Outlier Detection in Non-stationary Processes

Abstract

This paper presents a novel non-parametric procedure for detecting outliers in unit root models. The method utilizes order statistics functions of time series, making it more robust and adaptable to a wider range of scenarios without requiring distribution assumptions. The proposed method is evaluated using numerical and Monte Carlo simulations, showing high power in detecting additive outliers while maintaining a low false alarm rate. The method is also tested in scenarios with single or multiple outliers, and a study of real carbon dioxide emissions data from Venezuela confirms its effectiveness in detecting additive outliers in unit root models. The proposed procedure is shown to perform well in various distributions, which is an advantage over existing techniques. Overall, this new method offers significant advantages over current techniques and has the potential for wider application in outlier detection problems.