Tests for seasonal unit roots in panels of cross-sectionally correlated time series

Abstract

For panel models of cross-sectionally correlated time series, seasonal unit root tests are constructed for each seasonal frequency. The tests are based on instrumental variable estimators which are modifications of signs of the regressors. Cross-sectional correlation is controlled by rotating the system of time series using an estimated error covariance matrix. The limiting null distributions of the tests are chi-squared and are free from nuisance parameters arising from cross-sectional correlation. A Monte-Carlo experiment compares size and power performances of the proposed tests. © 2009 Taylor & Francis.