Finite sample results of Range-based integrated volatility estimation

In this paper we consider the finite-sample properties of Realized Range estimators of integrated volatility and we compare them to those of the Realized Volatility estimators when a sample of high-frequency data is observed. Simulated data are obtained from different generating mechanisms for the instantaneous volatility process, e.g. Ornstein-Uhlenbeck, long memory and jump processes. We analyze the impact that missing observations have on the Realized Range measures and we propose a simple correction in order to reduce the bias. We also evaluate the robustness of the different approaches considered when high-frequency prices are affected by bid-ask bounce and price discreteness. Simulation results confirm that realized range corrected for irregular sampling has lower bias while not increasing the estimator variance. The simulations also show how the degree of persistence in the estimated Integrated Variance series crucially depends on the sampling frequency adopted in the estimation and thus on the precision of the estimators. A brief empirical application with high-frequency IBM data is also included.