时变参数状态空间模型估计研究

在宏观经济和金融资本市场上广泛存在着非线性时变参数时间序列，而当前的研究主要关注静态参数状态空间模型的估计。本文通过引入变点分析，改进了静态参数的粒子学习滤波技术，提出了变点粒子学习滤波技术，用于估计时变参数状态空间模型。并且利用模拟实验同经典的变结构IMM滤波技术进行了对比，结果显示，本文提出的变点粒子学习滤波在动态模拟样本数据方面具有更大的优势。可以用于对股票价格和成交量的联合动态轨迹进行实时的模拟追踪。

The Study on the Time-Vary Parameter Estimation for State Space Model

There are numerous nonlinear time-vary time series in the field of macro-economy and financial capital market, however, the current study focus on mainly the static parameter state space model. The research improves the particle learning method for static parameter estimation, through introducing the change-point analysis, and proposes the change-point particle learning filter for the time-vary parameter estimation. Then, it compares the CPPL filter with the IMM filter via the random simulation trial, and the result shows that CPPL method has more advantages than the IMM filter in the parameter estimation for time-vary parameter state space model. The CPPL filter can be used to real-time track the simultaneous moving of stock price and trading volume.