Event Details

Paul Ormerod from Volterra Partners LLP and Alliance Business School, University of Manchester will give a presentation to the Management Science Department

Abstract: Over the past 30 years or so, microeconomics has made substantial progress. The same cannot be said of macroeconomics.

One of the problems is that a major task in theory was to import into macro, in the form of DSGE models, a highly simplified representation of the rational agent, at the very time that micro was developing much richer and more realistic models of (rational) agent behaviour.

The main issue, however, which I will discuss is empirical. Empirical evidence in macroeconomics is very fragile. Relationships, no matter how carefully estimated, break down frequently.

One manifestation of this is the failure of macroeconomic forecasting. The Survey of Professional Forecasters in the US, for example, has the track record of a wide variety of institutions and approaches going back to 1968. A regression of the SPF median forecast for GDP growth four quarters ahead on the actual has virtually zero explanatory power. There is no evidence that this is getting better over time.

I illustrate one reason for the instability of empirical relationships by plotting the vector field of a mildly nonlinear system of differential equations. The system away from equilibrium often behaves differently from the system near equilibrium.

The fundamental problem with empirical relationships in macro is that the data is dominated by noise rather than by signal. I illustrate this by the application of random matrix theory, a powerful signal processing tool, to time series data on both GDP growth and the change in inflation.