Past performance is no guarantee of future results... or is it?
Punters using statistical data to inform their betting face an inherent risk. This kind of data is, by its nature, a representation of what has happened in the past and, as the warning to any would-be investor goes, is no guarantee of what will happen in future.
Thankfully, though, whilst there are no guarantees, it seems the warning is overly cautious when it comes to tennis, as historical data seems to do a decent job of informing us of the relative likelihood on players winning future matches.
Players who win more, against better players, are more likely to go on to win again. Quantifying the value of matches won and lost takes some work (I've created and had to maintain an Elo ratings database, for example), but is within the capabilities of anyone who can use a spreadsheet and has some time on their hands.
We don't live in a statistically pure world
There are still times, though, when any stats-informed punter should be wary.
If tennis players lived in a hermetically-sealed world, where they had no interactions outside of the matches they played, my Elo ratings would be predictively infallible. Unfortunately (for me at least), players have lives outside of their tennis matches.
They get food poisoning, have affairs, get injured, and get divorced (although not always in that order). In short, all manner of unwanted life-pathogens can invade our world of sanitised statistics, making a mockery of them.
Mature markets and profitability
With these allusions to infectious disease, you can probably see where this is going.
In recent seasons my tennis profitability has reduced. Not dramatically, but noticeably so. I have put this down, in part, to market maturation. As the average operator in a market becomes more sophisticated - which, in financial markets at least, seems to happen over time - there will inevitably be a contraction in margins, as more participants compete to offer and take profitable positions.
I'm convinced, though, that a more significant factor in my reduced profitability has been the circumstances surrounding the upper echelons of the men's game in which I specialise. The aging of a dominant trio - who remain assertive but less reliably so - alongside a series of high-profile personal dramas and prolonged absences for various top-flight players - has constantly muddied the picture. It's rare, in recent seasons, to approach a tournament without some doubt around whether the numbers are to be trusted.
The season opener amid a pandemic
Enter the Australian Open 2021.
A global pandemic, which saw the 2020 tennis calendar decimated, means that the natural doubts that always surround the first major of the season (not knowing how the players have spent the preceding three months) are magnified.
Then there are the quarantine arrangements, which have seen players partially confined to their rooms, posting videos on social media of creative training regimes designed to supplement the five hours of allowed outside practice.
Against this backdrop, how can any stats-informed punter bet with any confidence?
Take confidence from the Brier number
It might not all be unwelcome news, however.
Measuring the accuracy of your betting model is good practice for any punter. Most will do this by monitoring their profit and loss. However, whilst this provides a valid indication of your overall betting effectiveness, unless you bet on every possible event within your chosen sport, it will not necessarily tell you how good your underlying statistics are.
For this reason, at the end of each tennis season, I always perform an autopsy on my tennis betting that includes calculating the Brier score for my Elo ratings.
A Brier score is a statistical tool that measures the accuracy of any probabilistic prediction. If my Elo ratings suggest ten players have a 60% chance of winning their matches, do six of them go on to win? A Brier score effectively asks this question for every tennis match for which I produced ratings over a season.
The impossible aim with Brier is to achieve a score of zero, which shows that there is no difference between your predictions and actual results. Over the years, my Brier score settles at around 0.22, with a range of around 0.15 to 0.30 tournament to tournament.
Historically, my Brier score for the Australian Open has been higher than the season average, no doubt accounted for by the natural inconsistencies introduced by it being the season opener.
However, amazingly (it was so unexpected, in fact, that I went back multiple times to re-run the analysis, convinced there must be an error), my Brier score for the end of the 2020 season was almost identical to my long-term average of 0.22.
The men's game, it seemed, hadn't been changed by the pandemic: the players had maintained their relative Elo form.
What does this mean for the Australian Open?
Well, usual season-opener nervousness aside, if that Brier score from last season is to be believed, the Elo numbers probably give us the best indication of what is likely to happen.
And the Elo numbers tell us that the top of the men's game has not been this compressed for two decades, with seven players covered by a mere 200 Elo points. To make these numbers more concrete, that means that if the best of those seven players, Novak Djokovic, played the worst, Stefanos Tsitsipas, the former would be the 1.341/3 favourite to the latter's 4.003/1.
Much will depend on the draw, which at the time of writing has yet to be made, but given these squeezed ratings, Djokovic looks poor value at 2.6413/8 to collect his ninth Australian title.
For all his inconsistencies, Rafa Nadal is interesting at 7.4013/2, as is the up-and-coming Andrey Rublev (40.0039/1), who closed the 2020 season with quarter-final appearances in both the French and US Open, and by my reckoning is now the fifth-best player in the world.