What do Sam Altman, Matt Bruenig, and the Sacramento Kings have in common? -- Are the polls wrong? -- CFTC vs. Everybody
That NASA market seems to be poorly named, as I'm pretty sure that NASA will have landed a person on the moon before 2025. Say, about 55 years before.
The blip in Putin's chances of remaining president around Sep 12th corresponds to the major counteroffensive Ukraine pulled off in Kharkiv oblast around that time, when they liberated tonnes of territory.
Thanks again to Substack for offering to support Manifold embeds, and to Scott for embedding them into this post! It's super gratifying (and humbling) to see our markets becoming elevated into the discourse around a bunch of current events; as always, please let us know if you have thoughts on how our site and our markets could be improved~
Also, for those of you specifically interested in the midterm elections, my teammates Stephen and Fede have created this neat election map, visualizing the markets for contentious senate and governor races: https://manifold.markets/midterms
Bonus: we're hosting a $500 forecasting tournament on the 2022 midterms! Check it out here: https://manifold.markets/group/us-2022-midterms/about
The samotsvety graph seems wrong/confusing? i.e. the pink section is way bigger than 1%
The "Karlstack gives more details" link is a duplicate link to Eisenberg's statement, it seems like the correct link should be https://karlstack.substack.com/p/exclusive-the-man-who-may-have-milked
16% chance of the Iranian regime collapsing by EOY 2022, 20% by 2023, and 34% by 2024? That seems like either a market that shouldn't be trusted or a bunch of predictions that it's especially unlikely to happen in 2023 specifically, no?
Apparently there's multiple manifold markets about me now, and they also spell my name different ways. Is there some rule that if I'm mentioned more than once the spelling must be different each time?
On the polling question, I'd point out that polls overestimated Republicans in 2018:
As for how to square that with the events in 2016 and 2020, a few possibilities:
* Polling error is random
* Polls have a systemic pro-Dem bias these days *and* an element of random error.
* Polls have a systemic pro-Dem bias when trump is on the ballot
An related question is, why didn't pollsters fix the issues from 2016 to 2020?
I seem to remember people complaining, after 2016, that the issue was not weighting by education, such that the polling models had a higher portion of the electorate be college grads than what the norm was. And they would post examples of polls still making this mistake in the run-up to 2020. Perhaps they fixed it now, or perhaps they will stubbornly refuse to ever fix it (or perhaps they compensated but then simultaneously the problem got worse).
For those of you more into prediction markets: Why are they suing to challenge laws instead of seeking to comply? The notice I read specifically mentioned safe harbors they could take advantage of. That wasn't just them being polite. That was the regulatory agency explaining what they needed to do to be allowed to continue to operate.
Prediction Markets are not gambling or investments. They're clearly games of skill with rewards at the end which are definitionally not gambling. Alternatively, if they must continue to operate as investments, CFR 17.2.227 provides another safe harbor but one with somewhat more complex tax implications.
Also, if the rationalist community somehow takes down the Administrative Procedures Act it will instantly become the hero of libertarians everywhere. Which would basically be necessary to argue that this finding was invalid. Chevron deference and all that.
Isn’t it deeply in Putin’s interest to have the West/NATO think the likelihood of a nuclear retaliation to any intervention is extremely high? Therefore, I imagine (if he ever thought other decision makers were looking at the prediction markets) he has various hackers/spammers manipulate those markets? Compared to the resources of e.g., Cozy Bear / Fancy Bear it doesn’t take much to manipulate most of these markets.
what about Womantic monday
I sent Kalshi a petty email too, not sorry for it. If they really sabotaged PredictIt they should feel bad!
My comment for the CFTC was just "Bring back PredictIt"
On the midterm election, I would point out that our most reliable source of election data, Ann Selzer's Iowa polls, point to a larger Democratic victory than either prediction markets or election forecasters, so count me strongly in the camp of "prediction markets are getting this one wrong".
Ann Selzer's final polls in 2016 and 2020 were both massive outliers, both predicted a large Trump victory, both had articles written about how you shouldn't believe outlier polls, and both were right on the money. Looking all the way back to 2008, she has a better election prediction track record than Nate Silver does: https://secondhandcartography.com/2022/10/15/ann-selzer/.
