The Rise of Nonbanks and the Quality of Financial Services: Evidence from Consumer Complaints

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) The Rise of Nonbanks and the Quality of Financial Services: Evidence from Consumer Complaints Ahmet Degerli and Jing Wang 2022-059 Please cite this paper as: Degerli, Ahmet, and Jing Wang (2022). “The Rise of Nonbanks and the Quality of Financial Services: Evidence from Consumer Complaints,” Finance and Economics Discussion Series 2022-059. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2022.059. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Rise of Nonbanks and the Quality of Financial ∗ Services: Evidence from Consumer Complaints Ahmet Degerli Jing Wang Federal Reserve Board University of Missouri ahmet.degerli@frb.gov jingwang@missouri.edu First draft: July 2021 This draft: July 2022 Abstract We show that as nonbanks’ market share increases in a local residential mortgage market, the quality of mortgage services in the market improves. Two instrumental variable analyses exploiting (1) stress tests conducted by the Federal Reserve, and (2) mortgage industry surety bonds required by each state confirm this finding. We find evidence that as nonbanks grow their market share, they develop a specialty in servicinglower-incomeborrowersandincreaseinvestmentintechnology, leadingtoimproved service quality. This improvement in service quality is more salient in counties with a higher percentage of minority populations. Keywords: Product quality, Mortgage lending, Banks, Nonbanks JEL classification: G21, G28, L13, L15 ∗WewouldliketothankJohnHowe, YouSukKim, MartienLubberink, KarenPence, LubomirPetrasek, Philipp Schnabl, and seminar and conference participants at the Federal Reserve Board, University of Missouri, the 34th Australasian Finance and Banking Conference, the 58th Eastern Finance Association annual meeting, and the Mortgage Bankers Association for helpful comments and suggestions. All errors are our own. The views expressed here are those of the authors and not necessarily those of the Federal Reserve Board, or the Federal Reserve System.

1 Introduction Nonbanks, including FinTech lenders, have been growing rapidly since the Global Financial Crisis (GFC). In particular, nonbanks have achieved a substantial presence in the U.S. residential mortgage market. For example, in 2018, they accounted for 47% of the mortgage originations and 35% of the post-origination mortgage servicing.1 This substantial growth of nonbanks raises one important question: Does the expansion of nonbanks in the mortgage market improve or impair consumer welfare? The existing literature examines this question by focusing on the origination stage of mortgages and investigating the effects of nonbanks on consumers’ access to and the cost of mortgages. Specifically, the literature provides evidence that nonbanks increase credit access for riskier and less creditworthy borrowers, whereas the evidence regarding the effects on the cost of mortgages is mixed (e.g., Buchak et al. 2018; Jagtiani et al. 2021). In this paper, we take a fresh look at this question by examining the quality of financial services associated with mortgages, not only at the origination stage but also at the postorigination servicing stage. Our goal is to understand the evolution of service quality in a local residential mortgage market as nonbanks increase their presence in this market. Such evidence complements the findings of prior studies in two important dimensions. First, it extends our understanding of the impact of nonbanks’ growth on consumer welfare beyond the origination stage of mortgages. Second, it sheds light on the interplay of the quantity and quality of financial services accompanying the rise of nonbanks. Such evidence should be especially relevant for policymakers and regulators, considering the ongoing discussions on tightening the regulation of nonbanks. Ex ante, it is not obvious whether and in which direction the service quality of the mortgage industry would change as nonbanks increase their share in the mortgage market. If 1These nonbanks’ market shares are computed based on a sample of mortgage lenders that are included in both the HMDA (Federal Financial Institutions Examination Council (U.S.), Home Mortgage Disclosure Act (Public Data)) and eMBS, Inc., eMBS Data (Mortgage-Backed Securities OnLine). 1

nonbanks’ increased market share is purely a result of regulatory arbitrage, service quality facing consumers in the mortgage market could deteriorate due to the less regulatory oversight of nonbanks compared to that of traditional banks (Fuster et al., 2021).2 However, as their market share grows, if nonbanks learn to better serve their customers through, for example, developing a specialty in servicing certain types of customers or investing in advanced technology, then service quality facing consumers in the mortgage market could improve. Ouranalysisrelatesthequalityoffinancialservicesassociatedwithmortgagesinacounty to the market share held by nonbanks in the county’s residential mortgage market. Admittedly, measuring the quality of financial services is challenging and subjective. The specific angle that we take to measure the service quality in the mortgage market follows recent literaturethatutilizesmortgage-relatedcomplaintsfiledwiththeConsumerFinancialProtection Bureau (CFPB) (e.g., Begley and Purnanandam 2021). These complaints capture an important and unique aspect of financial service quality in the residential mortgage market—the direct feedback from mortgage borrowers about their poor experiences with financial institutions. Typically, these complaints are the results of issues that customers could not obtain a resolution from financial institutions before the CFPB intervenes. Based on these complaints, our measure of service quality is the complaint ratio, which is the number of mortgage-related complaints filed with the CFPB divided by the number of outstanding mortgages in the local market. We follow prior literature (e.g., Buchak et al. 2018) and define nonbanks as lenders other than traditional deposit-taking banks. Our measure of nonbanks’ market share is based on loan-level mortgage origination data from HMDA and is calculated as the dollar amountofmortgagesoriginatedbynonbanksdividedbythetotaldollaramountofmortgages originated in the county. Although the CFPB data include both origination complaints and post-origination servicing complaints, we do not differentiate between nonbanks’ share of mortgage originations and that of post-origination servicing. This empirical choice is 2Using data on Consumer Financial Protection Bureau (CFPB) supervision and enforcement, Fuster et al. (2021) find that tighter regulatory oversight improves servicing practices. 2

appropriate because, despite the fact that comprehensive data on mortgage servicing are not available, nonbanks’ market share of mortgage originations and that of post-origination servicing in the local mortgage market are highly (above 90%) correlated.3 We find that as nonbanks’ market share increases in a county, the mortgage-related complaint ratio in the county decreases. The effect is economically large—a one standard deviation increase in nonbanks’ market share is associated with a 12% reduction in the mortgage-related complaint ratio in the county. Importantly, we obtain this result with county and year fixed effects. County fixed effects absorb any time-invariant county characteristics. For instance, counties may have different complaint ratios because they differ in their residents’ willingness or ability to file complaints. To the extent that the average willingness or ability of a county’s residents to file complaints is stable, county fixed effects absorb this potential heterogeneity. Year fixed effects are also instrumental if we consider the possibilities that consumers become more familiar with the CFPB and therefore file more complaints over time or that financial institutions become better at avoiding having complaints filed against them over time. Therefore, year fixed effects mitigate the concern that our results are due to the time trend of complaint ratios. Of course, nonbanks’ choices to expand business in specific counties are not random. In particular, time-variant county characteristics or local economic dynamics may affect these choices. Therefore, the main identification challenge is that county-specific economic conditions may be correlated with both the market share of nonbanks and the complaint ratio in the county, resulting in biased estimates. We conduct two instrumental variable (IV) analyses to address this identification challenge. In both IV analyses, we focus on the within-countychangeinnonbanks’marketsharesandexamineitseffectonthewithin-county change in complaint ratios. 3Using various samples, we show that the correlation between nonbanks’ market share of mortgage originationsandthatofpost-originationservicingisover90%. Althoughnonbanksmostlyusetheoriginateto-distribute model of lending, they frequently remain the servicing entity after the sale of mortgages to various third parties. Even in situations where nonbanks sell mortgage servicing rights, they typically sell those rights to other nonbanks. Therefore, mortgages that are originated by nonbanks are likely serviced by nonbanks. Please see Section 3.2 for more discussions. 3

The first IV analysis exploits the stress tests implemented by the Federal Reserve. Based on the requirement of the Dodd-Frank Act (DFA), the Federal Reserve conducts annual stress tests to assess if the selected group of large bank holding companies (i.e., BHCs) have sufficient capital to absorb losses if poor economic conditions were to occur. Previous literature documents that regulations, such as capital requirements, imposed on traditional deposit-taking banks are partially responsible for the rise of nonbanks (Buchak et al. 2018; Irani et al. 2021; Chernenko et al. 2022; Begley and Srinivasan 2022). In particular, prior studies show that the increased regulatory burdens due to stress tests lead to a reduction in bank lending, including residential mortgages (Covas 2017; Cort´es et al. 2020). This reduction in mortgage originations by stress-tested banks may facilitate the expansion of nonbanks in the local mortgage market. Therefore, nonbanks’ market share is likely to increase more aggressively in counties with a higher ex ante exposure to stress-tested banks, ceteris paribus. For this IV analysis, we use counties’ ex ante exposure to stress-tested banks as an instrument for the change in nonbanks’ market shares. We find that counties with a higher ex ante exposure to stress-tested banks experience higher growth in nonbanks’ market shares during our sample period, and consequently, these counties experience a larger reduction in their mortgage-related complaint ratio. The second IV analysis exploits the variation in mortgage industry surety bonds required by each state. When mortgage professionals apply for licenses to conduct business in a state, they are required to post surety bonds as part of the licensing application. The amount of surety bond required in a state reflects the costs of conducting mortgage business in the state and the difficulties in entering the mortgage market of the state. Therefore, nonbanks are likely to expand more aggressively in states with a lower surety bond requirement, ceteris paribus. For this IV analysis, we use states’ mortgage industry surety bond requirement as an instrument for the change in nonbanks’ market shares. We find that counties that reside in states with a lower amount of surety bond requirement experience higher growth in nonbanks’ market shares during our sample period, and subsequently, these counties 4

