Prediction Accuracy vs. Volatility: Why Upsets Still Matter

Upsets are part of why sports are fun—but they also carry information for a rating model. Zow Ratings tracks both prediction accuracy and volatility to understand not just outcomes, but how stable the model’s understanding of a sport is at any point in the season.

Prediction Accuracy

Prediction accuracy measures how often the model identifies the winner correctly based on current ratings. Improving accuracy usually means the model is learning team quality more effectively as results accumulate.

Volatility

Volatility measures how much ratings shift day-to-day. High volatility often occurs early in a season, when few results exist and each game carries more weight.

Why Upsets Still Matter

Even when accuracy rises, an upset can signal a real change—injuries, lineup changes, improving teams, or a mismatch the model hasn’t fully captured yet. Upsets also help reduce overconfidence and keep ratings responsive.

Related pages: Confidence Stats · Results Predictor