The favourite–longshot bias
Across a century of data and most betting markets, longshots are systematically overbet and favourites systematically underbet. What the effect is, how large it actually is, why it persists, and why knowing about it is not by itself a strategy.
The favourite–longshot bias is the tendency for long-priced selections to lose more than their odds suggest and short-priced ones to lose less. It has been measured in horse racing, football and most other betting markets for decades. The effect is real, well documented and small enough at ordinary prices that it does not by itself overcome bookmaker margin.
The shape of the effect
Grouping bets by price and comparing implied probability against observed frequency produces a consistent pattern:
| Price range | Implied probability | Observed frequency (typical) | Direction |
|---|---|---|---|
| 1.10–1.50 | 67–91% | Slightly higher | Underbet |
| 1.50–2.50 | 40–67% | About right | Roughly fair |
| 2.50–5.00 | 20–40% | About right or slightly lower | Roughly fair |
| 5.00–15.00 | 7–20% | Lower | Overbet |
| 15.00+ | under 7% | Substantially lower | Heavily overbet |
These are directional patterns from the literature, not fixed coefficients: the magnitude varies by sport, market and era. The consistent finding is the shape — flat through the middle, with the distortion concentrated at the extremes.
Why does the favourite–longshot bias persist?
- Risk-love at long prices. Some bettors value the small chance of a large payout above its expected value, exactly as with lottery tickets.
- Probability misperception. People systematically overestimate small probabilities — a robust finding in behavioural economics well outside betting.
- Bookmaker margin loading. Books apply more margin where demand is least sensitive, which is at long prices. Part of the measured "bias" is simply the price being worse there.
- Limited arbitrage. Correcting the bias means backing favourites in volume at low returns, which is capital-intensive and heavily restricted by books.
The third cause matters for interpretation. Some of what is measured as a behavioural bias is really a pricing decision — which means part of the effect cannot be captured by betting, only observed.
Why is the favourite–longshot bias not a strategy?
The obvious inference — back favourites, avoid longshots — fails on arithmetic. At odds of 1.30, a 2% edge from the bias against a 4% margin still loses. The bias narrows the gap; it does not close it.
Where it is genuinely useful is as a prior on your own model. If your model repeatedly finds large edges on selections priced above 10.00, the bias is a reason to suspect the model rather than the market: that is precisely the region where prices are least favourable and where a modelling error is most likely to look like an opportunity.
We apply this as a hard bound rather than a belief. Selections outside a configured price range are excluded from publication regardless of the computed edge, because the historical evidence at extreme prices does not support acting on model estimates there. The threshold is a setting, and changing it changes the version stamped on every subsequent signal — so past decisions remain explicable.
Check any of this against our record
Every signal CONSENSUS publishes carries the bookmaker odds fixed before the event starts and the settled result afterwards — including the drawdowns and the losing runs. The running total is on the front page and every entry is in the log.