Electronic Thesis and Dissertation Repository


Doctor of Philosophy




John Knight


This thesis consists of three chapters of interest to a portfolio manager. The first paper examines how the profitability of trading rules depends on volatility. In particular, a question of interest is whether one rule dominates all others regardless of the level of volatility, or whether it is more profitable to vary the choice of trading rule corresponding to volatility. Certain rules, such as the KST indicator using overbought/oversold levels, appear to excel under highly volatile conditions, while exponential moving average rules perform better with low volatility. In the second paper, a Value-at-Risk (VaR) model capable of producing accurate and robust forecasts is presented. In particular, the model presented here provides an extension to the VARLINEX model of Knight, Satchell, and Wang (2003) (hereafter KSW (2003)). The end result is a model capable of accurately forecasting VaR during the recent stock market crash (2008-09), as well as before and after the crash. The new model outperforms a benchmark model that had been successful prior to the crash, as well as the original VARLINEX model (KSW (2003)). The third paper explicitly spells out the link between independence tests and goodness-of-fit tests that are based on copula functions. However, the primary contribution is the development of a new copula-based goodness of fit test, which involves incorporating a weighting function in one of the test statistics proposed in Genest, Remillard, and Beaudoin (2009). Guidance is given in terms of how to choose an appropriate weighting function, and an application to Value-at-Risk forecasting is included.

Included in

Econometrics Commons