Abstract
For detecting an abrupt change when observations are continuous and independent, classical methods assume that the observations belong to a parametric family. Nonparametric methods have been devised, but usually they do not guarantee asymptotic optimality at a parametric model of choice. This article presents a nonparametric changepoint detection approach that guarantees compliance with a prespecified lower bound on the ARL to false alarm whatever the underlying distribution is, yet promises asymptotic first-order optimality if the true underlying distribution belongs to a suspected parametric family.
ACKNOWLEDGMENTS
This work was supported by a grant from the Israel Science Foundation and by the Marcy Bogen Chair of Statistics at the Hebrew University of Jerusalem.
Notes
Recommended by A. G. Tartakovsky