Volatility, Noise, and Market Microstructure: Econometric Analysis Using High-Frequency Data

Abstract

This dissertation contains four chapters. Chapter 1 gives an overall view of stochastic volatility and high-frequency financial econometrics. Chapter 2 proposes a new notion of realized autocovariance for stochastic volatility, which is an analogue of the standard autocovariance in a non-stationary non-ergodic continuous-time set- ting. I propose a plug-in type estimator for the realized autocovariance and show the central-limit theorem of it. Empirical applications are then provided.Chapter 3 concerns the “localized” estimation of volatility, namely spot volatility estimation, under two noisy settings. The first setting concerns an additive-noise multivariate price process, with regular sampling and conditional independence in the noise; the second setting further allows for random sampling and long-run dependence in the noise, but it requires a one-dimensional underlying price process. I propose different pre-averaging estimators of the spot volatility of asset price and that of the microstructure noise, and then provide the uniform rates of convergence for these estimators.Chapter 4 studies a semiparametric inference procedure for a finite-dimensional parameter in a continuous-time regression model in a large cross-section. The model studies the relationship between a dependent process and a possibly nonlinear trans- form of volatility over a fixed time span, and the coefficient is allowed to depend on a set of firm-specific characteristics. The construction of the estimator involves two steps: the nonparametric estimation of volatility processes, followed by a parametric second stage that uses the volatility estimates. I show that the estimator follows a central limit theorem and provide a feasible inference procedure based on a factor- analytic method. Lastly, I show in a realistically calibrated Monte Carlo setting that the performance of the inference procedure is reasonably good.Finally, Chapter 5 concludes.