Forecasting with a nonlinear dynamic model of stock returns and industrial production

Abstract

We model stock returns and industrial production as nonlinear and state-dependent, with dynamics depending on the sign and magnitude of past realization of returns and the growth of industrial production. We estimate various nonlinear models including smooth transition autoregressive models and examine their in-sample properties. We also conduct an out-of-sample forecasting exercise and compare the forecasting performance of the various nonlinear models with that of a linear model. For stock returns, we find that the linear model generally does as well or better than any of our nonlinear models, while for growth in industrial production, two of our nonlinear models outperformed the linear model.