Working Paper: CEPR ID: DP756
Authors: George Evans; Lucrezia Reichlin
Abstract: The Beveridge-Nelson (BN) technique provides a forecast-based method of decomposing a variable such as output, into trend and cycle when the variable is integrated of order one (I (1)). This paper considers the multivariate generalization of the BN decomposition when the information set includes other I (1) and/or stationary variables. We show that the relative importance of the cyclical component depends on the information set, and in particular that multivariate BN decompositions necessarily ascribe more importance to the cyclical component than does the univariate decomposition, provided the information set includes a variable which Granger-causes output. We illustrate the results for post-war data for the United States.
Keywords: trend; cycle; forecast; information; integrated series; granger causality; business cycles
JEL Codes: C32; E32
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| information set used for forecasting (C53) | cyclical component significance (E32) |
| multivariate BN decomposition (C39) | cyclical component significance (E32) |
| additional macroeconomic variables (E19) | cyclical component (E32) |
| variables that Granger-cause output (C29) | cyclical component significance (E32) |
| broader information set (D89) | forecasting of output growth (O40) |