Squash Match Probability, from Serve to Match
Give one number per player, the chance of winning a rally on your own serve, and this article shows you how that turns into game odds, best-of-3 or best-of-5 match odds, and a full game-score distribution under PAR-11 scoring.
What the calculator does
Modern squash uses point-a-rally scoring, called PAR-11. Every rally scores a point for whoever wins it, no matter who served. A game goes to 11 points, but you must win by 2, so at 10-10 play continues until one player leads by two. Matches are best of 3 or best of 5 games.
The calculator takes two inputs: Player A wins a fraction p of the rallies they serve, and Player B wins a fraction q of the rallies they serve. From those two numbers it builds the exact probability of every scoreline and rolls them up into game and match odds.
Here is the hook. Suppose A wins 53% of rallies on serve and B wins 50% on serve. That three-point edge sounds tiny. It is. But over a full best-of-5 match it grows into roughly a 58% chance of winning the match. Small per-rally edges compound, and the point of this tool is to show exactly how much.
When to use it, and when not
Use it when you have a stable estimate of each player's serve-rally win rate, from match charting or a season of data, and you want to translate that into match odds. It is well suited to comparing two known players, to seeing how sensitive the result is to a one-point change in rally skill, and to understanding why best-of-5 favors the stronger player more than best-of-3.
Do not use it as a prediction of a single match on a single day. Rally win rates drift with fatigue, tactics and the referee's let-and-stroke decisions. The model assumes both serve rates are fixed and known, which they never are exactly.
The serve edge in squash is genuinely small. Do not enter numbers like 70% unless you mean that one player wins 70% of every rally they serve, which would describe a total mismatch. Professional serve-rally rates sit only a few points above 50%.
The serve rotation and why it matters
In PAR scoring the winner of a rally serves the next one. So the server changes whenever the receiver wins a point. This is different from old hand-out scoring, where only the server could score.
The consequence is that the identity of the server drifts through a game. To handle this cleanly, the calculator tracks the state of the game as a pair: the score, and who is serving. From any state there are two moves. If A serves, A wins the rally with probability p and keeps serving, or loses it with probability 1-p and hands serve to B. If B serves, B wins with probability q.
The opening serve is decided by a spin of the racket, so the model averages over both possibilities: A serves first half the time, B the other half.
The math: a Markov chain to 11
Label a game state by the score and the server. Write P(a, b, s) for the probability that A eventually wins the game given the current score is a to b and player s is serving. The recursion for A serving is:
Here p is A's rally win probability on serve. If A wins the rally the score becomes a+1 to b and A keeps serving. If A loses, the score becomes a to b+1 and B serves next. The mirror equation for B serving replaces p with 1-q, since B holding serve means A loses the point:
The boundary conditions close the system. A has won when a \ge 11 and a - b \ge 2, giving P = 1. B has won when b \ge 11 and b - a \ge 2, giving P = 0.
The only awkward part is deuce. At 10-10 the game can run forever in principle, so you cannot just unroll the recursion. Instead you solve the two-point cycle directly. Let g be A's chance to win from deuce. From 10-10 the game reaches one of three outcomes over the next two rallies: A goes up two and wins, B goes up two and wins, or the score returns to deuce. Averaging over who serves at deuce gives a single linear equation in g, which you solve algebraically. The calculator does this exactly, so there is no truncation error.
From games to the match
Once you have a single number G, A's probability of winning one game, the match is a shorter Markov chain. A best-of-3 match is won by the first to 2 games, best-of-5 by the first to 3.
The term G^2 covers a 2-0 win. The term 2 G^2 (1-G) covers the three orderings of a 2-1 win. For best of 5 the same logic gives:
One subtlety: strictly, the per-game win probability G is not constant, because the winner of a game serves first in the next. The calculator averages the opening toss, which makes G the same for every game to a very good approximation, since the server edge is small. The same best-of-N formula also drives the Playoff Series Probability Calculator if you already have a single-game win probability.
Reproducing the demo: 53% versus 50%
Enter A serving at 53% (p = 0.53), B at 50% (q = 0.50), format best of 5, independent points. This is the demo button.
- Solve the game Markov chain. With these rates A wins a single game about 54.2% of the time. The edge per game is larger than the edge per rally because a game is a race and the leader tends to stay ahead.
- Feed G = 0.542 into the best-of-5 formula: 0.542^3 (1 + 3(0.458) + 6(0.458)^2).
- Compute the bracket: 1 + 1.374 + 1.259 = 3.633. And 0.542^3 = 0.1592.
- Multiply: 0.1592 \times 3.633 \approx 0.578.