experience a larger reduction in their mortgage-related complaint ratio. It is worth noting that stress tests and surety bonds may have a direct impact on service quality. To meet the exclusion restriction, we carefully design our IV strategies to include two features. First, as we emphasized earlier, we focus on the within-county change in nonbanks’ market shares and examine its effect on the within-county change in complaint ratios. Second, we construct the two instruments using information that is years preceding our sample period and remains constant during our sample period. As a result of these two features, even if the two IVs may directly influence the level of service quality, they are not expected to have a direct influence on the change in service quality during our sample period. In addition, we conduct robustness tests to strengthen our identification assumption. To investigate the reasons for the reduced county-level complaint ratio, we first compare the complaint ratios between nonbanks and traditional banks. We find that the average complaint ratio of nonbanks is much higher than that of traditional banks. Therefore, the cross-sectional difference in the complaint ratios between nonbanks and traditional banks cannotexplainthe reducedcounty-level complaint ratioasnonbanks’marketshare increases. Rather, we find that the complaint ratio of nonbanks decreases as their market share in the county increases, contributing to the overall reduction in the complaint ratio in the county. Specifically, focusing exclusively on mortgage complaints against nonbanks, we find that a one standard deviation increase in nonbanks’ market share is associated with an 18% reduction in the complaint ratio of nonbanks in the county. We confirm this finding using the two IV strategies. We explore two potential mechanisms through which nonbanks may improve their service qualityastheygainagreatermarketshare. First, weinvestigatewhethernonbanksarelikely to develop a specialty in servicing certain types of borrowers as their market share increases. Specialization would allow nonbanks to develop a deeper understanding of their customers’ needsandtobebetterequippedtosolvetheircustomers’issues, resultinginimprovedservice quality. Supportive of this conjecture, we find that nonbanks become increasingly focused 5

on lower-income borrowers as their market share increases. Specifically, as nonbanks’ market share increases in the local mortgage market, their average borrower income becomes lower and the dispersion of their borrowers’ incomes becomes smaller. Previous literature documents that the average borrower income of nonbanks is lower than that of traditional banks (Buchak et al. 2018). Our finding is novel and extends the prior finding by showing that as nonbanks increase their market share, their customers become even more concentrated among lower-income borrowers. Critical to this mechanism is whether nonbanks indeed become better at servicing lower-income borrowers. By examining detailed complaint issues, we find that complaints against nonbanks about issues related to difficulties in making mortgage payments—complaints that are likely filed by lower-income borrowers—are significantly reduced as nonbanks’ market share increases. Taken together, the evidence is consistent with the idea that, as their market share increases, nonbanks develop a specialty in offering better mortgage solutions to lower-income borrowers.4 Second, we investigate whether nonbanks are likely to invest in technology to improve their service quality as their market share increases (Fuster et al. 2019). We test this conjecture by first examining nonbanks’ incentives to invest. Nonbanks may have more incentives to invest if their increased market share is concentrated within a few lenders because these lenders are more likely to internalize the benefits of such investment (Petersen and Rajan 1995; Giannetti and Saidi 2019; Saidi and Streitz 2021). Also, nonbanks may have more incentives to invest if their market shares increase not only in the focal county but also in the entire nation, because the marginal cost of technology investment is decreasing and their increased market shares in the nation would allow them to take advantage of the economies of scale. Using two-stage regression models, we find evidence that the increased market share of nonbanks is positively associated with both sources of investment incentives, resulting in 4Ourinterpretationshouldnotbeseverelybiasedbythelowerfinancialliteracyoflower-incomeborrowers. We have shown that borrowers of nonbanks are not less likely to complain than borrowers of traditional banks, although the average income of nonbanks’ borrowers is lower than that of traditional banks’ borrowers. Also, prior literature shows that there are more mortgage-related complaints in areas with a higher proportion of poor populations (Begley and Purnanandam 2021). 6

a lower complaint ratio for nonbanks in the county. Lending support to this mechanism, we find that as nonbanks’ market share increases in a county, nonbanks’ use of technology (proxiedbynonbanks’demandforemployeeswithtechnology-relatedskills)indeedincreases. Furthermore, we find that complaints against nonbanks about issues related to the payment process—complaints that are likely resolved by upgrading technologies deployed in online payment platforms—are significantly reduced as nonbanks’ market share increases. Collectively, theevidencesuggeststhatnonbanksinvestmoreintechnologytoimprovetheirservice quality as their market share increases. Shiftingawayfromnonbanks,wecontinueourinvestigationintotheevolutionofmortgagerelated complaint ratios by examining traditional banks. We find that the mortgage-related complaint ratio of traditional banks also decreases as nonbanks’ market share increases in the county, contributing to the improved service quality in the local mortgage market. This result is potentially due to lender specialization following the rise of nonbanks. We find that traditional banks’ average borrower income increases as nonbanks’ market share increases in the county. Considering that higher-income borrowers are likely of higher credit quality and are less likely to file complaints (Begley and Purnanandam 2021), the evidence does not necessarily indicate an improvement in traditional banks’ service quality; rather, it may be thecasethatasnonbanks’marketshareincreasesinacounty, traditionalbanksinthecounty become increasingly focused on higher-income borrowers, resulting in a lower complaint ratio for traditional banks. The increased average income of traditional banks’ borrowers and the decreased average income of nonbanks’ borrowers together suggest that nonbanks and traditional banks become increasingly specialized in different types of borrowers as nonbanks grow in size. Importantly, this lender specialization benefits lower-income borrowers without hurting higher-income ones, consistent with an improvement in consumer welfare. Finally, we examine whether the improvement in service quality is more likely to benefit marginalized borrowers, such as minorities. Previous studies show that in the mortgage market, minorities are more likely to face discrimination in both accessing mortgages and 7

receiving high-quality mortgage services (e.g., Munnell et al. 1996; Bartlett et al. 2022; Begley and Purnanandam 2021). We find that the effect of the increased market share held by nonbanks on mortgage-related complaints is stronger for counties with a higher percentageofminoritypopulations. Thisfindinghighlightstheroleofnonbanksinalleviating discrimination in the quality of services received by minorities in the mortgage market. Taken together, our findings speak to the collection of recent studies on the consequences of the rise of nonbanks. This literature shows that the growth of nonbanks is important in substituting for bank lending when credit supply from banks contracts. In particular, the funding provided by nonbanks is critical for small business lending and homeownership by low-income households (e.g., Buchak et al. 2018; Gopal and Schnabl 2022; Gete and Reher 2021). Theliteraturealsoshowsthatthegrowthofnonbanksmayreducetheresilienceofthe credit market during market-wide stress because nonbanks lack access to insured liabilities (e.g., Irani et al. 2021; Drechsler et al. 2019). We complement this strand of literature by showing that the growth of the nonbank industry has a positive effect on the service quality in the mortgage market, leading to an improvement in consumer welfare; this channel of welfare improvement is particularly important for minority consumers. Our findings also provide new evidence on the interplay between the quantity and quality of financial services using the nonbank industry as a laboratory. Closely related to our paper is Begley and Purnanandam (2021). Focusing on the Community Reinvestment Act (CRA), Begley and Purnanandam (2021) show that when banks increase the quantity of mortgage supply due to regulatory requirements, their service quality (proxied by consumer complaints) declines. Their finding suggests that banks trade off the quantity of credit supply for the quality of such services. In contrast, our evidence suggests that when the mortgage supply increases due to market forces, financial institutions are able to develop relevant expertise and upscale investment, resulting in improved service quality. This contrasting evidence together is policy-relevant because, at the time of writing, policymakers 8

are considering overhauling the CRA to encompass nonbanks.5 2 Related Literature Nonbanks, i.e., financial intermediaries that provide banking services but do not rely on deposits as a funding source, have experienced marked growth in the consumer lending market since the recovery from the GFC. The consequences of this expansion have been of great interest to regulators and policymakers. So far, the literature documents both positive and negative effects of a growing nonbank industry. Regarding the positive effects, the literature shows that nonbanks play an important role in supplying credit, especially when credit supply by traditional deposit-taking banks is constrained. For example, Buchak et al. (2018) and Irani et al. (2021) show that, in response to the increased regulatory burdens, such as higher capital requirements, banks are likely to reduce credit supply while nonbanks are likely to fill in the gaps. By examining small business loans, Gopal and Schnabl (2022) show that lending by nonbanks substitutes for the reduction in lending by banks after the GFC. The literature also shows that nonbanks provide credit access for borrowers with lower credit quality (Buchak et al. 2018; Gete and Reher 2021; Jagtiani et al. 2021). Regarding the negative effects, theory suggests that nonbanks may impose significant fragility on the financial system because they lack insured liabilities in their funding structure (e.g., Plantin 2015). Empirically, Irani et al. (2021) show that, during the GFC, loans originated by nonbanks with relatively liquid liabilities were less likely to be rolled over and those loans also experienced greater price volatility. Our study contributes to this strand of literature by exploring a different dimension—the quality of financial services received by mortgage borrowers—to assess the consequences of the nonbank industry’s expansion. To understand the consequences of nonbanks’ presence, a collection of studies examine the extent to which nonbanks improve or impair the overall efficiency of the financial 5For example, please see https://www.wsj.com/articles/powell-highlights-slower-recovery-for-low-wageand-minority-workers-11620065926. 9

system (Philippon 2016). For example, focusing on the role of Fintech lenders in the Paycheck Protection Program (PPP) during the Covid-19 pandemic, the literature documents both positive and negative effects of Fintech lenders on allocating credit. On the one hand, Erel and Liebersohn (2020) show that Fintech complements banks during the pandemic by expanding financial services to ZIP codes where traditional banks do not have a strong presence. On the other hand, Griffin et al. (2021) show that misreporting is higher among FinTech loans than other loans in the PPP program. Furthermore, in the mortgage market, Gete and Reher (2021) show that nonbanks increase ZIP code-level home ownership, suggesting that nonbanks complement traditional banks by serving low credit quality borrowers. Our findings suggest that one channel through which nonbanks could improve the efficiency in the mortgage market is to improve service quality by developing expertise in servicing lower-income borrowers. This evidence is consistent with the market trend of the mortgage servicing industry, which indicates that nonbanks have increasingly focused on acquiring specialty and delinquent servicing.6 Our empirical investigation benefits from the prior literature on the impact of market concentration on banks’ lending behavior. Petersen and Rajan (1995) show that lenders are more likely to invest in building relationships with their borrowers if the credit market is concentrated because it is easier for these lenders to internalize the benefits of lending relationships. Giannetti and Saidi (2019) show that lenders’ credit concentration in an industry helps the industry receive more credit in distress because their lenders internalize negative spillovers when fire sales are likely to ensue (also see Saidi and Streitz 2021). Complementary to this literature, our study suggests that when nonbanks have a concentrated market share in a local residential mortgage market, they have more incentives to make investments to improve their service quality to customers because they are likely to internalize the benefits of future business. Anotherstrandofliteraturethatprovidesimportantguidancetoourresearchagendaand 6Please see https://www.mba.org/2015-press-releases/june/mba-pwc-us-release-the-changing-dynamicsof-the-mortgage-servicing-landscape. 10

