机器学习、文本大数据与Nowcasting即时预测  

Machine Learning,Text Big Data and Nowcasting

作　　者：谭辉 郇志坚[2] 普煜 Tan Hui;Huan Zhijian;Pu Yu(School of Finance,Xinjiang University of Finance and Economics;Xinjiang Uygur Autonomous Region Branch,the People's Bank of China)

机构地区：[1]新疆财经大学 [2]中国人民银行新疆维吾尔自治区分行

出　　处：《金融发展评论》2024年第5期59-74,共16页Financial Development Review

摘　　要：即时预测宏观经济具有重要的现实意义,Nowcasting通过实时数据分析,能够为决策者提供决策参考。本文针对传统计量经济学模型的不足,分析了机器学习方法在Nowcasting中非线性和时变性的优缺点。与传统数据相比,文本大数据具有多样性、异构性、大规模性、时频高等特征,提供了新信息源,在适当的学习方法下有助于提升预测的精度,为经济决策和市场预测提供更准确、及时的支持。Nowcasting macroeconomic trends is of significant practical importance,as real-time data analysis can provide valuable insights for decision-makers.This paper addresses the limitations of traditional econometric models and analyzes the strengths and weaknesses of machine learning models in Nowcasting,particularly in their ability to handle non-linearity and time-variability.Compared to traditional data,textual big data features diversity,heterogeneity,large scale,and high temporal frequency,offering new sources of information.Under appropriate learning methodologies,this can enhance the accuracy of forecasts,providing more accurate and timely support for economic decision-making and market predictions.

关 键 词：NOWCASTING 动态因子 机器学习 文本数据 

分 类 号：F224[经济管理—国民经济] TP181[自动化与计算机技术—控制理论与控制工程] TP391.1[自动化与计算机技术—控制科学与工程]