How Prediction Markets Actually Work, I Think

There has been a lot of talk about prediction markets, like Polymarket, heading into the U.S. election.

I myself am only learning much about them now. But I'll address what they are, how they work, and what so many people are getting wrong about them by pulling from these papers:

https://irep.ntu.ac.uk/id/eprint/35265/1/12800_Vaughan-Williams.pdf

https://www.biz.uiowa.edu/faculty/trietz/papers/forecasting.pdf

How they work in practice.

The first paper defines prediction markets as "judgement-based markets created for the purpose of making predictions."

Users buy "shares" representing different possible outcomes of an event, like the presidential election.

Typically users buy the share for some amount between $0 and $1. Winning shares payout $1, while losing shares become worthless. On Polymarket, users can currently buy a share of Trump winning for 60¢ or Harris for 40¢. Both pay out $1 if the user wins.

The theory is that this implies Trump has a 60% chance of winning, and Harris 40%.

But why?

The #1 mistake is believing the market implies Trump will receive 60% of the votes, and Harris 40%.

This logic is being employed by those claiming "the election is over" and that the market is telling us Trump is clearly going to win. This person believes Trump is receiving 60% of the votes, thus easily winning the election.

It may seem simple, but this mistake is pervasive on Twitter right now.

What this actually says is 60% of the money being bet on the outcome is being bet on Trump. Notice it's not 60% of the people betting, but 60% of the money that believes in this outcome.

Unlike the polls, participants have a financial stake in making accurate predictions. The theory is that the odds should correlate to the degree of confidence the bettors have in the particular outcome occurring.

The core of the theory, in my opinion, is that it pays people with insider information to reveal what they know.

The term "insider info" is often maligned. Here it simply means that someone has specific reason (information) to believe one candidate is more likely to win.

This market allows that individual to profit by betting on their knowledge. Revealing to the rest of us what information others have and the degree of confidence to which they believe it's correct.

A large bet on a candidate signals that someone has a high degree of confidence that candidate will win. There must be information available to them that suggests that's the case.

The market is meant to "aggregate the relevant information from different participants to help forecast future events."

But is this going to prove accurate?

First, understand that the market cannot tell us who is going to win, only what probability the candidates have.

Neither outcome will confirm or invalidate whether Polymarket is predictive, because it only tells us that one candidate currently has a 60% chance of winning. Events that have a 60% likelihood of occurring should occur...60% of the time.

We can only see, over time, if a specific market is proving to be predictive to the expected degree.

The accuracy of its predictions hinge on the size and representativeness of the Polymarket betting pool.

The main objection has been that Polymarket is not legally available to U.S. citizens.

Yes, you read that correctly. You cannot legally access Polymarket's betting functionality if you are an American. Which, at first glance, might seem to destroy its usefulness.

But does it really?

Remember that bettors are not revealing who they voted for, but who they believe will win.

The market dynamics remain the same. Non-American users would still bet according to the beliefs and information they have. If one believes Harris has a greater than 40% chance of winning, the expected outcome of buying shares at 40¢ is profitable regardless of where that user is from.

Those who disregard the results entirely are making the same mistake from above (that the bets are "votes" and therefore useless since they're from non-Americans).

But, these market participants may have less insider information than Americans would. In other words, many of the potential users with the most accurate info are barred from placing a bet.

Yes, Polymarket's predictive accuracy is likely diminished by the exclusion of U.S. participants. The market is missing information from those most likely to have it.

What about market manipulation?

This is where prediction markets excel.

Polls lack a financial penalty for manipulation, prediction markets have it built right in.

In order to manipulate the market, you have to keep betting on an outcome beyond it's likelihood of occurring. To make a candidate look more likely to win than he/she is, you must bet even when the odds result in a negative expected outcome for you.

Unless you succeed in swinging the election, you will lose money.

I'm skeptical that any current prediction market could swing the election even if it was manipulated.

This would require a participant to place large bets in an attempt to sway public opinion. But it would also require this to actually change people's votes or likelihood to vote. Very unlikely in my opinion.

If a manipulator pushes the market results in one direction, it becomes more lucrative to bet against them. The market had achieved equilibrium, but the prediction site is now paying more if the other candidate wins then the market needed to take the bet in the first place.

If it's true that the Trump/Harris likelihood of winning is 60/40, then any price less than 60¢ or 40¢ is profitable to buy at. The manipulator makes the price of the opposite bet more attractive for users to take.

This may be the killer reason these markets prove predictive on a long enough timeframe.

Will prediction markets prove right in the end?

The second paper claims evidence that "prediction markets are considerably more accurate long-run forecasting tools than polls across elections."

But, will Polymarket prove to be predictive in this election?

We don't know. We can't know if the sample size is large and representative enough (and we still won't know after the election).

But, over time, prediction markets might prove to be incredibly effective.