16:07 19/3/2019 RFS-OP-REVF180139.tex Page: 1900 19001938 Peer-to-Peer Lenders Versus Banks: Substitutes or Complements? Huan Tang HEC Paris This paper studies whether, in the consumer credit market, peer-to-peer (P2P) lending platforms serve as substitutes for banks or instead as complements. I develop a conceptual framework and derive testable predictions to distinguish between these two possibilities. Using a regulatory change as an exogenous shock to bank credit supply, I find that P2P lendingisasubstituteforbanklendingintermsofservinginfra-marginalbankborrowersyet complements bank lending with respect to small loans. These results indicate that the credit expansion resulting from P2P lending likely occurs only among borrowers who already have access to bank credit. (JEL D14, E51, G21, G23) Received June 1, 2017; editorial decision September 15, 2018 by Editor Andrew Karolyi. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online. Peer-to-peer(P2P)lendingplatforms,whichemergedaroundthe2008financial crisis, allow individuals and small businesses to borrow without the presence of traditional financial institutions. Following several years of exponential growth, this sector is now a significant supplier of credit to consumers. According to TransUnion, a U.S. credit reporting agency, technology-based lenders have become the biggest lender type and account for 30% of the unsecured installment loan sector in 2016. 1 YetlittleisknownabouttheborrowersegmentsthatP2Pplatformsserve.Do suchplatformsmeetthecreditdemandthatwouldotherwisebeunmetbybanks? Or do they compete with banks for the same clientele? This issue is important for assessing the implications of P2P lending expansion. If P2P platforms are complements to banks, they can make finance more inclusive by expanding I am grateful to Denis Gromb and Johan Hombert for their support and guidance. For valuable suggestions, I thank Itay Goldstein, Wei Jiang, and Andrew Karolyi (the editors), Tetyana Balyuk, Gerard Hoberg, Peng Liu, HongRu,GuillaumeVuillemey,XiaoyanZhang,andtheanonymousrefereesmentsfromparticipantsatthe RFS FinTech Workshops (Columbia, 2017; Cornell Tech, 2018), the HEC Brownbag seminar, the Tsinghua PBC SchoolofFinance,theACPRRegulationandSystemicRiskconference,CICF2018,SIF2018,andEFA2018are gratefully acknowledged. This research received financial support from the French National Research Agency (lAgence Nationale de la Recherche) through Project F-STAR (ANR-17-CE26-0007-01). Supplementary data can be found on The Review of Financial Studies Web site. Send correspondence to Huan Tang, HEC Paris, 1 rue de la libration, 78350 Jouy-en-Josas, France. Telephone: +33 1 39 67 96 36. E-mail: huan.tanghec.edu. 1 For details, see increasing-portfolio-risk-return-performance/. The Author(s) 2019. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For permissions, please e-mail: . doi:10.1093/rfs/hhy137 Downloaded from by Renmin University user on 05 December 2019 16:07 19/3/2019 RFS-OP-REVF180139.tex Page: 1901 19001938 Peer-to-Peer Lenders Versus Banks: Substitutes or Complements? credit access for borrowers who are underserved by the banking system. If instead they compete directly with banks, the credit expansion brought about by P2P platforms is probably limited to borrowers who have ready access to bank credit. This paper investigates whether P2P platforms and banks are substitutes or complements in the consumer credit market. The empirical challenge is that the econometrician does not observe whether P2P borrowers have access to equivalent bank lending. For instance, borrowers may voluntarily choose P2P credit over bank credit, or they may be forced to seek P2P loans after being denied access to bank loans. It is therefore not sufficientsimply by looking at the characteristics of P2P borrowersto determine whether most P2P borrowers are underserved by banks or instead are infra-marginal bank borrowers. To address this problem, I develop a conceptual framework in which P2P platforms may operate as substitutes for banks or instead as complements. I derive testable predictions about the effect of a negative shock to bank credit supply on the distribution of P2P borrower quality. The key identification assumption is that, if banks experience a shock that leads them to reduce credit supply, they tighten lending criteria. Borrowers at the lower end of the bank borrower quality distribution are therefore more likely to lose access to bank credit and hence migrate to P2P platforms. This results in a potential shift in the distribution of P2P borrower quality. The nature of the resulting change in the P2P borrower quality distribution depends on whether banks and P2P platforms are substitutes or complements. Supposefirstthattheyaresubstitutes.Inthatcase,thetwotypesoflendersserve the same clientele before the shock and face the same distribution of borrower quality. Following a negative