So when her most recent poll shows the incumbent Republican with only a 3 point lead, I think should make you significantly update towards the Democrats keeping control of the Senate.
> One of my hopes for forecasting is that it eventually becomes so well-validated that decision-makers can take these kinds of considerations into account: “Should we sent ATACMS missiles to Ukraine? It would have such-and-such benefits, but also increase the risk of nuclear escalation by 3.6%, is it worth it?”
The link to Dynomight at the end of the post is titled "Prediction market does not imply causation". Seems like Scott's interpretation here does rely on causation, and now I'm wondering if prediction markets are useful at all in informing decision making where the decisionmaker intends to affect the market outcome.
The quoted language from the PredictIt complaint is boilerplate. The reason it asks the court to say that the CFTC's actions were "arbitrary, capricious, an abuse of discretion, [and/or] otherwise not in accordance with law" and “without observance of procedure required by law" is that those are literally the phrases in the Administrative Procedure Act that describe when a court should vacate an agency's actions (https://www.law.cornell.edu/uscode/text/5/706). There are other grounds for doing so -- if the agency isn't authorized by Congress to do what it's doing, if the agency is acting unconstitutionally, etc. But these are two of the most important ones: that the agency's decision was irrational or that it didn't follow the procedures it was was required to. So I wouldn't read too much into that part of the complaint; that section of any complaint challenging any agency action will read similarly.
"Republicans distrust pollsters," it says.
I'm a liberal Democrat who has sworn off answering telephone polls of any kind. I used to do them and enjoy the process, but the pollsters have abused my hospitality. They ask long and interminable series of picayune questions a hair's-slice distance in topic from the surrounding questions, on topics on which I don't have a strong opinion and hardly know how to answer, especially when the same question is turned to show a different crystal facet. And they aren't allowed to skip or abbreviate a single thing, even if it's the same list given the Nth time. Or the questions are meaningless or too multivalent to answer, like "Do you approve of California's direction?" What does that even mean, anyway?
One time a pollster was asking me an endless series of picayune hair-slicing questions and my brain was getting tired out by the mere process, and I finally burst out, "How much longer is this going to go on?" The pollster assured me were we almost done. Turned out we were indeed almost done - with that section. There was at least one more whole section as long as the first to follow.
At that point I said, "No more," and hung up. And swore never to answer another telephone pollster, even if they beg me, which they have.
This is awesome, Scott!
One link is broken — "The Manifold Markets Team" should link to: https://comments.cftc.gov/PublicComments/ViewComment.aspx?id=70745
At risk of sounding like an idiot, can anyone explain to me how prediction markets are functionally different from sports betting in any meaningful sense? I’ve been wayyyy out of the loop on this, but from what I gather the %chance moves based on where people place their money just like the point spread in a sporting event would. Is this correct or am I missing something? At least on the surface where I stand as a newbie, It’s always explained in a language that makes it seem like a more complex system than what it actually is.
So prediction markets don't work for something like a full scale nuclear war. Theres no reason to bet on that because even if you win, you can't collect anyway.
> One of my hopes for forecasting is that it eventually becomes so well-validated that decision-makers can take these kinds of considerations into account: 'Should we sent ATACMS missiles to Ukraine? It would have such-and-such benefits, but also increase the risk of nuclear escalation by 3.6%, is it worth it?'
I'm having doubts about the IR theory baked into this.
Imagine that NATO decision-makers were using forecasts to calculate expected utility like that, and Russian spies found out. Couldn't Russia then announce that its nuclear missiles will launch automatically, if its satellites detect any shipments whatsoever crossing from NATO countries into Ukraine?
Prediction markets would react dramatically. With those new forecasts, wouldn't NATO decision-makers then cut off supplies to Ukraine?
Russia could then extend the same automatic launch procedure threat to another neighbor, then another, and so on.
Nuclear blackmail seems like an issue where maximizing expected utility is not achieved through calculating expected utility but through 1) acting on a qualitative understanding of how threats and credibility work, and 2) making a sharp distinction between defending the status quo and going on offense.
