Comparison of \(U\)-statistics in the change-point problem and in sequential change detection (Q5933706)

Two types (\(H_A^{(1)},H_A^{(2)}\)) of change point alternatives of i.i.d. data are considered. In the first case, for a sample of size \(n\), the change point \(\tau\) is to be retroactively determined. In the second case, the data come in one by one and decision is made after a random number of observations, i.e. a sequential test is performed. Using the large sample behaviour of the test statistics, the author compares the efficiency of the tests constructed for \(H_A^{(1)}\) with those constructed for \(H_A^{(2)}\). The main result is that for a certain class of local alternatives the sequential change point tests have higher efficiency.