First developed in 1982, the double Poisson model, where goals scored by each team are
assumed to be Poisson distributed with a mean depending on attacking and defensive
strengths, remains a popular choice for predicting football scores, despite the multitude
of newer methods that have been developed. This paper examines the pre-tournament
predictions made using this model for the Euro 2020 football tournament. These
predictions won the Royal Statistical Society’s prediction competition, demonstrating
that even this simple model can produce high-quality results. Moreover, the paper also
presents a range of novel analytic results which exactly quantify the conditions for the
existence and uniqueness of the solution to the equations for the model parameters.
After deriving these results, it provides a novel examination of a potential problem with
the model - the over-weighting of the results of weaker teams - and illustrates the
effectiveness of ignoring results against the weakest opposition. It also compares the
predictions with the actual results of Euro 2020, showing that they were extremely
accurate in predicting the number of goals scored. Finally, it considers the choice of
start date for the dataset, and illustrates that the choice made by the authors (which
was to start the dataset just after the previous major international tournament) was
close to optimal, at least in this case. The findings of this study give a better
understanding of the mathematical behaviour of the double Poisson model and provide
evidence for its effectiveness as a match prediction tool.