Elsevier

Decision Support Systems

Volume 136, September 2020, 113364
Decision Support Systems

Evaluating the credit risk of SMEs using legal judgments

https://doi.org/10.1016/j.dss.2020.113364Get rights and content

Highlights

  • Loan default prediction model with judgments brings economic benefits to a bank.

  • Compared with the baseline model, the AUC of model with judgments is 4.6% higher.

  • Not all judgments affect credit risk of SMEs.

  • The relative value of judgment amount influences predicting credit risk of SMEs.

Abstract

Loan application assessments of small and medium-sized enterprises (SMEs) are difficult because of information asymmetry. To mitigate the information asymmetry, this paper focuses on information found in legal judgments involving the company and its principles and combines this information with financial and firm-specific information to help evaluate the credit risk of SMEs. We propose a framework to identify legal judgments that are effective in predicting credit risk and extract relevant features that are contained within the effective legal judgments. Empirical evaluation shows that features extracted from effective legal judgments significantly improve the discrimination performance and granting performance of our model compared with the baseline model, which uses financial and firm-specific features only.

Introduction

Small and medium-sized enterprises (SMEs) significantly contribute to the economy by creating wealth and employment opportunities [16]. However, it is difficult for SMEs to get financial support from the credit market for further development, such as for expanding production capacity [25]. The main reason for this difficulty is information asymmetry between the SME and credit market, including an incomplete system of financial system records, information opacity, and so on [6,8]. Thus, the credit market prefers to lend money to large companies for higher profits and security rather than to SMEs [19,28].

To address the problem of information asymmetry, many studies have focused on non-financial information, such as the firm age, management style, number of employees, and characteristics of the board of directors, to evaluate credit risk [22,23]. Other studies have focused on the productive efficiency [29], business plan [1], and financial reports [34]. However, since SMEs lack mature management [9] and an information disclosure mechanism [5], there are significant challenges in obtaining non-financial information and verifying the authenticity of the information.

In this study, we use published legal judgments (which are referred to hereafter as judgments) involving the enterprise and/or its principles as non-financial information and combine it with the company's financial and firm-specific information (which are referred to hereafter as basic information) to evaluate the credit risk of manufacturing SMEs. Each judgment reflects a dispute between the enterprise and others that arose in the process of business operation. As we know, the legal sanction in a judgment has a more negative effect on SMEs than on large companies. For example, if a court orders an SME to pay RMB 5,000,000 to another party, the payment may be a hardship that hurts the operation of the SME, or even causes it to close down. However, the same payment may be affordable for a large company. Additionally, judgments include rich information that reflects the risk of default of SMEs to some extent. Consider a case in which an enterprise as defendant was sued for a private lending dispute; the case may reflect the poor credit record of the enterprise because of a lack of willingness to repay and/or an insufficient ability to repay. This information could be useful in evaluating the enterprise's credit risk. Since the judgments are openly available on the Internet, using judgments reduces information collection costs for banks and guarantees the authenticity of the information. The challenges in using judgments are the difficulty in identifying and selecting effective judgments, and the difficulty in extracting important features from the varying structures of the judgment text, features that can improve the performance of the prediction model.

Our study aims to address these challenges by using a framework consisting of three stages. First, based on the legal lexicon and structuring rules, we extract structured information such as the judgment code, date of judgment, lawsuit status of the loan applicant, cause of action, judgment result, and amount awarded. Then, based on taxonomy, we build a method to classify the judgments into four categories based on the lawsuit status and judgment result, and identify the influence of each judgment category. Third, we use the chi-squared test and logistic regression method to identify features with high predictive power that can be extracted from effective judgments. We add these features to prediction model for predicting the loan default probability of SMEs. The empirical evaluation shows that the discrimination performance and granting performance of our model are significantly improved by the addition of the judgment information.

This study makes several important contributions to research and practice. To the best of our knowledge, it is the first study to identify the utility of judgment text in predicting the credit risk of SMEs. We find that not all judgments affect credit risk; the judgments with an effect are the ones where the lawsuit status of the loan applicant and judgment results are negative for an applicant (the negative lawsuit status includes defendant, appellee, and so on; the negative judgment results include paying money, freezing assets, and so on). We use text mining techniques to extract features from the judgment text that can significantly improve the discrimination performance of the prediction model. We find that two variables—the number of judgments arising from disputes about loan contracts, and the ratio of the award amount to the yearly income of the company being greater than 12.15%—significantly increase the prediction power of the model over that of the benchmark. In addition, we test the granting performance of the models and examine whether including judgments in evaluating loan applications can bring economic benefits to a bank due to a lower default rate. These results demonstrate that the use of judgments is an effective and low-cost approach for banks and other lending institutions to reduce the losses caused by loan defaults.

