Outliers in time series: An empirical likelihood approach

Abstract

The empirical likelihood method is known to be a flexible and effective approach for testing hypotheses and building confidence regions in a nonparametric setting. This framework is adopted here for dealing with the outlier problem in time series where conventional distributional assumptions may be inappropriate in most cases. The procedure is illustrated by a simulation experiment. The results are also supported by the study of two well-known real-time series data: the fossil marine families extinction rates and the Nile river volume at Aswan 1871–1970.