Detecting big structural breaks in large factor models

Constant factor loadings is a standard assumption in the analysis of large dimensional factor models. Yet, this assumption may be restrictive unless parameter shifts are mild. In this paper we develop a new testing procedure to detect big breaks in factor loadings at either known or unknown dates. It is based upon testing for structural breaks in a regression of the first of the ¯r factors estimated by PC for the whole sample on the remaining r−1 factors, where r is chosen using Bai and Ng´s (2002) information criteria. We argue that this test is more powerful than other tests available in the literature on this issue.