the interpretation of results is the literature on scale economies in the financial sector. The existing literature supports the notion that the banking industry significantly benefits from scale economies (Hughes and Mester 2013). The benefits have become even more salient in recent years because of the high costs associated with adopting advanced technology in the banking industry, such as computing, telecommunication, automation, machine learning, and artificial intelligence (Mester 2010). In recent years, the U.S. banking industry has spent over $100 billion annually on technology investment.7 Fuster et al. (2019) show that advanced technology plays an important role in improving the operational efficiency of mortgage lenders. Our study provides evidence consistent with the notion that a large market share allows nonbanks to take advantage of scale economies in funding and utilizing advanced technology, reducing the likelihood of consumer complaints. Our study is closely related to the recent literature that goes beyond the quantitative measure of financial services and focuses on the quality of financial services. Using the CFPB data, Sedunov (2020) shows that consumer satisfaction is higher in counties with more small banks; Begley and Purnanandam (2021) show that mortgage-related complaints are higher inareaswithahigherproportionofpoorandminorityborrowersandinareastargetedbythe CRA. Using fraud conducted by financial advisers, Gurun et al. (2018) show that trust plays a critical role in the investment advisory industry; Egan et al. (2019) show that misconduct among financial advisers is extensive in the United States and unsophisticated consumers are the targets. Focusing on an important trend in the consumer lending market—the rise of nonbanks, we provide evidence that the expansion of nonbanks’ market share has a positive effect on the service quality received by consumers in the mortgage market. Our study is also related to the literature on discrimination facing minorities in the financial markets. This literature shows that minorities are more likely to be denied access to mortgages (Munnell et al. 1996), more likely to receive inferior financial services (Begley and Purnanandam 2021), more likely to experience debt collection judgments (LaVoice 7Please see https://www.wsj.com/articles/technology-is-banks-new-battleground-11568114378. 11

and Vamossy 2019), and less likely to achieve financial restitution after disputing financial services (Haendler and Heimer 2021). Recent studies explore whether nonbank lenders can alleviate discrimination against minorities. By examining interest-rate decisions in the mortgage market, Bartlett et al. (2022) show that Fintech lenders reduce, but do not eliminate, discrimination against Latinx/African-American borrowers, possibly because algorithms and bigdataenableFintechlenderstoextractrents. Byexaminingapprovalratesinthemortgage market, Giacoletti et al. (2021) show that nonbanks are more likely to have equal treatment of White and Black mortgage applicants than traditional deposit-taking banks. Our findings add to this strand of literature by shedding light on the potential of nonbanks to alleviate discrimination in the quality of services facing minorities in the mortgage market. 3 Data and sample 3.1 Data We obtain mortgage-related complaints from the CFPB. After the GFC, the DFA established the CFPB as a platform for consumers to file complaints against financial institutions with regard to a wide range of financial products, such as credit reports, debt collection, student loans, and mortgages. The CFPB has considerable authority to conduct investigations of the complaints. The resulting gathered facts are used to identify violations of federal consumer financial law and may be used in a public enforcement action. Since its inception until 2020, the CFPB has received over 1.5 million consumer complaints and provided over $13 billion in consumer relief. Among the complaints filed with the CFPB, 17% are related to mortgages. The majority ofthesecomplaintsconcernpost-originationmortgageservicing(suchaspaymentprocessing, loan modification, loan collection, and foreclosure) rather than mortgage originations. For most complaints, the CFPB reports the five-digit ZIP code of the filing consumer. For some complaints, the CFPB also includes a brief narrative of the complaint; in these cases, 12

the last two digits of the ZIP code are removed to protect the consumer’s anonymity. We restrict our analysis to only complaints with five-digit ZIP codes reported in the CFPB, to accurately match complaints in a county with the market share of nonbanks in the county. Our analysis, therefore, includes mortgage-related complaints that are associated with a five-digit ZIP code. We aggregate ZIP code-level complaints at the county level. Information on mortgage loans comes from the HMDA dataset, which provides substantial coverage of the U.S. residential mortgage market.8 The HMDA data contain detailed information on loan applications and originations, including the application year, application outcome, loan amount, loan type, loan purpose, the location of the mortgaged property, the income and demographic information of applicants, and lender identity. We classify HMDA lenders into traditional deposit-taking banks and nonbanks by using the Avery file, following Jagtiani et al. (2021). To construct our sample, following Loutskina and Strahan (2009), we keep only conventional home purchase loans originated for owner-occupied 1-to-4 family housings;9 we drop loans with missing information on loan size, applicant income, or location. We also drop loans smaller than $10 thousand or larger than $10 million. We exclude mortgages insured by the Federal Housing Administration (FHA) and mortgages guaranteed by the U.S. Department of Veterans Affairs (VA). This empirical choice is due to two reasons. First, FHA and VA loans are different from other mortgages in that they are issued to lower credit quality borrowers with explicit guarantee/insurance provided by government agencies; these differences per se are likely to lead to different service outcomes for FHA and VA loans compared with those for other mortgages.10 Second, nonbanks are the dominant mortgage originators for these loans during our sample period, reaching 80% by 2016 (Jagtiani et al., 2021); by comparison, in the conventional mortgage loan market, nonbanks originated 45% 8Avery et al. (2007) estimate that the coverage is about 80% in 2006. 9If we include all mortgages (including refinancing loans, loans for houses that are non-owner occupied, and loans for multi-family houses) in the sample, our main results remain robust. 10Confirming this conjecture, we find that although FHA and VA loans comprise 25% of the overall mortgage market in 2016, there are only 12.6% of the complaints filed against FHA or VA loans in 2016 in the CFPB database. 13

of the mortgages in 2016. Therefore, mixing the FHA and VA loan market with the conventional mortgage loan market would complicate the interpretation of the results about mortgage service quality.11 Regarding other data used in our analyses, we obtain county-level income per capita from the Bureau of Economic Analysis (BEA), county-level unemployment rates from the Bureau of Labor Statistics (BLS), and county-level minority populations from the 2010 Census files. 3.2 Sample and main variables Our sample includes 149,291 complaints filed between 2012 and 2018.12 Table 1 shows the distribution of issues associated with these complaints. Besides examining the total number of complaints, we also differentiate between complaints about the post-origination mortgage servicing and complaints about mortgage originations. Complaints about the postorigination mortgage servicing include complaints tagged by the CFPB under the following five categories: “Loan modification, collection, foreclosure,” “Loan servicing, payments, escrow account,” “Settlement process and costs,” “Struggling to pay mortgage,” and “Trouble duringpaymentprocess.”Complaintsaboutmortgageoriginationsincludecomplaintstagged by the CFPB under the following four categories: “Application, originator, mortgage broker,” “Applying for a mortgage or refinancing an existing mortgage,” “Credit decision / Underwriting,” and “Closing on a mortgage.” Four out of the top five issues in the sample are related to post-origination mortgage servicing. Less than 10% of the complaints are related to mortgage originations. For a county to be included in our sample, we require the median number of complaints in the county during our sample period to be at least one; as a result, our sample for the county-level analyses includes 7,178 county-year observations.13 Our measure of service quality is the county-year complaint ratio, which is defined as the 11If we include FHA and VA loans in the sample, our main results remain robust. 12Oursampleperiodstartsin2012becauseitisthefirstyearwithfull-yearcoverageintheCFPBdataset. 13Our results are robust to (1) requiring the median number of complaints in the county during our sample period to be more than one, or (2) including all counties with at least one complaint during the sample period. 14

number of complaints scaled by the number of outstanding mortgages. The scaling variable is the number of outstanding loans in a year rather than the number of loans originated in that year because, as shown in Table 1, many complaints are associated with the servicing of mortgages that may be originated in previous years. We proxy a county’s outstanding mortgages using the number of federal tax filings in the county with mortgage interest payments. For robustness, we alternatively use the total number of mortgages originated in that county during the last five years as a proxy for the county’s outstanding mortgages. Table 2 shows that an average county-year reports 19 complaints out of 29,304 outstandingmortgageloans,resultinginacomplaintratioof0.05%. Weobservealargecross-sectional variation in complaint ratios with an interquartile range from 0.02% to 0.07%. Nonbanks’ market share for a county-year is calculated as the dollar amount of mortgages originated in the county in year t-1 by nonbanks over the dollar amount of mortgages originated in the county in year t-1 by all lenders. For robustness, we compute an alternative measure of nonbanks’ market share using the number of mortgages rather than the dollar amount of mortgages. Figure 1 shows that there is a secular increase in the market share of nonbanks in the mortgage market after the GFC. The average market share of nonbanks in our sample is 32% with an interquartile range from 20% to 43%. It is important to note that our measure of nonbanks’ market share is based on information about mortgage originations. In interpreting our results, we use nonbanks’ market share of mortgage originations as a proxy for nonbanks’ overall market share of the mortgage industry, including both originations and post-origination servicing. This empirical choice is due to two considerations. First, comprehensive data on mortgage services are not available, making it impossible to conduct empirical analysis using granular observations. Second and more importantly, nonbanks’ market share of mortgage originations and that of post-origination servicing in the local mortgage market are highly correlated, making the use of origination information appropriate in measuring nonbanks’ overall presence in the mortgage market. For example, using a sample of mortgages sold to Freddie Mac and still 15

outstanding during our sample period (i.e., 2012–2018), the top left panel of Figure 2 shows the relationship between nonbanks’ nationwide market share of mortgage originations and that of post-origination servicing. The correlation between the two series is 0.99.14,15 The remaining panels of Figure 2 provide bin scatter plots, which show the relationship between nonbanks’ market share of mortgage originations and that of post-origination servicing at various geographical levels. Specifically, we group nonbanks’ market shares of mortgage originations into 10 bins for the state-level panel and 100 bins for the Metropolitan statistical areas (MSA) level and ZIP-code level panels; we plot the average nonbank market share of mortgage originations and that of post-origination servicing for each bin. Across all three panels,weseeaclosealignmentofnonbanks’marketsharesofthesetwobusinesses. Although nonbanks mostly use the originate-to-distribute model of lending, they remain the servicing entity for a sizable portion of the loans they sell. Within this sample of loans sold to Freddie Mac, the servicing rights of 32% of the loans remain held by the originating nonbanks; for the loans that the servicing rights are not held by the originating nonbanks, 57% are sold to another nonbank for servicing. These statistics imply that, in this sample, 71% of the loans that are originated by nonbanks are still serviced by nonbanks.16 Furthermore, during our sample period, traditional banks were limited in their ability to purchase mortgage servicing rights from nonbanks due to the Basel III capital requirements (LuxandGreene2015). Therefore, inoursample, mortgagesthatareoriginatedbynonbanks are likely serviced by nonbanks. Because the interpretation of our findings is based on the marketshareofallnonbanksinacounty,ratherthanthemarketshareofindividualnonbanks in the county, the transfer of mortgage servicing rights between nonbanks should not impact our interpretation. 14This sample is from Freddie Mac Single Family Loan-Level Dataset. The dataset reports the sellers of mortgages and we assume these sellers as the originators of the mortgages. 15We also use a sample of mortgage lenders that are included in both the HMDA and eMBS datasets between 2016 and 2018, and we find similar results; nationwide, the correlation between nonbanks’ market share of mortgage originations and that of post-origination servicing is 0.99. We thank You Suk Kim for providing this matched sample. 1632%+68%×0.57=71% 16

