Upgrading the Championship Projection Model

If you followed this site last year you’ll remember the championship projections model that simulated the remaining races of the season 50,000 times and returned the probability of every driver finishing in every position in the championship. This model was rooted in each driver’s average finishing position, average track position, consistency, and influenced by how many races had already been completed. For this season I have completely overhauled the projection system and I am going to run through a few of the major changes in this article.

The first major change is that championship projections and race simulations are now based on each driver’s Elo rating. Elo ratings are a ranking system I applied to IndyCar that show the relative strength of drivers and can be used to simulate races, matchups, and more. For a more in depth explanation of IndyCar Elo ratings, I encourage you to read this article. Elo ratings are the base of the system that inform how likely drivers are to win races or simply beat another driver in a race.

Every remaining race of the season is simulated using a Monte Carlo method. Points are awarded according to finishing position in the race just as they are in real life, including all double points races. At the end of the simulated season, all of the points earned by each driver are added to the points they already have in real life and the championship rankings are determined from that. This process is repeated for many seasons and all of the results are added up to estimate the probability of every driver finishing in any place in the standings.

There is another interesting change to the championship projections that I feel will result in a much more accurate system overall. The first race of each season is simulated using the most current Elo ratings for every driver. After the race, the Elo ratings are updated based on the results of that simulated race. This means that the Elo ratings will change throughout the season in order to reflect drivers getting better or worse. Overall, this results in much more variability in the projections which accounts for drivers going on streaks and getting better or getting into a slump and having their rating negatively affected. This is called running the simulation “hot” as the ratings are constantly updated and used for future races. At the end of the season, the Elo ratings are reset to the true current ratings for the start of the next simulated season.

Right now, the bonus points for earning pole position, leading a lap, and leading the most laps are not simulated. Since they are only 1, 1, and 2 points respectively, I am not too worried about not adjusting for these bonus points right now, but that might change as the end of the season gets closer.

This was a pretty short explanation of the major changes to the championship projection model, but if you would like more information or have a question please leave me a comment below or contact me on Twitter @thesingleseater.