Small sample testing for cointegration using the bootstrap approach

Recently, a number of authors have proposed using the bootstrap procedure for small sample testing procedure for unit roots and cointegration, based on bootstrapping the augmented (multivariate) Dickey–Fuller test. The present paper considers bootstrapping the Johansen test for an (unknown) number of cointegration relationships. This in effect amounts to undertaking the same type of Monte Carlo work that generated the Johansen tables of critical values, but for the purposes of calculating critical values relevant to a particular data set (based on unknown d.g.p.'s). However, our results suggest that the bootstrap test statistic has poor size properties.