摘要
针对具有非线性和不稳定性的时间序列,提出一种结合小波分解、滑动平均离散差分方程和马尔可夫方法的动态预测模型。利用小波多尺度分解将原时间序列分解到不同频率通道上,然后对分解出的低频近似小波系数利用滑动平均离散差分方程预测模型进行预测,并利用马尔可夫方法对时间序列的高频细节小波系数进行预测,最后将低频和高频的预测结果进行小波重构得到时间序列的实际预测值。原油价格的时间序列是一类典型的具有非线性和不稳定性的序列,利用此模型对WTI原油(周度)价格进行实证预测分析,分别预测WTI原油价格的整体变化趋势和周度实际原油价格。研究结果表明,此模型不但可以有效地预测时间序列的整体变化趋势,能从细节上对其进行有效的刻画,而且比其他基于小波的预测模型具有更高的预测精度。可以看出国际原油价格从整体上呈现周期性上涨趋势,并且不稳定的随机波动也会一直存在。
In view of unstable and nonlinear time series, a dynamic forecasting model was proposed in this paper, which integrated wavelet decomposition, Slip Discrete Difference Equation Prediction Model （SDDEPM） and Markov methods. In this model, the original time series were decomposed into different frequency channels by multi-scale wavelet. Then we predict the wavelet coefficients of the low-frequency approximation with the SDDEPM and predict the wavelet coefficients of high frequency details by Markov method, respectively. Therefore, predictive value of the original time series was obtained by wavelet reconstruction of the low and high frequency foresting results. Since oil price time series is a typical kind of unstable and nonlinear series, and the model was applied to forecasting WTI weekly crude oil prices. The research result shows that the proposed forecasting model could not only forecast holistic fluctuation frequency of time series effectively but also characterize the details of the time series. The prediction accuracy of this model is much higher than any other wavelet-based models. It seems that the international oil price presents a standing periodic rise with stochastic volatility.
出处
《管理科学》
CSSCI
北大核心
2011年第6期104-112,共9页
Journal of Management Science
基金
国家自然科学基金(70672027
71102139
71071002)
教育部新世纪优秀人才支持计划(NCET-07-0598)
安徽大学创新团队资助项目(KJTD001B
SKTD007B)~~
关键词
时间序列预测
离散差分方程预测模型
小波分解
油价预测
time series forecast
discrete difference equation prediction model
wavelet decomposition
oil price forecast
