A nonparametric test for a constant correlation matrix

TL;DR

This paper introduces a nonparametric test to detect changes in correlation matrices over time, requiring minimal assumptions and demonstrating strong finite-sample power, with applications to stock return data.

Contribution

It develops a novel nonparametric method for testing correlation matrix stability that works under weak dependence assumptions and provides theoretical and empirical validation.

Findings

The test effectively detects changes in correlation matrices.

It has good finite-sample power in simulations.

Applied to stock returns, it identifies periods of correlation shifts.

Abstract

We propose a nonparametric procedure to test for changes in correlation matrices at an unknown point in time. The new test requires only mild assumptions on the serial dependence structure and has considerable power in finite samples. We derive the asymptotic distribution under the null hypothesis of no change as well as local power results and apply the test to stock returns.