The goal of this project is to give context on the levels of basketball within Europe. Understanding the context of the league in which players and teams compete enables more accurate evaluations and, ultimately, more reliable projections.
The rankings and projections come from a blend of two separate models: a club model and a player model. The club model is a ridge regression on margin of victory; the player model is a RAPM-style ridge fit that measures each man’s impact on the score while he is on the floor. The result is a model that has backtested to beat the market in each of the past five seasons.
The club model
The club model was fitted on 61,413 games across 41 competitions, using mainly the international competitions as bridge leagues. Each new season, teams are regressed towards their league average, and then must re-earn their status from that current season’s games. Teams that play in multiple leagues are judged on their results in each, weighing the strength of the domestic league and the international competition alike. The result is a clear picture of the strength of each league over the past five seasons, as seen above, as well as projections built from the current season’s results and what each club is expected to be.
The player model
The player model is a RAPM ridge regression, fitted on play-by-play data from the international and the top domestic leagues. A minutes model projects each player’s minutes within his own team, so that his impact is assessed at the time he is actually expected to play. That impact is then broken down into offensive, defensive and net — his overall impact on the team while he is on the floor — with all three measured relative to an average player among the leagues the play-by-play comes from.