变点估计值对状态空间模型预测的影响分析

发布时间：2018-05-23 19:11

本文选题：状态空间模型 + 预测　；参考：《合肥工业大学》2012年硕士论文

【摘要】：本文主要研究了状态空间模型预测时，，样本序列的变点估计值对模型预测影响的问题。第一章叙述了变点和状态空间模型的研究背景及国内外研究现状，并对检测变点的几种经典方法作了简单介绍。第二章介绍了状态空间模型的定义和Kalman滤波，并给出了ARIMA模型转化为状态空间模型的标准形式的方法。第三章介绍了Γ分布参数变点的检测方法，并讨论了分布参数变点在状态空间模型预测中的应用。在应用方面，首先将上证A股指数收盘价序列转化，得到全涨收益率序列和全跌收益率序列，利用分布参数变点理论得到全涨收益率序列和全跌收益率序列的变点个数及其所处时间位置。然后根据变点位置的不同，分别对三个不同时段的上证A股指数收盘价序列建立状态空间模型。通过比较预测结果，得出变点越少，状态空间模型的预测精度越高的结论。最后在无变点的情况下比较了ARIMA、自回归与状态空间模型的预测结果，说明了状态空间模型具有更好的预测效果。
[Abstract]:In this paper, the influence of the change point estimation of the sample sequence on the prediction of the state space model is studied. In the first chapter, the research background of the change point and state space model and the current research situation at home and abroad are described, and several classical methods for detecting the change point are briefly introduced. In chapter 2, the definition of state space model and Kalman filter are introduced, and the method of transforming ARIMA model into state space model is given. In chapter 3, the detection method of parameter change point in 螕 distribution is introduced, and the application of variation point of distribution parameter in prediction of state space model is discussed. In terms of application, first of all, the closing price sequence of Shanghai A-share index is transformed to obtain the full-rise yield sequence and the full-fall yield sequence. By using the theory of change point of distribution parameter, the number of change points and the time position of all rise rate series and total fall return sequence are obtained. Then, according to the different position of the change points, the state-space model is established for the closing sequence of the A-share index in three different periods of time. By comparing the prediction results, it is concluded that the lower the change points, the higher the prediction accuracy of the state space model. Finally, the prediction results of Arima, autoregressive and state space models are compared in the case of no change points, which shows that the state space model has better prediction effect.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F832.51

【参考文献】