Research On The Effectiveness And Mechanism Of Fintech-enabled Retail Lending For Commercial Banks

China’s retail lending market not only has a large customer base and diverse demand,but also is dominated by “credit invisible” consumers,which has led to an acute problem of insufficient and unbalanced credit supply of traditional banks.In recent years,the deep integration of digital technologies,such as big data and machine learning with financial services,has greatly promoted the digital transformation of the banking industry.Commercial banks have become the primary driver of the high-quality development of Fintech in China.While previous literature on Fintech extensively focuses on the economic effect of Fintech industry development measured at the macro or regional level,and mainly takes non-bank Fintech lending platforms as the research object,there is a lack of quantitative measures of the bank-level Fintech and it also remains unclear whether and how Fintech enables commercial banks’ retail lending.Compared with P2 P lending and non-bank Fintech lending,commercial banks have the advantage of being able to use both non-traditional data with multidimensional heterogeneity but relatively weak financial attributes,and traditional credit report data with strong financial attributes.Based on classical theories such as information asymmetry,credit rationing and financial innovation,this dissertation uses unique account-level data from leading licensed lending institutions in China as well as bankyear-level panel observations,and employs methods and technologies including multiple regression,classification prediction,and natural language processing to investigate three levels of issues in a sequential and progressive manner: the role of two types of typical non-traditional data,i.e.,individual online shopping orders and smartphone device attributes,in Fintech lending with respect to default risk assessment and consumer access to credit;the mechanism through which Fintech lenders jointly leverage big data,machine learning,and traditional credit report data to improve access to credit for “credit invisible”consumers;the bank-specific,time-varying measure of Fintech development level,and its effects on bank performances and financial inclusion.The core contents and important conclusions are as follows:Firstly,licensed lending institutions can use both traditional and non-traditional data.Using a unique account-level data set of 46,607 borrowers,this dissertation examines the predictive power of two types of digital footprints,i.e.,E-commerce credit score and online shopping behavior,in default risk prediction,and explores how technologyenabled lending practices can improve credit availability after dividing all available information into four categories,including self-reported information,central bank credit information,e-commerce credit score and online shopping behaviors.The results indicate that these two types of digital footprints can improve default prediction accuracy by approximately 50%,with the E-commerce credit score exhibiting the strongest predictive power.Nevertheless,online shopping behaviors still contain incremental information content beyond the e-commerce credit score.The use of digital footprints in fintech lending can not only establish digital credit scores for “credit invisible” consumers but also accurately assess the creditworthiness of borrowers who are undervalued by the traditional credit scoring system.Secondly,given the protection of personal privacy by commercial banks under strong regulation,this dissertation uses device attribute information obtained from smartphones used for loan applications of 105,242 borrowers to construct two sets of less privacy-sensitive variables,namely,device value attributes and system update behaviors,and examines the role of privacy-insensitive smartphone device attributes in identifying loan default.The results show that these device attributes can provide incremental information for identifying loan defaults beyond the financial and demographic variables that are commonly used in models predicting default risks.Furthermore,device value attributes and system update behaviors can potentially be an easily accessible proxy for otherwise difficult to collect individual economic status and cognitive ability,respectively,leading to discernible incremental information.Further analysis also reveals that borrowers have incentives to manipulate device attribute data,and lenders dynamically learn from multiple lending transactions to make full use of the device information.Thirdly,in light of the salient deficiency of traditional credit histories for “credit invisible” consumers,this dissertation develops a novel measure to identify the degree of credit invisibility by leveraging an individual-level dataset of 61,477 borrowers and calculating the proportion of missing data in their credit reports obtained from the official credit bureau.By examining the role of non-traditional data and algorithms-based lending provision in mining the hidden information and economic characteristics behind the missing data,it devotes to reveal the mechanism through which big data and machine learning algorithms in conjunction with traditional credit reports,improve access to credit of “credit invisible” consumers.The