The favourite-longshot bias: why longshots cost more than they pay
Published
Betting markets are efficient enough to humble almost everyone — but they are not uniformly efficient. The best-documented wrinkle is the favourite-longshot bias: across decades of studies, first in horse racing and later in fixed-odds football markets, backing favourites at the offered price loses slowly, while backing longshots at the offered price loses fast. Not because favourites are secretly good bets, but because the price of a longshot systematically overstates its chances by more. If you bet, or build models, or just read prediction sites critically, this bias should change how you look at every long price you see.
What the favourite-longshot bias is
Strip the jargon and the claim is simple: the shorter the price, the closer it tends to sit to the true probability; the longer the price, the more it flatters the outcome’s real chances.
A bookmaker’s odds contain two things: an estimate of probability and a fee. The fee — the overround, or margin — is the amount by which the implied probabilities of all outcomes sum past 100%. (If margins are new to you, start with how to convert odds to implied probability, which walks through stripping them out.) The favourite-longshot bias is a statement about where that fee lives. It is not spread evenly across the outcomes. It is loaded, disproportionately, onto the longshot.
The consequence: two bettors backing opposite sides of the same match, at the same bookmaker, are paying very different prices for the privilege — and the one holding the underdog ticket usually doesn’t know it.
The math: same margin, very different burden
Take a hypothetical two-outcome market where the true probabilities are 75% and 25%. Fair odds would be 1.33 and 4.00.
A bookmaker running a 6% overround could shade both outcomes proportionally: implied probabilities of 79.5% and 26.5%, odds of about 1.26 and 3.77. Then both bets lose the same expected 5.7% per unit staked. That is the textbook picture — and it is not what real books look like.
What you typically see instead is something closer to 1.30 on the favourite and 3.45 on the longshot. Check the totals: implied probabilities of 76.9% and 29.0%, summing to 105.9% — the same 6% overround. But now run each bet against the true probabilities:
- Favourite at 1.30 with a true 75% chance: expected return 0.75 × 1.30 = 0.975, an expected loss of 2.5%.
- Longshot at 3.45 with a true 25% chance: expected return 0.25 × 3.45 = 0.86, an expected loss of about 14%.
Same fixture, same bookmaker, same headline margin — and the longshot backer’s expected loss is more than five times larger. The classic horse-racing studies found exactly this shape at scale: blanket-backing short favourites lost a few percent per bet, while blanket-backing extreme outsiders lost many times more. Fixed-odds football shows the same pattern, milder but persistent, and it grows with the length of the price.
Why the bias exists
Three explanations survive scrutiny, and they reinforce each other.
Bettors like lottery tickets. Recreational money is drawn to large payouts, and people systematically overweight small probabilities — a 12% chance feels like more than 12%. Demand for longshots is therefore less price-sensitive than demand for favourites, and a bookmaker facing inelastic demand does what any business does: charges more.
Longshot prices are where the bookmaker is most exposed. An error on a 1.30 favourite costs a bookmaker a few points of margin; the same relative error on a 15.00 longshot can be ruinous, and it is precisely the price sharp bettors attack when a book is slow to react. Padding long odds is cheap insurance against both model error and informed money.
Favourites get audited; longshots don’t. The heaviest betting volume — including the most price-sensitive, professional volume — concentrates on favourites and main lines, which keeps those prices honest. The far end of the odds range gets less scrutiny per price, so distortions live longer there.
None of this requires anyone to be foolish. It is the equilibrium of a market where one side of the book is bought for entertainment and the other side is priced for defence.
What it means if you use a prediction model
The bias has a sharp practical edge for anyone comparing model probabilities against market odds.
Raw implied probabilities overstate longshots by construction. If you take 3.45 at face value as “29%”, you inherit the bookmaker’s padding as if it were information. Any honest model-versus-market comparison has to strip the margin first — and because the margin is not distributed evenly, the naive proportional method understates favourites and overstates longshots. Margin-weighting methods that remove more juice from longer prices exist for exactly this reason.
“Value” found on longshots deserves double suspicion. A model that keeps flagging big prices as underpriced is making the single claim the market is best defended against. Sometimes it is right; far more often it has found the bias plus its own noise. This is one reason most detected “value bets” evaporate against closing odds — apparent edges cluster where margins, not mistakes, live.
Calibration must be checked by probability band, not overall. A model can look well-calibrated on average while being systematically overconfident about underdogs and underconfident about favourites — the same shape as the market’s own distortion. The test is to bucket predictions by stated probability and compare each bucket against what actually happened, which is precisely how a public, append-only record like our track record lets anyone audit a forecaster — including us — one odds band at a time.
How to spot the bias in any betting record
The favourite-longshot bias also gives you a fast diagnostic for any tipster or system whose history you can inspect:
- Split the record by odds band — say under 1.50, 1.50–2.50, 2.50–5.00, and above 5.00.
- Compute return on stakes per band, not win rate. A longshot-heavy record can show a dismal win rate while being genuinely profitable, and vice versa.
- Expect the slope. A record built by blindly taking prices will drift from mildly negative in the short bands to sharply negative in the long ones. A record that is flat or positive across bands — over a serious sample — is showing you something real.
- Distrust longshot-only glory. A highlight reel of 8.00 winners is the cheapest thing in betting to accumulate by luck and the most expensive to accumulate by skill.
The pattern is not a law of nature — its size varies by league, market and era, and it has occasionally reversed in some sports — so treat it as a prior, not a verdict. But it is one of the most replicated findings in the economics of betting, and the burden of proof sits with any record that claims to have escaped it.
The unglamorous summary: long odds are where the fee hides. Whether you bet them, model them, or merely admire them, price them as if the market has already charged you extra for the dream — because it has. And as always: no bias, and no model, turns betting into an income; stake only what you can afford to lose.