摘要：In this paper, we test the presence of stochastic trend in long series of US real GNP measured by Balke and Gordon (1989) and Romer (1989), using unit root tests robust against breaks and outliers. We apply two recent robust unit root tests proposed by Cavaliere and Georgiev (2009) and Lima and Xiao (2010), for which critical values are adapted to the small sample size and using optimal lag selection methods. The former is improved by selecting optimally GLS detrending parameter to make the test in small samples powerful. We obtain mixed results on the presence of a unit root in the GNP and GNP per capita series on the 1869-2007 period. Even if we cannot conclude on the pre-1929 period, probably because of the reconstruction on the data, we show that the US real GNP series is characterized by a deterministic trend on the post-1929 period.
