About the model
What is the underlying framework of the RatingTennis model?
The RatingTennis model is a modified version of the popular Elo rating system which was originally conceived as a method of rating the skill of chess players.
The Elo rating system is designed to assess the relative skill level of players such that it is possible to compare two players at any point in time even if they have never faced one another. This is done by assigning each player a rating score that dynamically adjusts after each game depending on whether they won or lost and the strength of the opponent they faced. How much each rating adjusts depends on how much of a surprise the result was to the rating system. For example, if a much stronger player defeats a weaker player, then the ratings for each player will not change drastically as the system views this as a predictable result. However, if instead that weaker player defeated the stronger player, the system will view this as an unpredictable result. Thus, the system will correct for this in the future by the assigning the weaker player a significant rating increase while assigning the stronger an equivalent rating decrease.
More information about the Elo rating system can be found here.
How does the RatingTennis model differ from a normal Elo rating system?
There are three key areas where the tennis-ratings model improves on the normal Elo rating system:
1. Assigns a separate rating for court surface type (Hard, Clay Grass)
Court surface can play a big part in tennis depending on the player and the match-up. An anecdotal example that will be familiar to most tennis fans would be Rafael Nadal and Roger Federer in their prime years. It wouldn't be any surprise if we told you that Nadal would be a significant favourite on Clay whereas Federer would be a significant favourite on Grass .. but by how much? The model answers this question by constructing separate rating scores for each court surface to reflect each players relative strength on each surface type.
2. The speed of ratings adjustment is a function of number of matches played
In tennis, the player base is always changing with a large number of players entering and retiring from the professional circuit every year. Thus, it is important to be able to be able to accurately rate new players as quickly as possible. The RatingTennis model accounts for this by assigning a rating change factor that starts off high for new players but decreases as the number of matches played increases.
3. The ratings adjustment depends on how much a player wins or loses by
The standard Elo rating system treats all wins and losses the same no matter how they occurred. However, tennis has a very unique scoring system where quite frequently the player that is playing the ‘better’ tennis may not win. A couple of lapses of bad tennis in a game could be the difference between winning and losing the match. The RatingTennis model assesses the magnitude of how much a player wins or loses by and incorporates this assessment into the ratings adjustment.
Are we able to see the underlying methodology?
Absolutely. The purpose of this site is to provide and encourage analysis of one of the most statistics rich sports in the world. To this end, the intention is for the RatingTennis model methodology to be fully up on the website shortly for those that are interested (formulas and all).
Can the model be used for betting?
Yes it can. The model was originally created for betting purposes with moderate success. However, tennis like most individual sports is more prone to individual factors that the tennis-ratings model does not take into account. For example, if a player is suffering from an injury but still plays, the model won’t take that into account. Using a combination of the model and qualitative factors has been something that has been very successful for the creator.