_02.jpg){kind=link}
Let’s start off with a couple of puzzles. I realize they aren’t the sort of things that keep most of you up at night, but they were odd enough to drive me to a flurry of coding, data analysis, and now blog writing.
On Wednesday, Ugo Humbert lost his first-round match in Rotterdam to Emil Ruusuvuori. It marked an unceremonious end to a hot streak for Humbert: He not only won the title in Marseille last week–launching himself into the Elo top ten–but he strung together 31 consecutive holds. 1,000 kilometers north, on a different indoor hard court, he got broken twice by a man ranked outside the top 50.
That’s the first puzzle: Why did the Frenchman lose? Again, it’s not that odd, as my Elo ratings gave Ruusuvuori a one-in-three shot to pull the upset. But it’s a match that Humbert should have won.
Head-scratcher number two: Why does Humbert always lose to Ruusuvuori? Wednesday’s decision marked their fifth meeting, and the Finn is undefeated. While the outcome is always close–Rotterdam was their fourth deciding set, and the other match went to two tiebreaks–the results are starting to get boring. Ruusuvuori is a solid player, and he is consistently able to blunt the Frenchman’s serve. But five in a row?
The answer to both mysteries is the same, and it’s more satisfying than I expected. Rotterdam is unusually slow for a hard court, especially indoors. Like most (or perhaps all) of the previous Humbert-Ruusuvuori venues, it plays slower than tour average. Just as important, Humbert’s game is unusually sensitive to surface speed. While that isn’t always true of big servers, he stands out as a fast-court specialist. We couldn’t have confidently predicted a Finnish upset, but we could have guessed that the Marseille champion would find this week’s tournament tougher going.
Rotterdam, it’s slow
The last time I published surface speed numbers, in late 2019, Rotterdam rated as the slowest indoor hard court on tour. Adjusting for the mix of players at the event, there were 10% fewer aces at the tournament than expected. It was a sharp decline from 2017 and 2018, when the venue sported more typically speedy indoor conditions.
Since then, the results have remained similar. Last year, the rate was 5% lower than expected, roughly tied with Stockholm as the slowest indoor surface on tour. Marseille, by contrast, gave players 12% more aces than usual.
There are limitations to using aces as a proxy for surface speed; I use aces because it’s the most relevant data that is widely available. Still, while you can quibble about the methodology or about a specific tournament’s place on the list, the overall rank order seems about right. Aces–adjusted for each event’s field–tell you much of the story.
With a growing mass of Match Charting Project data, we can do a little better. We have shot-by-shot logs for over one thousand matches since 2021. To compare conditions, I used my Serve Impact metric, which estimates how many points a player wins, directly or indirectly, because of his serve. It counts aces, other unreturned serves, and a fraction of the service points that take longer to decide. Depending on your motivation in measuring court speed, this isn’t perfect either: It doesn’t directly tell you anything about bounce height, for instance. But if you want to know what sort of players a tournament favors, Serve Impact gets you close.
By this more sophisticated metric, Rotterdam is… still slow. The venue takes away 4% of the points a player typically earns from his serve. Marseille and Montpellier each swing 7% in the other direction, Stockholm and Vienna provide a modest 3% boost, and Basel adds 8% to the server’s punch. With the exception of the short-lived tour stop in Gijon, Rotterdam has been the slowest indoor hard court of the 2020s. Even the clay in Lyon plays faster.
Here are the Serve Impact adjustments for the tournaments best represented in the dataset. Higher numbers mean faster conditions with more points decided based on the serve:
Tournament ServeImpact
Stuttgart 1.29
NextGen Finals 1.20
Tour Finals 1.16
Wimbledon 1.11
Shanghai Masters 1.11
Halle 1.10
Queen's Club 1.08
Basel 1.08
Washington 1.08
Dubai 1.07
Tournament ServeImpact
Antwerp 1.05
Gstaad 1.05
Australian Open 1.04
Davis Cup Finals 1.04
Cincinnati Masters 1.04
Paris Masters 1.03
Vienna 1.03
Miami Masters 1.02
Madrid Masters 1.01
US Open 1.01
Tournament ServeImpact
Canada Masters 1.00
Rotterdam 0.96
Indian Wells Masters 0.95
Rome Masters 0.92
Acapulco 0.87
Barcelona 0.87
Roland Garros 0.83
Monte Carlo Masters 0.83
Average Serve Impact is around 34%, so the 4% hit in Rotterdam knocks that down to about 32.6%. Humbert has an above-average serve, so the slow-court penalty is greater still. He isn’t going to win any awards for rallying prowess, especially against someone as sturdy as Ruusuvuori, so the points that he doesn’t secure with his serve will disproportionately go against him.
The first three meetings in the Humbert-Ruusuvuori head-to-head were on clay, at Roland Garros, Madrid, and Rome. The fourth came on grass, at ‘s-Hertogenbosch. It rates a bit faster from 2021-23 than Halle or Queen’s Club by the Serve Impact metric, though it rated as the slowest grass court on tour last year by my older ace-rate algorithm. Maybe it was less server-friendly in 2023, just in time for Humbert to be flummoxed once again.
Surface sensitivity
We tend to take for granted that players are suited to conditions in predictable ways. Big servers like fast surfaces, right? Broadly speaking, yes, but it’s not a hard-and-fast rule. Bounce height makes a difference, footwork matters, and some players are just more comfortable on some surfaces than others.
Armed with surface speed ratings, this is something we can test. If a player is particularly sensitive to conditions, each tournament’s Serve Impact rating should have a predictable influence on his match outcomes. I tried that for all tour regulars, controlling for player strength by using overall Elo ratings at the time of each match.
