 2016 (as many did) and 2020, surely NS measured the bias and corrected for it this time around. So... what's going on here?
I'm referencing the neighbor methodology used by Theo.
I'm still confused how this method is supposed to work.
I have no clue who any of my neighbours might vote for. I'm sure there's some loudmouths who can't help advertising their political allegiances to their neighbours, but none of them lives around me.
He says that it's worked before in other cases. But I don't know.
Andrew Gelman just wrote about that the other day: https://statmodeling.stat.columbia.edu/2024/11/09/polling-by-asking-people-about-their-neighbors-when-does-this-work/
It's unfortunately much more complicated then that.
If your poll in 2016 should have been Trump +5 and your poll in 2020 should have been Trump +1 what adjustment should you apply to your 2024 poll? Trump -3 because that's the naïve progression? Trump +3 because that's the average even if we have no reason to suspect the average will apply? +5 because he's not the incumbent? +1 because that represents people now having experienced a term of Trump?
Not to mention if you're a polling aggregator each of the polls you're aggregating has done independent attempts to correct for previous results. It should be noted though that 2020 and 2024 are pretty inline with historical polling, 2016 was the outlier in how far off the polls were.
This is the big issue here. The polling averages always reflect pollsters best efforts to correct for previous results. Attempting to correct for bias in the polling aggregate is applying a correction on top of the correction the aggregators are already trying to apply. In general, the direction of error hasn't been consistent over time. With enough sample size for elections with Trump specifically, polling aggregators would correct for that too, but we haven't had data to assume one.
I'm wondering here if it was the early vote effect. People in comments below are talking about the early votes, and I wonder if there was an expectation that this and mail-in votes would swing for Harris.
The (ahem) disagreement over the 2020 result was in part that Trump was (or looked to be) in the lead at the close of one day's counting, then the next morning suddenly it was Biden, and overnight all these mail-in ballots turned up to be counted?
So I'm now wondering if the expectation was "Harris is polling just a small gap ahead of Trump, but when the mail-in votes come in, those will be Democrat votes" and the assumption was that even if Trump led in early counts, once the rest of the ballots were counted Harris would pull ahead, and hence the polls were adjusted to take this into consideration?
What seems to be different this time, if I can trust the online sources hanging around in shady alleyways going "psst, wanna buy a post-election analysis?", is that the Republicans learned the lesson and made efforts to have people use early voting and mail-in ballots themselves, and this made the difference - it cut away the advantage from the last election, as well as being much tighter over watching counts, being present when returns were being made, and so on.
I feel Nate Silver has been a bit too cavalier about the recent consistency of the polling bias. In https://www.natesilver.net/p/why-harris-could-beat-her-polls he writes:
> However, the baseline assumption of the Silver Bulletin model is that while the polls could be wrong again — and in fact, they probably will be wrong to some degree — it’s extremely hard to predict the direction of the error. Empirically, there’s basically no correlation in polling error from one cycle to the next one.
But looking at the link in that quote (https://fivethirtyeight.com/features/2022-election-polling-accuracy/, table "Polling bias is pretty unpredictable from election to election") there was a pretty consistent bias in favor of democrats since 2013. Not fully, but calling it "basically no correlation" seems wrong to me. Maybe it was so in the past, but something having changed since Trump seems not implausible.
But misleading to quote that as his opinion, given he wrote paired articles making the case for each candidate
Problem is working out how much to compensate. And that'll vary by circumstances. You can't just add a blanket 2% if trump was underestimated by 2% or whatever
Silver's model generates a prior based on fundamentals (economy, incumbency, etc) and then adds polls to it, and the polls are weighted on their past reliability. But you still have the garbage in garbage out problem, if the polls were herding towards 50/50 (as he suggested https://www.natesilver.net/p/theres-more-herding-in-swing-state) then you can't tell what way to weight it to cancel out.
You only get a chance to correct your polls every 2 years, and those off-year elections have their own biases with the 4-year elections.
The state level errors are correlated, so you would expect the swing states to mostly go together. And Nate Silver's model actually had less probability on a PV/EC gap than the markets
The fact that Trump won by a lot doesn't really affect the correctness of the odds beforehand. When I'm about to bet on the outcome of a coinflip, it doesn't really matter whether the coin has been already flipped and is hidden from me or the flip is yet to happen, and it also doesn't matter whether it's "heads all the way or just 60% heads".
Correct odds are always relative to the amount of knowledge at the time. Give me two people at random and ask me who is more intelligent, I'll say it's 50-50 and that's a correct bet. If afterwards turns out that one of them beat the other on 9/10 of the tests we used, that doesn't make my previous odds incorrect.
Agreed, I understand that stuff about probability theory. But Trump's uniqueness as a political figure makes me think it's plausible that the shy Trump voter theory really might be right! We’ll never know because a larger sample size is impossible for clear reasons.
Exactly! This is the most convincing argument from me (apart from the early voting data). Timur Kuran (author of the famous book on preference falsification) also makes this argument: https://x.com/timurkuran/status/1850200912062566873
I'm very surprised by the popular vote, to be honest. Can anyone inform me about that? Either every single rural person turned out to vote this time, or the voters in the big population centres didn't. If it were the case of 2016 where Hillary was expected to win in a landslide, I'd understand the voters staying home because they thought a blue victory was a sure thing, but the polling did consistently show it as only a slim margin for Harris over Trump.
So it looks like the Harris campaign managed to not do their jobs properly (one of which seems to have been the emphasis on appealing to women voters over trying to appeal to men voters, and when they did try to appeal to guys, it was too flat-footed to work).
https://www.instagram.com/amalaclips/reel/DBOzSyhJkTQ/
"I eat carburetors for breakfast"? My friend, you look like you eat the entire truck (I can say that and it's not fat-shaming because I am a Person of Chonkitude myself). Granted, I only know Irish rural background myself, but if these guys are genuine American country folks, I'm a New York sophisticate in a size zero dress.
It seems that many latin americans decided to switch sides from Democrats to Republicans in this elections. Including in otherwise Democratic strongholds like NYC. https://www.bbc.com/news/articles/cze3yr77j9wo That drove up the popular vote. Trump also gained among working class african americans (although they still predominantly vote Democrat).
This makes a lot of sense to me and I've wondered why the usually christian (albeit catholic) and socially more conservative latinos vote for Democrats.
Now it seems there is no longer a big difference in political opinions between the working class with northern European ancestry and southern European ancestry (or mestizos) as both switched to Republicans (only the latter did so a few decades later).
By the way the advert is horrible and seems extremely ingenuine. And interestingly this is exactly what I thought about yesterday when I saw some images from campaigns of Harris and Trump. Trump is kitchy and obnoxious ... but Harris just seems fake and Trump does not. Of course he is really extremely fake in what he says. But it does not feel so cringeworthy as when I see Harris and Obama poiting fingers at each other in a "go girl" pose like they are 16 year old schoolgirls. I just don't believe them that this is really who they are (and if they were, it would perhaps be even worse). I believe Trump his behaviour ... It also shows he is a self-centered narcissist who talks bullshit but he is not pretending to be someone else. I would not vote for him if I were American but I would be very hesitant to vote for someone as fake as the current Democratic leadership.
Part of it, I think, is that with early Democratic Party you did have a strong strain of Catholicism from the Irish and Italians, so there was a common cultural background there to appeal to Latinos, as well as the reputation for being for the working man and the blue-collar workers.
As the Democrats slowly dropped that in favour of social liberalisation and chasing the college-educated vote, there were only the traditional links holding people to the party - we vote for the Democrats because our parents/grandparents voted for them, and they're the pro-immigrant party.
Catholicism is now probably, for the newer generations, a matter of 'cultural Catholic' or 'brought up Catholic' than deep faith, and for the first generation, again speaking as a cradle Catholic, there's a lot of folk piety but not much real understanding of the theology, as well as falling away to being influenced by the secular society around you (see all the polls done on contraception, divorce, abortion, etc.) The same way Pelosi and Biden can claim to be devout Catholics and still support abortion (sorry, is it 'reproductive justice' or 'healthcare rights' is the favoured term today? I can't keep up with how it switches).
So the cleavage point probably is "they're not representing the little guy like me" and Trump at least makes the mouth-sounds about bringing back good blue-collar jobs and immigrants have to follow the rules to get into America. Immigrant families who did follow the rules to get to America and become citizens don't like queue-jumpers any more than the rest of us do.
I think the Democratic party has generally taken the Hispanic vote for granted, on the presumption that the Hispanic vote should be convinced that the Republican party hates Latinos. It's a widespread opinion among the Democratic voter base that Republican animosity against illegal immigrants is a dogwhistle for hatred of Hispanics in general, but apparently not a particularly widespread opinion among Hispanic voters as a whole.
I think Trump is a bad candidate well beyond the usual scale of bad candidates, but the Democratic party had fallen pretty hard for the idea that the minority groups it's traditionally counted as constituents will vote based on identity politics rather than policy or cultural alignment, and while I don't think the average voter today actually does pay much attention to policy, a lot of minority groups are significantly more in cultural alignment with the Republican than Democratic party.
I think many people were disgusted by their choices for President, and it happened to favor Trump, as he had more support; people would turn out to vote FOR Trump, but if they were against Trump, it looks like they stayed home.
I base this on Trump winning about the same (maybe slightly fewer) votes in 2024 as he won in 2020, but Harris coming up short of Biden's 2020 by about 14 million votes. So lots of people didn't think Harris was worth voting for, but people still turned out to "protect" against a Harris presidency.
Low turnout in democratic safe states seems to explain a lot of it. Which is entirely sensible of you are efficiently allocating your get out the vote and persuasion efforts to swing states
A lot of Biden voters just stayed home.
I think Harris was pretty bad as a candidate and the fundamentals were against her, but she performed slightly better in the swing states, which is evidence her campaign was doing effective things.
Trump will win the popular vote, but don't use the current numbers to asses the margin. There's still millions of votes to be counted in NY and California.
Right. The markets currently put 25-35% on Trump breaking 2%. Definitely not 3%.
https://polymarket.com/event/popular-vote-margin-of-victory-in-presidential-election
https://kalshi.com/markets/popvotemov/popular-vote-margin-of-victory
One of the best criticisms I've seen of that ridiculous ad is that its entire purpose is to appeal to men, but it does not talk about a single issue men care about. "Be brave enough to vote for OUR interests" isn't how you win over a demographic.
> Either every single rural person turned out to vote this time, or the voters in the big population centres didn't.
Not even that, it appears that the voters in big population centres increasingly moved towards Trump. He didn't win New York City, but it went from 80-20 to 70-30, with the new votes coming from poorer neighbourhoods in Queens and the Bronx.
>So it looks like the Harris campaign managed to not do their jobs properly
The opposite seems to be the case, since her drop-off from 2020 in swing states (ie, where the campaign was operating) was substantially less than her drop-off overall.
When Silver would say that Trump/Harris were just a slight favorite, he was talking about the median outcome. Even a model that is extremely close at the median outcome (Trump vs. Harris winning) can have a wide outcome distribution, and I am certain that Nate Silver would admit ex ante that this was a strong possiblity (in fact Trump sweeping all swing states was the modal outcome).
>I am certain that Nate Silver would admit ex ante that this was a strong possiblity
You're right! https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/75961826.
Yes, but as I said here:
“Agreed, I understand that stuff about probability theory. But Trump's uniqueness as a political figure makes me think it's plausible that the shy Trump voter theory really might be right! We’ll never know because a larger sample size is impossible for clear reasons.”
Also, the fact that real money markets seemed to think this make this plausible.
He won the swing states by tiny margins. Given that polling methodology has a 6 point error margin, it would be very hard to predict elections that over and over get decided by a couple of percentage points. We're basically flying blindly in that scenario.
Nate Silver, if you were paying attention, said that the most likely outcome of this election was Donald Trump sweeping every swing state and the popular vote, and the second most likely outcome was Kamala Harris sweeping every swing state and the popular vote, and then a bunch of lesser probabilities after that.
I don't think there's anything wrong with Nate's reasoning or forecasting. I think the problem is with your misunderstanding of what a close election looks like. Because this is exactly what a close election looks like - the winner getting their safe states and most or all of the swing states, but *only* those states and only 51% of the popular vote.
Whether the winner absolutely sweeps the swing states, or whether some minor variation leaves one or two of them in the loser's camp, is lost in the noise. But if you care about that bit of noise, the Silver got it right albeit with very low confidence.
Fake money prediction markets only perform better sometimes because they are used by super high IQ nerds without (fore-)skin in the game
But in this case, Metaculus was wrong and Polymarket was right. Metaculus users are very left leaning, that's why they mispriced this completely
(i won 10k on Polymarket)
It's genuinely baffling that someone could look at a one-off event and say "Forecaster A said there was a 40% chance of rain, B said there was a 60% chance of rain, then it rained. So forecaster A was totally wrong."
You just can't use a one time result to assess the accuracy of a prediction market. Any *single* observation is consistent with a probabilistic estimate, you need to assess a range of predictions or a large number of observations.
Anti-semite!
But seriously, doesn't that make them more effective for their stated purpose? Technically the profit is an inefficiency in the system as it makes use of arbitrage.
I guess it does. Whatever system you have, limiting access to >130 IQ people always makes it more effective
Well, for something where intelligence is useful. I doubt, for example, movies with only actors with IQs over 130 would sell all that well.
Acting well takes a good deal of intelligence, for you must take on another role based on your understand of who that character is, and yet still maintain who you are as a separate identity. And switch between these identities on demand.
Right, but once you select for 2 percent of the population, you limit your ability to pick good-looking people as well, which movies always require.
I'm not saying smart people can't be good-looking; I'd actually expect the correlation to be weakly positive on account of decreased mutational load. But if you pick *only* from the top 2 percent of the population in IQ, the cutest person is going to be less cute than if you pick from the general population.
"limiting access to >130 IQ people always makes it more effective"
If your point is that you won't find enough people that look the part and also have >130 IQ, then that is reasonable, and I agree, for there is a lot of competition for intelligence across many fields. But if you could get them, then movies with such IQ actors should still sell well, for they should "be" the characters in addition to looking good while doing it.
> you limit your ability to pick good-looking people as well, which movies always require
This seems to be a Hollywood idiosyncrasy. Movies/TV shows in the UK don't particularly focus on casting lookers more so than ordinary looking people who can act really well.
Couldn't you make the case that, a day before resolution, when interest rates etc stop mattering, any prediction markets trades not at 0 or 1 are mispriced? Isn't the whole point here that, because there is a truth of the matter, no intermediate value could possibly be the true value? It's not clear to me that markets, in general, have "accurate prices", which I why I don't simultaneously trade in all possible markets for all possible things. There is only "worth my money". Or am I missing something about the intent of the post?
I don't think so. For example, if I'm going to flip a coin, any bet other than 50% is "mispriced", because we have enough evidence about the process to know that it ought to point to 50%.
I agree it's harder to come up with a clearly meaningful definition of mispriced for an election, but see https://www.astralcodexten.com/p/in-continued-defense-of-non-frequentist for why it should exist.
I agree 50% is the calibrated understanding of a coin flip, and I'm not frequentist myself. My comment is mostly about how these financial instruments are abstractions that strive to be related to probability, but aren't probabilities. The market structure is precisely why they might be valuable. But the market behavior is what we interact with, not the probability we are imagining. My initial response was, it is a breach of etiquette to *say* a market is mispriced, the appropriate etiquette is to *trade* that a market is mispriced. But if you acknowledge the trades aren't worth the hassle, then, what's left? The probability you infer from the market is not a product of that market. I don't mean to be pedantic here, it just seems the complaint isn't about mispricing, it's about real money markets not doing the secondary, aspirational thing you want -- letting you generate probabilities -- which, as you point out, can be achieved without any price mechanism at all.
I'm trying to make the case that "miscalibrated", the term related to predictions, is much better, I suppose. Perhaps I *am* being pedantic, but "mispriced" might not be a word it's possible to use when buyers and sellers agree on a trade.
This post is completely wrong about almost every single thing. Will write my own Substack article to refute it
Looking forward to it. The post basically says "if you had x prior, then you should update to y posterior after the given amount of evidence" in the context of metaculus and polymarket, and to me the math checks out. Curious how you are going to show that it's wrong.
Ok, I will respond to and cite your comment in my post
Thanks!
I'm not Alexej, but Scott's basic mistake, as others have pointed out, is acting as though the only information we got was that Trump won. We also learned that he won by a substantial margin, which is evidence that the probability of his losing was substantially less than .5, which is consistent with Polymarket but not with various of its competitors.
This is a good point, although I would hope Alexej would have more in order to conclude Scott is "completely wrong about almost every single thing", rather than just this one mistake, although it is a pretty significant one.
I'm not sure how you would quantify the amount Trump won by into Bayesian information though.
Totally valid concern to raise, however as I wrote in another comment, I don't think it pans out. Do you have an actual model for why (and how) the winning margin matters as evidence? I can think of a bunch of examples where it clearly doesn't.
Define "substantial margin".
Presumably if a candidate won the election by a single vote, or ten votes or 0.0001% of the votes, you would acknowledge that election as having been basically a coin toss.
By what percentage of the vote (nationwide or averaged over swing states) does a candidate have to win for you to say "the probability of that candidate winning was 55%"? What about 60% or 70%? At what popular vote margin do we retroactively assess the candidate as having had a 90 or 95% chance of winning?
I am genuinely curious as to what sort of math you might have for this. But without math, it's meaningless.
To add to that, even if a candidate has exactly one more vote than the other, then they have an exact 100% chance of winning (I know about electors, but there's always a vote that wins the election, let's go with that one). It's the rule, more votes win the election, there's no randomness in it, the ballott counters don't do random coin tosses.
So why do we have odds? They are determined in terms of the knowledge we have at the time. So any discussion that doesn't involve some concept of prior knowledge is moot.
thanks I like Scott but I feel like he was missing the ball on this one. It's like I told him to fetch but instead of bringing back the ball he brought back a kids old baseball glove. No! Stop! Don't chew on that! Go get the ball! I hope u help him find it.
Yeah, I almost never disagree with him. Now is the first time I *strongly* disagree with him on anything
Excellent.
One of the interesting things about tonal languages is that, on top of the complexity of tones, you will sometimes get regional dialects that just make up their own tones for common words, so you have to throw everything you know out the window when visiting a new town. In some cases, tonal patterns just feel inverted. The upwards 2nd tone is transposed with the falling 4th tone, and vice versa, across several common words. You can get really confused just by crossing a river. (河, hē, hé, hě, hè, in Hefei, Beijing, Wuhan, and Ürümqi, respectively).
Ok I was bluffing when I said "tonal languages." In fact, almost all languages are tonal, even English. We use them for grammar and emphasis instead of semantics. And we can flip our tones too. This will have hilarious or irritating results, depending on your perspective. In one interview, the Pythons recalled the somewhat symphonic nature of their arguments behind the scenes. When the Welsh Terry Jones was upset, he'd lilt up at the end of every sentence and get very high pitched. John Cleese, with his boarding-school RP, would slow his speech and go down at the end of every sentence. So you would get this cacophony of accelerating piccolo and descending stair step baritone when they were fighting over some sketch. As a result each side would very poorly judge when the other was getting riled up.
So it goes with linguistic registers. We don't really think too much about them, we don't teach them or talk about them, so they can get all mixed up and backwards.
I've noticed this happening with a different sort of register in English, if I can call it one: how we convey certainty.
Most of the people I work with use nuance to show conviction. They want to show off that they've really studied something, that they see both sides. They do so by hedging, steelmanning other positions, showing deference to other views (which, after all, might change their view on the margins, but pose no serious threat of reversal!).
MAGA vibes completely flummox this crowd, because that crowd doesn't play by those rules. Its leading voices often deploy hyperbole to demonstrate the same thing. "I've really studied this" becomes "I know it's true, I'm 100% right, and everybody who doesn't agree with every one of my points is completely wrong, (give or take an insult about their intelligence)."
This isn't a criticism. I'm just Attenboroughing here. I'm noticing two groups, who otherwise somehow navigate all the same cultural spaces just fine, yet somehow came up with two completely opposite and incompatible ways to demonstrate that they really believe what they say.
It is as though we are all trapped in some elaborate farce.
I first caught on to this when Trump was on some AM Christian radio show years ago. The host said, with genuine sympathy, "Hey you caught some flak for saying Obama was literally the founder of ISIS. But what I hear you saying there is that his policies ended up leading to a situation where..." At this point Trump cut him off and said "No. I meant Obama literally founded ISIS."
It took me a long time to realize what the hell was happening in that clip. I think I finally get it? Kinda? It honestly still sounds absurd to me, I have to strain to hear in the wavelength where it makes any kind of sense.
So I say MAGA opened my eyes (ears?) to this. But after noticing it a while, I no longer see it as political. I know rights who speak nuance, and lefts who exaggerate for emphasis (oof, or so I hope anyway).
It brings to mind the conflict vs. mistake dichotomy, but it's really not that. It is related though, because team emphasis could easily get mis-labeled as conflict theorists. They're just using an inverted register to say "I actually believe my position, I'm not just bullshitting you." And when you raise probing questions to clarify their main points, find limits, ask for some nuance or both-sides-ism, it gets interpreted as, "why don't you adopt a more bullshitting posture," or in other words, "nah I think you're just bullshitting." So they double down and turn up the emphasis another degree. They shout louder to get through whatever earplugs you seem to be wearing.
This creates an unstable feedback loop. Discussions disintegrate shortly after, both sides leave questioning the other's sanity or seriousness.
If this becomes a separate post instead of just a reply, I thought of this when I read Alexej.Gerstmaier's 7 Nov comment:
"This post is completely wrong about almost every single thing. Will write my own Substack article to refute it"
Which is perhaps the most copypastable comment I have ever read on the internet.
But also... ok hear me out... maybe just a great textbook on how to talk like you actually mean what you say.
If I were Scott or EY I'd have to give this dichotomy a catchy name I guess. Emphasis talkers and nuance whisperers? Eh, idk, feels a bit silly.
If anybody can think of a better name, say so and I'll dub you Thingnamer.
Very insightful comment! I once talked to an HR lady about the statement "Dogs are larger than cats". She insisted I was wrong while I called her autistic
Back in the early 2000s, I'd follow Hollywood Stock Exchange, a non-money prediction market for movie revenue, to get an idea which new film to review this week. (Editors don't like it when you review what turns out to be a dud that nobody is talking about instead of the new bombshell.)
Hollywood Box Office, which was owned the Wall Street firm Cantor Fitzgerald (which was devastated by 9/11 due to their offices being in the World Trade Center), was usefully accurate, in part because it encouraged insider trading. HSX encouraged people whose jobs gave them access to watch movies early (or see the daily rushes even earlier) to wager pretend money. Insider trading is a good thing when it comes to upcoming movies because it's good that more people hear about movies worth watching and, especially, more people hear what movies to avoid wasting their money and time watching.
Cantor Fitzgerald wanted to turn HSX into a real money futures trading exchange. But I never heard how insider trading was supposed to be handled with real money on the line. Would the SEC be in charge of prosecuting pool guys who overheard their clients raving about a movie they are working on? It seemed pointless.
Sorry, Hollywood Stock Exchange, not Hollywood Box Office.
The laws for insider trading don't apply by default to futures. (Unless the future is actually related to a security, for which insider trading laws do apply).
Box-office futures are a bit of a strange story in financial products. It was controversial whether they should be allowed to exist given the legal framework for commodities. They were approved in 2010 by the CFTC by a 3-2 vote, only to be banned one month later by Dodd-Frank.
Thanks.
The question is whether insider trading is a good thing or a bad thing. When you can't make money off of insider trading, as in the turn-of-the-century Hollywood Stock Exchange, it's clearly a (small scale) good thing because it reveals insider knowledge, which I found useful in choosing which upcoming movies to review.
But if you can cheat other people out of real money, it's more of a question.
I don't know the answer to that question.
My realization when I became a weekly movie reviewer in 2001 was that some movies were obviously pretty good and that other movies were obviously pretty bad, which came as a surprise to me. For example, to compare early 21st century sword and sandal movies, Ridley Scott's Gladiator is clearly pretty good and Oliver Stone's Alexander is clearly pretty bad.
But that was hard to predict ahead of time, except among insiders who'd seen the movies or at least footage from them.
It's less a matter of subtle aesthetic taste on the part of the critic than that making a good movie is hard, and even successful directors like Scott or Stone misfire at times. So some movies perceptibly hit on all 8 cylinders, while others obviously don't.
So, to my surprise, much of reviewing movies is pointing out that this movie works or that movie doesn't work. I thought it was going to be high level Pauline Kael vs. Andrew Sarris aesthetic argument, but a lot of going to previews of new releases is just reporting "This is awesome" or "This is awful."
Generally speaking I agree with your comments here but the example given is not apt because Oliver Stone’s past performance is certainly an indicator of his future results
Here is his Rotten Tomatoes page: https://www.rottentomatoes.com/celebrity/oliver_stone He had plenty of fresh movies prior to Alexander, but his most recent credits at that time were not.
Realistically speaking, and setting review sites aside that seem to believe 90% of movies are good, Stone has never made a single movie worth the ticket price
If everyone knows that insider trading is allowed, no one will get cheated. You know what you are getting yourself into.
