Forecasting with CGE models

Computable General Equilibrium (CGE) models are highly disaggregate models that encompass many goods, sectors and household types that are used to understand the consequences of various policies on the structure of an economy. Typical applications have been international trade reform and fiscal reform, with also recent applications to environmental issues. They are used extensively for "what if" scenarios, but how well they fare is little studied. The only retrospective I am aware of, on the predictions about the introduction of NAFTA by Timothy Kehoe has shown that CGE models did not predict at all what would happen.

Peter Dixon and Maureen Rimmer try to push the envelope even further with a CGE model and explore how good it is at plain forecasting. They take the USAGE model for the United States and claim that it removes about 40% of the forecasting error that plain extrapolations of trends from 1992 to 1998 into 1998 to 2005 would have given. That sounds great but it would still mean, for example, that a forecast 15.8% forecasted increase in consumption turned out to be 26%, no small difference. Or that exports would increase by 57%, while it was 24%. But the real comparison should be with other forecasting models, in particular dynamic ones.