The Effects of Additive Outliers and Measurement Errors when Testing for Structural Breaks in Variance

This paper discusses the asymptotic and finite-sample properties of CUSUM-based tests for detecting structural breaks in volatility in the presence of stochastic contamination, such as additive outliers or measurement errors. This analysis is particularly relevant for financial data, on which these tests are commonly used to detect variance breaks. In particular, we focus on the tests by Inclán and Tiao [IT] (1994) and Kokoszka and Leipus [KL] (1998, 2000), which have been intensively used in the applied literature. Our results are extensible to related procedures. We show that the asymptotic distribution of the IT test can largely be affected by sample contamination, whereas the distribution of the KL test remains invariant. Furthermore, the break-point estimator of the KL test renders consistent estimates. In spite of the good large-sample properties of this test, large additive outliers tend to generate power distortions or wrong break-date estimates in small samples.