Yeah, the debate about insider trading policy is interesting, because there are definitely tradeoffs with allowing or restricting it. There's variance among governments and markets about how to approach insider trading. AFAIK, in the US, only "securities" have the insider trading restrictions. Securities are generally the things you'd think of as investment products: stocks, bonds, etc. Other financial products, like commodities and derivatives thereof, have much less (if any) restriction on insider trading. Indeed, with commodities, insider trading is pretty much the whole reason for the existence of the markets: producers and consumers want to be able to hedge their risk via buying/selling futures.
Relevant to the topic at hand, prediction markets wouldn't be considered securities, so insider trading restrictions wouldn't apply by default. In some cases that's a good thing--you want the markets to aggregate information even from insiders. But in other cases, it could lead to perverse incentives or disorderly markets (e.g. if a person can both trade the contract and influence the outcome). I'm not up to speed on how prediction markets currently handle insider trading; although I expect it's a solvable problem, the right solution might not be obvious or straightforward.
> All of these factors gave me a 90% prior that Metaculus was better calibrated than Polymarket on the elections; now that Trump’s won, I update to 88%.
Your earlier point about not updating too much on a single coin flip should, I think, also make you update less on those factors. Effectively we can think of metaculus as having flipped five heads so far (more or less), so you should start out believing in metaculus but only by like 75% or so, not 90%.
I think the studies of why Metaculus is better includes many less-than-perfectly correlated results from each election, plus the evidence from the (many different) questions on my contest, plus the common sense evidence from Theo being one guy making the bet.
hm. I think you get ~2 bits per election for ~3 elections (the results aren't perfectly correlated but they are mostly correlated, and I don't think metaculus won literally every time? I'm ), maybe another 1-2 bits more from your contest (a lot more information but also it should be more discounted since the main advantages of election betting markets, like high liquidity and money involved, are missing). I don't know how much we should discount the one guy thing by (it seems more like inside view than outside view?). So call it about 8-10 bits overall. 10 bits gives us about 86% of Metaculus being better, but if we assume Metaculus lost one of those bets (and I think that's a fair assumption - the linked first sigma post does mention some races they did better on) that goes down to about 80%. Getting more precise than this requires a lot more research than I'm willing to do right now, but to a first approximation I think the 2024 election should drop us from about 80% to about 75% (which is what I get if I add in the 2024 prediction gap) that Metaculus is better.
Were you aware that Theo commissioned private polls, with his own methodology and at his own expense?
1. This means he's literally bringing information into the market that was unavailable to anyone else (and nobody would probably procure if there weren't a chance of recouping the investment; ie. non-money markets, poll aggregators)
2. If I say that this fact means that *my* prior is 90% that the money markets are giving a *more* accurate prediction than the other forecasters, so now my prior is 92% -- does that make me a good Bayesian?
3. The actual results indicate that the polls systemically underestimated Trump (again) -- if all the polls on Nov 3 happened to have landed on the actual state results +/- a random 3%, would you really think that the race was a coin toss? For reference, that would be Trump at +2 in GA & PA, +1 in WI & MI, +4 in NV/AZ, and +13 in Florida -- 60% seems a bit *low* on that map, to me.
Scott, what the fuck.
Why a rapidly half-assed stealth edit where you simply mention the private neighbor polls and go with “I’m just not impressed by private polling”?
Because a smart whale with non-public knowledge incentivized to pay to gain such knowledge and use it to affect market prices is kinda the perfect goddamn argument for why real money could outperform play money or non-market mechanisms.
Do we have any details about what those polls were exactly, what their scope was, their methodology, etc? If he asked 1,000 people using a poor methodology and not weighting for the location of the responders, I’m sure he’d get a “Trump solid win!” conclusion but that wouldn’t count as useful new information.
If I said “Theo was correct because he consulted an astrologist” that would technically be “new information” but you’d still challenge me. Same thing here - “Theo paid for polls!” alone doesn’t update me much without knowing the details OR seeing his track record winning money using such polling.
This is the key question imo. People often like to ignore the methodology of polls because who wants to spend their free time poring through the boring details? Yet, modern polling methodology contains quite a number of free parameters whose unjustified setting could easily show whatever result the poller wants to show. I also did not go over the details of the poll in question, but to me it does not sound at all unlikely that a smart generalist could outperform the old guard who simply just wants to do their job and not rock the boat that much with too suprising results..
The crux of our disagreement is the probability you assign to "a smart generalist could outperform the old guard" in the area of voter polling. I assign that <1% probability without seeing his exact methodology OR a long track record OR numerous insanely good bets in the same election. You assign it... a higher value? 50%?
This post doesn't mention that Theo secretly commissioned his own polling - it just references him "working off of really smart reasoning." Not only does that render the particulars of the post somewhat misleading, it also affects the underlying argument, somewhat. The argument is that a single very rich trader isn't that much smarter that many poorer traders who won't bet as much as him. But as this case demonstrates, the single rich trader has the resources and incentive to create non-public knowledge to leverage for his bets. So while real-money markets may still be less accurate than play-money ones due to the ability to be skewed by individuals, the ability for rich individuals to not only reason like anyone else, but create their own proprietary information mitigates the disadvantage of real-money markets.
I don't know whether the article changed since you commented, but I do see it mentioned in there.
It was a stealth edit.
He edited the post after I commented (but added a new typo there, in his haste), but I don't find his edit compelling. He now claims that "I don’t find the private polls very interesting either - the existence of private pollsters implies this happens often, and we shouldn’t expect these private polls to be massively better than the public ones."
But first of all, the article states that he commissioned a major pollster who he wasn't allowed to name, per his agreement with them. So this was almost certainly not some "private pollster," but rather a well known public pollster. As such, there's little reason to assume that this happens often. As far as we know, the business model of the major pollsters isn't dependent on conducting secret polls for individuals - why assume that something like this is then routine? The original post strongly implied that he hadn't read the article he was linking and this point keeps the question open of whether he has yet read the short article.
More significantly, even if Scott is right that secret polls aren't "massively better" than the public ones, they don't have to be! Polling averages are useful and even if the secret polls weren't better than the public ones, they still give a trader alpha that can allow him to beat the market, which is what Theo did.
The issue I raised therefore still stands. While it may be true that real-money markets are more volatile, since a single very rich person can afford to take a large gamble that poorer people can't effectively bet against if the market isn't liquid enough, the rich people have the ability and incentive to create their own information that smaller traders can't. Their info doesn't have to be massively better to allow them to make money off of trading.
As far as the editing, the original version of Scott's post was:
"—But didn’t Theo give a great explanation of his strategy to the Wall Street Journal, which proves he was working off of really smart reasoning?
Yes, but there were dozens of people who could give equally-plausible arguments for their positions before the election. These were divided half-and-half into intelligent-sounding pro-Kamala arguments and intelligent-sounding pro-Trump arguments, and Theo was a completely replacement-level example of the intelligent-sounding pro-Trump arguments. We should think of him as an example of an intelligent person with a good argument who got lucky, unlike the many other intelligent people with good arguments who didn’t.
(none of this is meant to knock Theo..."
I’m 90% bullshitting here but I believe the business model of many polling companies is, if not “secret polls for private individuals”, low-profile polls for private clients; I don’t see how those companies could be in business between elections otherwise. Market research and such.
>So this was almost certainly not some "private pollster," but rather a well known public pollster.
Fine, but given that the public polls had lots of correlated non-sampling error, why would a private pollster do better? They're doing the exact same thing as the public polls, so you would expect them to have the same nonresponse bias.
First of all, I pointed out in my comment that additional polls, even if no better than other polls, still create a more useful polling average that gave Theo an advantage that he capitalized on.
There's reason to think his private polls were particularly useful, though. First, the public pollsters were evidently herding, that is, their consistent 50/50 results were unlikely even were the actual electoral split 50/50, demonstrating that they were putting their thumbs on the scales. The reason for herding is presumably to avoid embarrassment if the results end up being very different. But that issue is mostly obviated when commissioning a private poll on the condition that the results remain private.
Second, Theo suspected that traditional polling wasn't capturing Trump support and that "neighbor polling" in which you ask whom people think their neighbors would vote for would do a better job. Whether or not he was right about polling methodology, since his polling method differed from the norm, he had particularly useful information.
That's an interesting point that I hadn't considered. Thank you.
My pleasure!
Do we have any details about what those polls were exactly, what their scope was, their methodology, etc? If he asked 1,000 people using a poor methodology and not weighting for the location of the responders, I’m sure he’d get a “Trump solid win!” conclusion but that wouldn’t count as useful new information.
As the linked article mentions, Theo commissioned "neighbor polls" in which people are asked how they think their neighbors will vote, instead of how they'll vote, themselves. Theo thought that this would better capture Trump support that people may be less inclined to reveal to pollsters. The details of the poll remain private per his agreement with the pollster.
Right but how do we know his method and sampling actually made sense?
We don't, just that he was willing to bet somewhere between 30-75 million dollars on what they said.
Why does it matter what form the "really smart reasoning" takes?
It's one guy, acknowledged as really smart, claiming that he can beat the market but with no prior track record and only a vague description of his methodology. That shouldn't move anyone's priors very much, and the details of the cleverness matter not at all to the reassessment of the priors.
"But long after Biden won the last election, PredictIt said there was a 9% chance *he* would be the next President; some commenters suggested that maybe he would #StopTheSteal and win a fair recount (aside from the inherent implausibility of this, some of the specific scenarios bettors placed money on required him to win California, where his campaign hadn’t even asked for a recount)."
based on reading this a couple times I think "he" here refers to Trump?
Thanks for the correction.
How does a $30 million bet move a $2 billion market so much? Why does that one bet move the market 10%?
Ik also very confused about this. I guess the $2 billion refers to the total trading volume, which also counts back-and-forth trading? The order book certainly didn't have $2 billion in it at any time.
https://fortune.com/crypto/2024/10/30/polymarket-trump-election-crypto-wash-trading-researchers/
> Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%. Now should we say that the “true” probability of Trump’s win was 90%?
No, because Theo's goal wasn't to drive up the odds, it was to take advantage of a mispricing. While he was confident enough to buy Trump at 60%, it's unlikely that he would have been willing to do so at 90% odds.
Didn't he accidentally buy some contracts at 90% odds or was that another polymarket whale
You can find all his trades at https://www.polymarketanalytics.com/?trader=0x1f2dd6d473f3e824cd2f8a89d9c69fb96f6ad0cf
Unless Theo thought the true odds were >90%, in which case buying Trump at $0.90 would still be positive EV from his perspective. To the extent that he was money-constrained, we don't know how high he might have pushed the price with more available capital.
anyone suffering from trans twitter brain rot might try dunking their head in the nearest bucket of cold water ive found this helps.
https://www.nbcnews.com/politics/2024-elections/exit-polls
We actually have data here, and it was the economy, maco. Mostly inflation; immigration was a distant second.
Abortion and preserving democracy were the big choices among people voting for Kamala, who after all almost won.
While I support a national healthcare system, including sex changes/gender-confirmation surgery where appropriate (some people are genuinely better off that way), at least for citizens... I have to admit the line on 'transgender operations for illegal immigrants in prison' was a pretty effective slogan, as it ties together anger over cultural changes, anger over illegal immigration, and anger over crime.
Thank you for the link! But I don’t see how the poll completely disproves maco’s claim, though: in the “issues you care about”, NBC did not ask about “cultural” topics such as, say, wokism or trans rights – was it an open-ended question?
I suppose you're right, and NBC, being MSM, would have good reason not get themselves in trouble by asking about that.
But I wonder if anyone else has asked the question!
Trannies think the world revolve around them. It doesn't. Literally almost no one cares. Yes it's always good political tactic to say the other party is aligning with the freakshows ( this is what many trans people appear as to the average uneducated voter im sorry) whether it is the satanists or the pot smokers or what have you. This strategy works. But if it wasn't transtrenders it would just be whatever other subversive unhinged movement infected the youth. It is not the trans themselves that are the unique issue it is what they represent to the voter. They are a symbol of degeneracy. This always exists in society and the left has been guilty of openly feeding into it as of late.
“Literally no one cares.”
Who just wrote a long answer focusing on that topic when I was asking about culture war in general? Called them a “symbol of moral degeneracy” (as opposed to many possible formulations highlighting their irrelevance, because a symbol is by definition significant)?
> a long answer
122 words is a "long" answer?
The debater’s answer would be that it’s still longer than not mentioning them at all or in passing, or in two sentences.
(It’s also longer than your answer, isn’t it? :))
The honest answer is that mobile screws a bit with my perception of length, especially far in a thread.
Maybe we try not to care, but it's put in our faces so much , especially if you are on the Internet much it feels as if "the trans are out In force" , so then we find ourselves dragged by its sheer oversaturation to dicuss it. This is how eric Weinstein and co. Found themselves expounding on trans issues for hours on podcasts. They were dragged in by the gravitational force once they entered the online sphere simply due to the outsized presence of trans individuals. Who are mostly AGP by the way. Trans Androphiles sleep with normal boys and dont spend their time commandeering every Internet message board known to man.
Spent an hour yesterday talking to a trans man about the election. He thinks the Twitter LGBT scene is ridiculous, because people seem to think their subgroup is the only group that matters. He hates wokeness. He was furious at radical lefties who voted for 3rd party candidates in swing states out of purism. Main concern about having Trump in office is that he has illegal immigrant family members in the US, and is worried about them. So how about reminding yourself that trans people are like other peoples -- some are childish and narcissistic, some are not. Most groups on Twitter look like self-important hysterical assholes. Twitter is designed to attract people and turn them into self-important hysterical assholes.
It was quite a choice, wasn't it? One side favored transgender surgeries for illegal immigrants in prison, while the other side would prefer to just castrate them.
You forgot the most important part of this. ...Why should taxpayers have to pay for this? In a country that doesn't even have a fully public health-care system like what Canada and Great Britain have. It really was an absolutely awful policy idea put forward by Harris, and not just because it was bad for her politically. It's because it's totally unwarranted given the current status of the American health care system. Why the heck should a non-citizen wanting expensive gender surgery take precedence over a citizen who needs heart surgery or cancer treatment in order to stay alive?
Trans are failing at the primary objective: be pretty girls! In Asia they are everywhere but they blend in and you barely bat an eye. Everyone would be far more willing to accept them if they succeeded in their primary mission. Instead they are blaming us because they did not live up to their name! If you are trans I shouldn't be able to spot u looking like a bigfoot from 100yards! Less drama and tears on twitter and more effort on transitioning please !
...What the actual fuck is wrong with you?
tim dillon show is a mainstream popular podcast that interviewed the vice president last week. he goes on rogan all the time. hes a popular GAY comedian. its funny a bit he did about a family saving up to buy their trans son a pussy for christmas on his podcast. what is wrong with sharing that. the news story about the kid dying was real and very tragic.
That's... not the part I was objecting to...
But honestly, it's nice to get some confirmation on the actual reason everyone hates trans people. It's fine though, deal with them as you wish. Nobody deserves happiness.
You know, I actually really like raunch, and I'm ok with hearing female genitals called pussies or cunts, and use those words myself both ias swears and to refer matter-of-factly to genitals. I'm a woman, by the way.But this joke offends me, so was trying to figure out why. And I did.
It's 2 things: The first is that it mixes sex and parents It makes me think about parents checking out their daughter's genital secretions. It's not that I disapprove of being transgressive in that particular way,it's that I just find it creepy and unpleasant, a real buzz-kill for my laughter.
The second is that it hinges on the idea that women should be ever-ready. As you probably know, even natural-born women do not have mucus running down their legs. They only secrete the lubricant when they are turned on. If you want frequently wet pussy in your life may I suggest that you work on getting way way WAY hotter than you are, and greatly improve your social skills. Until then carry a pocket pussy and a great big bottle of lube. If the bottle leaks, you too can leave trails of slime.
Banned.
Banned.
I found Mark Robinson's new account.
For some M to F trans people, being pretty may not be their primary objective. They may just want generic femaleness, or to ditch their male genitals, which have always felt to them like a grotesque lump that should not be there. And they may be attracted to women before transitioning and after, and prefer looking dykey. Also, transitioning is expensive and time-consuming, and most people, trans or not, are not in a position to travel to another country for surgery.
You aren't coming right out and complaining that it's jarring to see a big broad person with female attire and hairstyle, but I sort of think that's the root of your complaint. If so, hey, I get it. I find it jarring too. I find sexual ambiguity or change of gender jarring. I'm sure nature wired us all to distinguish between male and female, and of course the culture has too. It's natural to feel some recoil or confusion when confronted with someone who has characteristics of both genders in their appearance or behavior.
But I don't think it's reasonable to expect transgender people to spare you the experience of being jarred. What really works for weakening that reaction is just desensitization. Log enough time with trans people and their transness will just be a lot less salient, and you'll process them the way you do non-trans people: is the person an asset or a liability at work, do you like their sense of humor, do you share any interests, what do you agree or disagre with them about, etc.
Reading trans Twitter does not count as logging time with trans people. Most groups on Twitter are caricatures of themselves, and act like self-absorbed, hysterical assholes.
"Trump won because of my pet issue" might be true, but also said by everyone, including all the wrong people.
I think that giving sex changes to prisoners was dumb, probably cost her some thousands of votes, and was overall a symptom of her poor political skills. But people really hate inflation and being gaslit about it.
It's the economy, stupid.
on twitter why does the profile say "I'm a soup puppy kitten puddle slurpy mess goo bunny" and it looks like a boy that estrogenized his whole body to look like a preteen. I come across this profile aesthetic a lot flying the trans flag. I'm sympathetic to exploring gender to an extent but this is the kind of thing throwing some people off &_&
Banned.
I'm genuinely curious about this ban...is it because he used "tr***ies" and that is considered a slur?
That didn't help, but I might have let it slide if it had been in a thread about transgender issues or contained some interesting insight. Instead it was in a thread about prediction markets, and distracted from the topic with a one-sentence assertion of a very controversial opinion.
I've said before that I use some set of criteria like "true, kind, necessary". As far as I can tell this statement wasn't true (or at least the poster provided no evidence for it). The slur didn't help with kindness. And it was basically unrelated to what this thread was about. 0/3.
Thanks for the explanation!
Was it really mispriced? Early vote data showed about equal number of suburban and rural voter in Florida and Texas but missing MILLIONS of urban vote. it was inconceivable for them to make up this deficit on e-day and they didn't (both carried by 14 points flying In the face of any polls). Anyone looking at early voter data with an open mind seems to be much more confident about their predictions in certain swing sates like AZ and NV (that both had solid early voter data, nv being the most complete). The markets had these at -300+ at points because they were using early vote data.
For myself I bet 14k on Nevada on a sportsbetting site I use with very little fears even though I'm generally very risk averse and that was the largest bet by far i ever placed .
Stil waiting for the finalized results for the 12k pay out uwu
Other people were insisting the early voting numbers were good for Harris. Nate Silver had to talk them down here: https://www.natesilver.net/p/the-early-vote-doesnt-reliably-predict
(I don't have the full post, but I'm just talking about the content above the fold.)
Clearly a bunch of people wrongly thought the numbers were in their favor, while you thought it was in yours. What's the difference?
(Can't you see the party affiliation of the early voters? I think that would be the simplest proxy measure.)
You can’t know the party affiliation for certain, but you do know where they’re voting and that tracks with party affiliation pretty well. High early votes in rural counties (which are pretty much always lopsidedly Republican) with lower early votes in urban areas (lopsidedly Democrat) and you can make a good guess as to how enthusiastic the early voters are by party.
> Yes, but this system is meant to work in a world where amounts of money are at least somewhat even.
No! The system is meant to work in a world where the amount of money flowing through prediction markets is *a significant fraction of the economy.* The whole idea is if you can systematically make true predictions you can pull alpha out of the prediction market. You get a very *uneven* distribution of money, and that *correctly* drives your prediction strength up, because the market recognizes that you're a reliable source of accuracy. Even amounts of money would ruin prediction markets by turning them into democracy. The real problem is that being able to produce value for society has no reliable connection to predictive accuracy of bets. - Though, even then I have to say that being able to produce value via a banking job probably has a much higher connection to predictive accuracy than most other jobs, so maybe even in that regard the system worked as designed here.
I think 2% is a bit generous, we've seen that prediction markets have a rightwing bias so this is perfectly in line with expectations, I'll be updating 0.2%.
As to why prediction markets have a persistent bias when markets are supposed to correct for them: prediction markets don't just predict the future, they also *affect* the future e.g. if prediction markets predict that a certain policy will have a positive effect they're more likely to be implemented. This gives market participants an incentive to be more optimistic towards policies that if implemented would benefit them e.g. if Musk bets a million dollars to shift a policy (that would net him a billion dollars) towards a positive perception, then even if he loses that million dollars on the prediction market he more than makes up for it with the billion dollars he gains by having that policy implemented.
This brings us to the demographics of prediction markets, which is mostly rich white guys, people more likely to be rightwing. If you're a poor black woman who's inside a poverty trap living paycheck to paycheck you can't participate in the prediction market even if you want to. So even if you know that the policy that Musk wants to implement is just going to enrich rich white guys at the expense of the poor (which you might not know since living paycheck to paycheck doesn't give you a lot of time for policy analysis, but that's a whole different problem with prediction markets) you can't even correct it.
(for a more precise version of this argument see: https://bobjacobs.substack.com/p/conditional-prediction-markets-will )
The whale wasn't Elon, it was a French guy who won't be living under Trump's governance and just wanted to make money.
If there was a conditional prediction market, anybody with money who wanted to make more could seek out info from any "poor black woman" that had it.
I never claimed it was Elon that was just an example, why would you even think I believed that given that I didn't talk about the whale, did talk about a fictitious policy proposal, and that the post I'm commenting on literally talks on and on about Theo.
It's not that people with vested interests can't make money by predicting correctly, it's just that they can make even more money by predicting incorrectly, which means conditional prediction markets as a whole will, on average, be skewed in the interests of the wealthy.
I didn't claim you did claim it was Elon. I'm claiming that the facts of the case don't fit your hypothetical. As far as we know Theo derived no particular benefit from the outcome other than his bets paying off. They CAN'T "make even more money by predicting incorrectly", because they lose all the money on incorrect bets. Bet incorrectly and someone like Theo can figure it out and bet against you to take that money. Nobody on earth is rich enough to buy the pot from everyone else on earth willing to take their money.
I wasn't talking about the case/Theo, I didn't even mention them. And people CAN make even more money by predicting incorrectly if that causes a policy to be implemented which makes you more money, like I explained in my comment. It's not necessary for everyone to be on the same page to skew the result e.g. the scenario in the post.
You can't get the policy implemented if people who know better bet against you.
Even assuming perfect information (which as I pointed out is only true in the ancap's imagination and not in reality) this still isn't true since one group could outspend another group e.g. the scenario in the post
"Until early October, Metaculus (the top forecaster) was consistently +4% bluer than Polymarket (the top market)."
This is incorrect. I have a graph: https://photos.app.goo.gl/Tuab9CoLaL96cZpS9. The data points are only every few weeks, but it doesn't look remotely like a consistent +4 to Polymarket.
It follows that "If not for Theo, there’s no reason to think Polymarket would ever have shifted from its usual regime" is also wrong. There is no "usual regime". Theo's trading pattern is murky, but the accounts identified by the WSJ were not active in the week before the election. My own sense is that while Theo was responsible for the price spiking to 67c on 30 October, he didn't contbribute much to the price on 4 November. There were other people making big bets, and perhaps the argument goes through if you substitute "half a dozen whales" for "Theo", but the narrative where this was all about one French guy is too simplistic.
I remain of the view I expressed in the hidden thread, which I'll copy here so other people can see it:
There's more information here than whether the outcome occurred or not. If Trump had won by a very narrow margin, the result would be the same, but we would consider that a 50/50 prediction looked good. In the events which occurred, we are therefore bound to consider that it looks less good.
Sometimes, you predict something has a 10% chance of occurring, and it occurs, but the means by which it occurs was sufficiently peculiar that in retrospect you are content the prediction was sound. This is the opposite situation. It doesn't look like something that just happened to fall on one side of a knife-edge.
Theres robust early voter data for recent elections with the ability to make fine grained comparisons. If anyone looked at a state like Fl or TX the differences made it almost literally impossible for any compensatory eday vote to bring the results level with 2020 . In the swing state there were more nuance but this is the main factor that made the betting market more red than the poll aggregate. The early vote data portended underwhelming result with certain demos.
The post awards 10 percent probability to „Theo knows better than the rest of the world“, 90 percent to „Theo doesn‘t know better“, and then updates from 10-90 to 12-88 based on the election result. Scenarios such as this one, real markets incorporating early voter data better, are given 0 percent.
Why did Trump's odds on polymarket decrease from 66% to to 56% on Polymarket from October 29 to November 3?
With the election coming into close view, too much dumb, mainstream-poll-following money moved into the markets for the smart money to keep it in check . . .
(Meant to be a joke originally, but who knows. My own first speculation was a different scenario from Buckwheatloaf's, based on the fact that we know Nate Silver's model swung slightly towards Trump temporarily, which obviously couldn't have been caused by Theo: perhaps real-money markets interpreted the slightly shifting polls as the end of the Harris honeymoon effect and momentum for Trump, therefore experiencing a bigger swing than Silver's model, which does not consider momentum, as Gergő Tisza documented to me in the Mantic-Monday post's comment section.)
