PredictIt: Presidential Market is Increasingly Wrong

post by Zvi · 2020-10-18T22:40:00.968Z · LW · GW · 28 comments

Previously about PredictIt this election cycle: Free Money at PredictIt: 2020 General Election

Free Money at PredictIt?

It is important, on occasion, to state the obvious, on the record, at the proper time.

The prediction (alternatively, read: gambling) markets on the 2020 Presidential Election increasingly do not make sense. 

In particular, their movements over time do not make any sense.

Nate Silver’s model at 538, which puts Trump at 12.5% or so, does not take into account the possibility of anyone taking extraordinary measures to distort who is physically able to vote or to have their vote counted, or attempts to set aside the vote and appoint alternate electors or remain in office anyway, or anything like that. It is very reasonable to thus assign some probability to ‘Trump would lose an election with only a historically typical amount of shenanigans, but wins because of an ahistorically highly impactful level of shenanigans.’ You could also use different assumptions than Nate, and think the baseline situation is closer to even.

I even can imagine a world in which, right now, Donald Trump is currently 37% or so to win the Presidency – and I observe all the same things I’ve observed this year. In isolation, it’s possible. 

What I cannot explain, at all, is how this can be true if on June 20, three months ago, the market was 63-39, and now it’s 65-40, all but unchanged. Trump improved a bit, then got worse again, with Biden’s low being at 55. 

Note that on June 20, the 538 model had Trump at 22% to win (it began at 30% on June 1). The progression he offers makes sense and is consistent. The market’s doesn’t, and isn’t.

You cannot tell me that Trump’s situation is about as good as his situation at that time, at the start PredictIt’s longest quick lookback option.

Seriously, what the hell, market?

If you want to be super generous to Trump and assume that the market was correct because he has a fixed chance of winning ‘despite everything’ then we can put him at 38% on June 20, then put him at 29% to win now. That’s the hard upper bound, and I don’t think that’s reasonable – winning ‘despite everything’ gets harder as winning the regular way gets harder, so I think the ‘be afraid of shenanigans’ case caps out around 25%.

This is why I seriously considered betting on Trump months ago. The theory was, if things went well for him, I’d have a good bet. If things went poorly for him, I could cover my bets at little cost. One of my weaknesses as a gambler was always a reluctance to pull that trigger.

Many secondary markets are even more absurd than the primary one.

Biden is still 8-9% to drop out by 11/1 on 10/18. They thought he was 12% to drop out by 11/1 three months ago. If you can explain how those two agree with each other, please do so in the comments. You may wish to note “Challenge Accepted.” 

Time and again, I take what in retrospect was a very good bet when compared to doing nothing, but a very bad bet compared to waiting for an even better one. And it’s predictable.

It’s tough advice to follow, even when you know it is right, not to fire until you see the whites of their eyes.

In this case, if you really want to maximize your true expected value, my guess is that you should wait at least another week, and my gut tells me wait until two days before the election – you might get the best price of all on election day itself, but that’s definitely risky. 

Note that if you have the patience for it, some of the multi-way markets offer better prices than the main one by buying up all the “No” sides, and also allow you to get down additional funds. You can only bet $850/market, but there are at least five more or less clean variations of markets for who wins (including the female VP option, electoral college margin, popular vote margin, and party win), and if you’re willing to settle for close approximations like betting on Pennsylvania you can get a bunch more. Usually your prices on secondaries are a little worse then the primary price on Biden if they’re two-way, better if they’re multi-way. Thus, you can get down something like $10,000 at PredictIt if you want to and can take the odds hit, and you can use friends to further multiply that, but if you do that then you’ll be stuck with far too much funding there and nothing to do with it for a while, and facing expensive withdrawal fees. If you have a legal other way to bet, and can handle betting on that side the election, do take advantage of it.

If you want to stay side neutral, all the arbitrage opportunities not involving third party candidates that were there last time are still there. If anything, they’re bigger and over a smaller period of time.