What is the overlap in prediction markets for people who made bank on the Trump win of 2016? I feel like Prediction markets, especially the small volume ones that are around now, might be a little more right wing biased.
How accurate were the prediction markets in 2018/2020 vs. the result?
One of my hopes for forecasting is that it eventually becomes so well-validated that decision-makers can take these kinds of considerations into account: “Should we sent ATACMS missiles to Ukraine? It would have such-and-such benefits, but also increase the risk of nuclear escalation by 3.6%, is it worth it?”
Will that ever be possible? inasmuch as whether you get a nuclear war or not, it's a binary outcome and we're not (from a practical standpoint) in a multiverse with a distribution of results.
I kind of feel the same way when poll results are translated into "Trump has 17% chances of winning". "Oh that's almost a guaranteed loss" says the amateur. The role-player I am knows it's actually not that low - it'll happen. But how is it really helping when Trump actually win? Did he really have only 17% chance and got lucky? Or were the odds far higher?
We don't run elections thousands of time anymore than we run nuclear wars thousands of time. We either get one or we don't.
Here, the whole thing rests on 3 elements and 3 elements only. Will Putin use a nuke to try and save his regime? Given that he escalated to mobilisation rather than accepting defeat/declaring victory and go home, it seems quite possible.
If he does, will his generals follow his orders? Given that, apparently, at least a significant % of his inner circle is more hawkish than he is, that's not impossible at all.
If they do and detonate a nuke in Ukraine, we've been quite clear that NATO would intervene and, at minima, wipe out all Russian forces in Ukraine (though some uncertainty remains around that).
Would Putin (and/or his circle) accept defeat then rather than escalate to nuclear Armageddon? IDK.
> Suppose Russia is losing very badly in Ukraine. Putin, fearing a coup or revolution at home if he gives up, decides to use a tactical nuclear weapon, ie a “small” nuke more suited to winning battles than destroying cities. He nukes a Ukrainian battlefield position.
Haha, no, typical Western nonsense. You guys have no idea how Putin thinks. Why would he nuke Ukrainian soldiers in a war he is going to lose anyway? That does not make sense at all.
*If* he is going to drop a nuke on Ukraine, he will drop it on *civilians* instead. Conditional on the nuke being fired, 80% probability it will be fired on a city with population 50000 or more. That would bring him much greater popularity at home than destroying some ultimately irrelevant battlefield.
Please group all content regarding prediction markets (preferably at the bottom of posts) and clearly label them for an easy skip.
> I am not a lawyer, but it sounds kind of like they’re saying “the decision was bad, and the Administrative Procedure Act says regulators shouldn’t do bad things”. I am split between the part of me which hates government regulators doing bad things, and the part of me which feels like this is how you get a cover-your-ass-ocracy that never does anything at all without fifteen layers of paperwork and ten trillion dollars per action.
Weird. I don't agree that the argument here is "the decision was bad, and you're only supposed to do good things". There are plenty of laws that specify that only good things are allowed, and bad things will be illegal because they are bad. But I don't think this is in that category.
But I agree with the conclusion; this is exactly how you get a bureaucracy that isn't capable of doing anything except generating protective layers of non-action. I think the argument being made is "there's a process for changing the rules, and you didn't follow it, so your decision never actually took effect". And the validity our system bestows on that complaint ensures that the maximum speed at which things can happen is essentially a stop.
The APA isn't my primary area of expertise or practice (and so if anyone who practices with it more regularly wants to weight in please do)., but it's a little glib to summarize the allegations made under the APA as "the decision was bad, and the Administrative Procedure Act says regulators shouldn’t do bad things” -- while that's invariably the *driver* of these types of Complaints (and they are a source of significant bureaucratic friction), the specific complaint here is very much about procedure and touches on a pretty core part of the APA: specifically, that agencies generally can't just decide to do stuff without publicizing their rationale and, typically, having a period of public notice and public comment prior to issuance.
Agencies changing course on a whim are said to be acting in a manner that is "arbitrary and capricious" -- the basic idea is that agencies have a great deal of legislative power but they can't exercise it without jumping through enough hoops to indicate that there's a genuine rationale behind their decision and, moreover, one that amounts to at least a little more than "the president wants more / less regulation."
