There is no One Model to Rule Them All

I like the influence that physicists are having on economics. Moving towards agent-based modelling in some areas of the discipline is a great idea. But, in addition to lending their novel insights, some seem to enjoy piling on economics generally. Generally you have to take the good with the bad but Mark Buchanan’s latest article is so shockingly bad that I can’t help picking on it.

Buchanan’s article is intended as a critique of DSGE models. He first alleges that they are poor forecast models and uses their absence from Wall St as evidence of that. He then claims that they are useless for telling stories about the economy because they’re unfalsifiable.

On the first point he is half right: DSGE models have poor forecast performance. What he fails to point out is that there are no other models that do better! DSGE models are the core of the Bank of England’s COMPASS forecast model, the Reserve Bank of New Zealand’s KITT forecast model, the European Commission’s Smet-Wouters forecast model, and others. Sure, they struggle to accurately forecast more than one quarter ahead but they are equally as good as anything else out there.

So why aren’t they used on Wall St if they’re at the cutting edge of central bank forecasting? Largely because time series models provide equally good forecast performance and Wall St analysts have less need for the other benefits of a DSGE framework. Those benefits are its infinitely better performance in analysing competing policy options. The idea behind DSGE is to distil the economy down to ‘deep, structural’ factors; things that don’t change when the Government changes its mind. That is why they are useful for analysing what might happen if the Government did change its mind. Of course, that structural analysis takes a lot of extra effort and isn’t going to improve your forecasts but it is crucial for Government and central bank analysis. Different tasks require different tools.

Economists have spent a lot of time thinking about how to analyse policy questions and construct meaningful counterfactuals. The influence of physicists may well help us build better models but we shouldn’t be so bamboozled by their technical prowess that we throw out the expertise economists have spent over a century building.