I also think it is wise, as others have suggested, to top off your supplies in case the supply chain suffers a disruption after the election. The chances of this seem much smaller than they did in June (again, market is nuts) but the cost/benefit analysis on such actions is overwhelmingly positive. Make sure that if you can’t resupply for a few weeks, nothing terrible happens, have a rough idea what you’ll do if things start to get scary (in terms of a disputed election, not in terms of the wrong person wins), and expect to disregard all that most of the time.

This is not the first election with rather dumb prices that refused to update on time passing and the situation becoming lopsided. In 2008, McCain stayed close to 50% most of the way, despite an obviously degenerating position. Thank you, InTrade. In 2012, Romney was clearly priced too high given the stable nature of the polls and the way the electoral college stood. 

There are times when the market is nuts. This is one of those times. There are various forces contributing to that, but it is not purely a matter of people being size limited, as there are places for non-Americans to get down large amounts. 

Watching the election odds on November 3-4 will be interesting. Often when things transition from pre-game wagers to live betting during the game, the frame for deciding odds shifts, or you find out that the real odds you were starting from were somewhat different from what was offered for wagering. Other times, things are more consistent. Either way, the deltas for events when they happen are always interesting, and the stock market’s movements also will tell the story (the market itself will be closed, so you want to watch the ESZ0 Index contract and others like it). 

Or, of course, if you’d be sitting there terrified, do your best not to do that, and counterprogram something else and hope it’s over when you wake up. If you need someone’s permission to do that, I’m giving it to you, along with permission to not watch the second debate if it happens. You’re welcome.

28 comments

Comments sorted by top scores.

comment by philh · 2020-10-22T17:05:42.510Z · LW(p) · GW(p)

A related crazy-seeming market here is Trump's exit date. Will he complete his first term? Predictit says 85% yes, betfair says 90% yes. The only way I can see that not happening is if he loses the election and quits out of spite or whatever, and I wouldn't be confident enough to dismiss that out of hand. But Good judgment open doesn't think it's likely, 99% he completes his term. (I guess unless he quits on inauguration day?)

comment by Zvi · 2020-10-22T23:34:31.264Z · LW(p) · GW(p)

He might resign so Pence cod pardon him. I would be excited to sell 99s, but I would buy 85s.

comment by Dagon · 2020-10-22T23:12:58.613Z · LW(p) · GW(p)

I'd put "suicide or flight before inauguration day to avoid arrest " at 2-3%, and death by accident or illness at another percent (would have said closer to 5% a few weeks ago).  Not sure how anyone gets down to 85%, but 99% is too high.

comment by AllAmericanBreakfast · 2020-10-21T05:27:04.054Z · LW(p) · GW(p)

Too late!

From the supplementary comment to the rules for this bet:


Note to traders: As of this time, the contracts for Joe Biden and Donald Trump have reached the limit for the allowable number of traders.

comment by Zvi · 2020-10-21T11:05:47.152Z · LW(p) · GW(p)

I still find it so weird that this is true. In 2024, I recommend buying  or selling one contract in each major candidate during the primaries. 

For now, that means you have to use other markets, but using the Republican/Democratic market gets you your first $850 (or us the margin of victory markets if you want a better price, and there's two of those, which gets you to $2550), then you can get creative.

comment by AllAmericanBreakfast · 2020-10-21T05:52:08.332Z · LW(p) · GW(p)

And here I was calculating the EV of a Biden bet.

FWIW, after taking fees into account, you'd need to think Biden has better than a 70% chance to win the election to make it a rational bet at the current price of $0.64/share. If the average of 538 and The Economist's forecast (90% chance of Biden victory) is right under the assumptions of their model, then you'd need to think there was a less than 22% chance of some sort of Black Swan.

Could be electoral shenanigans (fraud, incompetence, interference, successful legal challenge) or catastrophe (Biden has an aneurysm) unaccounted for by the models of PredictIt and The Economist makes Trump win the electoral college, or makes neither of them/nobody win.

