MONEY GROWTH AND INFLATION IN THE UNITED STATES

Abstract

Specification tests reject a linear inflation forecasting model over the period 1959–2002. Based on this finding, we evaluate the out-of-sample inflation forecasts of a fully nonparametric model for 1994–2002. Our two main results are that: (i) nonlinear models produce much better forecasts than linear models, and (ii) including money growth in the nonparametric model yields marginal improvements, but including velocity reduces the mean squared forecast error by as much as 40%. A threshold model fits the data well over the full sample, offering an interpretation of our findings. We conclude that it is important to account for both nonlinearity and the behavior of monetary aggregates when forecasting inflation.