Abstract
The presence of deterministic or stochastic trend in US GDP has been a continuing debate in the literature of macroeconomics. Ben-David and Papell (Citation1995) found evidence in favour of trend stationarity using the secular sample of Maddison (Citation1991). More recently, Murray and Nelson (Citation2000) correctly criticized this finding arguing that the Maddision data are plagued with additive outliers (AO), which bias inference towards stationarity. Hence, they propose to set the secular sample aside and conduct inference using a more homogeneous but shorter time-span post-WWII sample. In this article we re-visit the Maddison data by employing a test that is robust against AO's. Our results suggest the US GDP can be modelled as trend stationary process
Notes
Note that we do not restrict xt to be stationary, that is, we can also assume that φ(L)=1−L.
This is the value of λ for the Gaussian distribution.
All the data are derived from The World Economy: Historical Statistics OECD Development Centre, Paris 2003, which contains detailed source notes.
The routine to perform the outlier detection test was written in Gauss code. To obtain the code you can send an E-mail to the authors.