Paul Krugman recently participated in a discussion about the current state of macroeconomics, particularly about the “dominant” paradigm of DSGE models and their predecessor, the IS-LM model. Using DSGE models by myself and disagreeing with basically all of what he said, let me comment on an especially unconvincing piece of reasoning:
“[…] how [do] we know that a modeling approach is truly useful. The answer, I’d suggest, is that we look for surprising successful predictions. So has there been anything like that in recent years? Yes: economists who knew and still took seriously good old-fashioned Hicksian IS-LM type analysis made some strong predictions after the financial crisis”.
In short: forget DSGE models and related stuff and go back to IS-LM because people using IS-LM recently made some “right” predictions. Is the exclusive focus on its “predictions”a reasonable criterion to judge (macro)economic models or how should they be judged else?
With respect to the former, let me give you an admittedly extreme example. Over the last few decades, we could have very well predicted the EU agricultural policy by assuming that the aim of policy makers was to reduce consumer welfare. Would you resort to this sort of model when discussing likely upcoming agricultural policies from an outsider perspective? I would not.
Can we take up another extreme and simply judge a model based on the realism of its assumptions? I suggest we can’t. “Let’s assume a representative agent who cares about consumption, leisure and wearing blue jeans” to take another extreme example. Would you reject a macroeconomic argument formulated with this agent based on the unrealism of the blue jean assumption (as long as the model is not intended to explain the jeans market). I would not because in this context, I regard this assumption to be irrelevant for the models conclusions/predictions. The difference with respect to the first example is then of course that in the former, I’m convinced that the assumption matters for the results, in the latter I´m convinced it does not.
So one cannot judge models solely by their predictions because the underlying assumptions in combination with the implied propagation mechanisms might be clearly implausible/unconvincing and important for the predictions. Assessing the latter in turn requires to dig into the model dynamics implying that one can also not base a judgement solely on the realism of the assumptions.
How can a reasonable criterion that takes the above findings under consideration then look like? As so beautifully described in McCloskey essay on the “Rhetoric of Economics”, (macro)economists are persuaders. What they do is “careful weighing of more or less good reasons to arrive at more or less probable or plausible conclusions – none too secure, but better than would be arrived by chance or unthinking impulse; it is the art of discovering warrantable beliefs and improving those beliefs in shared discourse […] (Booth 1961, see the above link p. 483 for the exact citation)”. The purpose of models, I’d argue is then to help organizing that discourse, help to structure your own thinking, clarify the debate and provide a framework to confront thought with data. In order to be useful for that, they need to be “persuasive” in the context they are used. The realism of the assumptions and the implied propagation mechanisms, their respective importance for the results and the model´s fit to data are all part of the subjective assessment with respect to that criterion.
How then about DSGE vs. IS-LM in policy debate? IS-LM and related models were mainly discarded in policy analysis because they incorporate reduced form parameters on e.g. the interest rate responsiveness of investment or the marginal propensity to consume which most economists were not convinced to be sufficiently independent of economic policy. This criticism is today as valid as it was 30 or 40 years ago, none of that has changed. All that has changed is that IS-LM made some “correct” (one may very well discuss the use of this word here but that´s not the purpose of this blog entry) predictions. Shall we go back to a model that was labelled as an unpersuasive tool although the main point of criticism is still valid – NO. Shall we use it now lacking a better alternative? Properly specified state-of-the-art DSGE models are a tool that outclasses IS-LM on virtually every aspect (YES even when it comes to rational expectations). For sake of shortening the entry, I will yet argue my case for that in a follow-up post.