Monitoring multivariate variance changes

Abstract

We propose a model-independent multivariate sequential procedure to monitor changes in the vector of componentwise unconditional variances in a sequence of $p$-variate random vectors. The asymptotic behavior of the detector is derived and consistency of the procedure stated. A detailed simulation study illustrates the performance of the procedure confronted with different types of data generating processes. We conclude with an application to the log returns of a group of DAX listed assets.