Soccer In-Play Prediction

Models based on birth-death processes with Poisson arrivals is the standard approach for in-play prediction of soccer. However, such models are typically not well aligned with goal expectancy – which they tend to systematically under-predict.

This project will test various extensions of in-play prediction of soccer with the objective of making unified predictions of outcomes (1X2) and the joint distribution of goals. Among the approaches are:

• Reflect uncertainty (error distribution) in probability estimates, and corresponding joint distribution of the two goal arrival rates.
• Bayesian models.
• Adjustments for time and goal difference.

Key tasks will be collecting data and coding and running analysis