4 Nonbanks’ market share and mortgage service quality Thegoalofourempiricalanalysisistounderstandtheeffectsofnonbanks’marketsharein a county on the quality of financial services associated with mortgages in the county. In this section, weprovideevidencethatnonbanks’marketshareinacountyisnegativelyassociated with the mortgage-related complaint ratio in the county. We conduct two instrumental variable analyses to support a causal link between the increase in nonbanks’ market share and the decrease in the mortgage-related complaint ratio. 4.1 Ordinary least squares (OLS) regression We start our analysis by estimating the following regression model as the baseline specification: Complaintratio = βNonbankshare +γControls +δ +η +(cid:15) , (1) c,t c,t−1 c,t−1 c t c,t where the dependent variable is the county-level complaint ratio, which is the total number of complaints in county c in year t scaled by the number of outstanding mortgage loans in the county in that year. Nonbankshare is the market share of nonbanks in the county’s c,t−1 residential mortgage market in year t-1. Controls includes county income per capita and c,t−1 unemployment rates to absorb time-varying differences across counties. δ indicates county c fixed effects and controls for the unobserved time-invariant differences across counties. η t indicates year fixed effects and controls for the time trend of complaint ratios and macrolevel shocks that may affect the complaint ratio. Standard errors are clustered at the county level. Table 3 presents the baseline estimation results. Column (1) shows that as the market share of nonbanks in a county’s mortgage market increases, the complaint ratio decreases. 17

The economic meaning of the coefficient is that, following a one standard deviation increase in the market share of nonbanks, the complaint ratio decreases by 0.006 percentage points, which is a 12% reduction from the average county-level complaint ratio.17 In columns (2) and (3), we classify complaints into complaints about post-origination mortgage servicing and those about mortgage originations, respectively. The results show that nonbanks’ market share is negatively associated with only complaints about post-origination mortgage servicing, not those about mortgage originations. This evidence suggests that our main finding is due to the reduction in complaints about the post-origination mortgage servicing rather than mortgage originations. This result may not be surprising since the majority of the mortgage-related complaints filed with the CFPB are about post-origination mortgage servicing. Nonetheless, this finding highlights an important distinction between our study and the existing studies that focus on the origination stage of mortgages to understand the impact of nonbanks on the mortgage market. We conduct a battery of robustness tests. First, we use an alternative measure of the complaint ratio, where a county’s outstanding mortgages are proxied by the total number of mortgages originated in that county during the last five years. Second, we use an alternative measure of nonbanks’ market share based on the number of mortgages rather than the dollar amount of mortgages. Third, to examine whether the results are driven by large nonbanks, we exclude Quicken Loans (i.e., the nonbank with the largest amount of loan originations in our sample) and the top 3 nonbanks, respectively. Fourth, we drop observations from 2012 to limit the impact of GFC on the result. The negative association between the complaint ratio and nonbanks’ market share survive all the robustness tests. These results are reported in Tables A2 to A5.18 170.1638×0.037=0.006;0.006/0.05=12% 18Inadditionalrobustnesstests,we(1)includeallmortgages(includingrefinancingloans,loansforhouses thatarenon-owneroccupied,andloansformulti-familyhousings)incalculatingnonbanks’marketshare,(2) includeFHAandVAloansinthesample, (3)requirethemediannumberofcomplaintsinthecountyduring our sample period to be more than one, and (4) include all counties with at least one complaint during the sampleperiod. Ourmainresultsremainrobust. Theresultsfromthesetestsarenottabulatedbutavailable upon request. 18

4.2 Instrumental variable analysis Thus far, we have documented a negative association between the county-level complaint ratio and nonbanks’ market share in the county. However, the presence of nonbanks in a county is not random. County characteristics or local economic dynamics may affect nonbanks’ decisions to expand in specific counties. Therefore, the main identification challenge to argue causality is that contemporaneous changes in a county’s economic conditions may be correlated with both the market share of nonbanks in the county and the complaint ratio of the county, resulting in biased estimates. In particular, in a scenario where we fail to control for relevant economic conditions, we may falsely attribute the changes in complaint ratios to the changes in nonbanks’ market shares, when the changes in complaint ratios are in fact due to the changes in the county’s economic conditions. To address this identification challenge, we conduct two instrumental variable (IV) analyses. 4.2.1 Instrumental variable analysis: stress tests The first IV analysis exploits the stress tests implemented by the Federal Reserve. The DFA requires Federal Reserve to conduct annual stress tests starting from 2009 to assess if large bank holding companies have sufficient capital to absorb losses under several stress scenarios. Covas (2017) shows that stress tests impose disproportionately high capital requirements on residential mortgages. Buchak et al. (2018) suggest that banks reduce mortgage lending in order to build capital buffers required by regulations, facilitating the expansion of nonbanks’ market share.19 Therefore, nonbanks’ market share is likely to increase more aggressively in counties with a higher ex ante exposure to stress-tested banks, ceteris paribus. For this IV analysis, we use counties’ ex ante exposure to stress-tested banks as an instrument for the change in nonbanks’ market shares in the local residential mortgage market. Specifically, we calculate the IV as the county’s mortgage market share in 2008 (i.e., the year right before the inception of stress tests) held by banks that would be subject to 19Please also see Irani et al. (2021), Chernenko et al. (2022), and Begley and Srinivasan (2022). 19

stress tests during our sample period. This IV analysis examines whether the instrumented within-county change in nonbanks’ market shares between 2012 and 2018 affects the withincounty change in the complaint ratios during the same period. We recognize that stress tests may directly affect the service quality in the local mortgage market through two channels. First, stress tests may affect the service quality of tested banks because of the increased regulatory oversight. Second, stress tests may further affect the service quality of non-tested banks through peer effects or market competition. However, these considerations should not result in a violation of the exclusion restriction in our empirical setup because our sample period starts in 2012, which is years after the inception of stress tests (i.e., 2009).20 During our sample period, the impact of stress tests on service quality is likely already incorporated in the observed level of service quality. Therefore, although the ex ante exposure to stress-tested banks may have a direct effect on the level of service quality in the local mortgage market soon after the inception of stress tests, it should not influence the within-county changes in service quality years after the inception of stress tests through channels other than its impact on the evolution of lender composition (i.e., banks versus nonbanks) in the county. In addition, we conduct two robustness tests of this IV strategy to ensure our interpretation is not contaminated by the direct impact of stress tests on service quality provided by banks and nonbanks. The details of these robustness tests are discussed at the end of this subsection. We estimate the following regression model: Complaintratio = βNonbankshare +γControls +(cid:15) , c,∆(2018−2012) c,∆(2017−2011) c,∆(2017−2011) c (2) where c,∆(2017−2011) indicates the within-county change in variables from 2011 to 2017, and Nonbankshare is instrumented by counties’ ex ante exposure to banks subc,∆(2017−2011) ject to stress tests. 20Prior to the passage of DFA in 2010, the Federal Reserve initiated the Supervisory Capital Assessment Program (SCAP) in 2009 to conduct stress tests on large U.S. bank holding companies. 20

Table 4 presents the regression results. The result from the first-stage regression (column 1)showsthatcountieswithahigherexanteexposuretostress-testedbanksindeedexperience higher growth in nonbanks’ market share. Consequently, the result from the second-stage regression (column 2) shows that these counties witness a larger reduction in the mortgagerelated complaint ratio. In column (3), we confirm that the results hold if we restrict complaints to only those related to post-origination mortgage servicing. Regarding the size of the coefficients, the IV estimates suggest a much larger effect of nonbanks’ market share on the complaint ratio than the OLS estimates. To strengthen our identifying assumption and ensure that our IV results are not due to the direct impact of stress tests on service quality, we conduct two robustness tests. First, we focus on a more recent change in complaint ratios (i.e., changes between 2014 and 2018). Considering that any possible direct effects of stress tests on service quality may take time to appear, the changes in complaint ratios during the early years of the stress tests may be partially due to stress tests directly. However, this is less likely to be a concern if we focus on changes in complaint ratios during the later years of our sample period. The results reported in 5 show that our findings remain robust when we focus on this more recent change in complaint ratios. Second, we limit our sample to counties, whose ex ante exposure to stress-tested banks is not influenced by the inclusion of more banks in the stress tests in the later years of our sample period. Because the number of stress-tested banks increased significantly in 2014,21 some counties’ exposure to stress-tested banks experienced a large increase after 2013, which may affect the service quality in these counties and hence the within-county change in complaint ratios that we focus on in our estimation. This should not be a concern if we drop counties with ex ante exposure to banks that have only become a participant in stress tests after 2013. The results reported in 6 show that our findings remain robust when we use this subsample of counties. 21Pre-2014,stresstestsonaverageinclude19BHCs; since2014,stresstestsonaverageincludemorethan 30 BHCs. 21