shock to bank credit supply, low-quality bank borrowers will migrate to P2P platforms. Hence the average P2P borrower quality should drop, and all the quantiles of the distribution of P2P borrower quality should decrease. Now suppose that P2P platforms instead complement banks by serving a low-quality borrower segment that is underserved by banks. In this case, the borrower pool of P2P platforms is of worse quality than that of banks. Following the shock to bank credit supply, borrowers switching from banks to P2P platforms will be of higher quality than existing P2P borrowers. The average P2P borrower quality should then improve, and all the quantiles of the distribution of P2P borrower quality should increase. Totestthesepredictionsempirically,Iexploitaregulatorychangethatcaused banks to tighten their lending criteria. In 2010, the Financial Accounting Standards Board (FASB) implemented a new regulation, FAS 166/167, that required banks to consolidate securitized offbalance sheet assets onto their balance sheets and, starting in the first quarter of 2011 (2011Q1), to include them in their risk-weighted assets. In aggregate, this caused banks to consolidate more than $600 billion of assets in 2011, over 80% of which 1901 Downloaded from by Renmin University user on 05 December 2019 16:07 19/3/2019 RFS-OP-REVF180139.tex Page: 1902 19001938 The Review of Financial Studies / v 32 n 5 2019 were revolving consumer loans. 2 This accounting change had a considerable effect on bank lending through its impact on regulatory capital. Banks with different amounts of offbalance sheet securitized assets that were subject to consolidation were heterogeneously affected. Previous studies show that affected banks reduced their small business lending and mortgage approval rates (Dou 2017; Dou, Ryan, and Xie 2017), while increasing their mortgage sales rates (Dou, Ryan, and Xie 2017) and the average quality of credit card loans (Tian and Zhang 2018). I conjecture that this shock affected local credit markets differently depending on the exposure of local banks to the regulation change. I classify banks as being treated if they held offbalance sheet assets subject to consolidationunderFAS166/167.Treatedmarketsaredefinedascountieswith at least one treated bank; other counties constitute the control group. I then examine the changes in the distribution of P2P borrower quality in treated markets. Thus the effect of the bank credit supply shock on the distribution of P2P borrower quality is identified by the variation in exposure to that shock across local markets. Using asset consolidation data obtained from Call Reports, which are mandated by the Federal Deposit Insurance Corporation (FDIC), I identify a total of 59 banks that consolidated assets under FAS 166/167. Data on P2P lending volume, loan application, and borrower characteristics are obtained fromLendingClub,alargeP2PplatformrepresentingmorethanhalfoftheU.S. market share of P2P-based personal loans. 3 I construct county-level variables using data on 880,346 loan applications and 93,159 loan originations during the period 20092012. The final sample consists of 1,908 treated and 1,025 untreated markets, all defined at the county level. The first step of my analysis is to examine the treatment effect of FAS 166/167 on P2P loan application and origination volumes. I find that, relative to the control group, treated markets experienced a disproportionate increase in P2P loan applications after 2011. Per 1,000 inhabitants in treated counties, an average of 0.07 more applications were made for an additional $1,108; this figure represents a 25% increase in the number of applications and a 39% increase in their respective dollar amounts. These results suggest that some borrowers who would otherwise have been served by banks turned to P2P platforms.Moreover,itappearsthatthisadditionaldemandwasatleastpartially satisfiedbyP2Pplatforms.Ifindthattreatedmarketsalsosawanincrease(over thecontrolgroup)inthetotalnumberanddollaramountofP2Ploans.Againper 1,000 county inhabitants, an average of 0.016 more P2P loans were originated 2 See the Board of Governors of the Federal Reserve Systems “Notes on Data” (released on April 9, 2010) of the data set “Assets and Liabilities of Commercial Banks in the United States H.8,” available at federalreserve.gov/releases/h8/h8notes.htm. 3 Source: 1902 Downloaded from by Renmin University user on 05 December 2019 16:07 19/3/2019 RFS-OP-REVF180139.tex Page: 1903 19001938 Peer-to-Peer Lenders Versus Banks: Substitutes or Complements? for an additional amount of $301. In comparison with the pre-shock level of originations, the number (amount) of loans increased by a factor of 1.1 (1.5). Next, I test predictions concerning the shocks effect on the distribution of P2P borrower quality. Toward that end, FICO scores are used as a measure of borrower quality. I find that, compared with the