The remainder of the paper is organized as follows. The relevant literature is reviewed in the next section. In section 3, we present the proposed framework for mining valuable information from judgments. We describe the empirical evaluation and analysis in section 4. Finally, we conclude this study by summarizing our contributions and discussing future research directions in section 5.

Section snippets

Literature review

Decisions on an SME's loan application by banks are generally based on the creditworthiness [35]. Calabrese et al. [3] argued that if a bank can better predict whether or not SMEs will default on their loans, it would be more efficient for the credit system. Thus, accurate assessment of credit risk plays a crucial role in solving the problem of loan availability for SMEs. Credit risk evaluation has focused on two aspects: determining the features impacting credit risk, and predicting the

Proposed framework

We propose a framework to extract valuable information from judgments and use this information to evaluate the credit risk of SMEs. The framework is illustrated in Fig. 1. To mine information from judgments, we use text mining methods that convert judgment documents into structured information. To identify which judgments are effective in predicting credit risk, we consider two aspects: time and judgment category. Regarding time, we analyze the time span between the date of the judgment and the

Data

We evaluated the proposed framework on a dataset collected from a commercial bank in the Anhui province of China, and from the websites “China Judgments Online” and “www.qcc.com.” To obtain the entire loan profile, we collected the credit loan records and financial data of SMEs that applied for a 12-month loan between 2015 and 2017 (ending between 2016 and 2018). We calculated ten financial features (see Table 2, No. 1 to No. 10) from one year of financial data before the loan application date.

Conclusion and future directions

In this article, we study how legal judgments related to an enterprise can be used to complement basic financial and other information in evaluating the credit risk of SMEs. We propose a framework for mining valuable information from judgments and incorporate this information into our loan default prediction model to help with decision making. Specifically, we examine the effect of four categories of judgments on the default prediction. We extract relevant features from selected judgments and

Acknowledgements

This work was funded by the National Natural Science Foundation of China (Grant Nos. 71731005, 71571059) and the Fundamental Research Funds for the Central Universities (Grant No. JZ2020HGQA0173).

Chang Yin is a doctoral student at the School of Management, Hefei University of Technology, China. Her research interests include credit risk evaluation of SMEs and data mining.

References (39)

Cited by (44)

  • Financial risk tolerance profiling from text

    2024, Information Processing and Management
View all citing articles on Scopus

Chang Yin is a doctoral student at the School of Management, Hefei University of Technology, China. Her research interests include credit risk evaluation of SMEs and data mining.

Cuiqing Jiang is a Professor at School of Management, Hefei University of Technology. He received his PhD degree in 2007 from Hefei University of Technology. His research interests include big data analytics and business intelligence, data mining and knowledge discovery, information systems, and financial technology (Fintech). He has published in such journals as Journal of Management Information Systems, European Journal of Operational Research, Information Sciences, Decision Support Systems, and International Journal of Production Research.

Hemant K. Jain is W. Max Finely Chair in Business, Free Enterprise and Capitalism and Professor of Data Analytics, in Gary W. Rollins College of Business at University of Tennessee Chattanooga. He is internationally acclaimed for his pioneering work on Effectiveness of Presentation of Product Information in E-Business Systems. His work has appeared in Information Systems Research, MIS Quarterly, IEEE Transactions on Software Engineering, Journal of MIS, IEEE Transactions on Systems Man and Cybernetics, Naval Research Quarterly, Decision Sciences, Decision Support Systems, Communications of ACM, and Information & Management. He served as Associate Editor-in-Chief of IEEE Transactions on Services Computing and as Associate Editor of Journal of AIS. Recently he served as Program Chair, of IEEE International Conference on Big Data. He received his Ph. D. in information system from Lehigh University, a M. Tech. from IIT Kharagpur, and B. E. University of Indore, India.

Zhao Wang is an Assistant Professor at the School of Management, Hefei University of Technology. He received his PhD degree in management science and engineering from that university. His research interests include data mining and credit scoring. He has published in such journals as Journal of Management Information Systems, European Journal of Operational Research, Annals of Operations Research, Electronic Commerce Research and Applications, and many others.

View full text