In practice, this can can easily amount to just requiring a lot of generation of paperwork and public comments that can be ignored, and the agencies can get away with a great deal so long as they do enough ass-covering, but in theory it's designed to limit *exactly* what PredictIt is alleging here -- random reversals in agency policy seemingly done at the whim of executive branch officials to whom a great deal of legislative power has been delegated.
Particularly in instances in which an agency pulls a complete 180 without any compelling factual developments indicating why the previous policy was inappropriate, courts are going to be generally be more skeptical of agency behavior and treat this as an instance in which throwing sand in the gears is salutary, and a court may end up reversing the agency's decision. In practice this can end up just being "come back with better documentation," but delay of implementation (e.g. until the next administration replaces all the top personnel with people who might have different ideas) can often be a win for the challenger in a practical sense.
In theory, the shortcut for agencies trying to get things past APA review instead of being stuck in APA-Complaint hell for four-to-eight years is to basically have an ironclad set of well-documented rationale for its decision, abide by notice and comment procedures, and then show these to the court who tells the plaintiffs to pound sand. Sadly I can't speak to how well that actually works in practice, but it seems that we do, in fact, get new regulations from time to time. The Executive seems to be basically the only branch of government that actually *does* anything these days.
This is not legal advice, I am not your lawyer, etc. etc.
Great analysis, as ways!
I would like to add another prediction market source about the Senate elections, on Futuur only 41% believes the Republicans will flip the US Senate elections
In short link link number four , one solution offered for some scenarios is to resolve a small number of prediction markets selected randomly. The downside is that a market which will get resolved 5% of the time will probably not offer a good ROI even if you can predict the outcome much better than your competitors.
One way to solve this would be to allow people to place their money on multiple markets at once and insure them against the case where more markets resolve than they have money for. If the probability of the market resolving is fixed in advance (like "we will fund 5% of the startup ideas randomly just to resolve the markets"), it would be trivial to calculate odds.
So an investor with 1$ could either invest 20$ in one market which resolves with p=5% (and lose everything 95% of the time) or they could place 1$ in 20$ different such markets (and pay some insurance to cover the case where more than one market resolves and they would be on the hook for more than 1$).
Good luck convincing anyone that this is not gambling, though.
Edit: also, could conditional markets for event B (nuke) given some event A (no fly zone) be modeled by just having markets predicting A, B, A AND B?
The Dynomight link in number four is by far the most interesting and important point here. The short summary is that conditional prediction markets just give correlation and if you want to use them to make decisions, you want causality. Seems obvious in retrospect. They also discuss how to work around this maybe.
The NASA market (that they'll land someone on the moon before the end of 2024) is utterly bonkers - there's not a snowballs hope in hell of NASA getting someone to the moon a year BEFORE they plan to.
Could metaculus not solicit donations (or some other way to acquire money) and then divide up the donations and award them to the best forecasters on a regular basis? Or is this loophole still blocked by law somehow?
The Kings Owner's "some reason" appears to be:
> I am also the founder and former CEO of the business intelligence software company TIBCO Software, which was one of the first companies to facilitate instant communication in financial markets. In 2016, I founded an early stage investment firm called Bow Capital. I have a long time interest in the domain of business prediction. In 2006, I wrote a book called The Power to Predict about the importance of anticipating the future for business success. In 2011, I followed up that book with The Two-Second Advantage: How We Succeed by Anticipating the Future–Just Enough. This belief that predicting the future is absolutely crucial to commercial success has convinced me to support Kalshi’s submission to the CFTC to list contracts on the outcome of Congressional elections.
I still don't understand how you measure prediction accuracy of percentage predictions for unique events. An event will either happen or not happen. You can say whether the event will or will not happen and then determine whether you were right or wrong. But if you say there is a 75% chance the event will happen and it does, do you say you were 50% right? Is this considered a valid measurement of success?
