Real-money prediction markets indeed did relatively poorly in 2022.
The further past shows a more complex story -- bettors were running slightly ahead of 538 from 2016 to 2020. With 2022 added, 538 gained a slightly edge. That chart and more are here, for anyone curious: https://www.maximumtruth.org/p/deep-dive-on-predicting-elections
I'm not familiar with the particulars here but surely this is misleading:
> Nate Silver is only one person, he has only one area of expertise, and you can’t hire him to predict random things for you (unless you’re rich and he’s bored). If Manifold can apply only-slightly-sub-Nate-Silver levels of analysis at scale to arbitrary topics, that’s a big deal.
Manifold is pretty good at elections *because* bettors defer to models like Nate's. Doing well on elections is little evidence for doing well on arbitrary topics. [Edit: the previous sentence refers to Manifold, not Nate.]
I think your last plot might be related to your fourth.
That is, now that Trump is looking reasonably likely to win the next US election, both Russia and Ukraine are unlikely to make a peace deal until that uncertainty (both whether he wins, and what his actual Ukraine policy will be) is resolved.
If the prediction markets are overvaluing a Trump win, perhaps that represents a kind of hedging where people are willing to take a loss if Biden wins because they think their lives will be good but want to gain some cash if Trump wins to make up for the worsening of their lives.
> Betting sites favored Biden to win all swing states, but Trump to win overall.
That makes it sound like there was one bet on whether Biden would win all swing states, but I think what really happened was a bunch of bets, one per state. If so, then FWIW, it wouldn't be inconsistent to think that Biden had better odds in each individual swing state but worse odds overall (insofar as he would have needed to win more swing states than Trump).
I just did an arbitrage strategy earlier today of buying Biden for 39c on Polymarket and Trump for 47c on PredictIt. This is almost guaranteed to turn 86c into $1.
The only cons are:
* Maybe one of them dies/goes to jail/drops out (but that might boost the odds of the surviving candidate)
* You need to fund an account on PredictIt and Polymarket
* Annoyance of reporting any gains on your taxes (is this gambling winnings or capital gains?) or a small risk of getting punished for not reporting
* Opportunity cost of not buying the S&P500 instead
Could you clarify how you intend to resolve this? How much of a factor would the portion of voters considering it a top issue be compared to your subjective impression of the political conversation?
I do think prediction markets show enormous promise, but they've got some growing pains to work through. Inefficiencies in those markets don't mean they're not valuable. And even highly liquid, sophisticated markets can have these kinds of irrational episodes where dumb money overwhelms smart money - the entire meme stock phenomenon is evidence of that.
Using decimal odds Trump is 1.83 on Bet365 and Biden is 2.75. That is an implied probability of 54.6% for Trump and 36.36% for Biden.
Stated differently, a $100 dollar bet on Biden would pay back $275 for a $175 profit.
If it is indeed a coin flip this is a massive value. My instincts are that the combined odds of him either losing to trump OR dropping out make that priced properly.
Of course a BET365 user I am likely in the same bubble that drove these odds to 2.75 in the first place.
I’m personally not sure how to feel about Neuralink - on one hand it (or any other complex BCI) could usher in an era as profoundly different to our world as the pre-smartphone era was to today. Yet I feel like there would likely be extremely nefarious or just plain bad uses that might completely wipe out that good (imagine advertising spam in one’s brain!). I’m not really sure which is stronger, or whether the development of BCIs will be similar to most modern technology, somewhere in the middle.
I'd say prediction markets did badly at the 2024 Presidential primaries, too. I guess it's possible that one day we could get a tell-all book revealing that Biden was planning not to run again and then changed his mind at some point, but that's probably not what happened, and even if it was true that wouldn't explain the low prices when he did officially announce. The low prices for Trump to win the Republican nomination are also probably because the market was wrong, not just because Trump got lucky. So not just a 2022 event.
I haven't checked if the markets were worse than Metaculus, to be fair.
I've been enjoying Manifold lately, but I do think it has a limitation in that it's hard to entice people to bet on markets they're not personally interested in. Most of the time and energy goes into the same handful of topics.
The ideal (it seems to me) would be a system that encourages people to dig in and research difficult topics to try and win mana. But it seems to me like the optimizing strategies right now don't really end up doing that.
>Maybe this is just because he has a higher chance of becoming President before the trials are over?
There are a variety of other options. Trump's old and could just die before he gets around to serving time. He could theoretically skip out on his bail, though it's not very likely. The ballot-removal shenanigans could escalate into civil war, which even if he loses might not wind up with him serving time (if he's assassinated or otherwise killed during it, or if he's executed for treason). Nuclear war could happen over Taiwan, which even if Trump doesn't die would see the Blue Tribe crippled.
Regarding "dumb money" in prediction markets - shouldn't it usually be the case that dumb predictions outnumber good ones? Prediction markets are a zero-sum game, and running the market incurs costs which need to be covered, so the median prediction should have slightly negative expected return. Stock markets attract smart money, but they have a positive expected value, since stocks are productive - even if your bets don't outperform the average, you are still likely to make money due to dividends and increased share values.
Nate Silver is more than sufficiently skilled to use manifold and metaculus to help him forecast e.g. tracking people with records of forecasting. 97% chance that Silver or his advisors can do more than enough spreadsheetmancy for that.
On the limits of real-money prediction markets: If you assume no cost for the effort of the research, their miscalibratipn still has to be large enough that betting on them would outperform the s&p in terms of risk-adjusted expected returns for smart money to want to bet on them.
For example if right now I can buy Biden at 40% but think he has a 45% chance to win, that's locking up money for almost a year for 10% returns - but I could get roughly the same expected returns by just buying SPY, and at a lower risk (since betting on election markets is all-or-nothing, which is very high risk). So you can expect them to be fixed by smart money arbitrage to within 10% of accurate, but not 5%.
RE: US election - you'd expect the markets to be misaligned because you have to tie money up for 3/4 of a year to correct them.
RE: liquidity & dumb money - there's a step below professional traders, which is professional gamblers. I work in this kind of area, and several of my friends have built statistical models for Eurovision Song Contest markets to answer questions like "how many points will Lithuania receive?"
As far as I'm aware, the big trading firms don't even bother betting on e.g. bookie odds for Premier League games, which have tens of millions of liquidity, but this is easily enough to sustain professional gamblers (and probably they're in the smaller/more niche markets even).
This affects the arbing problem of election odds too - as people professionalise, they'll create accounts on every real money prediction market and seek arb opportunities. Before that point - is it really worthwhile to spend your time looking for $10 here and there? The motivations for people who enjoy forecasting vs. people who enjoy finding edges in gambling markets aren't the same.
So the liquidity problem isn't as bad as it seems - we don't need prediction markets to attract Wall Street before these problems are solved. But there is likely at least one step change required before then, which is about attracting normal customers. Reach the point where arbing can get you an immediate $100, particularly if you can automate it, and you are set.
The shorter term solution would be if the big prediction markets had liquid markets like "What percentage win chance will this market have Biden at by the end of January", we would see quick agreement via arbing.
(I have Plans and Schemes to work on this, as it's similar to poker site liquidity, one of the few areas where I actually am an expert)
> Why? All the things voters might blame Biden for - inflation, Ukraine, Gaza - happened either well before or well after the early-2023 period when his numbers began to decline.
