New Economist argues that the shift from "coarse" large-scale macroeconometric models to "smooth" general equilibrium type models is only justified if forecasting accuracy improves. He notes: "In my experience, more theoretically sophisticated, medium-term macro models typically produce markedly inferior forecasts for the usual forecasting horizon of 6-18 months."
This pragmatic stance is quite sensible. What matters is that forecasts improve planning for the future. Presumably forecasts that are more accurate (after the fact!) implies systematically better planning.
My own fearful forecast: the smooth approach will eventually yield more reliable forecasts. The inferior performance of smooth models now, arises from as yet a faulty understanding of the nature of sticky prices and adjustment costs in the short-run. It is not yet clear how to efficiently extract Keynes from Walras.
Eventually though we need to get to Walras, and that is why I am betting on this particular horse. I am optimistic about achieving, in the next 10-15 years, a markedly better understanding of market rigidities. For example, a model with a Phillips curve equation is essentially forcing a black-box relationship between inflation and unemployment. Inside that box is some kind sticky wage story. A smooth model though would attempt to open that box and plug in some kind of sticky wage story; one that has - crucially - a happy ending in full employment.
It's the punchline which grabs you.