GPLO Model - What does that mean?
Curious about what I mean when I mention the GPLO Model? This introductory newsletter provides a description of the model and how it makes predictions.
A model for ranking MotoGP riders and making predictions
Hi! I'm Duston – A US-based MotoGP fan and professed nerd.
At the beginning of each race and throughout the season, I found myself wondering who had a good chance to score a win and who was on track to bring home the World Championship. Pundits and journalists would (occasionally) make gut-based guesses, but I was unable to find data and analytics to quantify a rider's chances of winning.
I used my expertise as a professional investor to build the GPLO Model - a predictive analytics model that uses historical data and machine learning to rank riders and forecast the outcome of individual races as well as the World Championship. Finally, a way for MotoGP nerds to unite!
If you—like myself—want to geek out on the details of how this works, keep reading. If you only want to know how your favorite rider will stack up next season, check out my website for the output from the Model.
Don’t forget to subscribe using the button below and also consider sharing. Want to tell me how much of a genius I am (or in the unlikely event that you disagree with my analysis)? Tweet me or leave a comment below.
The Model
At the risk of causing you to become cross-eyed, here is the general process of the Model:
Part 1: GPLO Ranking
The first step of the Model is to use historical race outcomes to rank riders.
To do this, I developed a proprietary adaption to the ELO ranking system. ELO was originally developed as a means of ranking chess players but has been adapted for other head-to-head competitions such as American Football.
ELO systems typically evaluate single outcome events, where a player has a specific outcome, either win, lose, or draw. The problem with MotoGP is there is a multiple person outcome (1st, 2nd, 3rd, etc.). To compensate for this, I adapted the system to compare expected rider performance to actual race results (finish position) and adjust GPLO accordingly. This skill score output, which I cleverly termed GPLO (a combination of ELO and GP – ha!) can be thought of as the current, point-in-time skill of the rider. With each race finish, a rider’s GPLO will update to reflect new results.
Don't worry--like all ELO rating systems, it is zero sum: GPLO gains are equal to GPLO losses in the same race. A closed point system makes rider scores comparable over time.
Part 2: Race Simulations
This is where things get really interesting—using race simulations to predict track-specific outcomes.
It all begins with the rider’s current GPLO – how the model thinks they stack up against the field. From here, the Model makes a number of track-specific adjustments, including:
Is the track in the rider's home country?
How has the rider historically finished at this track?
How does the bike perform at this track?
These track-specific adjustments are combined with starting GPLO to produce an adjusted GPLO or initial track ranking. Based on this adjusted GPLO, the Model uses an algorithm to calculate a race pace for each rider.
Using assumed rider pace, the model runs 2,000 simulations per race. This provides more than 1 million hypothetical laps… talk about a LOT of data! The Model uses simulated race times to rank riders and forecast the probability of each rider/position outcome.
Part 3: Point Assignment
Finally – assigning points!
The Model uses the concept of expected value to assign World Championship points. Expected value is a statistical concept that multiplies the probability of an outcome by the value of the outcome occurring. In this case, it multiplies the probability that a rider finishes in each position and the point value of that position (1st is 25, 2nd is 20, 3rd is 16, etc...). Expected points from each race are aggregated across the entire season to project out winner of the World Championship.
As races occur, forecasted results will be replaced by actual points scored and the Model will only forecast remaining races. Meaning as the season progresses, this value will more approximate the World Championship outcome.
Conclusion
At this point you probably would agree that I am a complete nerd.
If you somehow managed to get through all that without falling asleep then congrats! Otherwise you probably scrolled to the end to see the tl;dr.
TL;DR: The Model uses complex data analytics and machine learning to assess riders' abilities and runs simulations in order to make predictions on race outcomes and the World Championship.