What is Elo in tennis, and why it beats the rankings

Know who'd win, not who's ranked first.

Key facts

1
On the 2019 Davis Cup finals, Elo picked the winner in 72 % of matches and the ATP ranking in 60 %.
Jeff Sackmann, Tennis Abstract
2
In a benchmark of 11 prediction methods on 2,395 ATP matches from the 2014 season, Elo was the only public method competitive with the betting market.
Kovalchik (2016), Journal of Quantitative Analysis in Sports
3
In Elo, a 400-point rating gap corresponds to odds of about 10 to 1 for the higher-rated player.
4
FiveThirtyEight's tennis Elo ran on more than 265,000 matches since 1968, with K shrinking as a player's matches accumulate.
5
By Tennis Abstract's method, Nadal's clay Elo of about 2,550 in 2009 is the highest single-surface rating.
Tennis Abstract Elo reports
6
Peak career Elo by Tennis Abstract's method is about 2,525 for Djokovic (2015), about 2,524 for Federer (2007) and about 2,489 for Nadal (2013).
Tennis Abstract Elo reports

Why it works

Elo gives every player one number. The gap between two numbers becomes a win probability through a curve: the expected score for player A is 1 / (1 + 10^((R_B − R_A)/400)). A 400-point gap is about 10 to 1.

After the match, the surprise is what moves the ratings. The new rating is R + K × (S − E), where S is the result and E the expected score. An expected win moves little, an upset moves a lot, and K sets the scale. A high K makes ratings jumpy, a low K makes them smooth.

Surface-specific ratings fix an obvious flaw. A clay specialist and a grass specialist are nearly different players depending on where they play, so a surface Elo reads each surface on its own.

The rankings answer a different question. A player back from six months out still holds points from titles won before the layoff. Elo already knows what the matches since then said. With a 300-point surface Elo gap, the formula makes the higher-rated player about an 85 % favourite, whatever the ranking says.

The line to say on court
“The ranking says who earned the most. Elo says who'd win.”

Drill: The draw test

30 min · alone

Pick ten first-week matches from any tournament draw. Predict each winner from the ranking alone.
Predict each match again from the Tennis Abstract Elo report, using the surface rating where it's shown.
After the round, score both sets of picks.

Number to beat7 of 10 Elo picks correct, and at least one more than the ranking

Faults and fixes

Using overall Elo on clay

The surface changes who a player is.

FixRead the surface rating when it exists.

Reading a small gap as certainty

A 50-point gap barely makes a favourite on the curve.

FixConvert the gap through the curve before you call it.

Treating Elo as the verdict on the greatest player

It has one axis, no matchups and drifts across eras.

FixSay "peak by this method", never "best".

Questions players ask

Is Elo better than the ATP ranking for predicting matches?

On the evidence here, yes. Elo called 72 % of 2019 Davis Cup finals matches right against 60 % for the ATP ranking, and in Kovalchik's 2016 benchmark it was the only public method competitive with the betting market.

What does a 400-point Elo gap mean?

It means the higher-rated player is about a 10-to-1 favourite.

What is K in Elo?

K sets how far a rating moves after each match. A high K reacts fast and looks jagged, a low K is smooth. Some systems shrink K as a player's match count grows.

Who has the highest Elo rating in tennis history?

By Tennis Abstract's method, Djokovic peaked around 2,525 in 2015, just ahead of Federer at about 2,524 in 2007. Nadal's clay rating of about 2,550 in 2009 is the highest on a single surface. Peak values drift by method and source.

What this doesn't cover

Peak Elo values differ by method and source, so quote them as "by this method". K and surface blends are tuned, not derived. Elo assumes skill runs on one axis, so it can't see that matchups don't always follow the ratings, and it doesn't settle who was the greatest.

Sources

Jeff Sackmann, "An Introduction to Tennis Elo" (Tennis Abstract, 2019)
Kovalchik (2016), Journal of Quantitative Analysis in Sports
FiveThirtyEight, 2016 US Open forecast methodology
Tennis Abstract Elo reports

Know the gameWhat makes a big point in tennis: win probability and leverageTennis win probability is each player's live chance of winning the match, computed from the score and how often each wins points on serve. Leverage is how much one point moves that chance, and it's what makes a big point. A 30–40 break point at 5–6 in a deciding set carries huge leverage. 40–0 at 3–0 carries almost none.Know the gameWhich tennis stats actually predict who winsThe tennis stats that track the result are the quiet ones: second-serve points won, what happens by the fourth shot, and the few points with real leverage, like break points. Aces, winners and errors describe the last shot, not why the match was won. About two points in three end within four shots, and the box score never tells you that.Know the gameWhere tennis stats come from and when to trust themTennis stats come from a pipeline: tracking cameras like Hawk-Eye measure the ball, analysts turn the measurements into metrics, and open datasets let others check the work. To trust a stat, find its instrument, its denominator and the person or institution behind it. If you can't find all three, the number is a claim, not a measurement.Know the gameWhat made Federer, Nadal and Djokovic so goodFederer, Nadal and Djokovic shared one engine, the chain from hips to trunk to arm, and built three different answers on it. Federer took the ball early and stole time. Nadal hit a heavily spun left-handed forehand that landed high on a right-hander's backhand. Djokovic returned early and deep, turning the server's weapon into a neutral rally.