基于深度学习的企业财务困境预测方法研究

Research on Enterprise Financial Distress Prediction Method Based on Deep Learning

上市公司是金融市场中的主体,通过科学的方法对企业财务困境进行及时的预测、预警,避免企业陷入财务困境,对企业和整个金融市场来说意义重大。本文通过系统性的研究和分析,结合我国的金融市场情况,提出了基于深度学习方法的企业财务困境预测方法和模型,同时采用了具有较好的记忆能力和处理时间序列数据能力的LSTM神经网络模型,对ST企业的财务状况进行了分类预测,获取了企业年报中的文本信息、企业股票价格时间序列数据和非财务指标三种类型的实验数据进行了实验,将更多可能影响企业财务状况的因素纳入到实验中来。利用2020年ST和非ST企业的连续两年年度报告、连续三年的股票价格时间序列数据和指标数据进行训练,对2021年ST和非ST的企业进行财务困境的预测,并与其他常用的传统方法进行了对比分析。实验结果证明基于深度学习的方法具有更高的预测准确率和稳定性。

Listed companies are the main body in the financial market.It is of great significance for enterprises and the entire financial market to pay attention to and attach importance to the financial situation of enterprises,predict and warn the financial difficulties of enterprises in a timely manner through scientific methods,and avoid the financial difficulties of enterprises.Through systematic research and analysis,and combining with the situation of China's financial market,this paper proposes a method and model for predicting enterprise financial distress based on deep learning method.In this paper,the LSTM neural network model with good memory ability and the ability to process time series data is used to predict the financial situation of ST enterprises,and good prediction results are achieved in the experiment.We obtained three types of experimental data,namely text information in the annual report of enterprises,time series data of enterprise stock prices and non-financial indicators,and carried out the experiment,bringing more factors that may affect the financial situation of enterprises into the experiment.By using the annual reports of ST and non ST enterprises for two consecutive years in 2020,the time series data of stock prices and index data for three consecutive years for training,the financial distress of ST and non ST enterprises in 2021 is predicted,and compared with other commonly used traditional methods.The experimental results show that the method based on deep learning has higher prediction accuracy and stability.