Detecting Financial Data Dependence Structure by Averaging Mixture Copulas
2017/04/07
研討會日期 : 2017-04-07
時間 : 14:30
主講人 : Professor Xinyu Zhang
主持人 : Professor Chu-An Liu
地點 : Conference Room B110
演講者簡介 : Professor Zhang received his Ph.D. from Chinese Academy of Sciences in 2010. He is currently an Associate Research Fellow at Chinese Academy of Sciences. His research interests include Model Selection and Averaging, Forecasting Combination, Estimation of Differential Equation Models.
演講摘要 : A mixture copula is a linear combination of several individual copulas. It is able to generate dependence structures that do not belong to existing copula families. This makes it useful in modeling the dependence structures in financial data, as in empirical studies different pairs of markets may exhibit quite different dependence structures. Therefore, rather than selecting one single copula through certain criterion, we propose using a model averaging approach to estimate financial data dependence structure in a mixture copula framework. We select weights (for averaging) through a J-fold Cross-Validation procedure. We prove that the model averaging estimator is asymptotically optimal in the sense of minimizing a squared estimation loss. Simulation results show that the model averaging approach outperforms some competing methods when the working mixture model is misspecified. Using 12 years' daily returns of four developed economies' stock indexes (United States, United Kingdom, Hong Kong and Japan), we show that the model averaging approach gives more accurate estimation of their dependence structures than some competing methods.
