The Effects of Additive Outliers on Tests For Unit Roots and Cointegration

Abstract

This article discusses the properties of the univariate Dickey-Fuller test and the Johansen test for the cointegrating rank when there exist additive outlying observations in the time series. We provide analytical as well as numerical evidence that additive outliers may produce spurious stationarity. Hence the Dickey-Fuller test will reject a unit root too frequently and the Johansen test will indicate too many cointegrating vectors. The results easily generalize to models with ''temporary change'' outliers. Through an empirical example we discuss how additive and temporary change outliers can be detected in practice, and we show how dummy variables can be used to remove the influence of such extreme observations.