A lot of it might have been due to this one Selzer poll, honestly: https://www.desmoinesregister.com/story/news/politics/iowa-poll/2024/11/02/iowa-poll-kamala-harris-leads-donald-trump-2024-presidential-race/75354033007/
This poll instantly became a hot topic for a lot of poll watchers online Many conservatives immediately mocked it, finding it unbelievable. But perhaps it gave some renewed hope to Harris supporters, perhaps leading to some deciding to take the plunge on Polymarket.
maybe mostly emotions due to the selzer poll and the perceived possibility that all the previous analysis (regarding early voter data or other data) was somehow wrong. ppl just decided to hedge their bets due to mixed signals. in retrospect obviously just a wave of fear. i dont think there was any big manipulation in the markets, more of a genuine moment of confusion and doubt. luckily for me i got really good odds on nevada due to that. (i felt the fear, but i managed to stay strong in my convictions).
I'm slightly skeptical large bet on poly could we set the odds because Isn't there a lot of sportsbook with high(er) volume? It is easier by magnitudes to send ur BTC into sportsbools and place ur bet in an instant than try to use poly market "buy shares" (convoluted and confusing to anyone habituated to sportsbetting). Wouldnt the people who set the odds across sportsbooks be placing large bets in the millions reverberate to poly.
I don't know if I'd ever blanket compare two prediction markets in that way. The users have different cultures and will bet differently on different issues, and each issue has its own considerations. I think a lot of people were trying to arbitrage between polls and Polymarket, which ended up being a weakness... But also, if Polymarket hadn't been that close to 50/50, perhaps the incentive wouldn't have been there for a billionaire to commission his own, more accurate polling. I'm conflicted on whether this is a win or not for prediction markets, and given the results of that polling was hidden from the pubic, I'm conflicted on whether all of this is good for society. Ideally pollsters would just do better.
Easy for me to say, sure, but it's consistent enough that I was able to make money on the market in question based on the prediction that A) pollsters will continue to underestimate conservative sentiment, and B) lots of people are arbitraging between polls and prediction markets. I don't know what you'd call that. Reverse arbitrage?
You could make money by both active betting like what you did and risk-free arbitrage at the same time. You'll get larger returns doing no arbitrage, but you'll do that by accepting more risk.
I said it in the other thread, but betting markets are nothing but another data point. That data point only holds water if it reaches a critical mass and even then is just another data point.
Americans have proven three times in recent memory that we absolutely suck at predicting elections. Why disparage more data points? Polling was incredibly bad at predictive power this go round (landslide + popular vote), anecdotes in my county were spot on, betting markets were accurate.
Anonymous controlled spaces like reddit were very inaccurate, anonymous slightly less controlled spaces like X were more accurate.
All of this negated by Socrates: the beginning of wisdom is knowing you know nothing.
We know nothing Jon Snow
Re Bayesian updating: I made a little simulation: https://x.com/nic_kup/status/1854471561782128747
> I trust the normal operations of the market.
I'm somewhat confused on what you think the 'normal operations of the market' are? A big player confidently buying up a short-term contract seems like a very normal operation!
prediction markets are meant to weigh the wisdom of the crowd by each participants confidence, and if Theo was confident for the right reason, his vote should be outweighed! The argument that it was difficult for some to participate on Polymarket seems to also apply to why Theo wasn't buying on any other markets - clearly there were some frictions applying in the other direction too
"mid October" is when early voter data began to trickle in. To correlate the shift in the markets to a lone bettor before that seems like an oversight =.=
The sportsbet site also shifted at this time not just poly. It certainly is insulated from barrier to entry but I thought it still shifted with the other books not against it.
I'm not sure how big a factor this is, but if your currency is worth more in outcome A then outcome B, then real-money prediction markets make A seem relatively more likely.
Polymarkets is, what, USDC? If we assume that's worth the same as USD, the S&P 500 went up after the election. If traders expected that to happen (I didn't), then a Trump victory being "overpriced" is what we'd expect, though I'm not sure how much overpriced - looks like the index is up about 3% but idk offhand how to turn that into prices on prediction markets. (Supposing the market thinks there's an X% probability of a Trump victory, and value of USD in a Trump victory versus a Harris victory is Y% higher, we can probably get the market prices - but we know neither X nor Y, and surely we can't figure out both of them from the market prices. Maybe we can get plausible bounds?)
And betfair and smarkets are in GBP, where it looks like the FTSE all-share went up the day after but then back down, and I dunno how to interpret that.
At any rate, this is one theoretical reason to expect play-money markets to be more accurate.
Price is an indicator of how much value the buyers and sellers place on a thing.
If something is offered for sale, and the offer is taken, by definition the thing is not "mis-priced." Value is subjective and differs from person to person, which is what makes exchanges desirable.
The value of "Trump wins" shares on a prediction market was in the buyers' and sellers' expectations, not in how the bet resolved. Post-transaction regrets, absent force or fraud, are irrelevant to price as a marker for value.
My issue is that there is a serious disanalogy to the coin case. In that case, we're trying to adjust our credences to match some chancy disposition of a coin. There is, in that case, a "true" probability, and we want our credences to reflect that. In the election case, it's a mistake to think of the election as truly random binomial distribution with p = 0.5. Now that Trump won, I believe with something like 99%+ probability that if we reset time to the second the polls opened, Trump would win again, and again. On the p = 0.5 theory, if we reset time, he'd win half the time. No, we learnt the actual, non random state of the electorate. In such cases, it seems to me that higher credences (i.e., polymarket's higher odds) are very good for their theory of the case.
Analogy: You have two expert paleohistorian friends. One is 50% confident in asteroids-killed-the-dinosaurs theory and one is 90% confident. You can't tell who to trust, but the 50% gut has a general better track record. Ultimately, you come down (due to your ignorance) with a 50% credence. If the asteroid theory is subsequently conclusively demonstrated, I would *massively* defer (with no other information, and with no expertise of my own) to the 90% guy's general understanding of that whole situation.
Totally agree that a sample size of 1 (N=1) does not cause us to update much a prior probability that was around 1.
Even if Theo was consistently beating the markets by making lots of bets and winning in the long run, he still could have lost on Trump by random chance. (I sometimes do a bit where I argue that DARPA should derisk scientific research funding by diversifying grant money across multiple research projects in the same basic area; DARPA does, in fact, do this). You have to split your money across a lot of smaller bets each of which has > 0 expected utility.
There is a choice of different null hypotheses here ....
but my personal null hypothesis is that Theo, the degenerate gambler, correctly identified a mispricing in the prediction market and exploited it.
There’s not really any objective sense in which the shares can be mispriced, because there’s no “real price” to compare to. To the extent that there is, the “real price” was 100 cents, since that’s what they ended up being worth.
I think you mean something like “the best knowable guess for Trump’s victory chances was 50%”, but “knowable” is doing a lot of work there; an omniscient being would’ve probably put Trump’s victory chances at almost 100%.
I don’t think you can score particular probability estimates; you have to score over many events to get something meaningful.
It seems odd to pick a case where you ended up being wrong and say “no actually ex ante I was right here”. There’s not really a definition of “right” that can support this, and you just end up in a tangle of excuses. The point of being “right” is to make correct predictions, but this prediction was wrong.
Exactly. There are only correct prices - and this correct probability estimates for outcomes they haven’t happened yet - if there are correct values.
I think there are correct values, which I get makes me a minority here, but it’d worth nothing that “there are no true values” appears to be a phenomenon only in the 20th century and (in my opinion) the cause of so much of the chaos in the world today.
How do you determine the correct values, even in retrospect?
Just because there are correct values doesn't mean we know, or can know, those correct values. A particle has both a position and momentum at any given point in time, but you'll never know both, and arguably will never know either depending on how rigidly we define a "point".
> the cause of so much of the chaos in the world today
The world has always been chaotic.
>I think there are correct values, which I get makes me a minority here
It might also be the case that there are correct values, but, in most realistic situations, it is unrealistic to expect to be able to calculate them.
Look at what happened in the lab leak v. zoonosis debate. For a _retrospective_ analysis, on a _past_ event of the virus coming from a lab or from a wet market, with both presenters doing their best to present all of the evidence they had, there was still a factor of 50 difference in the odds calculated by the judges (even excluding the presenters).
> I think you mean something like “the best knowable guess for Trump’s victory chances was 50%”, but “knowable” is doing a lot of work there; an omniscient being would’ve probably put Trump’s victory chances at almost 100%.
I think you're leaving out "as of October". I think it's perfectly reasonable for someone to make the argument that an omniscient being might put Trump's chances, relative to October, at below 100%, and if we argue that "no, the omniscient being would know the future" then you'd also be trapped into arguing "this same omniscient being would, in October, place the odds of yesterday's Powerball drawing being '12 17 37 58 62 [4]' at 100%, ergo the oddsmakers were wrong in saying it was a 1 in 292.2 million longshot", which is just silly.
> It seems odd to pick a case where you ended up being wrong and say “no actually ex ante I was right here”. There’s not really a definition of “right” that can support this, and you just end up in a tangle of excuses.
Of course there's a definition, and it's trivially simple if you actually understand the actual arguments actually being made in actuality.
If you say "the odds of a standard Vegas roulette wheel landing on Red are 50-50", and I say "no, I believe the odds are actually 47.37-52.63", and we go to Caesar's Palace and spin a wheel, and it comes up Red, I was still right.
> The point of being “right” is to make correct predictions, but this prediction was wrong.
Then by your definition, the insane man on the Strip who told both of us that the wheels at Caesar's Palace have a 100% chance of landing on Red because that's the Lizardman's favorite color was right. He was not right. His (implied) prediction came true, but he was not right. That's not how right works.
Is it worth some discussion about what it means for an election market to be accurately priced, and relatedly about where the sources of randomness are?
Particularly when contrasted with a sports match for instance. If we hypothetically say that a football match is perfectly priced just before it starts, then that is the best we can do obviously, by definition, and most of the randomness is yet to ‘take place’, in the goings on of the actual match. If we repeated the game 100,000 times we would see results that agree in frequency with our pre-match ‘true price’. In this sense, the result of a single occurrence gives very little information about the price.
What would it mean for an election to be perfectly priced on the morning of the election? By this time, it would be reasonable to assume that most people have made up their mind on which way they intend to vote. As such, this source of randomness has been largely removed. In a certain sense, aren’t the voting numbers already largely ‘decided’ by the morning of the election? And most of the randomness left comes from our inability to know and sample this perfectly. But in a hypothetical world, the one in which the ‘true price’ lives, we are aware of everyone’s voting intentions in the morning. Shouldn’t this be taken into consideration when we talk about accuracy of the prediction markets? What do you mean by the statement that you now think there is an 88% chance Meticulus was more accurate?
If we repeated the day of the 5th November 100,000 times, wouldn’t it be very likely that Trump wins a very large proportion of these elections. In this sense, doesn’t a single occurrence give much more information about the price than a single occurrence of the football match. In it, something is revealed to us that was in a certain sense already true (what the state of play was at the start of the day).
Therefore, couldn’t we argue that the true price on the morning of the election of Trump winning is much closer to 100%? (Of course there is still a certain amount of randomness from those undecided, people actually making it to the polling station etc). Therefore, how can we argue that a market estimating, on the morning, the likelihood of a Trump win at 50% is more accurate than one that had it at 60%?
"If we repeated the day of the 5th November 100,000 times, wouldn’t it be very likely that Trump wins a very large proportion of these elections."
That's why Nate ran all the simulations:
https://www.natesilver.net/p/a-random-number-generator-determined
"At exactly midnight on Tuesday, I pressed the “go” button for the final time on our election model this year. I knew it was going to be close. I felt like I was spinning a roulette wheel. (Appropriate, I guess, in a year when I published a book about gambling.) We’d decided ahead of time to run 80,000 simulations instead of our usual 40K.
And after 80,000 simulations, Kamala Harris won the Electoral College in … 40,012 of them, or 50.015 percent. The remaining 39,988 were split between Trump (39,718) and no majority — a 269-269 tie — which practically speaking would probably be resolved for Trump in the U.S. House."
Maybe he should have run the 100,000 after all! Reality - how rude of it not to line up neatly with the beautiful mathematical models!
I'm not mocking Nate Silver here, by the way; he's been one of the best at covering this entire election and trying to parse out what polls are really saying, what bias is there, what weighting should be given, etc. And he did say that in the end, the result comes down to people going out and voting, and not what the polls predict.
But Nate's model is built on polls. There's fundamental uncertainty baked into it.
If the world woke up on Tuesday, November 5th 2024 100,000 consecutive times, lacking any 4th-wall-breaking awareness of their predicament, I think it's a safe assumption that far more than 50,000ish of those Tuesdays would end up with Trump winning Wisconsin.
Nate's model is based on very limited information. If we hopped in a time machine to 12:00 AM November 5th 2024 80,000 times, we would watch Trump win ~80,000 times in a row because in reality voters had already (and would already, based on the information they were fated to consume that day) decided their votes, and the votes would lead to a Trump win every time. The true value of a Trump win contract was ~100c and a Kamala win contract was ~0c, and in that sense everybody mispriced it heavily down, but Polymarket less so.
A little unfair. You can't tell from one guy making one big bet and winning if he's right or just lucky, so saying that the market was overpriced is the correct view. That's not mere cope.
I don't have the shining faith in prediction markets that others do, but at least this way they're getting a trial run and seeing how things can go wrong, before anyone tries using them as "let's set government policy that will affect the real world running of the country by the prediction market! what could possibly go wrong?"
See above for how it can do an oopsie 😀
Metaculus is the least calibrated market ( https://calibration.city/ )
His priors don't seem to be based on anything, trying to prove anything from predictit is silly, they have a betting limit per trader (which implies that whales are better for market accuracy...)
Market was right, Scott was wrong. he retroactively sets unrealistic priors so he doesn't have to update. My prior that my uncle is a fish is 1.00
It's so frustrating, I learnt about prediction markets from him, long brilliantly written essays how they work, why they work etc. and then in their biggest test, they were right! but because he assumed that it should be more like 50/50 we can now throw everything out the window
No, you are wrong.
Scott says what he bases his priors on (previous competitions), so its clearly not true his priors are based on nothing.
Furthermore, using calibration city for this is silly as it does not use the same set of questions for comparison. Manifold has thousands of "daily coinflip" and similar markets, it is very easy to reach perfect calibration on a coinflip (just predict 50%), while metaculus only has "interesting" questions (ie. A random user can't just make some easy to correctly predict question).
Personally, I use both Metaculus and Manifold and because I have a consistent profit on Manifold, I feel qualified to say that Metaculus is better.
Metaculus also has 1/10 of the number of bets analyzed. Compared to others. No wonder it is “less calibrated”.
I don’t know if Metaculus is accurate or not. My gut feeling is that the money markets are more accurate. But I do think polymarket in October was wrong.
Frankly, setting government policy by prediction markets is a dumb idea; they're only one tool among many. Of course IRL deep-pocketed interests would try to manipulate it.
I think the biggest real upside of futarchy is you can make (in the appropriate company) jokes about how there's no real argument intersex people are any better at government than anyone else.
('Futa' is also short for 'futanari', a genre of hentai involving people with external genitalia belonging to both sexes, often in anatomically impossible ways.)
My well of English has been permanently defiled, to quote "His Last Bow", by early exposure when I did get on to the innerwebs to the less savoury parts, hence every time I see "Futarchy" my mind automatically goes "Isn't that the Japanese porn thing about?"
Really do need to come up with a different name for it!
In Japan, 'bukkake' is a way of serving noodles.
Manipulation would attract more smart money betting against it. Such "dumb money" thus helps: https://statmodeling.stat.columbia.edu/2024/10/25/prediction-markets-and-the-need-for-dumb-money-as-well-as-smart-money/
You can only say it's "overpriced" if you have some standard other than the actual result, and I don't think you can have such confidence in them.
Not sure how many years you waited to make that comment, but banned.
"(T)he amount of money someone puts into Polymarket is exactly proportional to their trustworthiness. This is a fair approximation for the Polymarket algorithm to use; it’s the assumption that drives prediction markets"
I think this is the problem right here. Prediction markets (all markets, really) are not working off *trustworthiness*, they're working off *confidence*. Theo was very confident that his bet was correct so he put a ton of money on it.
People can be very confident that their prediction/bet is right, and put a ton of money on it, and lose their shirts. We know this from ordinary gambling and betting. Dressing up prediction markets as "one of the single best sources of truth" is what is blinding rationalists here.
Bob has a lot of confidence in his bets (let's call them that, if we're putting real money with them, instead of predictions) so he puts a lot of money on them. All the market can tell from that is that Bob is very confident. For the market to judge Bob as *trustworthy*, it (and we) would have to follow the pattern of his bets over time to see if Bob is generally right (better than chance) if he bets on "X will happen, Y will not happen".
If Bob gets a lot right at better than mere chance levels, then we can say Bob is trustworthy if he thinks Larry the Cucumber will win the presidency in 2028. If Bob makes one big bet, wins big, and peaces out, we have no idea if he's smart, or just lucky.
Theo seems to have had a process at work for his confidence that people were underestimating Trump's chances and overestimating Harris' chances, more than just going on a gut feeling.
But the problem again is that "oh but won't the smart money correct for the one big bet?" is that the people *do* think they're smart, so they *overthink*. "Hmm - this guy must be an idiot, why is he punting so hard on Trump? All the polls say it's way too close! Unless... does he know something I don't? Has he got inside info or a really stunning algorithm here?"
And since people on prediction markets tend not to be the average punter, they do tend to assume "other people on this market are running Bayes on all their decisions" and so they follow the money. If Theo bets big, maybe I should be betting big too. Even if it's contra the run of the polls, and maybe *especially* if it's contra the run of the polls, because prediction markets are for smart people and Trump supporters are not smart people, all the polling done about who is the most educated shows that! So he can't be a MAGA-head, so he must know something!
Human nature is a wonderful thing.
As for trustworthy smart people making predictions, Ann Selzer seems to have blown up her own reputation with her Iowa poll. Nate Silver was diplomatic about it, because plainly she is a big name as an expert whose polls are highly accurate for her area of expertise (how elections go in Iowa), but even he had to go "Nah, this is nuts":
https://www.natesilver.net/p/a-shocking-iowa-poll-means-somebody
"Releasing this poll took an incredible amount of guts because — let me state this as carefully as I can — if you had to play the odds, this time Selzer will probably be wrong. Harris’s chances of winning Iowa nearly doubled in our model from 9 percent to 17 percent tonight, which isn’t nothing. Polymarket shows a similar trend, moving from 6 percent to 18 percent after the survey. But that still places Harris’s odds at around 5:1 against.
The poll has a reasonable sample size: 808 likely voters. Still, the margin of error for the difference separating the candidates in a poll of that size is ±6.6 points. That means in theory, in 95 out of 100 cases, the “real” number should be somewhere between Trump +3.4 and Harris +9.6 if Selzer had surveyed every single Iowa voter instead of just an 808-person sample.
The theory, however, doesn’t capture all possible sources of polling error — especially the fact that the overwhelming majority of people who pollsters attempt to contact never complete the survey. And we ought to be good Bayesians here: even if Harris has a very good night on Tuesday, winning a state you lost last time by 8 points is a big ask."
https://eu.desmoinesregister.com/story/news/politics/iowa-poll/2024/11/02/iowa-poll-kamala-harris-leads-donald-trump-2024-presidential-race/75354033007/
"A new Des Moines Register/Mediacom Iowa Poll shows Vice President Harris leading former President Trump 47% to 44% among likely voters just days before a high-stakes election that appears deadlocked in key battleground states.
The results follow a September Iowa Poll that showed Trump with a 4-point lead over Harris and a June Iowa Poll showing him with an 18-point lead over Democratic President Joe Biden, who was the presumed Democratic nominee at the time.
“It’s hard for anybody to say they saw this coming,” said pollster J. Ann Selzer, president of Selzer & Co. “She has clearly leaped into a leading position.”
Results in Iowa?
https://eu.desmoinesregister.com/story/news/politics/elections/2024/11/05/election-2024-republicans-expect-donald-trump-win-iowa-over-surging-kamala-harris/75720978007/
"Former Republican President Donald Trump has once again cruised to victory in Iowa, tallying his third presidential win in the Hawkeye State — this time over Vice President Kamala Harris.
In the end, it wasn't close.
The Associated Press called the race about 9:40 p.m. Tuesday. According to partial, unofficial results released by the Iowa Secretary of State late Wednesday morning, Trump led Harris 56% to 43%."
In conclusion, nobody knows nuthin', I wish I had gone with my gut and punted on Trump, and stick with the established bookies if you want to win big.
I think Theo knew something. I hadn't heard he commissioned a private poll until recently, even though people had been talking about the French Trump whale earlier.
So what would people do to invest for the second Trump term? I'm thinking buy a house, since the tariffs (and deportations) are likely to cause inflation, and stocks over bonds, but I'm sure people here have some more thoughtful ideas. Energy? Finance? Small-caps?
A good bet if you think tariffs will be broad-spectrum would be to buy stock in companies that produce things in the US, as long as they also buy their raw materials/parts in the US. Their costs won't increase, so their pricing should get better relative to the competition, increasing sales and thus profits.
I have no particular companies in mind, though. We import a lot of stuff, and commerce is complicated.
Elections are not stochastic experiments. Market prices are not probabilities. The analogy of "voting with dollars" is more apt. If Joe's Exxon has gas at $3 on the corner of Fifth and Main, but Sally's Mobile has gas at $1.50 on 30th and Main, is anyone prepared to make a comment about the probability of finding gas at $2 at the corner of 2nd and Main?
Market prices give us information about what other people think the value of some good is. Mr. Market, as Benjamin Graham called it, has a mind of its own. Mr. Market combines the dollar votes of people into bids and asks. If someone comes along and says "the market price is wrong," my question is "compared to what."
Every day any number of people buy and sell stocks on the world's exchanges with a belief (or is it just a bet) that the price of some company is "too high" or "to low." Who knows how much of the action is "smart money" vs "dumb money"? Mr. Market doesn't know and doesn't care. Every dollar is just a vote. How are prediction markets any different? The more dollars voted, the greater weight the vote carries. That's the top, bottom and both sides of Mr. Market.
The Efficient Market Hypothesis proposes that market prices reflect all the information available and do so really quickly. Are prediction markets efficient? "Compared to what" is the only question that makes senses to me.
Scott, your coin flipped 3,000 times, and came up heads at least 2,000 times. It’s clearly not 50:50.
A good test for whether it was a fluke is looking at county-wide differences from the 2020 vote. If it truly was 50:50, we should have seen noise. But instead we saw a sea of red, because almost whole country shifted rightward.
I think that’s good evidence that the whale understood reality better than the play money markets, and the true probability of Trump victory was higher than 60%.
Yeah you clearly do not understand the math. County results are highly correlated.
True, but are they *that* highly correlated? I guess the key question is: "in what percent of Nate Silver's simulations were the county-wide differences as skewed as the actual result?"
Yes, they are very highly correlated. Why wouldn’t they be?
I feel like you're answering the wrong question. It is not "given the actual votes on Election Day, what are the odds of Trump winning?" Those are obviously 100% (minus whatever negligible amount you'd like). The question is, given the public and semi-public information available prior to November 5th (i.e. polls, analysis of the fundamentals, the method Theo was using, etc.), what are the odds of Trump winning? In other words, if you can imagine an equally likely outcome that looked the same beforehand but shifted the other way, then the amount and uniformity of Trump's win in practice does not really matter.
Another minor quibble is that since the Democrats' Electoral College advantage diminished by (I believe) about 3 points between 2020 and 2024, and the Ds won in 2020, a relatively uniform rightward shift is not incompatible with odds of 50-50.
I agree with your overall point about real-money markets doing weird things because arbitrage is harder.
FWIW I bet about $1500 and made about $300 as of this writing on Predictit. Last time I bet about $1000 and made about $100. I couldn't really use Kelly because I wasn't sure what my actual chances of winning were--I was making small bets and moving sides accordingly as news came in (same as everyone else on that site, I imagine), so stopping to record it would have cut into my winnings--I was arbitraging between 'Kamala next president' and 'next president is a woman', for example, which doesn't stay different long and often gets negated by the difference between buying and selling prices (I believe real traders call this the bid-ask spread?). I'm sure one of you folks could write code to store all movements or something.
I suspect the money on Kalshi's much smarter and I am not going to do well so I am not planning on a side hustle as a trader--something like 98% of people lose money on that. (Also, as you may have guessed, I have something of a problem with excess risk aversion.) But I thought people might find the experience interesting.
I couldn't bet on who specifically would be the next President on PredictIt because they weren't taking more bets there, so I had to bet on whether the next President would be a woman (which was priced multiple cents lower than Kamala).
Wow, I remember that from the last election. I think I was already in it at the time.
If I were starting out I'd probably go with Kalshi. I had $1000 in Predictit already (sunk cost) and the number of markets was small enough I could monitor them all. I also figured there was enough dumb money I could still have a small edge; Kalshi was going to be full of smart rationalist people with software they coded to calculate the optimal price based on AP feeds or something.
"Suppose you have a coin. You think there's a 90% chance it's fair and a 10% chance it’s biased 60/40 heads. Then you flip the coin and comes up heads. What should your new probability be? You would solve this with Bayes’ Theorem; the answer is 88% chance it’s fair, 12% chance it’s biased."
