In this paper we propose a test for seasonal unit roots which, under the presence of first order moving average errors, has a better empirical size and power than the Hyllerberg et al. (1990) (HEGY) test. Two alternatives are considered for this purpose. The first, suggested by Hall (1990), is based on instrumental variable methods. The second consists on developing a Hausman-type test (Hausman (1978) and Choi (1990, 1992)). Using Monte Carlo simulations, we show that the HEGY test with instrument variables does not represent an improvement with respect the original HEGY test under the presence of moving average. On the other hand, the Hausman test (HT) does present better properties than the HEGY test. The test shows correct size no matter what the value of the moving average coefficient is. In addition, for all cases studied, the HT test had better power than the HEGY test.
