The asymptotic behavior of some nonparametric change-point estimators (Q1178950)

The author studies the following change-point problem in a random seminorm framework: Let \(X_ 1,X_ 2,\dots,X_ n\) be a sequence of independent random variables, where \(X_ 1,X_ 2,\dots,X_{n\theta}\) have distribution \(P\), and \(X_{n\theta+1},X_{n\theta+2},\dots,X_ n\) have distribution \(Q\). The author proceeds to show that the nonparametric estimators proposed by \textit{E. Carlstein} [ibid. 16, No. 1, 188-197 (1988; Zbl 0637.62041)] and \textit{B. S. Darkhovskij} [Theory Probab. Appl. 21, 178-183 (1976); translation from Teor. Verojatn. Primen. 21, 180-184 (1976; Zbl 0397.62024)] are consistent with rate \(O_ p(n^{-1})\). The improvement of the convergence rate is based on Hotelling's exponential bound for a VC class. He derives the limiting distributions of certain estimators corresponding to four particular seminorms. The limiting distributions are determined by way of some functionals of two-sided Brownian motion with drift. He also compares the efficiencies of nonparametric estimators relative to some semiparametric estimators. Finally, the author proposes a bootstrap procedure for constructing a confidence set for the change point and then applies it to the Nile data.