Bayesian Sports Betting
Registration is now closed for Bayesian Sports Betting. The course will run for twelve weeks starting on Tuesday, January 4th, 2022.
Sign up to our mailing list below to be notified of new future cohorts of this course.
1. Learn small data techniques and Bayesian reasoning as applied to the real world of sports and early season data.
2. Network with other students and alumni on a private Slack workspace during and after the course is over.
“Small sample size” is a phrase that arises frequently in sports betting analytics. With major sports seasons lasting between 16 and 162 games, analysts often lack the luxury of fully reliable data sets. They must do whatever they can to extract as much predictive signal as possible from a sea of random noise. While most modern research focuses on machine learning and “big data”, the often neglected world of “small data” can be a much more important, and more challenging, topic for aspiring winning bettors. Drawing on material from probability theory and actuarial science, as well as real-world betting examples, this course will equip you with a set of tools so that you can do more with less, culminating in the study of a unique form of model developed by the instructor which is not found in any textbook or academic paper.
Matt Buchalter is the owner of Plus EV Sports Analytics, which offers consulting and education to bettors of all skill levels. Matt holds a Bachelor of Mathematics degree from the University of Waterloo, where he was awarded the Samuel Eckler Medal for highest academic standing in his graduating class. Trained as an actuary, he has spent his spare time applying quantitative methods to sports betting for the past 12 years. He has had success in the Canadian sports lotteries (which Wikipedia wrongly claims are unbeatable), North American horse racing and various prop betting markets in major American sports. He has been a guest on several leading podcasts in the betting space, and he has written articles for Pinnacle Sports and more recently for his own blog where his articles have been recommended by some of the world’s leading professional bettors. Matt has achieved worldwide acclaim in the field of sports betting analytics with specialized expertise in Bayesian analysis, applied probability, optimal bet sizing, betting market dynamics and evaluation of betting results. His teaching style blends academic theory and real-world practice while keeping the atmosphere casual and fun.
Although this course will cover some advanced topics, they will be taught in such a way that only a high school level mathematics background is required.
Some basic knowledge of statistical concepts (expected value, median, variance, etc) is recommended.
This course is an ideal next step for graduates of Analytics.Bet’s “Applied Analytics and Sports Betting” course who would like to continue their learning with a deeper exploration of a more specialized topic. However, this is absolutely not a requirement and no prior knowledge from that course will be presumed in this one.
Above all, students should come in with an open mind, a sense of intellectual curiosity and a willingness to learn.
What is Small Data?
• Empirical models: using the past to predict the future
• When empirical models tend to succeed, and when they tend to fail
• Learning the hidden “generator”
• Data = signal + noise and the law of large numbers
• What you can and can’t do with small sample sizes
• Defining the challenge of “small data”
Methods for Small Data
• The Efficient Market Hypothesis and regression to market
• Ballast methods – judgmental selection
• Ballast methods – data-driven selection
• Credibility methods
• Introduction to probability distributions & curve fitting
• Curve fitting using market implied probabilities for various types of distributions
• Implied assumptions, fragility/antifragility and tail events
• Curve fitting using empirical data and maximum likelihood estimation
• What is Bayes Theorem?
• Latent variables, epistemic uncertainty and process variance / parameter variance
• Using Bayes to “learn the generator”
• Mixed models and hyperparameters
• Conjugate priors
• Maximum posterior likelihood estimation
• Estimating covariate coefficients and hyperparameters together – “The Conjugate GLM”
Sports Betting Case Studies
• Early in-season projection of baseball statistics such as batting average
• Projection of NFL quarterback performance at every stage from rookie to veteran
• Analysis of basketball totals and props
• Updating power rankings in-season (applicable to any sport)
• Pricing “off-market” numbers for totals and props
Individual registration is available through the link below. Groups of 15 or more may contact firstname.lastname@example.org for information on custom-tailored private group sessions.