I wonder if some of this is actually the legal shenanigans. Trust me I don’t want Trump to win, not because of silly white supremacy, or Russian agent stuff - that’s a tale for idiots - it’s because of his volatility. We really don’t need that right now.
Anyway I’m still a bit put off by some of the legal stuff, is the award of $83m in anyway proportional to what happened here?
Seems to me that this could be used against anybody in the future - the republicans will surely use it. Clinton could hardly have survived the scrutiny.
(I’m not talking about the criminal or referral cases which probably have merit).
Regarding the problems with prediction markets I think what a lot of people (including the Asterisk article writer) are missing are the nitty gritty of how markets work:
- The headline probabilities the markets show are usually based on the last trade. So if after election day it is still showing Trump at 5% it doesn't necessarily mean people are willing to bet at those odds, more likely it is just that no-one has traded since outcome became clear (why would they?)
- A lot of markets are illiquid and have huge bid-ask spreads. For instance, in the PredictIt Trump-Biden market shown in the post, you could say it favors Trump but the bid-ask spread shows 46%-55% for him to win so I think it's more accurate to say it shows a tie.
- Markets have pretty high fees (PredictIt is 5% widthdrawal) so even if the bid-ask spread was 0 (which it is not) it might not make sense to bet against a 4% outcome even if you are 100% sure it won't happen.
- Cost of capital: if you think there's an opportunity to make 3% after all costs on a market that resolves in 1 year, a rational trader still wouldn't do it. You can just put the money in treasuries and get 4% in same period with zero risk.
- Sometimes there are technicalities about the resolution conditions of markets that make a big difference. I remember there was a market on 2020 election result on the now defunct platform Augur, that seemed to have opportunities for profitable bets on Biden after election day (I made a bet myself). But it ended up taking a long time to resolve (only in January I think) and for a while it actually looked that it could resolve as Invalid. This was because of the particular phrasing of the resolution condition it was unclear if it was unclear if certain technical conditions about electoral collage electors certifying the result were met by the resolution date. A naive observer would just look at the odds the market was still showing in December and decide prediction markets are dumb.
If you take all of these into account, it should be clear that prediction markets show a lot less irrationality than the Asterisk article claims. Of course, it does mean the markets are less useful than you'd naively think for the purpose of actually making predictions. To be fair, a lot of these issues will become less severe if the markets become more liquid.
Would be very funny if it turns out, that non-real-money prediction markets being about as good or even better than real money prediction markets, is not just a temporary bug and it's possible to have non-real-money financial markets that systemically beat the stock market.
The big question with Trump's election is likely going to be one of turn out. Trump turns out the left like nothing else but it's just hard to be all that motivated by a fear of Biden. What balances the odds are the chances that Biden has a big (literal, as in breaks a hip) fall or makes some dumb blunder before the election.
With Trump short of death or coma it's pretty much all priced in.
I suspect this turn out effect is relatively hard to handle in election polling as it's a likely voter effect that's canidate/election dependent.
Does this mean we'll never get the kind of large liquid prediction markets we need because the SEC (and similar) will never approve without stronger evidence of benefit which requires approval to reach sufficient capitalization?
>Johnson makes a pretty reasonable guess about the cause: lots of dumb money. People use their pocketbooks to root for their favorite candidate. Normally in a functioning market smart money would take the other side and set the final price, but the high transaction costs, long waits, and regulatory limits on prediction markets mean it’s not generally worth smart money’s time to correct the mispricing; Goldman Sachs isn’t going to hire statistics PhDs to make a model just so they can bet $850 on PredictIt.
I wonder if it is only about this. Some time ago I was reading Stieglitz from the early 1990s, and while I don't have the exact citation at hand, I remember he argued that there a priori can't be a complete set of markets about everything. The most naive form of the argument, number of possible states and contingencies there could be a market about is larger than number of people and time and attention available to them to gather information and trade on it. And this was the ol' good regular economic markets he was mostly concerned with, not prediction markets which can about arbitrary questions.
Fuzzily extrapolating to prediction markets, it looks prediction markets must always be wrong to some extent. For there to be an informational advantage for a knowledgeable trader to gain from, there must a mispricing. If starting a market is very cheap and gathering good information is effortful, there eventually is could be more markets than traders with information to trade. To get your market going, you'd need to compete only to get people to notice your market in the first place and then entice them to trade on it. In the best scenario, perhaps one could come up with some theoretical caps how badly mispriced prediction markets will be, statistically, under different assumptions.
Unrelatedly, maybe there is a reason why fake-money weird internet status points work so well compared to real money: people seldom price their hobby activities the same way they price paid-for work. In real-money markets, smart money eventually should start pricing in inflation and interest rates and opportunity costs compared to investing into other possible, more profitable markets. If the market rewards you with weird internet hobby points? People will take it as a hobby and essentially donate their time and effort.
> I think Paperclipalypse requires human extinction before 2050. It’s at 11%. But Metaculus’ direct “human extinction by 2100” market is only at 1.5%. Either I’m missing something, or something’s wrong. My guess: different populations of forecasters looking at each question.
By the resolution criteria for the first question, "If no resolution has occurred prior to 2050, the panel will meet and will vote on the option that is closest to the real world."
For example, if "only" 50% of humans will die, the answer will be Paperclipalypse, because that will be closer to the real world than other choices. But the "human extinction by 2100" question requires that all 100% of humans will die.
"Rather, the interaction between AI and Homo sapiens ends about the same way that the interaction between Homo sapiens and Neanderthals ended."
I'm going to take the opportunity to rant about this, not because I think this is making this particular error in all its glory, but because expressions of it are fairly common in AI discussions and as soon as we-as-a-collective *stop doing this* we can finally start to think about what it really means for intelligent species to coexist.
Homo sapiens are not a universally superior group that violently wiped out the Neandertals* when they met. Trying to directly compare H. sapiens and Neandertal intelligence is something of a fool's errand, but there's little good case that Neandertals were anywhere near as far behind anatomically modern humans in that respect as the image of them in the cultural consciousness is. We were smarter, yes -- but human intelligence has always had wide error bars. The two coexisted in Europe for thousands of years -- minimum estimates about 1.5k years, probably more double that. Modern humans outcompeted Neandertals over this period through some mix of competition for resources, interbreeding/assimilation (not quite the same, but lumped here), and violence. Neandertals were resource-hungry -- they had very meat-heavy diets -- and even all else equal would come out worse under a pure resource competition than an equivalent species with a lower demand for land-based game. (H. sapiens maybe ate more fish.) A model of interaction between two sophonts** where one rapidly and violently massacres the other because of their superior intelligence can't be one based on Neandertals.
"Sophont", then, takes you down some avenues. I think the case that cachalots ('sperm whales') are sapient in the same sense as humans, and possibly smarter than humans were at a comparable stage of evolution, is extremely strong. (See https://web.archive.org/web/20160322095648/http://cachelot.com/ for an introduction to the topic.) Humans have certainly exploited cachalots, to the point of what is from this perspective a particularly heinous genocide. When I talk about cachalot intelligence with people, this is often a tricky point for them to think about: how could humans have committed genocide against a species at least their equals? Wouldn't the whales just outsmart us? Well, they *did* fight whaling ships -- at least some whales (e.g. Mocha Dick) were able to recognize their patterns, seek them out, and attack them if they attacked -- but, no, human history shows little clear correlation between intelligence and the escaping of atrocity. Whalers exploited things like the desire of cachalot mothers to defend their children in order to wipe out whole pods; plenty of humans who wanted to cause one another harm have exploited similar desires.
