Estimating the `cyclical' components of GNP the difference between actual GNP and its trend level is important for various policy purposes: tax revenues and transfers and the budget deficit all depend on the level of GNP. The random walk component which is now perceived in many macroeconomic data series complicates the measurement of the business cycle, however. The appropriate statistical model for the GNP growth rate appears to be a stationary stochastic process, so it could be highly misleading to estimate the business cycle by fitting real GNP to a deterministic trend. This decomposition of GNP into trend and cycle, when GNP has a random walk component, is based on controversial assumptions concerning the structure of the economy.

In Discussion Paper No. 756, George Evans and Research Fellow Lucrezia Reichlin develop a method based on a forecasting perspective, which requires no such assumptions, which they apply to US quarterly data for 1949-90. They use the Beveridge and Nelson (BN) decomposition, which splits the series into a pure random walk with drift (the trend, or long-run forecast) and a stationary stochastic component (the cycle). The authors investigate the effect of including other economic data on the split between trend and cycle, to show that the univariate BN decomposition ascribes less importance than multivariate BN decompositions to the cycle, while enlarging the information set increases a lower bound for the cycle-trend variance ratio. They use the unemployment rate, the savings ratio, and indices of leading coincident indicators to forecast output growth. They then set up a (cointegrated) vector autoregression to estimate the cyclical component of GNP, which can be used to construct conditional forecasts of its growth. They compare these results with those obtained using fewer variables and find that the model that uses all the available information attributes the greatest importance to the cyclical component and also provides the smoothest estimate of the trend.

Information, Forecasts and Measurement of the Business Cycle
George Evans and Lucrezia Reichlin

Discussion Paper No. 756, January 1993 (IM)