Tennis fatigue and rest: do tired players really lose more?

“He played yesterday, he’ll be tired.” It is one of the most common reflexes before a prediction: you distrust the player who has played back-to-back matches and favour the one who arrives fresh. The idea is sensible, yet our numbers barely confirm it. What makes a player lose more than they should is almost the opposite: having played little.

We measured it across 38,123 ATP and WTA singles matches played between January 2022 and September 2026. For each match, we looked at how many days the player had rested and how many matches they had strung together, then compared the result with what their Elo rating predicted at equal level.

Playing yesterday costs very little

Take the players whose last match was yesterday, or the same day (the back-to-back rounds of a tournament), facing an opponent rested for two to thirteen days.

A point and a half less than expected: it goes in the direction of common sense, but is on the edge of chance (about one in fifty to get a gap like this by luck alone). Nothing that justifies writing off a player because they played yesterday.

Stringing matches together costs nothing measurable

The other way to talk about fatigue is to count matches played over the last fourteen days. Take those who had played six or more, facing opponents who had played three to five:

• Six matches or more: 52.5% wins (2,924 matches), exactly the rate Elo predicts (52.5%).
• Eight matches or more, facing opponents who had played seven or fewer: 52.9% wins against 53.6% expected, a gap of 0.7 points that cannot be told apart from chance (987 matches).

A crowded schedule is therefore not a signal on its own. A player who strings matches together is also, most of the time, a player who wins: their recent wins are already in their Elo.

What costs a lot: coming back after a break

Turn the question around. Instead of players who played too much, look at those who had not played for a long time, again facing an opponent rested two to thirteen days.

The contrast is stark: after a month or more without a match, a player wins 9.5 points less than their Elo predicts. A gap this size shows up by chance far less than once in a million. It is not rare: in 8% of matches (3,029 out of 38,123), one of the two players is back after thirty days or more, and in a quarter of matches (25%), after fourteen days or more.

The effect holds whichever way you slice it

We checked whether this could come from a single tour, period or surface. After thirty days or more, the player wins less than expected:

• on the ATP tour: 37.7% against 48.6% expected (−10.9 points, 849 matches);
• on the WTA tour: 39.4% against 47.6% (−8.2 points, 941 matches);
• from January 2022 to June 2024: −8.5 points, then from July 2024 to September 2026: −10.6 points;
• on hard court: −7.9 points; on clay: −15.0 points; on grass: −11.1 points (only 148 matches).

It is not about little-known players whose Elo would be poorly estimated either: among players above 1,900 Elo points, the gap is −9.9 points (964 matches). And the favourite who returns pays too: when the returning player had at least a 60% chance according to Elo, they won 64.8% of their matches instead of the expected 72.1% (596 matches). A 72% favourite on paper behaves, on return, like a 65% favourite.

The effect shrinks within one match

The second match back, meaning the player’s previous match was itself a return after thirty days or more, costs only 3.0 points (46.1% against 49.1%, 1,455 matches), against 9.5 for the first. The gap does not vanish entirely, but it shrinks by two thirds by the next match.

Why?

Our archive does not say why a player stopped. Several causes probably mix: an injury they have not fully recovered from, lack of match rhythm, a different preparation. One thing is certain about the mechanism: Elo does not move during a break. A player absent for a month keeps the rating they had when they left, while their current level is no longer quite the same.

Betting odds account for it, in part

Bookmaker odds account for this, though not entirely. Across the 18,796 matches in our sample with odds available, the same gap after thirty days or more (815 matches) is −12.0 points measured with Elo, but −4.8 points measured with the probability implied by the market odds. In other words, the market has already absorbed about 60% of the effect. Some remains, but on only 815 matches: too few to make a rule, especially as the odds used are averages.

This is consistent with a test we ran: adding actual rest days to the Pronocast model, along with other time-based variables, brought no gain (log-loss of 0.5991 against 0.5977, slightly worse). Our model already relies on the odds, which hold most of the information. That is the most likely explanation, not a proof.

How to read fatigue before a match

• Do not mark a player down because they played yesterday. The effect is a point and a half, on the edge of chance.
• Do not count back-to-back matches as a handicap. At six matches in fourteen days or more, players win as often as their Elo predicts.
• Be wary of a return after a month without a match, especially of a favourite: at equal Elo, the average loss is close to ten points of win probability, and it partly fades by the next match.
• Look at the date of the last match, not just the rating. Elo does not see it.
• Keep in mind that the odds already account for part of it. It is a piece of information for reading a match, not a money-making trick.

Pronocast looks at the number of matches played over fourteen days but, for now, not the number of rest days: our tests do not justify adding it. To see what the model does day to day, open today’s predictions, and the verified results page shows what it is worth once the matches are played. The full method is described on the algorithm page.

Method

Singles matches on the ATP and WTA tours, from 1 January 2022 to 24 September 2026, where both players had already played at least twenty matches: 38,123 matches, the same sample as our head-to-head study. Rest is the number of days since the player’s last singles match in our archive, all circuits combined (Challenger and ITF included) since January 2020; load is the number of singles matches played in the previous fourteen days. To compare at equal level, each player is set against a reference opponent (2 to 13 days of rest; 3 to 5 matches over fourteen days, or 7 matches or fewer for the comparison at eight matches or more), and their win rate is compared with the rate expected for their Elo probability, calibrated in 2.5-point bands on the whole sample. A match counts once per comparison. Gaps are judged with a z test on the sum of variances; the gaps of −9.5 points (30 days or more), −3.3 points (14 to 29 days) and −1.6 points (yesterday) have z values of −8.9, −5.1 and −2.4. Odds are the average of the bookmakers, converted to a probability without the margin. Breaks can have causes the archive does not know (injury, childbirth, scheduling choices): we measure an association, not a cause.