Creditworthiness Analysis in E-Financing Businesses - A Cross-Business Approach Kun Liang 1,2, Zhangxi Lin 2, Zelin Jia 2, Cuiqing Jiang 1,Jiangtao Qiu 2,3 1 Shcool of Management, Hefei University of Technology, 193 Tunxi Road, Hefei, 230009, China; 2 Jerry S. Rawls College of Business Administration, Texas Tech University, 703 Flint Ave, Lubbock, TX 79409, USA 3 School of Economic Information Engineering, Southwestern University of Finance and Economics, 55 Guanghuacun Road, Chengdu 610074, China liangkun_fd@163.com; zhangxi.lin@ttu.edu; zelin.jia@ttu.edu; jiangcuiq@163.com; qjt163@163.com Abstract. To cope with the challenge of data scarcity in creditworthiness analysis for e-financing business, this paper proposes a cross-business analysis approach based on the assumption of behavior consistency for client in different e-commerce environments. By this approach we can analyze individuals creditworthiness by associating financial data on lending platforms and crossbusiness non-financial data on social media. We conceived three creditworthiness assessment models, and conduct the experimental study on Ant Financial Co-Creation Data Platform. The results verify that our crossbusiness creditworthiness analysis approach is effective. Keywords: Online lending, Creditworthiness, Cross-business data, Modeling, Data mining 1 Introduction Following the trend of financial disintermediation, innovative e-financing businesses, such as P2P (Peer-to-Peer), P2B (Peer-to-Business), P2G (Private-to- Government), crowdfunding, and so on, provide diversified funding services directly to small businesses and consumers through various online platforms. For example, Alibaba launched its online microloan services in 2010, namely Aliloan, which has issued about 25 billion dollars loans by 2014, benefiting more than one million small and micro-sized enterprises [1]. In review an online loan application, a client s ability and willingness to fulfill contracts, i.e. his/her creditworthiness, is an important indicator [2], because creditworthiness analysis can effectively reduce the information asymmetry between financial suppliers and borrowers in the e-financing platform, and improve the accuracy of loan decisions. Existing creditworthiness analysis methods predicted individuals creditworthiness mainly based on the historical data formed in the lending business, such as payment history, credit usage, length of credit history; most of them are transactional financial
2 Kun Liang1,2, Zhangxi Lin2, Zelin Jia2, Cuiqing Jiang1,Jiangtao Qiu2,3 data [3, 4]. However, this kind of data is hard to obtain in e-financing businesses [3], while user-generated data are abundant but of non-financial, such as the social networking data in social media platforms and online reputation scores in electronic markets. These cross-business non-financial data can effectively reflect individuals creditworthiness from multiple perspectives [4]. For example, one person can play different roles in various e-commerce businesses. On one hand, he or she can be a borrower in an online lending business; on the other hand, he or she can also be a seller in a C2C market. In this context, many cross-business data, such as his or her online reputation scores formed in trading businesses (C2C transactions), can be used to assess his or her creditworthiness in the lending business [4]. This paper is intended to investigate that how individuals' creditworthiness-related non-financial data can be used in the e-financing business. Specifically, we study the following two questions. First, how effective the cross-business non-financial data can be used to analyze individuals creditworthiness in the e-financing business? Second, how to select reasonable cross-business non-financial indicators for creditworthiness analysis in e-financing businesses? The contribution of this study can be summarized as following: (1)We proposed a cross-business creditworthiness analysis approach to overcome the data scarcity problem of creditworthiness analysis in e-financing businesses; (2)We revealed that the social capital theory can be helpful in selecting reasonable cross-business nonfinancial data to improve the performance of creditworthiness analysis for e-financing businesses. 