"Different Structural Models Can Fit Aggregate Macroeconomic Data About Equally Well"
(p. 1149) There is an apparent lack of encompassing-forecasting and economic models that can explain the facts uniformly well across business cycles. This is perhaps an inevitable outcome given the changing nature of business cycles. The fact that business cycles are not all alike naturally means that variables that predict activity have a performance that is episodic. Notably, we find that term spreads were good predictors of economic activity in the 1970s and 1980s, but that credit spreads have fared better more recently. This is of course a challenge for forecasters, as we do (p. 1150) not know the origins of future business cycle fluctuations. Much needs to be learned to determine which and how financial variables are to be monitored in real time especially in an evolving economy when historical data do not provide adequate guidance.
Explanations for the Great Recessions usually involve some form of nonlinearity. The sudden nature of the downturn following the collapse of Lehman is consistent with nonlinearity being part of the transmission mechanism. At the same time, we lack robust evidence of nonlinearity from aggregate low-frequency macroeconomic data. Essentially, there is an identification issue as different structural models can fit aggregate macroeconomic data about equally well.
For the full article, see:
Ng, Serena, and Jonathan H. Wright. "Facts and Challenges from the Great Recession for Forecasting and Macroeconomic Modeling." Journal of Economic Literature 51, no. 4 (Dec. 2013): 1120-54.