Detecting structural changes in large portfolios

Abstract

Model-free tests for constant parameters often fail to detect structural changes in high dimensions. In practice, this corresponds to a portfolio with many assets and a reasonable long time series. We reduce the dimensionality of the problem by looking at a compressed panel of time series obtained by cluster analysis and the principal components of the data. With this procedure, we can extend tests for constant correlation matrix from a sub-portfolio to whole indices, which we exemplify using a major stock index.