We know that two sapient species coexisted for thousands of years before one went extinct under complex circumstances. There is good reason to think two coexist now, and some to suspect the one that caused more harm to the other is not the unambiguous intellectual superior. This is to say that bringing in another intelligence has very complex consequences, especially if you think that intelligence will be 'clearly superhuman', but you can't pattern a superhuman intelligence to the perceptions people have of previous sophont interactions and get anything clearly representative of reality. In particular, you can't extrapolate from the actions of H. sapiens to the actions of any given "superior species". This is one of those interesting circumstances where people try to correct from thinking other intelligent beings would be "like humans", and *still* correct to them being too humanlike. The real outcome of throwing a third intelligence into the mix is trickier, less dependent on whether that intelligence is clearly "smarter than" humans or not (because this probably doesn't play the primary role in human exploitation), and not extrapolatable from half-remembered impressions of Neandertals.
*I prefer the spelling without the h, because it expresses English pronunciation more clearly and is what the valley is called now.
**"Intelligent AI will be sapient" is another of those "your AI is still too humanlike" things. We have intelligent AI right now,*** which is not...sentient, at least. You could, if you want to get really fun, try to argue that "sapient but not sentient" is a coherent idea. At this point, it might not be impossible. But you'd need a very rigorous definition of sapience. I'm saying "sophonts" here just to address the fact that all organic Terran species past X intelligence level are sophonts.
***AGI, the letters defined like that, is here. People wrote their definitions in the 2010s of what they expected AGI to be. It's here. We have it. If you mean something more intelligent, use a different term -- it's not our fault everything turned out to be less correlated than even the "no, seriously, AGI will not be like humans" guys thought.
> What am I missing here? It’s got to be higher than 8%, right?
No. Putin is not interested in peace or ceasefire. He is fully ready to fight for as long as it takes. Especially now when he hopes for a Trump victory.
The popular uprising in Russia in 2024 is almost impossible (<1%). Russians are not that tired of the war.
Neither are Ukrainians. They realize that any peace agreement with territorial concessions only means that Russia will subsequently attack them again, and are also determined to fight for as long as it takes.
This war belongs to the same category as the Soviet–Afghan War or Vietnam war.
> What am I missing here? It’s got to be higher than 8%, right?
Both sides still hope for some kind of decisive victory that will ultimately resolve their dispute (i.e. other side's forces are in full rout, a change of power happens, and some kind of enforcible restrictions are placed on enemy's ability to restart the war later - for an Ukrainian victory, this would have to be (almost) instant NATO membership, for a Russian one, a bunch of Russian military bases on the remaining Ukrainian territory, if any).
I really don't understand what Ukraine hopes for (a direct NATO involvement?), but Russia has some hopes for drying out of Western funding for Ukraine, and if it doesn't happen, a win in battle of attrition. But even if US fails to provide any money or arms in 2024 at all (which is, I think, unlikely), Ukrainian forces may well survive into 2025 with reserves, EU help, and sheer size of the country. And who knows what will happen after US elections - Republicans might find renewed love for Ukraine once it's Trump's war to win or lose, and not Biden's (in case of Trump win). So "peace by 2025", or even a temporary ceasefire seems very unlikely - 8% is about right. For now, both sides still have enough people, vehicles, ammo and will to fight.
FWIW, assuming they are they are their respective party's nominees on election day, Swift Centre forecasters put 52% on Biden winning, 48% on Trump winning, and also forecasted on some interesting conditional questions: https://www.swiftcentre.org/biden-trump-rematch-chances/
I think it's really important to include a well-established real money market that comes from the betting world - they have very different audiences to those from the prediction world, and that affects the odds they generate.
Here's betfair's "exchange" market on the US Presidential election:
It has Trump at 2.24 (ie bet £1 get £2.24 if he wins) and Biden at 2.94 - that is to say that you can make a profit betting that one of them will win without specifying which one. 26% ROI on a bet that will run a little under a year.
This is being driven by the fact that the primaries aren't quite over - Haley has about a 4% chance of being president per the market, weirdly they have Michelle Obama at 7% (huh?), and various others at 1-2% (Newsom, RFK Jr, Philips, Harris).
I don't think the "dumb money" problem is solved by scale or relaxed betting limits, because it doesn't address the problem that prediction markets are inherently zero sum and the smart money needs to get paid. Prediction is hard work; if we haven't already maxed out the unpaid-hobbyist-predictor market we eventually will and probably through the unpaid markets like Metaculus. To do better, we're going to need a way for the smart money to place a positive expected value on investing in prediction markets.
In the case of e.g. the stock market, that's easy because the stock market is positive-sum; you're trading claims on the future income of profitable enterprises. Money comes in to the market from traders making their bets, but also from companies paying their dividends. If you smartly place the right bets, you wind up owning a nice dividend stream (which you can sell for immediate $$$ if you need that). It's certainly possible for smart money to prosper by betting on dumb money, but smart money can also prosper by accurately predicting which companies will be most profitable, and we get useful signal out of stock markets.
In a prediction market, money only comes in from traders placing bets, so smart money can only prosper by way of dumb(*) money -> smart money transactions. The smart move is always to figure out what dumb money is going to do, and do it first. So everybody in the market, smart and dumb alike, is going to be making the dumb-money moves. That doesn't give us a useful signal.
You could get around this by having someone altrustically fund markets with the expectation of losing money but getting good data. Or maybe the subsidy could come from people who would profit (outside the market) from good data, but that comes with an obvious free rider problem and it would probably be more effective for them to just privately hire a team of forecasters. There are other possibilities, but I think they're all pretty low probability.
* Dumb in market-profitability terms. If you have some other reason for investing, even just "I'm rich and this is fun", making unfavorable bets may not be absolutely dumb,
Re changes to manifold love, I think the best thing would be to bet on abstracted questions eg "odds heterosexual make and female both 30 years old reporting a good date" or "odds that two people who both say they like horror movies like each other", and aggregate those into giving the results for each person. (in a similar way to how OkCupid used to use the questions for compatibility score, but with the weightings set by the markets not by users choosing relative importance).
I don't imagine a lot of people are going to do substantial research into random pairs of strangers, but probably have opinions on general trends they'd enjoy testing .
I don't quite understand, are these prediction markets based solely off of people voting? If so, how are prediction markets valuable in situations where the real world is changed by people working towards something and scientific discoveries and not by voting in an election or something like that?
I wondered if Johnson's article covered bookies, and yes it does. But I cackled at this bit in the quoted study:
"We find that, although both markets appear to be inefficient in absorbing the new information contained in the vote outcomes, the betting market seems less inefficient than the FX market. "
That reinforces my inclination that if you're going to throw money around on predictions, stick with the bookies; you'll get better and more realistic odds, and it's gambling anyway no matter how you dress it up with fancy names. If you want prediction markets to take off, you have to open them to the general idiot, and that means dumb money does drive out smart money. Stick with Betfair and Paddy Power!
