一种基于变结构加权的时间序列预测模型研究

Time series forecasting model based on weighted variable structure

基于文献提出的优化权重方法,针对存在单个离散变结构下的时间序列预测问题,提出了一种变结构加权的ARMA模型.该方法利用优化权重对时间序列变结构前的数据进行调整,形成了新的时间序列,然后再对这个新的时间序列建立ARMA模型进行预测.蒙特卡罗模拟显示,该方法对存在单个离散变结构的时间序列具有较好的预测效果.实证结果表明中国股票市场的日收益率存在明显的变结构现象,并且在预测中运用加权ARMA模型能够明显改善预测效果.

For time series prediction under a single discrete break, a weighted variable structure ARMA model is proposed based on optimal weights used previously. Pre-break data are adjusted using optimal weights to form a new time series, ARMA model is then established for the forecast. Monte Carlo simulations are used to show good forecasting performance of this method for time series with a single discrete break. Empirical results show that the daily return rate of Chinese stock market has marked changes in variable structural enomenon, and forecasts based on weighted variable structure ARMA model deliver significant improvements.