Detecting change point for a sequence of random vectors under nonnormality (Q4516137)

The paper deals with testing and estimation of change points in mean vectors and covariance matrices for a sequence of random vectors. Let \(x_1,\ldots, x_{n}\) be a sequence of independent \(m\)-dimensional random vectors, and let \(\mu_1,\ldots, \mu_{n}\) and \(\Sigma_1,\ldots, \Sigma_{n}\) be the corresponding mean vectors and covariance matrices. The problem is to test the following hypothesis \(H_0:\;\mu_1=\ldots=\mu_{n}\) (unknown) and \(\Sigma_1=\ldots=\Sigma_{n}\) (unknown) versus the alternative \(H_1:\;\mu_1=\ldots=\mu_{k^{*}}\neq\mu_{k^{*}+1}=\ldots=\mu_{n}\) and \(\Sigma_1=\ldots=\Sigma_{k^{*}}\neq\Sigma_{k^{*}+1}=\ldots= \Sigma_{n}\), where \(1\leq k^{*}\leq n-1\) is the unknown position of the change point. The authors derive the likelihood procedure test statistic for \(H_0\) versus \(H_1\) and show that its asymptotic null distribution without normality is a Gumbel distribution in the case when the position of the change point is located between \(m\) and \(n-m\).