A Note on Hypothesis Testing Based on the Fully Modified Vector Autoregression

This paper investigates the sampling performance of hypothesis tests based on the fully modified vector autoregression (FM-VAR) that has recently been developed by Phillips (1995). The FM-VAR procedure is applicable without any prior knowledge about the number and location of unit roots. We consider Granger causality tests as a typical example to which the FM-VAR approach could usefully be applied. Through Monte Carlo experiments, we examine whether the rejection frequencies of the tests under the null hypothesis are close enough to a desired significance level for sample sizes that are typically available to economists.