Sports Analytics Weekly by kubeia.io - 35/2023

In the world of financial risk-taking, day traders in the stock market and horse bettors with bookmakers both engage in speculative activities. However, a notable contrast emerges when it comes to the regulatory treatment of these two groups. Day traders often find themselves enjoying a degree of freedom in their trading activities, with their accounts rarely facing limitations or closures. Conversely, horse bettors placing wagers with bookmakers often encounter more stringent restrictions and even closures of their accounts. This article aims to delve into the factors that contribute to this regulatory dichotomy, examining the mechanisms that govern these industries and the inherent differences that lead to divergent outcomes.

Is a goalkeeper who saves five goals above expectation on 100 xG better than a goalkeeper who saves four goals above expectation on 50 xG faced? So we start by breaking things down to their goals saved ratio, that is G/xG or G/PSxG. From there, we use empirical Bayes estimation. The short version of this is that we use the historical G/PSxG rates of MLS goalkeepers as the expectation for all goalkeeper performance, and the more sample size we get of their actual G/PSxG performance the further we move away from prior expectations.

Pass completion models are frequently used to evaluate QB performance, e.g. in our 2023 NFL Draft analysis. So what happens if we use these models to evaluate defensive performance instead? In this article, we’ll take a look at possible interpretations of these models from a defensive perspective.

While expected goals (xG) tell us about the quality of the shots—accounting for the context of a shot—they’re agnostic to player identity, so we need more info to tease out individual shooting ability. Shaw points out that one way to evaluate finishing “overperformance” is to divide a player’s count of goals (G) by their xG. A ratio of 1 indicates that a player is scoring as many goals as expected; a ratio greater than 1 indicates overperformance; and a ratio less than 1 indicates underperformance.But this measure (i.e. G/xG) wasn’t really the novelty of Shaw’s writing. Rather, it was his approach to adjusting this ratio for the fact that the volume and quality of shots vary a lot among players. Just comparing the empirical G/xG ratio of one player to another can be extremely misleading. (A direct comparison of the Crew’s Sean Zawadzki and Philly's Jose Martinez, who have completely different shot volumes and profiles, is statistically meaningless.) Using empirical Bayes (EB), one “shrinks” the G/xG ratio back closer to 1. While the exact amount of “shrinking” depends on how many shots one has taken, broadly, the G/xG ratio of those who have taken fewer shots is going to be adjusted more. After applying the EB adjustment, G/xG becomes more comparable across players, regardless of the number of shots they’ve taken.

In this article, I will share my experience of contributing to OpenCV, a renowned open-source library, despite having limited knowledge of C and understanding its architecture. I achieved this with the assistance of ChatGPT, a Large Language Model (LLM).

So, without further ado, here are 15 lessons, ideas, and frameworks from 15 years.

Mean-variance analysis is widely used in portfolio management to identify the best portfolio that makes an optimal trade-off between expected return and volatility. Yet, this method has its limitations, notably its vulnerability to estimation errors and its reliance on historical data. While shrinkage estimators and factor models have been introduced to improve estimation accuracy through bias-variance trade-offs, and the Black-Litterman model has been developed to integrate investor opinions, a unified framework combining three approaches has been lacking. Our study debuts a Bayesian blueprint that fuses shrinkage estimation with view inclusion, conceptualizing both as Bayesian updates. This model is then applied within the context of the Fama-French approach factor models, thereby integrating the advantages of each methodology. Finally, through a comprehensive empirical study in the US equity market spanning a decade, we show that the model outperforms both the simple 1/N1/N portfolio and the optimal portfolios based on sample estimators.

The MEENA model sparked an internal memo written by Noam Shazeer titled "MEENA Eats The World.” In this memo, he predicted many of the things that the rest of the world woke up to after the release of ChatGPT. The key takeaways were that language models would get increasingly integrated into our lives in a variety of ways, and that they would dominate the globally deployed FLOPS. Noam was so far ahead of his time when he wrote this, but it was mostly ignored or even laughed at by key decision makers.

In this edition of the newsletter, we direct our attention to one of the most prominent highlights of the summer: the release of the Llama 2 base and chat models, as well as CodeLlama, the latest highlights in the open-source AI large language model (LLM) landscape.Additionally, we delve into the leaked GPT-4 model details, discussing an analysis of its performance over time and covering emerging alternatives to the prevalent transformer-based LLMs.

Conventional xG models do not take into account the identity of the player taking the shot. You need a dataset comprising many thousands of shots to properly measure chance quality over all positions and situations, meaning that you must aggregate shots from many different players. Of course, you would expect attacking players of the quality of Harry Kane, Mo Salah or Eden Hazard to be more likely to convert a chance than less illustrious forwards, or the average defender or midfielder. Given that xG is calibrated based on the success rate of shots from mostly inferior players, we should therefore expect elite forwards, such as Kane, to outscore the xG total of their shots over a season. But to what extent?