Professor Yuanhua Feng from University of Paderborn gave a research talk on the issue of “Data-driven estimation of realized volatility under independent microstructure noise” on Sept. 24, 2013.

Professor Feng demonstrated that realized volatility (RV) is a model-free estimator of the daily integrated volatility (IV) based on high-frequency financial data. RV can be estimated in some simple ways. However, it is found that, if the data exhibit microstructure noise (MN), most of the simple definitions of RV are now inconsistent estimators of the IV. Different proposals are introduced to solve this problem. Most recently, Barndorf-Nielsen et sal. (2008, 2009 and 2011) introduced the realized kernels (RK), which are consistent estimates of the IV under give conditions. A crucial point to calculate the RV is the selection of the bandwidth. Our purpose is to propose an iterative plug-in algorithm for selecting the bandwidth for RK under the assumption that the MN are i.i.d. It is shown that the proposed algorithm is a fix-point search method and runs very quickly. To our knowledge this proposal is the first fully data-driven algorithm for selecting the bandwidth for RK. The nice properties of the proposed bandwidth selector are indicated by asymptotic results and application to high-frequency data of a few German and French firms within a period of several years.