Improving Mean-Reversion Strategy Performance: The Effect of Signal Thresholds under Volatility-Driven Market Noise

Abstract

Mean-reversion strategies are a cornerstone of quantitative trading but are vulnerable to microstructure noise and low-quality signals in minute-level markets. This paper studies signal thresholding as an explicit noise-filtering mechanism for a VWAP-based mean-reversion strategy. Using minute-level HDF5 market data and a Python and pandas backtesting framework, we compare fixed and volatility-adaptive thresholds in the range 0.1%–0.5%, evaluate performance under realistic transaction-cost assumptions, and test robustness across volatility regimes and assets. Initial experiments show that introducing a 0.1% signal threshold improved the Sharpe ratio relative to the original strategy, supporting the feasibility of the approach. The results indicate a nonlinear (inverted-U) relationship between the threshold level and risk-adjusted performance and support a volatility-adaptive threshold rule that balances signal quality and sample size.