时变β系数的卡尔曼滤波的仿真与估计  

Estimation and Simulation of Time-varying Coefficient β Based on Kalman Filter

作　　者：梅婷[1] 李建勋[1] 

机构地区：[1]上海交通大学自动化系

出　　处：《微型电脑应用》2010年第10期25-29,2,共5页Microcomputer Applications

基　　金：自然科学基金(20075157007)

摘　　要：旨在预测风险系数,我们对沪市10个行业的β系数的时变行为采用随机游走(RW)模型,在随机游走模型的回归变量前引进了未知参数,仿真预测的结果表明,引进未知系数后的模型,其绝对预测误差MAE和均方预测误差MSE都比引进未知系数前的模型更精确,改后的模型更适合描述沪市行业风险系数的时变行为。仿真过程是通过卡尔曼滤波的递推过程获取新息及其协方差阵,得到似然函数,从而推进参数估计。将估计的结果返回卡尔曼滤波更新过程,易得观测变量的预测值。This paper aims to predict systematic risk of Shanghai industrial stock returns. Ten industrial betas are modeling by a new model, which brings in two unknown coefficients based on Random Walk of time-varying systematic risk. Simulation results show that the new model whose forecast errors are more precise is more effective, the new model is more suitable to describe variability of systematic risk of Shanghai industrial stock returns. By the recursive process of kalman filter, the innovation and its covariance matrix and likelihood function are got, and then push on parameter estimation,return the result estimated to process of kalman filter, which make the prediction.

关 键 词：卡尔曼滤波 极大似然估计 随机游走模型 时变风险系数 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]