Favorite-Longshot Bias

How high-probability contracts are systematically underpriced in prediction markets

What is the favorite-longshot bias?

The favorite-longshot bias (FLB) is one of the most widely documented anomalies in betting and prediction markets. It describes the empirical finding that high-probability outcomes (favorites) tend to be underpriced, while low-probability outcomes (longshots) tend to be overpriced.

In practical terms: a contract priced at 90 cents — implying a 90% probability — may actually win 92-93% of the time. Conversely, a contract priced at 10 cents may win only 7-8% of the time. The mispricing is small on each individual trade, but it is persistent and has been observed across decades of data.

Academic origins

The bias was first documented by R.M. Griffith in 1949, who analyzed odds in American horse racing and found that bettors systematically overbet longshots relative to their actual win rates. Since then, the phenomenon has been replicated across dozens of markets and countries:

• Horse racing — documented in the US, UK, Australia, Hong Kong, and Japan
• Sports betting — observed in football, basketball, and soccer point-spread markets
• Prediction markets — found on platforms like InTrade, Iowa Electronic Markets, and PredictIt
• Lotteries — extreme longshot overpricing in state-run lottery games

Why does it exist?

Economists have proposed several explanations, and the debate remains open:

Risk-love / preference for skewness

Some participants are willing to overpay for longshots because they prefer the "lottery ticket" payoff profile — a small wager with a chance at a large payout. This is consistent with cumulative prospect theory (Kahneman & Tversky, 1979), which predicts overweighting of small probabilities.

Miscalibrated beliefs

Snowberg and Wolfers (2010) argue that the bias is primarily driven by participants who are simply bad at estimating probabilities near the extremes. Their analysis of racetrack data suggests misperception accounts for most of the effect, not risk preferences.

Limited arbitrage

Even if sophisticated traders identify the mispricing, position limits, transaction costs, and capital constraints may prevent them from fully correcting it. On platforms like Kalshi, per-market position limits and exchange fees create friction that can sustain the bias.

What the data shows

The size of the bias varies by market. In horse racing, Thaler and Ziemba (1988) found that favorites (odds-on) had positive expected returns after the track take in many datasets, while extreme longshots (50-to-1 or higher) were heavily overpriced.

In prediction markets specifically, Page and Clemen (2013) examined InTrade and the Iowa Electronic Markets and found that contracts above 80 cents were slightly better calibrated than contracts below 20 cents, but mispricing persisted at the extremes.

The edge is typically small — on the order of 1-3% per trade — which means fees, execution quality, and volume of trades all matter significantly.

Key considerations

• Transaction costs matter — Exchange fees can consume or exceed the edge if not carefully managed. On Kalshi, the fee structure should be factored into any expected-value calculation.
• Not guaranteed on any single trade — The bias is a statistical tendency across large samples. Individual contracts still resolve against the favorite regularly.
• Market-specific — The magnitude of the bias varies across categories. Weather, sports, and political markets may exhibit different levels of mispricing.
• Evolving — As more sophisticated participants enter prediction markets, the bias may shrink. Historical data may not perfectly predict future returns.