My projections simulate 5000 different league outcomes. The outcome of each match is determined randomly – but each team’s average attack and defence strength comes from my adjusted goals rating system. This is a simple method for rating teams’ performance, worked-out by combining Goals, Shots-on-target and Shots, for and against each team.
The projected outcomes depend on 2 inputs:
a) Actual points accumulated to date. As the season progresses, actual points become increasingly important to the final outcome.
b) The relative strength of each team. This is used to simulate the results of future matches.
Both of these are updated each time I run a projection, so any changes from week to week can be due to both a) and b).
I use my long-term adjusted goals rating system for b) above. This is an objective measure, based solely on a team’s actual performance, which has worked reasonably well in predicting future performance in previous seasons (rationale here).
But the model’s not perfect. In particular, because the long-term adjusted goals ratings use data from this season and last season – they don’t quickly adjust to changes (e.g. change of manager, transfer/injury of key players or investment in team). So, where there are genuine reasons for a team to perform above its long-term average – these factors won’t be reflected in the ratings.
As with all my projections, the main purpose is to aid understanding (e.g. try to help explain when things don’t happen as expected or why market odds are different). The model’s logic, construction or inputs have not been independently tested.