Free Money at PredictIt: 2020 General Election

post by Zvi · 2020-09-14T14:40:01.941Z · LW · GW · 3 comments

Contents

  Presidential Popular Vote Margin of Victory
  Electoral College Margin of Victory
  Timing of the Election Call
  What are some other juicy targets?
None
3 comments

Previously: Free Money at PredictIt?

It’s time for another look at PredictIt. Is there free money? What are our best options for free money? 

The short answer is that there is free money if and only if you have available capital at PredictIt. 

There is no free money if your plan is to deposit to make the wager and then withdraw. That hits you with a 5% withdrawal fee, wiping out your profits. 

Let’s look at the major markets first, then scour for minor ones.

As with the last such post, despite the fact that we can’t discuss these prices without discussing the potential for a stolen election, let’s be clear: No advocacy for or against any candidate or party in the comments. Any such comments will be deleted reign-of-terror style. That’s not what this is about.

General Election 

Prices are where you would sell if you traded right away by hitting the bid.

Joe Biden 58


Donald Trump 44

Kamala Harris 3

Hillary Clinton 2

Mike Pence 1

That adds up to 108. You pay 10% on winnings. If you take the relative prices here at face value, you’d pay roughly 5.9 cents in fees, leaving a profit of 2.1 cents without need to tie up capital. 

You also get a freeroll to win all bets if none of those five win. That’s probably under a 1% shot but every little bit helps. It’s good to win the weird outcomes, and win them big. Note that many people are betting on candidates, including at other sites, and paying >100% combined for Biden and Trump. Not only does the house always win if they choose to balance their books, the house can sweep.

The labor and cognitive costs of doing this arb if you’re not already set up aren’t worth it as such on their own, but it’s good to note this is there. If you derive satisfaction from taking free money, I approve.

I’m not taking this because I already took it during the primary, and thus can’t take it again. 

What do I think of the baseline claim here, that considered as a two-way race Biden is 57% to win, or BetFair’s 54% for that same question? Note that the secondary candidates favor the Democrats in both cases, so the two-way races are more like 59% and 55% respectively for the Democratic side.

That depends on what exactly you mean by ‘win the election.’ The rules say ‘the winner of the presidential election’ but that really doesn’t clear it up this year. Whereas the BetFair rules say ‘next president.’ Which if we interpret literally makes Mike Pence at 175:1 a screaming buy! If Trump dies in office or resigns or otherwise leaves before ending his term, then Pence is the ‘next president’ without winning the election, and that’s definitely a >1% chance. It also makes it even better to sell the Democratic candidates, since they can win the election and lose the wager. Always read the rules carefully!

The real question for PredictIt is what happens in a disputed election where both Trump and Biden claim victory. Is it who ends up serving the term as president? Is it something else? I’d want to know.

The only way the odds here make sense is putting a substantial chance on an outright stolen or fraudulent election. That’s not electoral college versus popular vote split. That’s not ordinary standard vote suppression. Nor is vigorous litigation of close elections enough either. This would need to be Stalin-level asking of who counts the votes. Hacking voting machines, destroying or ignoring uncounted ballots, declaring victory in spite of the vote count and sending your own electors and stuff like that. 

It’s hard to properly price that. Without it, the 538 model’s current 76% for Biden doesn’t sound unreasonable to me. Is there more or less than a one in six chance of Trump successfully outright stealing the election? 

They don’t offer a market on that one. It’s the missing market, the most interesting one of all in many ways. I am a skeptic that the chances are anywhere near that high. Don’t get me wrong. The chances are still way way way too high, especially since there’s also the scenarios where he tries and fails and that’s no picnic either. I’m taking these scenarios seriously enough that I’m staying in Warwick with fully stocked up emergency supplies on election night, and only after a concession planning my return to New York City. 

It makes sense to take the arbitrage here rather than back a side. If you want to back a side, you can do so elsewhere at better prices.

