Sensitivity analysis applied to tilting methodologies
摘要
This paper addresses the challenge of constructing factor portfolios that accurately reflect asset characteristics while remaining robust to noise. We examine the variation of the portfolio weights produced on noisy versions of the original signal while targeting a fixed unit of signal exposure. It is proposed a framework to evaluate how well portfolio strategies handle noise by analyzing how portfolio weights vary with noisy versions of the original signal, while maintaining a consistent level of signal exposure. Assuming normal distribution for signal and noise, we analytically derive the distribution of portfolio weight deviations. These findings are supported by simulations using more realistic skewed, fat-tailed, and correlated signals. The framework is then applied to various common portfolio strategies and factors.