Long memory and persistence in aggregate output

Abstract

We examine persistence in U.S. aggregate output by estimating fractionally integrated ARIMA models. These models provide better low-frequency approximations to the Wold representation than previous stochastic specifications, and earlier results on the importance of a permanent component emerge as special cases. We find evidence of long memory, which induces persistence, though this long memory need not be associated with a unit root. Our point estimates indicate that macroeconomic shocks, while persistent, are distinctly less persistent than many earlier studies suggest; however, confidence intervals associated with the long-run response are quite wide.