Win Rate Regression: Teams That Outperform xG

## The Overperforming Team Problem A team finishes the first 10 matches of the season with 8 wins, a +15 goal difference, and a top-table position. The market prices them as genuine title contenders. But their xG data tells a different story: they have scored 12 goals on just 8 expected goals (150% conversion), and their goalkeeper has made 6 saves rated in the top 5% for difficulty. They are significantly outperforming their underlying performance. ## The Statistical Expectation Teams that significantly outperform their xG in one period are expected to revert toward their xG performance in the next period. This is regression to the mean — not because the team "gets worse" but because the lucky components (high conversion, exceptional saves) do not sustain. ## Quantifying the Regression Research on football performance data shows: - Goal conversion rates in excess of xG predict lower conversion in subsequent matches with 60–70% frequency - Goalkeeper save percentages above expected predict lower save percentage in subsequent matches - Point tallies significantly above xPoints (expected points from xG) predict lower points per game in subsequent matches The regression is not guaranteed in any individual match — it is a statistical tendency over many matches. ## The Betting Opportunity When the market overvalues a team because of outperformance metrics (ignoring the xG data), a value opportunity exists: - The overperforming team is overpriced - Their opponent (or "Under" on total goals) may be underpriced The analytical edge: use xG-based ratings rather than results-based ratings. The market's heavy weight on results creates systematic mispricing of teams in hot or cold sequences. ## The Timing Challenge Regression is not always immediate. An overperforming team can continue overperforming for 5–10 matches before the statistical tendency asserts itself. Betting against form too early can produce multiple losing bets before the regression arrives. Risk management (standard staking, not increasing stakes in anticipation of regression) is essential.