无抵押金融科技贷款对抵押贷款市场的溢出效应（英）.docx

Home Is Where Your FinTech Loan Is∗

Tamanna Singh Dubey†
First Draft: October, 2024 This Draft: March, 2025

Abstract
I study the spillover effect of unsecured FinTech lending on mortgage markets. Using quasi-exogenous variation in LendingClub loan activity due to its partnership with the BancAlliance consortium of community banks in February 2015, I present causal evidence that increase in unsecured personal loans by LendingClub resulted in a significant increase in overall mortgage activity in an area. This spillover is more pronounced for new home purchase activity versus mortgage refinancings, and for borrowers who face larger information frictions in the mortgage market. Despite the increase in mortgage lending, I show that mortgage default rates did not increase. Overall, my findings uncover a new benefit of FinTech lending and provide important inputs to ongoing policy debates in the unsecured FinTech lending market.

Keywords: FinTech lending, Mortgage lending, Information frictions
JEL Classification: G20, G21, G23

∗I am especially grateful to my dissertation chair, Amiyatosh Purnanandam, for guidance and support. I
am also grateful to my committee members, Edward Kim, Jeremy Kress, Uday Rajan, and Melvin Stephens. I thank Snehal Banerjee, Sugato Bhattacharya, Pulak Ghosh, Ankit Kalda, George Korniotis, M P Narayanan, Paolo Pasquariello, Srinivasaraghavan Sriram, Boris Vallee, and participants at the Federal Reserve Board of Governors, Indian School of Business, Texas A&M University, University of Hong Kong, University of Miami, University of Michigan, and Warwick University for their helpful comments on the paper. I thank the Mitsui Life Financial Research Center for providing financial support for this research.
†Ross School of Business, University of Michigan; email: [e-mail].

One of the primary objectives of financial intermediaries is to reduce information frictions in credit markets. Traditionally, they have achieved this by using a variety of screening devices such as requiring borrowers to post collateral, or by offering them a menu of contracts. While these devices allow lenders to distinguish between borrowers who are creditworthy and those who are not, some still face financing frictions and are denied credit (Gorton and Winton, 2003). Mechanisms and institutional arrangements that alleviate these frictions have a first order impact on the efficiency of credit markets, as well as on the welfare of borrowers. Can recent advances in lending technologies, such as the significant rise of FinTech lenders, improve subsequent outcomes in credit markets for these borrowers?
To address this larger question, I examine the impact of unsecured FinTech lending on borrowers’ access to the U.S. mortgage market. While several studies have explored the impact of FinTech lending within its own market, my paper is among the first to examine its spillover effects on the mortgage market in a causal manner. On one hand, FinTech lenders can alleviate credit market frictions by using superior screening technology and alternative data, thereby improving credit access for new and existing borrowers through cheaper debt consolidation and the provision of new loans (Berg, Burg, Gombovi´c, and Puri, 2020; Di Maggio, Ratnadiwakara, and Carmichael, 2022). On the other hand, if FinTech lenders are merely “skimming” the best borrowers from the existing pool, or if borrowers use FinTech loans for conspicuous consumption, their presence might not lead to any substantial improvement in borrower outcomes (Di Maggio and Yao, 2021). Therefore, my study sheds light on an important and unresolved empirical question.
In addition, my study is important for two other reasons. First, unsecured FinTech lending has grown tenfold over the past decade, while the mortgage market represents a substantial share of household debt in the U.S., making this an economically significant setting.1,2 Second,
1Source: Transunion
2Iacoviello (2011) find that two-thirds of the wealth of a median U.S. household consists of housing wealth. For the bottom 90% wealth group, Smith, Zidar, and Zwick (2021) show that 62% of non-pension wealth is composed of housing.