This is flawed, because Trump did not just win, he won by A LOT.
Prediction markets didn't give the odds of Trump winning by a lot as opposed to just winning, but we can infer the odds they would probably have given, from the odds they give for Trump just winning.
Imagine there's a guy who represent Polymarket and another who represent the others. If we had asked them what are the odds of Trump winning by this much (again, as opposed to just winning), we would get answers that differ greatly, because of how bell curves are shaped.
Not a small difference like 50% as opposed to 60%.
Therefore I think the outcome vindicates Polymarket by a lot more than 2%.
Was there a market for "Trump wins the popular vote"? That would seem to be better calibrated on the size of the win than just "Trump next President".
Even better would be the market for "Trump wins popular vote by >3 points"
If you think Trumps wins the popular vote by >3 points, you should bet on it. There's still time:
https://polymarket.com/event/popular-vote-margin-of-victory-in-presidential-election
There were definitely sports betting markets for it, I remember watching that particular one change over the course of the night. I don't know where it started before counting began, but it went from a reasonably long shot to a dead certainty as more votes were counted.
I think this post illustrates exactly why the whole prediction market idea disgusts me: like most of capitalism, it exists in a world without any connection to morality and is only considered within itself and the milieu of money, regardless of the world actions that it is reflecting. Bleh.
I know you (Scott) love it and are way into the idea of future prediction, but honestly every time I read one of your posts about betting on prediction, it somehow turns my stomach.
But what do I know, I'm just a penniless old guy who seems to have nothing in common with people who have and play with money, more and more as the years go by. I guess your world is entirely different.
For example, I read this: "After Republican candidate Donald Trump won the U.S. Presidency, the price of Bitcoin saw a sharp upward move that saw it hit a new all-time high at $76,482 as market optimism over the potential introductions of a friendlier regulatory environment in the U.S." and I realize that the "friendlier regulatory environments" are exactly this same playground, and also exactly what I believe are responsible for huge wealth disparities in the world, leading to scarcity, fear and ultimately to autocratic governments. Yuck.
Wealth disparities are caused by the existence of huge amounts of wealth. If no wealth existed, there couldn't be any disparity. The poor in a wealthy country like the US are far wealthier than the poor of the Third World, which is why people all over the world want to move to the US. Rather than scarcity, there is increasing abundance (hence the prevalence of obesity, which is now being fixed by semaglutides).
What's immoral about it?
Ignoring the entire context of the human society in which it exists.
When I sleep I also ignore the entire context of human society. Is this also immoral? This question probably sounds like a lame gotcha to you, but it is absolutely honest: I have no idea what is the meaning of your previous comment, and I would guess a lot of of other people are also confused, so feel free to elaborate for more clarity.
I'll try. What I mean is that playing with money regardless of the context (in this case betting on election outcomes that will have global affect) separates the monetary aspect of it from the greater sphere of the subject, which can only insulate the players from being effected by the good or bad that comes from it. In this particular example, I see it as the same as, for example, investing in companies that have detrimental environmental effect upon the world but being happy about making money, disregarding that these companies and indeed the entire stock market have an adverse effect on wealth disparity overall. Or perhaps like betting on dog racing, disregarding the well-being of the dogs themselves.
I'm both jet-lagged and extremely concerned about global repercussions to the *situation* that these bets are riding on, so perhaps emotional, but seriously: my reaction to reading the post was nausea.
I find that those who play with money almost always disregard the very fact that their "play" is made available to them by a constant inequality in wealth among people, and most people ignore the context in which that wealth exists in the first place.
If that doesn't explain why I posted this, I'm not sure I can.
It sounds like you don't like Donald Trump and are upset that some people are making money out of his win?
I've had the same reaction to most prediction market posts.
Seeming example of this: https://x.com/RichardHanania/status/1842940892048703513. It's easy to gesture at issues without even roughly quantifying how those issues override the benefits. While it's conceivable that betting markets affect outcomes negatively since perhaps by introducing valuable information, someone could make a decision on its basis, that's ultimately just an argument against information! Sure, a campaign could adjust its practices in light of prediction market behavior, but the same goes for punditry, inasmuch as the punditry contains a useful signal.
Similarly, the follow up comments note that capitalism has led to undesirable outcomes like income inequality. Sure! You can't have much inequality, when everyone is a subsistence farmer. Everything is bad for someone (vaccines are bad for business, if you're a coroner). And it's trivial to string words together alluding to the downsides of anything, without making a meaningful argument about that thing actually being bad.
It still hits me as crass.
Hi, I think you'll enjoy my YouTube channel. I won't share it here so as not to spam but I get exactly how you feel. Clicking my avatar will get you to my substack to my youtube. Check out especially those videos that are more than a month old. I engaged in a serious experiment and the results are such that I'm undergoing a radical transformation and my future videos or whatever are likely/hopefully to be far less moral polemics with practical intent. The experiment didn't result in the vindication of goodness and truth that I preferred it would but I am very happy to have done it so thoroughly that I can now rule it out. I have something much better and more fun in mind.
You may be able to learn things from my experiment too If you do I would be pleased to hear from you.
Making an honest profit is about the most moral thing in the world.
Frankly, I'm not certain that all economics isn't a zero-sum game. Profit implies loss. Maybe there is no such thing as "honest" profit in any situation, someone or something loses. For example, in extraction-based capitalism, they don't even count the damage to the environment as a "cost", hence it's all "honest profit." Similarly much of globalization has meant more diesel ships at sea, worse working environments *somewhere*, etc etc, for that "honest" profit.
You don’t understand producer surplus or consumer surplus or efficiency gains and tech advancement from incentives and competition.
Trade offs and externalities are a thing, but guess what, damaging the environment predates modern capitalism.
"Frankly, I'm not certain that all economics isn't a zero-sum game. Profit implies loss."
Consider the amount of wealth in the world 50, 100, 200, and 1000 years ago, and the amount of wealth today. Where are you thinking all that wealth came from?
as above: "Trade offs and externalities are a thing, but guess what, damaging the environment predates modern capitalism."
edit in an afterthought: where do I think all that wealth came from? off the back of people and places that had no say in the matter and got nothing in return. It didn't come from nowhere.
>where do I think all that wealth came from? off the back of people and places that had no say in the matter and got nothing in return.
That is _spectacularly_ wrong. Almost all of the increased wealth from the last few hundred years came from technological innovation, using the _same_ ores and land and worker-hours to fulfill more human wants and needs.
I don't understand. "Same ores and land"? They aren't being recycled? The increases in extraction and harvesting of environmental resources, especially in the last few hundred years, are now causing environmental and climatic catastrophes, causing more human suffering at the trade-off of more monetary wealth. There may be more people who have more money, but there are increasingly more people in general, which means more suffering despite the apparent wealth of "things". You're saying that wage workers now are better off now than slaves, sure, but having an iphone doesn't make a person "better off", just a different level of slavery. (Ok, that's hyperbolic, but you understand.)
I mean, I know Hans Rosling's work, but his work itself is paradoxical—he tracks the decrease in bad and increase in good, but ignores increases in bad. Just saying that tech innovation brings up the overall level of human wealth ignores the environmental and psychological costs.
Yup https://westhunt.wordpress.com/2017/08/03/chained-nixies/
> Profit implies loss.
Er, no it doesn't. This is one of the most fundamental things in economics.
Ok, I'm dumb then. You can all go to bed happy.
Profit does not imply loss. The most basic form of profit is, a farmer with a tiny bit of land and a kernel of corn plants the kernel, nurtures it for a season, and harvests several ears of corn. He has made a profit. There has been no loss.
If you insist on money changing hands before it counts as "profit", the farmer sells half the corn for three obols. He has made a monetary profit of three obols. He has made a profit of three obols. And someone else has bought dinner. There has been no loss. At some point the farmer will spend his obols on a spool of yarn. Now the yarn-seller has made a profit, and the farmer has bought the makings of a new pair of socks. There has still been no loss. Only the gain of some corn, some yarn, and whatever the corn-buyer made and sold to get the money to buy the corn. The world, and everyone in it, has become richer.
It is possible for one person to profit from the loss of another. It is not *necessary* that he do so, and by blinding yourself to all other forms of profit you miss what the rest of us are trying to do here. Which is, for the most part, trying to find a way to buy accurate predictions of important future events. That's profit, and it requires no loss.
This is an example of what I was initially posting about, what you are describing is only the monetary value of the things involved in an entire system. The farmer's time and effort (and I don't mean time=money, but time as experiential, an element of their life—and not labor as work=money, ie wages, but labor as effort that might be applied to other usages. Maybe the farmer really was only happy when they danced, but couldn't due to the needs of raising the crop), plus of course the water, the land, the entirety of the environment involved that was being changed to a crop bearing strata.
There are enormous amounts of things "lost", things that are never counted in the economy of money. The economies of the "useless" (for example the arts) make huge impacts on the qualities of life. That is the context I am speaking of here.
I think this is why I wrote that I "am not convinced" that economics—as a gestalt— isn't a zero sum game, resulting in equilibrium like any other physical process. Discounting all variables that aren't able to be quantified to dollars *at the moment of transaction* ignores a great deal of the overall system. It's well known that extraction-based capitalism ignores the externals, like the environmental costs, but they come back to haunt us in time. Beyond what I mention above regarding the farmer, there is the use of land over time: if the farmer plants corn over and over, the land ceases being arable due to the monoculture sucking out nutrients. I feel like the entire game of money ignores so much else.
->This is also why I flinch at these posts on prediction markets. "Will somebody use a nuclear device in 2025? Yay! I win some money!"
If the "loss" you're concerned about is that we can't all spend all our time dancing or whatever and having food, clothing, shelter, etc, miraculously appear for us, then I don't think that's a useful framing. You can't lose what you never had, and we never had those things. Well, not since we were kicked out of the Garden of Eden, I suppose, but it's hard to blame that one on the profit-seeking capitalists.
You're sort of ignoring the environmental impacts mentioned.
Also: we did have those things. As children. (At least I hope we all did/do).
I don't really bet on or use prediction markets, but I still find them useful. I mostly comment and read forums in fairly Democratically-aligned spaces, and I saw lots of information last week that was hard to gauge in terms of its accuracy. People were spending hours dissecting early vote results, the Seltzer +3 poll, etc etc. Some people were predicting a Harris landslide. See here [1] for a typical example of this. Whenever I saw this, rather than trying to engage with a complex statistical argument, I looked at 538 and PredictIt, and found that the odds were still about 50/50. An enormous amount of election coverage is essentially bullshit that doesn't matter or isn't predictive. This is why I find it useful: as a summarization of what's happening. I think it's important to separate the message and the messenger here.
[1]: https://app.vantagedatahouse.com/analysis/TheBlowoutNoOneSeesComing-1
Agreed! I, also, am just an onlooker towards the markets, but I find it very useful when someone is claiming "X is inevitable" or "Y is impossible", to look at the markets and be able to say: Well, this market says the odds of X are 40:60 and that market says 55:45, so people who have skin in the game are betting reasonably close to 50:50. As you said,
>An enormous amount of election coverage is essentially bullshit
I'm not sure I agree. If the election indeed would have been close (as in 2020), there would indeed have been no way to predict it over chance. But this was a shellacking. There were signs that this was coming, and there was presumably a way to discern those signs from the Kamala noise.
I'm not sure a biased coin is the right model for an election. The main characteristic of a (biased or non-biased) coin toss is that the physics of its flight path are so complicated to compute that even a human with a computer cannot predict its outcome.
But surely, an election result is in principle more predictable than that? With an expensive effort you could send people into Iowa, to get to know the people there and understand their values and beliefs and predict successfully how they will vote.
I'm not saying Théo's efforts were on this level, but I don't know, one is at least on shaky ground when criticizing the man who commissioned his private polls and who turned out to be right. I don't find it so unthinkable that the incentive of winning $30 million made him do a better at getting good polls than anyone else. (Hmm, though what about Wall Street? Wouldn't they also have a large monetary incentive to find out who will win?)
Also, I fully agree that Polymarket didn't have *enough* sophisticated traders to really trust their odds, but given that the most sophisticated trader made it go towards Trump (the winning outcome), it could well be that with more traders of higher sophistication, the market would have been even more skewed to the winning outcome, because, again, I would argue that an election outcome is something that is knowable, as opposed to the outcome of a coin flip.
This is a general way to model how much information you do and don't have - with a fiar coin you can't (barring an implausible level of physics modeling) reduce entropy below one bit of uncertainty, but with an election good polling and modeling often can - but still not to zero (no one predicts every election exactly right, there's always unknown factors).
"But surely, an election result is in principle more predictable than that? With an expensive effort you could send people into Iowa, to get to know the people there and understand their values and beliefs and predict successfully how they will vote."
This strikes me as about as impractical as solving the Navier-Stokes equations and setting up an elaborate diagnostic suite of flow-visualization and -quantization instruments to determine which way the coin will flip. In principle, at least in the classical limit, the coin-flip is absolutely deterministic. In practice, it involves forces that we can only assess collectively and statistically.
The same is true of elections, with any plausible level of election-forecasting resources.
I know Scott says the other big money markets were arbitraged which is why they were closer to Polymarket than PredictIt was. However they didn't seem to be arbitraged very well. At times after the French trader Theo entered, there were extended periods where there was at least a 4% difference between Polymarket and Betfair. How is that big of a spread possible? Commissions can't be that high can they?
Bertie has a 2% commission on winnings, Polymarket doesn’t have any, but there are costs in moving money in and out of crypto. There is also currency risk and opportunity costs of locking the money up in bets, increasing the further before the election you place the bets.
I believe the markets on Polymarket did correct on election day, completely removing the Theo effect. The popular vote market was trading at 73-74c Kamala the afternoon of election day, and the president market was trading at 39-42c.
I will say that Polymarket has always had a Trump bias, so many traders on Poly were expecting a Trump surge near election day. On November 2, the election market was trading at around 45-55, then on Monday and Tuesday the Trump surge happened (Theo barely traded on Monday).
Since Theo was actually correct on everything, I would insist on scare-quotes around "correct" for any bet against him.
An election isn't like a coin flip. It's not a binary yes/no. The fact that Trump didn't just win but swept all the swing states should increase your trust in Polymarket a lot more than if it ended up being a close win for Trump.
I don’t think this is right even though it’s intuitive. swing state sweep was among the highest probability outcomes for the Silver model, because it expects correlated polling error like we saw. Very impressive sub-prediction! Overall probability was still 50-50.
Isn't it obvious in hindsight that such an error was more likely to occur in favor of Trump than Harris? To me, the Polymarket betters seem smarter than Silver for having predicted when he didn't.
I don’t think it was obvious ex ante. Pollsters claimed they were trying to correct for their repeated Trump underestimates, which were embarrassing.
What odds did Silver give to a Trump swing state sweep? I know manifold gave 22% for a Trump OR harris sweep.
just looked it up. 20% Trump sweep, highest probability. 13% Harris sweep, next highest probability. More impressive than I thought!
Very cool
AFAIK he didn't give odds for a swing state sweep, but the single most likely outcome according to his model was Trump 312 - Harris 226 (which matches the NYT's current projection for the final result).
The night before, it had about a 28% chance of popular vote victory and 20% chance of every swing state going Trump. Both of these came true. However, these numbers are basically the same as Nate Silver's modeling.
I think this analysis ignores the "wisdom" part of the wisdom of crowds.
When you looked at something like Nate Silver or the RCP polling average or 538 you would come to a number around 50%. But those numbers were only based on polls.
Trump had beaten his poll numbers twice already, Silver and others assured us that this was a random coin flip and could just as easily go the other way for Harris. But there is no real reason to believe that. The data on polling errors is quite short and decidedly mixed and does not control for the fact that usually the candidate changed
I think introspection tells you the errors with Trump would not be completely fixed. When I model they type of person that works for a polling firm I don' think of a person who makes big adjustments to methodology but instead makes gradual adjustments. This type of data was outside of the "numbers".
My view is that you had to start with the Silver numbers as a baseline and shade them towards Trump. I was 60-40 Trump for most of the time leading up to the election. Polymarket over shot that a couple of times and I would have bought Harris but would up right at that number on election day.
A lot of this analysis falls apart when we look at Kalshi, which is US-regulated, open to entities like trading firms, and has done nearly $1B of election market volume in the last month or so. Kalshi's presidential market orderbook in particular has been consistently deep and traded entirely by US individuals *and entities* with deep pockets and high confidence (SIG being perhaps the most widely-known market maker on the site). As much as I like the idea that these are niche hobby markets for wealthy nerds, it's not really the case anymore. We're really entering a new age of prediction market accuracy.
Interesting reading this, just a completely different mindset and view of the world to my own.
My strong default would be the real money markets would be far better as in every other area of life, putting money on the line forces people to be more rational, and those that aren’t don’t generally last very long (poker, hedge funds etc). So very surprising to see research showing non money markets doing better. My best guess is this might be when the money prediction markets were very small and illiquid, but incredibly hard to see something outperforming markets on the US election, with probably a few hundred $m bet in total.
In the medium term if it is true that the non-money markets outperform, imagine this will disappear as people bet on the money markets using the information from the non-money ones.
Regarding hedge funds - I guess they aren’t called hedge funds but know a few people who work for companies that make money by using data and analytics to place bets on sports and politics.
Finally take issue a bit with the coin flip metaphor (although agree shouldn’t update much from one event). With a coin flip (in theory) you can’t figure out before the toss anything that will help you predict it. In contrast the morning of an election (particularly one with a decent margin like this one) the result is pretty knowable if you had the brains and maybe a couple of hundred million to figure it out (I would guess you could call it the right way maybe 80-90% of the time), with the major sources of unpredictability being weather, last minute political surprises and some variations in turnout that might be hard/impossible to predict in advance. No one actually does this but some people get closer than others (private polling, new data approaches, inside information etc). So the fact that the money prediction markets got closer to the ‘real’ odds of maybe 85-15 does count for something.
But do agree that this is only one event so doesn’t count for much, and also only applies to an election where there is a decent margin of victory. In contrast in 2000 or 2016 the ‘real’ odds would have been much closer to 50-50, as they were so tight.
"My strong default would be the real money markets would be far better as in every other area of life, putting money on the line forces people to be more rational, and those that aren’t don’t generally last very long"
OK, so what's the average trading lifespan of the average prediction-market player? Polymarket is only four years old; has there even been time for the market to flush out the irrational players?
And if you think that's inevitable, check out any casino in Las Vegas. Lots of people betting lots of money, most of them losing, very very few of them truly expert card-counters or poker sharks who can realisticaly expect +EV, Has betting real money "forced" them to be rational?
Arguably they are being rational, in that they are enjoying themselves and paying for entertainment that one enjoys is rational. But once you acknowledge that people aren't always trying to Maximize Moar Moneyz, that the point of making money is that you can then *spend* it on other things that they value, then you can't trust a mechanism based on the postulate that everybody is trying to rationalize their expected monetary return. You can't even trust "if they don't rationally maximize their expected return they'll go broke and purge themselves from the sample", because almost all of them will make almost all of their money elsewhere and can afford to lose indefinitely in this market if they're having fun with it or otherwise securing some valuable-to-them reward.
I think you're seriously underestimating the amount of smart money that was available to correct any obvious mispricing. There are billions of non-Americans in the world, and many of us follow your politics (arguably too intently). We're much less likely to have relevant non-public information, but we have access to basically all of the public information. And we wouldn't have had to use Polymarket; if we bet enough to seriously move any real-money market, arbitragers would have showed up and done their thing.
>I have never heard of a hedge fund betting on a prediction market and my guess is that it would either not be legal or require too much compliance paperwork to be worth it. I hope this changes!
Apparently the main thing is just that they have better options. Even if they think they have uniquely good information on the result they can just invest/short stocks that they expect to go up or down as a result. And the downside risk is much lower
Polymarket in particular has a lot more third party risk, and systemic risk around crypto, than most funds would be comfortable with.
> Polymarket in particular has a lot more third party risk
Blockchains eliminate third-party risk. They're trustless: the smart contract always does what it says it will do. There is no third party to trust. (Of course there could be a bug in a contract, or something, but Polymarket has been around for a long time and there's a lot of free money for someone if they can find a bug, and they haven't so far.)
I was using the term slightly loosely. Yes on paper it eliminates third party risk. In practise looking at it from the perspective of a fund you have the questions of: How likely are people to continue to want to buy the token I get out at the end of it. How much do I trust that the person who wrote the contract did so competently and non-maliciously. How much do I trust that my belief that it is difficult or impossible to hack the contract is true? What legal protections if any do I have? And if I like the answers to these can I convince my boss and risk management team?
> How likely are people to continue to want to buy the token I get out at the end of it.
They use USDC, which is a regulated stablecoin collateralized by treasuries and similar which are held by a traditional custodian, so that's not a major worry.
Not sure I'd call Circle a traditional custodian since almost all of what they do is crypto related. USDC briefly lost its peg to the dollar when silicon valley bank collapsed, if I was a fund I'd consider that a problematic precedent.
That's not me saying that the risk is super high. But as a fund you would want to compare it to the other options you have for using your capital. If you can invest in more traditional implements that have the same correlation with the election result, why bother.
Who decides which outcome wins on Polymarket? Often it's this weird Rube Goldberg machine called UMA. A very clever system for reaching decentralized consensus about events, but also it's definitely weird and extra risk.
That may well be the biggest risk, but they haven't done anything clearly wrong yet... resolution of some of these questions can be gray areas, is the worst I've seen.
Yeah mechanisms are like that are fine when it's mostly a hobbyist thing. But if if anyone with serious resources is involved and has an incentive to cheat you have a problem.
For a uma you just need to buy up enough of the tokens that you can swing the result. Which is potentially easier on a contested issue like the election.
My understanding is that UMA worked fine in the 2020 election despite Trump insisting he won and accusing it of being stolen.
Hedge funds come in all shapes and sizes. I think that there's room for small hedge funds (a couple of dudes with a seven or eight figure chunk of money) to clean up in this field.
If I believed Scott's core claim here that fake-money prediction markets are systematically more accurate than real-money prediction markets then I'd be quitting my job right now and starting up such a hedge fund because trading the diff between real money markets and metaculus would be ridiculously easy.
The question is not whether it's possible to make money, it's how much money you can make at what risk level relative to what else you can do with your capital. It doesn't matter if there are $1 bills on the ground if I picking them up you lose the opportunity to pick up $100 bills. And given the current rates of growth of the stock market etc competition is high.
I don't know about specifically the play money markets, but plenty of people have made money in the past by taking 538 or other public models results and betting that. That's why the market tends to converge towards them outside odd situations
As a heavy prediction market user, I think this argument is directionally correct. Theo's market impact created a substantially wider gap between real-money and fake-money markets that never really closed.
This runs contrary to most people's intuitions about how markets work, but I think the important thing to understand is that 5% of edge on a one-off event (i.e. the additional "arb" that was created from Theo's impact) is just not that much. Furthermore, although this was a breakout year for prediction markets, there isn't actually *that* much money hunting for opportunities here. It would have taken tens of millions of dollars to reverse the additional spread that Theo created, and while there were some traders willing to do it, the total bankroll willing to correct Theo's impact was just not on the order of Theo's own risk appetite.
The one caveat I would add is that I think we should distinguish the claim that "Polymarket's odds on election day were wrong" from the claim that "Metaculus/etc's odds on election day were right". I think it's very plausible that real-money markets were correct prior to Theo's arrival, or at the very least that the true probability was somewhere in the 4-5% gap between real- and fake-money markets at that time. But given that this gap widened by another 5% or more, *and it did so purely as the result of one trader's impact*, I think it's very reasonable to claim that Polymarket was mispriced on election night in a literal sense.
Did PredictIt take a loss because there was more money on Trump than Harris? It seems like this wouldn’t be an issue if prediction markets were more like the stock market (finite amount of action available) than a bookie.
As I see it:
1. the best polls said the election was 50-50. Small random factors can move it either way.
2. the reality was that virtually every demographic, and in every swing state, Trump overperformed his last election. Small random factors made no difference at all. In theory, maybe all these decisions were made by voters on election day and pollsters couldn't pick that up ---but that seems highly unlikely. The clearest story is that the pollsters missed exactly the kind of thing they should have easily picked up. If some demographic groups were moving more toward Trump and others against him - that can be tricky to resolve. But when almost every group did that, how can pollsters miss it?
3. Before the election I wondered why more people didn't bet against the French whale, given where the polls were. But now, of course, the question is why more people didnt bet with him. I don't mean in the trivial sense (Trumps probability of winning is now 100). I mean, in a more efficient betting market I would expect such a broad based move to generate levels much higher than 60% prior to the election.
Yeah, how (why) did the pollsters miss it. It's not a single coin toss. It's like in every state they got it wrong. And the money markets got it right!
I don't know why you'd still refer to it as "mispricing", when betting against "Theo" would have lost money and moved the market away from what turned out to be the correct result. And perhaps I'm just ignorant of how many people are commissioning private polls, but that doesn't seem "replacement level" to me. How many other Polymarket bettors were doing that? It also seems incorrect to say that Polymarket absent Theo is "normal market operations" while with him it isn't. The market includes all the participants, including whales.