I suspect the rise in "NO" in September 12 is somehow connected to Russian regional elections (they lasted from September 9 to September 11). But it is really strange, I'd update in the opposite direction: the results of these elections really favoured United Russia (the party that has the majority in Duma, which supports Putin, and is the most popular). It is not fully clear to me what caused these results: changes in minds (according to polls, the support of UR has grown with the beginning of the war), arrests of oppositional municipal deputies (and denial of their registration) or falsifications (distant electronic elections might be easy to falsificate). Maybe people, who bought "NO" expected post-electoral protests?
(Please keep in mind that English isn't my first language, I'm sorry for the mistakes I make.)
I have observed some ... interesting behavior with 538, lately, in which the list of polls used as supporting evidence appears to be getting modified retroactively, and a tendency for the adjustments made to the polls to be -away- from the mean of the polls (which if you're being rigorous, you don't change your methodology midway through, so great - but also, maybe this is a sign that your methodology wasn't great to begin with).
My personal guess at the moment is that Republicans are going to do far better than the polls may currently suggest, and the polls will show an apparently-mysterious sudden Republican shift over the next couple of weeks. (Basically, I think pollsters are telling people what they want to hear, but this can only continue until we get near the election, when they will correct - possibly even overcorrect - so they appear accurate after the fact.)
Out of curiosity, I did a super simple regression to see how far off betting markets think 538's forecast is.
It looks like in the key Senate races, they expect Republican candidates to outperform the 538 forecast by about 1.8 percentage points on average.
"Tactical nukes wouldn’t really be useful in Ukraine; the battle lines are too spread out and there’s no single place where a nuclear explosion could take out a substantial portion of Ukraine’s forces. "
This is misleading. It's been about thirty years since tactical nuclear weapons were expected to destroy enemy forces on or near the battle lines. Tactical nuclear weapons are expected to be used against logistics hubs, command centers, airbases, and other such targets behind the battle lines. I would expect that Ukraine has been trying to spread those out, but many of them are tied to things like roads, railroads, and runways that are built in peacetime to exploit economies of scale and concentration.
We've seen Russia's offensives grind to a halt because the Ukrainians kept blowing up their headquarters and supply depots; Russia hasn't been able to do the same because they're critically low on precision-guided missiles to do it with. But if they can achieve the same results with missiles that don't need to be precise to be devastatingly effective, they can plausibly stop Ukraine's counteroffensives and then some.
If you want to argue that this sort of thing is "strategic" rather than "tactical", you can, but you'll be using different definitions than most of the people you are talking to and that leads to unnecessary confusion.
Fwiw, reading this on October 21, some of the prediction markets no longer match the text, which presumably will become more and more true. Is it too annoying to put a screenshot of the market at the time as well or write the probability in each case (you did in several). Or maybe it doesn't matter because this isn't meant to be read days or weeks later?
In every case I've noticed, back to 1980, US Presidential opinion polls overestimate the Democrat's strength significantly. 2020: 8.4% (https://www.ajc.com/opinion/opinion-a-way-to-improve-political-polling/HUIH2PG5OFDR3BLIPENT4Z6UFA/). Looking at Gallup (https://en.wikipedia.org/wiki/Polling_for_United_States_presidential_elections), I see more variety -- the overestimate can be down to 0% or even a slight underestimate. But what I always see is that the D is a shoo-in -- even in 1984! And Gore was supposed to beat Bush easily. We all knew Clinton would defeat Trump (https://en.wikipedia.org/wiki/Nationwide_opinion_polling_for_the_2016_United_States_presidential_election#Two-way_race). My guess after seeing Gallup's results is that although there is a slight bias toward D doing better than he really will, this is exaggerated by which polls I actually saw reported by news media.
Since this goes so far back, I can't credit that it's because R's distrust pollsters. (In 1980, what had they to fear?) Or poll fatigue, which I think really kicked in in 2012. Or undersampling cell phone users. It's just guessing, but I suspect discrepancies in poll results is from the already known phenomenon that R voters are more likely to actually get to the voting booth on Election Day. So D's would show up stronger in polling. Then of course we can expect news sites to be more likely to highlight polls if it makes their side look strong. If you were really doing your best to report accurately, and get proportions of D's and R's and I's right in your polling, you still couldn't very well adjust for D's being less committed to voting except by weighting D's answers less, which looks pretty unfair. So I'll tentatively attribute the difference to, well, not malice.