  • Certainly that chance is not 0%. But if it were, your EV for a $1,000 bet under these assumptions would be winning $287.
  • If you think there's a 5% Black Swan chance (85% chance of Biden victory), your EV is $216.
  • If you think there's a 10% Black Swan chance, your EV is $144.
  • If you think there's a 15% Black Swan chance, your EV is $73.
  • If you think there's a 20% Black Swan chance, your EV is about nil.
comment by Liam Donovan (liam-donovan-1) · 2020-10-25T23:27:49.072Z · LW(p) · GW(p)

FWIW,  understanding is that the 538 model is meant to account for the chance of catastrophe but not the chance of electoral interference; i.e. it's meant to model the result of a hypothetical fair election held on Nov. 3. 

comment by Liam Donovan (liam-donovan-1) · 2020-10-24T20:05:49.915Z · LW(p) · GW(p)

There's still plenty of proxy markets on PredictIt where you can bet on Biden, like "will the winner of the {MA,SC,TX} Democratic primary win the election?" or "Will the next VP be a woman?". 

comment by Annapurna (jorge-velez) · 2020-10-19T02:59:39.704Z · LW(p) · GW(p)

I just did a quick model of the 2016 elections, Clinton's FiveThiryEight prediction model probability of winning versus Clinton's percentage chance of winning based on betting odds for a range of dates. Here are the results:

August 18, 2016

Betting Market odds of winning: 81.82%

538 Prediction model odds of winning: 86.40%

September 26, 2016*

Betting Market odds of winning: 68.25%

538 Prediction model odds of winning: 54.80%

*This was debate night that year

October 25, 2016

Betting Market odds of winning: 84.62%

538 Prediction model odds of winning: 85.00%

November 1, 2016

Betting Market odds of winning: 73.33%

538 Prediction model odds of winning: 71.20%

comment by philh · 2020-10-20T16:10:56.725Z · LW(p) · GW(p)

What I cannot explain, at all, is how this can be true if on June 20, three months ago, the market was 63-39, and now it’s 65-40, all but unchanged. Trump improved a bit, then got worse again, with Biden’s low being at 55.

Note that on June 20, the 538 model had Trump at 22% to win (it began at 30% on June 1). The progression he offers makes sense and is consistent. The market’s doesn’t, and isn’t.

Just as a note, three months ago was July 20th, and that seems to be the date you used for the market. (The market was more like 55-45 on June 20th.) The 538 forecast was indeed 22% for Trump on June 20th, and had actually gone up to 25% by July 20th, but I don't think that changes much.

comment by bpunk · 2020-10-19T12:05:36.661Z · LW(p) · GW(p)

I started to sign up for Predictit and there was a note that any money deposited is charged a 5% withdrawal fee. Maybe that effects the odds since people who would be placing accurate bets factor the withdrawal fee into their potential winnings. 

comment by Oscar_Cunningham · 2020-10-19T06:19:16.587Z · LW(p) · GW(p)

Maybe there's just some new information in Trump's favour that you don't know about yet?

comment by siclabomines · 2020-10-19T16:43:49.853Z · LW(p) · GW(p)

I doubt that kind of hidden information can affect PredictIt betting odds as it limits the amount each person can bet. 

 

There is bias or Zvi is reaching wrong conclusions with the same info.

comment by philh · 2020-10-19T11:54:36.091Z · LW(p) · GW(p)

That might explain a recent sudden divergence. I don't think it explains the trend Zvi describes in the post.

comment by deluks917 · 2020-10-19T01:25:35.024Z · LW(p) · GW(p)

Do you know if mybookie is legit. They are offering a ridiculous 1K signup bonus if you use crypto (you have to bet about 10K as I understand it to get the money out). The odds are worse than offshore sites but that is a lot of free value on top of an amazing bet.

