Goalkeeper Performance and Regression

## The Goalkeeper Regression Phenomenon Goalkeeper performance is among the most volatile statistics in football. A goalkeeper can "save" a team for a run of matches with exceptional performances — but statistical regression means this performance level almost never sustains. ## Why Goalkeeper Stats Are Volatile A typical goalkeeper faces 3–6 shots per match that would score if not saved (on target, above xG threshold). Each save is a binary outcome with significant randomness. Over 10 matches, a goalkeeper can make 6–8 more saves than their long-run expected rate — appearing "in form" — before regression normalises the numbers. ## Post-Save Percentage Analysis Compare a goalkeeper's current save percentage (saves / shots on target) to their career average and league average: - Current: 82% - Career average: 71% - League average: 73% This goalkeeper is performing 11 percentage points above career average. The probability they sustain this for the next 10 matches: approximately 12%. Regression toward 71–73% is the base expectation. ## The Market's Goalkeeper Blind Spot The betting market is generally poor at incorporating goalkeeper performance regression into pricing. A team with a "goalkeeper in form" often has its odds shortened based on recent clean sheets — without the market questioning whether the clean sheets reflect genuine defensive quality or goalkeeper outperformance. ## The Betting Application When a team's recent defensive results are significantly better than their xGA (expected goals against): 1. Calculate how much of the defensive overperformance is attributable to goalkeeper luck vs genuine defensive improvement 2. If goalkeeper luck is the primary driver: expect regression, which is not priced in 3. Over/Under goals markets (favour the Over) and defensive team win markets (favour the opponent) become relatively better value ## The Caveat Goalkeeper quality is genuinely variable. A new elite goalkeeper can genuinely be better than his predecessor — the improvement is real, not just luck. Always separate genuine quality change from performance deviation around a stable baseline.