Staff Report 92

Specifying Vector Autoregressions for Macroeconomic Forecasting

Robert B. Litterman

Published March 1, 1984

This paper describes a Bayesian specification procedure used to generate a vector autoregressive model for forecasting macroeconomic variables. The specification search is over parameters of a prior. This quasi-Bayesian approach is viewed as a flexible tool for constructing a filter which optimally extracts information about the future from a set of macroeconomic data. The procedure is applied to a set of data and a consistent improvement in forecasting performance is documented.

Download Paper (pdf)