On the detection of structural changes in volatility dynamics with applications
This study aimed to address a series of issues regarding both the CUSUM-type tests that have been developed to detect a potentially existing break in the conditional variance of an economic process, and their corresponding breakdate estimators that are used to identify the time of occurrence of that break. The approach taken has been based on an extensive Monte-Carlo simulation to evaluate the size, power and accuracy of the corresponding breakdate estimator of a set of 27 test variants for approximately 450 data generating processes, some homoskedastic but most conditionally heteroskedastic. The first set of issues is concerned with the finite sample properties of the tests, while the second one focuses on the choice of the appropriate test which depends on the assumptions made regarding the conditional variance of the underlying process. A third set of issues concerns the ability of the tests to detect breaks in the conditional mean of the process as well as the conditional variance. In addition to investigating the theoretical properties of the tests, four applications are included amongst which are a note on the work of Andreou and Ghyssels (2002), who perform a Monte-Carlo analysis with features common to those of this study, albeit with some discrepancies in the results, and three empirical studies that signify the importance of testing for structural changes in processes prior to any further analysis and provide suggestions about robustifying results obtained by the use of the CUSUM-type tests.