What is the favourite-longshot bias in horse racing?

Where it was first noticed

In 1949 Richard Griffith published a short paper in the American Journal of Psychology on how American racegoers set odds. His interest was psychological rather than financial. Odds express what a crowd believes; the proportion of winners at those odds expresses what actually happened. Griffith wanted to know whether the two lined up, and found that they did not, in a way that was consistent rather than random.

Mukhtar Ali took the question into economics in 1977 in the Journal of Political Economy, treating the pattern as something to be explained by bettors' probability and utility estimates rather than merely described. By 1988 Richard Thaler and William Ziemba were writing it up in the Journal of Economic Perspectives as one of the standing anomalies of the field, in a series whose whole purpose was to catalogue the places where real behaviour and clean theory come apart.

That is an unusual pedigree for a racing observation. Most things punters believe about racing have been examined by nobody. This one has been examined by economists for three quarters of a century.

Two explanations, and which one the evidence favoured

There have always been two broad stories for why the bias exists.

The first is that bettors simply like risk. On this account nobody is mistaken about anything. A punter knowingly accepts a worse expected return on the 25/1 shot because the shape of the bet appeals to them, in the way a lottery ticket appeals despite everyone knowing the odds are dreadful.

The second is that bettors misperceive the probabilities. On this account the punter is not buying excitement at a known price. They genuinely believe the outsider's chance is better than it is, because very small probabilities are hard to feel accurately.

Erik Snowberg and Justin Wolfers tested the two against each other and published the result in the Journal of Political Economy in 2010. Their approach was to see whether a model fitted to simple win betting could also account for how the same people bet into compound pools. They reported evidence favouring misperceptions of probability, in the manner suggested by Prospect Theory, over risk-love.

The difference matters more than it sounds. If the bias is a taste, it is a preference and there is nothing to correct. If it is a misperception, it is an error, and errors are the kind of thing a market can learn to price out.

What we found in our own data

We hold a governed study of 60,671 British and Irish races with a complete Betfair book and a single clear favourite. Races where two runners shared the lowest price are held out entirely, all 300 of them, because picking one of two identical prices would invent a favourite the market never named.

The headline number is the one everybody expects. The favourite won 35.0% of those races and did not win 65.0% of them. People find that surprising, and it is not surprising at all: a horse priced around a one-in-three chance is expected to lose about two times in three. That is what the price said it would do.

The number we actually care about is the second one. Comparing the wins that happened against the wins the favourites' own prices implied gives 0.997, where 1.000 would mean the outcomes matched the prices exactly. Across sixty thousand races, the market's own expectation of its favourites was almost precisely what the racing delivered.

Why an old finding is not a modern edge

Here is the trap. A punter reads that the favourite-longshot bias is real, well documented and decades old, and concludes that backing favourites must therefore be profitable. The reasoning feels airtight and it has a hole in the middle of it.

A statistical bias in prices is not the same thing as money on the table. Three things sit between them. The margin has to be small enough that a thin edge survives it. The bias has to still exist now, in this market, rather than in the market somebody measured in 1977. And enough of it has to remain after everybody else who has read the same literature has already bet on it.

Prices are not a photograph of a fixed truth. They are the current state of an argument between people who are also reading the research. A well publicised inefficiency is, by construction, one of the least likely to still be there.

This is the same shape of argument we make about the draw, which is a real effect and a worthless standalone edge at the same time.

What we take from it

We are not in the business of telling you the favourite is overpriced or underpriced. On our data, over a large sample, it is priced about right on average, and that is a genuinely useful thing to know because it tells you where not to look.

The interesting question is not whether favourites win often enough. It is which particular races put a favourite under a kind of pressure the price has not fully absorbed. That is a question about race structure, pace and field shape rather than about odds, and it is the question we spend our time on.

We have no validated betting edge, and we say so on every page where it could matter. What we have is a measurement instrument and a public record of what it has and has not found.

How often does the favourite win is the full study. Why favourites lose takes the question apart by field size and price. What is Betfair Starting Price explains the price everything above is measured against.

See how the same discipline is applied to today's card.