摘要
研究了可用于非线性时间序列建模的异方差混合双自回归模型(heteroscedastic mixture double- autoregressive model,HMDAR),给出了HMDAR模型的平稳性条件,利用ECM(expectation conditional maximization)算法来估计模型的参数,运用BIC(Bayes information criterion)准则来选择模型.HMDAR模型条件分布富于变化的特征使它能够对具有非对称或多峰分布的序列进行建模,将HMDAR模型应用于几个模拟和实际数据集均得到了较为满意的结果,特别是对波动较大的序列,HMDAR模型能比其他模型更好地捕捉到数据序列的特征.
The heteroscedastic mixture double-autoregressive (HMDAR) model, which is designed to model nonlinear time series, is investigated in this paper. Firstly, the stationary conditions are derived. Secondly, the parameters are estimated via expectation conditional maximization (ECM) algorithm. Thirdly, the Bayes information criterion (BIC) is used to select the model. The varied feature of conditional distributions makes the HMDAR model capable of modeling time series with asymmetric or multimodal distribution. Finally, the model was applied to several simulated and real data sets with satisfactory results. Especially to a highly time-variant series, the HMDAR model shows better performance than other competing models in data capturing.
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
2006年第6期879-885,共7页
Control Theory & Applications
基金
国家自然科学基金(60375003)
国家航空基金(03153059).
关键词
异方差混合双自回归模型
平稳性
BIC准则
ECM算法
非对称分布
多峰分布
条件异方差
heteroscedastic mixture double-autoregressive model
stationarity
Bayes information criterion
expectation conditional maximization (ECM)algorithm
asymmetric distribution
multimodal distribution
conditional heteroscedasticity