How are the forecasts calculated?
The tournament forecasts entirely depend on the difference between the two players’ Elo Ratings. The difference in Elo Ratings in plugged into a quadratic equation which was created from a quadratic regression, and then adjusted to better align with the ranking system. A 15 point difference in Elo Ratings indicates that the higher rated player has a 61% chance of winning. A 30 point difference indicates a 71% chance, a 45 point difference indicates an 80% chance, a 60 point difference indicates an 87% chance, a 75 point difference indicates a 92% chance, and a 90 point difference indicates a 96% chance.
Do the forecasts include exhaustion and/or matchup?
The forecasts are sometimes changed to factor in exhaustion and matchup but do not currently include those factors in the equation.
How do the Elo Ratings work and what is the four-part algorithm?
The Elo Ratings only take into account the quality of the two players, rather than the round or score of the match. There are four different algorithms that modify a player’s Elo Rating after he has played a match: one for winning against someone below you, winning against someone ahead of you, losing against someone below you, or losing to someone ahead of you. The algorithms consider the difference between the two players’ Elo Ratings.
Why are the rankings different?
My Elo Ratings is different than other ranking systems because it values consistency and wins against higher rated players. The ratings give more weight to Grand Slams than other tournaments.
The next step towards advancing the Elo Ratings is generating surface-specific Elo Ratings to forecast matches more accurately depending on the surface. I am also working on ways to develop the forecast equation to include other factors, such as time spent on court and the head to head record of the two players.