Regression to the mean is the statistical phenomenon where extreme performance tends to move back toward the average over subsequent observations. In sports betting, markets often underweight regression, creating exploitable patterns.
Classic examples: - A team covers 7 of its last 8 games: the market may shade them as a "hot team," but their true cover rate hasn't changed. The next game, regression suggests a revert toward ~50% cover rate. - A quarterback throws 4 touchdowns: the next game's TD prop line may be set too high, overvaluing recent performance. - A kicker hits 3 field goals from 50+: subsequent prop lines for FG make may not regress properly.
Why markets overreact: Recency bias is universal in human judgment, and betting public opinion is overwhelmingly recent-form driven. Books exploit this — or at least don't fully correct for it — because balanced action is more valuable to them than perfect pricing.
Modeling approach: Use regression-aware features. Weighted rolling averages that decay recency overweight decay too fast. Bayesian priors anchored to position/team averages provide automatic shrinkage toward the mean.
The key check: When does recent outperformance reflect genuine skill change vs. noise? Injury return (genuine), scheme change (genuine), opponent quality (matchup), or fluky fumble recoveries (noise) require different treatment.