Saturday, February 7, 2009

Structural papers in seminars

As part of our job search process - yes, Michigan is hiring in economics - we have had several speakers through presenting "structural" papers in various sub-fields such as labor, industrial organization and international macro. Sitting through these job talks has crystallized some thoughts I have been having about good ways, and less good ways, to answer seminar questions about such papers.

For non-economist readers, a "structural" paper in economics lays out a complete, formal mathematical model of some aspect of economic behavior (or, in the case of macro, perhaps of the entire economy). These models typically then get estimated or calibrated (don't ask - that is a topic for another day) and their implications for our understanding of individual behavior and/or of potential policy reform are discussed.

The key issue is that structural models are quite complex, both to write down and to estimate. As a result, much of the game in writing such a paper lies in figuring out which aspects of the phenomenon under study to explicitly include in the model and which to assume away or at least grossly simplify. For example, in some particular area of behavior there might to 30 factors that theory, the existing empirical literature and casual introspection suggest might be important. Intellectual and computational factors may, however, mean that the model that gets written down can include only, say, 10 of these. So the trick for the researcher, then, is to pick the "correct" 10 out of 30, where this choice will be guided both by considerations of relative importance in some absolute sense and by which aspects the behavior being studied the researcher most wants to emphasize (which might depend on, for example, a particular policy proposal of interest).

In seminar presentations of such papers, many questions from the audience will be of the form "why did you leave out factor #16" from your model?" A common response to such questions from presenters can be paraphrased as "I had already included 10 factors and so my model was full." In my view, audiences should not tolerate this type of answer. The answer you want, and which seminar audiences should demand, would take the form "The reason I did not include #16 is that it is less important than #8, which was the marginal factor that I did include. It is less important than #8 because of X, Y and Z," where X, Y and Z are reasons based on the existing theoretical and empirical literature or that have to do with point that the author is trying to make with the model. Such answers do a much better job of showing that the researcher has thought seriously about what to include and what not to include in his or her model.