Introduction to Tennis Elos

Tennis Pilot

Tuesday, December 17, 2019 11:26 AM UTC

Tuesday, Dec. 17, 2019 11:26 AM UTC

Elo is a superior rating system to the standard ATP and WTA ranking formulas
<p>Anyone who follows tennis will be aware of how the ranking system works. Every tournament has a certain amount of ranking points on offer. Players are ordered from 1 to over 1500 based on who has earned the most ranking points over the last 52 weeks. Nice and easy.</p><p>However, this ranking system is very simplistic. The only metrics measured are matches won and tournament level. In other words, who you play is ignored, the most important metric of all. A first round win over the world number 1 is treated the same as a first round win over the world number 1500. Surely it makes sense to take individual player quality into account? Enter the Elo ranking system.</p><p>Designed by Arpad Elo, the Elo system is based on statistical estimation. As an article from Betfair states, ‘Elo works by assigning a rating to each player. When two players play each other, the players’ win probabilities are given by their difference in Elo rating’. The Elo Wiki page notes: ‘A player's Elo rating is represented by a number which may change depending on the outcome of rated games played. After every game, the winning player takes points from the losing one’. The higher a player’s Elo, the more points gained if you win. Conversely, the lower a player’s Elo, the more points you lose in defeat. In other words, you gain or lose points relative to your opponent’s ability.</p><p>The key question is simply ‘does it work?’. The answer is yes. In fact, the Elo system works very well compared to most tennis rating systems. Stephanie Kovalchik analysed 11 different models and found that the Elo system was best, with a 70% accuracy in predicting match outcomes on the ATP tour in 2014. For comparison, the bookies odds correctly predicted 76.1% of matches at the US Open in 2014 compared to 73.5% for the Elo, so it was not far off.</p><p>To avoid confusion, I should note that there isn’t one standard ‘Elo system’ used by all. The underlying concept is shared across all Elos but the whole edge in match prediction tends to come from customisation. Most notably, an aspect called the K factor is pivotal as it determines how much the Elo can change following a match result.</p><p>I won’t go into depth concerning the maths of how an Elo is calculated here given there is a wealth of literature on the topic online. Most notably, Jeff Sackmann of Tennis Abstract has done several deep dives into tennis Elos and has a regularly updated set of tennis Elo rankings created using his own Elo.</p><p>One of Jeff’s most interesting stats is peak all time Elo. No prizes for guessing the top 3. The highest recorded Elo in men’s tennis is Novak Djokovic in Miami 2016, followed by Federer in Dubai 2007 and Nadal in Madrid 2009.</p><p>It may surprise readers that several other areas which are not taken into account include match up, tournament, winning margin, fitness and surface. It is impressive that the Elo system can produce such excellent results without being sensitive to metrics that we would normally see as crucial.</p><p>The bookies still have an edge over a simple Elo, which is unsurprising. Their algorithmic pricing will take many more variables into account and surely does use some form of opponent sensitive ranking system at its heart. Nonetheless, for anyone serious about making an automated betting strategy, it seems likely that a modified Elo is the way to go, especially one which takes those factors – match up, tournament, winning margin, fitness and surface – into account.</p><p><a href=";a=f339dce1-88de-4994-ab6a-c24e019bdf3b&amp;f=3"><img alt src="" style="width:100%;height:38px" /></a></p>
comment here