Across 52,521 one-home matches, home players beat their ELO expectation by 2.0 percentage points on the ATP side and 1.0 on the WTA side. Against the closing price the effect mostly disappears: the devigged residual is +1.41pp pooled and clears zero on neither tour alone. We filed it as rating calibration, not edge.
Home advantage is the most comfortable belief in sport. Crowds lift the local player, travel wears out the visitor, and everyone can name the night it looked exactly like that. Tennis should be the clean place to test it: no home stadiums in the structural sense, a global calendar, and the same two players who will swap the home label a few weeks later in another country.
So we tested it on our own database. The answer has three layers. Home advantage is real: +2.0 percentage points against rating expectation for ATP players, +1.0 for WTA, on 52,521 one-home matches. The naive version of the measurement gets the sign wrong, because home wildcards drag the average down. And the layer that decides whether any of this earns money: the market already prices about half of it, and what survives is too thin to stake. This is the study that did not clear our own bar, published anyway, because a research program that only reports its wins is a marketing program.
The published record on tennis home advantage is genuinely split. The reference point is Koning, who modelled professional matches with a consistent logit specification and concluded that significant home advantage exists for men while "the performance of women tennis players appears to be unaffected by home advantage" [1]. A decade later, Wunderlich, Corten, Furley and Memmert looked at German team tennis across levels and found the opposite headline: home advantage that exists independent of competition level, gender and even the COVID crowd restrictions [2]. Two careful papers, two conclusions about the same phenomenon. That usually means the effect is small enough for sample and setting to decide the verdict, which is exactly the situation where a bigger panel helps.
We grade against ELO expectation rather than raw win rates, for the same reason as every study in this series: home players are not average players, so "home players win a lot" confounds selection with advantage. ELO methods rank among the most accurate published tennis forecasters in the one head-to-head comparison of the field [3], and our specific blend is the measured optimum from our surface-ELO study.
Our database stores no venue country, so we derived it from tournament names through a reviewable 700-venue map: 91.7% of matches resolved to a host country, and the rest were excluded, never guessed. A match enters the sample when exactly one player is playing in their own country (52,521 matches). Matches where both players are home cancel by construction (16,248), and the 101,563 neither-home matches become the control pool: whatever miscalibration our rating has at a given ELO gap, we subtract it from the home cells at the same gap, so the home effect is measured net of the rating's own habits. Both players need at least 20 prior rated matches. The reason for that burn-in is the methodological story of this study, and it gets its own section.
| cell | n | vs ELO, adjusted | 95% CI |
|---|---|---|---|
| ALL | 52,521 | +1.59pp | [+1.20, +1.98] |
| ATP | 33,387 | +1.97pp | [+1.47, +2.46] |
| WTA | 19,134 | +0.94pp | [+0.31, +1.58] |
Both tours are positive and significant, and the confidence intervals are disjoint: the ATP effect is roughly double the WTA effect. That resolves the Koning-Wunderlich disagreement into a gradient rather than a contradiction. It also explains how both papers could be right about their own data. Our WTA main-tour cell alone is +0.38pp on 3,411 matches, statistically nothing, which is the slice closest to what tour-level studies like Koning's could see [1]. The WTA significance in our headline is carried by volume at ITF level (+1.31pp on 14,067 matches), the same lower-level territory where the German team-tennis data found its effect [2]. Small effect, big panels needed, settings matter.
| tier | n | vs ELO, adjusted | 95% CI |
|---|---|---|---|
| ATP main tour | 4,363 | +2.03pp | [+0.65, +3.40] |
| Challenger (M) | 14,158 | +1.77pp | [+1.00, +2.53] |
| ITF (M) | 14,866 | +2.14pp | [+1.41, +2.87] |
| WTA main tour | 3,411 | +0.38pp | [−1.12, +1.87] |
| WTA 125 | 1,656 | −0.97pp | [−3.15, +1.22] |
| ITF (W) | 14,067 | +1.31pp | [+0.56, +2.05] |
The ATP effect is strikingly flat across tiers: main tour, Challenger and ITF all sit near +2pp. Whatever produces it is not a big-stadium phenomenon.
Run the same study with no burn-in requirement and home advantage disappears: ATP reads −0.22pp and WTA reads −1.30pp. Naively measured, home players underperform.
The mechanism is wildcards. Tournaments hand main-draw and qualifying spots to local players, so lightly-rated locals are funnelled into the home cell. A player with a handful of rated matches still carries our tier debut prior (1250 at ITF, 1350 at Challenger, 1500 at tour level), and that prior is not a measurement. Their rating is systematically too high the day the wildcard puts them on court, they lose more often than the number says, and the loss lands entirely in the home column. Sweep the burn-in threshold and the measured advantage climbs monotonically until about 20 prior rated matches, then goes flat: a data-quality boundary, not a tuned knob.
The favourite/underdog split says the same thing from another angle. ATP home favourites and home underdogs both sit at +1.97pp adjusted. WTA home favourites get +2.16pp while WTA home underdogs get +0.08pp, and the WTA underdog cell is where wildcard selection concentrates. Any study that measures home advantage on raw results without confronting who gets put on court at home is measuring tournament wildcard policy as much as psychology.