The site is way crappier if you are not using BTC for various reasons.

comment by Liam Donovan (liam-donovan-1) · 2020-10-19T03:05:11.950Z · LW(p) · GW(p)

They are not legit at all; use bookmaker.eu instead. (also every bookie has insane rollover requirements for signup bonuses)

comment by Zvi · 2020-10-19T22:24:46.160Z · LW(p) · GW(p)

Indeed do not accept sign up bonuses unless you know how to roll them over reasonably (if size is big enough I can help with that).

comment by deluks917 · 2020-10-19T01:22:31.203Z · LW(p) · GW(p)

Obviously, we agree on the fundamentals. But what exactly is the argument for waiting? Maybe polls will get closer?

comment by Annapurna (jorge-velez) · 2020-10-19T01:28:26.481Z · LW(p) · GW(p)

I believe Zvi is arguing that since the betting market hasn't moved much, despite the perdiction odds of Trump winning deteriorating, there is no reason to believe the betting market will make any moves the closer we get to election. The closer we get to election night, assuming the discrepancy stays constant, the bigger the value on the Biden winning / Trump losing bet. 

comment by deluks917 · 2020-10-19T01:48:05.215Z · LW(p) · GW(p)

Its only 16 days from the election. I am not so worried about the time value of money.

comment by philh · 2020-10-19T12:02:57.881Z · LW(p) · GW(p)

Super not an expert, saying it loud so I can be corrected if wrong:

I don't think time value of money is the main thing here. The observed pattern seems to be that as the election draws closer, people get more information but the market stubbornly refuses to do so. If that pattern continues, then people get more edge as time goes on, meaning future bets will be more advantageous than current bets.

If your strategy is something like "put $100 on Biden as long as I think his odds are more than 5% better than the market thinks" this might not make much difference; waiting only helps in case Biden's odds-according-to-you suddenly drop a lot. But if you're going to bet different amounts depending on the gap, then waiting also helps in case Biden's odds-according-to-you drop a little. (I think if they go up, you can just put more money in, so waiting hasn't gained you anything. But you have to have some probability that they drop.)

comment by ike · 2020-10-19T22:49:06.111Z · LW(p) · GW(p)

If the odds according to you drop, you can also just sell your position.

comment by crl826 · 2020-10-19T00:32:29.517Z · LW(p) · GW(p)

Since the outcome isn't known, how can you be so sure that it is wrong either way?

How is this different than saying a particular stock price is "wrong"? 

comment by arxhy · 2020-10-19T02:38:59.131Z · LW(p) · GW(p)

"Wrong" as in "less likely to match reality." Not very much is certain, but that doesn't mean we are forbidden from talking about certainty.

comment by crl826 · 2020-10-22T02:25:16.000Z · LW(p) · GW(p)

Sure, but the whole point of prediction markets is to capture knowledge from individuals that wouldn't otherwise be available.

If everyone just bets what the polls say, then why have the market at all?

The response could just as easily be "why are the polls wrong?"

comment by orthonormal · 2020-10-25T20:37:39.147Z · LW(p) · GW(p)

To steelman the odds' consistency (though I agree with you that the market isn't really reflecting careful thinking from enough people), Biden is farther ahead in the 538 projection now than he was before, but on the other hand, Trump has completely gotten away with refusing to commit to a peaceful transfer of power. Even if that's not the most surprising thing in the world (how far indeed we have fallen), it wasn't at 100% two months ago.

comment by philh · 2020-10-25T21:47:59.654Z · LW(p) · GW(p)

I take it you mean "people might be betting on the possibility that Trump wins the election, as forecasts predict, but remains president by refusing to concede"?

Betfair's fine print excludes that possibility from the market:

This market will be settled according to the candidate that has the most projected Electoral College votes won at the 2020 presidential election. Any subsequent events such as a ‘faithless elector’ will have no effect on the settlement of this market. In the event that no Presidential candidate receives a majority of the projected Electoral College votes, this market will be settled on the person chosen as President in accordance with the procedures set out by the Twelfth Amendment to the United States Constitution.

I don't have PredictIt's fine print in front of me, but IIRC it's similar but less explicit.

comment by shminux · 2020-10-19T01:03:41.615Z · LW(p) · GW(p)

Something is only a sure bet if there is a way to Dutch-book it. Not clear to me if any of the bets you suggest work like that.