﻿ 基于Z-Score指标决策树的财务风险预警模型研究 Financial Risk Early Warning Model of Decision Tree Research Based on Z-Score Indicators

Modern Management
Vol.07 No.05(2017), Article ID:21959,9 pages
10.12677/MM.2017.75026

Financial Risk Early Warning Model of Decision Tree Research Based on Z-Score Indicators

Yunyi Yang, Zhonghe Sun, Jing Lu

IT Center, China National Audit Office, Beijing

Received: Aug. 16th, 2017; accepted: Aug. 31st, 2017; published: Sep. 7th, 2017

ABSTRACT

Financial risk early warning model is one of the primary means of forecasting financial crisis. Z-score model has been applied widely since it has precise formula and good practicability. However, as regard to the adequacy of using Z-score model to forecast financial crisis for domestic companies, there is a dispute. Therefore, we collect financial statements data from 2007-2013 of Shanghai listed companies, and prove that Z scores between ST companies and non-ST companies are linearly inseparable. It reveals that Z-score model with linear separable principle in nature, has its own limitation. Hence, this paper brings up a financial risk early warning model of decision tree based on Z-score indicators and it solves the problem of Z-score model. The empirical research shows that the decision model given by this paper can forecast financial crisis three years before it happens with 75.37% accuracy, and with 95.45% accuracy one year ahead of crisis occurance.

Keywords:Financial Risk Early Warning, Z-Score, Decision Tree

Copyright © 2017 by authors and Hans Publishers Inc.

1. 引言

Z分数模型以其严谨的公式和较强的操作性被广泛应用。美国学者Edward I. Altman (1968)选取1946~1965年间的33家破产公司和33家正常经营的公司，提取出5种最具共同预测能力的财务比率，建立起类似于回归方程式的鉴别函数——Z分数模型 [2] 。其中，Z分数为五项财务指标的加权和，推荐以1.81为是否有破产危机的临界值，如果Z值低于1.81，则公司存在很大的破产危险。

2. Z分数预警模型适用性分歧

Z分数模型的基本原理是用基于多重指标的线性多元函数识别有财务风险的公司。经典Z分数模型的表达式为：Z = 1.2X1 + 1.4X2 + 3.3X3 + 0.6X4 + 0.99X5。其中：Z为判别函数值；X1 = 营运资产/总资产；X2 = 留存收益/总资产；X3 = 息税前利润/总资产；X4 = 权益市Z场价值/负债账面总价值；X5 = 销售收入/总资产。奥特曼提出判断公司破产的临界值：若Z > 2.675，表明公司的财务状况良好，发生破产的可能性较小；若Z < 1.81，则公司存在很大的破产危险，即1.81是公司破产发生的定量界定。

3. 基于上市公司财报的Z分数模型有效性分析

3.1. 数据准备

3.2. Z分数模型有效性实证分析

Table 1. Overview of adopted data sets

Figure 1. Prediction result of enterprise financial risk in 2012

Table 2. Overview of prediction result of Z-Score model in 2012

4. 基于决策树的财务风险预警模型

4.1. 基于决策树的财务风险预警模型

4.2. 基于决策树的财务风险预警模型规则解释

Figure 2. Decision tree under the imbalanced data

Table 3. Rules of model for financial risk judgment

4.3. 基于决策树的财务风险预警模型验证

Table 4. Prediction result of decision tree under balanced and imbalanced data with the duplicated rare instances

58.33%；提前2年预测的准确率为88.54%，覆盖率为63.91%；提前3年预测的准确率为75.37%，覆盖率为63.52%。在参数平衡数据状态下，从表4的模型验证结果看出，结果与非平衡数据状态下类似，在未来三年内的无破产危机公司的预测都比较准确，准确率和覆盖率都达95%以上。而对于有破产危机公司的预测，数据2倍平衡下的结果比数据非平衡下的结果略胜一筹。虽然准确率要比非平衡数据的结果稍差一些，但在覆盖率上能达到70%以上。提前1年预测破产的准确率为75.71%，覆盖率为73.61%；提前两年预测破产的准确率为75.74%，覆盖率为77.44%；提前三年预测破产的准确率为62.63%，为74.84%。总体看来，在参数2倍平衡状态下的模型在准确率和覆盖率两方面表现更均衡一些，但值得注意的是，随着预测年份的推移，两种模型的准确度都有所下降。

5. 总结

Financial Risk Early Warning Model of Decision Tree Research Based on Z-Score Indicators[J]. 现代管理, 2017, 07(05): 185-193. http://dx.doi.org/10.12677/MM.2017.75026

1. 1. 常树春, 范文雨. 基于Z-score模型的汽车行业上市公司财务风险实证分析[J]. 商业会计, 2013(13): 89-91.

2. 2. 陈珍静. Z-score模型在我国上市公司财务预警中适用性的探讨——基于交通运输设备制造业的实证分析[J]. 国际商务财会, 2011(4): 60-63.

3. 3. 胡剑. 论企业财务风险的防范[J]. 会计师, 2014(4x): 47-48.

4. 4. 李艳. 基于Z-Score模型对我国社会服务业上市公司破产预测的实证分析[J]. 商, 2012(7): 84, 109.

5. 5. 邰丽娜. Atman’s Z-score模型对我国上市公司的适用性分析[J]. 财会研究, 2011(24): 49-51.

6. 6. 王芳云. 运用Z计分模型对上市公司财务风险的实证研究[J]. 山东纺织经济, 2005(5): 42-44.

7. 7. 向德伟. 运用“Z记分法”评价上市公司经营风险的实证研究[J]. 会计研究, 2002(11): 53-57.

8. 8. 徐秀渠. Altman’s Z-Score模型在企业风险管理中的应用研究[J]. 经济经纬, 2010(4): 103-106.

9. 9. 叶海英. 对上市公司财务预警的实证研究——Z计分模型[J]. 中国外资, 2011(24): 84-85.

10. 10. 张楚堂, 杨志强. 基于Z-score模型的国有公司财务分析[J]. 武汉理工大学学报, 2009(15): 145-149.

11. 11. 张蔚虹, 朱海霞. Z-Score模型对科技型上市公司财务风险预警的适用性检验[J]. 科技管理研究, 2012, 32(14): 228-231.

12. 12. 周剑涛, 庞文凤. Z-SCORE财务预警模型在上市公司应用的实证研究[J]. 财会通讯, 2009(24): 25-27.

13. 13. Altman, I.E. (1968) Financial Ratios Discriminant Analysis and the Prediction of Corporate Bankruptcy. Journal of Finance, 4, 589-609.

14. 14. Han. J. and Kamber, M. (2000) Data Mining: Concepts and Techniques.