39

as FinTech loans are typically smaller than mortgages, my study provides insight into whether small, cash-flow-based lending can help facilitate access to larger, collateral-based loans. This can be particularly relevant when collateral alone is insufficient to address information frictions between lenders and borrowers (Besanko and Thakor, 1987; Benmelech, 2024). Thus, improvements in traditional credit metrics, such as repayment history, debt-to-income ratio, and credit scores resulting from cash-flow-based lending, could further enhance borrowers’ access to collateral-based loans.
I empirically examine the effect of unsecured FinTech loan originations, measured using LendingClub data, on the outcomes in the mortgage market, measured using HMDA data, during 2010 to 2019. I combine these two datasets and aggregate them at 3-digit zip code level, or zip3 level, to analyze how the growth of unsecured FinTech lending influences mortgage market outcomes across over 900 zip3 areas in the US. Since borrowers often receive preliminary indications from lenders about their likelihood of obtaining a mortgage before they formally apply, I use both the number of mortgage applications and the number of loans originated as measures of mortgage activity in each area. To control for any time-invariant local characteristics, I employ zip3 area fixed effects. Additionally, I include the interaction of state and year fixed effects to account for any time-varying changes in state-level factors, such as economic growth, fiscal policy, regulation, and demographics. Furthermore, by regressing mortgage activity on lagged values of LendingClub loan activity, I address potential concerns about reverse causality between the two markets. I find a positive correlation between LendingClub loan activity and overall mortgage activity in an area in this baseline specification.
Is this relationship causal? Unobserved time-varying factors, such as economic growth and job opportunity, may simultaneously drive both FinTech lending and mortgage activity within an area. To address this endogeneity concern, I employ a novel identification strategy based on the February 2015 partnership between LendingClub and the BancAlliance network, one of the largest consortium of community banks in the US. A key aspect of this partnership

was that BancAlliance member banks would redirect their unsecured personal loan customers to the LendingClub platform, enabling LendingClub to significantly expand its borrower base in areas where these banks operated. Since the partnership with the BancAlliance network was established at the national-level, LendingClub could not selectively target specific areas to expand its borrower base. This makes the increase in LendingClub loan activity exogenous to mortgage outcomes and growth at the zip3 level. Furthermore, this arrangement also prevented LendingClub from selling mortgage or other products to bank customers.3 BancAlliance member banks, who were primarily small community banks, entered the partnership to invest in securitized LendingClub loans in order to diversify their credit portfolios with unsecured consumer loans rather than expand mortgage lending. This quasi-exogenous expansion of LendingClub loan activity provides a unique opportunity to causally identify the impact of unsecured FinTech lending on mortgage market outcomes in an area.
At the time, BancAlliance had over 200 member banks across 39 states, with member banks being small community banks with assets between $200 million and $10 billion. To appreciate the scale of this network, it is helpful to note that if all its member banks formed a single entity, the BancAlliance network would rank fourth in the country by branch count and seventeenth by asset size.4 Areas where BancAlliance member banks operated during the 2015 partnership with LendingClub are designated as treated areas, while all other areas in the U.S. serve as control areas. I use a standard difference-in-differences specification to compare outcomes in LendingClub and mortgage activity before and after 2015 across treated and control areas. I find that the number of LendingClub loans increased by 9.25% in treated areas compared to control areas after the 2015 BancAlliance partnership, highlighting the significant effect of this partnership on LendingClub loan activity. In my main test using the same difference-in-differences specification, I show that mortgage activity in treated areas increased by nearly 3% relative to control areas after 2015. Moreover, I show that before
3LendingClub And Smaller Banks In Unlikely Partnership, Wall Street Journal, June 2015
4LendingClub’s Newest Deal Fuels Investor Excitement, Wall Street Journal, February 2015

