Unit Roots in Real GNP: Do We Know, and Do We Care?
NBER Working Paper No. 3130
No, and maybe not. [additional text from author's introduction] To us, the possibility of providing a compelling case that real GMP is either trend or difference stationary seems extremely small, certainly on the basis of post-war data. This is because there is only one difference between these two types of processes and that difference is completely summarized by the answer to the question. How much should an innovation to real GMP affect the optimal forecast of real GMP into the infinite future? If the answer is zero, then real GMP is trend stationary. If the answer is not zero, then real GMP is difference stationary. The competing hypotheses have no other testable differences. Once we pose the question in this way, it seems clear that economists ought to be extremely skeptical of any argument that purports to support one view or the other. Simply put, it's hard to believe that a mere 40 years of data contain any evidence on the only experiment that is relevant.