摘要
汇率的变化具有非线性、非平稳、连续变化等特点,传统离散点值预测会损失其波动信息,也无法表示和估计其波动的区间范围。考虑到海量的互联网非结构化数据为汇率预测提供了大量的有效信息,本文提出一种非结构化数据驱动的混合二次分解汇率区间多尺度组合预测方法。首先,从百度指数中提取汇率相关的非结构化数据,利用奇异谱分析(SSA)对其进行去噪,并通过主成分分析(PCA)实现非结构数据的降维。其次,为了充分提取非结构数据和汇率区间序列的有效信息,基于集成经验模态分解(EEMD)、变模态分解(VMD)、奇异谱分析(SSA)和小波变换(WT)对汇率区间序列进行混合二次分解,得到多个不同分解路径下的汇率区间分解结果。然后,针对汇率区间分解得到的高频、低频和趋势序列的特征,基于非结构化数据,建立长短期记忆模型(LSTM)、支持向量回归(SVR)和BP神经网络(BPNN)三种单项预测方法分别对其进行预测,并对结果进行集成,得到不同分解路径下的汇率区间预测值。最后,利用最优加权组合方法对不同分解路径下的汇率区间预测值进行组合,得到汇率区间的最终预测结果。为验证该组合预测方法的有效性,本文对2018-2020年的美元兑人民币日汇率区间时间序列进行实证预测分析,结果表明,本文方法适用于具有高噪声的汇率区间预测,与已有方法相比具有更高的精确度和良好的适用性。
The foreign exchange rate has the characteristics of non-linear,non-stationary,continuous change,etc.The traditional forecasting method of point value will lose its fluctuation information.In addition,the single data decomposition method has its inherent defects,there is the problem of incomplete decomposition,and the choice of decomposition method is uncertain.At the same time,massive unstructured data on the Internet provide a large amount of effective information for exchange rate prediction,but there is still a lack of systematic research on how to use unstructured data for exchange rate interval prediction.In order to resolve these problems,in this paper,an unstructured data driven hybrid secondary decomposition multi-scale exchange rate interval combination forecast approach is proposed.First,singular spectrum analysis(SSA)and principal component analysis(PCA)are employed to de-noise and de-dimension the data respectively.Second,the hybrid secondary decomposition method is used to divide exchange rate interval sequence and unstructured data,and decomposition results through multiple different decomposition paths are obtained.Third,the long-and short-term memory model(LSTM),support vector regression(SVR)and BP neural network(BPNN)are utilized to model and forecast high frequency sequence,low frequency sequence and trend item,respectively.At the same time,the results are integrated to get the predicted exchange rate for each decomposition path.Finally,the optimal weighted combination method is employed to combine the different predicted values,and the final predicted results are obtained.To verify the superiority of the proposed approach,an empirical forecast analysis is conducted on the interval time series of the daily USD/RMB exchange rate from 2018 to 2020.The experimental results show that the proposed method is suitable for the exchange rate range prediction with high noise,and has higher accuracy and good applicability compared with the existing methods.Based on the effective use of unstructured dat
作者
刘金培
罗瑞
陈华友
陶志富
LIU Jin-pei;LUO Rui;CHEN Hua-you;TAO Zhi-fu(School of Business,Anhui University,Hefei 230601,China;School of Big Data and Statistics,Anhui University,Hefei 230601,China)
出处
《中国管理科学》
CSSCI
CSCD
北大核心
2023年第6期60-70,共11页
Chinese Journal of Management Science
基金
国家自然科学基金资助项目(72071001,72271002,72001001,71871001)
教育部人文社会科学研究规划基金资助项目(20YJAZH066,21YJCZH148)
安徽省自然科学基金资助项目(2008085MG226,1908085J03)
安徽省高校优秀青年人才支持计划重点项目(gxyqZD2022001)。
