Where Prediction Markets Break Down

TL;DR: if you expect an embryonic prediction market to tell you the result of a highly-politicized event when there are perverse incentives at play, and a significant disparity between amount-at-risk and expected outcome, you are going to have a bad time.

I suspect the 2024 US Election will be a sensational case study of when prediction markets break down.

For starters.. in theory the “money at risk” factor makes it more accurate than polls, but that only matters inasmuch as the bettor caring about the money they are throwing at it.

Therefore, anyone with deep enough pockets compared to the current market open interest could make a bet that skews the book, if they decide it is valuable for them to do so. Say… they consider headlines of “X ahead by Y points” as having publicity value.

In that scenario, anyone pouring money into Polymarket could easily discount it as dollar-cost averaging into their preferred outcome as a marketing expense.

This means those markets are taking “rational player” to mean solely “rational as per the payout of this specific market”, when there might be other factors at play when someone lays down a bet, because they expect an out-of-band outcome (something that results because of their bet, but that isn’t a direct Polymarket payout).

Then there is the fact that Polymarket neither caps the amount per bettor, nor can limit bets to people who are likely to affect the actual outcome (eg. I could bet whatever I want to on the US election result, if the odds are favorable, while not being an actual voter).

If you do look at those that either cap the amount per bettor, or do some KYC, they present a very different US election outcome from Polymarket’s currently results.

Yes, these are still susceptible to Sybil attacks, but they raise the bar; and no, I am not arguing for “KYC all the things!” - I am pointing out that different systems will have different trade-offs, and will attract different players.

It’s also not meant as a knock against prediction markets. I expect that, once they get large enough, prediction markets will be an outside context problem for traditional polling approaches.

Published: 2024-10-16