On the Existence and Interpretation of the "Unit Root" in U.S. Gnp

We use the revised estimates of U.S. GNP constructed by Christina Romer (1989) to assess the time-series properties of U.S. output per capita over the past century. We reject at conventional significance levels the null that output is a random walk in favor of the alternative that output is a stationary autoregressive process about a linear deterministic trend. The difference between the lack of persistence of output shocks either before \VWII or over the entire century, on the one hand, and the strong signs of persistence of output shocks found by Campbell and Mankiw (1987) and by Nelson and Plosser (1982) for more recent periods is striking. It suggests to us a Keynesian interpretation of the large unit root in post-WWII U.S. output: perhaps post-WWII output shocks appear persistent because automatic stabilizers and other demand-management policies have substantially damped the transitory fluctuations that made up the pre-WWH Bums-Mitchell business cycle