war-gar-decomposition

You can estimate hockey WAR and GAR by applying RAPM ridge regression to shift-level data, using duration-weighting and strength-state separation to isolate individual skater impact from teammates and opponents.

RAPM ridge regression is a statistical method that estimates a hockey player's isolated impact by penalizing large coefficients, using sparse design matrices from shift data to separate teammate and opponent effects across strength states.

GAR component breakdown splits total value into even-strength offense, even-strength defense, power play, penalty kill, penalties drawn, and penalties taken, applying strength-state separation to shift data for each distinct component.

Yes, you can convert GAR into WAR and then into surplus value, making the estimates useful for trade analysis, contract grading, and player evaluation by quantifying a skater's value above their salary.

You need shift-level performance data to construct sparse design matrices, which are then processed using duration-weighted regression and aging-curve adjustments to estimate skater impact by strength state.

You validate WAR estimates by applying regularization guardrails, enforcing sample-size thresholds, and running team-level sanity checks to ensure the player valuations remain interpretable and stable.