Deterministic seasonality versus seasonal fractional integration

Abstract

We propose in this article the use of a testing procedure due to Robinson (J. Amer. Statist. Assoc. 89 (1994) 1420) for testing deterministic seasonality versus seasonal fractional integration. A new statistic, based on the score principle, is developed to simultaneously test both the order of integration of the seasonal component and the need of seasonal dummies. Both tests have standard null and local limit distributions. However, finite-sample critical values of the tests are computed, and experiments based on Monte Carlo show that the sizes of the asymptotic tests are too large, these larger sizes being also associated with some superior rejection frequencies compared with the finite-sample-based tests. Using quarterly data for real consumption and income in Canada, the UK and Japan, the results show that both variables are seasonally fractionally integrated for the three countries without need of deterministic seasonal dummies. We also find evidence that the series may be seasonally fractionally cointegrated.