Estimating Latent Factors Based on Statistical Data Analysis

Abstract

Abstract In recent years, statistical methods have been widely used to estimate latent risk factors that affect the prices of financial assets. This paper develops new estimators for asset pricing factors by introducing dependence measure--distance covariance, that can identify nonlinear dependence. We combined distance covariance with Principal Component Analysis (PCA) and Risk-Premium PCA (RPPCA) and made contrast analysis based on Chinese market data. RPPCA, as a new method, shows strong applicability and detects factors with high Sharpe-ratio efficiently. Moreover, distance covariance produces better performance than covariance in PCA as a factor estimator, which illustrates the superiority of the distance covariance. Finally, the most striking results revealed by the study is that RPPCA including distance covariance of residuals outperforms others with a smaller pricing error and a significantly large Sharpe-ratio.