The 2001 US recession: what did recession prediction models tell us?

BIS Working Papers No 148
March 2004
How predictable was the recent US recession? This paper evaluates the accuracy of several recession prediction models. In particular, traditional rule-of-thumb models using the composite index of leading indicators (CLI), Neftçi's sequential probability model, a probit model, and Stock and Watson's experimental recession indexes are compared. Despite the relatively mild depth of the recession, the models using the CLI performed particularly well. The results are robust across different types of models and with respect to the use of real-time data. The strong real-time performance stands at odds with earlier sceptical claims about the marginal usefulness of the CLI in predicting cyclical turning points, and complements the results in the earlier research of Filardo (1999). At a more conceptual level, the paper provides general support to the classical business cycle view that turning points of business cycles from expansion to recession are complex, possibly endogenous and nonlinear, phenomena. The results also suggest that the impressive insights of Geoffrey Moore into the theory and construction of the CLI will continue to shape our understanding of business cycles well into the future.