On the ranks of tests having null of cointegration: a Monte Carlo comparison

Abstract

The null of cointegration tests for testing the existence of cointegration are available in literature in great diversity. The selection of a particular test from all these available tests is the crucial problem and often researchers face this. This study is carried out to solve this problem by comparing eight tests on basis two properties of size and power using a new proposed methodology of rank scores. Three different specifications of deterministic component and four sample sizes are considered. It is concluded that if asymptotic critical values are used then it results into an uncontrolled empirical size. While, if the simulated critical values are used, then the empirical size is controlled around nominal size. Moreover, on basis of power, a simple test, which is based on KPSS statistic, is the sole better performer for all of the different 132 different cases of data generations considered in the study.