Introducing Formula 1 Power Predictions
With this post I’m happy to announce the introduction of F1-Predictor power predictions – the culmination of a 3-year effort to systematically collect and analyze historical F1 data to identify ‘golden’ potential. You can find the power predictions on power.f1-predictor.com where you can also see their historical performance. Although power.f1-predictor.com shows the historical betting performance of my predictions, the power predictions are not betting tips. This website was created for entertainment purposes only and is meant to highlight certain predictions, not to provide sport betting tips. Betting is discouraged. I have, on purpose, hidden certain information from the upcoming race table (e.g. betting odds are not shown) – therefore, these predictions cannot be used for sports betting. Nonetheless, I will describe below how this project came to life.
When I started this blog in early 2017, I never thought it would lead me to F1 betting. To be honest, I had never bet, till 2020, even a single euro on sports. I had no idea what you could bet on F1, how you do that, let alone how to identify winning strategies. However, in 2018 I wanted to give it a try but, as an extremely risk-averse person, I wanted to be sure I would not lose money. So, I needed to do a deep analysis of the betting market and the historical odds.
I quickly found out that there were no historical F1 betting odds available. I ended up creating a few crawlers and started collecting data from several betting companies. Fast forward to mid-2020 and I had collected the betting odds for the past two seasons.
It turned out that in F1 there are several things you can bet on. From long-range bets such as the winner of the upcoming championship to race-specific things like whether a driver is going to win the race or finish in the top 3, 6 or 10 places, whether someone is going to DNF, do the fastest lap, driver head-to-head battles etc. Different betting companies would offer different markets but the most common ones were about betting on a driver finishing in the top 1, 3, 6, or 10 places in a given race. That’s why my efforts so far have been focused on these markets.
After understanding the markets, I had to make my own predictions. To this end, I created a few ML models that directly predicted the probability of each driver finishing in the top-N positions for each race. I was now able to compare by own predictions against the betting odds. The odds correspond to an implied probability. For instance an odd equal to 1.5 implies a 1/1.5=66.7% probability of the event happening.
Reading more about sports betting, I came to know that the only realistic way to make long-term gain is the so-called value betting. Simply put, when value betting, you will be placing bets that have a larger chance of winning than implied by their odds. As an example, if the odds are 1.5 (implied probability 66.7%) but the true probability, as calculated by my own ML models, is 80%, then this is a value bet.
This meant that my calculated probabilities had to be a true indication of the real probabilities of the events. In machine learning, this is called probability calibration. Although I will discuss about it in a separate blog post, I had to make sure that my models where well-calibrated.
Having calculated the ‘true’ chance of the events and the betting odds, I started coming up with various betting strategies which I was back-testing using the collected data. Some strategies focused on a high hit rate (i.e. percentage of bets won) while others focused on higher ROI. Of course, there is a whole spectrum of strategies between these two aspects. I also found out that I could get higher value if I bet on races before the qualifying and the practice sessions have been completed. This makes results more unpredictable but with a higher return.
I quickly realized that I could had made a decent ROI had I bet in the past seasons. So, I continued collecting the betting odds in 2020 and I’ve created a dashboard where I will be presenting the historical betting performance of my predictions. You can access the dashboard on power.f1-predictor.com.
Before accessing my power predictions, please read carefully the legal disclaimer and the Power Predictions Explained blog post.
Betting is discouraged.
6 thoughts on “Introducing Formula 1 Power Predictions”
I think Merc is gonna roll out the calvary tomorrow now that they can smell the gold.
Ver and Redbull did some sneaky stuff to get the pole today, but overall Ham performed better.
My feeling is that Ham is gonna win tomorrow.
BUT! The last six races on this track the pole position took the victory.
.. So honestly it’s a complete 50/50 race I think.
H vil and Merc will clutch it
Hi Asterios- this is fascinating information! This has inspired a few of my classmates and I to also build an F1 model for a data products class.
Would it be possible for you to share the historical betting odds data with us?
Happy to connect to explain our purpose in further detail.
Best,
Chris
Hi Chris,
It’s great you found my blog valuable. Can you please share via email how you are planning to use the betting odds?
Hi Stergios- thanks for getting back to me. I could not find your email address on the page, but have left mine with the comment. We are in a designing data products class and have built a model to predict top 6 and top 10 finishes for F1 races. We have built a model with high precision rates, but without historical betting data, we cannot prove if this is of any value – whether we are producing odds better than book makers.
I would be happy to share more details on our model and our methodology as well.
Im interested too in the data for the same reason, create an interesting project on F1 with python to learn the modelling and data science. I also left the email in the comment, thanks!