2 A cross-business creditworthiness analysis approach In this section, we propose a cross-business creditworthiness analysis approach based on the consistent attribute of creditworthiness. Adelson et al (2009) proposed that creditworthiness can be understood as the relative ranking of default frequency, and this relative ranking is consistent in different business and scenarios [5]. In addition, individuals creditworthiness is their ability and willingness to fulfill contract. Although individuals may play different roles and have to fulfill different contracts in various businesses, their creditworthiness is influenced by some common factors, such as their morality and sense of responsibility. Therefore, one person s creditworthiness is consistent in different businesses. We select individuals who are borrowers in the online lending market, and also sellers in the C2C market. Assuming that an individual s creditworthiness is consistent across different online platforms [5], we analyze these individuals creditworthiness in lending business by considering their behaviors in online commodity transactions, which could be reflected by their reputation and social capital. An individual s default behaviors could destroy his/her reputation that was previously accumulated as the intangible asset [6]. From this sense, the online reputation can effectively reflect one s creditworthiness. Furthermore, maintaining good creditworthiness can help one to gain more social resource such as friends, job opportunity, and so on [4]. Based on this idea, we conceive three models to analyze
Creditworthiness Analysis in E-Financing Businesses - A Cross-Business Approach 3 individuals creditworthiness in e-financing businesses with different indicator sets (see Figure 1). Model 1 analyzes individuals creditworthiness based on the indicator set which only contain financial factors (Fa); Model 2 counts online reputation factors (Fb) into the indicator set; Model 3 further adds social capital factors (Fc) into the indicator set. We are to study whether we can analyze individuals creditworthiness in e-financing business more accuracy by adding cross-business non-financial factors (Fb and Fc). Fig. 1. A cross-business creditworthiness analysis approach Table 1. The performance of different creditworthiness assessment models Models FN TN FP TP TPR TNR Cost DT 1 321 1293 161 205 0.3897 0.8893 3.7723 DT 2 250 1203 251 276 0.5247 0.8274 3.0243 DT 3 239 1173 281 287 0.5456 0.8067 2.9195 NN 1 320 1331 123 206 0.3916 0.9154 3.7348 NN 2 295 1328 126 231 0.4392 0.9133 3.4517 NN 3 283 1301 153 243 0.4620 0.8948 3.3334 LR 1 345 1336 118 181 0.3441 0.9188 4.0165 LR 2 302 1328 126 224 0.4259 0.9133 3.5315 LR 3 291 1329 125 235 0.4468 0.9140 3.4054 SVM 1 421 1401 53 105 0.1996 0.9635 4.8387 SVM 2 363 1369 85 163 0.3099 0.9415 4.1991 SVM 3 340 1351 103 186 0.3536 0.9292 3.9492 Model n (include DT n, NN n, LR n, and SVM n), n=1,2,3. Model 1, Model 2, and Model 3 refer to the models that use the indicator sets of Fa, Fa+Fb, Fa+Fb+Fc, respectively. For example, DT 1 is the decision tree model which adopt only financial indicators (Fa)
4 Kun Liang1,2, Zhangxi Lin2, Zelin Jia2, Cuiqing Jiang1,Jiangtao Qiu2,3 3 Experimental study We conduct an experimental study in Ant Financial Co-Creation Data Platform, which became available in June 2015 complying with the agreement between Ant Financial and Tongji University, to validate the feasible of proposed cross-business creditworthiness analysis approach. We compiled a dataset of 6,598 observations containing 366 variables dated 20 April, 2015. Each observation represents an individual s records in both online lending market and C2C market. The target variable, i.e. the creditworthiness of an individual is set either good or bad, determined by whether there is a late payment record in his/her loan history. We select four classification technologies to construct our creditworthiness assessment model, i.e. Logistic Regression (LR), Decision Tree (DT), Support Vector Machine (SVM), and the Neural Network (NN). We use three criteria to evaluate the performance of each model with different indicator sets. They are TPR, TNR and Cost, which were defined in [7]. TPR is the percentage of correctly classified bad creditworthiness. TNR is the percentage of correctly classified good creditworthiness. Cost is a aggregative criteria consider both TPR and TNR. The results for the comparison of different models and indicator sets are summarized in Table 1. In Table 1, the TPR of Model 2 is higher than that of Model 1. This result indicates that the model can identify more bad creditworthiness individuals when add the reputation factors. However, the TNR of Model 2 is slightly less than that of Model 1. This result shows that the ability to recognize