cannonstats.com - Adam Rae Voge, Scott Willis
Breaking down and ranking the 7 possibilities.
statsbomb.com - Matt Edwards
I love analyzing Offensive Linemen through the lens of StatsBomb engagement data. No proxies, no guesswork, real-boy data.
statsbomb.com - StatsBomb
We thought a nice way to finish would be to look back on some of the best articles that have been published on the site over the ten years. Choosing just ten wasn't easy, but those we've selected were chosen either for their foresight, expertise and invention, or just because they perfectly represent a particular moment in football analytics history.
gillerinvestments.com - Graham Giller
The Markets Aren’t Normal. Continuous Time Finance Isn’t Real . The Sharpe Ratio Sucks. Backtesting Sucks. Mean Variance Optimization Sucks
mlcontests.com
We summarise the state of the competitive landscape and analyse the 300 competitions that took place in 2023. Plus a deep dive analysis of 60 winning solutions to figure out the best strategies to win at competitive ML.
soundcloud.com
This week on Bet the Process, Jeff and Rufus chat football (soccer) with Sarah Rudd, former head of analytics at Arsenal, about her beginnings in the industry, the challenges faced by women and Americans in the field, and her decision to start a consulting business in the sports and data space.
let-all.com - Georgy Noarov and Aaron Roth
Calibration is a popular tool for uncertainty quantification. But what exactly is it good for? It turns out a lot! If predictions are calibrated, then for any downstream decision maker, it is a dominant strategy amongst all policies to treat the predictions as correct and act accordingly. This is a strong sense in which calibrated predictions are “trustworthy”. In all sorts of scenarios, this property implies desirable downstream guarantees. But calibration is hard to achieve in high-dimensional settings. Fortunately, in many concrete applications, full calibration is overkill. In a series of blog posts, we’ll describe a program, laid out more formally in [NRRX23], that aims to identify weaker conditions than calibration that suffice to give strong guarantees for particular downstream decision-making problems, and show how to achieve them efficiently.
