Author(s) :

Arnav Rajadhyaksha.

Volume/Issue :

Abstract :

This study evaluates whether dynamic investment scaling based on spread magnitude enhances the performance of pairs trading strategies. Using daily closing prices of U.S. equities across eleven sectors from 2021 to March 2026, pairs are identified through a two-stage selection process combining zerocrossing frequency and the Augmented Dickey–Fuller (ADF) tests applied to rolling z-score–normalized spreads. Pairs are formed over a 12-month formation period and evaluated over the subsequent 3-month trading windows. Trading rules incorporate threshold-based entry, exit, and stop-loss conditions, while position sizes are adjusted dynamically using a parameterized scaling function (k-values) as spreads change. Sensitivity analysis suggests that intermediate entry thresholds (z ≈ 1.25) balance trade frequency and signal quality, while wider stop-loss thresholds (z ≈ 3.5) help mitigate extreme losses. Results show that no single scaling parameter consistently maximizes returns across sectors or time periods. Regression analysis indicates that scaling parameters are not statistically significant predictors of returns, whereas ADF statistics and selected sector indicators exhibit significance. However, the explanatory power remains limited (adjusted R² ≈ 0.03), consistent with the noisy nature of financial returns. In contrast, scaling parameters are significantly associated with maximum drawdown, suggesting an effect on downside risk rather than expected returns. Overall, dynamic investment scaling does not materially improve returns but can reduce drawdowns. Strategy performance is more strongly driven by pair selection characteristics, particularly spread stationarity, and sector-specific factors. These results should be interpreted with caution given simplifying assumptions such as excluding transaction costs and relatively limited out-of-sample evaluation data.