During this year's U.S. presidential elections, I have repeatedly encountered discussions about prediction markets, where individuals trade on the election's outcome. These discussions are often accompanied by an uncritical argument that suggests prediction markets are better than polls: "Polls just represent what people want the outcome to be, while prediction markets represent what people believe the outcome will be".
I am surprised by this lack of critical analysis.
What are prediction markets? They are markets (such as predictIt.org), where traders buy and sell contracts, such as: "If X wins the election, you receive $1; if X does not win, you receive $0". Once the election results are announced, the value of this contract will either be $0 or $1. Prior to that, the value fluctuates somewhere in between. Since traders invest their own money, there is an expectation that they will be motivated to price these contracts "right".
What does the "right" price tell us? In the best-case scenario, we can hope to derive a probability from the prediction market: what is the likelihood of X winning the election? However, it's important that this probability doesn't tell us about the margin of victory. For example, the probability of winning may be very high, while the actual margin of victory could still be very slim. This characteristic makes it challenging to easily disprove the market's prediction. The market might indicate a 95% chance of X winning, but X could still lose. In such a case, the market may still have been accurate, as there was indeed a 5% chance of X losing.
At what point in time do prediction markets represent the probability? Is it one month, one hour, or one second before voting concludes? Or perhaps after voting has finished? For understandable reasons, prediction markets fluctuate over time, and during a given period, they may forecast one election outcome to be more likely and later predict the exact opposite. One could argue that this behaviour already proves that the market is not not working for the purpose of predicting the outcome.
Going further, the theory that "prediction markets represent what people believe the outcome will be" may actually be quite inaccurate. Instead, these markets might simply represent what people believe others believe will happen next (along the lines of a "Keynesian beauty contest"). A trader might buy at a low price because they anticipate the price will rise, based on their belief that others will buy soon after. They don't have to wait for the election to occur to sell. Such a market price has little to do with the actual election outcome.
For a prediction to have any chance of accuracy, it must be based on facts. What facts does a trader have access to? Primarily their own knowledge and data from polls. They could conduct their own private polls, but given the trading limits applicable on many prediction markets, this would hardly be cost-effective. Ultimately, prediction markets reflect the limited knowledge of traders. More important than that knowledge being limited, it's probably also heavily biased towards the insights of traders. Only a very small fraction of the eligible voting population is trading in prediction markets. So in the best case, at any given point in time: "prediction markets represent what traders believe will happen next".
Other reasons for trading in prediction markets may include hedging against specific election outcomes. In these cases, the trader is not informing the market but rather relying on it as a gauge of risk. In more autocratic situations, one could even imagine a candidate themselves (or their cronies) hedging against the possibility of losing. In this context, "prediction markets represent what people fear they will lose given a certain outcome".
In conclusion, while I find prediction markets to be a fascinating concept, considering the way they are currently implemented, I have significant doubts about their usefulness for gaining an early understanding of election outcomes.