Compare to the pure two-way market, which is 61-43, for the same implied price of 59% for democrats, but without an opportunity for Free Money. This is another way of illustrating that the true free money here is on the secondary candidates, especially Clinton. Selling Biden and Trump is only a way to then free up capital.

Again, prices are where you can sell these on demand.

Dem 10.5%+ 15

Dem 9-10.5% 10

Dem 7.5-9% 12

Dem 6-7.5% 12

Dem 4.5-6% 12

Dem 3-4.5% 11

Dem 1.5-3% 9

Dem 0-1.5% 7

GOP 0-1.5% 7

GOP 1.5-3% 4

GOP 3-4.5% 3

GOP 4.5-6% 2

GOP 6-7.5% 2

GOP 7.5-9% 1

GOP 9-10.5% 1

GOP 10.5%+ 3

That adds up to 111, so you can definitely take some free money if you’re interested, or you can try to be selective. Note that this curve centers nicely around Biden by about 6.5%. 

This roughly agrees with Trump’s 20% in 2020 Predictions | Will Trump win the popular vote in 2020?

If you look at the distribution, Democrats by 10.5%+ seems cheap. Following the slope on the other side, it’s clear this should be priced much higher. Presumably people think this is because there’s no way, there’s too much partisanship. I can’t agree. The bottom could easily fall out in any number of ways. That doesn’t mean that 15% is cheap, but I’m skeptical that this general level of variance and this median are right, and there’s only a 15% chance of a blowout. 538 agrees, and sees a 30% chance that Biden wins by 10 or more, whereas here you can get 9%+ for only 25%. 

Note also that a true Biden blowout isn’t that likely to be ‘brought back’ by fraud. If Biden wins by this much, the election can’t be stolen, so doing brazen things to reduce the apparent margin also seems not worthwhile. 

The GOP by 10.5%+ also sticks out. Why is that so big? To me this one makes relative sense, again on the outright fraud principle. There are two ways to steal an election. You can pull a (alleged) 1960, and not steal one more vote than you have to. Or you can pull a Stalin or Hussein, and claim a huge landslide not caring that no one believes you. I wouldn’t go out of my way to sell that possibility. If anything, the GOP by 6-9% seems decidedly less likely than 10.5%+, to me. Something very strange would have to happen to get things to move that much, so at that point, the broader range is a lot more attractive to me.

Thus, given how crazy this market could get later, and given I already tied up my funds, I’m going to take the arbitrage here, at least not yet. I might take it later, but for now I want to reserve the right to make a better play. 

Next question is, what does this market really say about Biden’s chances?

In a fair election, 538 thinks Biden has an 11% chance of winning the popular vote but losing the electoral college. If Biden was worse off generically, that number would only go slightly higher, say 13%. However, if the GOP decides to steal only in selected states, and only a 1960-style amount, or do things like halt counting before mail ballots are counted, that number could go much higher.  

This chart has a 55% chance (after normalization) for Biden to win by 4.5% or more, and an 18% chance of winning by 1.5% to 4.5%. I think it’s reasonable to say he wins almost all the times in the first bucket, and about half the time in the second bucket, so it’s implying 64% or so chance of victory. That’s substantially different from the presidential market, so either the two have diverged, or a bunch of probability mass is in ‘Trump brazenly steals only the tipping point states,’ or the evaluation here by the Federal Election Commission could go to Biden while Trump ‘wins the election’ anyway through theft. 

There are a lot of ways to play this.

Electoral College Margin of Victory

GOP 280+ 4

GOP 210-279 3

GOP 150-209 5

GOP 100-149 9

GOP 60-99 10

GOP 30-59 7

GOP 10-29 5

GOP 0-9 4

Dems 1-9 2

Dems 10-29 4

Dems 30-59 6

Dems 60-99 9

Dems 100-149 16

Dems 150-209 12

Dems 210-279 9

Dems 280+ 7

That adds to 110, which is again an opportunity for free money without tying up capital. It’s also, again, another chance to get money down on a side while expressing an additional opinion. The reason not to take this is in case you want to save it for later.

