INTRODUCTION
How can tennis Elo ratings outperform ATP rankings? Our advanced model relies on three key indicators to measure player level more effectively. Learn how it works.
Tennis Elo Ratings
ADVANCED ATP MODEL
* Demo Chart *
| LEVEL | SURFACE | PLAYER | PRE - MATCH | MATCH STATS | VARS | MATCH RATING |
NEW RATING |
W L |
||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RATING | RD | GAMES PRED | SETS | COEF | S1 | S2 | S3 | GAMES WON | LV | DIF | ||||||
| CASE 1: THE FAVOURITE WINS UNDER THE GAME PREDICTION | ||||||||||||||||
| 5.20 | GRASS | SCHWARTZMAN | 4.07 | 1.13 | 0.39 | 2 | 1 | 5 | 6 | - | 0.44 | -0.52 | 0.05 | -0.23 | 3.84 | L |
| DRAPER | 6.32 | -1.13 | 0.61 | - | - | 7 | 7 | - | 0.56 | 0.52 | -0.05 | 0.23 | 6.55 | W | ||
| CASE 2: THE FAVOURITE WINS OVER THE GAME PREDICTION | ||||||||||||||||
| 7.41 | CLAY | SCHWARTZMAN | 8.34 | -0.93 | 0.59 | 2 | 1 | 6 | 6 | - | 0.86 | 0.74 | 0.26 | 0.50 | 8.84 | W |
| MARTINEZ | 6.48 | 0.93 | 0.41 | - | - | 1 | 1 | - | 0.14 | -0.74 | -0.26 | -0.50 | 5.98 | L | ||
| CASE 3: THE UNDERDOG WINS | ||||||||||||||||
| 8.00 | HARD | SCHWARTZMAN | 7.22 | 0.78 | 0.42 | 3 | 1 | 6 | 6 | 6 | 0.58 | 0.80 | 0.16 | 0.48 | 7.70 | W |
| TSITSIPAS | 8.78 | -0.78 | 0.58 | - | - | 7 | 3 | 3 | 0.42 | -0.80 | -0.16 | -0.48 | 8.30 | L | ||
* LEVEL: Average Rating | * RD: Rating Difference | * GAMES PRED: Games Prediction | * LV: Level / 10 | * DIF: Games Won - Games Pred | * NEW RATING: Rating + Match Rating
PRE-CONSIDERATIONS
1. Is winning a match 7/6 7/6 the same as 6/0 6/0? Likewise, is it the same to lose a close match against a Top 10 as it is versus a very low-ranked player?
1.1. What if we go beyond traditional points-based rankings and focus players' evaluations on these critical parameters: match results, opponent strength, and game differential?
2. Let’s approach this with a dynamic ATP leveling model, providing a cutting-edge perspective on tennis performance compared to the conventional Elo Rating System.
GLOSSARY
1. RATING represents the estimated level of any player on a scale from 0 to 10 (with lower or higher values in very specific cases), though it may vary depending on the surface. *We chose Schwartzman as an example, as his level differed significantly across surfaces.
1.1. LEVEL is the average rating of both players.
2. GAMES PREDICTION is the expected number of games each player is projected to win based on their ratings.
2.1. This is supported by the following premise: players compete within a range of 10 points, since larger gaps would diminish competitive balance. Therefore, a 10-point difference may correspond to an expected result of 6/0 6/0, representing the maximum competitive gap.
2.2. Matches outside this range are adjusted to a maximum difference of 10.
3. SETS COEFFICIENT evaluates match reliability based on the number of completed sets (1 set: 0.5 / 2 or 3 sets: 1).
4. GAMES WON represents the total number of games won by each player, expressed on a normalized scale from 0 to 1.
5. NEW RATING is then calculated as the sum of the pre-match rating and the match rating coefficient.
<MATCH RATING> FORMULA
LV + DIF
2
* SETS_COEF
1. The main variable is LV (= LEVEL / 10). Players compete to gain the LV coefficient through each match. Consequently, higher-ranked players experience more frequent rating fluctuations than lower-ranked players, allowing the model to adjust more quickly in cases of overrating.
2. The second variable is DIF, defined as the difference between games won and games prediction. This variable adds an extra adjustment factor, helping balance player ratings while encouraging competitiveness. Since every game counts, each match reaches a higher intensity.
2.1. However, a potential limitation may arise in singular cases where the match rating could be negative despite winning the match. In these situations, negative values can be corrected to 0.
3. Finally, SETS_COEF is the last variable, but it is also essential for rating reliability, as not all matches are concluded due to retirements or injuries.
EXAMPLE ON GRASS
SCHWARTZMAN
5/7 6/7
DRAPER
4.07
Pre-Match Ratings
6.32
39%
Games Prediction
61%
44%
Games Won
56%
3.84
New ELO Ratings
6.55