I don't think Scott thought through that part, much. He added that part after it was pointed out to him that Theo commissioned secret polls, but the original post wasn't written with that understanding. See https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76002149.
To steelman the original version of his post (which didn't realize that Theo had his own polls), though, one could claim that "normal market operations" have enough liquidity to prevent individuals from significantly affecting the markets, as a whole. They therefore more effectively approximate the aggregate opinion. While a few large players moving markets against the consensus of the rest of the market is possible and is consistent with normal market operations, in practice, markets typically more closely resemble the situation in play-money markets in which single individuals don't move the markets as much.
This would therefore not be a fundamental distinction between prediction markets and other markets, but instead just a feature of the liquidity of these particular markets due to the factors that Scott lists later in the piece which limit participation in Polymarket.
> I don't know why you'd still refer to it as "mispricing", when betting against "Theo" would have lost money and moved the market away from what turned out to be the correct result.
I beleive the idea is that there is some correct ex ante assessment of the probabilities given avalible information (public and private in this case), and that even though Theo happened to have won, the market still may have been mispriced. If someone offers you 60-40 odds on a fair coin to land heads, the market is mispriced — even if the coin does in fact land heads, meaning the bet loses you money and moves the market away from the outcome that happened in this case.
Someone says "I bet that coin will come up heads because it's a double-headed coin!", bets enough money that lots of other people say "he must know something!", such that the local prediction market gives 90% odds of the coin coming up heads.
The coin comes up heads. But on examination, it was a perfectly ordinary coin. Was the prediction market "mispriced"?
You can define "mispriced" and "properly priced" to make it so, but if you do you'll have defined a market that can't be determined to be "properly priced" until after the fact and can't be trusted to make predictions because it sometimes gives you 90/10 odds on a fair coin toss.
Scott is using the terms in a way that points to potentially useful results.
There is no coin that was examined to have 1 vs 2 heads after the fact. There were multiple polls and disagreements about which ones were reliable and what kind of bias existed in them. Someone already replied to me about the nature of uncertainty in this case, so I'll link to that https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76203508#comment-76266715 "Theo" really did go out and collect info that others didn't possess.
I think some of your arguments imply that you shouldnt listen to prediction markets at all. If you think prediction markets only work when they weigh people approximately equally, you should just ask people and weight their opinion approximately equally. If youre going to discount market movements without legible justification, you should just believe whatever is legibly justified to you. So, I dont think theres a reason to trust this prediction market less then usual. The track record matters, and arguments like the above may be reasons to trust them less *in general*, but I think arguments to discount a particular market based on how its going are almost always wrong.
COPE.
It sounds to me the non-money prediction markets were mis-priced by 50 cents, and the money markets by only 40 cents - thanks to Theo.
You seem to believe too much that the consensus of the opinions of intelligent people must be correct, and not enough that the data is the way you find out what's correct. We know there are big problems with the polls: shy Trump voters have happened three times in a row now, and we've observed similar things going back to at least the exit polls on Obama's first election (when exit polls stopped being good predictors). We know that the excessive clustering of polls is statistically impossible, so the pollsters must be fudging them. The Polymarket Whale wasn't just a guy with a different argument that should be counted equally to other arguments, and he didn't just do a "private poll:" he commissioned a "neighbor poll" to correct for the shy Trump voter problem. You don't seem to believe fully in markets, when you say that everyone's "votes" should count equally, not based on how much they bet. The Whale had a history of successful financial betting, high conviction, and better evidence than most people, so his vote *should* count more. I do believe all those things make markets work better than majority wins. Your pre-election article struck me as the worst of yours I've read; hearing you call the Whale a just a degenerate gambler was a low point. I've also seen you have the same blind spot before (heavily weighting the opinions of people rather than the direct data): when you missed Biden's cognitive decline, while others just watched the video over the years. I get the impression that some political emotionalism may have clouded your judgment recently as well. Another point you didn't make in this article: volume *matters* in market efficiency... and Polymarket had enough volume to be efficient this time. You point out problems with PredictIt, without mentioning that it has low volume limits. Also, hedge funds and big American banks' foreign subsidiaries *did* bet on Polymarket this time, according to Haseeb Qureshi (he is an investor in Polymarket, among other things). That 2022 data needs to be be updated, now that we finally have some volume to make markets efficient. You mention "the personal opinion of one whale," but that *is* part of the "normal operations of the market." People with better judgment, like the Whale, that look at evidence, not theories, that got richer over time in the business of betting, should have more weight in order for markets to be efficient... that's essential to how markets work, part of what makes markets accurate, and I believe markets are superior to intellect. The smart money/whales take money from the stupid money over time, making markets more accurate (another example: the play at a new poker room improves in quality over time as winners take money from losers, as the latter mostly get better or stop playing). People have always had theories (intellect), since before the scientific method; I'm glad that now we have the scientific method and markets, where being wrong has a cost (unlike areas like governance and academia).
Do we have any details about what those polls were exactly, what their scope was, their methodology, etc? If he asked 1,000 people using a poor methodology and not weighting for the location of the responders, I’m sure he’d get a “Trump solid win!” conclusion but that wouldn’t count as useful new information.
If I said “Theo was correct because he consulted an astrologist” that would technically be “new information” but you’d still challenge me. Same thing here - “Theo paid for polls!” alone doesn’t update me much without knowing the details OR seeing his track record winning money using such polling.
It's about the type of polls: neighbor polls versus conventional ones.
An excerpt from the WSJ article, based on their conversations with Théo:
Polls failed to account for the “shy Trump voter effect,” Théo said. Either Trump backers were reluctant to tell pollsters that they supported the former president, or they didn’t want to participate in polls, Théo wrote. To solve this problem, Théo argued that pollsters should use what are known as neighbor polls that ask respondents which candidates they expect their neighbors to support.... Théo cited a handful of publicly released polls conducted in September using the neighbor method alongside the traditional method. These polls showed Harris’s support was several percentage points lower when respondents were asked who their neighbors would vote for, compared with the result that came from directly asking which candidate they supported.... To Théo, this was evidence that pollsters were—once again—underestimating Trump’s support.... At the time that Théo made those wagers, bettors on Polymarket were assessing the chances of a Trump popular-vote victory at less than 40%.... In an email, he told the Journal that he had commissioned his own surveys to measure the neighbor effect, using a major pollster whom he declined to name.... Théo declined to share those surveys, saying his agreement with the pollster required him to keep the results private.
You seem to believe too much in the opinion of one intelligent person who has a clever idea which tickles your fancy and who was right once, compared to the consensus of all the many other intelligent people with vast quantities of data gathered the boring old-fashioned way.
Which, yes, is known to be biased, but biased in a way that can be studied and understood and to some imperfect extent compensated for by intelligent people who devote themselves to the task. Are you under the impression that Polymarket Theo's clever idea produces infallibly error-free data? Do you even understand what the sources of error and bias are in that case? Does Theo, even?
No, I believe in markets. And Theo wouldn't have had enough money or conviction to move markets that much if he'd only been "right once." (See the example above about poker rooms).
Before this election, people acted like someone who disagreed with the consensus must be wrong (or even illegally manipulating the market)... that was an emotional reaction, not a rational one: what people wanted to believe. There was a lot of emotion in this stuff, which makes for inefficient markets = free money to other people. Good gamblers take advantage of other people's emotions (or are influenced less by emotions themselves). It might be the single most important thing in playing poker, for example.
Those "intelligent people who devote themselves to the task" *are* Theo, and others like him. The rest are amateurs that he takes money from. It reminds me of the old joke about the economics and finance professors walking down the sidewalk when they see a $20 bill on the ground; the finance professor goes to pick it up and the economics professor says "what are you doing? if there were a $20 dollar bill there it would already have been picked up!" Theo's the guy doing the work of picking it up, so that most people never see a $20 bill on the sidewalk. The market will always be closer to reality than the ivory tower, where people don't pay for being wrong. Theo doesn't have to be infallible; markets just have to be better than every other solution (and they are).
I mentioned a couple of the sources of bias above, at least one of which Theo used.
If Polymarket and co were so mispriced, how much did you earn by trading against the mispricing?
Well, many people made money betting on when Kamala's odds shifted from 34% on October 29, to 44% on November 3. Presumably some of them were rationalists, given I saw some rationalists on twitter talking about the mispricing.
That four day shift implies one of the following:
- Polymarket's odds can be mispriced/fluctuating for reasons not related to the market topic (e.g. a whale investing)
- Trump's chances got worse in those four days for some reason.
- Someone learned new information that changed Trump's odds during that time
While perhaps all these contributed, I find the first option most significant, because I don't see the specific facts that would make 2 or 3 more likely to be significant.
If you'd have asked me before the election, I would have said I thought Trump was favored 55-45.
As would I have, and as a result I lost money on Kamala making what I thought was a +EV investment just like the rest of the "smart money". However, I think this article is much more about the election day odds than the pricing movement in the preceding week.
My point is the that the pricing movement in the preceding week implies something about how reliable prediction markets are for elections in general.
"But on the specific literal level, Polymarket was mispriced last Monday."
Yes, not being facetious, but on the specific, literal level it should have been 100% Trump. I defend probabilistic reasoning under uncertainty, but the election has to be interpreted as a deterministic event, and one that could theoretically be accurately determined beforehand.
Some people, for probably a combination of good and bad reasons, mediated by good or bad epistemics, used more accurate proxies and measurements to assess this, while others (including myself) failed to do so, believing incorrectly that it was a "toss-up". Metaculus and poll aggregators just had the wrong combination of proxies and measurements.
I agree we have to be Bayesian about people being right for good or bad reasons, and we may never know if there was anyone genuinely well-calibrated. But it does seem that the "toss-up" people were wrong in some fundamental way that they wouldn't have been for, say, a sports prediction.
In some sense, almost everything that happens is deterministic. It's pretty clear that most events we care about are not influenced by actual quantum randomness. With this is mind, predictions that aren't basically certain reflect our ignorance, inability to model the world well, and lack of infinite computing power.
> But it does seem that the "toss-up" people were wrong in some fundamental way
Given the above, everyone is wrong some fundamental way. The only way we can evaluate prediction sources empirically is by measuring calibration (i.e. consistency with themselves) and comparing different prediction sources (e.g. using some proper scoring rule). Even then, it's hard to evaluate a prediction with few other similar predictions, for example one for a presidential race.
So, back to the point, I think Scott is right that while empirics are important in evaluating predictions, we should also value theory heavily in valuing prediction sources. If someone used astrology to predict 13 elections in a row, I wouldn't particularly value that as prediction source, because it has no theoretical backing that would lead me to believe its likely to be right in the future.
Conversely, Nate Silver's method of averaging the results of a bunch of sources asking people who they are voting for, adding uncertainty based on past polling errors, adjusting for bias, etc. etc. seems like a pretty reasonable way to predict an election to me. When people say it makes bad predictions as a method, my question is "in comparison to what". It seems rather pointless to say its bad without recommending an alternative method, because *calibrated predictions can only be compared in relation to each other*. Though I would probably trust prediction aggregators over Nate alone.
To be more specific about polymarket, the level of volatility doesn't make sense to me based on what happened this election. And, for example, on the eve of the election 538 had "Trump winning popular vote" at 29% and polymarket at only 26% (though polymarket did have higher odds two weeks beforehand). So perhaps prediction markets overreact to news or are overly certain?
I'm not sure if this directly responds to your comment, but these are some thoughts it spurred.
Good comment. To add a thought to your thought about how to evaluate predictors not merely based on their results, but also whether their methodology is using astrology or more like Nate's, I'll mention that top forecasters often say they are a better judge of their peer's forecasting ability than their peer's forecasting track record. I agree with that, and think it's mainly because you can only get so much information from a small sample of a few hundred or few thousand forecasts (track record). Examining how those forecasters make forecasts provides a lot more information, and one skilled forecaster can learn a lot from that about a fellow forecaster's strength as a forecaster.
"But it does seem that the "toss-up" people were wrong in some fundamental way that they wouldn't have been for, say, a sports prediction."
Yes, and the difference is that there is more "irreducible uncertainty" (Philip Tetlock's term) in a sporting game than in a Presidential election.
"Yes, not being facetious, but on the specific, literal level it should have been 100% Trump."
The world is not necessarily deterministic--there is quantum randomness--but as for whether this would ever cause Harris to win >0% of elections if we re-ran a simulation of the universe a million times starting one week before the election is unclear to me.
I support using a notion of "true probability" that is meaningful in forecasting by never quite being 100% or 0%. This is achieved by adding random noise to the hypothetical simulated re-runs of the universe. We re-run the soccer game or re-run the election, but tweak the universe in some random way at some point in time before the game or before the election. So for example, a week before the election we make someone's car tire get punctured and go flat when it hits a rock on their drive to work, rather than just holding together as happened in reality. Or we cause one of the soccer players to sleep in for 5 minutes later or set their alarm for 5 minutes later than they otherwise would have a week before the game. These tweaks would undoubtedly lead to big changes in the long-term, and meaningfully affect the frequency of wins in the million universe simulation of the election and soccer game.
In this sense we might say that the "true probability" of a team winning a soccer game is low, say 60%, even if that team in fact went on to win. We know this because we believe the two teams were about equal in strength and random-tweaking of the simulations would actually change who wins a significant amount of the time.
With elections though, if you add the random noise only shortly before the election, say a week-or-so, then that's probably no enough to make Harris win a significant percentage of the time, given the margins Trump actually won by. As such, I think that the "true probability" (in the sense that I mean it in which the random noise is added a week before the election) that Trump won is likely >90%, but not quite 100%. If we ran a billion simulations, Harris would likely win some of them. (Or if you disagree with that, imagine that the random noise happened a few months before the election. At some point, you'd start giving Harris nontrivial chance of winning some of the elections in some of the simulations of the universe.)
Having said all this, the reality is that the best subjective forecast we can make on a given question is almost always more uncertainty than the "true probability" given what Philip Tetlock calls "irreducible uncertainty" in the world.
So perhaps an efficient betting market for the soccer game with a true probability of 60% would have a market price of 51%. And similarly there is probably no rational way that the betting market for Trump could have been priced at >90% (that would require irrational actors, betting on unjustified true beliefs; irreducible uncertainty in the world would prevent them from being able to know that Trump was that likely to win).
Could it be the case that the most informed best forecast one could have made on Trump winning given the irreducible uncertainty in the world was less than 60%? Could it have been less than 50%? I think both are possible, but very unlikely.
See my top-level comment elaborating on my critique of Scott's post for further explanation: https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76061286
Help an average Joe understand why prediction markets are a good thing.
No participant in a market is passive. Everyone is an actor. If a person bets money on a certain outcome, they will naturally do what they can to influence things in a direction that's favorable to them. They will do this unconsciously if not consciously.
There are significant consequences to mistakenly regarding prediction markets like passive polls. It opens up the public mind to undue influence when a whale bets a certain way. Anyone who understands market dynamics knows about the momentum of sentiment when things start to go a certain way. Yes, there are opportunities for arbitrage, but history tells of how speculation can bubble to a harmful degree.
Market dynamics are one thing when based on some underlying value. But with prediction markets, we're betting on the future. We're betting on mass phenomena. We're often betting on something we are both a part of, and a part of perceiving.
Aren't we just letting Moloch having even more control or influence over us? More power and sway for money to influence and shape mass social behavior?
Are markets the solution for everything?
What might we be running from?
Scott wrote a FAQ on prediction markets that you can easily find on this very website.
My understanding is that the prediction markets enthusiasts of this website believe that markets are indeed the solution to most assessment problems.
It’s not unsound: an ordinary stock price reflect the expectation of (suitably discounted) future earnings. So betting on the stock market is already a prediction problem.
The basic idea is that people want more money. Therefore, by putting money at stake, you incentivize people to spot mistakes and correct them.
Because of this incentive, this should imply that a sufficiently large prediction market gives you an actual prediction, even without anyone giving you any odds.
It is supposed to be accurate because anyone believing otherwise can (expect to) make money simply by betting.
More generally, believing markets to be the solution to just about every problem is not particularly rare here. I’ve come to suspect that it may be an American thing. After all, they literally spend their whole lives betting that the market will go up (so that investing their economies into pension funds to fund their retirements makes sense). Apparently, on the long term, it hasn’t failed them yet.
(What would be convincing alternatives?)
This line of argument concedes that markets do fail sometimes, but only in the predictable ways studied by classical economists for a while, with well-known remedies.
I am personally skeptical and suspect that there’s a lot of motivated reasoning going on in there. That freer-markets are a lot easier to support when you’re affluent, things are going well for you, and you win off of it.
(Which, as an old Scott reminds us, is *not* the majority situation. Sure, we’re better off than subsistance peasants, but that’s not too relevant.)
Prediction markets are good because they aggregate the most accurate information, thus helping us to predict things. Attempting to influence voters is fair game in politics, but I'm unaware of prediction market speculators like "Theo" doing that because that wouldn't be the most efficient way for him to use his money to make more money. Remember that the Harris campaign had something like 3 times the funds of the Trump campaign and still lost.
> But with prediction markets, we're betting on the future.
That describes a LOT of speculative markets, such as the stock market. Someone could short a stock and try to sabotage the corporation, but that basically never happens.
Does Polymarket show the bets people are making? I might assess the prediction differently if I knew that one big bet moved it that much. This seems like a good partial solution at least.
Another point in your favor is the very large post-election move in financial markets, which suggests that those massive markets were very much unsure of a Trump victory.
I'd only expect the size of the move to be ~20% larger if everyone is updating from 50% confident than if everyone is updating from 60% confident, so I don't think the vague magnitude of the move proves much.
I don't really understand this paragraph: "Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%. Now should we say that the “true” probability of Trump’s win was 90%? Imagine he spent half his money on a yacht just before the election and only had enough money left to move the markets up to 56%. Now should we say that the “true” probability of Trump’s win was 56%? Why should the true probability of Trump’s win depend on whether a French guy bought a yacht or not?"
Surely the true probability of Trump winning should have been 100%, since he did in fact win, and the fact that the markets were trading at less than that means there's a lot more accuracy out there to be gained?
See my comment that clarifies this. I support a notion of "true probability" that is meaningful by never quite being 100% or 0%, but also reflects the fact that the best subjective forecast we can make on a given question is almost always more uncertainty than the "true probability" given what Philip Tetlock calls "irreducible uncertainty" in the world.
My comment: https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76061286
Three other comments on the reasonableness of a larger move given a hypothetical richer Theo:
a) Echoing other commenters, if the available information, even with additional polling, is insufficient to push the expected probability up past e.g. 70%, then Theo++ will stop buying when the price reaches that point
b) weak point: A richer Theo++, if self-made, is presumably at least marginally smarter than the real world Theo (or luckier? Teela Brown gene :-) ? ) and their opinion should count more heavily
c) stronger point: real Theo commissioned an additional poll to gather additional real information. Theo++ has a larger budget, and could commission more polls, or larger polls. At the very least, the 1/sqrt(N) random error should go down.
Isn't there something like a p-value to say how unlikely the observed election results were in Nate's model? Was it an extreme tail event?
I remember Nate writing a model deep dive that listed this outcome (Trump sweeps all 7 swing states) as the modal outcome, out of the 128 possible outcomes for the 7 swing states. Don’t remember the exact probability he assigned to this outcome but I think it was around 7-10% of all results. The next most common individual outcome was Harris sweeping all 7 swing states, I think that was around 5%
Yes. In this post: https://x.com/NateSilver538/status/1848060370268938538 he had a Trump sweep at 24.7% and a Harris sweep at 16%.
Yeah, that’s it. Guess I remembered the probabilities being a lot more uniformly distributed than they were.
Hi Scott!
Any thoughts on IBKR's new Forecast Trader?
https://forecasttrader.interactivebrokers.com/eventtrader/#/markets
It allows US residents to buy real-money contracts on US election results.
https://www.interactivebrokers.com/lib/cstools/faq/#/content/685271469
Kind regards,
David
Great post! This would be a very good reading to discuss in a PhD finance class, because it is so good for thinking about market efficiency. I took a 6-student class with Fischer Black at MIT in 1979 based on 30 to 40 questions like "Why do corporations pay dividends?" (and longer ones), and I can see him putting a betting market question and your post on the syllabus. He would stand at a lectern in his 3-piece suit (no other prof wore them, or even suits usually) listening and writing little notes, and then maybe give his opinion.
One part of the discussion would be why you can't do arbitrage between betting markets-- place hedging bets that reduce the risk to zero. Probably the answer is transaction costs plus that if you place bets big enough to make it worthwhile to set up the apparatus, you move the markets too much because they're thin.
I think transactions costs are probably a big part of the story. The other part is probably slight differences in resolution criteria that make it difficult to price those differences accurately.
I buy some of this. But I think it's interesting that there wasn't a multimillionaire in Canada, the UK, Japan, etc who decided to take the other side of this bet, given the pricing. Although I'm sure some did the arbitrage trade (I wanted to, but didn't have time to set up my wallets).
But I suppose also from a statistical analysis perspective, it's not crazy to throw out an outlier if that's having a large effect on the variable you're measuring. Since it's clear the biggest whale really did have an outsized effect on price, I think it's reasonable to question if Polymarket was really efficient. Had it been open to US persons, I think we could trust it more.
What fraction of the world's multimillionaires do you think actively follow Polymarket?
> Second, [real money markets have a long history of giving weird results.](https://asteriskmag.com/issues/05/prediction-markets-have-an-elections-problem-jeremiah-johnson?utm_source=substack&utm_medium=email)
It seems like part of the problem is that some people bet their identity rather than betting as insurance + arbitrage. Are there any plans to educate people on how to think better about betting? Otherwise lowering the access barrier may not result in better results, if people bet their identity rather than their beliefs.
Dumb money acts to subsidize smart money: https://statmodeling.stat.columbia.edu/2024/10/25/prediction-markets-and-the-need-for-dumb-money-as-well-as-smart-money/
Nate Silver always states a probability and then people knock him for a specific outcome that he explicitly said was possible, it must drive him nuts.
I think he's developed a thick skin over the years (and a low regard for most people's understanding of probability).
"Yes, but there were dozens of people who could give equally-plausible arguments for their positions before the election. These were divided half-and-half into intelligent-sounding pro-Kamala arguments and intelligent-sounding pro-Trump arguments"
Nate Silver disagrees:
"To the extent there was anything that resembled actual insight, it was that it was much easier to come up with a list of reasons that Trump would win — inflation, immigration, Joe Biden wanting to be president until he was 86, the illiberal backlash around the world — than to make the same list for Kamala Harris. I tried and could maybe come up with 6 or 7 good points for her, but not 24. So that may have weighed on my “mental model” of the race."
https://www.natesilver.net/p/the-story-of-trumps-win-was-foretold
(His list of 24 reasons for a Trump victory: https://www.natesilver.net/p/24-reasons-that-trump-could-win)
"Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%. Now should we say that the “true” probability of Trump’s win was 90%?"
This assumes that Theo would have continued to buy at that high a price. It's possible that he thought probability of Trump win was, say, 70%, in which case he would have bid it up to that but no higher.
> But long after Biden won the last election, PredictIt said there was a 9% chance Trump would be the next President.
This is an understatement. If I recall correctly, the Trump vs Biden election market was at 17% Trump in late Nov/early Dec. 2020, and the add all the "Trump win" results in the "What will be the electoral college margin market" it was at 23%. I know because I played some, uh, Minecraft during that time period.
It seems like all these prediction models were fundamentally flawed. Does anyone really think if we simulated Election Day 100 times, Kamala would not have won 40 times?
Funny enough, ChatGPT and Claude3.5 both accurately predicted the election outcome with non polling-based tools.
https://open.substack.com/pub/davidkingsley/p/can-ai-forecast-the-presidency-part?r=2hmql2&utm_medium=ios
If we were sampling from the ensemble of possible universes which produced polling data, market prices, etc, that exactly matched what we saw right before the election, then I believe Kamala would most likely have won somewhere between 40 and 50 times.
Several people here seem to be arguing that we live in what we now know is a single deterministic universe(*) where it was ~100% certain that Donald Trump would win the election, and that forecasts should be judged correct by how confident they were that Trump would win. This may be true, but it is irrelevant for predictive purposes. We may live in a deterministic universe where the outcome of the election is preordained, but we don't know *which* such universe we live in. There are many possible deterministic universes in which the polling results were exactly as we saw and Kamala Harris won the election, and many such universes in which Donald Trump won.
"Prediction", is the art of figuring out the odds that we live in a particular subset of universes, and its useful accuracy is in how closely the predicted odds match the fraction of possible deterministic universes in which the stated outcome occurred. And, given the available data and analysis at the time, the best answer was something like "slightly better than 50% that we live in one of the universes where Donald Trump is destined to win, and about 20% one where he sweeps all the swing states".
* Or nigh-deterministic if we assume e.g. that people's decisions are fixed on election day but a large number of freak buttefly-effect accidents preventing some voters from going to the polls might have had a miniscule chance of changing the outcome.
Cool!
