"Who wins tonight?" and "who wins the tournament?" sound like the same question. They are not. The first is about one match between two known players. The second is about up to seven rounds and matchups nobody can name yet, because they depend on results that have not happened.
This article explains, in plain language, how our tournament simulations answer the second question, and why they need a different tool from our daily match predictions. The WTA simulations work the same way.
Two questions, two tools
Our daily tennis predictions come from neural networks that study each scheduled match in detail: recent form, serve and return numbers, surface history and much more about the two players on court. That detail exists only for a match that is actually scheduled, not for a quarterfinal next Friday between two players still to be decided.
A tournament forecast has to price every possible matchup, including ones that may never happen. For those it uses player ratings: overall Elo, Glicko-2, where each player finished last season, and surface ratings for the tournament's surface. A model fitted on more than 26,000 past ATP matches turns the gaps between two players' ratings into a win probability, so any pair of players always has one. The one exception is the next match already on the schedule: there we use the daily match model's probability, because it reads current form better than the ratings do.
Then we play the tournament 20,000 times
- Keep what already happened. Played matches are facts and their winners always advance.
- Flip weighted coins. Every match still to be played is decided by chance, weighted by those probabilities. A player with a 70% chance wins 70% of the runs and loses 30%, because upsets are part of tennis.
- Record the champion, reset, repeat. 20,000 times.
After that it is just counting. Won 6,100 of 20,000 simulated tournaments? The title chance is 30.5%. Reached the semifinal in 13,400? A 67% chance of making the last four. The same 20,000 brackets answer every question at once: title odds, the chance of reaching each round, even how likely a particular final is.
Why not just pick the best player?
Because the draw is half the story. Take two players of equal strength. One has qualifiers and an out-of-form seed in his section, the other has to beat three strong seeds to reach the semifinal. Same ability, very different title chances, and the simulation sees it because it plays out both paths thousands of times.
It is also why a player's odds can rise without him hitting a ball: when a dangerous rival loses on the other side of the draw, the next run already reflects the easier road.
What "30%" means
A 30% title chance means that, if this tournament were played from today's situation many times, that player would win about 3 of every 10. It does not mean he will lose, and we were not wrong if he does. Surprise champions happen, and an honest forecast gives them small but real numbers in advance.
A probability is only useful if it is calibrated: things we give 40% should happen about 40% of the time. We re-forecast 30 ATP tournaments from 2023 to 2025 from their opening day, with a model that only knew what was known then. The chances of reaching each round came within one or two points of what actually happened, for every range of probabilities. How we check calibration for every model is on the model transparency page.
The track record in numbers
The same historical test, in numbers: 30 ATP tournaments (2023–2025), every one re-forecast from its opening day with a model trained only on data through 2022:
| Segment | Tournaments | Champion = our top pick | Champion in our top 5 |
|---|---|---|---|
| All tournaments | 30 | 33% | 67% |
| Grand Slams | 12 | 42% | 100% |
| Masters 1000 | 10 | 30% | 50% |
| ATP 500 | 5 | 20% | 40% |
| ATP 250 | 3 | 33% | 33% |
For context: in a 128-player draw, naming the champion at random succeeds 0.8% of the time, always picking the #1 seed succeeds 21% of the time, and always picking the highest-rated player 30%. The simulation's edge over simply backing the favourite comes from pricing the draw. The live track-record section always shows the latest audited numbers. They update whenever the engine is retrained and re-tested, never by hand.
Forecasts that sharpen round by round
After every real match the bracket changes: winners advance, losers drop to zero and everyone else's path gets clearer. Each daily run adds the new results, refreshes the ratings and plays all 20,000 tournaments again, so the numbers during the quarterfinals reflect everything that has already happened.
See it live
Title odds and round-by-round chances for every ATP tournament in play, updated daily.
Open the tournament simulationsThe fine print
- Qualifiers not yet known play with a default rating until their name is in the draw.
- Withdrawals and walkovers count once they appear in the official draw.
- Randomness is small. With 20,000 runs the numbers move by only a few tenths of a point from chance alone. A change from one day to the next means new information.