Forecasting consumption, income and real interest rates from alternative state space models

The well-known difficulties in forecasting real interest rates are addressed by a comparison between two classes of models: (i) where real interest rates and income are exogeneously determined from a vector autoregression as inputs to a state space model determining consumption, and (ii) where consumption, income and real interest rates are all endogenous. The first model resembles Hall and Flavin, while the second model approximates the dynamics of a real business cycle model similar to Christiano. The authors find that the second model yields superior forecasts of real interest rates. Furthermore, greater smoothness of consumption in the second model has implications for some topical issues of real business cycles and policy.