Trend adjustment prior to testing for the cointegrating rank of a vector autoregressive process (Q2744944)

The authors develop the following approach to testing for the cointegrating rank of a vector autoregressive process according to which: (a) the deterministic trend is removed in the first step, (b) then the test statistics are applied to the trend-adjusted data. For this purpose a simple generalized least squares method is developed for estimating trend parameters and versions of likelihood ratio (LR) and Lagrange multiplier (LM) type tests based on trend-adjusted date are proposed.NEWLINENEWLINENEWLINEThe asymptotic properties of LR and LM statistics are investigated under the null hypothesis and under local alternatives. Small sample path properties of the new tests are investigated by means of a small simulation experiment. The simulation study demonstrated that tests based on trend-adjusted data have considerably more power in small samples than the competing standard LR test. Generally the LR type versions of the new tests outperform the LM type versions. Since in some cases the standard LR tests are superior to the new tests in terms of power it is recommended that the old and new tests are used simultaneously in practice.