摘要
本文进一步研究了用于非线性时间序列建模的均值异方差混合转移分布(Expectation het-eroscedastic mixture transition distribution,EHMTD)模型。该模型的条件方差定义为以往观测值的非线性函数,讨论并得到了EHMTD模型的平稳性条件。运用ECM(Expectation conditional maximization)算法估计模型的参数,利用BIC(Bayes information criterion)准则选择模型。最后将EHMTD模型应用于一组金融数据,结果表明EHMTD模型的条件分布灵活多变,能够对序列的非对称、多峰等非Gauss特征进行描述。
The expectation heteroscedastic mixture transition distribution （EHMTD） model for modeling nolinear time series is further studied in this paper. The component＇s conditional variance is defined as a nonlinear function of the time series given its history. Stationary conditions of the model are derived. A simple expectation conditional maximization （ECM） algorithm is designed and works well for estimation. The Bayes information criterion （BIC） is used to select the model. The model is applied to a financial time series with satisfactory results. The shape changing feature of conditional distributions of the EHMTD model makes the model capable of modeling time series with asymmetry or multimodal.
出处
《工程数学学报》
CSCD
北大核心
2008年第6期1051-1058,共8页
Chinese Journal of Engineering Mathematics
基金
教育部重点科研基金(03I53059)
航空科学资助项目(03I53059)
关键词
平稳性
ECM算法
非对称
多峰
厚尾
stationarity
ECM algorithm
heteroscedastic
asymmetric
multimodal
