Astral Codex Ten

Smarkets

13 min read

Smarkets

Article

Smarkets is a recurring organization in the Astral Codex Ten archive, appearing 2 times across 2 issues between July 02, 2024 and November 05, 2024. The archive places it in contexts such as ""their model is based on results from Betfair, Smarkets, PredictIt, and Polymarket.""; “Their model is based on results from … Smarkets”; “Smarkets got to Trump 59% on 10/16”. It most often appears alongside Betfair, CFTC, Democrats.

Metadata

  • Category: Organizations
  • Mention count: 2
  • Issue count: 2
  • First seen: July 02, 2024
  • Last seen: November 05, 2024

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Prediction Markets Suggest Replacing Biden
July 02, 2024 · Original source
I assume they chose these three because they’re the only ones discussed enough to have enough data. I am following their lead. I appreciate John and Maxim’s work, but I’m not completely comfortable trusting it. Their model is based on results from Betfair, Smarkets, PredictIt, and Polymarket. But I don’t know much about the first two (as an American, I’m banned from even reading Betfair), and the latter two are notoriously bad at partisan political questions. They usually overestimate Republicans’ chances, partly because Democrats’ opposition to online political betting has turned the pool of online political bettors disproportionately red. While a fluid and easily-accessible prediction market should be able to avoid biases like these, neither PredictIt nor Polymarket really qualifies. The CFTC, which regulates prediction markets, has crippled both - PredictIt has very low maximum investments per market, and Polymarket is crypto-only and banned for US citizens. These have prevented their biases from being corrected and made both of them perform relatively weakly in head-to-head contests. And Stossel/Lott’s focus on betting sites automatically excludes two of the biggest and most historically accurate forecasting engines from their calculation - Metaculus and Manifold. In order to get numbers I trusted more than theirs, I looked at Metaculus, Manifold, PredictIt, and Polymarket, weighting each by how much I trusted it. Here’s what I found: The Biden number is about 4% higher than Nate Silver’s model over the same time period; see below for why that might be. [EDIT 7/2/24: Original version had a miscalculation which decreased everyone’s odds by about 10%. Above version should be correct.] You can find my sources at the bottom of the post. “Explicit” odds are based on questions like “What are the chances of Biden winning if he is the nominee?” “Implied” odds were generated by combining the questions “What is the chance of Biden being the nominee?” and “What is the chance of Biden winning?”; this is safe enough with Biden, but with unlikely nominees like Newsom, some of the percentages can get small enough that they start running into small-number-biases and become less trustworthy. I’ve weighted each market’s explicit calculation higher than their implicit one to compensate. A possible objection to these results: conditional probabilities don’t exactly reflect the intuitive concept of decision-making. That is, we’re not asking “We want to know whether or not to keep Biden, so what are the chances that he’ll win if we do?”, we’re asking the market for the chance that he’ll win, in the set of worlds where people decide to keep him for other reasons. We should expect this to overestimate his performance. That is, imagine that tomorrow, Biden has completely recovered, he easily wins his next debate with Trump, and everyone agrees the most recent debate was just a fluke - in that world, he is both more likely to be nominated and more likely to win. Alternatively, if tomorrow he gets much worse and can’t even speak in full sentences, he’s much less likely to be nominated and much more likely to lose. Since the real world includes both those possibilities, restricting ourselves to the set of worlds where he gets nominated means we’re overestimating the chance that he wins. There are similar-albeit-less-severe problems with other candidates - if we choose Newsom, that might be because he won some kind of debate or process versus Harris and all the other potential replacements. Overall I expect this to be mostly correct, but probably overestimate Biden’s chances by a percent or two relative to others. Along with these three candidates, Metaculus had an explicit “should the Democrats replace Biden?” question: Manifold also asks how Democrats will do if they replace Biden (without specifying a particular replacement): We can compare this to their Biden market… …and find that once again, they expect replacing Biden to go better (though I think 51% is just cope). At the Manifest prediction market conference in early June, I interviewed Nate Silver: …and asked him for his probability that the Democrats would win this election, versus his probability that the Democrats would win conditional on Biden not being the nominee (specifically “drops dead tomorrow of natural causes”). He said 40-45% chance normally, 50% chance without Biden. This was before the debate, but I think it matches the markets’ opinion that switching candidates would help the Democrats’ chances - and this has only become more true since the debate. On the other hand, polls asking people how they would vote in possible matchups don’t show any advantage of alternate candidates