论文部分内容阅读
基于低频金融数据的预测,在时间上具有长期性,依赖于整体经济环境,不能形成短期内的准确预测.但是由于高频金融时间序列具有非线性、非平稳性以及其特有的日历效应等特性,传统的ARMA模型也无法得到满意的预测结果.本文提出基于小波多分辨率分析的预测方法,将收益率数据分为高频部分(周期性)与低频部分(趋势性),对拆分后的序列进行重构,并对重构后得到的数据分别建立ARMA模型.实证研究表明,小波多分辨率分析能很好地滤出日内效应,由于股指期货独特的市场特征,应将分解层数定为3,分解重构模型可以提高预测精度.
The forecast based on low-frequency financial data is long-term in time and depends on the overall economic environment and can not form accurate forecasts in the short term. However, the characteristics of high-frequency financial time series are non-linear, non-stationary and their unique calendar effects , The traditional ARMA model can not get a satisfactory prediction result.This paper presents a prediction method based on wavelet multi-resolution analysis, the data is divided into high frequency part (periodic) and low frequency part (trend), after the split ARMA model is established respectively for the reconstructed data.Experimental results show that the wavelet multiresolution analysis can filter out the intraday effect very well.Due to the unique market characteristics of stock index futures, As 3, decomposition and reconstruction model can improve the prediction accuracy.