Estimating the quadratic covariation matrix for asynchronously observed high frequency stock returns corrupted by additive measurement error

This paper studies the estimation problem of the covariance matrices of asset returns in the presence of microstructure noise and asynchronicity between the observations across different assets. Motivated by Malliavin and Mancino (2002, 2009) we propose a new Fourier domain based estimator of multivariate ex-post volatility, which we call the Fourier Realized Kernel (FRK). An advantage of this approach is that no explicit time alignment is required unlike the time domain based methods widely adopted in the existing literature. We derive the large sample properties and establish asymptotic normality of our estimator under some general conditions that allow for both temporal and cross-sectional correlations in the measurement error process. Our results can be viewed as Frequency domain extension of the asymptotic theories for the multivariate realized kernel estimator of Barndorff-Nielsen et al. (2011). We show in extensive simulations that our method outperforms the time domain estimators when two assets with different liquidity are traded asynchronously