As to the results of the EU referendum in the study, I think nobody quite believed the Tories would cock it up *that* badly, especially after they won on the Scottish independence referendum earlier. Nobody was expecting "Leave" to win, least of all David Cameron and his flying monkeys, but there you go.
Just as nobody expected Trump to be the Republican candidate in 2016, much less win the thing, but there you go. And nobody expected 2024 to be Trump versus Biden, Round II, but there you go. At this stage, 50/50 is the most prudent thing you can go for if you're not going to simply throw up your hands and go "well what next, we've had plague and war, when is the giant meteor of death going to strike?"
538 (new Disney rebranding!) kicked out/allowed Nate Silver to depart, so I'm not expecting much from them going forward - I also cackled at this part in the Johnson article:
"FiveThirtyEight (now 538) typically releases three election models with varying degrees of complexity — Deluxe, Classic, and Light. The score for FiveThirtyEight shown in this graphic is from their Deluxe model. If the Classic model had been chosen instead, FiveThirtyEight would have outperformed even Metaculus. Light was nearly identical in score to Deluxe."
Translation: when they stuck to Nate's original model, they did best. Now they're trying to gouge more money out of you for fancy 'deluxe' model and it's not as good since they bounced the guy. Heh heh heh.
Has anyone a history of how the prediction markets handled Ron DeSantis (e.g. did they give him a good chance of winning, did that change over time as he performed or didn't in the debates, etc.)?
"I haven’t been following the Trump Legal Issues Cinematic Universe, so I was surprised to see “Trump won’t go to jail” climbing so consistently."
I think it's *because* there are so many cases against him right now, it's overkill. It's reaching "arson, murder and jaywalking" levels. And the Georgia prosecution over election racketeering is being hampered by revelations that the DA has been a naughty girl (her Wikipedia page studiously neglects to mention any of the scandal-mongering but the NYT is not as fastidious):
Yet again, a cautionary tale of "if you're gonna dump your missus for the new squeeze, get the divorce out of the way *first* and try to do it on good terms instead of stiffing her on the money" 😀 His ex is spitting feathers and spilling the tea in her divorce, which is how we learned about Nathan and Fani getting cosy in personal and professional life:
It may be possible for the AI Future to be both "Futurama" and "AI-Dystopia" depending on whether you're an Excelceite (a member of the neo-upper-class living in a vast pleasure-dome) or you're part of the displaced hoards of lower-class depth-grobblers who will live underground in tiered cities. I imagine, however, that most of the actual endless toiling for nuggets of neo-plasmin will be done by robots.
"All the things voters might blame Biden for - inflation, Ukraine, Gaza - happened either well before or well after the early-2023 period when his numbers began to decline."
...you forget the largest elephant in the room: The ongoing massive immigration under Biden. Deeply unpopular, in particular among unskilled US American workers - and among those who try to unionize unskilled US American workers. (Europeans would highly likely have voted for Orban-clones everywhere, if we had US levels of irregular migration to EU borders. You seem totally crazy to us. Strike a deal with Mexico to limit immigration from further South, the way we have done with Turkey and other transit-countries, for God's sake.)
...Add to this the medium-sized elephants: Trump will highly likely reverse the Green Shift and start fracing again, which is popular among those who feel the sting of high energy prices. Plus that he is seen as less woke-friendly. And that he is unlikely to start new wars, plus is likely to ramp up protectionism of US-based industries, in particular if China is the competitor. Biden does some of this, but Trump has more "ownership" of these issues.
The strongest argument against Trump is that he tried a coup when he lost. That hopefully matters to people when election day comes. However, if Biden does not get his act together on the substantial elephant-ssues, my hunch is that a majority of US voters will cross their fingers, hold their noses and take their chances with Trump II. How strong a hunch? 0.65.
Betting on extinction seems like a good example of a niche where betting sites aren’t very useful. If you bet on the world ending, and it does, the people running the prediction market won’t be around to pay you out (and you probably won’t be around either).
While PredictIt is hobbled by the $850 cap, Polymarket isn't. A potential explanatory factor for Polymarket's unreliability is that its resolutions are not determined by any reliable authority, but by a stake-weighted vote of anonymous stakeholders on a crypto blockchain. This has exactly the result you'd expect: when they can get away with it, a small group of wealthy token owners will conspire to resolve a market incorrectly in order to turn a profit. This has happened multiple times.
I'm sorry, what actual evidential basis is there for Biden being even or ahead in some markets? The polls are overwhelmingly against him; history is against him (Trump has greatly overperformed his polls twice and Biden has underpeformed once); several ecomomic and foreign policy indicators are against him, and the "Keys to the White House" system is looking bad for him as well.
And what's in his favour? As Scott says, wishful thinking among left-leaning tech types, and...anything else?
It's hard to take seriously the claimed potential reliability of these markets when I can't see what hard/quantitative evidence exists that the Biden numbers could be based on.
Trump is doing well in polls vs biden now, but that lead will predictably decline when the election gets closer and the media get around to aiming all their cannons at trump
I read yesterday that Trump has declared war on Taylor Swift. This is a Rubicon moment and does not augur well. Biden will have to pull something very special out of the hat to lose. Declare war on Yemen probably won't be enough. If I can get good odds I might go for sell out Taiwan.
Came back to update that the day after the Hur special counsel report Bide has dropped to 3.75 which is 26.7%!! Michelle Obama is no 6.50 and Newson 13.00. This all seems bonkers and I can't imagine the value here is anywhere other than on Biden.
If you are in London, we are doing a prediction market dating show in support of manifold.love on Feb 10th: https://lu.ma/qp9q3tmy
Why would anyone bet on a Paperclipalypse outcome? If that outcome happens, definitionally you won't be around to collect.
Real-money prediction markets indeed did relatively poorly in 2022.
The further past shows a more complex story -- bettors were running slightly ahead of 538 from 2016 to 2020. With 2022 added, 538 gained a slightly edge. That chart and more are here, for anyone curious: https://www.maximumtruth.org/p/deep-dive-on-predicting-elections
I'm not familiar with the particulars here but surely this is misleading:
> Nate Silver is only one person, he has only one area of expertise, and you can’t hire him to predict random things for you (unless you’re rich and he’s bored). If Manifold can apply only-slightly-sub-Nate-Silver levels of analysis at scale to arbitrary topics, that’s a big deal.
Manifold is pretty good at elections *because* bettors defer to models like Nate's. Doing well on elections is little evidence for doing well on arbitrary topics. [Edit: the previous sentence refers to Manifold, not Nate.]
I think your last plot might be related to your fourth.
That is, now that Trump is looking reasonably likely to win the next US election, both Russia and Ukraine are unlikely to make a peace deal until that uncertainty (both whether he wins, and what his actual Ukraine policy will be) is resolved.
If the prediction markets are overvaluing a Trump win, perhaps that represents a kind of hedging where people are willing to take a loss if Biden wins because they think their lives will be good but want to gain some cash if Trump wins to make up for the worsening of their lives.
> Betting sites favored Biden to win all swing states, but Trump to win overall.