control group, treated markets experienced a decrease in all the quantiles of the distribution. On average, the ten deciles decreased by 4 points, that is, 2% of the difference between the maximum and the minimum of the FICO scores of LendingClub borrowers. Most prominently, the top four deciles decreased by an average of 8 points. According to this result, P2P platforms act as substitutes for banks. Tests exploiting the frequency distribution of P2P borrower quality lend additional support to this interpretation. If bank borrowers migrating to P2P platforms are of worse quality than existing P2P borrowers, then a higher frequency should be observed only in the low-FICO range of the P2P borrower quality distribution. This is indeed what I find. The number of originations increased by 1.9 times among borrowers with FICO scores below 710, or the 45th percentile of the pre-shock FICO scores of LendingClub borrowers; in contrast, there was no significant change in the number of originations at the upper end of the distribution. TheFICOscoreisafairlycoarsemeasureofborrowerquality,sobanksoften use additional information to assess default risk. For example, data collected under the Home Mortgage Disclosure Act (HMDA) show that the primary reasons for mortgage application denials include credit history, debt-to-income ratio, and length of employment. I therefore construct another measure of borrower quality that combines FICO score, debt-to-income ratio, and length of employment. More specifically, I estimate an ordered probit model in which the dependent variable is the application outcome (rejected, qualified but not funded, or qualified and funded) and the explanatory variables are those three borrower characteristics. The model estimates allow me to construct a single- dimension cardinal measure: the predicted borrower quality, normalized to be between 0 and 1. Using this new measure in the quantile and frequency tests delivers similar results as before. First, the predicted borrower quality distribution in treated markets experienced a decline, with respect to the control group, in all the quantiles. In particular, the average reduction in the ten deciles is 0.02, or 2% of the difference between the maximum and the minimum of the predicted borrower quality. Second, frequency tests show that the increase in the number of originations was driven by borrowers whose predicted quality is below the 20th percentile. The overall number of originations among those low-quality borrowers increased by a factor of 1.3. Once again, the results are consistent with the hypothesis that P2P platforms operate as substitutes for banks. Overall, these findings indicate that P2P platforms are substitutes for banks in that they serve the same borrower population. However, the technological advantageofP2Pplatformsmayallowthemtooperateascomplementstobanks 1903 Downloaded from by Renmin University user on 05 December 2019 16:07 19/3/2019 RFS-OP-REVF180139.tex Page: 1904 19001938 The Review of Financial Studies / v 32 n 5 2019 intermsofloansize.Thatis,inlightoftheirlowfixedcostoforiginatingloans, P2P platforms may be able to focus on providing smaller loans than banks, and therefore operate as complements in this segment of the market. To test this hypothesis, I repeat the analysis for the distribution of P2P loan size. I find that borrowers migrating from banks to P2P platforms applied for larger loans than did preexisting P2P borrowers. Relative to the control group, the average P2P loan size in treated markets increased by 9.6%, or $1,066, and almost all the quantiles of the loan size distribution exhibited an increase. In particular, the top two quantiles, the 85th and 95th, increased significantly by $1,563 and $3,870, respectively. Moreover, frequency tests show that the number of originations quadrupled for loans larger than the 80th percentile of the pre-shock loan size, whereas the number of smaller loans did not change significantly. This outcome is consistent with P2P platforms operating as complements to banks in the small loan market. Taken together, the evidence suggests that infra-marginal bank borrowers and small borrowers are the most likely to benefit from the expansion of P2P lending. On the one hand, peer-to-peer credit appears to be fungible with bank credit for borrowers who could have been served by banks, and on the other hand, the loans offered by P2P platforms tend to be relatively smaller. MuchoftheemergingliteratureonP2Plendingfocusesoninvestorbehavior in relation to borrower characteristics such as appearance, disclosures, and social networks (Herzenstein, Sonenshein, and Dholakia 2011; Duarte, Siegel, and Young 2012; Michels 2012; Lin, Prabhala, and Viswanathan 2013; FreedmanandJin2017).Severalstudiesdocumentherdingbyinvestors(Zhang and Liu 2012; Chuprinin and Hu 2016; Kim and Viswanathan 2016) or the adverse incentives of investors whe