Spurious Long-Horizon Regression in Econometrics

This paper extends recent research on the behaviour of the t-statistic in a long-horizon regression (LHR). We assume that the explanatory and dependent variables are generated according to the following models: a linear trend stationary process, a broken trend stationary process, a unit root process, and a process with a double unit root. We show that, both asymptotically and in finite samples, the presence of spurious LHR depends on the assumed model for the variables. We propose an asymptotically correct inferential procedure for testing the null hypothesis of no relationship in a LHR, which works whether the variables have a long-run relationship or not. Our theoretical results are applied to an international data set on money and output in order to test for long-run monetary neutrality. Under our new approach and using bootstrap methods, we find that neutrality holds for all countries.