Equity Strategy Backtesting: Luck or Edge? The MinervaScore as a Statistical Robustness Grade

Battle Count: 54

Directly addresses one of the most critical problems in quantitative trading: distinguishing genuine alpha from backtest overfitting. The paper provides a practical, auditable validation framework for trading strategies with explicit multiple-testing correction, regime stability assessment, and verdict-consistent reporting. The 359,062 production records and pre-registered real-market study demonstrate practical deployment. The framework is specifically designed for the equity strategy backtesting pipeline and addresses selection bias, overfitting, and evidence sufficiency - all core concerns for quantitative traders.

The construction involves multiple stages: five validation gates with different statistical foundations (DSR with pre-Φ statistics, CSCV-based PBO, bootstrap SPA, MinTRL formula, regime composite), two-scale margin encoding (logit, pre-Φ, tanh), correlation-adjusted weighted inverse-normal aggregation, conservative offset, normal CDF mapping, and a two-band percentile display with frozen quantile knots. Requires access to CPCV candidate-by-fold matrices, effective trial counts via eigenvalue decorrelation, and calibration constants from a production population. The fail-closed design and versioned calibration add operational complexity.

The paper provides extensive reproducibility details: production scoring module at frozen calibration commit, released derivation script for calibration constants, self-contained synthetic study scripts importing production gate functions, pre-registered analysis plans committed before execution, cryptographic attestation (SHA-256) on batch exports, 81 frozen quantile knots, and all numbers produced by scripts. However, the platform (Minerva) is proprietary and the full production data (359,062 records) is not publicly available. The synthetic environment is self-contained but the real-market data is platform-specific.