Conclusion
This paper examines the performance of the multiple period drift-independent Yang and Zhang (2000) volatility estimator, the YZ estimator and the demeaned squared returns in detecting sudden breaks in volatility using the IT-ICSS algorithm by means of Monte Carlo simulation experiments. Using data generating processes from sequence of i.i.d. random numbers (the Gaussian, the Student’s t, the double exponential, the gamma-mixture and the generalised error distributions), the generalised autoregressive conditional heteroskedasticity model, the stochastic volatility model and the fractionally integrated GARCH model, this study assesses the size and power properties of the YZ estimator and the demeaned squared returns. The findings from Monte Carlo simulation experiments indicate that the YZ estimator exhibits outstanding size and power characteristics when used with the IT-ICSS algorithm. However, the demeaned squared return exhibits oversized behaviour and severe size distortion for most of the data generating processes taken for simulation experiments. This indicates that the IT-ICSS algorithm can detect appropriate sudden breaks in the YZ estimator; however, the sudden breaks detected in the demeaned squared returns may be spurious. To confirm the findings of simulation experiments, this study applies the IT-ICSS algorithm on the YZ estimator and the demeaned squared returns of the USD/Euro, the USD/Japanese yen and the USD/GBP exchange rates to detect sudden changes in the respective volatility proxies. The empirical findings indicate that most of the sudden breaks detected in the YZ estimator can be related to major macroeconomic events. On the other hand, the ITICSS algorithm detects too many breaks in the demeaned squared returns, and most of the detected breaks cannot be related to any macroeconomic events and are probably spurious.