That makes it sound like there was one bet on whether Biden would win all swing states, but I think what really happened was a bunch of bets, one per state. If so, then FWIW, it wouldn't be inconsistent to think that Biden had better odds in each individual swing state but worse odds overall (insofar as he would have needed to win more swing states than Trump).
I just did an arbitrage strategy earlier today of buying Biden for 39c on Polymarket and Trump for 47c on PredictIt. This is almost guaranteed to turn 86c into $1.
The only cons are:
* Maybe one of them dies/goes to jail/drops out (but that might boost the odds of the surviving candidate)
* You need to fund an account on PredictIt and Polymarket
* Annoyance of reporting any gains on your taxes (is this gambling winnings or capital gains?) or a small risk of getting punished for not reporting
* Opportunity cost of not buying the S&P500 instead
* Max bet of $850 on PredictIt
https://manifold.markets/ScottAlexander/in-2028-will-ai-be-at-least-as-big?r=QmVuamFtaW5Ja3V0YQ
Could you clarify how you intend to resolve this? How much of a factor would the portion of voters considering it a top issue be compared to your subjective impression of the political conversation?
Thanks for the shout out! I think the link you used might be broken, here's a working link to the article - https://asteriskmag.com/issues/05/prediction-markets-have-an-elections-problem-jeremiah-johnson
I do think prediction markets show enormous promise, but they've got some growing pains to work through. Inefficiencies in those markets don't mean they're not valuable. And even highly liquid, sophisticated markets can have these kinds of irrational episodes where dumb money overwhelms smart money - the entire meme stock phenomenon is evidence of that.
Using decimal odds Trump is 1.83 on Bet365 and Biden is 2.75. That is an implied probability of 54.6% for Trump and 36.36% for Biden.
Stated differently, a $100 dollar bet on Biden would pay back $275 for a $175 profit.
If it is indeed a coin flip this is a massive value. My instincts are that the combined odds of him either losing to trump OR dropping out make that priced properly.
Of course a BET365 user I am likely in the same bubble that drove these odds to 2.75 in the first place.
Here are the next four:
Michelle Obama 10.00
Nicky Haley 17.00
Newsom 23.00
Kennedy jr. 29.00
I’m personally not sure how to feel about Neuralink - on one hand it (or any other complex BCI) could usher in an era as profoundly different to our world as the pre-smartphone era was to today. Yet I feel like there would likely be extremely nefarious or just plain bad uses that might completely wipe out that good (imagine advertising spam in one’s brain!). I’m not really sure which is stronger, or whether the development of BCIs will be similar to most modern technology, somewhere in the middle.
I'd say prediction markets did badly at the 2024 Presidential primaries, too. I guess it's possible that one day we could get a tell-all book revealing that Biden was planning not to run again and then changed his mind at some point, but that's probably not what happened, and even if it was true that wouldn't explain the low prices when he did officially announce. The low prices for Trump to win the Republican nomination are also probably because the market was wrong, not just because Trump got lucky. So not just a 2022 event.
I haven't checked if the markets were worse than Metaculus, to be fair.
I've been enjoying Manifold lately, but I do think it has a limitation in that it's hard to entice people to bet on markets they're not personally interested in. Most of the time and energy goes into the same handful of topics.
The ideal (it seems to me) would be a system that encourages people to dig in and research difficult topics to try and win mana. But it seems to me like the optimizing strategies right now don't really end up doing that.
37% chance of an AI fizzle? Ouch! But, yeah, I'm still waiting for GPT4 to get the "Which inorganic compounds are gases at STP?" close to right. :-(
>Maybe this is just because he has a higher chance of becoming President before the trials are over?
There are a variety of other options. Trump's old and could just die before he gets around to serving time. He could theoretically skip out on his bail, though it's not very likely. The ballot-removal shenanigans could escalate into civil war, which even if he loses might not wind up with him serving time (if he's assassinated or otherwise killed during it, or if he's executed for treason). Nuclear war could happen over Taiwan, which even if Trump doesn't die would see the Blue Tribe crippled.
Regarding "dumb money" in prediction markets - shouldn't it usually be the case that dumb predictions outnumber good ones? Prediction markets are a zero-sum game, and running the market incurs costs which need to be covered, so the median prediction should have slightly negative expected return. Stock markets attract smart money, but they have a positive expected value, since stocks are productive - even if your bets don't outperform the average, you are still likely to make money due to dividends and increased share values.
I'm calling hax.
Nate Silver is more than sufficiently skilled to use manifold and metaculus to help him forecast e.g. tracking people with records of forecasting. 97% chance that Silver or his advisors can do more than enough spreadsheetmancy for that.
On the limits of real-money prediction markets: If you assume no cost for the effort of the research, their miscalibratipn still has to be large enough that betting on them would outperform the s&p in terms of risk-adjusted expected returns for smart money to want to bet on them.
For example if right now I can buy Biden at 40% but think he has a 45% chance to win, that's locking up money for almost a year for 10% returns - but I could get roughly the same expected returns by just buying SPY, and at a lower risk (since betting on election markets is all-or-nothing, which is very high risk). So you can expect them to be fixed by smart money arbitrage to within 10% of accurate, but not 5%.
(This gap should narrow closer to election day).
RE: US election - you'd expect the markets to be misaligned because you have to tie money up for 3/4 of a year to correct them.
RE: liquidity & dumb money - there's a step below professional traders, which is professional gamblers. I work in this kind of area, and several of my friends have built statistical models for Eurovision Song Contest markets to answer questions like "how many points will Lithuania receive?"
As far as I'm aware, the big trading firms don't even bother betting on e.g. bookie odds for Premier League games, which have tens of millions of liquidity, but this is easily enough to sustain professional gamblers (and probably they're in the smaller/more niche markets even).
This affects the arbing problem of election odds too - as people professionalise, they'll create accounts on every real money prediction market and seek arb opportunities. Before that point - is it really worthwhile to spend your time looking for $10 here and there? The motivations for people who enjoy forecasting vs. people who enjoy finding edges in gambling markets aren't the same.
So the liquidity problem isn't as bad as it seems - we don't need prediction markets to attract Wall Street before these problems are solved. But there is likely at least one step change required before then, which is about attracting normal customers. Reach the point where arbing can get you an immediate $100, particularly if you can automate it, and you are set.
The shorter term solution would be if the big prediction markets had liquid markets like "What percentage win chance will this market have Biden at by the end of January", we would see quick agreement via arbing.
(I have Plans and Schemes to work on this, as it's similar to poker site liquidity, one of the few areas where I actually am an expert)
> Why? All the things voters might blame Biden for - inflation, Ukraine, Gaza - happened either well before or well after the early-2023 period when his numbers began to decline.
I wonder if some of this is actually the legal shenanigans. Trust me I don’t want Trump to win, not because of silly white supremacy, or Russian agent stuff - that’s a tale for idiots - it’s because of his volatility. We really don’t need that right now.
Anyway I’m still a bit put off by some of the legal stuff, is the award of $83m in anyway proportional to what happened here?
Seems to me that this could be used against anybody in the future - the republicans will surely use it. Clinton could hardly have survived the scrutiny.
(I’m not talking about the criminal or referral cases which probably have merit).