The Democratic wins add to 63, the Republican wins to 47. Thus, this predicts a 57-43 distribution, slightly better for the Republican side. 

The 538 model’s best prediction is Democrats by 122, which is right in the middle of the highest probability group, but not at the median of the market distribution.

Note that these groupings are not the same size. If we normalize for that, we see a very broad distribution of outcomes all seeming similarly likely. 

Timing of the Election Call

Hats off to PredictIt for the definition. If CNN and Fox News both call the election for the same person, that’s a pretty good proxy for it being over. That doesn’t mean Trump might not dispute it anyway (or even Biden) but it’s as good a definition as was available.

As always, I only include prices you can sell at. Assume it costs one extra to buy instead of sell.

November 3 21

November 4 31

November 5 10

November 6-7 7

November 8-9 5

November 10-16 7

November 17-23 5

November 24-30 5

December 1-14 7

After December 14 11

That adds to 109. Again, enjoy your cash.

That is one scary chart. There’s less than a 50% chance that the election will be resolved even a full day after the election. There’s a 10+% chance it won’t be solved over a month later, which likely means it’s going to congress or worse – and again, Trump could well fight on even after Fox News gives up. 

I don’t know that much about when various votes are likely to come in from mail ballots, but neither does anyone else. The one seeming ‘error’ here is that November 8-9 is 5% for a 2-day window, then you have a 7% chance for a 7-day window, then the next 7-day windows are 5% each. Unless there’s a specific reason, that’s a very weird distribution. 

A good question to ask is, how does this line up with the margin of victory?

If the margin of victory is >6% on either side, I’d assume the networks would be able to make the call on November 3-4. That’s a 55% chance according to that market. And there’s a decent chance they can make the call with a smaller margin. If the GOP is legitimately winning the popular vote outright but less than 6%, that’s arguably a >10% chance, and realistically Biden almost certainly is toasty enough that CNN should give up on November 4, although they plausibly hold out past midnight on election night. 

Consider that this market has only a 19% normalized chance (buy costs 22%) that there’s a call on election night, but there’s a higher chance than that for Biden to win by more than 9%. If he does that, I have a hard time believing there’s no call by midnight.

Thus, my gut reaction is that this market is not totally crazy, but it’s somewhat too skeptical of resolving things in the first two days. If anything, I’d think it’s too optimistic then about resolving things quickly or on November 5, provided they’re still in the air on the 4th, but there’s a reasonable argument for ‘once Biden takes the vote lead in the tipping point state it is over, but Fox News won’t call the race until he does take that lead, and it takes a few days to count enough mail in ballots.’ But 10% for that one day, or about 20% of the time given we get that far, does seem like a lot. 

Thus, if I was thinking about what to sell and what to keep, my inclination would be to keep the nightmare scenarios and the quick resolutions, and sell the stuff in between. 

What are some other juicy targets?

Will Biden drop out by 11/1?

You can buy the ‘No’ at 90%. That’s crazy. He’s not going anywhere. 

One way to know it’s crazy is that time is going by, and the number isn’t being discounted. When I got into the No at 10% a week ago, I got the same price. In late August, the price was the same. On July 2 it was 9%! 

I made a good trade. But there was a much better trade available. Which was to buy Biden dropping out in early July, then sell it now for the same price or higher. 

Between now and then, Biden has looked relatively healthy, he has established and maintained a solid poling lead, and has had no scandals that could possibly push him towards dropping out. There’s no way his chances of dropping are more than half what they were in early July, let along higher. 

2020 Election Predictions | State with smallest MOV in 2020?

I won’t list the odds except to note that they add to 111, and you get some bonus states as a freeroll. Nothing in the relative rankings looks crazy to me. My guess is the value is on selling the states trading high and the ones trading super low, and keeping the ones in the middle, if you don’t want to just take free money.

Will the winner of the popular vote also win the Electoral College?

