Tuesday, February 2, 2016

Estimated Dynamic NTRP Ratings for the ATP - How much higher rated is Novak than you?

I've toyed with calculating my Estimated Dynamic NTRP Ratings for professional tennis in the past, calculating ratings for just the 2014 men and women at the Australian Open a few years ago.  But my roots in calculating ratings for professional tennis go back to my teen years when I was calculating ratings for players from just their play at the Grand Slams.  But I digress.

More to the present, I am now calculating what ATP player's dynamic NTRP ratings would be from their play in ATP (main draw and qualies), Challenger, and Futures events going back a number of years.  Given the differences between league play and tournament play at the professional level, I would not say these ratings are perfect by any stretch, but they are interesting to look at.

The ratings below are a little preliminary, I'm still fine tuning things, but wanted to share some initial data right away for people to see and comment on.

Before we look at some ratings, a few caveats.

First, these ratings are just from singles results, no doubles matches are included.

Second, the relative difference between players listed is correct/accurate, but since this set of data is not connected to league players, the difference between ratings I calculate from league play and these is a guess.  What I did was to try to ensure that the top-400 or so were considered 6.5 or higher since the USTA guidelines state a 6.5 or 7.0 is considered world-class.

Third, the dynamic NTRP algorithm is very biased towards your most recent results.  So a high rated player that has had a couple unexpected losses due to injury, playing their worst surface, etc. may have a lower rating than you'd expect.  Similarly someone who has pulled one of those upsets or is playing on their best surface may be rated higher than expected.

Fourth, these ratings include results through the end of 2015, but do not include the Australian Open or other January tournaments.

With that said, on to the ratings!  Here are the current top-20.

1Novak Djokovic7.07
2Roger Federer6.94
3Rafael Nadal6.85
4Andy Murray6.78
5Stanislas Wawrinka6.71
6Tomas Berdych6.70
7David Ferrer6.69
8Marcos Baghdatis6.68
9Kei Nishikori6.66
10Marin Cilic6.64
11Hyeon Chung6.63
12Steve Johnson6.63
13Joao Sousa6.62
14John Isner6.61
15Richard Gasquet6.60
16Jack Sock6.60
17Milos Raonic6.60
18Gilles Simon6.59
19Roberto Bautista Agut6.59
20Jo Wilfried Tsonga6.58

No surprise who is rated the highest, and Novak is #1 is by a pretty healthy margin of 0.13 based in part to Novak's win in the final at the Tour finals.  Federer and Nadal are next on the strength of their Tour Finals final/semi, but there is a 0.09 gap between them, another 0.07 to Murray, and 0.07 again to Wawrinka.  After that it does get a lot closer.

It is surising to see Baghdatis and Chung so high, but Chung did have some good results to end 2015 winning a Challenger and making the quarters at Shenzhen playing Cilic close, and while not included in the above he beat Groth in straights and played Cilic very close again in Brisbane before a respectable, albeit straight sets, loss to Novak at the Aussie Open.  Baghdatis had a run to the semis in Stockholm to end his 2015.

By the scale set by the USTA, Novak would actually be a "7.5" while the rest shown would be 7.0s.  Again, this is somewhat arbitrary given where I starting things out.  But the gap between players is representative of the algorithm and you see there is nearly a full level (0.5) between #1 and #20 which I would not have expected.  There is just 0.13 between #5 and #20 though which shows the separation of the top-4 from those behind.

It will be interesting to see how these change after the Australian Open and other January events are incorporated.  Because of the third caveat above, players can move up or down a fair amount with a few better or worse than expected results, see Baghdatis and Chung above.

Stay tuned for more.

Attribution: Match data is courtesy Jeff Sackmann / Tennis Abstract.