Measuring and Forecasting Financial Market Volatility using High-Frequency Data Defended on Friday, 11 January 2013
This dissertation consists of three studies on the use of intraday asset price data for accurate measurement and forecasting of financial market volatility. Chapter 2 proposes a refined heuristic bias-correction for the two time scales realized range-based volatility estimator in the presence of bid-ask bounce and non-trading. The merits are illustrated through simulations and an empirical forecasting application. Chapter 3 introduces a novel approach for estimating the covariance between asset returns using intraday high-low price ranges. The realized co-range estimator compares favourably to the realized covariance for plausible levels of microstructure noise and non-synchronous trading. The estimator is successfully implemented in a volatility timing strategy that deals with constructing mean-variance efficient asset allocation portfolios from stock, bond and gold futures. Chapter 4 introduces a mixed-frequency factor model for vast-dimensional covariance estimation. This original approach combines the use of high- and low-frequency data with a linear factor structure. We propose the use of highly liquid ETFs -- that are essentially free of microstructure frictions -- as factors such that factor covariances can be estimated with high precision from ultra-high-frequency data. The factor loadings are estimated from low-frequency data to bypass the potentially severe impacts of noise for individual stocks and to circumvent non-synchronicity issues between returns on stocks and liquid factors. Theoretical, simulation and empirical results illustrate that the mixed-frequency factor model is excellent, both compared to low-frequency factor models and to popular realized covariance estimators based on high-frequency data.
high-frequency data; realized volatility; realized range; factor models; microstructure noise; bias-correction; forecasting; mixed frequencies; vast dimensional covariance matrix