基于MF-LSTM的上市公司现金流预测研究  

Research on Cash Flow Forecast of Listed Companies Based on MF-LSTM

作　　者：江泽茹 王冉冉[2,3] JIANG Ze-ru;WANG Ran-ran(School of Government,Peking University,Beijing 100871,China;National Engineering Laboratory for Big Data Analysis and Application,Peking University,Beijing 100871,China;Chongqing Research Institute of Big Data,Peking University,Chongqing 401121,China)

机构地区：[1]北京大学政府管理学院,北京100871 [2]北京大学大数据分析与应用技术国家工程实验室,北京100871 [3]北京大学重庆大数据研究院,重庆401121

出　　处：《数理统计与管理》2024年第3期512-526,共15页Journal of Applied Statistics and Management

摘　　要：企业现金流量很大程度上反映着企业的生存和发展能力,现金流的预测和分析对于投资者和市场管理者都具有十分重要的现实意义。本文利用我国A股上市公司2012年至2019年的财务数据和账户数据,提出混合频率长短时记忆神经网络模型(Mixed Frequency Long Short Term Memory,MF-LSTM)对上市公司现金流量进行预测。通过神经网络结构设计,本文将不同频率数据进行了有效混合,预测结果明显优于传统时间序列模型。另外,实验结果还验证了企业账户数据是现金流预测的有效解释变量。Cash fow can reflect survival and development potential of companies.For both investors and managers,the prediction and analysis of cash fow is of great practical significance.Based on financial data and account data of China's A-share listed companies from 2012 to 2019,this paper designs Mixed Frequency Long Short Term Memory(MF-LSTM)neural networks to predict cash fow of listed companies.By designing the structure of the network,mixed frequency information is combined effectively.Compared with traditional time series models,our models enjoy better performance significantly.In addition,experimental results show that account data is an effective explanatory variable for cash flow prediction.

关 键 词：现金流 时间序列预测 混频模型 LSTM 

分 类 号：F832.5[经济管理—金融学] F275.5[理学—概率论与数理统计] O212[理学—数学] TP183[自动化与计算机技术—控制理论与控制工程]