2015, treated and control areas exhibited parallel trends in both mortgage applications and loan originations.
To directly link LendingClub loans to mortgages, I use a two-stage regression specification using instrumented difference-in-differences (DDIV) approach (Duflo, 2001; Hudson, Hull, and Liebersohn, 2017). The first stage is a standard difference-in-differences specification, as described earlier, where I estimate the impact of the BancAlliance partnership on LendingClub loan activity in an area. In the second stage, I assess the sensitivity of mortgage activity to unsecured FinTech loan activity, instrumented by the 2015 BancAlliance partnership from the first stage. I observe two key findings. First, there is a statistically significant positive spillover effect from LendingClub loan activity to mortgage activity within an area. On average, a 1% increase in LendingClub loans led to a 30-40 basis point increase in mortgage applications and loans. Second, this effect is smallest immediately after the increase in LendingClub loans and grows larger over time. This pattern is consistent with the gradual improvement in borrowers’ traditional credit metrics, driven by FinTech lenders’ ability to identify creditworthy borrowers and alleviate information frictions even beyond the unsecured loan market. Additionally, using the same difference-in-differences specification, I find that mortgage interest rates declined by 1.62% in treated areas compared to control areas after 2015. This result, combined with the observed increase in mortgage quantity, suggests an outward shift in the credit supply curve in the mortgage market following the quasi-exogenous increase in LendingClub loan activity, as instrumented through its 2015 partnership with the BancAlliance network.
What is the magnitude of this spillover effect? In the DDIV specification, I find that a one percent increase in LendingClub loans results in approximately a 30-40 basis point increase in mortgage activity in an area. Given that the average number of LendingClub loans is 335, and the average number of mortgage loans is 10,570 in an area in my sample, the DDIV specification suggests that three additional LendingClub loans lead to about 32 extra mortgage loans. This indicates a multiplier effect in the spillover from LendingClub

loan activity to mortgage activity.

There are four possible explanations for this multiplier effect. First, borrowers vetted and approved by the LendingClub platform may use the loan terms they receive as leverage to obtain an unsecured loan from a bank.5 Second, banks may interpret a FinTech platform’s decision to offer a loan as a positive signal, prompting them to extend an unsecured loan to these borrowers (Balyuk, 2023). Third, once a LendingClub borrower secures a mortgage, they may decide to refinance it—sometimes more than once—resulting in multiple loans for the same borrower in the mortgage market data. Finally, the IV estimates here represent the effect of the treatment on the subset of areas that are affected by the instrument, i.e. the compliers. This group may respond more strongly to the treatment than the average area captured by the baseline OLS specification, making the IV estimate larger. Consequently, the large value of elasticity here is in line with the interpretation that because the complier areas have a higher sensitivity to treatment (Jiang, 2017).
What is the economic mechanism behind my findings? FinTech lenders can use alternative data and their superior processing technology to identify creditworthy borrowers in the unsecured loan market, and consequently arm them with cheaper debt consolidation loans that help them improve their traditional credit metrics, which is crucial to facilitate the spillover effect from unsecured loan market to secured mortgage market. I check this mechanism using two tests. In these tests, I examine the varying levels of information asymmetry across both borrower types and mortgage products in the mortgage market. First, different mortgage products also present varying degrees of information barriers. For example, new home purchases typically involve greater information asymmetry for borrowers to overcome, whereas refinancings generally carry lower information frictions due to the borrower’s ability to secure a mortgage in the past. Second, some borrowers, such as middle-income individuals, are more likely to face higher information frictions compared to other borrowers.
5Obtaining loan terms from LendingClub would not only improve the informativeness of borrowers but
also reduce their search costs. This would further aid borrowers to get favourable unsecured loan contracts from traditional lenders (Argyle, Nadauld, and Palmer, 2023).

In the first test, I consider the varying levels of information asymmetry in the mortgage market based on the type of mortgage product. For new home purchases, the information gap between lenders and borrowers should be larger compared to refinancings, where borrowers are already repaying an existing mortgage. Therefore, I compare the sensitivity of new home purchase versus refinancings to LendingClub loan activity in an area. I find that new home purchases are more sensitive to LendingClub loan activity than refinancing activity. Using a triple differences regression specification to compare treated versus control areas before and after the 2015 BancAlliance partnership, I show that, when compared to refinancings, new home purchase applications increased by 3.5% and loan originations increased by 4.6% in treated areas post 2015, versus control areas.
The second test examines the impact of alleviating information asymmetry across different income bands in the mortgage market. I find that, conditional on borrowers applying for a mortgage, middle-income borrowers are the most sensitive to increases in LendingClub loan activity in an area. This result aligns with the info