Long-held beliefs are hard to shift, writes Jack Houghton, who has been reassessing his reliance on surface-specific ratings ahead of Wimbledon
"For a while now I've started to doubt the value of knowing how a player has fared on different surfaces..."
Elo doubts begin to surface
It's no surprise to regular readers of this column that I'm a fan of Elo ratings, having used them for several seasons to trade profitably on the men's game. Of particular use have been the surface-specific ratings I keep. Being able to filter player performance according to whether the Elo points they have gained have come on grass, clay, or hard courts has given me - or so I thought - an important edge.
Beginning the analysis for this year's men's singles at Wimbledon, then, my first action was to compare overall ratings to grass ratings. That analysis was plagued by disquiet, though. Because for a while now I've started to doubt the value of knowing how a player has fared on different surfaces.
Beware our childhood biases
If I'm honest, thinking back all those years ago, when first putting together my ratings system, I can't remember what my rationale was for including the ability to filter by playing surface. Perhaps, growing up watching tennis in an era where the likes of Becker, Sampras and Edberg dominated at Wimbledon, and a bunch of Spanish and South American players won the untelevised French Open, I learnt that the grass game and clay game might as well be different sports, and so constructed my ratings accordingly, without questioning whether it was necessary.
We probably all have these biases, that are based on what we first learned when becoming sports fans, that we need to be wary of. For me, I still start my analysis of the Cheltenham Gold Cup with the adage ringing in my ears that it is a race won by out-and-out stayers. I still assume that winning the toss in the Boat Race gives one crew an advantage over the other. And I still look first at Aston Villa's odds in the FA Cup first, because my uncle always said to me that it was their lucky competition.
Even though, as an adult, I've rationally discounted these boyhood folk-punting nonsenses, and they no longer influence the bets I place, it's striking that they still rattle about in my brain, yelling for attention. Which makes me wonder how many others are clattering about in there, shouting their mouths of, that I haven't yet learnt to ignore.
Or, worse still, which ones are masquerading as sensible, cogent thoughts - like the idea that surface is a principal factor in assessing a tennis player's form - that will take some serious sleuthing to root out.
Losses provide invaluable feedback
The wonderful thing about punting, of course, is that it's easy to get honest feedback on the quality of your thinking. Long-term profit means you're broadly getting things right; long-term losses suggest you're not. And here's the thing that it's hard to admit: for the last four years, whilst still profitable, my tennis betting margin has been steadily diminishing.
There could be several reasons for this - more competitive markets, with more savvy punters, pushing margins down, for example - but being professional in how you approach betting means constantly reassessing your methods to check that they are sound.
Is the playing surface irrelevant?
And writing a preview for the French Open, where I once again suggested the waning dominance of the Big Three, I had what to many will seem an obvious thought: if Federer, Nadal and Djokovic have managed to be dominant across all four majors, for so long, then maybe this has less to do with their versatility, which had always been my assumption, and more to do with the different surfaces no longer being played in a way that made them all that dissimilar.
To test this, I did two pieces of analysis.
First, I took the last eight years of results in ATP events from my database and compared these to the predicted results for those matches according to my ratings, both the overall Elo and the surface-specific one. I did this by grouping predicted results into average percentiles. I then charted the results on to a scatter diagram - predicted results (both overall Elo and surface-specific Elo) versus actual results.
If my ratings had been perfect, the lines would have drawn themselves on top of each other. Or put another way, looking at all the times my ratings had predicted a player to have a 60% chance of winning a match would have resulted in those players, on average, winning 60% of the time. And so on, up-and-down the percentile bands. The bigger the distance between the lines, however, and the worse my ratings would have been shown to be.
The death of surface-specific ratings
Here's the thing. In the analysis described above, the overall Elo rating performs well. It slightly undervalues favourites and overvalues outsiders, but not significantly. The surface-specific ratings, though, are all over the place. I might as well have been doing a dance to the punting gods and praying they impart their betting wisdom, rather than wasting time analysing court surface. Symbolic sacrifices to watching deities may have been more successful for picking winners.
Rallying round the idea that the majors are all the same
Second, I nicked a spreadsheet from a friend who has analysed the rally length at the four majors over time. The data is a bit crude, because there is less data available the further you go back, but it nonetheless shows something interesting. When I was growing up watching Sampras and his brethren, I was right to think that the grass game was different. At the time, the average rally length hovered around two shots, with rallies at the French Open lasting three-times as long.
Nowadays, however, there's hardly any difference. At the US Open, rallies last around five-and-a-half shots on average. At Wimbledon, it's a little over four. The other two majors lie in the middle of these extremes. So, for all the obvious visual differences between the surfaces, the tennis we see is largely the same, when it comes to rally length, at least.
Now that's a fairly rudimentary bit of analysis, and there is a debate to be had about what has caused the convergence between the different surfaces. Whatever the outcome of that debate, though, it seems clear that spending time tinkering with surface-specific ratings isn't going to help much when it comes to bet selection.
Ignore what's happened since Paris
And it's also why getting wrapped up in debates about what the top players have done since the French Open is fairly pointless, too.
In the Wimbledon previews I've read, much is being made of the choice made by Djokovic, Nadal and Thiem to rest from competition after their exertions in Paris. Whilst they have been convalescing, Federer has reappeared, resplendent as always with the backdrop of a manicured green court, strolling to his 10th Halle title, with only Tsonga asking him any searching questions.
This time in the media spotlight has caused a tightening of Federer's Wimbledon odds. He's now a 4.407/2 shot. Looking at his grass-only Elo rating, those odds look value, as Federer is only a few points below Djokovic, and far clear of Nadal.
However, as we've seen, surface schmurface. Looking instead at the overall Elo ratings, a more complex picture emerges, with Federer and Nadal now virtually level with world-leader Djokovic. Had Federer avoided those two in his half of the draw, I would have made him favourite to take his ninth Wimbledon title, but as we've seen before, draws can make or break a player's chance of tournament success, and a likely Federer-Nadal semi-final collision makes me wary of supporting either of them.
Djokovic is blessed by the draw
Changing our long-held beliefs is difficult, and even if we think we are taking an analytical approach, sometimes the choices we make in our analyses are flawed because of the very assumptions upon which we base them. For my part, I assumed that surface was important, because it always had been in my youth. Djokovic, the rightful 2.506/4 favourite with a favourable draw, is likely to prove that it isn't any more.
Back Novak Djokovic @ 2.506/4