Comment on “ An unbiased measure of realized variance ” and “ Realized variance and market microstructure noise ”

If efficient asset prices follow a semi-martingale and are perfectly observed, their quadratic variation can be measured accurately from the sum of a large number of squared returns sampled over very finely spaced intervals, i.e., realized variance (Andersen et al., 2003, and Barndorff-Nielsen and Shephard, 2002). With the emergence of high-frequency data, it seems that we should be able to identify volatility rather easily. However, this identification hinges on being able to observe the true (or efficient) price process. Unfortunately, observed asset prices are affected by market microstructure effects, such as discreteness, different prices for buyers and sellers, “price-impact” of trades, and so forth. If we think of observed prices as efficient prices plus market microstructure noise contaminations, then we face an interesting as well as complex econometric task in using high-frequency data to estimate quadratic variation from noisy observed asset price data. Hansen and Lunde make several nice contributions to this growing literature. First, they document empirical evidence regarding the dynamic features of the noise including evidence about the nature of the dependence between the efficient price and microstructure noise. Second, they propose a clever procedure to remove microstructure-induced biases from realized variance estimates. We start with the former.