4.2.2 Instrumental variable analysis: mortgage industry surety bonds The second IV analysis exploits the variation in mortgage industry surety bonds required by each state. These requirements are imposed in order to protect mortgage borrowers against dishonest lending practices. To conduct mortgage business in a state, mortgage professionals, such as mortgage brokers, lenders, originators, and servicers, need to apply for licenses from the state and are required to abide by all state laws and regulations. In particular, as part of the licensing application, mortgage professionals (i.e., the principals) are required to post surety bonds through a surety bond company. In case of noncompliance (such as predatory lending behavior), surety bond company will promptly pay for all costs and damages and then seek full compensation, including expenses, from the principals. The annual premium charged by the surety bond company is determined by the amount of surety bond required, the expected value of claims against the principal, and the probability of collecting full compensation from the principal. As a result, before issuing the bond, the surety bond company may conduct detailed screening of the bond applicants, making surety bond a significant barrier to entry in states with a high amount of surety bond requirement, especially for entities with few years in the business (Kleiner and Todd (2009). Therefore, the amount of surety bond required in a state reflects the costs of conducting mortgage business in the state and the difficulties in entering the mortgage market of the state. This implies that nonbanks should expand more aggressively in states with lower surety bond requirements, ceteris paribus (the relevance condition). ForthisIVanalysis, weusestates’minimumamountofsuretybondimposedonmortgage brokers as an instrument for the change in nonbanks’ market shares in the local residential mortgage market.22 Same as the first IV strategy, this IV analysis examines whether the instrumented within-county change in nonbanks’ market shares between 2012 and 2018 af- 22Surety bonds imposed on mortgage brokers are the most prevalent across states compared to those imposed on other mortgage professionals; such information, therefore, provides a clear comparison of the costs of conducting mortgage business across states. Typically, the amount of surety bond imposed on mortgage lenders and servicers is usually the same as that imposed on mortgage brokers. 22

fects the within-county change in the complaint ratios during the same period. One might argue that the level of surety bond requirement of a state may reflect the overall toughness of the state’s mortgage regulations, which might have a direct effect on the level of service quality in the mortgage industry of the state. However, surety bond requirements should not affect the within-county changes in service quality as long as there are no changes in these regulations over time (the exclusion condition). Therefore, for this IV analysis, we restrict our sample to states that have not experienced any changes in the amount of bond required since 2010.23 We re-estimate equation (2) but use the log amount of the mortgage broker surety bond required by each state to instrument for Nonbank . Table 7 presents the regresc,∆(2017−2011) sion results. Column (1) reports the result from the first-stage regression. It shows that counties that reside in states with a lower amount of surety bond requirement experience higher growth in nonbanks’ market share. These counties, as shown in the results from the second stage regression reported in column (2), witness a larger reduction in the mortgagerelated complaint ratio subsequently. Column (3) shows that the results hold if we restrict complaints to only those related to post-origination mortgage servicing. 5 Evaluate potential explanations for the improved service quality Inthissection,weexploretheexplanationsfortheimprovedservicequalityaccompanying the rise of nonbanks in the local residential mortgage market. We also investigate whether the effect of nonbanks’ market share on service quality differs across counties based on the percentage of minority populations. 23To identify states with no changes in surety bond requirement for mortgage brokers during our sample period,wecollecteachstate’smortgagebrokersuretybondrequirementatthetimeofwriting(i.e,September 2021) and compare it with the mortgage broker surety bond requirement as of 2010 available here. 23

5.1 Mortgage service quality: nonbanks vs. traditional banks Onepotentialexplanationforthedecreaseinthecomplaintratioisthatnonbanksprovide better services than traditional banks on average. If so, the increase in nonbanks’ market share in a county would reduce the complaint ratio in the county. To examine whether this is the case, we conduct a univariate comparison of complaint ratios between nonbanks and traditional banks that are located in the same county. We find that the average complaint ratio of nonbanks (0.46%) is much higher than that of traditional banks (0.27%). This comparison suggests that the cross-sectional difference between the complaint ratios of nonbanks and that of traditional banks cannot explain the reduced county-level complaint ratio as nonbanks’ market share increases. We interpret this finding as consistent with the differences in borrower characteristics betweennonbanksandtraditionalbanksdocumentedinpriorliterature. AsshowninBuchak et al. (2018), nonbanks are more likely to serve riskier, less creditworthy, and lower-income borrowers. Considering that such borrowers are more likely to have difficulties in making mortgage payments, and therefore, more likely to file a complaint (Begley and Purnanandam 2021), it is not surprising that nonbanks have higher complaint ratios than traditional banks onaverage. Thesefindingsarealsoconsistentwiththedifferencesinbusinessmodelsbetween nonbanks and traditional banks. It is typical for nonbanks to finance their entire originations through securitization and the originate-to-distribute model; whereas traditional banks still hold between 30% and 50% of their originations on their balance sheets (Buchak et al. 2018). As a result, traditional banks may be more concerned with loan performance, and consequently, more incentivized to provide good services. Overall,theresultindicatesthatthereductioninthecomplaintratiofollowinganincrease in nonbanks’ market share is not due to the cross-sectional difference in the service quality between nonbanks and traditional banks. 24

5.2 Nonbanks’ market share and mortgage-related complaints against nonbanks 5.2.1 OLS and IV analysis In this section, we examine if the decrease in the county-level complaint ratio is due to an improvement in service quality by nonbanks as their market share increases. We start evaluating this explanation by studying the evolution of complaint ratios for nonbanks. Specifically, we exclude the complaints against traditional banks and calculate the countylevel complaint ratio for nonbanks, which is the number of complaints against nonbanks divided by the number of outstanding loans originated by nonbanks. Since the federal tax filings do not let us distinguish between mortgages granted by nonbanks and those granted by traditional banks, we proxy the denominator by the total number of mortgages originated by nonbanks in the county during the last five years. We conduct both the OLS and the IV analyses. Table 8 reports the regression results. The OLS estimates reported in column (1) show that as nonbanks hold a larger market share in a county, their complaint ratio in the county decrease—a one standard deviation increase in nonbanks’ market share is associated with a 0.083 percentage point reduction in nonbanks’ complaint ratio. This magnitude of reduction is economically meaningful as it represents 18% of the average county-level complaint ratio for nonbanks.24 Columns (2) and (3) report the results from the IV regressions using the two IV strategies discussed in Section 4.2. The results show that the instrumented change in nonbanks’ market shares is associated with a reduction in nonbanks’ complaint ratio. The results (reported in columns (4)-(6)) remain similar if we focus on only complaints about the post-origination mortgage servicing. Taken together, the results in Table 8 suggest that nonbanks indeed improve their service quality as they increase their market share in a county. One might wonder if these results are due to the transfer of mortgage servicing rights 24The average county-level complaint ratio for nonbanks is 0.46%. 25

from nonbanks to traditional banks. Although the information on the transfer of mortgage servicingrightsisnotavailabletous,wedonotbelievethatthetransferofmortgageservicing rights from nonbanks to traditional banks is driving our results. As discussed in 3.2, during oursampleperiod,mortgagesthatareoriginatedbynonbanksarelikelyservicedbynonbanks because traditional banks are not likely to buy mortgage servicing rights from nonbanks due to the Basel III capital requirements (Lux and Greene 2015). By contrast, nonbanks, which do not face the same regulations, are more likely to purchase mortgage servicing rights from traditional banks, especially the mortgage servicing rights of lower-quality loans. Therefore, our results, if anything, may underestimate the extent to which nonbanks improve their service quality during our sample period. 5.2.2 The mechanisms through which nonbanks improve their service quality We explore two mechanisms through which nonbanks may improve their service quality as they increase their market share. The first mechanism that we explore is product specialization. If nonbanks develop a specialty in servicing certain types of borrowers as their market share increases, such specialization may allow them to provide higher-quality services to their customers. The second mechanism that we explore is nonbanks’ investment in technology. Fuster et al. (2019) show that the adoption of technology play an important role in improving the efficiency of the U.S. mortgage market. If nonbanks make such investments as their market share grows, their service quality may improve as a result. 5.2.2.1 Product Specialization Regarding the first mechanism, we start by exploring the borrower profiles of nonbanks and estimate the following regression: Outcome = βNonbankshare +δ +η +(cid:15) , (3) c,t c,t−1 c t c,t 26

where Outcome is either the average income of nonbanks’ borrowers in a county-year or c,t the income dispersion of these borrowers (i.e., the interquartile range of borrower incomes scaled by the average income). Nonbank is the lagged market share of nonbanks in the c,t−1 county’s residential mortgage market. δ indicates county fixed effects and η indicates year c t fixed effects. Column (1) of Table 9 reports the results for the average income. We find a negative and significant coefficient on Nonbankshare , indicating that as nonbanks’ market share c,t−1 increases in the local mortgage market, the average income of their borrower base decreases. Column (2) shows the result for the income dispersion. We find a lower income dispersion as nonbanks’ market share increases. Taken together, the evidence suggests that nonbanks become increasingly focused on lower-income borrowers as nonbanks’ market share grows. Previous literature documents that the average borrower income of nonbanks is lower than that of traditional banks (Buchak et al. 2018). Our evidence is different from this literature in that our evidence is not about the difference between the average borrower income of nonbanks and that of banks. Instead, we show that nonbanks’ customers become more concentrated among lower-income borrowers as nonbanks’ market share increases. Throughoutourpaper,weinterpretthedecreasedcomplaintratioasevidenceofimproved service quality. One might argue that lower-income borrowers may be of a lower level of financial literacy so they do not know that they can complain to the CFPB or how to do so. We do not believe that this possibility could bias our results and interpretation in a meaningful way because previous literature shows that there are more mortgage-related complaints in areas with a higher proportion of poor populations (Begley and Purnanandam 2021). Furthermore, the cross-sectional comparison of complaint ratios between nonbanks and banks in Section 5.1 implies that nonbanks’ borrowers are not less likely to file complains although the average borrower income of nonbanks is much lower than that of traditional banks. Critical to this first mechanism is whether nonbanks indeed become better at servicing 27

