A new cusum-type test for the null of cointregration with a model-free limiting null distribution

Resumen

We propose a new and simple to compute semiparametric CUSUM-type statistic based on the sequence of centered and squared OLS residuals from the estimation of a singleequation cointegrating regression model as the basis to test the null hypothesis of cointegration against no cointegration. The main novelty of this testing procedure is that, besides very simple corrections for serial correlation and endogeneity of the integrated regressors and the only use of OLS residuals, the non-standard limiting null distribution is invariant to the number and type of components appearing in the estimated regression. We derive such a limiting null distribution, establish its consistency rate under no cointegration and also present some numerical results obtained by simulation to illustrate its finite-sample size and power properties