Discovery of a 13-Sharpe OOS Factor: Drift Regimes Unlock Hidden Cross-Sectional Predictability

Battle Count: 79

This paper is directly and highly relevant to quantitative trading. It presents a complete cross-sectional equity strategy with explicit signal construction, portfolio implementation, risk management, walk-forward validation, capacity analysis, and operational requirements. The regime-conditioning framework (binary gate on drift regimes) represents a novel approach to factor activation. The paper addresses practical concerns including transaction costs, market impact, liquidity stress, kill-switches, and sector neutrality. The Sharpe ratio of 13+ OOS, while extraordinary and warranting skepticism, is presented with extensive robustness testing. The strategy is designed for institutional implementation with $100-500M capacity.

The core signal construction is relatively simple: inverse price percentile ranks, 10-day reversal z-scores, and a 63-day positive-day fraction threshold. Portfolio construction involves standard market-neutral long-short with volatility scaling. However, practical implementation requires: daily data feeds for S&P 500 constituents, real-time regime calculation, order execution infrastructure (DMA/prime broker), risk monitoring systems, kill-switch logic, and reconciliation. The paper estimates $500k-$1M annual operational cost with 1-2 FTE oversight. The algorithmic complexity is low but operational infrastructure requirements are moderate.

The paper provides detailed formulas for signal construction (value percentile rank, reversal z-score, drift regime filter), portfolio construction (market-neutral long-short, volatility scaling), walk-forward validation protocol (5-year train, 1-year test), transaction cost assumptions (0.6bp), and kill-switch parameters. However, no code or repository is provided. The use of current S&P 500 constituents introduces survivorship bias that is acknowledged but not corrected. The exact data processing pipeline and implementation details for the 20-year daily dataset are not fully specified.