Optimal prediction of cyclical downturns

Abstract

In the conventional way of visualizing economic time series, the prediction problem does not dichotomize into predicting turning points and predicting conditional means. Yet, the prediction of turning points has always been considered a separate issue in Business Cycle literature. The paper first discusses these issues and then describes a model where the prediction of turning points does separate from the prediction of conditional means. Using the model an 'optimal' way of predicting turning points is derived and applied to U.S. data. The paper uses time series methodology and the theory of optimal stopping times.