Prediction markets match people who place wagers backing events to happen with people who wager that the event will not happen. Promoted in popular books by Surowiecki (2004) and Sunstein (2006), these markets have long been considered useful for their ability to summarise the public’s information, but their development was limited by regulatory restrictions in the US. A 2008 article by 22 high-profile social scientists, including four economics Nobel prize winners (Arrow et al. 2008), recommended easing these restrictions and argued for “the promise of prediction markets”.

While prediction markets like PredictIt or the Iowa Electronic Markets have been around for many years, the number of events you could wager on and the potential stake size both increased dramatically in the early 2020s when two new platforms, Kalshi and Polymarket, entered the market. Kalshi was approved to operate in the US as a designated contract market (DCM) in 2020 by the Commodity Futures Trading Commission (CFTC), free from the strict limits on stake sizes that were imposed on previous legal platforms. Polymarket, by contrast, is a crypto-based market, and in July 2025 a US arm of its business, Polymarket US, was designated as a CFTC-regulated DCM.

Kalshi is a quote-driven market where makers can post offers into an orderbook and takers pick the best offer. These offers range from 1 cents to 99 cents. For example, if someone buys a 30 cent contract backing an event to happen (a ‘Yes contract’), their 30 cent and the other side’s 70 cent payment for their ‘No contract’ are held by Kalshi, with $1 awarded to whichever side’s prediction turns out to be correct.

Kalshi’s prices are often interpreted as the probability of an event occurring. In a recent paper (Bürgi et al. 2025), we study the accuracy of Kalshi’s prices as predictions and assess biases.

Accuracy of predictions

One key motivation of supporting prediction markets is their alleged accuracy relative to surveys. This is due to participants having a stake in the outcome. One anecdotal example is the 2024 US presidential election, where prediction markets correctly predicted a Trump win while polls suggested a toss-up.  In our data, we find that contract prices tend to broadly reflect win percentages, meaning a 50 cent contract wins around 50% of the time. In addition, contracts tend to become more accurate as the contract expiry date is approached.

Favourite-longshot bias and the return on contracts

We analyse over 300,000 contracts and their outcomes, obtained from Kalshi’s application programming interface (API). While contract prices broadly reflect their probability of winning, we find that cheap contracts win less often than their price suggests and more expensive contracts win slightly more often. For example, a 10 cent contract will win less often than 10% of the time (see Figure 1).

Figure 1 Win percentages sorted by price

This is known as the favourite-longshot bias. This bias has commonly been reported for sports betting markets (Ottaviani and Sørensen 2008) but has not generally been a found in previous studies of older prediction markets. We find strong evidence for this bias across many different sub-samples of the data – for example, across contracts on politics, on entertainment, and on economic data releases.  The pattern is also present across different sub-samples for total market trading volume and transaction size.

The implication of the bias for participants is that those who purchase cheap contracts have highly negative returns on their investments, while those who buy expensive contracts get small positive returns. For example, if a 5 cent contract only wins 2% of the time, it loses 60% of the money invested. The 95 cent contract that is on the other side wins 98% of the time, which gives its investors a small positive profit rate. The result is that, even though Kalshi’s market involves zero-sum profits (prior to Kalshi’s fees), the average pre-fee return on a contract is still minus 20%.

Figure 2 Post-fee return across price range sub-samples for 300,000 contracts

Returns for makers and takers

The trade-level data provided by Kalshi show which side of each trade was taken by the maker and which by the taker. There is a striking difference between the returns earned by makers and takers.

The evidence shows a clear favourite-longshot bias pattern for both sides, but the pattern is much stronger for takers than for makers, with far bigger losses for takers on their longshot contracts. Takers lose almost 32% on average, while makers have an average loss of about 10%.

A matching model to explain the results

We explain our results using a model based on Whelan (2025). In the model, people with differing beliefs select into taking different positions on the market, choosing to be either a maker or taker, factoring in that only a certain fraction of offers end up being matched. We find that the model explains the data on Kalshi contract prices well, capturing both the better returns for makers and the favourite-longshot bias pattern (see Figure 3).

Two aspects of the model explain the data. The first is takers generally being too optimistic about their contract’s chances of success. They decide to accept the best available offer rather than seek a better price as a maker. They prefer the certainty of a positive subjective return to the uncertain possibility of a contract with a higher return. This demand for certainty ends up coming with a cost. They are affected by a form of the ‘winner’s curse’ in which the successful bidder in an auction pays too much.

Second, we incorporate a distortion such that people tend to over-estimate small probabilities, a well-documented pattern in the literature on probability judgement (Kahneman and Tversky 1979). The calibrated distortion that matches the data is modest. For example, it implies that when the true probability of an event is 20%, the average belief among the public is 23%. This leads both makers and takers to accept unprofitable trades on cheap contracts.

Figure 3 Returns for makers and takers with modest overoptimism

Conclusions

Previous evidence had shown prediction markets could produce highly accurate prices, but historical markets were limited in scope and had very small trading volumes. Kalshi runs a very wide range of markets and has uncapped trading volume – features proponents argued would improve predictive performance. Our results show that, as of yet, this promise has not been fulfilled.

These findings have implications for the use of prediction markets in policy and business contexts. While they are clearly a useful tool for aggregating information, our results suggest that Kalshi’s prices should not be interpreted as unbiased probability estimates.

References

Arrow, K J, R Forsythe, M Gorham et al. (2008), “The Promise of Prediction Markets,” Science 320(5878): 877-878.

Bürgi, C, W Deng and K Whelan (2026), “Makers and Takers: The Economics of the Kalshi Prediction Market”, CEPR Discussion Paper No. 20631.

Kahneman, D and A Tversky (1979), “Prospect Theory: An Analysis of Decision under Risk”, Econometrica 47: 263–291.

Ottaviani, M and P N Sørensen (2008), “The Favorite–Longshot Bias: An Overview of the Main Explanations,” in Handbook of Sports and Lottery Markets, Elsevier.

Sunstein, C (2006), Infotopia: How Many Minds Produce Knowledge, Oxford University Press.

Surowiecki, J (2004), The Wisdom of Crowds, Abacus.

Whelan, K (2025), “Agreeing to Disagree: The Economics of Betting Exchanges”, CEPR Discussion Paper No. 20633.