P500 TELEBASEBALL Loading...
GLOSSARY
BETTING & RISK MANAGEMENT
NET UNITS
+
Standardized position sizing relative to bankroll. Enables consistent risk scaling and performance attribution independent of bet size variance.
WINS/LOSSES
+
Total number of winning and losing bets over the season.
YIELD
+
Return on investment (ROI) expressed as a percentage of total handle. The definitive efficiency metric that reveals true edge beyond raw win rate.
SHARPE RATIO
+
(μ - r₀) / σ
Risk-adjusted return metric: (mean return - risk-free rate) / standard deviation. Validates that profitability isn't achieved through excessive variance.
Two strategies both average +10% yield. Strategy A has consistent weekly returns (low variance), Sharpe = 2.0. Strategy B swings wildly between big wins and losses (high variance), Sharpe = 0.5. Strategy A is superior—same returns with less risk.
KELLY CRITERION
+
(bp - q) / b
Optimal bet sizing formula: (bp - q) / b, where b=odds, p=win probability, q=loss probability. Maximizes logarithmic utility over infinite horizon.
MODEL PERFORMANCE METRICS
ACCURACY
+
(TP + TN) / (TP + TN + FP + FN)
Classification accuracy: (TP + TN) / (TP + TN + FP + FN). Baseline comparison against Vegas closing line consensus to validate edge extraction.
LOG LOSS
+
-Σ(y·log(ŷ) + (1-y)·log(1-ŷ))
Logarithmic scoring rule: -Σ(y·log(ŷ) + (1-y)·log(1-ŷ)). Penalizes both incorrect predictions and overconfident miscalibrations. Lower is superior.
If we predict 75% win probability for the Yankees but they lose, log loss = -log(0.25) = 1.39. If we predicted 60% and they lost, log loss = -log(0.40) = 0.92. The model is penalized more for being confidently wrong.
BRIER SCORE
+
(1/n)Σ(ŷ - y)²
Mean squared error of probability forecasts: (1/n)Σ(ŷ - y)². Measures calibration quality—how well predicted probabilities match empirical frequencies.
If we predict 70% win probability for 100 games and exactly 70 win, Brier = 0.21. If only 60 win, Brier = 0.25. Lower scores indicate better-calibrated probabilities that match actual outcomes.
AUC-ROC
+
Area under the receiver operating characteristic curve. Quantifies discriminative power—the model's ability to rank positive instances above negatives across all thresholds.
AUC of 0.85 means when comparing a random winning bet to a random losing bet, the model ranks the winner higher 85% of the time. Perfect separation (AUC=1.0) means all winners ranked above all losers.
F1 / F2 SCORE
+
F1 = 2·(precision·recall)/(precision+recall)
Harmonic means balancing precision and recall. F1 weights them equally; F2 emphasizes recall (β=2), prioritizing detection over false positive avoidance.
F1: If we recommend 20 bets, 15 win (precision=75%) and we catch 15 of 20 true edges (recall=75%), F1=75%. F2: Prioritizes catching more edges even if it means more false positives—better for maximizing opportunity capture.
TELEBASEBALL • ANALYTICS DECODED