lower-income borrowers. To shed light on this question, we take a close look at the specific issues of the complaints. Specially, we examine complaints about “Loan modification, collection, foreclosure” and “Struggling to pay mortgage,” both of which are related to situations in which borrowers may have financial difficulties in making mortgage payments. In particular, borrowers may not have the financial resources to make mortgage payments based on the original mortgage terms. If nonbanks become more experienced in providing mortgage services to lower-income borrowers, complaints regarding these two issues should be reduced. Column (1) in Table 10 reports evidence supportive of this mechanism. It shows that as nonbanks’ market share increases, complaints that are more likely filed by borrowers who lack financial resources are significantly reduced. Collectively, our findings are consistent with the notion that as nonbanks’ market share increases, they develop a specialty in servicing lower-income borrowers, hence improving their service quality towards these borrowers. 5.2.2.2 Investment in technology Regarding the second mechanism, we start by examining whether nonbanks’ have enhanced incentives to invest in technology as their market share increases. We estimate the following two-stage regression: Incentives = β Nonbankshare +γControls +δ +η +(cid:15) (1ststage) Nonbanks,c,t 0 c,t c,t c t c,t (cid:92) Complaintratio = β Incentives +γControls +δ +η +(cid:15) (2ndstage). c,t 1 Nonbanks,c,t−1 c,t−1 c t c,t (4) Incentives indicates two potential sources of incentives for nonbanks to invest in Nonbanks,c,t technology. Regarding the first source of investment incentives, we propose that nonbanks may have heightened incentives to invest if the increased nonbank market share is concentrated within a few nonbanks, because these nonbanks are more likely to internalize the benefit of such investment (Petersen and Rajan 1995; Giannetti and Saidi 2019; Saidi and Streitz 2021). We measure this source of incentives using the market concentration within 28

nonbanks. Specifically, we calculate the Herfindahl–Hirschman Index (HHI) of nonbanks in the county.25 This variable captures whether a small number of nonbanks dominate the increased market share held by nonbanks or a large number of nonbanks are competing with each other for the increased market share held by nonbanks.26 Regarding the second source of investment incentives, we propose that nonbanks may have heightened incentives to invest in technology if they have a large nation-wide market share because they can take advantage of the economies of scale and the marginal costs of technology investment are decreasing. We measure this source of incentives using the market share in the nation held by all the nonbanks that originate loans in the county. The national market share is calculated as the dollar amount of mortgages originated by the nonbanks in year t-1 scaled by the total dollar amount of mortgages originated in the nation in year t-1. This variable captures whether nonbanks increase their presence at the national level while increasing their market share in the focal county or they retract from othercounties. Controls includescounty-yearlevelincomepercapitaandunemployment c,t−1 rate. δ indicates county fixed effects and η indicates year fixed effects. c t Table 11 reports the regression results for the first source of investment incentives. Column (1) reports the result for the first stage and indicates that the market share of nonbanks becomes more concentrated as nonbanks’ overall market share in the county increases. This result indicates that the main driver of the increase in nonbanks’ market share is the increased presence of a small number of nonbanks in the county rather than an increased number of nonbanks in the county. The second stage result in column (2) suggests that the increased market concentration within nonbanks, predicted by the higher market share of nonbanks in the county, is associated with a reduction in the complaint ratio for nonbanks in the county. Column (3) shows that the result holds for complaints about post-origination 25We calculate the summation of the squares of each nonbank’s market share of all mortgages originated by nonbanks in the county. 26For instance, if there are traditional banks and only one nonbank in the county mortgage market, this measurewouldtakethevalueofonebecauseallthemortgagesoriginatedbynonbanksareoriginatedbythe only nonbank in the county. As the number of nonbanks increases, the measure approaches zero. 29

mortgage servicing. Table 12 reports the regression results for the second source of investment incentives. Column (1) reports the results for the first stage and shows that as nonbanks’ market share increases in a county, these nonbanks’ market share in the nation increases as well. In other words, theincreasedpresenceofnonbanksinacountyisnotduetotheshiftofresourcesfrom other counties. The second stage result in column (2) suggests that the increased market share at the national level, predicted by the higher market share of nonbanks in the county, is associated with a reduction in the complaint ratio for nonbanks in the county. Column (3) shows that the result holds for complaints about post-origination mortgage servicing. Do nonbanks indeed increase their investment in technology as their market share increases? Using job posting data from Burning Glass Technologies (BGT) during our sample period, we find evidence supportive of this notion. BGT collects the near-universe online job postings. Aparticularlyusefulfeatureofthisdataisthatitcompilestheskillrequirementsin thejob postings. Ifnonbanksincreaseinvestments intechnology, their demandforemployees with technology-related skills should also increase. To examine whether this is the case, we aggregate the number of job postings in a county that require technology-related skills by nonbanks, and regress it on the market share of nonbanks in the county. We categorize a skill as a technology-related skill if it is tagged by the BGT as an information technology (IT) skill or a software skill. Table 13 reports the results. With different specifications, we consistently observe a positive and significant association between nonbanks’ market share and their demand for technology-related skills. In terms of economic significance, column (4) indicates that a one standard deviation increase in nonbanks’ market share in a county is associated with an increase in the number of nonbanks’ job postings in the county that require technology-related skills by 13%. To assess whether the effect on skill requirements is due to a few large fintech lenders, In column (5), we exclude Quicken Loans and Guaranteed Rate Inc., two of the largest fintech nonbanks, and our result remains robust. This evidence suggests that nonbanks’ use of technology increases as their market share increases, 30

potentially helping nonbanks provide higher-quality services to their mortgage customers.27 Finally, we examine the specific issues of the complaints against nonbanks. Among all the issues listed in Table 1, “Trouble during payment process” is most likely to benefit from investment in technology because this issue is likely resolved by improvements in nonbanks’ infrastructure, such as upgrades of the technologies deployed in their online payment platforms. Therefore, if increased investment in technology can at least partially explain our finding, complaints regarding this issue should be reduced as nonbanks increase their market share. Column (2) of Table 10 reports evidence supportive of this mechanism. It shows that as nonbanks’ market share increases, complaints that are more likely due to technical issues in the payment process are significantly reduced. In addition, columns (3) and (4) of Table 10 show that as nonbanks’ market share increases, complaints regarding post-origination mortgage servicing that are not related to the two identified mechanisms do not experience significant reductions. The evidence, therefore, provides further support to the two mechanisms that we propose in this paper. 5.3 Nonbanks’ market share and mortgage-related complaints against traditional banks In this section, we study the effect of the increased market share of nonbanks on the complaint ratio of traditional banks. We are interested in understanding if the improved service quality of nonbanks comes at the price of a deterioration in the service quality of traditional banks. We regress the complaint ratio of traditional banks on nonbanks’ market share. The complaint ratio of traditional banks is calculated as the number of mortgage complaints against traditional banks divided by the number of outstanding loans originated bytraditionalbanks; thedenominatorisproxiedbythetotalnumberofmortgagesoriginated by traditional banks in that county during the last five years. 27The positions advertised in these job postings are likely filled because the finance industry has one of the highest vacancy-employment conversion rates probably because wages in the finance industry are higher than those in the other industries (Abis and Veldkamp 2020). 31

The results are reported in Table 14. In column (1), we find that the complaint ratio of traditional banks decreases as nonbanks account for a larger market share. Column (2) shows that this result holds using the stress test IV strategy described in Section 4.2. In column (3), we restrict our sample to only traditional banks that have never undertaken a stress test in the IV analysis and our finding remains robust. Column (4) shows that the result also holds when we use the surety bond IV strategy described in Section 4.2. Columns (5) to (8) repeat the above exercises using only complaints about post-origination mortgage servicing and the results are similar to those reported in the first four columns. The results from Table 14 imply that the reduced complaint ratio of traditional banks also contributes to the reduced complaint ratio in the county. However, we are cautious to conclude that banks improve their service quality as nonbanks’ market share increases. The reason is that the reduced complaint ratio of traditional banks might be due to the change in the composition of these banks’ borrower base. Table 15 shows that as nonbanks’ market share increases in the local mortgage market, the average income of traditional banks’ borrowers increases albeit the income dispersion does not change. Begley and Purnanandam (2021) show that, on average, higher-income borrowers receive higher-quality mortgage services, hence are less likely to file complaints against their lenders. The increased average income of banks’ borrowers (reported in Table 15) and the decreased average income of nonbanks’ borrowers (reported in Table 9) together suggest that nonbanks and banks become increasingly specialized in different types of borrowers as nonbanks grow in size. Importantly, this lender specialization benefits lower-income borrowers without hurting higher-income ones, consistent with an improvement in consumer welfare. 5.4 Effects of nonbanks’ market share conditional on minority populations A highly relevant topic among studies on the mortgage market is the treatment of minority borrowers. Previous papers show that minorities are more likely to experience dis- 32

crimination in the mortgage market in terms of pricing, access to credit, and the quality of mortgage services (e.g., Munnell et al. 1996; Bartlett et al. 2022; Begley and Purnanandam 2021). We contribute to this strand of literature by examining whether the rise of nonbanks in the mortgage market has a disproportionately larger positive effect on the service quality received by minority borrowers. For this analysis, we estimate the following regression model: Complaintratio = β Nonbankshare +β Nonbank ×Minoritypopulation c,t 0 c,t−1 1 c,t−1 c (5) +γControls +δ +η +(cid:15) , c,t−1 c t c,t where the dependent variable is the county-level complaint ratio. Nonbankshare is the c,t−1 laggedmarketshareofnonbanksinthecounty’sresidentialmortgagemarket. Minoritypopulation c is the percentage of minority populations in the county from the 2010 Census. We measure the minority population in two ways: the Hispanic population and the Non-white population. All other variables are defined the same as in equation (1). The variable of interest is the interaction term, the coefficient on which implies whether the increased market share of nonbanks has a stronger effect on the complaint ratio in counties with more minorities. Table 16 reports the results. Across the four columns, the coefficients on the interaction termareallnegativeandsignificantatthe1%level, indicatingthattheeffectoftheincreased market share held by nonbanks on mortgage-related complaints is more pronounced for counties with a higher percentage of minority populations. This finding supports the role of nonbanks in alleviating discrimination in the quality of services received by minorities in the mortgage market. 6 Conclusion In this paper, we study the consequences of the rise of nonbanks on the service quality in the residential mortgage market. The specific aspect of service quality that we focus on 33

