Il test bootstrap esterno per la ricerca di radici unitarie in presenza di outliers

Abstract

The Dickey-Fuller test is a usefull statistical tool to detect unit roots in time series. This paper compare the robustness of the Dickey-Fuller test and its bootstrap extention from two different point of view. On one hand we study the conseguences of the violations of the residuals omoschedasticity and incorrelation hypotheses, on the other hand we analize the effect of the additive and innovative outliers contamination of time series. The results point out that the bootstrap test give us better performace than the Dickey-Fuller one when the time series is contaminated by outliers and in some cases also when the residuals are correlated or heteroschedastic.