Predicting a crisis: a DSGE perspective

The last global recession raised concerns about the ability of macroeconomists to predict crises of such magnitude. Certainly, the forecasting is not the main focus of macroeconomics. I would say that, at least nowadays, it is of rather marginal interest to academic macroeconomists. A proof in this sense is provided by the very low number of publications related to macroeconomic forecasting.

I discussed some issues related to the topic of predicting the crisis in this previous post. Here, I rather focus on the issue of whether DSGE models could have predicted the crisis. Del Negro et al. (2013) think that they could. They took a rather standard DSGE model (in the tradition of Smets and Wouters 2003, 2007) which they augmented with financial frictions (to be more precise, a financial accelerator a la Bernanke, Gertler and Gilchrist, 1999). When they performed out of sample forecasts starting with 2008, they found that their model can replicate in a very good manner the path of output and in a quite acceptable manner the dynamics of inflation. While this may seem great news for the DSGE approach, there are many issues which remain unanswered (I admit that the paper is preliminary, however the directions in which the paper is developed are quite clear).

First of all, although the model predicts in a reasonable manner the last crisis, we don’t know how it would perform for the other recessions. Should we think that all recessions are alike? If this model is useful in predicting only a certain type of recessions, then, it might not be that useful in the future.

Second, the model can predict the crisis only once the new information about the Lehman collapse is available, however a major criticism of mainstream economics was that it did not foresaw the crisis (including the Lehman collapse) in due time. What the model does is to show that once certain conditions (for a possible recession!) are available, then a DSGE model could predict a crisis.

Third, financial frictions came into fashion exactly in the aftermath of the crisis, so the model they proposed was not quite standard although it comprised many standard assumptions. On top of that, there are currently numerous ways to introduce a financial and a banking sector in a DSGE model. Would all these alternatives yield the same results?

Not at last, the models generally used by central banks are much more complicated than the Smets – Wouters plus financial frictions model in this paper. Would adding certain features affect the forecasting performance of this model? What about the case of small open economies (the papers used a closed economy)?

In conclusion, the contribution by Del Negro et al. (2013) is certainly interesting but does not offer a pertinent answer to whether the crisis could have been predicted, since it only shows that when certain financial information became available, a model could have predicted the macroeconomic dynamics of US. Moreover, it does not save the DSGE approach from criticism, since what they showed was that a certain DSGE model (which would never be used by everybody in this exact specification) could have predicted the recession conditional on some information available. The research is also limited by the lack of exercises for other recessions. If the authors think that this model would apply only for the last recession, then why would we expect that somebody could predict the next crisis using a DSGE model (probably based on assumptions tailored for the specific next recession…)?

Leave a Reply

Your email address will not be published. Required fields are marked *