Efficient Estimation of Integrated Volatility and Related Processes

We derive nonparametric bounds for inference about functionals of high-frequency volatility, in particular, integrated power variance. In the absence of microstructure noise, we find that standard Realized Variance attains the nonparametric efficiency bound, also in case of unequally spaced random observation times. For higher powers, e.g., integrated quarticity, the block-based procedures of Mykland and Zhang (2009) can get arbitrarily close to the nonparametric bounds in case of equally spaced observations. The estimator in Jacod and Rosenbaum (2013) is efficient, also at non-constant volatility, still for equally spaced data. For unequally spaced data, we provide an estimator, similar to that of Kristensen (2010), that can get arbitrarily close to the nonparametric bound. Finally, contrary to public opinion, we demonstrate that parametric information about the functional form of volatility generally leads to a decreased lower bound, unless the volatility process is piecewise constant