Crisis Date Identification and Testing for Contagion across Stock Markets

Abstract

This paper studies whether the choice of the crisis start dates affects the magnitude of contagion estimates. Contagion models generally use exogenously determined crisis start date by relying on event-based markers. We conduct structural break tests and endogenously determine the start dates of the global financial crisis for markets in three regions. We then estimate models with regime switching that incorporates these start dates to test for contagion. We present evidence in favour of contagion through correlation and coskewness. Finally, we evaluate whether there are differences in estimates based on contagion models with exogenously or endogenously determined crisis start dates. We find that there are substantial differences in estimates and that the estimation error in correlation is trivial, but enormous for coskewness. We show that properly identifying the crisis start date through econometric tests is crucial for avoiding potential bias from sample selection and estimation errors induced by this bias.