Forecasting test cricket match outcomes in play

This paper forecasts match outcomes in test cricket in play, session by session. Match outcome probabilities at the start of each session are forecast using a sequence of multinomial logistic regression models. These probabilities can assist a team captain or management in considering a certain aggressive or defensive batting strategy for the coming session. We investigate how the outcome probabilities (of a win, draw, or loss) and covariate effects vary session by session. The covariates fall into two categories, pre-match effects (strengths of teams, a ground effect, home field advantage, outcome of the toss) and in-play effects (score or lead, overs-used, overs-remaining, run-rate, and wicket resources used). The results indicate that the lead has a small effect on the match outcome early on but is dominant later; pre-match team strengths, ground effect and home field advantage are important predictors of a win early on; and wicket resources used remains important throughout a match.