This market has a whopping 29% chance that the winner of the electoral college didn’t win the popular vote. That would add up to a 49% chance for Trump to win. This seems like the best way to bet on Biden in a two-way. 

https://www.predictit.org/markets/detail/4614/Will-Hillary-Clinton-run-for-president-in-2020

No. She won’t. Yet somehow they won’t give up thinking she will, and keep not discounting this much for all the time that goes by. I bet the no on this a long, long time ago, early in the primary, and I’m only 1% to the good right now. There’s a lesson in that. You can still buy the No at 95%, and it’s cheap at 99% if you don’t care about tying up capital.

Will Nancy Pelosi become Acting US President on January 20?

This is people living out their fantasies. I’m not sure if it’s outright impossible, but I do know this is not an 8% chance. Wow.

2020 Election Predictions | Will a woman be president?

Think Kamala Harris at 3% is overpriced? Here you can sell her at 6%. Technically you also have to sell all other women, but who is second in probability? Jo Jorgensen? 

Will Michelle Obama run for president in 2020?

Get it through your thick skulls, everyone, that Michelle Obama hates politics and will never run. Yes, she would win, so I get why you dream, and yes everyone says they won’t run until they do, but no she won’t ever run. You can only get 3% on this at this point, so probably not worth the capital. Again, people dreaming. Still, it’s more plausible than Hillary Clinton.

2020 Election Predictions | Will Donald Trump drop out before November 1?

It’s 4%, same as a few weeks ago. Not budging. Not worth touching, but another dumb line.

2020 Election Predictions | Will John Kasich run in 2020?

Still can sell 1% if you’d like! Let no one say interest on your money is dead. Same with Paul Ryan.

2020 Election Predictions | 2020 Iowa Winner Elected President

Another 1% still available, because people want to free up capital and/or forget that this only pays out if the democratic winner wins, so it’s actually 0% to be Yes.

There are also a bunch of states available. I’m not going to go into them all, but there are some good overpriced long shots to sell if you’d like.

Impeachment Predictions | Will the Senate convict Donald Trump in his first term?

4% is still available. That’s on top of 11% for him to resign  in his first term. Much better than his 13% to not complete his first term, since that’s only two of the ways he can leave office. 

There are a lot of Senate races and House races. I’m not up on the situations enough to know the right stuff, and I want to get this out right after writing so the prices don’t change. 

Similarly, I see a bunch of value in the US Government section and the World section, but no free money. 

Adding it all up, it looks like there are a decent number of full-arbitrage markets, plus a large number of ways to earn decent capital returns by betting on a No. The issue, as always, is that either you withdraw the money after the election, or you have to keep it there. I’ve been rolling it, leaving the money on the site, but I only have a few thousand. That leaves me short of being able to sell everything I want to sell at a moment like this. But it was never about actually maximizing every dollar as such. It’s about using the whole thing as a learning exercise. 

Have fun, everyone, but don’t take these market prices too seriously. 

3 comments

Comments sorted by top scores.

comment by Pattern · 2020-09-15T16:32:45.380Z · LW(p) · GW(p)

Thus, given how crazy this market could get later, and given I already tied up my funds, I’m [not] going to take the arbitrage here, at least not yet. I might take it later, but for now I want to reserve the right to make a better play.

 

Thanks for writing this. How prediction markets work in practice is interesting.

comment by philh · 2020-09-14T15:56:02.088Z · LW(p) · GW(p)

Whereas the BetFair rules say ‘next president.’

Note that betfair has both an exchange and a sportsbook, and you linked the sportsbook. Their exchange market for this question is also called "next president", but does have more detailed rules. In particular:

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'm surprised there's nothing like that (that I can see) on the sportsbook.

Replies from: Zvi
comment by Zvi · 2020-09-14T16:09:49.752Z · LW(p) · GW(p)

Yeah, that's still somewhat confusing, but makes it clear that they mean the election. I'll edit to reflect that.

(I've now edited the original, mods please update - I gotta go and now is not a good time for me to learn how to properly paste the HTML or what not)