摘要
摘要
近两年来由于疫情的持续影响以及中美两国经贸摩擦不断升级，我国企业所处的
市场环境竞争也越发激烈，对企业的经营管理，特别是财务危机管理的要求也越来越
高。面对瞬息万变的市场环境，诱发企业出现财务危机的因素也更加多种多样，所以
如何构建更加科学有效并且高精度的财务危机预警模型对于企业实现更高质量的财
务管理目标，就显得尤为重要。目前已有的财务危机预警模型由于受到数据分析技术
发展的制约，大都基于简单的 Z分数模型、支持向量机、BP神经网络模型等，在构
建模型时会存在样本资料选择范围不够广、指标体系选取不全面、建模过程受人为因
素影响较大等不足，因此在大数据时代的背景下，针对上述问题本文利用人工智能领
域中的深度学习技术来构建一个预测性能更高的财务危机预警模型。
首先在回顾梳理国内外研究者进行的相关研究成果的基础上，从以下五个维度：
偿债能力、运营能力、盈利能力、企业成长能力和现金流能力来从 WAND数据库中
选取反映上市公司财务状况的财务指标，同时又选取其他同样可以反映企业财务状况
变化的非财务指标，搭建了包含“所属行业”这一虚拟变量在内的 48个指标的指标
变量体系，其次在此基础上，筛选出沪深两市 1999年至2020年的 749家 A股上市
公司 T-2和 T-3年的财务报表数据作为样本数据进行研究，然后在验证了财务指标在
行业间存在显著性差异的基础上分别建立了基于行业差异性和卷积神经网络的上市
公司财务预警模型、不考虑行业差异性的基于卷积神经网络的财务预警模型以及构建
基于行业差异性的 BP神经网络财务预警模型。
实证研究结果表明，基于行业差异性和卷积神经网络的财务预警模型预测准确率
为 95.11%，较不考虑行业差异性的卷积神经网络预警模型的预测准确率 91.56%和基
于行业差异性的 BP神经网络财务预警模型的预测准确率 87.33%都有了明显的提升。
验证了新模型具有更准确的预警评估性能，并进一步提升了财务危机预警模型的稳定
性和可适用性。
关键词：行业差异；卷积神经网络；上市公司；财务预警模型
I
Abstract
Abstract
In the past two years, due to the continued impact of the epidemic and the escalating
economic and trade friction between China and the United States, the market environment
in which Chinese enterprises operate has become more and more competitive, and the
requirements for enterprise management, especially financial crisis management, have
become higher and higher. In the face of the rapidly changing market environment, there
are many different factors that can trigger a financial crisis, so it is especially important to
build a more scientificand effective financial crisis earlywarning model with high
accuracy forenterprises to achievehigher quality financialmanagement goals.The
existing financial crisis early warning models are mostly based on simple Z-score model,
support vector machine andBP neural network model due to theconstraints of the
development of data analysis technology, and there are shortcomings in the construction of
the model such as the sample data selection is not wide enough, the selection of index
system is not comprehensive, and the modeling process is influenced by human factors, etc.
Therefore, in the background of the big data era, this paper uses the deep learning in the
field of artificial intelligence to address the above problems. Therefore, in the context of
the big data era, this paper uses the deep learning technology in the field of artificial
intelligence tobuild afinancial crisisearlywarning modelwith higherprediction
performance.
Firstly, on the basis of reviewing the relevant research results conducted by domestic
and foreign researchers, the financial indicators reflecting the financial status of listed
companies were selected from the WAND database based on the following five dimensions:
solvency, operational capacity, profitability, enterprise growthcapacity and cash flow
capacity, while other non-financial indicators reflecting the changes in the financial status
of enterprises were also selected to build a system of 48 indicators including the dummy
variable "industry". On this basis, the financial data of 749 A-share listed companies from
1999 to 2020 in Shanghai and Shenzhen were selected as sample data for the years T-2 and
T-3, and then the financial indicators were verified to be significantly different among
industries. Then, basedon the verification of thesignificant differences of financial
indicators betweenindustries, thefinancial earlywarning modelbased on industry
differences and convolutional neural network, the financial early warning model based on
convolutional neural network without considering industry differences, and the financial
II
Abstract
early warningmodelbased onindustry differenceswithBP neuralnetwork were
established respectively.
The results of the empirical study show that the prediction accuracy of the financial
early warning model based on industry variability and convolutional neural network is
95.11%, which is a significant improvement over the prediction accuracy of 91.56% of the
convolutional neural network early warning model without considering industry variability
and the prediction accuracy of 87.33% of the BP neural network financial early warning
model based on industry variability. It verifies that the new model has more accurate early
warning evaluation performance and further improves the stability and applicability of the
financial crisis early warning model.
Key words: industry differences; Convolutional neural network; listed company; Financial
early warning model
III