Essays · 15 November 2026 · 7 minute read

The favourite-longshot bias: why cheap prediction market contracts lose money

If you'd bought every Kalshi contract priced at 10¢ or less, you'd have lost over 60% of your money once fees came off, while the ones priced above 70¢ would have made you a small profit. That's the favourite-longshot bias (Americans spell it "favorite-longshot bias"), and racecourses have been showing the same thing since the 1940s, which is that long shots cost too much for how rarely they win and favourites cost a little too little.

What did the Kalshi study find?

Three economists at University College Dublin, Constantin Bürgi, Wanying Deng and Karl Whelan, wanted to know whether Kalshi's prices could be trusted, so they pulled its trades through the API from its launch in 2021 up to April 2025. They kept contracts that had traded at least $1,000 and threw out any whose final spread was over 20¢, which left them with 313,972 prices across 46,282 contracts, and then they did the simple thing that hardly anyone bothers with and counted how often each price actually won.

If the prices were fair, a 5¢ contract would win about 5% of the time, but the cheap contracts won less often than their price said they should and the expensive ones won more often. Above 50¢ the return turned slightly positive, and above 70¢ it stayed positive even after fees (it's small, but it's too steady to be luck). Across every contract the average came to about minus 20% before fees. They also saw the bias shrinking over the years, which you'd expect as bigger traders turn up, though it hadn't gone by the end of their data.

How can a 5¢ contract lose 40% of your money?

The paper uses a neat example that's worth taking a minute to work through, even with no fees at all. Imagine a contract priced at 5¢ that wins 3% of the time, so it pays out a dollar three times in every hundred, and each 5¢ you spend brings back only 3¢ on average, which means two fifths of your money disappears. On the screen those two probabilities look almost identical, since there's only a couple of percentage points between them, but measured against such a small price that couple of points becomes a painful proportion of everything you've paid.

Now flip this round and consider a contract priced at 95¢ that wins 98% of the time, which returns 98¢ on average for your 95¢ and leaves you a little over 3% ahead. The mistake in the price is three percentage points here, which is bigger than before, but it's spread over a price nineteen times larger, so the return barely seems to be affected. That's why small misjudgements about unlikely outcomes are so expensive compared with similar misjudgements about likely ones, and it explains why the losses accumulate at the cheap end.

Do fees make long shots worse?

While the study ran, Kalshi charged whoever took an offer $0.07 × P × (1 − P) per contract, where P is the price in dollars, and the total was rounded up to the next cent. If you measure that fee in cents, the largest is at 50¢, but what you really care about is the fee as a proportion of what you're paying, and dividing through gives 7% × (1 − P), which is largest on the cheapest contracts. We've calculated it for an order of 100 contracts at five different prices, so you can see how quickly the proportion grows as the price falls:

If you look at the top and bottom rows, you'd pay the same 34¢ in fees either way, but in one case it's on $5 and in the other it's on $95, and in our 3% example that fee pushes the loss from 40% to about 44%. Kalshi has charged makers as well since April 2025 and fees now vary from one market to the next, so it's worth checking the one you're on, though any fee shaped like P × (1 − P) still bites hardest at the cheap end once you set it against your stake.

Has the same bias shown up elsewhere?

It has, and for a long time. R. M. Griffith spotted it at American tracks in 1949, and later Erik Snowberg and Justin Wolfers went through every horse start in US racing from 1992 to 2001 and found that if you'd backed every horse at 100/1 or longer you'd have lost about 61%. Picking horses at random lost 23%, while backing the favourite in every race lost just 5.5%, and they found the same pattern in Australia and Britain, where bookmakers set some or all of the odds.

Put that next to Kalshi and it's a close match, because seventy-odd years on and in a very different market, the people buying the longest shots are still losing well over half their money.

Why does it happen?

Snowberg and Wolfers tested two ideas, one being that bettors just like risk and will pay extra for a big payout, and the other that they get small chances wrong, so a 2% chance feels more like 5% to them. Bets on two or three horses at once (exactas, quinellas and trifectas) let them tell the two apart, since each idea prices those bets a bit differently, and getting small chances wrong fitted the prices far better.

The Kalshi paper adds a twist that comes from how the exchange works, where every trade has a maker who posts an offer and a taker who accepts it. Takers tend to hold the strongest views, because they're the ones who'll pay to trade right now, and the authors show that a bit of disagreement between traders plus a slight habit of overrating small chances is enough to produce the whole pattern. Makers did better than takers, though you'll find the bias in both.

How do you use the bias to win?

We'd start with where the money went. If long shots are overpriced on average then the other side of them is cheap on average, and the other side of a 5¢ Yes is a 95¢ No, which is where the study found its profits. In the paper's example a 95¢ contract that wins 98% of the time makes a little over 3%, and while that isn't much on one trade, it adds up across lots of markets and it's the kind of edge you can keep repeating.

You still need judgement, though, because a 95¢ contract loses everything if the upset happens, so you want markets where you've a real reason to think the upset is rarer than 5%, and you'll want to size each one so a single shock doesn't hurt you (module 4 is all about sizing). It's worth posting your own offers where you can, since makers did better than takers, and you should check the fee too, because on a trade that only makes 3% a big fee can eat most of it.

If you still fancy a long shot, make it earn your money first. If you think a 4¢ contract is really 8%, write down why before you buy, because most traders lean the other way, and while single markets do get mispriced in both directions (so a cheap contract can be a good buy), you'll want a reason you can point to. If reading a price as a chance still feels slow, module 1 of the course goes through it, and you can try it on the practice markets without spending anything.

Where the numbers come from

The Kalshi figures are from Makers and Takers: The Economics of the Kalshi Prediction Market by Bürgi, Deng and Whelan (University College Dublin, January 2026, first out as UCD working paper WP25/19), and if you trade on Kalshi it's worth your time, with the chart of returns by price band being the bit we'd read first. The racing figures are from Snowberg and Wolfers in the Journal of Political Economy (volume 118, 2010), which you can read free as NBER working paper 15923, and Griffith's 1949 study is in The American Journal of Psychology (volume 62, pages 290 to 294). For today's fees you'll want Kalshi's fee schedule, since they've changed since the study. We worked out every fee and return on this page with a short script, and it matches both of the worked examples in the Kalshi paper.