Consumer finance exploded in 2017, which not only drove economic growth but also triggered a number of risks and challenges. Therefore, it is more essential for financial institutions to figure out how to regulate and optimize the business and assess the default risk of borrowers. Focusing on this topic, the session of Credit Risk Model for Individual Loan Contest at 2018 SAIF MF International Youth Leadership Finance Summit (2018 IYLFS) organized by Shanghai Jiao Tong University Shanghai Advanced Institute of Finance (SAIF) was successfully concluded on January 30th, 2018. Outstanding players from seven top-class universities in China went at the event hammer and tongs, joined by the spectacular presentations of SAIF MF16 and MF18 students, looking out for the future of China’s individual credit business.
2018 IYLFS assembled a jury of both scholars and industry leaders. Prof. Xiaomeng Lu, Assistant Professor of Finance, SAIF; Mr. Honghui Zhu, Researchers of China Academy of Financial Research (CAFR), SAIF; Mr. Fugong Wang, Vice President, Iuicity and Mr. Jiayue Chen, Head of Risk Controlling & Modelling Team, Kuainiu Group are invited as honorable judges for this session.
PKU Team: Profitability determines the availability of credit to defaulters
The modeling of Peking University (PKU) Team aims at identifying the loans with potential default risk, which is completed through five processes of pre-processing, factor analysis, log analysis, machine learning and model selection. The model demonstrates plenty of innovative highlights in terms of data identification, interpretation and processing.
In practice, financial institutions can use this model to decide whether a loan shall be granted to a past defaulter, depending on profitability instead of overdue performance. Moreover, three criteria are assessed in parallel for more effective application, that is, collection of personal information, description of applicant and follow-up and payment reminder after application.
THU Team: Anti-fraud is highly potential while it is important to prevent moral hazards in data collection
Tsinghua University (THU) Team built up two personal credit risk models. The first model is a linear model based on log regression, which explains how different factors affect results. And the second is a machine learning model, which is nonlinear and able to provide more and more precise forecasts.
The machine learning model is capable to extract more information related to social network, online shopping and market. Moreover, the application calls for more customized models, while it is important to prevent moral hazards in data collection. The other demanding concern is the collaboration between theoretical model and human operators.
SHUFE Team: Today’s competition relies on the depth of understanding on customers and the volume of data collection
Shanghai University of Finance and Economics (SHUFE) Team pointed out that the size of credit in China will keep on rising in future. Therefore, it is critical to establish a consumer credit model, which will help financial institutions measure credit risks. Today’s competition relies on the depth of understanding and the volume of data collection rather than mere understanding on customers.
SHUFE Team’s model represents two major benefits. First, it proves that XGBoost can better improve model performance. And second, it picks up several essential factors, such as communication behaviors, Sesame Rating and trading behaviors. In the future, it is possible to focus more efforts on these three items to upgrade the quality of data sources.
SJTU Team: It is imperative to build up machine-learning-based personal credit model
At present, though the default rate and cost of debt remain high in the personal credit market, the common credit rating models suffer data unavailability and inflexibility. Therefore, Shanghai Jiao Tong University (SJTU) Team set up a new model based on monitored machine learning, which is trained by XGBoost, processes data under PGA and provides the economic implications of various data. Compared to traditional models, it boasts accuracy, flexibility and robustness.
RUC Team: Data collection and sharing is key to the promising credit market in China
By comparing the development of China’s credit business to that in US, the Renmin University of China (RUC) Team believes the personal credit business in China is extremely promising, and thus built up a model inspired by ZestFinanc in US.
It processes data by setting up an infrastructure, performs log regression in four ways of machine learning, balances data sets with EBCA technique, converts raw data into characters through characteristic engineering and utilizes it in real-world credit decision-making. It encourages credit rating agencies to collect and share more data, and adapt the model by market scenarios, while the white box helps applicants with comparatively poor characters to improve their application results.
