Research on FinTech

Consumer Attraction and Retention

1. Mobile Effects on Interrelated Two-Sided Financial Decisions: Comprehensive Evidence from Field Experiments on Peer-to-Peer Lending Platforms (with Sihan Fang, Eric Hyeokkoo Kwon, Yingjie Zhang). Under review. [Paper link] [NET Institute Summer Grant 2023]

In the finance sector where borrower and lender behaviors are interrelated, quality and quantity on one side can impact decision-making processes on the other side. We investigate how mobile channels shape the behaviors of both borrowers and lenders in the context of peer-to-peer lending platforms. We also examine the corresponding impacts on credit risk management and economic returns. The empirical results illustrate that mobile borrowers are more likely to terminate loan submissions, especially during peak commuting hours. Meanwhile, mobile lenders demonstrate a greater tendency to approve loan applications within a shorter period. Surprisingly, we observe no change in the quality of submitted or approved loans. Considering the improved debt collection capabilities afforded the platform by the mobile channel, we recognize that mobile adoption enhances profit.  

2. Being Emotionally Supportive: Exploring the Value of Emotional Appeals Retargeting in Recovering Consumers of FinTech Services (with Xianghua Lu, Hui Yang, Peter Yun Zhang). Production and Operations Management, 2024, 33(1): 166–183. [Paper link] 

This study explores the value of two types of emotional appeals retargeting strategies tailored for financial services: empathetic support and privacy commitment retargeting. Our field experiment involving a microloan platform reveals that empathetic support retargeting successfully recovers 11.93% of consumers, whereas privacy commitment retargeting recovers 17.27% of consumers. These effects translate to 5.19 and 10.53 percentage-point increases in the rate of return, respectively, over the no-message operation. This study further evaluates the quality of consumers retained by these retargeting strategies and finds that consumers called back by the empathetic support and privacy commitment retargeting present high credibility, with loan approval rates 11.4 and 13.69 percentage points higher than those of nonabandoning consumers, respectively.  

Credit Risk Assessment and Loan Decision (Improve both Financial Profitability and Inclusiveness)

3. Profit vs. Equality? The Case of Financial Risk Assessment and A New Perspective on Alternative Data (with Yingjie Zhang, Beibei Li). MIS Quarterly, 2023, 47(4): 1517-1556. [Paper link] [ICIS 2019 Best Paper Award; Facebook Economic Opportunity and Digital Platforms Research Grant 2019] 

We investigated how alternative data from smartphones and social media can help mitigate potential financial inequality while preserving business profitability in the context of financial credit risk assessment. We found that profiling user financial risk using smartphone activities is 1.3 times more effective in improving financial inclusion than using online social media information (23.05% better vs. 18.11%), and nearly 1.3 times more effective in improving business profitability (42% better vs. 33%). Surprisingly, we found that using consumers' online shopping activities for credit risk profiling can hurt financial inclusion. Furthermore, we investigated potential explanations for financial inclusion improvements. Our findings suggest that alternative data, especially users' smartphone activities, not only demonstrate higher ubiquity but also appear to be more orthogonal to conventional sensitive demographic attributes. This, in turn, can help mitigate statistical bias driven by the unobserved factors or underrepresentative training samples in machine-based risk assessment processes.  

4.  Inclusive Decision Making via Contrastive Learning and Domain Adaptation (with Xiyang Hu, Yan Huang, Beibei Li). Under 3rd round revision. [Paper link]

We propose a novel multi-task adversarial deep learning model that explicitly considers both profit and fairness for credit risk modeling. Our approach makes use of sophisticated machine learning (ML) models' ability to triangulate (i.e., infer the sensitive group affiliation by using only permissible features), which is often deemed "troublesome" in fair ML research, in a positive way to improve fairness. In the big data era, with the availability of fine-grained user-behavior data, triangulation becomes more and more concerning due to features' predictive power for sensitive group attributes. Our proposed model addresses this important problem. Our model significantly outperforms regular deep neural networks and the currently most popular credit risk model XGBoost, in terms of improving lenders’ profits and promoting fairness in allocating financial resources between borrower gender groups. Our empirical findings also demonstrate that the traditional AUC accuracy metric does not always align with a model's performance on the borrowers’ welfare and lenders' profits, which are the objectives the decision-maker intends to maximize/improve. This result highlights the benefit of incorporating the profit function into ML models' objective function for algorithmic decision-making.

5.  Turning Triangulation into Opportunity: A Dual-Adversary Algorithm for Fair and Profitable Credit Risk Modeling (with Xiyang Hu, Yan Huang, Beibei Li). Under review. [Paper link]

Traditional machine learning algorithms for credit evaluation face the selective labels problem, as the training data usually only contains default outcome labels from approved loan applications that tend to represent borrowers with more favorable socioeconomic characteristics. Consequently, these algorithms struggle to effectively generalize to disadvantaged borrowers. To overcome this problem, we introduce a Transformer-based loan screening model that leverages self-supervised contrastive learning and domain adaptation. Our model uses contrastive learning to train our feature extractor on unapproved (unlabeled) loan applications and employs domain adaptation to generalize the performance of our label predictor. We evaluate our approach on a real-world micro-lending dataset and demonstrate its effectiveness. The results show that our approach significantly enhances inclusion in funding decisions, while simultaneously improving loan screening accuracy and lender profit by 7.10% and 8.95%, respectively. Additionally, we find that incorporating test data and labeling a small ratio of it further enhances model performance.

