Ciao from Italy 
I’m looking for a thinking & coding buddy with pymc in the field of soccer sport betting.
I’m a full time developer in IBM Iseries (aged 57) but I’d like to share my weekend project with someone interested in PYMC & sport betting.
I’have tried with Stan and with Machine Learning and a lot of code in python has been developed (scraping e bets portfolio selection with cvxpy).
My objective:
- learn better Bayesian data analysis with pymc (following Richard McElreath “Statistical Rethinking” book);
- gain better coding habit in python (not the language in my daily coding activities);
- gaining fresh idea and momentum;
- maybe (!) gain money from the results of this effort.
Confidential information: all my previous attempts, in the long run,
have produced less money (losses) 
Hi,
I’m also working on sports but I have not ventured into soccer. I’d be happy to collaborate if possible and discuss sports betting with you.
great. I Will message you. Thank you very much.
Giàmpaolo
You might find my blog post here interesting: Accounting for Game Situation When Using xG to Evaluate Team Strength
Using this model, I was breaking even in backtests in the Premier League. My next step was to evaluate different window lengths for form.
But personally, I think a model-only approach won’t work in the big leagues. There is too much information that oddsmakers and other bettors have that either requires a significant investment of time to track (e.g. injuries) or is tough to quantify (the tendency of Tottenham (my favorite team) to find ways not to win).
good luck!
3 Likes
Hi @DanWeitzenfeld,
I am studying your post; it is very interesting.
Ok on misalignment of information between me and bookmaker, they know the “wisdom of the crowd”.
I have two questions for you:
-
in his first video (L1: Introduction (State of Bayes Lecture Series)), Maxim Kochurov (@ferrine) says that Bayes model is the best if your objective is different from “fit-predict” mantra of machine learning. Maybe machine learning models are better for sport prediction than bayesian model? But machine learning models can produce well calibrated probabilities?
-
even if I can arrive at the perfect bayesian model, will bookmaker let me run free in sport betting or it will be much probable that they will ban me?
Ciao from Italy: forza Milan 
@Gianpaolo_Galli Hey! I have also began to look at pymc soccer team strength modelling. How far did you get in your journey? I have managed to create a decent Dixon Coles model with time decay applied, but struggling to now implement xG as well as goals and also league hierarchy effects.
@porterhouse Hi, not far away. Almost stopped 
Lost momentum. Sorry.
@Gianpaolo_Galli What sort of model structure did you go for?
Thank you, did you ever explore adding xG?