>If we were sampling from the ensemble of possible universes which produced polling data, market prices, etc, that exactly matched what we saw right before the election, then I believe Kamala would most likely have won somewhere between 40 and 50 times. [out of 100, presumably]
Reminds me of a statistical mechanics class from half a century ago, making a distinction between a macrostate, where we know the pressure, temperature, volume, and maybe a handful of other measurements of a sample of a gas, and microstates, where we know the individual quantum states (including translational states in a box...) of each individual molecule in the gas. Good old microstate ensembles! :-)
The presupposition that these polls were modeling anything approximately our universe should have been dispelled midway through election night. The Iowa Ann Selzer poll was 16 points off. There was no 40-50% chance of Kamala winning whereas if we simulated people going to vote that day 100 times 40-50 we would have had the opposite results.
The polling data was systematically flawed and only made to appear a close election through nonrepresentative sampling and arbitrary weighting. I am not saying we live in a deterministic universe. I'm saying that there were a million red flags that must be ignored to claim these represent the voting population. It is a prime example of a garbage-in-garbage-out models.
One of the best function of prediction markets is that - like any good price signal - it cuts through the motivated reasoning and lies and gives ordinary people access to reality. On election night, even as late as 10pm the TV networks hadn't called states like GA and NC that everyone paying real attention knew were over, because of the percentages on Polymarket. Somebody watching CNN would've had to read tone and body language and have an amount of political knowledge/insight that we may take for granted here but that probably less than 5% of the American people possess. But somebody just checking Polymarket regularly all night knew this thing was all over but the shouting by 10pm.
And it performs that service very well even if it's off by 10%, which is meaningful to bettors but to normies who just want to know "is this thing the news media is yapping to me about something I should take seriously?" doesn't need to be that precise. Just as a consumer in a small town doesn't need to know the price of pork as accurately as a grocer does, but still benefits from having a device in his phone that shows him what the walmart in the next town over is charging. I think this is a great tool for average people to cut through the manipulation aimed at them.
Watching Steve Kornacki on MSNBC you were hearing, dozens of times, about counties in GA and NC, "Harris is not performing at Biden 2020 levels and the remaining vote out is likely tilted R." It was clear by, I think, about 9ET where it was going. There are detailed guidelines for "calling a state" but there was more than body language available. That's the kind of info the Polymarket people were presumably working from.
I largely agree, but I want to defend the mainstream media a little bit. Even putting aside any political bias/preferences, the mainstream media has two strong reasons to call states very late:
1. Ratings. Ratings probably spike briefly right after a critical state is called, but ratings will likely decline if/when people think the election is basically over (and for people on the east coast, they can finally go to bed).
2. Nothing is more embarrassing for the media on an election night than calling a critical state and then having to reverse it not long after.
So basically, even if we had a magically unbiased mainstream media who truly didn't care which party/candidate won, they'd have two good reasons to drag out the calling of states. Thankfully, Polymarket does not have the same concerns. And so yes, for those who don't want to stay up all-night or try to read between the lines of what news anchors are saying, they can just look at Polymarket and do some simple math.
Why was it so important for people to know that GA and NC have been called by 9pm vs 10pm? Other than pure political entertainment value, why does it matter?
This also ignores the possibility of Polymarket calling a state too early in a more complicated situation, such as what happened in 2020 when Trump was leading initially but then Biden overtook him as the night went on in certain states.
<i>we’re asking whether to trust the normal operations of the prediction market vs. the personal opinion of one whale</i>
That's an interesting definition of "normal operations of a market". Having large traders is part of the normal operation of any money market. It is also one of the reasons why these markets are relatively efficient. Large traders are (sometimes) better informed because for them there are stronger economic incentives to be better informed. This has certainly played a role here (Theo commissioned his own private polls, if that's not adding information to the market I do not know what is.) So that's what a normal operation of a market looks like.
At to the claimed better accuracy of fake money markets vs. real money markets, I am not sure it is measured correctly. Real money markets binary event prices do not equal implied probabilities for low and high implied probability events, for some pretty simple reasons (interest rates, credit risks, frictions.) So when one takes these prices and discovers that they do not match realised probabilities, a part of mismatch is due to the incorrect price to probability mapping.
>Suppose that Theo had done even better at his banking job, was 10x as rich, and could move Polymarket up to 90%
This is very, very wrong. Theo didn't move Polymarket to 90% because he didn't think Trump had a 90% chance of winning; he felt ~60% was a fair price and set limit orders to floor the market, slowly accumulating volume. Prediction markets are not binary, you are not supposed to just decide what you think will happen and go all in. Instead, you invest if you think the probability of something happening is higher than the probability expressed in the market. Theo did set the price, yes, but to the probability he thought was correct at whatever volume would fill it, not to an arbitrary probability based on a 100% certainty of a Trump win * his bankroll.
This massively under-rates Theo. It wasn't just his opinion or "a completely replacement-level example of the intelligent-sounding pro-Trump arguments"
From an interview he gave later:
> As Théo celebrated the returns on Election Night, he disclosed another piece of the analysis behind his successful wager. In an email, he told the Journal that he had commissioned his own surveys to measure the neighbor effect, using a major pollster whom he declined to name. The results, he wrote, “were mind blowing to the favor of Trump!”
He didn't just think the polls were wrong, he literally sought disconfirming information by funding an independent poll using what he believed was a better methodology. Then, he bet on the private information he received from that poll.
I speculate that the ~$40MM was probably not just his money, he probably got some external funding as well. Almost certainly he got the results and his interpretation cross-checked by multiple other experts. Even someone with a few hundred million of net worth doesn't just throw down $40MM on a whim.
This is literally exactly how prediction markets are supposed to work. You have a very informed bettor whose private information is incorporated into the probabilities.
Right. If George Soros in his trading prime had made this bet instead of some dude named "Theo", we would believe that Soros had moved the market to the more correct price, because Soros had a phenomenal track record for trading as a whale who moves markets. Yet Theo supposedly was a great financial trader himself. His lack of prior fame makes all the difference somehow. Soros would have been right, but "Theo" got lucky.
It could be that Theo is well-known (at least in France) under his real name, but trading on a crypto platform with pseudonymous accounts obscures that.
Two related observations:
(1) Calibration city says Metaculus is less well-calibrated than other markers and tends to be biased towards 50% https://calibration.city/calibration
(2) On election night, Metaculus was useless since it does not move quickly in response to new information. Manifold, Kalshi, and Polymarket all converged on >90% Trump while TV pundits on my other screen were all talking about "multiple paths to victory" etc etc.
So from a purely popcorn-munching, I-know-the-result-sooner-than-my-friends perspective, you've gotta admit the real money markets are the better option!
Scott, I don’t know if you’ve been reading Rajiv Sethi’s Substack, “Imperfect Information,” but a lot of what you’ve been saying lately regarding potential prediction market liabilities has been explored there. The main difference is Sethi’s often talking in reference to the relative utility of models vs. prediction markets. From his most recent post:
“Markets have some significant advantages over models—they can take on board novel sources of information, can deal with historically unprecedented situations, and can respond very rapidly to evolving events. But they also have potential shortcomings as forecasting mechanisms. They can come to be dominated by a few large traders whose views may be idiosyncratic, they are subject to overreaction and correction, and when they reference events whose objective likelihood of occurrence depends on subjective assessments of this probability, they create incentives for manipulation.”
Were markets actually "dominated by a few large traders" recently?
I think Sethi’s claim is not that they definitely were, but that “they can come to be.” I take it you don’t agree with him that prediction markets are at risk of this.
I think it depends on the market. Polymarket is open to people everywhere but the US, and there was a lot of attention on this election, so even "Theo" only provided a small portion of the total. PredictIt was less accurate because it's more restricted, which is precisely what Sethi said could potentially make it more accurate by not having such influence from large traders:
https://rajivsethi.substack.com/p/the-limits-of-arbitrage
It should be noted that I also mused along Sethi's lines earlier (but as sort of a Devil's Advocate for an argument), and afterward noted that the actual outcome falsified that theory in that case: https://x.com/TeaGeeGeePea/status/1854870049359630447
If these markets get big enough, the primary use for many users will be hedging election outcome risk. In fact, I think it would be the main benefit of these markets maturing. For example, investors in solar companies could have bet on Trump to win to protect against stock losses.
1. Run for President promising terrible policies that will destroy all businesses
2. Wait for billions of dollars to flow into the prediction markets hedging against your possible win
3. Place a large bet that you won't win
4. Drop out of the race (or flame out spectacularly in a scandal)
5. Profit
I posted one comment about Theo being under-rated, but I think the description of "rich techie with $1M net worth should reasonably bet around $2k" is also incorrect.
If you have extreme confidence in your bet - or even better, if you have some edge - the right move is not to bet $2k or even $166k. The right move is **get more money**.
If you had an amazing but capital-intensive startup idea, you wouldn't say "hey let's see how far I can take this on $2k", you would go talk to a bunch of VCs and raise a few million dollars. Similarly, if you have actual edge in the betting markets and there's sufficient market liquidity, the play is to go and raise external money from investors or set up some kind of capital pooling with other, wealthier people. There are entire industries dedicated to this kind of thing: hedge funds, private equity, proprietary trading firms, etc.
If you have actual, real edge and not "Kamala wins is just like, your opinion, man" then you should be able to raise reasonable amounts of money with a few weeks to months of lead time (depending on your personal network).
If you don't have real edge, then it's totally appropriate to bet $2k money and book it as entertainment rather than investing.
Given how well-versed this crowd is on this topic, I hesitate to bring up this old paper, but I think it provides some theoretical arguments about the limitations of prediction markets, even without whales, that are almost never engaged directly in such discussions:
https://www.nber.org/system/files/working_papers/w10359/w10359.pdf
Key quote relevant to "whales":
"The concluding Section 4 observes that the information about beliefs contained in equations (1) and (2) is fragile. If budgets are not statistically independent of beliefs, prices in prediction markets are determined by the joint distribution of expectations and budgets rather than by the distribution of expectations alone. If traders are risk averse, price reflects their risk preferences as well as their expectations and budgets.
For these and other reasons, I suggest that direct measurement of expectations may be preferable to inference on expectations from prices in prediction markets."
I strongly disagree with the claim that Polymarket's price that Trump would win was mispriced in the direction of being too high on e.g. November 4th.
I explained why in a series of comments I wrote on the Facebook post of a friend who shared Scott's post. Copying them here (sorry for not polishing them up, but I think they're quite comprehensible despite being written sloppily):
----
Scott says several things I disagree with in this post and it would take a full post to adequately explain why I disagree with all of it.
Listing out some issues:
(1) He claims that the markets were "mispriced" without explaining what he means by this.
(2) Later he references the concept of a "'true' probability", which he does not define.
(1b) By "mispriced" does he mean that the market price does not match the "'true' probability"? Or does he mean that the market price does not reflect the most-informed credence that humans can come up with (a different probability)?
(1c) His "mispriced" claim seems to be that the markets should have been priced closer to Metaculus's ~50% than their actual ~60% Trump price. (1b continued) Whether he thinks this because he thinks Metaculus was accurately forecasting the "'true' probability" or because he thinks Metaculus' forecast was the best forecast one could make given the information available to the world before the election is unclear.
(1d) In either case, I claim that Scott is wrong to claim that the markets ought to have been priced below 60%.
And now (because it's easier), to just say some related beliefs I hold without responding directly to Scott's post:
I believe that looking back on how the election unfolded provides us with evidence that the "true probability" of Trump winning was >90%.
Note that I've been trying to come up with a single coherence concept of "true probability" for years that works for what I mean in forecasting conversations like this one while at the same time not contradicting our knowledge of physics and how it is roughly deterministic (or random with quantum mechanics).
I don't have a single definition, so what I will say is that I mean something like "if we were to reset the Universe to the state it was in one week before the election and had some random noise (like a rock that almost punctures someone's tire in real history actually causing them to get a flat, or vice versa) to make sure that they did not all play out exactly the same (due to determinism, if quantum randomness doesn't actually make a difference), then, in what percentage of simulations would Trump win the election? "True probability" is probably a misnomer given this meaning, but this captures our meaning, similar to how we want "true probability" to mean something such that "the true probability that the D6 will land on 6 when I toss it right now is 1/6th", even though if we had enough information about how the die was oriented in your hand and how your brain worked we could predict the exact way in which you would toss it and how it would tumble through the air and hit the table and land with high probability (since quantum randomness probably doesn't affect this, unless it does with cognition somehow, but you get might point), meaning that 1/6th probably isn't *actually* the "true probability" and the true probability is probably exactly or almost exactly 1 or 0.
Note that had PA been decisive and had Trump won by 1 vote or 10,000 votes in PA, I think that the random noise injected into the simulations a week prior to the election would be enough to make it so that Harris would have won in >10% of elections in the simulated worlds that started a week before the election. But the further the election is from being close, the less of a difference that the random noise makes, and with what actually happened I think it's reasonable to predict that Trump would win in >90% of the simulated-world-elections (which I'm calling "true probability" even though my definition is arbitrary and "true probably" shouldn't have an arbitrary definition.)
All this said, might there a good reason why forecasts or market prices were only ~50-60%, even if I'm right that the "true probability" was >90% that Trump won?
Yes.
There is (what Philip Tetlock calls) "irreducible uncertainty" in the world. For forecasters trying to predict how a D6 will land, this may result in them making forecasts close to 1/6th that it lands on 6, even though they have video footage of the die flying through the air. This is because their uncertainty in the measurements introduces too much noise, and if they're slightly wrong about how the die will strike the table then that will completely change how it will land. So even though someone with a perfect knowledge of physics and the state of the world when they make their forecast midflight could more accurately predict how the die will land, the forecasters knowledge is limited and they cannot possibly know better. There is inherent uncertainty that they can't get rid of given the information they have access to.
Was 50% or 60% the best forecast one could make given the irreducible uncertainty, and assuming I'm right that the "true probability" was >90%?
Not necessarily.
On November 4th I wrote:
"My 2/10 low information inside view judgment is that Trump is about 65% likely to win PA and the election. My all-things-considered view is basically 50%.
However, notably, after about 10 hours of thinking about who will win in the last week, I don't know if I actually trust Nate and prediction markets to be doing a good job. I suspect that there may be well-informed people in the world who *know* that the markets are wrong and have justified "true" beliefs that one candidate is >65% likely to win. Such people presumably have a lot of money on the line, but not enough to more [sic] [move] the market prices far from 50%." (https://www.facebook.com/.../pfbid02kHsKK9oWUKGo3bvhs9wV9...)
In other words, I thought before the election and I think now that people more informed than me (I said I was only 2/10 informed) might be justified in making their all-things-considered believe >65% in favor of a particular candidate (unlike me, who meta-updated downward from my inside view toward believing the markets and better informed forecasters; not my "50%" in the quote was not meant to say that I believed the forecasters/Nate more than the markets, just that I didn't trust my inside view judgment of 65% Trump due to the markets and forecast aggregations assigning lower probabilities and me knowing I was relatively uninformed.)
Why do I think that? In short, because forecasters aren't close to being at their peak strength, and political prediction markets are nowhere near as efficient as e.g. stock prices. As Scott said, there probably weren't any big professional trading firms betting on Polymarket, let alone many like in stock markets. It just wouldn't be surprising if a very-informed person or group was able to correctly identifying that the prevailing wisdom of November 4th about how likely Trump was going to win was wrong.
So, expanding on "not necessarily" above: Could Trump-wins have been priced too highly on Polymarket despite Trump having a "true probability" of >90% of winning?
Yes, but (1) Scott and I and all of us reading this don't know if that's the case.
And (2) us learning now that Trump won and had a >90% "true probability" of winning is strong evidence that the hypothetical very informed forecasters I mentioned (whom may have actually existed, as I said on Nov 4th I thought was ~50% likely) were people who were at >65% Trump rather than people who were at <35% Trump. And thus it's more likely that ~60% market prices were too low rather than too high, given the best forecasts that could be made given irreducible uncertainty.
Scott also says a lot of stuff re the French whale on Polymarket, some of which I'd critique, but I don't think it's as important as the fundamental mistake Scott is making (claiming that Polymarket's price on Trump winning was too high).
strong agree
I'm reminded of that scene in hpmor
If you want to criticize your opponents' reasoning for being incorrect, at least wait for them to actually be factually mistaken
It feels very strange for him to be complaining about a probability estimate that was directionally correct compared to the one scott proposes
If lottery tickets are actually negative sum, then you will get lots of opportunities to criticize people who buy them and then lose money
It feels like Scott is currently in the position of somebody who has been pro-lottery his entire life, and then suddenly posted his very first criticism of the lottery the day after winning the jackpot
Maybe he's even correct, but seriously?
Well actually I'd push back on this. I think there is a true thing that Scott could have said in his post that is not far from what he actually said.
If we consider a different question where there is more irreducible uncertainty in the than there is for a Presidential Election, such as a soccer game or the role of a die (see this comment-reply of mine above to someone else: https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76065426), such that learning the result that actually happened only causes us to update our estimate of the "true probability" that the event was going to happen to, say, 60%, rather than >90%, then I could see a post very similar to Scott's post being reasonable.
Suppose that Polymarket put the result of the soccer game at 70% that Team A wins, and Metaculus puts it at 55%. Then Team A wins. Scott could then write a post saying "Hey people, just because Team A won doesn't mean that Polymarket's 70% was better than Metaculus' 55% forecast. I still think Polymarket's 70% was price too high and that Metaculus' forecast was better."
This is related to what Tetlock calls the "wrong side of maybe" fallacy. (Hey, I've been thinking about this for 8+ years: https://imgur.com/gallery/wrong-side-of-maybe-fallacy-uG3rgGU ). Just because a forecast was on the wrong-side-of-maybe doesn't mean it was wrong or bad. It's evidence that it was bad, but not decisive evidence, and there are cases where it is not actually bad.
More generally it's also true that just because a forecast is less-close-to-the-result-that-actually-happened doesn't decisively mean that it's worse than a forecast that was closer to what actually happened. Again, it's generally *evidence* that it's worse, but that's not definitively the case, and there are exceptions. E.g. People widely regard the Dilbert guy Scott Adams as being overconfident when he stated 99% Trump in 2016.
If Polymarket had been at 99% Trump for a week before the election, yet the result had turned out exactly as it had (same number of votes everywhere, Nate and Metaculus thinking ~50%, etc), then saying that Polymarket had Trump winning priced too highly would be reasonable.
I think you can only conclude that with a larger sample size of predictions. If outcomes that Polymarket assigns a 60% probability to actually happen 60% of the time, then Polymarket is well-calibrated.
"If outcomes that Polymarket assigns a 60% probability to actually happen 60% of the time, then Polymarket is well-calibrated."
Yes, that is the definition of "well-calibrated".
However, "well-calibrated" does not mean that the Trump actually wins 60% of the time, nor does it mean that 60% is the best forecast a highly-informed, rational forecaster can make on the question.
"I think you can only conclude that with a larger sample size of predictions."
I'm not sure that I know what you are referring to by "that". Is it my last statement? That: "If Polymarket had been at 99% Trump for a week before the election [...] then saying that Polymarket had Trump winning priced too highly would be reasonable."
If so, let me clarify that it'd be reasonable because we'd know that market actors, like Scott Adams, didn't have the actual knowledge to know that Trump would win 99+% of the time, even if it was actually the case that he would win that much. Market actors have epistemic uncertainty that would prevent them from knowing that. So we'd know that the market price was irrational, even if it happened to correspond to the true frequency of Trump winning.
If the market is well-calibrated, then if it assigns 60%, I conclude that the available evidence is consistent with a 60% probability.
I don't think Scott Adams was taking 100-to-1 bets. I think he was BSing.
"If the market is well-calibrated, then if it assigns 60%, I conclude that the available evidence is consistent with a 60% probability."
It sounds like your claim is that the available evidence is consistent with the aleatoric probability that Trump would win was 60%. I wrote a concise argument against that just now in this comment[1], which I'll copy here:
"The January 2024 Less Wrong post "Will quantum randomness affect the 2028 election?" makes clear that uncertainty about election outcomes is almost entirely epistemic uncertainty, not aleatoric uncertainty. Thus, whichever outcome happens is very strong evidence of what the true aleatoric probability of that outcome happening was. Specifically, it is very strong evidence that the outcome that happened was very likely to happen. Therefore, a day before the 2024 election Trump was very likely to win (>99%). Therefore, Polymarket's ~60% price was priced too low, not too high. https://www.lesswrong.com/posts/HT8SZAykDWwwqPrZn/will-quantum-randomness-affect-the-2028-election "
Here is an explanation of what "aleatoric uncertainty" is, from this source[2]:
"So aleatoric uncertainty is, if I tell you I’m going to flip a coin, you’re roughly 50-50 on whether the coin will come up heads. But you have almost no epistemic uncertainty, because you’re quite confident that your distribution is correct. And there isn’t a way to reduce it further with a practical amount of knowledge. If you knew the whole state of the world, you could get it to zero, but you don’t. So in that situation, you have all aleatoric uncertainty and no epistemic uncertainty. So in that case, you don’t expect to learn more by getting more data about the problem. And in the epistemic case, I flip the coin, it lands, I close my hand and now I’m going to reveal the data point, what the coin is. And so I’m still 50-50, but I have no aleatoric uncertainty and all epistemic uncertainty. I just don’t know the answer, but there is definitely an answer to be learned."
[1] https://www.astralcodexten.com/p/congrats-to-polymarket-but-i-still/comment/76263538
[2] https://www.lesswrong.com/posts/bLr68nrLSwgzqLpzu/axrp-episode-16-preparing-for-debate-ai-with-geoffrey-irving
(disclamer: didn't read Scott's article, got here via your comment on FB)
(1b) does he mean that the market price does not reflect the most-informed credence that humans can come up with?
Surely yes? (basis: it's always seemed to me that Scott and I conceive of probability the same way, i.e. if I have already flipped a coin but no one has looked at it, we should say the probability of heads is 50.5% or so, notwithstanding that the "real" probability is 100% or 0%). Or perhaps something further in this direction, like: almost nobody had better information than Nate Silver, and if a few people did, they probably didn't have the resources to move the market that much. (by analogy: one man has seen the coin, but not shared the results with anyone, and isn't rich enough to move the market on it)
The best argument in favor of "markets priced correctly" that comes to my mind is "wisdom of crowds": if enough low-information people in swing states were betting, that might aggregate enough information to set the price correctly even if none of the individual betters had enough information to do so.
"Surely yes? (basis: it's always seemed to me that Scott and I conceive of probability the same way, i.e. if I have already flipped a coin but no one has looked at it, we should say the probability of heads is 50.5% or so, notwithstanding that the "real" probability is 100% or 0%)"
I agree with this. (Note that "real" probability would more appropriately be called "aleatoric" probability)
"Or perhaps something further in this direction, like: almost nobody had better information than Nate Silver, and if a few people did, they probably didn't have the resources to move the market that much."
I think you're assuming unjustifiably that Nate's forecast was near the best quality possible. Even if he was very informed, he may not have applied the information he had most appropriately to form a forecast. (Two forecasters given access to the same information about a forecasting question often have different credences, due to different background knowledge from their life, different methodologies, etc.)
"and if a few people did, they probably didn't have the resources to move the market that much. (by analogy: one man has seen the coin, but not shared the results with anyone, and isn't rich enough to move the market on it)"
The French trader moving the market toward Trump may have been less informed and irrational relative to Nate, and may have effectively been betting on Trump randomly and just got lucky. We can stipulate that that was the case and I claim that still wouldn't show that the "correct market price" should have been less than 60% rather than greater.
I'll make a stronger claim: Even if *all* the Metaculus forecasters were better informed forecasters who were stronger / higher skilled at forecasting in general than *all* the Polymarket traders, that still wouldn't show that the "correct market price" ought to have been closer to the Metaculus aggregation than what the Polymarket market price actually was.
Why? Because the Metaculus traders, despite being higher skilled and more informed forecasters than the Polymarket traders were likely still not making the best estimates possible, despite inevitable epistemic uncertainty due to irreducible uncertainty in the world that would prevent them from ever being able to be as confident in the outcome as they would be if they had zero epistemic uncertainty and their credences matched the aleatoric uncertainty perfectly.
So the fact that the Polymarket price may have been largely what it was due to low-information traders is actually pretty irrelevant to the question of whether "correct market price" ought to have been what the Metaculus aggregation was or something else.
We don't know precisely how much the best possible forecast could have reduced epistemic uncertainty, but I don't think we have any reason to believe that reality tricks people into doing worse than chance on this question when they try to become very informed. On the contrary, this is the sort of normal forecasting question where we'd expect it to be possible to be able to get some evidence in the right direction. This implies that the "correct market price" should be >50%.
But how much more than 50%? I don't know precisely, but I'd be surprised if one couldn't get beyond >60%. As I wrote on November 4th, I suspected that there actually were highly informed people who were justifiably above >65% in favor of a particular candidate, though I didn't know which candidate. In other words, before the election I believed that it was possible to reduce epistemic uncertainty by enough to get over 15% of the way away from 50%. And that was before I had the benefit of hindsight of knowing that aleatoric uncertainty was >99%, and thought the election could have been close such that aleatoric uncertainty would have been <90% in one direction or another.
The fact that Nate and Metaculus was only at 50% does not at all show that it wasn't possible to reduce epistemic uncertainty enough to reach a credence of >65% (including both epistemic uncertainty and aleatoric uncertainty).
