Forecasting Tail Risk in Financial Time Series Using a Mixture of Distribution Approach
2010/08/03
研討會日期 : 2010-08-03
時間 : 15:10
主講人 : Professor Mike Ka Pui So
地點 : B110
演講者簡介 : Mike Ka Pui So got his Ph.D. in Statistics from University of Hong Kong in 1996. He is currently serving as Associate Professor of Information System, Business Statistics and Operations Managementat Hong Kong University of Science and Technology. His research interest is in nonlinear time series analysis, financial time series modeling, market volatility study, modeling the dynamic structure of economic data and risk management.
演講摘要 : Financial return distribution is well recognized for its fat-tail behavior and its tail asymmetry. One natural question is whether the tail asymmetry is still significant after accounting for the conditional heteroskedasticity in returns. In this paper, we propose a mixture of distribution approach under a GARCH framework to capture the tail asymmetry. A key idea is to use the Peak-over-Threshold method of extreme value theories to construct a mixture distribution for the innovation in GARCH models. This mixture distribution combines a common distribution, like asymmetric normal, and two generalized Pareto distributions for the two tail parts to capture both the leptokurtosis and the tail asymmetry. A Bayesian approach is adopted to estimate unknown parameters by using Markov Chain Monte Carlo (MCMC) methods. A Bayesian test for tail asymmetry is also established. We perform simulations to analyze the performance of the MCMC and the effectiveness of using a mixture distribution to approximate common distributions. We also study the performance of our approach in forecasting volatility and the tail risk, like Value at Risk and expected shortfall, using real data.
