Inference in possibly integrated vector autoregressive models: some finite sample evidence

This paper deals with hypothesis testing in vector autoregressive (VAR) models that may contain some unit roots. We consider situations in which the researcher’s goal is not detecting the presence (absence) of unit roots or their location (i.e., cointegrating relations), but testing some economic hypotheses expressed as coefficient restrictions of VAR models. We investigate the finite sample performance of three testing procedures that are applicable in such situations. After a brief review of these procedures in a general setup, we focus on Granger causality tests as a typical example. We then compare their sampling performance through extensive Monte Carlo simulations.