Electronic Thesis and Dissertation Repository

Degree

Doctor of Philosophy

Program

Economics

Supervisor

John Knight

Abstract

The thesis consists of three essays dealing with the modeling of volatility in financial markets, trade durations, and Value-at-Risk (VaR). The first essay models nonlinearities in the return series to estimate time-varying volatility by incorporating both regime changes and jumps. Two types of regime-switching GARCH-jump models with autoregressive jump intensity are presented. The first model follows the traditional Markov regime-switching model proposed in Hamilton (1989). As the unknown regimes in the Markov model lead to difficulty in forecasting, a threshold GARCH-jump model, in which regimes are known after observing the threshold variable in the previous period, is also proposed. The second essay models the intraday durations between two adjacent trade transactions by considering the impact of unaccounted structural changes on parameter estimates. Monte Carlo simulations show that the observed high persistence in trade durations can be spurious and caused by unaccounted structural changes in the data generating process. The third essay investigates the use of realized moments in VaR forecasting, which is an important issue in risk management. Many VaR models rely only on the mean and volatility and ignore higher moments of returns, which leads to underestimation of VaR due to the unaccounted fat tail property of the return series. By applying the Cornish-Fisher expansion to incorporate realized higher moments constructed from high frequency data, the proposed realized moment models outperform the realized volatility model and the traditional RiskMetrics model, especially during the financial crisis period (2008-09).


Included in

Econometrics Commons

Share

COinS