Tennis betting research from our 300,000-match database. Each study states its sample, method and caveats, including the results that went against us.
Home players beat their rating by +2.0pp (ATP) and +1.0pp (WTA) on 52,521 matches. The market already prices about half of it. No edge claimed.
One in 28 matches ends in retirement. Base rates by tier, age and surface, and a 2.24x prior-retirement hazard, after two fixes to our detector.
Players really do underperform after winning a title, by 3.5pp on 64,669 transitions. The cause is ELO overshoot, not emotion: runners-up dip the same.
Layoff rust measured on 238,716 player-matches: returning players underperform ELO by 0.9pp after 8-14 days off, 9.8pp past a year, fading over 4-6 matches.
No marathon cliff: 218,890 player-matches show tennis fatigue is a smooth games gradient, an easy win is a +3.4pp bonus, and we correct our own -4.5pp rule.
Surface-split tennis ELO loses to the pooled rating on 62,744 matches, both tours. The best mix is 0.7 overall plus 0.3 surface.
Landing more first serves wins the match 55.9% of the time. Winning more first-serve points wins it 84.9%. 14,082 ATP and WTA matches.
The market's favorite wins 71.5% and its odds are near-perfectly calibrated. Our model doesn't beat the close. Why 80% accuracy claims prove nothing.
Our 'beat the close 72%' stat included the bookmaker vig. Against the vig-stripped exchange close, our closing-line value is negative.
Lefties win 46.3% vs 46.8% implied. Across 802 players we found fewer mispriced outliers than chance predicts. Style is not a betting edge.
On 209,000 matches the rested player wins just 52.9%; a three-set win the day before costs about 4.5 points. Refined by our later gradient study.
Across 21,917 matches, flat-backing heavy favourites lost 3.2% and longshots lost 34.8%. The bookmaker margin is loaded onto underdogs.
We graded 72 matches our model declined: flat-betting them would have lost 13.6%. A CLV claim in the first version did not survive devigging.
The H2H leader wins 59% of rematches, but against the betting odds the angle adds nothing: 36.0% actual vs 36.7% implied.