Realized Variance and IID Market Microstructure Noise

We analyze the properties of a bias-corrected realized variance (RV) in the presence of iid market microstructure noise. The bias correction is based on the first-order autocorrelation of intraday returns and we derive the optimal sampling frequency as defined by the mean squared error (MSE) criterion. The bias-corrected RV is benchmarked to the standard measure of RV and an empirical analysis shows that the former can reduce the MSE by 50%-90%. Our empirical analysis also shows that the iid noise assumption does not hold in practice. While this need not affect the RVs that are based on low-frequency intraday returns, it has important implications for those based on high-frequency returns