FDU Team: Rate model performance by precision, recall rate and accuracy
Fudan University (FDU) Team rates model performance by precision, recall rate and accuracy. The team wisely chose a threshold value, aiming to better distinguishing false positive and negative results, which features the rating accuracy of 85% in terms of log model. Therefore, financial institutions can see potential loss more precisely and pick up the best cutoff line, while taking consideration of a number of dependent and major variables in actual applications.
NJU Team: Sophisticated personal credit risk models help financial institutions minimize risks
The first of three models created by Nanjing University (NJU) Team focuses on qualitative analysis, which identifies the potential of overdue payment and the quality of a customer. The 2nd model forecasts overdue days through machine learning and in turn liquidity risk. And the 3rd model rates and profiles good customers so that financial institutions engage in targeted marketing for better profitability.
According to NJU Team, sophisticated models help financial institutions minimize risks and accelerate their FinTech transformation.
SAIF MF16/MF18 Team as a non-competitor: A credit rating model with valuable economic potential
As all available credit rating models are linear without more sophisticated quantitative techniques, SAIF MF Team presented its innovative version, which engages in characteristic engineering in the first place, adjusts the parameters of Random Forest Model, achieves an aggregation of simple averages, optimizes the weights with genetic algorithm and eventually arrives at the Numericon. Revealing valuable economic potential, it may be developed into a close-loop system to transform its performance improvement into real profitability.
Under the guidance of SAIF faulty and mentors from China Academy of Financial Research, SAIF MF Team benefits from closer communication with trade practitioners. Therefore, its presentation revealed more depth and insights than other teams, while its scenarios were more practical.
Along with the rapid growth of consumer credit and the rising NPL rate, financial institutions call for quality credit rating models. The presentations of all eight teams were exciting and inspiring. Within a short framework, the players developed new credit models through programming, experimentation, data analytics in different dimensions and reflection from various perspectives. Thanks to the remarkable platform structured by SAIF at this event, we believe that these active and effective explorations and ideas will foster the healthy evolution of China’s credit market and the financial sector at large.
“We are impressed by the presentations of all participants. Their performances have exceeded even our expectations. Hope that participants will know more about market needs after the summit and put their innovative thought into practice”, commented by judges from the industry.
Concluding remarks by Prof. Zhan Jiang, Faculty Director of Master of Finance Program, SAIF
Insightful comments from judges
At last, RUC Team won the competition, followed by SHUFE Team and SJTU Team as the 2nd and 3rd places respectively.
Mr. Jie Pan, Deputy Dean of SAIF, presented awards to the winners
About 2018 IYLFS
2018 International Youth Leadership Finance Summit (2018 IYLFS) centered on the hotspot domain of FinTech and exchanged on the three areas of Robo-advisor, Data Mining and Blockchain. The final competition focused on four topics of “Rotation Strategy within the Banking Sector”, “Design of Blockchain Application”, “Quantitative Asset Allocation Using Black-Litterman Model” and “Credit Risk Model for Individual Loan”.
After several rounds of keen competition, 130 outstanding undergraduates from 25 top-class universities and colleges across the world were shortlisted in the final among over 1,000 participants from world-class institutions, including not only local C9 and 985 universities like Peking University, Tsinghua University, Zhejiang University, Nankai University, Fudan University and Shanghai Jiao Tong University, but also international schools, such as London School of Economics and Political Science, University of Toronto, University of Waterloo, Pennsylvania State University, University of British Columbia, National University of Singapore, University of Hong Kong, The Chinese University of Hong Kong, Taiwan University and National Chengchi University.
Launched by SAIF MF Program in 2013, IYLFS is designed to providing a platform of collective learning and advancement for students from all kinds of higher education institutions, which will foster their capabilities of communication and organization. Moreover, it also aims to improving their understanding on financial theories and practices through presentation, competition and face-to-face dialogue with business leaders and scholars in the financial sector.