the good creditworthiness individuals has a little decrease when add the reputation factors. Due to the different misclassification cost of good and bad creditworthiness [7], we use the criterion of Cost to determine the Models comprehensive ability to differentiate the good and bad creditworthiness. In general, Model that has a lower Cost is better in creditworthiness analysis. In Table 1, the Cost of Model 1 is higher than that of Model 2, which indicates that individuals reputation in Alibaba C2C business has significant prediction ability to their creditworthiness in Ali-loan business. Similarly, the Cost of Model 2 is higher than that of Model 3, which indicates that individuals social capital in Sina micro-blog (reflected the social capital accumulated in C2C transaction) has significant prediction ability to their creditworthiness in Ali-loan business. Compared to the performance of Model 1-3, we revealed that individuals creditworthiness in an e-financing business can be predicted by their creditworthiness in the trading business which is analyzed by their reputation and social capital (cross-business non-financial factors) formed in C2C transaction activities. This result verifies the consistent attribute of creditworthiness [5]. Now, we can answer the two questions proposed in the Introduction. First, we can analyze individuals creditworthiness in the e-financing business based on the cross-business non-financial data. Second, we can select reasonable cross-business non-financial indicators according to the reputation and social capital theory.
Creditworthiness Analysis in E-Financing Businesses - A Cross-Business Approach 5 4 Conclusions In order to solve the data scarcity problem in creditworthiness analysis for the e- financing business, this study proposed a cross-business creditworthiness analysis approach according to the consistent attribute of creditworthiness. This paper makes several contributions to the literature. We proposed a new creditworthiness analysis approach for e-commerce businesses, which analyze individuals creditworthiness in one business by considering their creditworthiness in the other business. In addition, we enlarge the application scope of reputation and social capital theory to crossbusiness creditworthiness analysis. From the perspective of practices, we provide a better creditworthiness assessment model for e-financing businesses, which can effectively improve the accuracy of loan decisions. The methodology of crossbusiness creditworthiness analysis can also be used in other e-commerce businesses, such as P2P lending. From this sense, we could provide e-commerce platforms with more insights about individuals creditworthiness from various business perspectives. However, this research is limited and must be further expanded. In this paper, evaluation indicators are restricted to structured data, and there is a lack of relevant indicators to reflect the relational aspect of social capital in creditworthiness analysis. In next step, we will analyze the creditworthiness through the relationship types of social network ties and indirect social network ties, which can effectively reflect the effect of relational aspect of social capital on creditworthiness. 5 References 1. The actual financing cost is only 6.7 %: Ali finance shocks the private lending, http://business.sohu.com/20130320/n369488586.shtml 2. Safi, R., & Lin, Z. (2014). Using non-financial data to assess the creditworthiness of businesses in online trade. PACIS Proceedings. 3. Wang, Y., Li, S., & Lin, Z. (2013, July). Revealing Key Non-financial Factors for Online Credit-Scoring in e-financing. In Service Systems and Service Management (ICSSSM), 2013 10th International Conference on (pp. 547-552). IEEE. 4. Lin, Z., Whinston, A. B., & Fan, S. (2015). Harnessing Internet finance with innovative cyber credit management. Financial Innovation, 1(1), 1-24. 5. Adelson, M., Ravimohan, R., Griep, C., Jacob, D., Coughlin, P., Bukspan, N., & Wyss, D. (2009). Understanding Standard & Poor s Rating Definitions.Standard & Poor's. http://www.standardandpoors.com/spf/delivery/assets/files/understanding_ Rating_Definitions. pdf. 6. Van den Bogaerd, M., & Aerts, W. (2015). Does media reputation affect properties of accounts payable? European Management Journal, 33(1), 19-29. 7. Lessmann, S., Baesens, B., Seow, H. V., & Thomas, L. C. (2015). Benchmarking state-of-the-art classification algorithms for credit scoring: A ten-year update. European Journal of Operational Research.