An Empirical Study on New Model-Free Multi-output Variable Selection Methods

Chatterjee’sAnsari, J. Lütkebohmert, E. Rockel, M. rank correlation has attracted a lot of attention since it is able to characterize independence between a random variable \(Y\) and a random vector \(\textbf{X}\) as well as perfect directed dependence of \(Y\) on \(\textbf{X}.\) Further, it satisfies an information gain inequality and also characterizes conditional independence, which—thanks to a fast estimator—allows a model-free variable selection in almost linear time. In this paper, we compare two multi-output variable selection methods based on recently investigated extensions of Chatterjee’s rank correlation to a multivariate output vector \(\textbf{Y}.\) In an empirical study, we investigate the relevance of the five Fama French factors to explain returns of six big tech companies over the time period from 2013 to 2024.