On Ukraine - I think people have generally overestimated its short- and medium-term prospects at least since fall 2022: the "front line change" market for 2023 (https://manifold.markets/PS/will-the-front-line-in-ukraine-chan-3a82fb352136) started dropping fast in August and then again in September, and the situation was pretty much the same for the first half of 2023 (https://manifold.markets/PS/will-the-front-line-in-ukraine-chan-0591b16ee4cd). I'd guess part of the reason was the same as for the Trump markets - most people on Manifold are pro-Ukraine.
Regarding the problems with prediction markets I think what a lot of people (including the Asterisk article writer) are missing are the nitty gritty of how markets work:
- The headline probabilities the markets show are usually based on the last trade. So if after election day it is still showing Trump at 5% it doesn't necessarily mean people are willing to bet at those odds, more likely it is just that no-one has traded since outcome became clear (why would they?)
- A lot of markets are illiquid and have huge bid-ask spreads. For instance, in the PredictIt Trump-Biden market shown in the post, you could say it favors Trump but the bid-ask spread shows 46%-55% for him to win so I think it's more accurate to say it shows a tie.
- Markets have pretty high fees (PredictIt is 5% widthdrawal) so even if the bid-ask spread was 0 (which it is not) it might not make sense to bet against a 4% outcome even if you are 100% sure it won't happen.
- Cost of capital: if you think there's an opportunity to make 3% after all costs on a market that resolves in 1 year, a rational trader still wouldn't do it. You can just put the money in treasuries and get 4% in same period with zero risk.
- Sometimes there are technicalities about the resolution conditions of markets that make a big difference. I remember there was a market on 2020 election result on the now defunct platform Augur, that seemed to have opportunities for profitable bets on Biden after election day (I made a bet myself). But it ended up taking a long time to resolve (only in January I think) and for a while it actually looked that it could resolve as Invalid. This was because of the particular phrasing of the resolution condition it was unclear if it was unclear if certain technical conditions about electoral collage electors certifying the result were met by the resolution date. A naive observer would just look at the odds the market was still showing in December and decide prediction markets are dumb.
If you take all of these into account, it should be clear that prediction markets show a lot less irrationality than the Asterisk article claims. Of course, it does mean the markets are less useful than you'd naively think for the purpose of actually making predictions. To be fair, a lot of these issues will become less severe if the markets become more liquid.
Would be very funny if it turns out, that non-real-money prediction markets being about as good or even better than real money prediction markets, is not just a temporary bug and it's possible to have non-real-money financial markets that systemically beat the stock market.
The big question with Trump's election is likely going to be one of turn out. Trump turns out the left like nothing else but it's just hard to be all that motivated by a fear of Biden. What balances the odds are the chances that Biden has a big (literal, as in breaks a hip) fall or makes some dumb blunder before the election.
With Trump short of death or coma it's pretty much all priced in.
I suspect this turn out effect is relatively hard to handle in election polling as it's a likely voter effect that's canidate/election dependent.
The "Border" one seems weird - the pre-2022 borders *did* include more than Crimea, namely the Russian-occupied parts of Donbas.
Does this mean we'll never get the kind of large liquid prediction markets we need because the SEC (and similar) will never approve without stronger evidence of benefit which requires approval to reach sufficient capitalization?
>Johnson makes a pretty reasonable guess about the cause: lots of dumb money. People use their pocketbooks to root for their favorite candidate. Normally in a functioning market smart money would take the other side and set the final price, but the high transaction costs, long waits, and regulatory limits on prediction markets mean it’s not generally worth smart money’s time to correct the mispricing; Goldman Sachs isn’t going to hire statistics PhDs to make a model just so they can bet $850 on PredictIt.
I wonder if it is only about this. Some time ago I was reading Stieglitz from the early 1990s, and while I don't have the exact citation at hand, I remember he argued that there a priori can't be a complete set of markets about everything. The most naive form of the argument, number of possible states and contingencies there could be a market about is larger than number of people and time and attention available to them to gather information and trade on it. And this was the ol' good regular economic markets he was mostly concerned with, not prediction markets which can about arbitrary questions.
Fuzzily extrapolating to prediction markets, it looks prediction markets must always be wrong to some extent. For there to be an informational advantage for a knowledgeable trader to gain from, there must a mispricing. If starting a market is very cheap and gathering good information is effortful, there eventually is could be more markets than traders with information to trade. To get your market going, you'd need to compete only to get people to notice your market in the first place and then entice them to trade on it. In the best scenario, perhaps one could come up with some theoretical caps how badly mispriced prediction markets will be, statistically, under different assumptions.
Unrelatedly, maybe there is a reason why fake-money weird internet status points work so well compared to real money: people seldom price their hobby activities the same way they price paid-for work. In real-money markets, smart money eventually should start pricing in inflation and interest rates and opportunity costs compared to investing into other possible, more profitable markets. If the market rewards you with weird internet hobby points? People will take it as a hobby and essentially donate their time and effort.
> I think Paperclipalypse requires human extinction before 2050. It’s at 11%. But Metaculus’ direct “human extinction by 2100” market is only at 1.5%. Either I’m missing something, or something’s wrong. My guess: different populations of forecasters looking at each question.
By the resolution criteria for the first question, "If no resolution has occurred prior to 2050, the panel will meet and will vote on the option that is closest to the real world."
For example, if "only" 50% of humans will die, the answer will be Paperclipalypse, because that will be closer to the real world than other choices. But the "human extinction by 2100" question requires that all 100% of humans will die.
"Rather, the interaction between AI and Homo sapiens ends about the same way that the interaction between Homo sapiens and Neanderthals ended."
I'm going to take the opportunity to rant about this, not because I think this is making this particular error in all its glory, but because expressions of it are fairly common in AI discussions and as soon as we-as-a-collective *stop doing this* we can finally start to think about what it really means for intelligent species to coexist.
Homo sapiens are not a universally superior group that violently wiped out the Neandertals* when they met. Trying to directly compare H. sapiens and Neandertal intelligence is something of a fool's errand, but there's little good case that Neandertals were anywhere near as far behind anatomically modern humans in that respect as the image of them in the cultural consciousness is. We were smarter, yes -- but human intelligence has always had wide error bars. The two coexisted in Europe for thousands of years -- minimum estimates about 1.5k years, probably more double that. Modern humans outcompeted Neandertals over this period through some mix of competition for resources, interbreeding/assimilation (not quite the same, but lumped here), and violence. Neandertals were resource-hungry -- they had very meat-heavy diets -- and even all else equal would come out worse under a pure resource competition than an equivalent species with a lower demand for land-based game. (H. sapiens maybe ate more fish.) A model of interaction between two sophonts** where one rapidly and violently massacres the other because of their superior intelligence can't be one based on Neandertals.
"Sophont", then, takes you down some avenues. I think the case that cachalots ('sperm whales') are sapient in the same sense as humans, and possibly smarter than humans were at a comparable stage of evolution, is extremely strong. (See https://web.archive.org/web/20160322095648/http://cachelot.com/ for an introduction to the topic.) Humans have certainly exploited cachalots, to the point of what is from this perspective a particularly heinous genocide. When I talk about cachalot intelligence with people, this is often a tricky point for them to think about: how could humans have committed genocide against a species at least their equals? Wouldn't the whales just outsmart us? Well, they *did* fight whaling ships -- at least some whales (e.g. Mocha Dick) were able to recognize their patterns, seek them out, and attack them if they attacked -- but, no, human history shows little clear correlation between intelligence and the escaping of atrocity. Whalers exploited things like the desire of cachalot mothers to defend their children in order to wipe out whole pods; plenty of humans who wanted to cause one another harm have exploited similar desires.
