Detection of Changes in Panel Data Models with Stationary Regressors

We consider a panel regression model with cross-sectional dimension N. The aim is to test, based on T observations, whether the intercept in the model remains unchanged throughout the observation period. The test procedure involves the use of a CUSUM-type statistic derived from a quasi-likelihood argument. We derive the limit null distribution of the test under strong mixing and stationarity conditions on the errors and regressors, and show that the results remain valid in the presence of weak and dominating cross-sectional dependence. We also propose a self-normalized version of the test which is convenient from a practical perspective in that the estimation of long-run variances is avoided entirely. The theoretical results are supported by a simulation study that indicates that the tests work well in the case of small to moderate sample sizes. An illustrative application of the procedure to US mutual fund data demonstrates the relevance of the proposed procedure in financial settings.