A Model Selection Approach to detect Seasonal Unit Roots

The popular 'airline' model for a seasonal time series assumes that a variable needs double differencing, i.e. first and seasonal (or annual) differencing. The resultant time series can usually be described by a low order moving average model with estimated roots close to the unit circle. This latter feature complicates the standard autoregression-based tests for (seasonal) unit roots which are often used in practice.<br> In this paper we propose an alternative route to detect seasonal unit roots by analyzing (versions of) the basic structural model [BSM]. This BSM can generate data which are (approximately) observationally equivalent to data generated from an airline model. Using Monte Carlo simulations, we show that our method works very well. We illustrate our approach for a large set of macroeconomic time series variables.