Creator: Kilian, Lutz and Ohanian, Lee E. Series: Staff report (Federal Reserve Bank of Minneapolis. Research Department) Number: 244 Abstract:
Unit root tests against trend break alternatives are based on the premise that the dating of the trend breaks coincides with major economic events with permanent effects on economic activity, such as wars and depressions. Standard economic theory, however, suggests that these events have large transitory, rather than permanent, effects on economic activity. Conventional unit root tests against trend break alternatives based on linear ARIMA models do not capture these transitory effects and can result in severely distorted inference. We quantify the size distortions for a simple model in which the effects of wars and depressions can reasonably be interpreted as transitory. Monte Carlo simulations show that in moderate samples, the widely used Zivot-Andrews (1992) test mistakes transitory dynamics for trend breaks with high probability. We conclude that these tests should be used only if there are no plausible economic explanations for apparent trend breaks in the data.
Keyword: Transitory Shocks, Unit Roots, and Trend-Breaks Subject (JEL): C15 - Statistical Simulation Methods: General, E32 - Business Fluctuations; Cycles, and C22 - Single Equation Models; Single Variables: Time-Series Models; Dynamic Quantile Regressions; Dynamic Treatment Effect Models; Diffusion Processes