Economists call for changes to forecasting models

18 June 2014

Finance Graph
© Unknown

A radical overhaul of macroeconomic forecasting models is needed to avoid a repeat of their failures during the 2008 financial crisis, argue Sir David Hendry and Professor Grayham Mizon in a new paper.

The paper exposes, for the first time since the financial crisis, why the models that failed the world in the 2008 crash are fundamentally flawed, rather than in need of adjustment. In a technical publication Sir David and Grayham Mizon show that we now understand why so called dynamic stochastic general equilibrium (DSGE) models, a feature of forecasting systems used by central banks, fail just when they are most needed, primarily because of the mathematical theory on which they are based.

This theory presumes that economic decision rules are not affected by unanticipated future changes. But, the authors argue, crises alter the predicted impacts of shocks, and create events whose effects cannot be calculated using standard mathematical techniques.

While DSGE models take into account regular economic randomness, they are vulnerable to "extrinsic unpredictability", that is, to probability outcomes that cannot be calculated in advance. And in a crisis, economic agents and governments change their plans, causing further shifts which are again not taken into account by DSGE models. While such changes have long been the downfall of economic forecasts, the authors demonstrate how they cause DSGE model analyses to fail at the very moment they are most needed.

To avoid a future debacle, say the authors, empirical evidence must play a much larger role in the formulation of macroeconomic models, as opposed to their being primarily theory-based. A new way of doing this, developed by members of the research group directed by Sir David, is to embed economic theory models in a much broader framework, not just imposing the theory. Valid insights from theory can be retained while allowing for additional empirical features of the real world, such as previous unanticipated shifts, other forces that might alter decisions, and different reaction speeds as events unfold.