非参数随机条件持续期模型及其迭代算法  

The Non-parametric Stochastic Conditional Duration Model and the Iteration Algorithm

作　　者：孙艳[1] 何建敏[1] 周伟[1] 

机构地区：[1]东南大学经济管理学院

出　　处：《统计研究》2011年第8期103-110,共8页Statistical Research

基　　金：国家自然科学基金项目"流动性调整期望损失La-ES和最优变现策略"(70671025)资助;国家自然科学基金项目"基于复杂网络的银行间传染性风险及其演化模型研究"(71071034)资助

摘　　要：随机条件持续期(SCD)模型能有效刻画超高频时间序列中持续期的变化,但该模型假定期望持续期生成机制固定,且模型参数估计存在一定的困难。文章在不假定条件均值形式和冲击项分布的基础上结合核估计方法提出了非参数SCD模型及其迭代求解方法。然后,基于TEACD(1,1)模型生成的模拟数据,将非参数SCD模型与用卡尔漫滤波进行伪似然估计的参数SCD模型和用Gibbs抽样进行马尔科夫蒙特卡罗估计的参数SCD模型的拟合效果进行比较,实证表明在大样本条件下非参数SCD模型的拟合效果与用MCMC估计的参数SCD模型的拟合结果相差不大,但明显优于用QML估计的参数SCD模型的拟合结果,且非参数SCD模型能为参数SCD模型的参数设定提供参考。The SCD model can effectively describe the changes of the durations in the ultra-high time series,but it is based on the assumption that the financial durations are generated by a fixed mechanism, and the estimation of the parameters is not easy. Combing with the kernel estimation, this paper proposes the nonparametric SCD model and its iteration algorithm neither relies on the form of the conditional mean of the duration nor the distribution of the error term. Then,based on the simulated data generated by the TEACD （1,1）model, the non-parametric SCD model is compared with the parametric SCD model which is estimated by the quasi-maximum likelihood with the using of the Kalman filter and by the Markov chain Monte Carlo with the using of the Gibbs sampling,proving that the non-parametric SCD model performs as well as the parametric SCD model by the MCMC estimation and yields better fitted effects than parametric SCD model by the QML estimation with large samples and proposes a valuable preliminary suggestion on the choice of the parametric specification.

关 键 词：SCD模型 持续期 伪似然估计 MCMC估计 核估计 

分 类 号：F222.3[经济管理—国民经济]