So I think the "correct market price", i.e. the price that the market should have been at if it were reflective of the credences of forecasters who had reduced their epistemic uncertainty as much as possible given inevitable uncertainty in the world, was probably about 65-90%, not 60%, and definitely not 50%. This 65%-90% is still much more uncertainty than a zero epistemic uncertainty that matches the true aleatoric uncertainty (what you called the "real probability"), but it's not so uncertain as to say that humans can't reduce any epistemic uncertainty they have when it comes to election results.
Here's a question I haven't seen much discussion of:
Should the 2024 election results make one update, to however small a degree, on whether the 2020 election was honest and accurate? Not merely the fact that Trump won, but more micro-data, like numbers of registered voters participating in particular districts, etc. Is there now new reason to hold that some of the 2020 results look less credible?
Looking at county by county data https://www.nytimes.com/interactive/2024/11/05/us/elections/results-president.html what strikes me is the uniformity of the shift. If there was an election-turning level fraud in 2020 then we'd probably expect to see the swing being particularly concentrated in particular states and counties, but as far as I can figure out we don't.
This doesn't disprove all 2020 fraud but it places limits on it.
If the Dems couldn't steal the 2024 election when they were running the government, how could they do so in 2020 when they weren't?
I like Yudkowsky's idea of not critiquing people's behavior on the occasion they turned out to be right. How about following Theo's wallet for a bit longer to wait until his method fails?
'I have never heard of a hedge fund betting on a prediction market....' Well, I can tell you from personal experience that SIG was very involved in one of the busiest markets. Maybe update your priors on that one. :)
Your argument assumes that all we learned on election day was that Trump won. What we actually learned was that he won by a substantial margin, took all the battleground states, lost NY state by 12% instead of 20%, took Texas with an even larger margin than he lost NY by. That's additional information that you are ignoring.
If Silver was right, the probability of Trump winning was about .5 but the probability of his winning by this much was considerably less than that. If Polymarket was right, his winning by this much was not surprising, since it is in a world where his expected margin is significantly more than zero that the probability it will be negative is only .4.
Even per Silver, the Trump margin wasn't actually that unexpected. See: https://x.com/NateSilver538/status/1848060370268938538.
A fair point, but I don't think that is true of other forecasters and I don't see how you can deduce it from the non-money markets. How much Scott underestimated the change in estimated reliability from this event is a much more complicated problem than I am prepared to deal with. My point was that he left out an important part of the information. I think Silver's odds on a popular majority were about 25%, for what that is worth — I don't know what the Polymarket odds were.
You can see Polymarket on the popular vote win here: https://polymarket.com/event/presidential-election-popular-vote-winner-2024?tid=1731020014331.
It shows Trump at 24% on October 1, 34% on October 21, 36% on November 1, 27% on the afternoon of November 5, over 99% by the early morning of November 6.
I correctly called this election with a much simpler methodology: I just read whatever the smart libtards were predicting and adjusted their takes by how biased I think they are. Scott, pretty smart, so I'll read his stuff, but since he has a slight liberal bias I'll give him a liberal mind virus score of 4/10. Nate Silver, also smart, but his liberal mind virus score reaches 5/10. etc. This gave me a result pretty close to reality.
The difficult part, of course, was giving the LMV score to each pundit. I went with ordering them around from least to most libtard and just going +1 +1 +1.
It's saddening seeing Scott fall for all the same biases he has fought his entire life against.
'All the other real-money markets rose close to Polymarket’s level because of arbitrage'
1) This is not how the arbitrage works, in fact you have it backwards here. Because PolyM was the biggest market, arbitrageurs would be *selling* Trump there heavily and buying Trump elsewhere, spread out among the other markets. They would be pushing Trump odds down below 60%. Not raising the others as high! Whether the fair price 'should' have been 57% is partially unknowable.
2) The WSJ article wrote that French Whale had put in $30m into Trump wins POTUS, but even raise that to $60m. The handle on PolyM was over $3.6 *Billion*. The Whale was a Minnow in the grand scheme of things.
3) You fundamentally misunderstand the nature of +EV and Kelly betting, it does not matter that this opportunity was a 'one-off' event [which it's not, there's always elections and other betting] you should have bet Half-Kelly [to account for less than perfect info] and not a random amount 97% smaller, if you were interested in maximizing your +EV.
The 'this is only happening once, so I can't bet what I should' is a well-known fallacy. Each horse race on each track every day is a unique event, if you have a 10% edge on GreyMareCryptoHODL but bet far less than you should because she is only racing at Aqueduct, one time, ever, versus these 9 other horses than you are doing it totally and completely wrong.
An easy example to show this is a coin flip where the true odds of Heads are 99.999% and Tails is 0.001%. I offer you the choice to bet up to $1m on heads, but only once. If you're only going to bet $5k or $10k, you fundamentally misunderstand the nature of +ev gambling, odds, and risk of ruin.
Are you making an invalid counterfactual?
You state that "If Theo hadn’t bet on Polymarket, it would still have been at ~54%." How are you calculating this price?
I assume you are not factoring in that others buy into the market based on the price. That is if he had not bet, then the prices would have been different and others may have invested.
Meh.
$5M bet by a billionaires is pocket change.
The amount of bet is not meaningful unless the money bet is literally life changing.
Equally, comparing Metaculus to Nate Silver is comparing different tribe's voodoo witch doctors.
What is spectacularly clear in this election is that the anti-Trump biases in particular, and anti-conservative biases in general, in many of the polls are spectacular.
For example: Trump won the popular vote by 2%. Ipsos, Morning Consult, Forbes and Marist - the top "left" polls, predicted a Harris popular vote win between 2% to 4%. These incompetent morons' own fine print say the margin of error is 3% - so every single one of them was wrong and well outside their own margin of error.
These idiots' polls swung the overall averages to +0.1% Harris as of the evening before election day.
Is it any wonder the prediction markets were screwed? They were relying on accuracy from these incompetent outfits.
This gets worse if you drill down. In the pre-election ACX publication - I noted that no candidate has won only 1 or 2 of the 3 Rust Belt swing states since 1988. I take zero credit for this - it was Rich Baris of People's Pundit who pointed this out.
Yet from November 3 to the end of November 5 - Michigan and Wisconsin had Harris favored by over 1.5 to 1 vs. Trump (i.e. Trump under 40% chance to win). Why? Because the polls said so - heavily skewed by the above polls. This despite Pennsylvania and all the other Swing States showing Trump in the lead DESPITE the garbage pollsters.
Nor is this a one election phenomenon.
In my same post - I noted the same day election poll results then (i.e. 2024) vs. the same day in 2020 and 2016 - and linked directly to the RCP page showing this information. I then contrasted the actual results vs. what the polls showed in the previous elections.
It is now the 3rd Presidential election in a row that the RCP average has been skewed by the above garbage polls to favor the Democrat candidate by a huge amount (i.e. way above margin of error) - and so anyone relying on them (most people) would act accordingly.
GIGO
As for Nate Silver: I am looking forward to see if his one foray into putting his money where his mouth is - which he lost spectacularly - will result in him actually paying:
https://www.outkick.com/analysis/nate-silver-about-lose-100000-bet-trump-would-not-win-florida-more-than-8-points
Nate Silver bet $100K publicly that Trump would win Florida by 8 percentage points or less.
Trump actually won Florida by over 13%.
Epic Fail.
Agreed. Nate Silver is full of hot air and Scott is incredibly biased.
The actual electoral outcome was the one he calculated as the modal (single most likely). https://www.natesilver.net/p/the-model-exactly-predicted-the-most
Something I think consistently gets left out when looking at the efficacy of prediction markets is market manipulation. It could be beneficial for someone to make a huge bet for no rational reason at all, but simply because it benefits them to move the markets.
For example, someone with a lot of money who wants Trump to win might make a large bet just to move the markets, to make things look hopeless in a way that affects turnout.
Or, someone might engage in a classic "pump and dump" scheme. Invest heavily in one candidate (for no reason at all) to get it to look like that person is surely winning, let other traders jump on the bandwagon, and then sell before the results are in.
"Efficient market hypothesis" folks tend to accidentally assume that markets are infinitely deep, and that's just not the case. Even real markets are not immune to this. Large Wall Street funds might have incentive to push markets around occasionally, even large markets like the S&P! If you can afford to push the price a bit to influence some exotic options contract you're in, and the cost to push the market is less than the profit you're going to make, why not?
It's enough of a problem that there are rules around market manipulation, and those rules get broken all the time.
It feels like rationalists think there is some infinitely deep pocket out there, that has perfect information, and is ready and willing to correct any mistake the market is making. That's simply not the case.
There must be two sides to every trade. Either insider trading is rampant, in which case you need a lot of fools betting against the insiders, or insider trading is rare or impossible, in which case you just have a bunch of people trying their best at guessing an unknown outcome.
Agreed. While there is a lot of money in the election betting markets - a Presidential campaign in which $2.6 billion was spent makes say, $30 million used to skew numbers is little more than a single small state ad spend.
There were lots of fools betting. The biggest bettor appears not to be someone trying to manipulate an outcome, but merely a French speculator who saw what he concluded was a mispricing he could make lots of money on.
Have we discussed the bias created by one's own desires yet?
Scott wanted Harris to win, and he thinks the process which gave Harris higher odds was more accurate. I wanted Trump to win, and I think the process which gave Trump higher odds was more accurate.
The urge to believe the set of numbers that tells you the story you want to believe is strong. I don't know how to correct for it.
I do.
Whenever I know that I prefer one side over another - in any probabilistic situation - I add a reasonable few petcentage points to the other side.
This is most effective when I'm at 50/50. I choose the one that I would less prefer.
I largely agree for similar reasons, but I would go further: I'm skeptical that either probability is all that useful a signal for an event that only happens once, unless the odds get pretty extreme. Suppose we have two scenarios:
* It will rain on wedding day
* It will be sunny on wedding day
Whether it's a 50-50 chance or 60-40 chance, you have to plan for both. It's not until it gets to 90-10 or higher that maybe you can think about not planning much for the unlikely scenario, if it doesn't matter much if you're wrong. And the more severe the consequences, the more you have to consider rare scenarios. A 10% chance of death is pretty darned high. Better write your will.
In situations that are similar to an roughly-even election, numeric odds are not much more useful than saying "it could go either way" for planning. (This doesn't apply for gambling or other situations where you can make repeated bets.)
Perhaps I'm biased by working as a programmer. We write code to handle rare events without doing any formal probability calculations. What would be the point? Even if you estimated the odds correctly, such estimates are fragile.
As an individual, I have a hard time getting excited by the difference between 50:50 odds and 60:40 odds for this last election.
Now, admittedly, I have _relatively_ little at stake here - and the part that is at stake (e.g. whether there is a nuclear exchange) would not be well-constrained by even perfectly knowing the results of the election in advance.
Now, there are public policy questions where the difference between 50:50 and 60:40 for e.g. the odds for another pandemic in N years matters.
For me as an individual, even for fairly dire events, the odds have to reach something like 90:10 before I feel comfortable disregarding the low probability side.
To phrase it as a question: _Other_ than for the question of whether to place a bet, are there _any_ current prediction markets where anyone reading this would take some personal action differently depending on whether the market odds were 50:50 instead of 60:40?
You are interested in the question: Is Polymarket more reliable (or some variant of that).
But a different question is: why was a blowout predicted by polls (and poll-adjacent, like Nate Silver) as 50%?
The link between the two is Theo, and not in the sense that he guessed lucky, but in the sense that he seemed to realize that the polling industry was as morally and competently bankrupt as the news industry, and put together a somewhat more functional alternative. Polling is a wasteland because of
- push polls and similar nonsense, which mean that anyone who isn't terminally lonely or braindead hits "end call" the moment they realize the caller is doing political polling
- our betters running our institutions have made many people scared, even in the context of talking to a stranger, of voicing opinions that might get them cancelled.
Theo worked out a scheme for dealing with the second of these, if not the first, the "neighbor" method, and got the results he got.
https://www.business-standard.com/world-news/how-a-french-trader-predicted-trump-s-victory-by-asking-about-the-neighbour-124110701151_1.html
The obvious takeaway is that some sort of alternative to current polling is required, in the sense that X is an alternative to the pathologies of the traditional news media.
Markets may play some of this role, but plenty of people who are not comfortable with the specifics of how a market works, or just don't care enough, are happy to answer questions (IF the questions are honest, a big if -- as I said, push polls, or even the GSS which is clearly a push poll even if the people in charge of it are so philosophically ignorant and clueless about the rest of the country that they can't see it; and if they don't feel there's some sort of blowback possible).
I'm not sure where this goes. Maybe X-Polls, if in 4 years X is considered trusted enough by a wide enough range of people? Maybe polls as short games on ad-supported platforms? But I think the current industry is kaput, they lost their last shreds of credibility this election. Even those, like Nate Silver, who probably aren't deliberately biased, just did not do enough to cope with the social environment in the country - in spite of the fact that this environment was clear to even a French outsider...
Polls did pretty well though. It wasn't an unusual polling error at all.
Say I have a coin that I suspect of being biased, so I make a market to figure it out.
An anonymous whale spends millions of dollars performing a detailed scan of the coin, builds computer model of the coin, and flips the simulated coin 1000 times. Then he bets heavily that the coin is biased, which substantially moves the market in that direction.
In this scenario, nobody would argue that the market is now mispriced. Right? The whale introduced data into the market. The whale’s model is incorporated into the wisdom of crowds that is the market. Another way of saying this is that the whale *knows that the coin is biased* and he is informing us of this fact, up to the limit of the amount of money he’s comfortable betting.
Is it possible that a different whale comes along and bets an equal or greater amount in the opposite direction? Yes, but it’s *less likely* because the coin is in fact biased and proper study and analysis of the coin should indicate that. An expensive study that says the coin is fair (or biased in the other direction, or whatever makes sense in this metaphor) would imply that someone spent a lot of money and made a huge mistake in their analysis. Can this happen? Sure. Mistakes, mismeasurements, and misjudgments are always possible. But you have to think that in a reasonable-ish world, the results of analyses should correlate with the actual state of the actual coin.
The thing about elections is that the outcome is predictable. We call such things “stochastic” as a shorthand way of indicating that we can’t practically know all the details that would be required to get the right answer. But Omega would simply figure out the answer, but building a detailed simulation of the entire Earth and everyone on it, and running it. A really expensive study commissioned by a French whale is not quite at that level but it’s somewhere on the spectrum between Omega and “some guy’s opinion.”
What bothers me is how whales are getting access to crypto. Is this what those guys are doing with the LLM that speaks whale?
How did the whale validate the code that he used to model the coin flip?
If the answer is "I dunno, but the idea of modeling the coin flip is clever and neat and he clearly put a lot of work into it and is confident of the result", then Oh Lord No. That's the formula that, in other contexts, gives us the Replication Crisis. If that's the formula that will set the price in a prediction market, then that market's predictions will at best be *slightly* better than random chance.
Typically you would do a sensitivity analysis to understand the impact of your assumptions on the outcomes. This is how modelers across all domains understand their model uncertainty, and their measurement uncertainty.
Perhaps he would conduct the same procedure on a coin known to be fair and verify that the simulation approach calculates that this coin is in fact fair.
Since this is a metaphor for the actual polling approach that took place, the truth is that we don’t know what he did to validate or analyze the model, all we know is that he put up the money and then won.
I think you sort of have to have a prior that is something like “don’t assume maximum uniform incompetence at all scales at all times.” It’s possible that the computer modeling experts simulating the coin flip forgot to convert from Imperial to SI and got a flagrantly stupid result, which nobody bothered to check for some reason. But there’s a kind of conceptual convexity to a lot of science/engineering tasks that means if you make a significant error in one part of a model that will show up as a weird result or inconsistency elsewhere in the model, and then you are forced to track it down. Perhaps this convexity doesn’t exist in soft science domains. It would be “hard to miss” such a mistake in the coin flip simulation example. You would notice problems and inconsistencies, unless you were just very incompetent, and if you’re very incompetent, why do you have tens of millions of dollars to bet on this question? Is wealth uncorrelated or negatively correlated with competence? Probably not, because if it were so, prediction markets wouldn’t work at all.
Finally, I think maybe your comment here and Scott’s OP indicate a mindset that is perhaps excessively outside-view. If you were sitting at a bar talking to a friend and they casually described the “neighbor polling” method and then told you that it suggested a massive advantage for Trump, you would probably not reply with the sort of Cartesian skepticism that I’m seeing in these comments — you would probably have said, “Oh word?” and adjusted your prior.
Tbh is seems to me that the consistent generalized for me of your argument would imply that prediction markets cannot possibly work as methods of information aggregation because people are sometimes mistaken. I’m sure that’s not what you mean or what you think. But if you have an epistemic carveout for “whales,” i.e. “bettors who have resources and confidence,” then on net you’ll just end up throwing away Bayes points.
"I think you sort of have to have a prior that is something like “don’t assume maximum uniform incompetence at all scales at all times.” It’s possible that the computer modeling experts simulating the coin flip forgot to convert from Imperial to SI and got a flagrantly stupid result, which nobody bothered to check for some reason. "
Have you ever done any serious computer modeling? Because I have. I am, in fact, taking a break from a fairly substantial modeling project to wander through the ACX garden right now. And I think I'm on pretty solid ground in expecting a *lot* of incompetence in the field, albeit not uniformly distributed.
Oh, and the errors that matter aren't usually silly mistakes like "Oops we forgot to convert to SI", but "Hey, the (n,7Li) reaction cross section is really tiny under these conditions, we don't need to waste clock cycles tracking it, do we?" Serious modeling, or polling, is filled with the opportunity to make mistakes like that. And I don't count it as incompetence when people make those mistakes. But it all too frequently crosses over into incompetence when people place high confidence in an untested model and e.g. neglect to fully evacuate the test area around Bikini. Betting forty million dollars on a single run of your untested model would be a lesser degree of incompetence, to be sure.
"If you were sitting at a bar talking to a friend and they casually described the “neighbor polling” method and then told you that it suggested a massive advantage for Trump, you would probably not reply with the sort of Cartesian skepticism that I’m seeing in these comments — you would probably have said, “Oh word?” and adjusted your prior."
I can assure you that while I would in such a situation politely say "Oh, that's really clever!", in roughly the same way I'd say that to a Las Vegas gambler explaining to me that he has a System that will let him Beat the House. I might ask about the details of validation and testing, but if those aren't forthcoming I will be changing the subject and the only prior I will be adjusting is the one about his being an overconfident fool.
That doesn't, and shouldn't, change if I then see him place a single bet on Red at the roulette wheel and win.
You have activated my trap card. I am a computer modeling specialist with a PhD in programming computer models of various house al systems in Fortran and a 15 year career in applied simulation.
Unfortunately though you didn’t say much here that I disagree with so I don’t have too much of a rebuttal.
I’ve seen a lot of shoddy modeling in my day. The Imperial to SI remark was referencing the Mars mission that was lost due to exactly that error. My intent wasn’t to say that it’s impossible to do modeling badly, and I can see how it came across that way.
On the other hand, I’ve also seen a lot of incredibly good modeling work that was disregarded because the answer output by the model disagreed with the institutional wisdom of the organization that commissioned the model. I think I have personally seen probably ten billion dollars wasted in my own field due to people not taking seriously the conclusions of simulation, not for any principled reason but because “simulation is untrustworthy.” And anybody can always endlessly nitpick the assumptions of a given model, because even career modelers like me will admit that all models are wrong. You do the best you can with the tools and time at your disposal and if the client or stakeholder wants an excuse to disregard the work, they will always find it.
This is why I am somewhat sensitive to the blanket skepticism that some lone nerd might be able to beat the house. Lone nerds beat the house all the time. We live in a world dominated by lone nerds who made insane bets and won. It is the main way of getting ahead in this world.
Final remark here is that the original conversation wasn’t really about modeling so much as data gathering. The French Whale performed a new kind of survey, basically. So metaphorically this is less like making a computer simulation of a coin flip and perhaps more like using a specific tool that simply measures the imbalance of the coin’s center of mass, in a circumstance where nobody else has done that sort of analysis. Viewed this way the French Whale is very directly and straightforwardly adding data to the wisdom of the market. It seems that if this character had bet $1000 instead of tens of millions we wouldn’t be having this conversation at all, and that fact seems to indicate that the claim is something more like “he was overconfident” (not necessarily true if he is very rich) or “this bet distorted the odds too much” (he was correct; he won the bet, so his bet *correctly updated* the odds in this case.)
I don’t know if you’re familiar with Yudkowsky’s
>"Hey, the (n,7Li) reaction cross section is really tiny under these conditions, we don't need to waste clock cycles tracking it, do we?"
Bravo, Bravo! :-)
I like this piece, but it overlooks one crucial fact -- how much you should update your probabilities ought to depending on the scale of victory. Election outcome isn't binary -- win or loss. To take an illustrative extreme example: suppose your prior is 90% that probability is 50-50 and 10% that it's 60-40. Then, election results come out, and Trump has swept 48 states. Your prior ought to be updated with greater magnitude than a closer victory.
So, here's the elephant in the room: what do y'all think Trump *will* bring...?
...Why are you acting like he didn't tell everyone exactly what he would do? There's nothing stopping him anymore. No need to be concerned with being re-elected (legitimately), immune to consequence, full control over every branch of government, and most importantly, he has the consent of the public. This is what the people want. Would you deny them catharsis?
Can you be specific? And no, "Trump will end democracy" is not specific.
Prosecuting political opposition, bombing Mexican cartels, ending the Ukraine war by ending support to Ukraine, a bunch of tariffs, using military to stop protests and crime without consent of the states, deporting as many illegal immigrants as possible (and legal immigrants if they're undesirables), performing an ideological purge of federal agencies, and a bunch of other stuff I'm probably forgetting.
You can do a lot in four years when you actually have ambition.
>Prosecuting political opposition
I doubt this will happen on a large scale. His track record on this one is already weak(muh swamp, muh lock her up). But he will try to get a bunch of people that particularly annoy him fired(he did this in the past).
>bombing Mexican cartels
This sounds really based. I doubt he will do this though. Or, if he does, it will be probably a performative bombing with consent from the Mexican government.(Military operation decisions in his first administration always involved a lot of talk with generals, and they were mostly successful in dissuading him from doing truly unhinged stuff).
>ending support for Ukraine
Possible. But he sends mixed signals on the Ukraine issue all the time. According to him he will simultaneously save Ukraine by being the toughest president against Putin in history, save everybody by signing the best deal in the history of deals, and kill NATO because the European countries don't give enough money to the alliance.
>bunch of tariffs
It can happen, it would suck. But the GOP may dissuade him, it is not clear that the rest of the party is on board with his populist economic stuff.
>using military to stop protests and crime without consent of the states
This sounds based. I hope he does this. Him being a pussy during the George Floyd riots and refusing to send the military probably hurt his image. Also, the US military is pretty well trained compared to your average cop, they will not go around slaughtering random protesters, they will probably end the chaos pretty efficiently.
>deporting as many illegal immigrants as possible
Probably will happen. It is the second campaign he runs promising to deport everybody, if he fails to deliver AGAIN, I can see even hardcore Trumpers turning on him.
>performing an ideological purge of federal agencies
Elon will probably champion the idea of firing a bunch of people, this seems to be the idea behind the DOGE thing(holy cringe name). Trump will probably just want to fill a bunch of positions with his most loyal buddies(like he did the first time). We'll see who wins. At the end of the first month we either have a Milei style purge, or Dana White as Secretary of State(maybe both?).
Of course, there is the fact that Trump is really unpredictable, so he can very well find very creative ways to do shitty stuff that we will never see coming and can very well be worse than all the stuff you're expecting. But, I'll maintain that most of the stuff you're predicting won't happen.
>His track record on this one is already weak(muh swamp, muh lock her up).
Agreed. I come to the same conclusion on this one because of the same datum.
"I can see even hardcore Trumpers turning on him"
I cannot see this. Hardcore Trumpers will find some reason to support anything he does, just as hardcore Democrats will support the candidate under any circumstances.
"the idea behind the DOGE thing"
The DOGE will do what a DOGE does, when in the presence of a Donald doing diplomacy, that is.
But he'll do some bad stuff too, right?
...You're free to gloat all you want. I just hope you're prepared to deal with the consequences of your actions. Just because you're on the winning side doesn't mean you won't be forced to make sacrifices.
It was more of a grim joke. I actually don't think he'll be good for the country.
I have given up on Ukraine and would be OK with bombing Mexican cartels, though, especially if he gets the OK from Mexico. I think even the Mexicans (probably *especially* the Mexicans) are sick of them.
>Just because you're on the winning side doesn't mean you won't be forced to make sacrifices.
nit: I gritted my teeth and pushed the button for Trump, but I don't think of that as being on the winning side. I think of him as being _marginally_ less bad than Harris (assuming that her administration would have more-or-less been a continuation of the Biden/Harris administration), largely because of Harris's anti-free-speech statements and corresponding Biden/Harris administration actions. But I don't gloat to have a liar in the White House, one who suggested looking into injecting disinfectant, with the implications of that for Trump's common sense.
He watched a lot of TV in his first term. He is not energetic, not smart, not well organized, not practical and not good at finding people who are. He's good at picking people who'll do what he wants, but not at figuring out who actually knows how to manage this and that huge, complicated novel task.
He is very energetic, specially for someone his age. But he doesn't seem to like boring policy stuff, he saves his energy for rallies, speeches and golf(the stuff he actually enjoys).
