Abstract
This article investigates whether the US unemployment rate is best described as a unit-root or mean-reverting process. An out-of-sample forecast exercise is conducted in which the performance of an autoregressive (AR) model with an imposed unit root is compared with that of a mean-reverting AR model. A bootstrap distribution for the relative root mean square forecast error is generated and provides strong support for mean reversion in the US unemployment rate.
Notes
1This opinion has also been expressed by, for example, MacDonald and Taylor (Citation1992) and Diebold and Mariano (Citation1995).