基于新闻情感分析和区间分解的汇率预测研究  

Research on exchange rate forecasting based on news sentiment analysis and interval decomposition

作　　者：刘金培[1] 储娜 罗瑞 陶志富[2] 陈华友[2] LIU Jinpei;CHU Na;LUO Rui;TAO Zhifu;CHEN Huayou(School of Business,Anhui University,Hefei 230601,China;School of Big Data and Statistics,Anhui University,Hefei 230601,China)

机构地区：[1]安徽大学商学院,安徽合肥230601 [2]安徽大学大数据与统计学院,安徽合肥230601

出　　处：《安徽大学学报（自然科学版）》2024年第1期1-10,共10页Journal of Anhui University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(72071001,72001001,72371001);教育部人文社会科学规划项目(20YJAZH066,21YJCZH148);安徽省自然科学基金资助项目(2008085MG226,2108085MG239);安徽省高校优秀青年人才项目(gxyqZD2022001)。

摘　　要：汇率序列具有非线性和连续变化等特点,其细节波动是一系列事件和新闻综合影响的结果.然而,现有区间预测模型难以量化重大事件和公众情绪的影响,导致其缺乏广泛的适用性,且传统区间分解方法存在上下界混叠的缺陷.因此,论文从新冠疫情冲击出发,提出一种基于新闻情感分析和区间分解的汇率波动实时预测模型.首先,基于Snownlp情感词典对外汇新闻文本进行情感分析,获得相应的情感分数.另外,构建全球恐惧指数(the global fear index,简称GFI)以量化新冠疫情的影响,并将其与芝加哥期权交易所波动率(the Chicago board options exchange volatility index,简称VIX指数)相结合作为汇率的影响因素.然后,提出一种新的区间经验模态分解(interval empirical mode decomposition,简称IEMD)方法对区间汇率序列进行多尺度分解,并根据样本熵重构得到高、中、低频区间序列和残差项.其次,利用极限学习机(extreme learning machine,简称ELM)、多层感知机(multi-layer perceptron,简称MLP)、随机森林(random forest,简称RF)和二次曲面支持向量回归(quadric surface support vector regression,简称QSSVR)分别对不同特征的子序列进行组合预测,以提高预测结果的准确性和稳定性.最后,利用论文方法对美元兑人民币、澳元兑人民币和瑞士法郎兑人民币3种汇率进行实证预测分析,结果表明,论文模型适用于重大事件影响下的汇率区间波动预测,与现有方法相比具有较高的预测精度.Exchange rate series have the characteristics of non-linearity and continuous changes,and the detailed fluctuation is influenced by a series of events and news.However,the existing interval prediction model is difficult to quantify the impact of major events and public sentiment,resulting in its lack of extensive applicability.Meanwhile,the traditional interval decomposition method has the defect of overlapping upper and lower bounds.Therefore,starting from the impact of the COVID-19,a real-time forecasting model based on news sentiment analysis and interval decomposition for exchange rate fluctuations was proposed.First,based on Snownlp emotion dictionary,emotion analysis was carried out for foreign news text,and corresponding emotion scores were obtained.In addition,the global fear index(GFI)was constructed to quantify the impact of the COVID-19,and combined with the Chicago board options exchange volatility index(VIX)as the influencing factor of the exchange rate.Second,a new interval empirical mode decomposition(IEMD)method was proposed to decompose the interval exchange rate series at multiple scales,and reconstruct the high,middle and low frequency interval series and residual terms according to the sample entropy.Third,the combination prediction of sub-sequences with different characteristics was carried out by using extreme learning machine(ELM),multi-layer perceptron(MLP),random forest(RF)and quadric surface support vector regression(QSSVR)to improve the accuracy and stability of the prediction results.Finally,we used the proposed model to make empirical prediction and analysis on USD/CNY,AUD/CNY and CHF/CNY.The results showed that the model in this paper was suitable for forecasting the fluctuation of exchange rate range under the influence of major events,and had higher prediction accuracy compared with the existing methods.

关 键 词：汇率预测 情感分析 区间经验模态分解 二次曲面支持向量回归 

分 类 号：O212[理学—概率论与数理统计]