COVID gives the crowd theory a natural experiment, and the result is awkward for it. In the closed-crowd window from March 2020 to June 2021, ATP home advantage did not vanish. It was +4.97pp [+1.11, +8.83] on 640 matches, point-higher than the +1.91pp of normal times, though the band is wide. The WTA closed-crowd cell (196 matches) is too thin to read. The German team-tennis study reached the same conclusion from its own COVID window [2]. If the advantage survives empty stands, the residual explanations are the boring ones: sleeping in your own bed, no long-haul travel, familiar courts, familiar food, scheduling built around you. Crowds are the most visible correlate and, on this evidence, not the load-bearing one.
Host country cells scatter around the mean the way small cells should, with a few that clear it: the United Kingdom (+4.68pp on 1,427 ATP matches), Australia for WTA players (+5.07pp on 1,020), the USA on both tours. We read those as ordinary small-sample spread around a real +2pp mean until a larger panel says otherwise.
Everything above is a statement about our rating. A bettable edge is a statement about the price. We took the window where we hold closing odds (February to July 2026), kept matches with exactly one home player and 20+ rated matches on both sides, and asked the same question twice: did home players beat our ELO expectation, and did they beat the devigged market expectation?
| tour | n | vs our ELO | vs market |
|---|---|---|---|
| ATP | 3,072 | +2.71pp [+1.10, +4.32] | +1.04pp [−0.54, +2.62] |
| WTA | 1,717 | +2.57pp [+0.42, +4.71] | +2.07pp [−0.02, +4.16] |
| ALL | 4,789 | +2.66pp [+1.37, +3.95] | +1.41pp [+0.15, +2.67] |
Against our own rating the effect replicates out of sample, stronger if anything. Against the market, roughly half of it is gone before we arrive: bookmakers already move their lines for home players. The residual +1.41pp clears zero pooled, by a hair, and clears nothing on either tour alone, on a five-month window. We have been here before: every style effect we hunted was priced before we got there, and the honest reading of this table is "suggestive, unproven".
Nothing that stakes money. Our framework rule is that a new contextual signal must beat the closing price before it can touch a bet, and this one has not. Home advantage goes into the model the same way the layoff curve did not: it is queued as a rating-calibration variant (our ELO demonstrably underestimates home players by about 2pp, and a rating should not carry a known bias), with an explicit no-edge annotation so no analyst leads a betting thesis with it. If the priced residual is still positive when the odds panel is three times this size, that is a different conversation, and we will publish the update either way.
Method: 302,945 country-resolvable matches, ATP and WTA, all tiers; 8,165 of 10,102 tournaments name-resolved to a host country (venues with no fixed country, ambiguous names and unmapped names dropped; 91.7% match coverage). Sample is matches with exactly one home player and 20+ prior rated matches for both players (n=52,521). Effect is observed minus ELO-expected home-player win rate; the adjusted column additionally subtracts the mean residual of neither-home control matches (n=101,563) in the same pre-match ELO-gap bin, removing the rating's own miscalibration at that gap. p-values are two-sided z-tests. The no-burn-in and burn-in-sweep numbers are from the 2026-08-12 run of the same command recorded in our research program memo; the market check uses the 2026-02-12 to 2026-07-17 odds slice with devigged closing prices, both sides 20+ rated matches, exactly one home player (n=4,789), measured 2026-08-12; the vs-ELO arm of it was re-verified 2026-08-14 on the same dates. All other cells are from the 2026-08-14 rerun.
References:
[1] R. H. Koning, "Home advantage in professional tennis," Journal of Sports Sciences, vol. 29, no. 1, pp. 19–27, 2011, doi:10.1080/02640414.2010.516762.
[2] F. Wunderlich, M. Corten, P. Furley, and D. Memmert, "Home advantage in tennis exists independent of competition level, gender and COVID-19 restrictions: evidence from German team tennis competitions," International Journal of Sport and Exercise Psychology, vol. 22, no. 7, 2024 (online 2023), doi:10.1080/1612197X.2023.2235592.
[3] S. A. Kovalchik, "Searching for the GOAT of tennis win prediction," Journal of Quantitative Analysis in Sports, vol. 12, no. 3, pp. 127–138, 2016.
Yes, and it is measurable: home players beat their rating expectation by +2.0 percentage points on the ATP side (33,387 matches) and +1.0 on the WTA side (19,134 matches), significant on both tours. The effect is roughly flat across main tour, Challenger and ITF levels for men. But it only appears once lightly-rated home wildcards are excluded; measured naively, wildcard selection flips the sign.
The crowd is not the load-bearing part. In the COVID closed-crowd window (March 2020 to June 2021) ATP home advantage did not disappear: +5.0pp on 640 matches, wide interval, but no collapse. Independent German team-tennis research found the same. Travel, sleep, court familiarity and scheduling are the better candidates.
Mostly no. Against the devigged closing price, home players came in at +1.41pp pooled on 4,789 matches in 2026, which barely clears zero and clears it on neither tour separately. The market already prices about half the effect. We queued it as a rating calibration with an explicit no-edge annotation, and nothing in our framework stakes on it.
See today's picks — published before the match, graded in public →