We know that two sapient species coexisted for thousands of years before one went extinct under complex circumstances. There is good reason to think two coexist now, and some to suspect the one that caused more harm to the other is not the unambiguous intellectual superior. This is to say that bringing in another intelligence has very complex consequences, especially if you think that intelligence will be 'clearly superhuman', but you can't pattern a superhuman intelligence to the perceptions people have of previous sophont interactions and get anything clearly representative of reality. In particular, you can't extrapolate from the actions of H. sapiens to the actions of any given "superior species". This is one of those interesting circumstances where people try to correct from thinking other intelligent beings would be "like humans", and *still* correct to them being too humanlike. The real outcome of throwing a third intelligence into the mix is trickier, less dependent on whether that intelligence is clearly "smarter than" humans or not (because this probably doesn't play the primary role in human exploitation), and not extrapolatable from half-remembered impressions of Neandertals.
*I prefer the spelling without the h, because it expresses English pronunciation more clearly and is what the valley is called now.
**"Intelligent AI will be sapient" is another of those "your AI is still too humanlike" things. We have intelligent AI right now,*** which is not...sentient, at least. You could, if you want to get really fun, try to argue that "sapient but not sentient" is a coherent idea. At this point, it might not be impossible. But you'd need a very rigorous definition of sapience. I'm saying "sophonts" here just to address the fact that all organic Terran species past X intelligence level are sophonts.
***AGI, the letters defined like that, is here. People wrote their definitions in the 2010s of what they expected AGI to be. It's here. We have it. If you mean something more intelligent, use a different term -- it's not our fault everything turned out to be less correlated than even the "no, seriously, AGI will not be like humans" guys thought.
> Will the Ukraine war end by 2025
> What am I missing here? It’s got to be higher than 8%, right?
No. Putin is not interested in peace or ceasefire. He is fully ready to fight for as long as it takes. Especially now when he hopes for a Trump victory.
The popular uprising in Russia in 2024 is almost impossible (<1%). Russians are not that tired of the war.
Neither are Ukrainians. They realize that any peace agreement with territorial concessions only means that Russia will subsequently attack them again, and are also determined to fight for as long as it takes.
This war belongs to the same category as the Soviet–Afghan War or Vietnam war.
> Will the Ukraine war end by 2025
> What am I missing here? It’s got to be higher than 8%, right?
Both sides still hope for some kind of decisive victory that will ultimately resolve their dispute (i.e. other side's forces are in full rout, a change of power happens, and some kind of enforcible restrictions are placed on enemy's ability to restart the war later - for an Ukrainian victory, this would have to be (almost) instant NATO membership, for a Russian one, a bunch of Russian military bases on the remaining Ukrainian territory, if any).
I really don't understand what Ukraine hopes for (a direct NATO involvement?), but Russia has some hopes for drying out of Western funding for Ukraine, and if it doesn't happen, a win in battle of attrition. But even if US fails to provide any money or arms in 2024 at all (which is, I think, unlikely), Ukrainian forces may well survive into 2025 with reserves, EU help, and sheer size of the country. And who knows what will happen after US elections - Republicans might find renewed love for Ukraine once it's Trump's war to win or lose, and not Biden's (in case of Trump win). So "peace by 2025", or even a temporary ceasefire seems very unlikely - 8% is about right. For now, both sides still have enough people, vehicles, ammo and will to fight.
FWIW, assuming they are they are their respective party's nominees on election day, Swift Centre forecasters put 52% on Biden winning, 48% on Trump winning, and also forecasted on some interesting conditional questions: https://www.swiftcentre.org/biden-trump-rematch-chances/
I think it's really important to include a well-established real money market that comes from the betting world - they have very different audiences to those from the prediction world, and that affects the odds they generate.
Here's betfair's "exchange" market on the US Presidential election:
https://www.betfair.com/exchange/plus/politics/market/1.176878927
It has Trump at 2.24 (ie bet £1 get £2.24 if he wins) and Biden at 2.94 - that is to say that you can make a profit betting that one of them will win without specifying which one. 26% ROI on a bet that will run a little under a year.
This is being driven by the fact that the primaries aren't quite over - Haley has about a 4% chance of being president per the market, weirdly they have Michelle Obama at 7% (huh?), and various others at 1-2% (Newsom, RFK Jr, Philips, Harris).
I don't think the "dumb money" problem is solved by scale or relaxed betting limits, because it doesn't address the problem that prediction markets are inherently zero sum and the smart money needs to get paid. Prediction is hard work; if we haven't already maxed out the unpaid-hobbyist-predictor market we eventually will and probably through the unpaid markets like Metaculus. To do better, we're going to need a way for the smart money to place a positive expected value on investing in prediction markets.
In the case of e.g. the stock market, that's easy because the stock market is positive-sum; you're trading claims on the future income of profitable enterprises. Money comes in to the market from traders making their bets, but also from companies paying their dividends. If you smartly place the right bets, you wind up owning a nice dividend stream (which you can sell for immediate $$$ if you need that). It's certainly possible for smart money to prosper by betting on dumb money, but smart money can also prosper by accurately predicting which companies will be most profitable, and we get useful signal out of stock markets.
In a prediction market, money only comes in from traders placing bets, so smart money can only prosper by way of dumb(*) money -> smart money transactions. The smart move is always to figure out what dumb money is going to do, and do it first. So everybody in the market, smart and dumb alike, is going to be making the dumb-money moves. That doesn't give us a useful signal.
You could get around this by having someone altrustically fund markets with the expectation of losing money but getting good data. Or maybe the subsidy could come from people who would profit (outside the market) from good data, but that comes with an obvious free rider problem and it would probably be more effective for them to just privately hire a team of forecasters. There are other possibilities, but I think they're all pretty low probability.
* Dumb in market-profitability terms. If you have some other reason for investing, even just "I'm rich and this is fun", making unfavorable bets may not be absolutely dumb,
Re changes to manifold love, I think the best thing would be to bet on abstracted questions eg "odds heterosexual make and female both 30 years old reporting a good date" or "odds that two people who both say they like horror movies like each other", and aggregate those into giving the results for each person. (in a similar way to how OkCupid used to use the questions for compatibility score, but with the weightings set by the markets not by users choosing relative importance).
I don't imagine a lot of people are going to do substantial research into random pairs of strangers, but probably have opinions on general trends they'd enjoy testing .
I don't quite understand, are these prediction markets based solely off of people voting? If so, how are prediction markets valuable in situations where the real world is changed by people working towards something and scientific discoveries and not by voting in an election or something like that?
I wondered if Johnson's article covered bookies, and yes it does. But I cackled at this bit in the quoted study:
"We find that, although both markets appear to be inefficient in absorbing the new information contained in the vote outcomes, the betting market seems less inefficient than the FX market. "
That reinforces my inclination that if you're going to throw money around on predictions, stick with the bookies; you'll get better and more realistic odds, and it's gambling anyway no matter how you dress it up with fancy names. If you want prediction markets to take off, you have to open them to the general idiot, and that means dumb money does drive out smart money. Stick with Betfair and Paddy Power!