So a three-point rally edge becomes a 4.2-point game edge becomes a 7.8-point match edge. Each stage of the competition amplifies the gap. For best of 3 the same G gives about 55.8%, less amplification because there are fewer games for skill to assert itself.
Reading the game-score distribution
The tool also reports how the match is likely to be won, not just who wins. With A at 53% and B at 50% in best of 5, the game-score outcomes look roughly like the table below.
| Result | Winner | Probability |
|---|---|---|
| 3-0 | A | 0.159 |
| 3-1 | A | 0.219 |
| 3-2 | A | 0.200 |
| 0-3 | B | 0.097 |
| 1-3 | B | 0.161 |
| 2-3 | B | 0.164 |
The six numbers sum to 1. Add A's three rows to recover 0.578. Notice the closest score, 3-2 either way, carries about 0.364 of the probability. Squash matches between near-equal players are usually decided in the fifth game, and the distribution shows why.
The game-score distribution is where a small skill edge is most visible. A's 3-0 chance (0.159) is well above B's (0.097), a ratio of about 1.6, even though the match-win ratio is only 1.37. Dominant results are more sensitive to skill than close ones.
Momentum mode and its limits
The default model treats every rally as independent. Real players ride streaks. Momentum mode simulates 20,000 full matches point by point, and after each rally it nudges the winner's next-rally chance up by the momentum strength and the loser's down by the same amount.
At strength 0.03 the effect on match odds is small, usually under one percentage point, because streaks help and hurt each player symmetrically over a long match. At 0.1 you see fatter tails: more 3-0 sweeps and more improbable comebacks, but the average winner barely changes. Momentum reshapes the score distribution more than it moves the headline probability.
Run the independent model first. It is exact and instant. Switch to momentum only to check how streakiness widens the score distribution, and expect a wait of a few seconds while 20,000 matches simulate.
Common mistakes
Confusing serve-point rate with overall rally rate is the frequent one. Under PAR scoring your serve rate and your receive rate can differ, and the model uses only serve rates because the winner-serves rotation ties them together. If you have overall rally win share instead, the number you enter should be close but not identical.
Another is reading too much into a single figure. A 58% match probability means A loses 42 matches in 100. Over a five-match weekend, A winning three and losing two is completely ordinary. The number describes a long-run frequency, not a guarantee.
Finally, do not treat momentum strength as a measured quantity. It is a knob, not a fitted parameter. Charting data that pins momentum precisely is rare, so use 0.03 as a plausible default and read the range of outcomes, not a single decimal.
Related tools
The same Markov-chain approach drives the racket-sport calculators on this site. For serve-point scoring with games, sets and tiebreaks, see the Tennis Match Probability Calculator. For rally scoring in other sports, try the Volleyball Match Probability Calculator, the Table Tennis Match Probability Calculator, and the Badminton Match Probability Calculator with its 30-point cap. If you prefer to start from a rating rather than a serve rate, the Elo Rating Calculator converts two ratings into an expected score.
Frequently asked questions
Why does a 3% serve edge become a 7.8% match edge?
Because a match is a race repeated many times. Each rally is a coin flip weighted slightly toward A. Averaging over a game (about 15 to 25 rallies) and then over up to five games lets the small bias accumulate. The stronger player wins more of the long races than the short ones.
Does the winner-serves rule favor the server?
Only slightly. Under PAR scoring the server edge is a couple of points at most, since the receiver can score too. The model captures it exactly by tracking who serves in every state, then averaging the opening toss.
What if I only know overall rally win percentage?
Use it as an approximation for both serve rates, or split it: your serve rate is a little above your overall rate and your receive rate a little below. For near-even players the difference in the final match probability is under a percentage point.
Why is best of 5 better for the stronger player than best of 3?
More games mean more chances for the true skill gap to show. In the demo, A's match odds rise from 55.8% in best of 3 to 57.8% in best of 5. Longer formats reward the favorite.
Is the independent-points model realistic?
It is a clean baseline that matches aggregate results well. Points are not perfectly independent, which is why momentum mode exists, but at realistic streak levels the headline match probability barely moves. Trust the exact model for who wins and use momentum to explore how.
Related tools
Elo Rating Calculator
Expected score and rating change for any two Elo ratings and K-factor.
Playoff Series Probability Calculator
Chance of winning a best-of-3/5/7 series from your single-game win probability.
Soccer Score Simulator (Poisson)
Win/draw/loss and most likely scorelines from two expected-goals numbers.
Badminton Match Probability Calculator
From serve-rally win % to game and match probability with the 30-point cap.
League Season Simulator
Monte Carlo the rest of the season: title, top-N and relegation odds from the current table.
Table Tennis Match Probability Calculator
From serve-point win % to game, deuce and match probability for best-of-3/5/7.