is the mortgage-related complaints filed with the CFPB as a percentage of all outstanding mortgages (i.e., the complaint ratio). We find that as nonbanks increase their market share in a county, the complaint ratio of the county decreases. To verify that the relation between nonbanks’marketshareandmortgage-relatedcomplaintsisindeedcausal,weconducttwoIV analyses. Weinstrumentthechangeinnonbanks’marketsharesinthelocalmortgagemarket using(1)thecounty’sexanteexposuretobankssubjecttoFederalReserve’sstresstests, and (2) mortgage industry surety bonds required by each state. We find that the instrumented increase in nonbanks’ market share is associated with a reduction in the mortgage-related complaint ratio. Despite the fact that nonbanks have a higher complaint ratio than traditional banks on average, we find that nonbanks significantly improve their service quality as their market share increases, contributing to the reduction in the complaint ratio at the county level. For this improvement in service quality, we provide evidence consistent with two explanations. First, as nonbanks increase their market share, they develop a specialty in servicing lower-income borrowers. Second, as nonbanks increase their market share, they make more investments in technology. Moreover, as nonbanks’ market share grows, traditional banks increasingly focus on higher-income borrowers, and their complaint ratio decreases as well. Therefore, the evidence together suggests an improvement in consumer welfare. We also find that the improvements in service quality are more likely to benefit marginalized borrowers, such as minorities, who are more likely to receive low-quality financial services. As far as we are aware, our paper is the first to focus on the service quality of mortgages to understand the consequences of nonbanks’ expansion on consumer welfare. Focusing on financial products other than mortgages, such as cryptocurrencies and credit cards, existing studies show that nonbank financial companies, fintech companies in particular, may harm consumers (Calem 2022). We find that in the residential mortgage market, although nonbanks on average have a higher complaint ratio than traditional banks, their expansion reduces their complaint ratio, leading to a reduction in the overall complaint ratio at the 34

county level. Our findings suggest that future policies and regulations of nonbanks should consider the effect of market shares on service quality. Our findings echo Philippon (2016) in thatthenon-traditionalpartofthefinancialsectorhasthepotentialtoimprovetheefficiency of the financial system. 35

References S. Abis and L. Veldkamp. The changing economics of knowledge production. Available at SSRN 3570130, 2020. R.Avery, K.Brevoort, andG.Canner. Opportunitiesandissuesinusinghmdadata. Journal of Real Estate Research, 29(4):351–380, 2007. R. Bartlett, A. Morse, R. Stanton, and N. Wallace. Consumer-lending discrimination in the fintech era. Technical Report 1, 2022. T. A. Begley and A. Purnanandam. Color and credit: Race, regulation, and the quality of financial services. Journal of Financial Economics, 2021. T.A.BegleyandK.Srinivasan. Smallbanklendingintheeraoffintechandshadowbanking: a sideshow? The Review of Financial Studies forthcoming, 2022. G. Buchak, G. Matvos, T. Piskorski, and A. Seru. Fintech, regulatory arbitrage, and the rise of shadow banks. Journal of Financial Economics, 130(3):453–483, 2018. P. Calem. Lax supervision of fintechs harms consumers. Available at https://bpi.com/laxsupervision-of-fintechs-harms-consumers/, 2022. S. Chernenko, I. Erel, and R. Prilmeier. Why do firms borrow directly from nonbanks? The Review of Financial Studies forthcoming, 2022. K. R. Cort´es, Y. Demyanyk, L. Li, E. Loutskina, and P. E. Strahan. Stress tests and small business lending. Journal of Financial Economics, 136(1):260–279, 2020. F. Covas. The capital allocation inherent in the federal reserve’s capital stress test. Research Note, January, 31, 2017. I. Drechsler, A. Savov, and P. Schnabl. How monetary policy shaped the housing boom. Technical report, National Bureau of Economic Research, 2019. 36

M. Egan, G. Matvos, and A. Seru. The market for financial adviser misconduct. Journal of Political Economy, 127(1):233–295, 2019. I. Erel and J. Liebersohn. Does fintech substitute for banks? evidence from the paycheck protection program. Technical report, National Bureau of Economic Research, 2020. A.Fuster, M.Plosser, P.Schnabl, andJ.Vickery. Theroleoftechnologyinmortgagelending. The Review of Financial Studies, 32(5):1854–1899, 2019. A. Fuster, M. C. Plosser, and J. I. Vickery. Does cfpb oversight crimp credit? 2021. P. Gete and M. Reher. Mortgage securitization and shadow bank lending. The Review of Financial Studies, 34(5):2236–2274, 2021. M. Giacoletti, R. Heimer, and E. G. Yu. Using high-frequency evaluations to estimate discrimination: Evidence from mortgage loan officers. Available at SSRN 3795547, 2021. M. Giannetti and F. Saidi. Shock propagation and banking structure. The Review of Financial Studies, 32(7):2499–2540, 2019. M. Gopal and P. Schnabl. The rise of finance companies and fintech lenders in small business lending. The Review of Financial Studies forthcoming, 2022. J. M. Griffin, S. Kruger, and P. Mahajan. Did fintech lenders facilitate ppp fraud? Available at SSRN 3906395, 2021. U. G. Gurun, N. Stoffman, and S. E. Yonker. Trust busting: The effect of fraud on investor behavior. The Review of Financial Studies, 31(4):1341–1376, 2018. C. Haendler and R. Heimer. The financial restitution gap in consumer finance: Insights from complaints filed with the cfpb. Available at SSRN 3766485, 2021. 37

J. P. Hughes and L. J. Mester. Who said large banks don’t experience scale economies? evidence from a risk-return-driven cost function. Journal of Financial Intermediation, 22 (4):559–585, 2013. R. M. Irani, R. Iyer, R. R. Meisenzahl, and J.-L. Peydro. The rise of shadow banking: Evidencefromcapitalregulation. The Review of Financial Studies, 34(5):2181–2235, 2021. J. Jagtiani, L. Lambie-Hanson, and T. Lambie-Hanson. Fintech lending and mortgage credit access. The Journal of FinTech, 1(01):2050004, 2021. M. M. Kleiner and R. M. Todd. Mortgage broker regulations that matter: Analyzing earnings, employment, and outcomes for consumers. In Studies of labor market intermediation, pages 183–231. University of Chicago Press, 2009. J. LaVoice and D. F. Vamossy. Racial disparities in debt collection. Available at SSRN 3465203, 2019. E. Loutskina and P. E. Strahan. Securitization and the declining impact of bank finance on loan supply: Evidence from mortgage originations. The Journal of Finance, 64(2): 861–889, 2009. M. Lux and R. Greene. What’s behind the non-bank mortgage boom? M-RCBG Associate Working Paper, (42), 2015. L. J. Mester. Scale economies in banking and financial regulatory reform. The Region, 24: 10–13, 2010. A. H. Munnell, G. M. Tootell, L. E. Browne, and J. McEneaney. Mortgage lending in boston: Interpreting hmda data. The American Economic Review, pages 25–53, 1996. M. A. Petersen and R. G. Rajan. The effect of credit market competition on lending relationships. The Quarterly Journal of Economics, 110(2):407–443, 1995. 38

T. Philippon. The fintech opportunity. Technical report, National Bureau of Economic Research, 2016. G. Plantin. Shadow banking and bank capital regulation. The Review of Financial Studies, 28(1):146–175, 2015. F. Saidi and D. Streitz. Bank concentration and product market competition. The Review of Financial Studies, 34(10):4999–5035, 2021. J. Sedunov. Small banks and consumer satisfaction. Journal of Corporate Finance, 60: 101517, 2020. 39

Figure 1: Nonbanks’ market share This figure plots the median and inter-quartile range of the county-level nonbank market shares between 2008 and 2018. Nonbanks’ market share for a county-year is calculated as the dollar amount of mortgages originated in the county-year by nonbanks over the dollar amount of mortgages originated in the county year by all lenders. Source: Authors’ calculations based on HMDA (Federal Financial Institutions Examination Council (U.S.), Home Mortgage Disclosure Act (Public Data)). 40

Figure 2: Nonbanks’ market share: Originations vs. post-origination servicing This figure plots the relationship between nonbanks’ market share of mortgage originations and that of post-origination servicing at different geographical levels, using the sample of mortgages sold to Freddie Mac and still outstanding during our sample period (i.e., 2012–2018). The top left panel shows nonbanks’ nationwide market shares, the top right panel shows nonbanks’ state-level market shares, the bottom left panel shows nonbanks’ MSA-level market shares, and the bottom right panel shows nonbanks’ 3-digit ZIP code-level market shares. We group nonbanks’ market shares of mortgage originations into 10 bins for the state-level panel and 100 bins for the MSA and ZIP-code level panels, and plot the average nonbank market share of mortgage originations and that of post-origination servicing for each bin. The market shares are trimmed at the 5% level. Source: Authors’calculationsbasedonHMDA(FederalFinancialInstitutionsExaminationCouncil (U.S.), Home Mortgage Disclosure Act (Public Data)) and Freddie Mac Single Family Loan-Level Dataset. 41

Table 1: Summary statistics: Complaints from the CFPB This table reports the distribution of issues associated with the mortgage-related complaints filed with the CFPB. Issue Frequency Percent (%) Loan modification, collection, foreclosure 67,993 45.54 Loan servicing, payments, escrow account 40,985 27.45 Trouble during payment process 9,230 6.18 Application, originator, mortgage broker 8,486 5.68 Struggling to pay mortgage 8,328 5.58 Other 4,479 3.00 Settlement process and costs 4,044 2.71 Credit decision / Underwriting 2,674 1.79 Applying for a mortgage or refinancing an existing mortgage 1,596 1.07 Closing on a mortgage 974 0.65 Issues related to credit report 502 0.34 Total 149,291 42

Table 2: Sample summary statistics The main sample that we use in our analysis is a product of matching county-level complaint data set from CFPB with county-level mortgage market data from HMDA, supplemented with county characteristics from BEA, BLS, and 2010 Census. This table reports the countyyear-level summary statistics. Mean SD 25th perc. Median 75th perc. Complaint ratio (%) 0.05 0.06 0.02 0.04 0.07 # of complaints 19.47 65.27 1.00 4.00 13.00 # of outstanding mortgages 29,304 57,392 4,550 10,966 29,500 Nonbank market share (%) 32.17 16.38 19.67 30.67 43.39 Income per capita (tho. $) 44.05 13.78 36.15 40.98 48.00 Unemployment rate (%) 5.91 2.26 4.32 5.50 7.05 Mortgage market HHI 0.07 0.05 0.04 0.06 0.09 Hispanic ratio (%) 10.12 12.74 2.89 5.52 11.74 Non-white ratio (%) 17.09 14.51 6.64 12.06 23.22 Obs. (County × Year) 7,178 43

