An iterative plug-in algorithm for realized kernels

Realized kernels introduced by Barndorff-Nielsen et al. (2008) are consistent estimators of the daily integrated volatility in the presence of microstructure noise. A crucial problem by applying realized kernels is the selection of the bandwidth. This paper proposes an iterative plug-in algorithm to solve this problem under independent microstructure noise, which adapts the idea of Gasser et al. (1991) in nonparametric regression to the current context. It is shown that the selected bandwidth is consistent up to a bias factor due to the use of a biased formula of the asymptotically optimal bandwidth. The nice practical performance of the proposal is illustrated by application to data of a few German and French firms within a period of several years. Further analysis of the obtained realized kernels using a most recently proposed exponential SEMIFAR model is also discussed.