The serve dominance gap
The most consequential structural difference between the tours is serve dominance. On hard court, ATP men hold serve roughly 80% of the time; WTA women hold around 65%. That single statistic ripples through every part of model design:
- WTA features the much higher break rate, so points-won metrics compress less.
- ATP set scores are more bimodal (6-3 / 7-6 dominate); WTA set scores spread wider.
- Best-of-5 (men's slams) compounds serve-hold dominance further; women's slams are best-of-3.
Surface effects differ between tours
Clay-court WTA is the closest thing tennis has to a coin flip — break frequencies are high enough that match outcomes are dominated by mental and form factors more than structural skill gap. On the same surface, ATP top-30 vs top-100 is far more deterministic.
Where TIPERO splits the model
TIPERO's calibration is genuinely separate per tour:
- Separate Elo tables (
data/ratings_atp.csv,ratings_wta.csv) with surface-specific ratings. - Separate ace-rate calibration (
ace_calibration_{atp,wta}.csv). - WTA-specific calibration overlay (the V4 logistic regression isotonic-blends differently per tour).
- Separate tier policies — clay-WTA is largely banned because of the structural noise.
The mistake of pooling ATP + WTA data
A single model trained on combined ATP + WTA data ends up calibrated to neither — it learns an averaged serve-dominance prior that's too high for WTA and too low for ATP. The fix is to either fit two models, or include a tour indicator and an interaction term with every serve-related feature.
What changes per surface within each tour
| Tour × Surface | Hold rate | Implication |
|---|---|---|
| ATP hard | ~80% | Best-of-3 outcomes near coin-flip on similar Elo; tiebreaks frequent. |
| ATP clay | ~74% | More breaks; favouring ground-game players over big servers. |
| ATP grass | ~83% | Serve dominance peaks; tiebreaks dominate set outcomes. |
| WTA hard | ~65% | Higher variance; break exchanges common. |
| WTA clay | ~58% | Near-random — TIPERO bans this tier for most strategy. |
| WTA grass | ~68% | Best-of-3 with tiebreaks; favourites win less often than odds suggest. |
Bottom line
ATP and WTA share the rules but not the structure. A model that doesn't split them out is leaving real edge on the table. TIPERO has run separate calibrations from day one, and the publicly tracked profit reflects that.
See TIPERO's ATP + WTA picks →