Bayesian analysis of long memory and persistence using ARFIMA models

This paper provides a Bayesian analysis of Autoregressive Fractionally Integrated Moving Average (ARFIMA) models. We discuss in detail inference on impulse responsens, and show how Bayesian methods can be used to (i) test ARFIMA models against ARIMA alternatives and (ii) take model uncertainty into account when making inferences on quantities of interest. Our methods are then used to investigate the persistence properties of real U.S. GNP.