Discriminating between nonstationary and nearly nonstationary time series models: A simulation study

This paper describes empirical evidence that it may often be possible to very simply discriminate between (homogeneous once-integrated) nonstationary time series models and nearly nonstationary approximations to them. The authors show that, in such a situation, there can be a feature of the serial correlation structure which enables the pair of models to be distinguished with considerable confidence. This observation appears to be useful in the context of the Box-Jenkins identification of time series models; and they believe it opens up a promising field for new research, complementary to the currently well-established tests for unit roots.