over Biden. Here’s the only post-debate poll I could find: And if Biden does need to be replaced, Democrats mostly support Harris, who the prediction markets find least promising: Maybe Democrats are the wrong people to ask - they’re already going to vote Biden, so you want someone who’s more attractive to independents. Of course, in a normal primary it would be Democrats making the decision. But if elites are going to do something behind closed doors, maybe they should take advantage and choose the candidate most likely to win, for once. I think these polls are the strongest objection to the prediction markets’ verdict. You could make an argument where prediction market users are mostly educated liberal white males, and even though they’re incentivized to honestly determine what ordinary people think, they’re too out-of-touch with ordinary people to do so effectively. Or they might be over-fixating on “voters don’t like Biden’s senility” without considering that, even if voters didn’t know Biden was currently senile before Thursday, they probably guessed that he would become senile sometime in his four-year term, and had basically accepted that his aides would do the hard work. Maybe they prefer a well-known likeable incumbent over an unknown quantity (and the unknown quantity’s potential new/weird aides), even if the well-known likeable incumbent is senile. Maybe elites know more than we do about how hard it is to inject a new candidate at the last moment, how dangerous it is to have someone who hasn’t been thoroughly vetted for scandals, et cetera. Still, for now I trust the prediction markets. I think replacing Biden would add ~10 prcentage points to the Democrats’ chance of victory. At the end of this post, I’ll list the prediction markets I’m using as sources. But before then, a brief interlude of: Fuzzy Subjective Human Factors I Am Not Really Qualified To Talk About Many people on Twitter are asking “how could anyone possibly have been stupid enough to not realize that Biden was senile?” I was that stupid. I didn’t say it openly, because I’m at least smart enough to have a high threshold for giving my opinion on political things I don’t know much about. But I thought it in my heart. So in case the people asking “how could anyone have been that stupid?” actually want an explanation, here’s my former reasoning. Republicans have been accusing Biden of being senile (and the Democrats of hiding it) for at least five years now. Before the 2020 debates, they were excited that this was when they could finally prove once and for all that Biden was senile. Then Biden did fine, and they retreated to “well he’s senile but they have some secret drug they’re giving him, just during debates, that makes him look fine”. Notice this is from 2020; according to polls, he did win the debate that year (source) I think a lot about experimental cognitive enhancement drugs, and I can say with confidence that nothing like that exists. Stimulants can help people with mild dementia be more active and motivated, but they don’t really improve cognition directly, and they can’t make a demented person temporarily lucid. Still, for the past four years, every time Biden was going to do something - a press conference, a State of the Union, whatever - the Republicans would say “ha, this time is going to be the proof that he’s senile!” And then he would always do fine, and they would retreat back to “I guess he used the secret drug this time too”. The satire site Babylon Bee had some funny articles about this: Babylon Bee, after Biden gave a good State of the Union speech earlier this year. Meanwhile, the Democrats were spreading the alternate narrative that Trump was senile. This one has gotten less press, because I don’t know how many people really believed it. But it came up occasionally, along with out-of-context video snippets where Trump said or did something dumb or meandering. Of course, anybody with a presidential candidate’s level of public exposure will have a few gaffes. Even if they don’t, you can always deceptively crop something so it looks like they did. Wait, why is a psychoanalyst getting quoted as a top expert in dementia? (source) I didn’t know you could diagnose someone via Change.org petition, but 2544 people who claim to be licensed professionals can’t be wrong! So with the constant attempts to prove that both candidates were senile, the constant demonstration by both candidates that they weren’t, and the constant retreat into conspiracy theories of “I guess he used the magic drug again but we’ll get him next time!”, I just tuned out this entire category of thing. And I guess I kept it tuned out longer than I should have, whoops. Reversed stupidity is not intelligence. Even if liars are saying something for their usual liar reasons, it can still be true. For twenty years, people spread false rumors that Castro was on his deathbed, but this didn’t make Castro immortal. In the same way, I should have figured out that even if I couldn’t trust any particular claim that Biden was senile, the prior for an 81 year old becoming senile was still high. But I guess I assumed that if he was becoming senile, some Democratic elites would have secret knowledge about it, and they couldn’t possibly be so stupid as to deny it while also scheduling him for a debate where it would inevitably come out. So I figured the Democratic elites who were closest to him thought he was doing well, and I trusted them more than the people who had been wrong every time for the past five years. I’m still confused what those elites were thinking. Reading the news coverage for the past few days (including some video clips from a post-debate rally where he seemed noticeably better) it seems like some combination of: He has good days and bad days, and they were hoping this would be a good day.
