I recently read an interesting paper that was published last year, by Igor Barbosada Costa, Leandro Balby Marinho and Carlos Eduardo Santos Pires: Forecasting football results and exploiting betting markets: The case of “both teams to score . In the paper they try out different approaches for predicting the probability of both teams to score at least one goal (“both teams to score”, or BTTS for short). A really cool thing about the paper is that they actually used my goalmodel R package . This is the first
I have to say that I personally prefer to model the scorelines directly using Poisson-like models, rather than trying to fit a classifier to the outcome you want to predict, whether it is over/under, win/draw/lose, or both-teams-to-score. Although I can’t say I have any particularly rational arguments in favour of the goal models, I like the fact that you can use the goal model to compute the probabilities of any outcome you want from the same model, without having to fit separate models for every outcome y
Okay, so how can you compute the probability of both teams to score using the Poisson based models from goalmodel ? Here’s what the paper says: