Prediction markets rely on efficiency, but efficiency cannot guarantee.
Author: Benjamin Sturisky, Delphi Digital Researcher
Compiled by: Shaofaye123, Foresight News
The structure of prediction markets can play a role. However, they rely on many different components, so they cannot consistently provide accurate probabilities.
These systems relying on complete market efficiency is unrealistic.
In the first article on prediction markets, I briefly introduced how prediction markets can serve as a source of truth under a cloudy market. I also listed three fallacies that hinder specific markets from reaching true probabilities. This second article attempts to delve deeper into these three fallacies: bias bias, hedging bias, and time bias.
Market Efficiency
Market efficiency is crucial to the accuracy of prediction markets, because without efficiency, there will be probability biases.
Here is the purest example of market efficiency:
- Establish a market based on coin flips, with a market maker selling coin flips at 55c odds. The market maker actually gains a 10% edge from each coin flip, as he sells at 0.55 for a 0.5 payoff. In this example, the buyer expects to lose 5 cents per coin flip.
- Another market maker sees the market and wants to participate. He offers a lower price than the other seller, setting the odds at 52.5c. His edge per coin flip is 5%, and the buyer's expected loss per coin flip is 2.5 cents.
- A third market maker joins in, trading at 51c. His edge per coin flip is 2%, and the buyer expects to lose 1 cent per coin flip.
The key is that in an efficient market, profit opportunities will diminish until the risk premium is reached. For coin flips, since the outcome is highly predictable, the risk premium is very low, so the market will be very efficient (+/- ~1 basis point). However, for things with higher outcome uncertainty, like insurance (e.g., a forest fire destroying a community), the risk premium is larger, requiring a greater gap between expected cost and insurance price to ensure profitability for the insurance company.
Bias Bias
If there is no pure market efficiency, the predictions of prediction markets will be biased (usually upward biased).
When people observe the market, they tend to be biased towards outcomes that benefit them. This indirectly prices the probability of that event occurring higher than the actual probability (e.g., Chelsea fans are more likely to bid on Chelsea winning the Champions League than Arsenal fans).
The problem in inefficient markets is that no one is willing to bid the Chelsea price back down to the "true" probability.
I also want to use a real-world example related to everyone's favorite topic: the US presidential election.
Currently, Polymarket's prediction for Trump YES is around 57%, and for Harris YES around 39.5%.
How does this compare to other prediction tools?
- Silver Bulletin: Trump (56.9%) and Harris (42.5%).
- Manifold Markets: Trump (54%) and Harris (43%).
- Metaculus: Trump (55%) and Harris (45%).
- PredictIT: Harris (51%) and Trump (50%).
Polymarket's core user base is crypto-leaning right-wing. This is evident as Polymarket has Trump's winning probability higher than any other prediction tool/market.
Polymarket is the most liquid prediction market globally, with over $460 million in total volume for this election. If any market is efficient, it should be this one. But it is not efficient nonetheless.
If prediction markets rely on efficiency, but cannot revert to true probabilities when biases distort the odds, should they be used as a source of probabilities?
Time Bias
Prediction market efficiency is not as simple as the coin flip scenario above. If someone wants to bring the market back to true probabilities, the edge they gain must be worth it.
If a market is 1% upward biased, but resolves in six months, the person hedging the edge won't bother arbitraging the market back to true probabilities. This is because 1% over six months is equivalent to 2% per year, which is below the risk-free rate.
The only way for such a market to revert to true probabilities is if someone is interested in the opposite direction.
So the market won't reflect efficiency until the bias increases or the resolution time decreases (playing market maker and beating the risk-free rate is +EV).
Hedging Bias
Hedging can distort actual probabilities by pushing the odds higher or lower.
Here's an example of how hedging can manipulate prediction market probabilities:
- A trader buys $1 million worth of SPY EOD call options on the morning of an FOMC meeting.
- The trader believes a rate cut will increase SPY, while no change will decrease SPY. The market currently prices these two outcomes at 50:50.
- Shortly before the decision, the trader gets cold feet and wants to reduce directional risk. He doesn't want to sell the SPY calls because the liquidity is relatively poor (remember, this is a theoretical example).
- To solve this, the trader buys $200,000 worth of the NO on the rate change market, pushing the probability of a rate cut to 48/52.
- If the market consensus is 50:50, but the prediction market is 48/52, market efficiency would require the trader to buy the YES shares until the market reverts to 50:50. But this doesn't always happen.
There are several reasons why this market may not revert to the true 50/50 probability.
The most obvious one is: there may not be traders willing to take on the directional risk of arbitraging the market to gain a tiny edge.
Unlike the infinitely repeatable coin flip, FOMC meetings only occur 12 times per year. This infrequency significantly increases the risk premium, as each event has a major impact.
The EV formula below shows a 48 cent investment has an expected average return of 2 cents.
EV = (.5 * 1) + (.5 * 0) - .48 = 0.02
Given the low frequency of FOMC meetings, we may not find traders willing to take on the directional risk of this position. Additionally, since this market inefficiency is caused by a one-time hedge, this specific market opportunity is unlikely to arise again at the next FOMC meeting. Ignoring external market hedges/uses (which don't always exist), arbitraging this market is effectively like buying a single coin flip at 48 cents.
The second reason is theoretical, highlighting information asymmetry. If the prediction market is used as the sole true source of event probabilities, traders may be unwilling to arbitrage the market, as they don't know if the bidder has information they can't access. They can't know if the bidder is just trying to hedge their SPY calls. This significantly changes the model, as now the traders not only need to be willing to take on directional risk, but also bet that the 52 cent bidder doesn't have asymmetric information.
My Take
I'm quite bullish on prediction markets. However, using them as the sole truth of probabilities is mistaken.
They are excellent at information discovery - I believe prediction markets will become the "go-to" place to view real-time odds on any event. At the same time, I disagree with the view that they are always fully accurate.
For large events, I think adding error bands to predictions would help address biases caused by biases, hedging, or time.
