A Comprehensive Method for Integrated (Realized) Volatility Estimation

In this dissertation, a comprehensive kernel-based estimator, PCA Kernel (PK), is studied to estimate the integrated (realized) volatility under the effect of a more realistic and complex market micro-structure noise. Challenging situations, such as irregular diurnal sampling times, moving-average (MA) long-memory noises, inhomogeneous latent log-price processes, jumps, diurnal pattern trading volumes, and non-linear trading information related noises, are considered to test the performance of the proposed estimator. The Principal Components Analysis (PCA) approach is implemented to improve the estimator's stability, and the linear, polynomial, and Gaussian kernels are applied in the reproducing kernel Hilbert spaces to test their de-noise capacities. The kernel-based estimator dramatically improves the efficiency and accuracy of volatility estimation in terms of both the empirical variance and bias.