6. Leveraging Multi-source Heterogeneous Data for Financial Risk Prediction: A Novel Hybrid-Strategy-based Self-adaptive Method (with Gang Wang, Gang Chen, Huimin Zhao, Feng Zhang, Shanlin Yang).  MIS Quarterly, 2021, 45(4): 1949-1998. [Paper link]

The study proposes a new feature-sparsity learning method to adaptively integrate multisource heterogeneous soft features with hard features and a proposed improved evidential reasoning rule to adaptively aggregate base classifier predictions, thereby alleviating both the declarative bias and the procedural bias of the learning process. Evaluation in two cases at the individual level (concerning borrowers at a P2P lending platform) and the company level (concerning listed companies in the stock market) showed that, compared with relying solely on hard features, effectively incorporating multisource heterogeneous soft features using our proposed method enabled earlier prediction of financial risks with desirable performance.  

Behavior Tracing and Repayment Reminding

7. Prosocial Compliance in P2P Lending: A Natural Field Experiment (with Ninghua Du, Lingfang Ivy Li, Xianghua Lu). Management Science, 2020, 66(1): 315-333. [Paper link]

We implement behavioral mechanisms in a natural field experiment to increase loan repayment rates on a peer-to-peer (P2P) lending website. The results show that text message reminders that convey lenders’ positive expectations considerably increase the likelihood that borrowers will repay their loans, whereas reminders emphasizing the adverse consequences of failure to repay loans do not have enduring effects. Our experiment results in an increase in loan repayments in the sample period.

8.  From Credit to Savings: Spillover Effects of Credit on Individual Consumption with High-resolution Transaction Data (with Zenan Zhou, Jiayan Han, Yingjie Zhang). Reject & resubmit. [Paper link] [Tencent Research Grant 2021]

The study examines how individual users respond to credit from both a "vertical" (i.e., time-varying) and a "horizontal" (i.e., spillover) perspectives. The results showed that credit users' consumption amounts significantly expands, by 65.4%, after gaining access to credit in the short term. However, over the long term, those users would ultimately curtail that initial consumption expansion to only 16.9% in order to cope with financial constraints. We also reveal and quantify the spillover effects of credit on consumption financed through savings channels. We rationalize users’ consumption-behavior changes after credit activation by drawing on the regulatory focus theory. Our empirical examination, having separated users into promotion- and prevention-focused types, demonstrates diverse consumption patterns (in both the "vertical" and "horizontal" dimensions) between them, thus validating our theoretical foundations.

Post-loan Management and Debt Recovery

9. When Less Is More? Deep Reinforcement Learning-Based Optimization of Debt Collection (with Cenying Tracy Yang, Kan Xu, Beibei Li, Xianghua Lu). Under major revision. [Paper link]

We apply a deep reinforcement learning (DRL) algorithm to examine whether "harsh" debt actions are necessary and derived optimal collection strategies on a fine-grained dataset. Installment-level and loan-level optimized results reduce the frequency of harsh actions by 49.05% and 60.19%, respectively. This suggests that microloan platforms should deploy collection actions more cautiously. More interestingly, we show that installment-level and loan-level optimizations suggest somewhat different debt collection patterns across installments in a loan: installment-level optimization overall recommends avoiding the use of any harsh actions; by contrast, loan-level optimization recommends using very few actions in early stages but increasing the intensity of applications of (harsher) actions in late stages. Generally, loan-level optimization yields a higher recovery rate and greater economic gains than installment-level optimization or other commonly-used debt-collection strategies. This is probably owes to the fact that our DRL algorithm could capture the potential correlations across installments when optimizing at the loan level. Despite the overall superiority of loan-level optimization, our heterogeneity analysis further reveals that installment (loan)-level optimization should be used when borrowers with good socio-economic backgrounds fail to repay early (late) installments in a loan duration. Otherwise, the original intense strategy is more effective.

10.  Can Social Notifications Help to Mitigate Payment Delinquency in P2P lending? A Randomized Field Experiment (with Xianghua Lu, Chong Alex Wang, Ruofan Wu). Production and Operations Management, 2021, 30(8): 2564-2585. [Paper link]

We study whether and how a platform can leverage on borrowers’ social connections and use automatic social notifications in regulating repayment behavior. Our results indicate that notifying social contacts of a delinquent regarding the overdue payment significantly improves the repayment rate. Compared with the control group in which no notification messages are sent to social contacts, notification-triggered social sanctions and social support reduce the default rate by more than 50% in both core-circle and peripheral-circle groups. Furthermore, we find that social notifications targeted at peripheral-circle social contacts are only effective in the short term, and its effectiveness decreases with repeated use. By contrast, social notifications targeted at core-circle social contacts have a lasting effect.