Traders confused: Prediction markets on Biden's Odds Diverge

If prediction markets are so awesome, how can platforms disagree significantly?

An interesting example of divergence is Biden’s chance to be the Democrat’s nominee. Joshua started a great discussion:

Manifold thinks Biden is 90% likely to be the Democratic Party nominee. Metaculus says it's 94%.

Meanwhile, Polymarket thinks this is only 78% likely. Other real-money markets are similarly low, with ElectionBettingOdds's average putting it at 71.4%.

In the 538 Politics Podcast this week, the 538 crew mostly agreed with current prediction market prices on other 2024 questions but universally agreed that Biden's chances were much higher than 78%. Nathaniel Rakich, who I might trust more than anyone else on this, said it should be 95%.

He “misused” the multiple choice market feature, so everybody could collaboratively rank the opinions and in the end everybody got their money back. Easy to manipulate, but for the sake of a discussion there are worse methods.

The top answers.

Let’s take a look at the top three, since number 4 is significantly lower (67% vs 56%).

Probability distortion (from prospect theory): people risk seek when faced with low prob. gains and high prob. losses, and are risk averse in the opposite circumstance. Therefore, low prob. events are systemically overvalued by money sites.

An argument that Polymarket is wrong: People overestimate the low probabilities of other candidates like Newsom (10% vs 2%) and Harris (16% vs 2%). Likewise, people underestimate high probabilities (80% vs 94%). The well-known book Thinking, Fast and Slow made Prospect Theory popular. While the replication crisis criticized parts of the book, prospect theory replicated.

The argument doesn’t explain why the real-money markets are different. It just says traders are worse there for some sciency reason.

You don't get loans on Polymarket!

An argument that Polymarket is wrong: Without loans, traders might have more profitable markets available and bet too little on this market.

This market resolves in August. A return of investment of 10% in 8 months is roughly the same as an all-world index ETF. That makes it nice for diversification but not a great opportunity. For contrast, on Manifold you get your money loaned back in small amounts and Metaculus has no money constraints at all.

This is my favorite hypothesis.

I can imagine a workaround: A prediction market platform puts the trader’s money into an all-world index ETF and then they trade (micro-) shares on predictions. Everybody gets the market returns plus prediction returns on top.

Real-money prediction markets better incentivize accurate reporting of beliefs than Manifold. (They historically outperformed Manifold in liquid markets where there was disagreement.)

An argument that Polymarket is right: With real money at stake, traders should be more careful, thus more accurate. With over $3Mio traded on Polymarket, there is quite some liquidity involved.

In the comments, people cited Taylor Swift as Person of the Year and the LK-99 room-temperature superconductor cases. I can’t see a clear difference there.

Calibration City does not show a clear advantage for anyone. You can make anyone win with slight tweaking of filter criterias.

It is hard to compare platforms. One challenge is that comparing scores only makes sense if the same predictions are used. If one platform works with harder questions, it would not be a fair comparison. Maybe after the US election, someone makes an in depth evaluation, because there could be a good set of equivalent questions about it.

Each prediction market is arbitraging the different answers to add up to 100, so positive bets on random candidates are inadvertently negative bets on Biden

This is the conclusion of the market creator, who probably read all comments and arguments more closely than anybody else, and an argument that Polymarket is wrong: Many people bet on a candidate to signal support irrespective of the chances.

It's not shocking to me that there are a handful of people out there who would each bet their one long-shot candidate up to 5% on the off-chance that Biden dies. But if four different people are willing to bet up four different long-shots to 5%, suddenly that's 20% of the probability space and Biden ends up at ~80% even if those four different people all think he should be at 90%+.

This argument is very close to the first one (probability distribution) but the root cause is different. Supporting your desired candidate is fine, though I doubt that betting them up on a prediction market is an effective method.

---

Probably until next week, my divergent readers! 😊