Deterministic Seasonality in Dickey-Fuller Tests: Should We Care?

This paper investigates the properties of Dickey-Fuller tests for seasonally unadjusted quarterly data when deterministic seasonality is present but it is neglected in the test regression. While for the random walk case the answer is straightforward, an extensive Monte Carlo study has to be performed for more realistic processes and testing strategies. The most important conclusion is that the common perception that deterministic seasonality has nothing to do with the long-run properties of the data is incorrect. Further numerical evidence on the shortcomings of the general-to-specific t-sig lag selection method is also presented.