Table 3: Nonbanks’ market share and mortgage-related complaints This table presents the OLS estimates from the regression of the county-level complaint ratio on nonbanks’ market share in the county. The dependent variable is Complaint ratio, which is the number of mortgage-related complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year. Column (1) includes all mortgage-related complaints; column (2) includes complaints about postorigination mortgage servicing; column (3) includes complaints about mortgage originations. Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio (1) (2) (3) All complaints Complaints about Complaints about servicing originations Nonbank market share -0.037** -0.042** 0.005 (0.017) (0.016) (0.005) log(Income per capita) -0.056*** -0.046** -0.009* (0.021) (0.020) (0.005) Unemployment rate 0.284*** 0.306*** -0.022 (0.085) (0.082) (0.027) Year FE Yes Yes Yes County FE Yes Yes Yes Observations 7,178 7,178 7,178 R-squared 0.568 0.556 0.353 44

Table 4: IV analysis using stress tests This table presents the results of a regression model in which we regress the change in county-level complaint ratios from 2012 to 2018 on the change in nonbanks’ market shares in the county from 2011 to 2017, which is instrumented by the county’s ex ante exposure to banks subject to stress tests. The dependent variable in the first-stage regression is the change in nonbank market shares. The dependent variable in the second-stage regressions is the change in complaint ratios; complaint ratio is calculated as the number of mortgagerelated complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year; in column (2), the dependent variable is calculated using all mortgage-related complaints; in column (3), the dependent variable is calculated using complaints about post-origination mortgage servicing. Stress test exposure is the percentage of the county’s mortgage market that is served by banks subject to stress testsrightbeforetheinceptionofstresstests. Income per capitaisfromBEA;Unemployment rate is from BLS. All variables are defined in Table A1. ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, and ∗p < 0.1. First stage Second stage Second stage (1) (2) (3) All complaints Complaints about servicing Stress test exposure 0.228*** (0.028) ∆Nonbank market share -0.457*** -0.433*** (0.090) (0.085) ∆log(Income per capita) -0.186*** -0.118*** -0.100** (0.058) (0.042) (0.040) ∆Unemployment rate -0.443** 0.115 0.151 (0.214) (0.169) (0.161) Observations 710 710 710 F-stat 63.80 45

Table 5: IV analysis using stress tests: The change in complaint ratios between 2014 and 2018 This table presents the results of a regression model in which we regress the change in county-level complaint ratios from 2014 to 2018 on the change in nonbanks’ market shares in the county from 2013 to 2017, which is instrumented by the county’s ex ante exposure to banks subject to stress tests. The dependent variable in the first-stage regression is the change in nonbank market shares. The dependent variable in the second-stage regressions is the change in complaint ratios; complaint ratio is calculated as the number of mortgagerelated complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year; in column (2), the dependent variable is calculated using all mortgage-related complaints; in column (3), the dependent variable is calculated using complaints about post-origination mortgage servicing. Stress test exposure is the percentage of the county’s mortgage market that is served by banks subject to stress testsrightbeforetheinceptionofstresstests. Income per capitaisfromBEA;Unemployment rate is from BLS. All variables are defined in Table A1. ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, and ∗p < 0.1. First stage Second stage Second stage (1) (2) (3) All complaints Complaints about servicing Stress test exposure 0.126*** (0.021) ∆Nonbank market share -0.457*** -0.433*** (0.205) (0.196) ∆log(Income per capita) -0.113** -0.194*** -0.186*** (0.054) (0.064) (0.061) ∆Unemployment rate -0.191 -0.155 -0.153 (0.231) (0.287) (0.275) Observations 770 770 770 F-stat 35.11 46

Table 6: IV analysis using stress tests: Excluding counties with ex ante exposure to banks that have become a participant in stress tests after 2013 This table presents the results of a regression model in which we regress the change in county-level complaint ratios on the change in nonbanks’ market shares in the county from 2011 to 2017, which is instrumented by the county’s ex ante exposure to banks subject to stress tests. The counties that have ex ante exposure to banks that are newly added to the stress test list since 2014 are excluded from the sample. The dependent variable in the first-stage regression is the change in nonbank market shares. The dependent variable in the second-stage regressions is the change in complaint ratios; in columns (2) and (3), the change is computed during our entire sample period (i.e., between 2012 and 2018); in column (4), the change is computed for more recent years (i.e., between 2014 and 2018). Complaint ratio is calculated as the number of mortgage-related complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year; in column (2), the dependent variable is calculated using all mortgage-related complaints; in columns (3) and (4), the dependent variable is calculated using complaints about post-origination mortgage servicing. Stress test exposure is the percentage of the county’s mortgage market that is served by banks subject to stress tests right before the inception of stress tests. Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. First stage Second stage Second stage Second stage (1) (2) (3) (4) All complaints Complaints Complaints about servicing about servicing Stress test exposure 0.268*** (0.043) ∆Nonbank market share -0.379*** -0.356*** -0.630*** (0.135) (0.129) (0.223) ∆log(Income per capita) 0.166* 0.082 0.100 -0.005 (0.095) (0.086) (0.081) (0.089) ∆Unemployment rate -0.086 0.336 0.401 0.270 (0.315) (0.270) (0.257) (0.388) Observations 283 283 283 324 F-stat 39.46 47

Table 7: IV analysis using mortgage industry surety bond requirements This table presents the results of a regression model in which we regress the change in county-level complaint ratios from 2012 to 2018 on the change in nonbanks’ market shares in the county from 2011 to 2017, which is instrumented by the log of the dollar amount of mortgage broker bond required at the state level. The dependent variable in the firststage regression is the change in nonbank market shares. The dependent variable in the second-stage regressions is the change in complaint ratios; complaint ratio is calculated as the number of mortgage-related complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year; In column (2), the dependent variable is calculated using all mortgage-related complaints; in column (3), the dependent variable is calculated using complaints about post-origination mortgage servicing. Surety bond requirement is the average mortgage industry bond requirement in the state (dollars in thousands). Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. First stage Second stage Second stage (1) (2) (3) All complaints Complaints about servicing Surety bond requirement -0.012*** (0.003) ∆Nonbank market share -0.528** -0.545*** (0.207) (0.204) ∆log(Income per capita) -0.259*** -0.194** -0.189** (0.066) (0.076) (0.074) ∆Unemployment rate -0.848*** -0.014 0.002 (0.229) (0.266) (0.262) Observations 536 536 536 F-stat 14.11 48

Table 8: Nonbanks’ market share and mortgage-related complaints against nonbanks This table presents the OLS (columns (1) and (4)) and the second-stage of IV estimates (columns (2)-(3) and (5)-(6)) from the regression of the county-level complaint ratio of nonbanksonnonbanks’marketshareinthecounty. Columns(1)-(3)includeallmortgage-related complaints; columns (4)-(6) include complaints about post-origination mortgage servicing. In the IV analyses, all variables are within-county changes during our sample period; the changeinnonbanks’marketsharesisinstrumentedbythecounty’sexanteexposuretobanks subject to stress tests (columns (2) and (5)) and by the log of the dollar amount of mortgage broker surety bond required at the state level (columns (3) and (6)). The dependent variable in columns (1) and (4) is complaint ratio, which is the number of mortgage-related complaintsfiledagainstnonbanksinayearscaledbyallmortgagesoriginatedbynonbanksin the previous five years. The dependent variable in columns (2)-(3) and (5)-(6) is the change in complaint ratios. Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. All complaints Complaints about servicing OLS Stress Surety OLS Stress Surety Test IV Bond IV Test IV Bond IV (1) (2) (3) (4) (5) (6) Nonbank market share -0.507** -0.512** (0.254) (0.257) ∆Nonbank market share -1.925*** -2.578** -1.797*** -2.793** (0.557) (1.173) (0.542) (1.186) log(Income per capita) -0.581 -0.383 (0.427) (0.411) ∆log(Income per capita) -0.318 -0.856* -0.230 -0.857* (0.269) (0.437) (0.262) (0.442) Unemployment rate 2.823* 2.737* (1.468) (1.450) ∆Unemployment rate 0.019 -1.156 -0.183 -1.484 (1.069) (1.459) (1.041) (1.476) Year FE Yes No No Yes No No County FE Yes No No Yes No No Observations 7,178 725 550 7,178 725 550 49

Table 9: Borrower income of nonbanks This table presents OLS estimates from the regression of borrower income of nonbanks on nonbanks’ market share in the county. The dependent variable in column (1) is Average Income, which is the average income of nonbanks’ borrowers in the county; the dependent variable in column (2) is Income dispersion, which is the interquartile range for the incomes of nonbanks’ borrowers in the county scaled by Average Income. All variables are defined in Table A1. ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Average income Income dispersion (1) (2) Nonbank market share -0.069* -0.070** (0.041) (0.030) Year FE Yes Yes County FE Yes Yes Observations 7,185 7,178 R-squared 0.81 0.35 50

Table 10: Detailed complaint issues against nonbanks This table presents the OLS estimates from the regression of the county-level complaint ratio on nonbanks’ market share in the county. The dependent variable is Complaint ratio, which is the number of mortgage-related complaints filed in a year scaled by all mortgages originated by nonbanks in the previous five years. Column (1)includes complaints about “trouble during payment process;” column (2) includes complaints about “loan modification, collection, foreclosure” and “struggling to pay mortgage;” column (3) includes complaints about “loan servicing, payments, escrow account;” column (4)includes complaints about “settlement process and costs.” Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio (1) (2) (3) (4) Payment Payment Escrow Settlement Difficulty Processing Account Related Complaints Complaints Complaints Complaints Nonbank market share -0.352* -0.106** -0.040 -0.013 (0.212) (0.046) (0.124) (0.029) log(Income per capita) -0.068 -0.188 -0.152 0.024 (0.279) (0.118) (0.192) (0.039) