Measuring the Relevance of Factors on Cross-Sectional Returns with Decision Trees

Abstract

This study is concerned with new ways to identify and analyse the factors on cross-sectional returns in financial markets with respect to their time-variability. Therefore, classification and regression trees and conventional regression models are applied. This study uses data on the S&P 500 from 1999 to 2019. Empirical findings show high time variability of factors on cross-sectional returns. The high level of time-variability is not dependent on the applied model. It is also shown that CARTs and conventional regression models have low power when it comes to identifying the factors on cross-sectional returns or predicting the returns themself.