APBRmetrics

Hi everyone,
ok here is all my code and data that I have been working on for the last 3-4 years.


I do not have much time (nor interest) in pursuing this any more, since my focus has moved on to other things.


If someone has any use for it please go ahead and try it out.
It is in python.

https://github.com/vzografos/NBA_prediction

Any high level takeways having worked on this for a while? Looks like you've tried out a lot of different staking strategies.

A lot.

1) Markets (exchanges) are very efficient. Not in terms of accuracy of prediction but in terms of probabilities. So they might get 40-45% of the games wrong but the market/crowd are very good in giving high probabilities on the games they get right and low probabilities on the games they get wrong. Not an easy task to accomplish

2) The best staking methods are the Kelly (& derivatives, like the Drawdown regularised) where you also allow to take positions against the market (Short). Reinforcement learning staking methods should be better than the Kelly but didnt have the time to fully investigate.

3)If you do not have an edge over the market no staking method is going to help you

4) If you want to make money consistently focus on A) Prediction of player boxscore scores (average over the season) B) Prediction of lineups and C) incoprorpate other expert information (i.e. sportsbooks, bookies etc) into your game outcome prediction method. Sportbooks usually have a team of very good data scientists behind doing their predictions

5) Never bet against a sportsbook. Always try against an exchange (e.g Betfair, Matchbook etc). I dont know if you have those in the US