By Warwick J. McKibbin and Andrew Stoeckel
Macro models have come under criticism for their ability to understand or predict major economic events such as the global financial crisis and its aftermath. Some of that criticism is warranted; but, in our view, much is not. This paper contributes to the debate over the adequacy of benchmark DSGE models by showing how three extensions, which are features that have characterized the global economy since the early 2000s, are necessary to improve our understanding of global shocks and policy insights. The three extensions are to acknowledge and model the entire global economy and the linkage through trade and capital flows; to allow for a wider range of relative price variability by moving to multiple sector models rather than a single good model; and to allow for changes in risk perceptions which propagate through financial markets and adjustments in the real economy. These extensions add some complexity to large scale macromodels, but without them policy models can oversimplify things, allowing misinterpretations of shocks and therefore costly policy mistakes to occur. Using oversimplified models to explain a complex world makes it more likely there will be “puzzles”. The usefulness of these extensions is demonstrated in two ways; first, by briefly revisiting some historical shocks to show how outcomes can be interpreted that make sense within a more complex DSGE framework; then, by making a contemporary assessment of the implications from the proposed large fiscal stimulus and the bans on immigration by the Trump administration which have both sectoral and macroeconomic implications that interact.
A frequent criticism of DSGE models is that the simplest model cannot address all sorts of empirical regularities (“puzzles”) that often can be dealt with through appropriate extensions of the model. I think this paper is trying to make the point that some of those extensions should be systematically added to the canonical DSGE model. I do not think I can agree with that. First, adding all sorts of bells and whistles to a model makes it very difficult to understand. As a researcher we should strive to understand why something happens, not just observe that it happens. Models are abstractions of the reality that keep just what is needed to understand the empirical phenomenon. Of course, one can debate whether the assumption of the model are right, and this is what peer review, for example, is supposed to do. Second, having a too complex model makes it difficult to solve. A model that goes after a simple question and need a cluster to run is not a good model. Say you what to study the impact of unemployment insurance extension. Having trade and capital flow linkage in a global economy is then going to be of second or third order importance. Stay with the simplest possible model that can answer your research question.