文章摘要
熊景华,茹璟.基于随机森林算法和模糊信息粒化的汇率预测组合模型研究[J].数量经济技术经济研究,2021,(1):135-156
基于随机森林算法和模糊信息粒化的汇率预测组合模型研究
Research on Exchange Rate Forecasting Integrated Model Based on Random Forest and Fuzzy Information Granulation
  
DOI:
中文关键词: 汇率预测  随机森林算法  模糊信息粒化  组合模型
英文关键词: Exchange Rate Forecasting  Random Forest  Fuzzy Information Granulation  Integrated model
基金项目:本文获得国家社会科学基金重大项目“粤港澳大湾区跨境资本流动与金融风险防范研究”(19ZDA093)的资助。
作者单位
熊景华 华南理工大学经济与金融学院武汉大学水利水电学院 
茹璟 华南理工大学经济与金融学院 
中文摘要:
      研究目标:本文致力于利用智能算法,提升汇率预测精度,为跨境资本流动和汇率风险防范提供参考依据。研究方法:引入机器学习前沿领域的随机森林算法RF,结合模糊信息粒化FIG,利用参数和非参数模型构建具有更精准汇率预测能力的组合模型。研究发现:基于RF和FIG构建的组合模型其汇率预测精度分别优于常用的ARIMA、GARCH、SVR单一模型15.5%、16.7%和36.6%;模糊信息粒化可以实现汇率的变尺度的区间预测,且区间均值比点预测更加精准;模糊粒化窗口为季度时模型表现最好,模型误差随窗口宽度先增后减。研究创新:一是利用随机森林算法合理分配权重将三种单一模型导入汇率预测组合模型,经过PCA降维和FIG处理,增强汇率预测精度。二是利用模糊信息粒化将汇率波动预测从点预测转化为区间预测,并可通过修改时间窗变换预测的时间周期。研究价值:随着人民币国际化和汇率市场化不断深化,汇率双向波动加强,精准预测汇率对宏观决策和微观主体都具有重要意义。
英文摘要:
      Research Objectives: This paper is committed to using intelligent algorithm to improve the accuracy of exchange rate prediction and to provide reference for cross-border capital flow and exchange rate risk prevention. Research Methods: Random Forest (RF) and Fuzzy Information Granulation (FIG) in machine learning frontier field are introduced to build an integrated model with more accurate prediction ability of exchange rate.Research Findings: The results show that the accuracy of the integrated model based on RF and FIG is better than that of the commonly used ARIMA, GARCH and SVR models 15.5%, 16.7% and 36.6% respectively; the model based on the training of fuzzy information particles can predict the exchange rate fluctuation range rather than fitted value; the model performs best with the quarterly granulation window, and the error increases followed by decreasing trend as window grows. Research Innovations: The first advantage of Random Forest (RF) is used to reasonably allocate variable weights and distribute the predicted values of three single models into the integrated model, combining with PCA and FIG method, thus enhancing the accuracy of exchange rate prediction. The second advantage is to use Fuzzy Information Granulation to transform the forecast of exchange rate fluctuation from point forecast to interval prediction, and the time period of prediction can be changed by modifying the time window. Research Value:With the deepening of RMB internationalization and exchange rate marketization, the fluctuation of exchange rate is strengthened. Accurate prediction of exchange rate is of great significance to both macro decision-making and micro economic activities.
查看全文