Politicians never keep their campaign promises and Trump never keeps ANY promises. No doubt among the many things you said he would do there are some he never had any intention of doing and there are some he genuinely wants to do. That's why OP's question would actually be fun to speculate about.
There were lots of things he said he'd do. He couldn't do all of them even if he met no legal resistance or political pushback. Four years is nowhere near enough time. Deporting 15 million people, for instance, is a huge job. The logistical problems alone are huge.
I'm excited about who could be in his cabinet. I like JD Vance, and RFK Jr. and Tulsi Gabbard..... Call me crazy if you must.
“You shouldn’t update much on dramatic events!” That depends very much on the event. A baseball game, no. A boxing match, absolutely yes. Just because the odds are 50% for each does not mean that the predictions are equivalent. And an election is a lot more like a boxing match.
In bayes terms, a fair coin flip has priors of 100% 50/50. A boxing match has priors of 50% 90/10 and 50% 10/90.
It’s a serious conceptual error to equate the two, and very common.
Can you construct an example of someone validly updating a lot on a boxing match? I still don't get why it's so difference from baseball.
Baseball has notoriously small edges for a great team. This years World Series winners, the dodgers, only won 60% of their 162 regular season games. If a baseball team wins a matchup it doesn’t tell you very much whether they will win the next matchup.
In boxing, like chess, the better boxer wins very often. Unlike chess, bouts are so rare that it is difficult to assess how two boxers compare unless they actually fight. So when two boxers fight for the first time, odds will be relatively even, but their next match the odds will be heavily favoring the one who won previously.
Several really terrible takes here:
A) predict it is not liquid. Liquidity is important for accuracy
B) we don’t know if the person who lost 5 million was hedging. My gut says they’re hedged.
C) the whale is known and he literally ran his own polls before the election which is why he was so sure. This is exactly why non money prediction markets are garbage. It’s just a bunch of people full of hot air like Nate silver speculating.
C)
Nate Sliver, I believe, lost a $100K bet on the election (FL Trump margin). Not sure if he put more money elsewhere, but suggesting he is just full of hot air is incorrect.
I don't think his proposed counterparty actually agreed to that bet.
oopsies on his part lol
Nate Silver is following the polls, if his data is 'off' for some reason, what do you want him to do? (shade them in the direction he would like?) (I guess he needs to find better data!)
Find a real job.
This reminds me of a similar story, except it ended with the whale losing.
See this paper: https://dx.doi.org/10.2139/ssrn.2322420
Rothschild, David M. and Sethi, Rajiv, Trading Strategies and Market Microstructure: Evidence from a Prediction Market (November 22, 2015). The Journal of Prediction Markets
The paper looked at Intrade's trade history for the 2012 election, finding that a single whale (codenamed “Trader B”) spent >$300K on election day betting on Romney, which propped up his winning odds from 20% to 30%. (Yeah, markets had less liquidity back then.)
Romney lost by a wide margin. Part of that loss was to arbitrageurs from other venues like Betfair.
The paper puts forth a few hypotheses:
“This leaves three possibilities: (i) the trader was convinced that Romney was underpriced throughout the period and was expressing a price view, (ii) he was hedging an exposure held elsewhere, or (iii) he was attempting to distort prices in the market for some purpose.”
You talk about adjusting priors based on “Theo” doing (i), but suppose if his goals included (ii) and/or (iii), is the same adjustment warranted? A banker should have large exposures in other financial assets, in addition to the yachts.
"prediction markets work because we expect nobody to have vastly more money than anyone else, giving everyone a fair chance to compare their opinions"
Given the actual inequality in the world, why wouldn't this cast prediction markets into serious doubt? It sounds like they are premised on something that is simply false.
It's a bit strange that people writing about "real money markets" ignore every sportsbook around the world. Sportsbooks took millions and millions of bets, more than enough to offset one French whale, and landed at 60%+ for Trump on the morning of the election.
Even if you ignore them, Robinhood alone created an election derivatives market that had hundreds of millions of dollars in volume and started the day with Trump at 63c. There are more real money markets, Horatio, than are dreamt of in your Manifest conference.
This is correct
Oh dear, I have no idea, but I'm nominating it for comment of the week.
I suspect a lot of the guys who do this stuff aren't big into sports. (Myself included.) Stereotypes are that way for a reason. :)
BTW, Googling 'dating in the second person' gives you articles on polyamory. Which I know you are an advocate of, but is not the subject of your blog. :)
I've enjoyed your past stuff, also BTW, but have aged out of the market and it's no longer applicable. But the 'dating as trading' metaphor is interesting; I spent way too much time trying to figure out what archetype to try to turn myself into. Had I read your article 20 or even 30 years ago, things might have been different.
Good luck with your blog!
Dude, take the L
The January 2024 Less Wrong post "Will quantum randomness affect the 2028 election?" makes clear that uncertainty about election outcomes is almost entirely epistemic uncertainty, not aleatoric uncertainty. Thus, whichever outcome happens is very strong evidence of what the true aleatoric probability of that outcome happening was. Specifically, it is very strong evidence that the outcome that happened was very likely to happen. Therefore, a day before the 2024 election Trump was very likely to win (>99%). Therefore, Polymarket's ~60% price was priced too low, not too high. https://www.lesswrong.com/posts/HT8SZAykDWwwqPrZn/will-quantum-randomness-affect-the-2028-election
I agree that election outcome uncertainty is epistemic, but that doesn’t mean that we should judge the market result by whether it aligns more closely with “objective” probability, whatever that even means! Heck, if you believe the universe is deterministic, then Trump winning was always 100%. But that bears no relevance whatsoever for whether betting for Trump on polymarket at 60% odds was the right bet. That is always going to be fundamentally a question about epistemic probability. Saying otherwise strikes me as very confused (plus it goes against the Bayesian framework that Scott adopts here).
So I see the points in the article about the initial perturbations in chaotic systems having huge effects as being irrelevant, because our epistemic data doesn’t come close to establishing the initial conditions with any sort of precision to allow us to predict those outcomes.
I think we should interpret questions like “were the true probabilities 50/50 or 60/40?” to be about the ratios of possible nearby universes where Trump or Harris won. In short, we have limited epistemic data about physical reality, and this data constrains the way our physical universe can be. But our data is also massively underdetermined, in the sense that there are a lot of ways the actual universe can turn out to be configured in a way that is still entirely consistent with our epistemic data. Questions about correct probabilities are thus just questions about the configuration space of nearby possible universes where all the epistemic facts are held fixed.
To say that the “true” odds were 50/50 is thus just to say that “holding all the epistemic facts before the election fixed, the ratio of possible ways to configure the universe where either Trump or Harris won comes out 50/50. There are just as many possible configuration spaces where Trump won as where Harris won, given the epistemic data”.
Of course there are questions about how to sort the configuration space, and I don’t have easy answers here, but I think in the main that’s broadly what people are gesturing at when they talk about these kinds of probabilities.
"That is always going to be fundamentally a question about epistemic probability. Saying otherwise strikes me as very confused (plus it goes against the Bayesian framework that Scott adopts here)."
Yeah, I agree with you, sorry for not making that clear--my initial comment was trying too hard to be concise and I forgot to clarify that I agree that "what the market price ought to be" (what Scott Alexander was debating) is not equal to the aleatoric uncertainty, but rather is more uncertain than the aleatoric uncertainty due to inevitable epistemic uncertainty that will remain.
It wouldn't be useful to say that market prices ought to exactly correspond to aleatoric probabilities, because that would be impossible with irrational traders given that real world traders are inevitably going to have some epistemic uncertainty. So I agree with you on the goal posts here.
"Heck, if you believe the universe is deterministic, then Trump winning was always 100%. But that bears no relevance whatsoever for whether betting for Trump on polymarket at 60% odds was the right bet."
Well it does bear relevance, because it's evidence of what the best forecast possible (given inevitable irreducible uncertainty in the world) was.
If aleatoric uncertainty was >99% Trump (as I claim the actual election outcome shows us is true), then the best possible forecast (reflecting both epistemic uncertainty and aleatoric uncertainty combined) was very likely to be >50% (since we have no reason to believe with >50% confidence that the universe fools people into doing worse than chance on this question when they become informed) and probably something like ~65%-90% (as I wrote in another comment elsewhere here).
It's still going to be <90% rather than >99% because of the irreducible uncertainty in the world, causing some remaining epistemic uncertainty even for the highest skilled most informed forecaster possible.
But saying that it's only 60% or less seems implausible to me, and we certainly can't establish that it's that low just by pointing out that Polymarket had a big low-information pro-Trump trader, or that Metaculus was at ~50% or <60% and Metaculus is pretty good.
The reality is that even people on Metaculus who we think were relatively informed, or even people like Nate Silver, were likely far from the limits of what is epistemically possible in terms of being informed and rational-at-making-forecasts-given-the-information available.
"I think we should interpret questions like “were the true probabilities 50/50 or 60/40?” to be about the ratios of possible nearby universes where Trump or Harris won."
Yes, I agree, if by "true probability" you mean "aleatoric probability". And note that before the election I had some weight, perhaps ~5-30% on the aleatoric probability being 10-90% Trump. If the election had wound up very close, e.g. PA being decisive and Trump winning by 1,000 votes, I think we'd be right to conclude then that the true aleatoric probability was not >99% Trump, but rather something like 10-90% Trump. (A good example is the 2000 election: The aleatoric probability the day before the election that Gore would win was probably 10%-90%, not <1%.)
I think that in forecasting contexts we can just ignore the hard-to-figure out quantum perturbation stuff (I don't know much about quantum mechanics) by using thought experiments that introduce noise to simulations. The way I was thinking about this is:
Suppose we re-run the universe a million times starting a week before the 2024 Presidential election, however we ensure that even if the universe is deterministic that we don't get the same outcome 100% of the time, by introducing small amounts of noise. For example, we cause a person to sleep-in 5 minutes later than they actually did in reality. Or we cause someone to get a flat tire on their way to work. Or we cause a tire that did get punctured to not get punctured instead. Etc.
Due to chaos / the butterfly effect, I think these sorts of interventions would be extremely likely to change the amount of votes that each candidate receives. In close elections, these interventions would even change who won some of the time. But they wouldn't change it so much that Harris would win almost half of the elections. I still think that in 2024 Trump would have won >>90%, probably <99%, and maybe even >99.9%. But not 100%. Less than 100%. (And if you think 100% still, then let's chance the thought experiment so that the simulations start a month before the election rather than a week, etc, to reduce the aleatoric uncertainty by enough that even in a million simulations with different bits of noise injected into each one, the same candidate wouldn't win all one million of them.)
"But our data is also massively underdetermined, in the sense that there are a lot of ways the actual universe can turn out to be configured in a way that is still entirely consistent with our epistemic data. Questions about correct probabilities are thus just questions about the configuration space of nearby possible universes where all the epistemic facts are held fixed."
I think we should distinguish between two questions: (1) What forecast should a forecaster make given the information that they have, and (2) What is the best possible forecast a forecaster can make given irreducible uncertainty in the world?
Re (1), a low-information forecaster should probably have had a credence of ~40-60% Trump, and would have been luckily-overconfident if they had had a credence of >60% Trump. We can say that this person's forecast was irrational if it was >60% given that they had sufficiently low information. They had no way of knowing the odds were >60%.
We can say this about the low-information forecaster *even if* it is true that the most informed forecaster possible could obtain a rational forecast of, say, 70% Trump. As I said previously in this comment, I think the answer to (2) is much more likely to be ~70% than <60%.
In other words, I think that the most confident the most informed person could have been in favor of Trump without being irrationally overconfident was probably ~70% (or ~65%-90% as I said before).
This does not mean that the French whale was not overconfident. This does not mean that Metaculus forecasters were necessarily underconfident--they could have just been ignorant, low-information forecasters assigning probabilities rationally given their high epistemic uncertainty that they didn't reduce as far as was physically possible.
"Questions about correct probabilities are thus just questions about the configuration space of nearby possible universes where all the epistemic facts are held fixed."
"To say that the “true” odds were 50/50 is thus just to say that “holding all the epistemic facts before the election fixed, the ratio of possible ways to configure the universe where either Trump or Harris won comes out 50/50. There are just as many possible configuration spaces where Trump won as where Harris won, given the epistemic data”."
So if by "correct" here and "true" odds you meant "rational, given the information I actually have", I agree with you.
But I don't think that's what Scott is saying in his post when he claims that the odds should have been and Polymarket was thus over-priced.
What I think Scott means by "true" when he says (your) "the “true” odds were 50/50" is "what the market price ideally should be, i.e. what it would be if it were determined by maximally-informed rational traders who reduced their epistemic uncertainty as far as possible, up to the limit of irreducible uncertainty."
And I don't think you can establish that without either showing that the person(s) you're talking about who were at 50/50 were maximally informed. And I don't think they were. I think maximally informed people would have been at ~65-90%.
>Suppose we re-run the universe a million times starting a week before the 2024 Presidential election, however we ensure that even if the universe is deterministic that we don't get the same outcome 100% of the time, by introducing small amounts of noise. For example, we cause a person to sleep-in 5 minutes later than they actually did in reality. Or we cause someone to get a flat tire on their way to work. Or we cause a tire that did get punctured to not get punctured instead. Etc.
>Due to chaos / the butterfly effect, I think these sorts of interventions would be extremely likely to change the amount of votes that each candidate receives. In close elections, these interventions would even change who won some of the time. But they wouldn't change it so much that Harris would win almost half of the elections.
Agreed re noise and chaotic dynamics. And, even setting aside quantum mechanics, the effects of thermal noise in a classical universe do very similar things...
>And I don't think you can establish that without either showing that the person(s) you're talking about who were at 50/50 were maximally informed.
There is one sense in which "maximally informed" is ill-defined. The number of people one could poll depends on the budget one has for it. To pick an even more extreme case, given an infinite budget, one could pay voters to tell their opinion (not to _change_ it, this isn't vote buying), and pay to establish a credible case that one will not disclose their answers and will not take any action for or against them based on their answers. In the limit, one polls every single eligible voter.
Now, this still will miss the voter that gets the flat tire, or the voter who changes their mind between the poll and the final vote.
But ... still, this is possible in the real world (no omniscient omegas involved). So, how close to this does one have to get to count as maximally informed?
"So, how close to this does one have to get to count as maximally informed?"
Good question, and I agree that my "maximally informed" was ill-defined. What I meant was that a "maximally informed" person is a person who reduces their epistemic certainty to near-zero and is only left with aleatoric uncertainty.
How close to zero epistemic uncertainty it possible to get?
In one sense, your person with an infinite budget who polls everyone will have extremely little epistemic uncertainty, and will only fail to predict who fails to get to the polls on election day do to a surprise wrench in their plans, etc. (Note that I think this hypothetical person would have been able to get to 90+% Trump.)
And in another sense, your hypothetical person is more informed than it was actually possible for anyone to be, since no one actually had the budget to go interview every eligible voter in a trustworthy way to get them to disclose who they planned to vote for and how likely they were to actually end up voting in a way for that person to reveal that information honestly.
So I think what I meant by "maximally informed" was actually a lower bar than your hypothetical. I meant basically 'as in formed as possible' in a practical way, rather than according to the laws of physics, or some higher standard.
Perhaps another way to say this is to point out that whoever you think was the most informed in reality (Nate Silver and his team?) weren't tackling this problem with a $100M budget, and probably weren't even tackling it with a $10M budget, but if they were, they would have been better informed. (They obviously weren't tackling it with the near-infinite budget that'd be required to gain the trust of every eligible voter and get them to honestly share with you the knowledge they have about who they plan to vote for, but that's too high a standard). Spending a $100M budget on hiring a team of forecasters and support to make a massive effort to forecast the 2024 election would have at least been realistically plausible. E.g. Someone who cared to make that happen could do that for the 2028 election. There'd be no guarantee that the project would make a good forecast just because it has a large budget, but it'd be much better equipped to become more informed and figure out the truth of who is more likely to win in an election that would not wind up being that close.
Had the 2024 election wound up like the 2020 election, even a team with a $100M budget might have struggled to determine the likely winner. Though perhaps that's because even in retrospect determining who was most likely to win the 2000 election is hard. But given that the 2024 election wasn't very close, I expect that very informed people (e.g. with a $100M budget and strong forecasting skills in general, would have been able to figure out that Trump was >60% likely to win, and probably >70% likely to win.
Presumably an extremely informed group that settled on e.g. 75% Trump before the 2020 election wouldn't have been literally maximally informed, but they'd be close enough for my purposes. And mainly I just want to say that I think it was theoretically possible for a group to become informed enough to come to a 75% forecast for Trump like that without being irrationally overconfident.
Many Thanks!
>Presumably an extremely informed group that settled on e.g. 75% Trump before the 2020 election wouldn't have been literally maximally informed, but they'd be close enough for my purposes. And mainly I just want to say that I think it was theoretically possible for a group to become informed enough to come to a 75% forecast for Trump like that without being irrationally overconfident.
That seems reasonable.
I think there are at least two other plausible candidates for a "maximally informed" bettor.
The higher bar one would be someone/some group that had access to all of the data from all of the polls that had actually been made. This is arguable, since some of the polls (such as Theo's) were completely private, and some (most?) of the polls that disclosed their aggregate results weren't disclosing their full raw data. So, these considerations make this a somewhat unreasonable standard (though still below my hypothetical of a pollster with an infinite budget :-) ).
The lower bar would be whichever person/group had the "most" poll data actually available to them. One of the snags in this bar is that the measures of the amount of poll data aren't quite comparable. The set of polled voters available to one group are unlikely to be a strict superset of the set available to some other group. If one group polled 20,000 voters, and another polled only 15,000 but did a better job of matching their polled voters to the likely-to-actually-vote population, who has "more" data?
Reasonable thoughts. Note though that poll data is far from the only kind of data that can inform a forecaster about the probability of Trump winning.
E.g. Forecasters could have learned about the probability of Trump winning by watching the Presidential debate, by watching Trump's interview on the Jake Paul podcast or the Joe Rogan podcast or Harris interviews, by seeing what kind of political discussions people are having on Twitter, by seeing what kind of reasons people give for supporting or opposing Trump, etc, by seeing how these things compare to what Biden looked like during his 2020 election campaign, by looking at voter registration numbers by party in different states compared to the 2016 and 2020 election, by observing whether your known Republican neighbor bothers to put up a Trump sign this year or not, by reading political philosophy and finding out whay Jason Brennan means by "hobits" in a political context, etc. All of these things provide information.
Your "That seems reasonable" reply suggests to me that you agree with me that Scott's claim that Polymarket had Trump-wins priced too highly is wrong. Is that the case?
Happy birthday Scott - you have brought me great joy, solace, reassurance and fascination at many moments in my life. I appreciate you and what you do
One factual error that bother me:
Other prediction markets did move in sync with Polymarket, the distance was usually 2-4%, the biggest gap I found was Preidctit vs polymarket oct 29th, about 7%... and it was quickly corrected.
I didn't arbitrage myself, for the same reasons many people didn't. Too much risk (of markets going under), too much fiscal hassle (banks and tax authorities tend to look at your really funny when really large sums leave and enter accounts and these are events I'd rather avoid because e.g. increasing risk of going up a KYC tier or getting thoroughly audited is just not wroth it), too little gain (fees eat at profit), and too much hassle (you have to keep buying/selling every day, very few orders are at market price)
This is not a *failure* of prediction markets as much as a combination of a few failures of scale, distribution and communication style that will be smoothed out:
1. Different ways markets turn books into probabilities
2. Fees
3. Limits on bets
4. Limits on betting legally from certain jurisdictions
5. Trust by people with money in the platform
6. Financial hassle
5 and 6 looks like "Yes you could make a 3-4% return a month by arbitraging x vs y but you need a multinational structure, some pretty on-top lawyers and accountant, and should still account for a large risk of one of these orgs going bankrupt or rogue. And at that point many other arbitrages are equally tempting and (even though this seems surprising), there isn't enough leverage in the world to arbitrage all markets.
I suspect that many people made loads of money by arbitraging and I expect this will be less of an issue as time goes on and more money floods the markets & track records get established & insurance offerings pop up.
Not sure if this is the thread to discuss predictions generally, but I just checked Allan Lichtman's particular 2024 prediction (https://hdsr.mitpress.mit.edu/pub/kcnosyqm/release/1), and it must be maddening to Lichtman that he could have predicted a Trump win if he had just utilized his own (flawed) model rigorously and objectively! Lichtman says that there were four flipped keys against the Dems, but it seems obvious to me that Policy key and Foreign/Military Success key should have been flipped as well: Biden didn't achieve a New Deal level policy change (insofar as I've understood this is *not* supposed to be an easy key to flip), and Ukraine is not a Foreign/Military Success, as the Ukraine/West coalition has not, uhh, actually won the war.
Yes. His 13 Keys model is still very good. He just allowed too much personal political bias in how he applied it this time. But then, that is an edge that a purely mathematical model would have. I think it's easier to put personal bias aside when you're just dealing with poll data and/or prediction market data, and not having to judge if a candidate is charismatic/hero-like or had a major policy achievement.
I think 3 other keys are at least in doubt: 1. Contest (sitting President did have to be forced off the ticket, even if then party unity resulted)
2. Scandal (same events, really: a sitting President becoming senile, this being covered up, and then resolved behind closed doors)
3. Challenger charisma (DJT fails part of the definition because he doesn’t appeal across the board, but he smashes the other part which is being a once-in-a-generation transformational figure)
ISTM in all these cases there’s some typical-minding going on: AL perhaps doesn’t see those things as I have described them, so he doesn’t notice other people do see them that way.
Good points. A sitting President having to be forced off the ticket strongly suggests *some* sort of scandal, like you said. In this case, the scandal was the covering up of Biden's declining mental faculties.
Thanks!
Theo misunderstood the poll; he thinks the NYT numbers are fake because they don't match the raw data, but that's because they used weighting.
https://x.com/michael_wiebe/status/1854963945209315640
Neighbor polling. Sounds interesting. Do you know of any good links... I find a lot of trivial stuff about the whale Theo.
Oh, and I think Scott missed the point of this election. Money markets will rule, because some people will go to all sorts of lengths if money is involved.
Linked elsewhere in the thread: https://statmodeling.stat.columbia.edu/2024/11/09/polling-by-asking-people-about-their-neighbors-when-does-this-work/.
I overlaid the main events on the probability chart and you can see that Trump’s odds never went above 52% after Harris got the nomination … that is until Musk started paying swing-state voters to register and Trump’s odds then rose from 49% to 64% (right before the MSG rally which knocked them back down to 54%, which was the opposite of the actual Latino result)
https://www.canva.com/design/DAGV0Izxii8/KQEKVstU8cmFMI1QreSugg/view?utm_content=DAGV0Izxii8&utm_campaign=designshare&utm_medium=link&utm_source=editor
What do you mean by “lost synchrony”? AFAIK all financial markets stayed within transaction cost bounds of each other.
Matt Levine's writing is generally delightful, he provides some commentary on prediction markets in the wake as well:
https://www.bloomberg.com/opinion/articles/2024-11-07/prediction-markets-are-a-thing-now
He cites this Whitaker and Mazlish article about obstacles to prediction market popularity:
https://worksinprogress.co/issue/why-prediction-markets-arent-popular/
Which argues that prediction markets (currently) lack some of the features and players common in other markets, stalling their adoption.
The Whitaker and Mazlish is a long article, difficult to summarize, but does invite one to wonder if one could catalyze adoption rates by playing with the right levers.
It inspires this very Levine-ish line:
"“Prediction markets are not a natural savings device” seems obviously true, and a good reason to think that prediction markets will never compete with stock markets for size and liquidity and efficiency. (However: “Design a product for retail investors who want to use prediction markets as a savings device” strikes me as a very good exercise for the reader, and I expect that within two years I will be writing about a buy-and-hold prediction-markets exchange-traded fund.)"
Caveat, Levine very handwavily endorses the default narrative on Théo. But his article also brings a lot of whimsy, which if not grating, ends up fungible with accuracy to some exchange rate.
He closes by noting:
"Oh meanwhile, amusingly, Polymarket might be illegal in France, where Théo is apparently located, oops..."
I recommend the whole article though. It's paywalled, but his column is free if you just sign up for his mailing list like it's 1999.
I get the feeling that Scott's punishing us with this dry spell for electing Trump.
I didn't vote and had no plans to vote ( https://youtu.be/vJriNA37xng ) but had I have known it would effect our intrepid host I wouldn't have been so cavalier about TheMostImportantElectionInHistory™.
More seriously, Scott, if you're actually hurting over this I hope your legendary resiliency asserts itself and you bounce back with vigor and joy.
"I have never heard of a hedge fund betting on a prediction market and my guess is that it would either not be legal or require too much compliance paperwork to be worth it. I hope this changes!". Actually, Susquehanna trades on Kalshi (https://www.bloomberg.com/news/articles/2024-04-03/susquehanna-starts-trading-desk-for-event-contracts-on-kalshi?embedded-checkout=true), Polymarket would be more complicated for as US-based firm I guess. It will be interesting to see if more firms will set up prediction markets trading desks.
Is there a possibility that the mispricing was primarily due to the rare occurrence of a US election, and can isn't it obvious then to keep expecting more mispricing and arbitrages due to the relative infrequency of these events?
hmm whales are good and provide market liquidity. id disagree with ur whole thesis.