正态逆高斯扩散模型的MCMC估计.pdf

15 2 系 统 工 程 理 论 方 法 应 用 V ol. 15 N o . 2 2006 4 SY ST EM S EN GI NEER IN G- T HE O RY M ET HO DO LO G Y A PP L ICA T IO N S A pr . 2006 : 1005-2542( 2006 02-0133-06 MCMC 胡素华, 　张　彤, 　张世英 ( , 300072 使用贝叶斯方法估计了正态逆高斯扩散 型, 该方法首先使用Euler 方法对连续过程进行离散 化, 用离散过程的似然函数做为 型参数的近似似然函数。证明了MCM C 方法是分析正态逆高斯扩散 型 的有效工具, 由M CMC 方法抽样所得的后验分布可以用来进行统计推断。 拟试验表明: 正态逆高斯扩散 能够体现资产收益的许多经验特征, 如泰勒效应、尖峰厚尾等。 : MCM C; 正态逆高斯扩散; 广义抛物线扩散; 贝叶斯方法 : F 830. 91: A Estimation of Normal Inverse Gaussian Dif fusion Using MCMC Method H U Su-hua, ZHA N G Tong, ZHA N G Shi-ying ( Scho ol o f Management , T ianjin Univ. , T ianjin 300072, China ( Abst ract In t his paper w e propose a Bay esian met hod t o est imat e t he normal inverse Gau ssian N IG dif- fu sion model. T he approach is based o n t he Markov chain M ont e Carlo ( M CMC m et hod w it h t he likeli- ho od of t he discredit ed pro cess as t he appr oxim at e post erio r l ikeliho od . We demo nst rate t hat t he M CM C met ho d provides a u seful t oo l in analyzing N IG dif f usion . In part icular, quantit ies of post erior dist ribu - tio ns obt ained fro m the M CMC output s can be u sed f or st atist ical inference. T he MCM C m et hod is based on Euler schem e. Our simulat ion st udy show s t hat t he N IG dif fu sio n exhibit s many of the st yl iz ed fact s abo ut asset ret ur ns docum ent ed in t he discrete-t ime financial eco nom et rics lit erature, such as the T ayl or eff ect , a slo wl y declining aut ocorrelat ion f unct ion of the squar ed ret ur ns , and t hick t ails . Key words: M ar ko v chain M ont e Car lo; normal inv er se Gau ssian diff u sion ; generaliz ed hy perbol