Linear prediction and estimation methods for regression models with stationary stochastic coefficients

Abstract

A linear regression model is proposed in which the coefficient vector is a weakly stationary multivariate stochastic process. The model provides a convinient representation of a general class of nonstationary processes. Prediction and estimation methods are proposed that are linear and relatively easy to compute. The proposed procedures are illustrated by estimation of time-varying GNP multipliers of several macro policy instruments over the period 1891-1970. The results are compatible with theoretical priors and suggest that predictability of policy outcomes depends on the mixture of policy instruments.