Research Foundations
HandicapLab operates as an empirical sports analytics research terminal. We reject black-box betting tipster models in favor of rigorous statistical evaluation.
1. The Dixon-Coles Bivariate Poisson Model
Standard Poisson models assume independence between home and away goals scored. However, empirical football research (Dixon & Coles, 1997) shows low scores (0-0, 1-0, 0-1, 1-1) exhibit significant interdependence.
Our model applies a bivariate correction factor ρ (rho) to adjust probability density for low-scoring scorelines:
In our EPIC 56 calibration tournament, ρ was fitted per out-of-sample walk-forward fold, locking champion parameter ρ = -0.05.
2. Expected Value (EV), CLV & Brier Score
Expected Value (EV)
EV computes the mathematical return per unit staked across all settlement states (Full Win, Half Win, Push, Half Loss, Full Loss). An edge exists only when model fair probability exceeds devigged market probability.
Closing Line Value (CLV)
Closing Line Value measures whether taken odds exceed the final closing price at kickoff, using Pinnacle as the ground-truth benchmark. Beating the closing line is the gold standard for statistical edge in liquid markets.
Brier Calibration Score
We measure probabilistic accuracy using the Brier score, penalizing overconfidence and rewarding well-calibrated probabilities.
3. What "NOT VALIDATED" Means
In our historical backtest spanning 2015-2026 across European leagues (7,225 matches), the champion Dixon-Coles model achieved:
- Realized ROI: -2.30% (flat unit staking)
- Mean CLV: -0.0311% (not statistically significant, p=0.555)
- Status: RESEARCH_ONLY / NOT_VALIDATED
Because the model does not beat the bookmaker margin after vig, we publish all inferences openly for scientific research and live track-record compilation, without claiming betting profitability.
4. Data Governance, Provenance & Free-First Architecture
HandicapLab enforces an uncompromising Zero-Synthetic Data Invariant. We separate all analytical assets into three verifiable data classes:
Final scorelines (FTHG, FTAG), kickoff dates, and team identities sourced from API-Football and Football-Data.co.uk. Absolute ground truth.
Deterministic outcome resolution: Over/Under 2.5 (FTHG + FTAG) and BTTS (FTHG > 0 && FTAG > 0). Evaluates probability calibration directly against pitch facts without synthetic odds.
Historical closing odds (Pinnacle PCAHH / PC>2.5 from Football-Data.co.uk) and live market feeds (OddsPAPI v4). Used strictly for CLV benchmarking and EV calculation.
Honest Research Disclosure: In historical backtests, Asian Handicap lines and Over/Under 2.5 lines have full Pinnacle closing price coverage. Both Teams to Score (BTTS) and alternate totals are evaluated on Class B match-fact distributions for statistical calibration; live BTTS signals activate only when real bookmaker prices meet our minimum edge threshold.
Responsible Gambling Notice
This is a research project. Nothing here constitutes betting advice. Past performance does not guarantee future results. All content is for informational and educational purposes only. Gamble responsibly and only where legal.