资源描述
NBERWORKINGPAPERSERIES FINTECHBORROWERS: LAXSCREENINGORCREAMSKIMMING? MarcoDiMaggio VincentYao WorkingPaper28021 nber/papers/w28021 NATIONALBUREAUOFECONOMICRESEARCH 1050MassachusettsAvenue Cambridge,MA02138 October2020 Theviewsexpressedhereinarethoseoftheauthorsanddonotnecessarilyreflecttheviewsofthe NationalBureauofEconomicResearch. NBERworkingpapersarecirculatedfordiscussionandcommentpurposes.Theyhavenotbeen peerreviewedorbeensubjecttothereviewbytheNBERBoardofDirectorsthataccompanies official NBERpublications. 2020byMarcoDiMaggioandVincentYao.Allrightsreserved.Shortsectionsoftext,notto exceed two paragraphs, may be quoted without explicit permission provided that full credit, includingnotice, isgiventothesourceFintechBorrowers:LaxScreeningorCreamSkimming? MarcoDiMaggioandVincentYao NBERWorkingPaperNo.28021 October2020 JELNo.G21,G23,G4 ABSTRACT We study the personal credit market using unique individuallevel data covering fintech and traditional lenders.Weshowthatfintechlendersacquiremarketsharebyfirstlendingtohigher riskborrowers andthentosaferborrowers,andmainlyrelyonhardinformationtomakecredit decisions. Fintech borrowers are significantly more likely to default than neighbor individuals withthesamecharacteristics borrowingfromtraditionalfinancialinstitutions.Furthermore,they tendtoexperienceonlyashortlivedreductioninthecostofcredit,becausetheirindebtedness increasesmorethannonfintechborrowers afewmonthsafterloanorigination.However,fintech lenderspricingstrategiesarelikelytotakethis intoaccount. MarcoDiMaggio HarvardBusinessSchool BakerLibrary265 SoldiersField Boston,MA02163 andNBER mdimaggiohbs.edu VincentYao GeorgiaStateUniversity J. MackRobinsonCollegeofBusiness 35BroadStreetNW Atlanta,GA30303 wyao2gsu.edu1. Introduction Financial markets have recently witnessed a disruptive force: the rise of online intermediaries and, more generally, ntech companies, i.e., rms that apply technology to improve nancial activities. Fintech companies have targeted the consumer credit market, which is one of the largest credit markets, with outstanding credit of $3.8 trillion in 2018 (FED, 2018) and their market share has been predicted to increase to 20% by 2020 (Transunion, 2017). Therefore, it is important to understand how these new intermediaries aect households borrowing and consumption decisions. Given their increasing popularity, there are natural questions to ask: who borrows from ntech lenders? Do ntech lenders serve individuals underserved by the traditional banking system or are they able to attract the most credit-worthy borrowers? Do these loans help borrowers build a better credit history? Some observers argue that ntech lenders might be able to operate where the banks do not nd it protable. 1 This might be because they face signicantly lower xed costs, e.g., they do not have branches, or because they are less strictly regulated, which might allow them to adopt laxer lending standards. 2 Thus, the entry by ntech lenders might alleviate credit frictions, such as credit rationing due to information asymmetries (Stiglitz and Weiss, 1981) or imperfect competition (Ausubel, 1991; Parlour and Rajan, 2001). This might result in access to credit for nancially constrained households or lower nancing costs for those who switch from traditional institutions to new online lenders. On the contrary, the use of dierent data and tools might enable ntech lenders to capture the most creditworthy borrowers, which might result in lowering the average quality of the pool of households borrowing from banks. Thus, how the market equilibrium looks remains an empirical question. Ideally, to investigate these issues, one would need individual-level data on borrowers 1 For instance, Jamie Dimon told investors in 2014 that: There are hundreds of startups with a lot of brains and money working on various alternatives to traditional banking. The ones you read about most are in the lending business, whereby the rms can lend to individuals and small businesses very quickly and these entities believe eectively by using Big Data to enhance credit underwriting. They are very good at reducing the pain points in that they can make loans in minutes, which might take banks weeks. (JP Morgan Chase annual report, 2014) 2 Fintech lenders are generally regulated by the Consumer Financial Protection Bureau and state regulators, rather than by the Federal Reserve or the Oce of Comptroller of Currency (OCC). 