Semiparametric estimation and inference for trending I(d) and related processes

This paper deals with estimation and hypothesis testing in stationary and nonstationary models with a linear trend. Using semiparametric estimators, we obtain asymptotic confidence intervals for mean, trend, and memory parameters. The confidence intervals are applicable for a wide class of processes (including some nonlinear processes), exhibit high coverage accuracy and are easy to implement. We also develop joint hypothesis testing for these parameters, when the alternative for the memory parameter is one-sided, but the ones for the deterministic components are two-sided. We use our results to show that US GDP has less memory than is implied by a unit root, and that it evolves around a deterministic trend. This result has important implications for macroeconomic stabilization policies.