study finds that the use of big data and machine learning can effectively identify high-quality borrowers among the pool of creditinvisible populations by mining the hidden information behind the missing data of their credit reports.Furthermore,once these high-quality borrowers obtain(first-time)Fintech loans,their subsequent access to credit from traditional lending institutions will be also improved.This suggests that Fintech lending has a positive externality in terms of financial inclusion by establishing credit records for “credit invisible” individuals.Finally,different from existing literature that mostly focuses on the impact of nonbank Fintech firms and Fintech industry development level,in which the key Fintech metrics are measured at the macro or regional level,this dissertation first develops a bankspecific,time-varying measure of Fintech development level by exploiting over 170,000 public news texts,and examines its effect on bank performance and financial inclusion.The results show that the development of Fintech can promote the operation,service,and risk control capabilities,thus enhancing bank performance.However,while Fintech spurs the growth of bank loan volume,it does not reduce loan interest rates,which indicates a potential “broad but not cheap” problem in promoting financial inclusion.The main contributions of this dissertation are as follows:(1)Given the availability of data,the dissertation exploits two types of typical nontraditional data,i.e.,individual online shopping orders and smartphone device attributes,and analyses their capabilities of identifying and predicting loan defaults.It explores the economic mechanism that underlies the connections between these characteristics extracted from non-traditional data and loan default behaviors from the perspectives of being easily accessible proxies for economic status,personal traits,preference habits,behavioral patterns,and cognitive abilities that are difficult to collect and directly observe,and provides corresponding empirical supporting evidence.These results offer important guidance on how to understand the improvement of screening ability based on the big data and machine learning algorithms adopted by commercial banks.(2)Compared to non-bank Fintech lending platforms,commercial banks can use both traditional credit report data and non-traditional data to assess the credit qualities of borrowers in retail lending.To our knowledge,the dissertation is among the first to use the proportion of missing data in borrowers’ credit reports to measure their invisibility level by exploiting the characteristics of the retail lending market,in which “credit invisible” consumers are predominant.Through the perspective of big data and machine learning algorithms that can uncover hidden information behind missing credit data,this dissertation reveals the plausible mechanisms through which Fintech lenders improve access to credit for credit invisibles.Furthermore,this dissertation provides new insights into the positive externalities of Fintech lending in enhancing financial inclusion via promoting the establishment of credit systems for credit invisibles.(3)Previous literature has constructed Fintech indices at the macro or regional level and documented extensive impacts of Fintech industry development on households,enterprises,commercial banks,and the financial system.To do so,it mostly focuses on non-bank Fintech platforms and tends to regard commercial banks as disrupted objects.However,these studies ignore the active Fintech efforts made by commercial banks in recent years.There has been relatively little research devoted to measuring the development level of Fintech at the bank level and investigating its empowering effects on commercial banks.This dissertation employs various natural language processing techniques to develop a bank-specific,time-varying,multi-dimensional Fintech index and investigates the empowering effects with respect to bank performance and financial inclusion.The construction of the commercial bank Fintech index not only provides a based dataset for subsequent research on commercial banks’ Fintech but also offers methodology guidance for other text analysis-based research.(4)Based on the reality of the widespread application of digital technologies such as big data and machine learning,this dissertation also explores the decision-making behaviors and strategic responses of borrowers and lenders in the retail lending market.These findings not only provide an in-depth understanding of the micro-behavior of participants in Fintech lending but also offer important guidance for the decision-making practices of Fintech lending institutions.In the research on the default identification ability of mobile device information,the results suggest that borrowers have the motivation to manipulate non-traditional data in order to get good credit offers,and lending institutions will use such information effectively through dynamic learning from multiple lending transactions.In the research on how Fintech improves the access to credit for “credit invisible” consumers,the findings indicate that traditional lending institutions will also timely use the credit records of “credit invisible” consumers produced by Fintech lending institutions to update their credit decisions.