Inline links: https://substackcdn.com/image/fetch/$s_!3oIv!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F31e3710e-6231-49c6-989b-1bc16f70fdf7_833x288.png, https://substackcdn.com/image/fetch/$s_!Fr-Z!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa7933bcc-fc24-4043-9064-47a958e24497_787x306.png, https://manifold.markets/HamishTodd/conditional-upon-the-democratic-par, Here’s, https://substackcdn.com/image/fetch/$s_!Q3fs!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ff99e148f-4051-40f9-b0c6-22eb10289854_1500x1091.png, https://substackcdn.com/image/fetch/$s_!MXOh!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7d694e0c-d306-4578-a55d-5314ca965a83_1500x1246.png, https://substackcdn.com/image/fetch/$s_!qt9P!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa11d1a27-cfc2-4708-9db1-e344b391ac9a_1288x758.png, source, think a lot about experimental cognitive enhancement drugs, https://substackcdn.com/image/fetch/$s_!shq2!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F92d67bb2-5940-44ba-ad21-df822276280a_781x829.png, Babylon Bee, https://substackcdn.com/image/fetch/$s_!2y5u!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa986c7ac-16c2-4987-b520-8f6ddf9ad13d_757x690.png, https://substackcdn.com/image/fetch/$s_!SN0G!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6068601f-0eef-4a1c-b885-19f5712decf1_751x668.png, https://substackcdn.com/image/fetch/$s_!vL01!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fcb26c689-1e81-422b-9493-ae44cac59125_858x814.png, source, https://substackcdn.com/image/fetch/$s_!6ssu!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F760276bb-2f91-4545-bbf5-d3f9fe911de7_975x654.png, Change.org petition, Reversed stupidity is not intelligence
Mantic Monday: Judgment Day
November 05, 2024 · Original source
On October 14th, Polymarket gave Donald Trump 54% odds of winning, compared to Nate Silver’s 49% and Metaculus’ 45%. Whatever, everyone knows Polymarket has a small right-wing bias, and 5% isn’t too bad. Three days later, it had risen from 54% to 61%, despite no news and no change for Metaculus or Nate, bringing the Polymarket/Silver spread to an unprecedented 11%. What happened? This is the rare prediction market story where the answers are already in the New York Times and the Wall Street Journal: one really rich guy put $30 million on Trump (a recent followup by Jorge Velez claims it’s actually more like $75 million). Although he prefers to remain anonymous, reporters have talked to him and are able to reveal that he’s French, goes by “Theo”, is a former banker, and has no insider connections. He just a normal rich guy who really thinks Trump will win. This is exactly the sort of shock that prediction markets are supposed to be resilient against. Instead, the market stayed at 61% for days, swung even higher for a while, finally fell back down two weeks later, then went back up again. What happened? The simplest story would be insufficient liquidity: there just weren’t enough people to gather the $75 million it would take to bet against Theo. This is superficially plausible: Polymarket requires crypto and bans Americans, so the mispricing couldn’t be corrected until enough crypto-literate, American-election-following foreigners showed up to bet $75 million. That’s a tall order, and maybe it took two weeks. But the simple story seems wrong. Other real-money markets rose approximately in tandem with Polymarket. For example, Smarkets got to Trump 59% on 10/16, and peaked at 64% on 10/30. Kalshi followed a similar path. Both tracked Polymarket, not Nate Silver or Metaculus (neither of whom ever went above Trump 55% since Harris joined the race). So I think the remaining stories are: Theo made his giant bet on Polymarket. By coincidence, at the same time, bettors everywhere massively overcounted a few good polls for Trump and started a feeding frenzy on pro-Trump shares. This made all other markets gain, and Polymarket stay at its Theo-caused peak, until a few bad polls for Trump brought everyone back to reality last week.
Inline links: New York Times, Wall Street Journal, followup, Smarkets, https://substackcdn.com/image/fetch/$s_!8twW!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F25ccd651-3404-446a-a7cb-5747af316038_1180x315.png
Theo made his giant bet on Polymarket. Everyone else did some kind of galaxy-brained meme stock herding behavior. Maybe bettors, both on Polymarket and Smarkets, thought that Theo might make other big bets, or that other people would falsely expect this, or that other people would get excited and follow the vibes - and they hoped to profit off of those people and sell before the election.
[EDIT, thanks to commenter Chastity for the explanation] There wasn’t enough liquidity to get Polymarket back down, but traders looking for free money arbitraged with Polymarket-Kamala plus Smarkets/Kalshi-Trump. Polymarket is so much bigger than the other two (and Theo kept driving it back up), so this just looked like Smarkets and Kalshi going all the way up to Polymarket levels.
Inline links: commenter Chastity