Exact test for breaks in covariance in multivariate regressions

We propose an exact test for breaks in covariance in multivariate regressions. The test is based on the LR criterion from Anderson ([Anderson, T.W. (1971), The Statistical Analysis of Time Series, John Wiley and Sons, New York.], chapter 10), extended to account for regression covariates and unknown break dates, under general parametric Gaussian and possibly non-Gaussian distributional assumptions.