Essays on value at risk and asset price bubbles

Abstract

This thesis describes a series of investigations into the reliability of different financial risk models for measuring downside risks during financial crises caused by the bursting of asset price bubbles. It also provides further insight into modelling asset price series with periodically collapsing asset price bubbles. We start by reviewing the volatility models that are commonly used for quantifying downside financial risks in Chapter 2. The characteristics of several important univariate and multivariate autoregressive conditional heteroscedasticity family volatility models are reviewed. In Chapter 3, we apply the volatility models to identify the direction of volatility spillover effects among stock markets. The financial markets considered in this study include Japan, China, Hong Kong, Germany, the United Kingdom, Spain, the United States, Canada, and Brazil. Findings from both the dynamic conditional correlation (DCC) and asymmetric DCC models show that the asymmetric volatility spillover effect is highly significant among financial markets, while the asymmetric correlation spillover effect is not. Financial contagion will be reflected in price volatility, but not in correlations. Subsequently we move to testing different Value-at-Risk (VaR) approaches using market data. We define 1 June 2008 to 1 June 2009 as the financial crisis period. We study nine hypothetical single-stock portfolios and nine hypothetical multiple-asset portfolios in the nine countries considered. Both the univariate and multivariate VaR approaches are tested and the results show that the long memory RiskMetrics2006 model outperforms all other univariate methods, while the Glosten-Jagannathan-Runkle DCC model performs well among the multivariate VaR models. Next, in Chapter 5 we use simulations to explore the characteristics of financial asset price bubbles. Evans (1991) proposed a model for investigating asset price movements with periodically collapsing explosive bubbles. We modify and extend this model to make it more realistic; as a result the modified model better controls the growth and collapse of bubbles, while exhibiting volatility clustering. In the simulation tests, the RiskMetrics VaR model performs well during financial turmoil. Finally, we discuss the sup-augmented Dickey-Fuller test (SADF) and the generalized SADF test for identifying and date-stamping asset price bubbles in financial time series. Unlike in Chapter 4 (where we use personal judgement to define financial bubble periods), pre- and post-burst periods are defined here based on the identification results of the asset price bubbles’ origination and termination dates from the backward SADF test. Our empirical results show that the criticism that VaR models fail in crisis periods is statistically invalid.