Econometric Analysis of Financial Markets Using High-Frequency Data

Abstract

This dissertation employs high-frequency data and techniques to examine various topics in financial markets. Chapter 1 compares forward regression model with eight statistical/practical trading exchange rate models in terms of forecasting foreign exchange rates. Superior forecast power of the forward regression model is found at horizons longer than one month, and the superiority enhances as forecast horizon lengthens. This indicates that fundamentals matter in the long run. Chapter 2 examines inter-market information spillovers across the US, Japan, Asia ex-Japan, and Europe. The non-parametric, realized volatility method based on intra-day exchange-traded funds data is used to avoid model misspecification in volatility estimation. Uni-directional volatility spillovers from the US to other markets are observed. Chapter 3 compares the realized volatility method with traditional multivariate GARCH method in terms of information transmission detection and portfolio optimization. The realized volatility method detects uni-directional volatility spillovers from the US to Japan index funds, while the GARCH method finds no spillovers between the two funds. In addition, the optimized portfolios based on the realized volatility method outperforms those based on the GARCH method in terms of minimizing portfolio risks (measured by annualized standard deviations) or maximizing portfolio return-to-risk ratios at various horizons.