The Effect of Varying Length Var Models on the Maximum Likelihood Estimates of Cointegrating Vectors

This paper takes the model of Drobney and Hall (1989), which investigated nondurable consumption in the United Kingdom within a standard Engle-Granger framework, and applies the Johansen procedure to it. It then goes on to investigate the effect of changing the order of the vector autoregressive system on both the estimates of the cointegrating vector and on the test procedures given by the maximum likelihood approach. The conclusions are that while the small sample properties of the maximum likelihood estimator seem good, the power of the test statistics seems to be highly dependent on the specification of the vector autoregression.