2characteristics, including information about their liabilities, recorded not only at the time of the loan application but over time; furthermore, it would be critical to have a benchmark to assess ntech borrowers performance, e.g., similar individuals borrowing from other insti- tutions. This paper investigates these issues using novel and unique panel data from one of the three main credit bureaus in the country, which allows us to overcome these challenges. The key novelties of the data are the ability to distinguish between traditional and ntech lenders; information about the terms of the loans, and the richness of the data which include information about all borrowers liabilities, as well as some demographic information about the borrowers. In contrast to existing studies on Fintech lenders, we are able to include in our analysis multiple lenders, rather than focusing, for instance, on Lending Club, and our data is a monthly borrower-level panel rather than a cross section of loan applications. Furthermore, in contrast to previous studies, we observe a natural benchmark: individuals borrowing at the same time from traditional lending institutions. While this data covers multiple types of loans, we focus on unsecured personal loans for two key reasons. First, personal credit is one of the fastest-growing segments of the consumer credit market, and it has been the subject of particular interest from ntech lenders. Second, these loans are unsecured loans, which make them more easily comparable across lenders, because the contract is standard and the only terms are the loan size, the maturity and the interest rate (each of which we observe). Since very little is known about the market for consumer credit, we start by investigating its key features and how this market has evolved in the last several years. One key question is how ntech lenders successfully increased their footprint while also facing the competition of signicantly more established nancial institutions. There are two potentially successful market penetration strategies. New lenders might target credit constraints borrowers, which has the advantage of providing them with more data to improve their credit models and time to increase their brand recognition to attract better borrowers, but it is likely to come at the cost of higher defaults. In contrast, new lenders might try to focus on the most creditworthy borrowers, winning them over by oering them better terms than those oered by traditional 3banks, this is likely to result in slower volume growth and lower protability but it can prove to the market the better technology in identifying higher-quality borrowers. We test this hypothesis by dierentiating between lenders based on how much time they have been operating in a specic market, dened at the state level. We nd that ntech lenders tend to start with less creditworthy individuals and then increase their market share by extending credit to better borrowers. 3 Most of the ntech borrowers have credit scores in the mid-range between 640 and 720. They tend to have a higher number of accounts and exhibit a higher credit utilization ratio, which suggests that they already have plenty of access to credit, and that one of the potential reasons to apply for a ntech loan is to consolidate higher-rate credit card debts. In all specications, to absorb any time-varying credit demand shock at the local level, such as changes in house prices or in employment opportunities, or heterogeneous diusion of these new lenders, we control for zip code by month xed eects. We can further exploit the granularity of the data to explore the main loan features. In particular, we test whether the loan features oered by ntech and traditional institutions dier signicantly. We nd that ntech lenders charge on average higher rates, about 3% higher, to lower score borrowers. 4 We investigate whether this pricing dierence depends on the ntech lenders market share and nd that the dierence in pricing is lower for regions where ntech lenders originate less than 20% of loans, i.e., ntech lenders are more cautious in charging higher rates in areas where they still have room to grow. Instead, we nd that ntech lenders oer a better deal than non-ntech to higher score individuals with a 1.5% lower rate. Finally, if data processing is one of the key dierences between ntech and traditional lenders, one might think that the rates would dier depending on how much information the credit report contains. We nd that the rate dierential is the largest for borrowers with thin les, as ntech charge a premium of more than 5% compared to non ntech lenders. This evidence suggests that the way hard information contained in the credit report is 3 Alternatively, this result can be demand-driven as at the beginning ntech lenders might only attract bor- rowers that are not able to obtain credit from other lending institutions. However, once their reputation is established, ntech lenders might be able to attract higher quality borrowers 4 The rate does not include potential additional fees which we do not observe. 