Reply to our discussants

Abstract

Let us begin by thanking our eight discussants. It is an unexpected blessing to have received careful and extensive consideration from such a constructive and fair-minded group. Our reaction to their comments is perhaps better understood with some autobiographical perspective. As is probably evident to any reader, we are not econometricians. This is not to offer an excuse for any shortcomings of our work, but to explain that it was motivated by interests somewhat different than the typical interests of econometricians. We are both applied monetary/macroeconomists, and one of us is particularly interested in the philosophical and methodological problems of empirical economics. We have both done applied work on causal inference in macroeconomics, using concepts and techniques substantially different from those that originate in the work of Clive Granger. In our causal investigations, we have used LSE specification procedures. Our experience has been, contrary to our initial expectation, that the validity of these procedures, which we took to be fairly well established, was more often questioned by referees than was the validity of the causal methodologies that we proposed, and which we regarded as novel. We were faced with a dual problem: on the one hand convincing referees and editors that our applied work made sense; and, on the other hand, assuaging the doubts that were naturally raised in our own minds by intelligent critics. Thus, while we had a natural predisposition towards the LSE approach, we believe that we were genuinely ready to acknowledge whatever results we obtained. Interestingly, while LSE econometricians have generally greeted our paper enthusiastically, opponents of the LSE approach often see the results that we report—even taken at face value—as lending little comfort to LSE practitioners. The key thing to remember is that our goal was not to provide a generalized tool for specification search, but to ask whether it worked in a case in which we thought the LSE approach should work, if it could ever work at all. The five commentaries raise a large number of points. We will not attempt to address them all. Rather, in the remainder of the reply, we would like to consider only some of the larger themes.