CAJ | 학술논문

本文以2005～2009年首次被实施ST的中小上市公司为研究对象，综合利用财务指标和公司治理指标并采用考虑极端值样本的稳健logistic回归构建了财务失败预警模型。研究结果显示，公司治理指标对中小上市公司的财务失败预警具有重要作用，在财务指标基础上加入公司治理指标可以提高财务失败的预测效果，而这种预测效果的提升在排除了含有财务指标极端值样本的情况下表现得更加显著。研究结论对开展中小企业信用风险评估问题具有一定的参考价值。
본문이2005～2009년수차피실시ST적중소상시공사위연구대상，종합이용재무지표화공사치리지표병채용고필겁단치양본적은건logistic회귀구건료재무실패예경모형。연구결과현시，공사치리지표대중소상시공사적재무실패예경구유중요작용，재재무지표기출상가입공사치리지표가이제고재무실패적예측효과，이저충예측효과적제승재배제료함유재무지표겁단치양본적정황하표현득경가현저。연구결론대개전중소기업신용풍험평고문제구유일정적삼고개치。
In recent years, the problem that it＇s hard to financing for small-and-medium enterprises （SMEs） has been caused high attention. One of approaches to solve this problem is enable commercial banks to increase loans for SMEs. However, Chinese commercial banks have being short of a suitable credit risk measure model for SMEs for a long time, which constraint their willingness of providing loans for SMEs. Therefore, it is of great im- portance to develop a suitable credit risk measure model for SMEs for Chinese commercial banks. Although many domestic researchers had been modeling credit default risk prediction, their models mainly focus on big enterprises but rare models are focus on SMEs. However, some recent literatures show that there exists credit risk heterogeneity between big enterprises and SMEs. The models of credit default risk based on big enterprises are not adaptable to SMEs, it is necessary to model credit default risk models for SMEs separately. Currently, logistic model has become the most popular approach in modeling credit default risk, but logistic regression is easily susceptible for outliers sample and the parameters estimation will be caused a deviation, which will reduce its default prediction ability of out-of-sample. Moreover, this estimation deviation will be larger for SMEs because of SMEs＇ data are often far from prefect theoretically. To solve this problem, we introduce a robust logistic regression developed by Atkinson and Riani （2001,2006） to improve the estimation of parameters in logis- tic model. Besides, only financial ratios are considered in most current domestic literatures, while non-financial information such as corporate governance factors are ignored. However, some researchers argue that corporate governance factors may affect the quality of financial information of companies, and thus the effectiveness of measuring credit default risk. So, it is necessary to take the corporate governance information into account when modeling credit default risk. Inspired by existing literatures above, we take financial distress events （Special Treatment events ） as the proxy indictor for credit default risk. Based on the sample of first-ST listed companies during the period of 2005 2009, we employ both financial ratios and corporate governance factors to model a robust logistic model of credit default for SMEs, and compare its prediction power with logistic model. The empirical result shows that whether include corporate governance indictors or not, the robust logistic model performs better than the logistic model in both discrimination power within sample and prediction power out-of-sample. The empirical result also shows that the prediction power will be improved when model specification takes corporate government indicators into account, because it eliminates the financial information distortion caused by potential earnings management behavior of companies. This paper is organized as follows. In Section II, we introduce the principle and estimation procedure of robust logistic regression. The principle of robust logistic regression is that we eliminate outliers sample with logistic regression, then employ logistic regression again based on the remaining sample excluding outliers, and thus obtain the parameters estimation. In order to minimize the ＂masking effect＂ in estimation procedure of robust logistic regression, we use a forward search approach to detect the potential outliers suggested by Atkinson and Riani （2001, 2006 ）. In Section III, we discuss the data source and variables application. In Section IV, we establish an empirical work under the model specification of robust logistic model, and compare its prediction power with logistic mod- el. In Section V, we come to a conclusion. In sum, robust logistic model eliminates the disadvantage influence of outliers sample for parameters estima- tion, and compared with logistic model, the parameters estimation is more stable and the prediction power is higher in robust logistic model. Prediction power will achieve best when both financial ratios and corporate governance indictors are included altogether in robust logistic model. Our conclusion shows that financial distress risk and credit risk of SMEs not only depend on financial condition, but also on corporate governance condition, and that corporate governance information cannot be ignored when modeling financial distress prediction and credit risk for SMEs. Although our findings based only on the sample of listed SMEs, the conclusion can be also equally applied to unlisted SMEs, which has credit pricing implications for commercial banks.