Integrated Covariance Estimation using High-frequency Data in the Presence of Noise

We analyze the effects of nonsynchronicity and market microstructure noise on realized covariance type estimators. Hayashi and Yoshida (2005) propose a simple estimator that resolves the problem of nonsynchronicity and is unbiased and consistent for the integrated covariance in the absence of noise. When noise is present, however, we find that this estimator is biased, and show how the bias can be corrected for. Ultimately, we propose a subsampling version of the bias-corrected estimator which improves its efficiency. Empirically, we find that the usual assumption of a martingale price process plus an independently and identically distributed (i.i.d.) noise does not describe the dynamics of the observed price process across stocks, which confirms the practical relevance of our general noise specification and the estimation techniques we propose. Finally, a simulation experiment is carried out to complement the theoretical results. Copyright 2007, Oxford University Press.