The resulting numbers are an abstraction on top of an abstraction, so they’re a bit difficult to get your head around. I’ve tried to simplify matters by rendering them in terms of Elo points. A player who is very sensitive to surface and does better on hard courts is, effectively, a better player in faster conditions. The ‘Sensitivity’ numbers given here are the benefit–denominated in Elo points–of each single percentage point that a surface is faster than average. For players who like it slow, negative numbers express the same idea, the Elo-point advantage of a one-percentage-point slowdown.
Here is the list of all players with at least 100 tour-level matches since 2021, plus Rafael Nadal:
Player Sensitivity
Tallon Griekspoor 11.1
Ugo Humbert 9.5
Richard Gasquet 9.1
Novak Djokovic 8.7
Adrian Mannarino 7.9
Sebastian Korda 4.9
Jordan Thompson 4.3
Matteo Berrettini 4.0
Aslan Karatsev 3.6
Tommy Paul 3.4
Marcos Giron 2.9
Marton Fucsovics 2.9
Marin Cilic 2.6
Felix Auger-Aliassime 2.1
Hubert Hurkacz 1.7
Player Sensitivity
Frances Tiafoe 1.6
Carlos Alcaraz 1.4
Emil Ruusuvuori 1.3
Brandon Nakashima 1.3
Cristian Garin 0.6
Alexander Zverev 0.5
Alexander Bublik 0.5
Ilya Ivashka 0.0
Arthur Rinderknech -0.1
Taylor Fritz -0.3
Jan Lennard Struff -0.3
Lorenzo Sonego -0.4
Mackenzie Mcdonald -0.5
Andy Murray -0.9
Grigor Dimitrov -1.1
Player Sensitivity
Roberto Bautista Agut -1.2
Alex de Minaur -1.2
Karen Khachanov -1.4
Jannik Sinner -1.4
Yoshihito Nishioka -1.5
Miomir Kecmanovic -1.9
Andrey Rublev -2.2
Daniel Evans -2.2
Cameron Norrie -2.5
Holger Rune -2.9
Roberto Carballes Baena -3.0
Botic van de Zandschulp -3.1
Daniil Medvedev -3.4
Denis Shapovalov -3.5
Sebastian Baez -3.7
Player Sensitivity
Laslo Djere -4.1
Dusan Lajovic -4.1
Pablo Carreno Busta -4.4
Jaume Munar -4.6
Fabio Fognini -4.8
Nikoloz Basilashvili -4.9
Casper Ruud -5.0
Diego Schwartzman -5.4
Francisco Cerundolo -5.9
Alexei Popyrin -6.4
Albert Ramos -6.8
Rafael Nadal -9.9
Alejandro Davidovich Fokina -10.1
Stefanos Tsitsipas -10.2
Lorenzo Musetti -11.2
There’s Ugo! He’s not quite as surface sensitive as Tallon Griekspoor, but a couple of points is within the margin of error. A sensitivity rating of 9.5 means that Humbert is about 100 Elo points worse in Rotterdam than he is Marseille, as long as I’ve accurately estimated the server-friendliness of the respective playing conditions. Ruusuvuori may also like it faster, but only marginally so; he’s effectively neutral.
Keen-eyed readers may have noted that I earlier referred to “overall” Elo. I’m not using surface-specific Elo ratings here, because I don’t want to adjust for surface twice. Surface-specific ratings already capture some of this: Humbert’s hElo (for hard courts) is 120 points higher than his cElo (for clay courts), which tallies reasonably well with these more fine-grained distinctions. What hElo and cElo can’t tell us, though, is how much his (or anyone else’s) performance will vary on the same surface, depending on the conditions at each specific venue.
It’s easy to get lost in the weeds of Elo-based forecasting calculations, but it’s important to remember they are just tools to help measure a real-world phenomenon. Not every big server is equally at sea on clay; some dirtballers are less dependent on slow conditions than others. Small differences in surface speed are, for most matchups, a minor consideration. But for some players, conditions matter a lot. Ugo Humbert likes his surfaces fast, as much as almost anyone else on tour. In Rotterdam, the conditions did not cooperate.
* * *
Subscribe to the blog to receive each new post by email:
Do you think Ruusuvuori will also upset Griekspoor in Rotterdam?
My Elo forecast had Griekspoor at 50.9% going in, so it wasn’t much of an upset either way, but yeah, these numbers would have tilted things in Emil’s favor. Doesn’t look as insightful after today’s match though…
Hi Jeff, I wonder how this current tournament in Los Cabos shakes out regarding speed? I’m not sure what the court is built on, or what lies beneath it, but, wow, it’s playing as if the players are on clay, with huge bounces. Just curious what your numbers tell you.
yeah, ace rate says it’s slow — about the same as Madrid Masters, a tiny bit faster than Indian Wells (which always rates as very slow for hard)
Interesting to see the spread of 2 handed backhands in the court speed sensitivity table.
I saw Tsitsipas and Musetti at the bottom and thought “That makes sense as they need more time on the backhand side”, then I saw Dimitrov a bit higher up and thought “Well he has the big serve which will be better on fast courts and the slice when he is rushed”.
But then what is Gasquet doing at +9.1, I wonder what it is about his game that makes the difference in him actually doing better on faster courts?
yeah, I had a similar reaction seeing tsitsipas and musetti there, good point about gasquet. It might just be a fluke — these are all numbers from the last few years, so maybe it isn’t representative of Gasquet’s whole career. But it might be that he really does prepare better/faster than other one-handers do.
Is ace rate per tournament publically available anywhere?
Not that I know of
This is awesome