ISEMSeminar: AnEfficient Bayesian Inference for the State Space Models with the StochasticVolatility

Time:2:00pm-3:30pm,Oct 23, 2013

Place: ISEM Conference Room (3rdfloor, Chengming Building)

题目：状态空间随机波动模型的贝氏估计

主讲人：黄宇凡（国际经管学院助理教授）

时间：2013年10月23日（周三）下午2:00-3:30

地点：国际经管学院会议室（诚明楼三层）

主办方：国际经管学院

An Efficient BayesianInference for the State Space Models with the Stochastic Volatility

Abstract: Thispaper is concerned with Bayesian estimation methods for linear state spacemodels with stochastic volatility. I find that the Gibbs sampler, referred toas the auxiliary mixture sampler, may not converge even in a simple case. Ipropose a new sampler, referred to as the marginalized mixture sampler, whichis implemented without the need to simulate the mixture indicator variable. Twomodified versions of the auxiliary mixture sampler are also proposed. Usingsimulated data, I show that all three algorithms converge to the posteriordistribution covering the true values of the parameters. Meanwhile, themarginalized mixture sampler outperforms the other two by a factor of 4 to 11.

状态空间随机波动模型的贝氏估计

摘要：本文探讨线性的空间状态随机波动模型之贝氏估计法。模型中非常态分配的随机干扰项可用多个常态分配的溷合来趋近，此方法亦是由Kimet al. (1998)所发展。文献中常用辅助混合抽样法(auxiliarymixture sampler)是使用dataaugmentation的方法进行贝氏估计，而本文提出一个不需要使用dataaugmentation的方法--边缘混合抽样法(marginalmixture sampler)。模拟实验显示出边缘混合抽样法比辅助混合抽样法更有效率4至11倍。

主讲人简介

黄宇凡，首都经济贸易大学国际经管学院助理教授。本科毕业于国立政治大学，硕士毕业于台湾大学，博士毕业于华盛顿大学。主要研究方向为Empirical Macroeconomics和Applications of Bayesian Inference。