4used might be a key factor in explaining the dierences in lenders pricing decisions. However, ntech lenders advertise the use of alternative data not present on the credit report, e.g., rent payment history, utility bills or education, to provide a more accurate assessment of a consumers nancial behavior. We shed some light on this issue by regressing the interest rate on borrowers characteristics, allowing for both linear and non-linear eects, and con- trolling for time-varying shocks by including zip-code by month xed eects. The R 2 from these regressions measures how much of the variation in interest rates is explained by the observable borrower characteristics across lender type. Somewhat surprisingly, we nd that the information in the credit report is able to explain most of the variation in interest rates for ntech lenders, while this is not the case for traditional lenders. To our knowledge, this is the rst study showing that ntech lenders, much more than traditional lenders, focus their credit decisions on the most salient hard-information gures found on the credit report, suggesting a soft information deciency for these new lenders. We then turn to our main results and examine whether ntech loans exhibit dierent per- formance than loans granted by traditional institutions in the 15 months following origination. We nd that ntech loans are signicantly more likely to be in default, even when we include a full set of borrowers credit characteristics, as well as loan features and zip code by month xed eects. In other words, these results are not driven by time-varying local heterogeneity as we are comparing similar borrowers getting loans from ntech and non-ntech in the same month and zip code with similar terms. The results are also economically meaningful, in fact, we nd that the ntech loans exhibit a 1.1% higher default probability, which is large compared to the sample mean of 1.4%. In addition, the relative underperformance persists for our entire time window starting in month ve after origination. One potential concern with this analysis is that borrower heterogeneity between those that have a ntech loan and those who do not might be aecting our ndings. We address it in multiple ways. First, we present our results for three more restricted samples for which we match ntech borrowers to non-ntech ones using a propensity score matching, a manual cri- teria, and an entropy balance methodology. We match on a full set of demographic and credit 5characteristics pre-origination that are likely to aect the loan performance. All specications conrm our main ndings. Furthermore, we exploit the panel nature of the data and show that the results are robust to the inclusion of borrower xed eects, which allows us to capture time-invariant observable and unobservable characteristics of the borrower. We also provide results highlighting how this relative underperformance diers based on borrowers character- istics. We nd that ntech borrowers default more than matched non-ntech borrowers when they have a low credit score and a thin credit le, suggesting that rather than being able to identify the invisible prime borrowers, the identied individuals perform worse than those borrowing from traditional institutions. We show that these results are not driven by one specic period of time in our sample, as they hold even when we distinguish between dierent cohorts of borrowers. Next, we analyze what the reason for this higher delinquency probability might be. One possibility is that the ntech borrowers are using the additional funds not to consolidate their debts, but rather to support additional expenditures. We nd several pieces of evidence con- rming this result. Specically, both their total indebtedness and revolving balance increase more than their matched individuals borrowing from traditional institutions after loan origi- nation. Also, we nd that ntech borrowers are more likely to purchase a car in the rst few months after origination. This makes them overextended and more likely to default. In fact, we also nd that the delinquency rate on any type of account for ntech borrowers is higher. This type of behavior results in their credit score rst increasing, due to the initial partial debt consolidation, but then falling steadily after the rst quarter post loan origination. We also explore the heterogeneity of this result and nd that borrowers with low credit score, high interest rate and a thin le are the subgroups where the results are the strongest. The evidence on the soft information deciency and this heterogeneity based on the length of the credit report further suggests that ntech lenders are exposed to adverse selection. For instance, ntech borrowers might be weaker in their nancial management skills which would translate into higher default rates. However, by predominantly relying on hard information, ntech lenders are likely to miss this source of risk and so end up giving credit to borrowers 6that would have been rejected by a traditional lender. At this point we can ask whether ntech lenders are likely to take this dierent behavior into account, e.g., by charging a higher interest rates, which could compensate for the higher default probability. We investigate this issue by examining whether ntech and non-ntech lenders interest rates are predictors of ex-post loan performance. Specically, we model de- faults as a function
展开阅读全文