As to the results of the EU referendum in the study, I think nobody quite believed the Tories would cock it up *that* badly, especially after they won on the Scottish independence referendum earlier. Nobody was expecting "Leave" to win, least of all David Cameron and his flying monkeys, but there you go.
Just as nobody expected Trump to be the Republican candidate in 2016, much less win the thing, but there you go. And nobody expected 2024 to be Trump versus Biden, Round II, but there you go. At this stage, 50/50 is the most prudent thing you can go for if you're not going to simply throw up your hands and go "well what next, we've had plague and war, when is the giant meteor of death going to strike?"
538 (new Disney rebranding!) kicked out/allowed Nate Silver to depart, so I'm not expecting much from them going forward - I also cackled at this part in the Johnson article:
"FiveThirtyEight (now 538) typically releases three election models with varying degrees of complexity — Deluxe, Classic, and Light. The score for FiveThirtyEight shown in this graphic is from their Deluxe model. If the Classic model had been chosen instead, FiveThirtyEight would have outperformed even Metaculus. Light was nearly identical in score to Deluxe."
Translation: when they stuck to Nate's original model, they did best. Now they're trying to gouge more money out of you for fancy 'deluxe' model and it's not as good since they bounced the guy. Heh heh heh.
Has anyone a history of how the prediction markets handled Ron DeSantis (e.g. did they give him a good chance of winning, did that change over time as he performed or didn't in the debates, etc.)?
"I haven’t been following the Trump Legal Issues Cinematic Universe, so I was surprised to see “Trump won’t go to jail” climbing so consistently."
I think it's *because* there are so many cases against him right now, it's overkill. It's reaching "arson, murder and jaywalking" levels. And the Georgia prosecution over election racketeering is being hampered by revelations that the DA has been a naughty girl (her Wikipedia page studiously neglects to mention any of the scandal-mongering but the NYT is not as fastidious):
https://www.nytimes.com/article/fani-willis-georgia-trump.html
Yet again, a cautionary tale of "if you're gonna dump your missus for the new squeeze, get the divorce out of the way *first* and try to do it on good terms instead of stiffing her on the money" 😀 His ex is spitting feathers and spilling the tea in her divorce, which is how we learned about Nathan and Fani getting cosy in personal and professional life:
https://www.theguardian.com/us-news/2024/jan/19/fani-willis-travel-paid-nathan-wade-trump-georgia-case
On the Trump-Biden question, The Economist got a significantly better result for the Democratic candidate by asking the Good Judgment Project: https://www.economist.com/the-world-ahead/2023/11/13/what-the-superforecasters-predict-for-major-events-in-2024 ; 65-35 Democratic-Republican odds.
It may be possible for the AI Future to be both "Futurama" and "AI-Dystopia" depending on whether you're an Excelceite (a member of the neo-upper-class living in a vast pleasure-dome) or you're part of the displaced hoards of lower-class depth-grobblers who will live underground in tiered cities. I imagine, however, that most of the actual endless toiling for nuggets of neo-plasmin will be done by robots.
I'm not personally a believer in Paperclipalypse, but, in principal it could (a);eventually happen (b) happen after 2050.
Might take longer than 2050 for either of
A) AI becomes truly dangerous, e.g.if it's going to need another theoretical breakthrough
B) for AI to become so embedded in society that it's just impossible to turn it off, even after it turns rogue.
This is "slow takeoff" paperclipalypse, versus fast takeoff where e.g. OpenAi creates something that kills us all tomorrow.
The AsteriskMag link at the very top is broken.
For anyone wondering about the massive spike in "Will Bard overtake ChatGPT", it's because Bard just jumped to (almost) the top of Chatbot Arena
https://www.reddit.com/r/LocalLLaMA/comments/1abqzsj/chatbot_arena_leaderboard_updated_latest_bard/
"All the things voters might blame Biden for - inflation, Ukraine, Gaza - happened either well before or well after the early-2023 period when his numbers began to decline."
...you forget the largest elephant in the room: The ongoing massive immigration under Biden. Deeply unpopular, in particular among unskilled US American workers - and among those who try to unionize unskilled US American workers. (Europeans would highly likely have voted for Orban-clones everywhere, if we had US levels of irregular migration to EU borders. You seem totally crazy to us. Strike a deal with Mexico to limit immigration from further South, the way we have done with Turkey and other transit-countries, for God's sake.)
...Add to this the medium-sized elephants: Trump will highly likely reverse the Green Shift and start fracing again, which is popular among those who feel the sting of high energy prices. Plus that he is seen as less woke-friendly. And that he is unlikely to start new wars, plus is likely to ramp up protectionism of US-based industries, in particular if China is the competitor. Biden does some of this, but Trump has more "ownership" of these issues.
The strongest argument against Trump is that he tried a coup when he lost. That hopefully matters to people when election day comes. However, if Biden does not get his act together on the substantial elephant-ssues, my hunch is that a majority of US voters will cross their fingers, hold their noses and take their chances with Trump II. How strong a hunch? 0.65.
Betting on extinction seems like a good example of a niche where betting sites aren’t very useful. If you bet on the world ending, and it does, the people running the prediction market won’t be around to pay you out (and you probably won’t be around either).
While PredictIt is hobbled by the $850 cap, Polymarket isn't. A potential explanatory factor for Polymarket's unreliability is that its resolutions are not determined by any reliable authority, but by a stake-weighted vote of anonymous stakeholders on a crypto blockchain. This has exactly the result you'd expect: when they can get away with it, a small group of wealthy token owners will conspire to resolve a market incorrectly in order to turn a profit. This has happened multiple times.
I'm sorry, what actual evidential basis is there for Biden being even or ahead in some markets? The polls are overwhelmingly against him; history is against him (Trump has greatly overperformed his polls twice and Biden has underpeformed once); several ecomomic and foreign policy indicators are against him, and the "Keys to the White House" system is looking bad for him as well.
And what's in his favour? As Scott says, wishful thinking among left-leaning tech types, and...anything else?
It's hard to take seriously the claimed potential reliability of these markets when I can't see what hard/quantitative evidence exists that the Biden numbers could be based on.
>The real odds are 50-50.
Do you endorse this choice of words? Maybe I'm nitpicking, but I did get kinda stuck on the phrase "real odds".
"a revolution that’s entirely comparable to the scientific, industrial, or information revolutions"
One of these things is not like the other, one of these things doesn't belong
https://lukemuehlhauser.com/there-was-only-one-industrial-revolution/
Trump is doing well in polls vs biden now, but that lead will predictably decline when the election gets closer and the media get around to aiming all their cannons at trump
I read yesterday that Trump has declared war on Taylor Swift. This is a Rubicon moment and does not augur well. Biden will have to pull something very special out of the hat to lose. Declare war on Yemen probably won't be enough. If I can get good odds I might go for sell out Taiwan.
Came back to update that the day after the Hur special counsel report Bide has dropped to 3.75 which is 26.7%!! Michelle Obama is no 6.50 and Newson 13.00. This all seems bonkers and I can't imagine the value here is anywhere other than on Biden.
Nate Silver has at least 3 areas of expertise :)