PCA Residual Mean Reversion

Code Stratmill research code

Summary

The document outlines an equities statistical arbitrage method that uses principal component analysis to estimate common return factors. Asset returns are standardized, PCA components provide factor weights, and regressions of returns on factor returns produce residual series and coefficients. The residuals are treated as mean reverting; an estimated reversion speed filters candidate portfolios, while an S-score expresses deviation from an estimated centered mean. Threshold rules open and close long or short positions, and regression coefficients and factor weights are combined to form portfolio targets.

The code comments describe a rolling workflow: refresh factor weights on a correlation lookback, estimate residuals over a separate window, and update positions from the resulting scores. It references a published statistical arbitrage paper and gives parameter guidance, but presents no backtest results or transaction cost analysis. The supplied source is visibly truncated, and its signal construction contains implementation details that would need review before use. PCA factors and residual mean reversion may vary across samples, so the method’s profitability and robustness cannot be inferred from this code description alone.

Key ideas

PCA is used to represent common return drivers across a stock universe.
Regressing each asset’s returns on the PCA factor returns yields residuals and coefficients for portfolio construction.
Residual processes are filtered by estimated mean reversion speed and converted into S-scores for trade entry and exit.
The strategy combines qualifying eigen portfolios into time-varying target weights.
The excerpt provides no empirical performance evidence and is incomplete, so the implementation requires validation.

PCA Statistical Arbitrage with Residual Mean Reversion

Summary

Key ideas

Tags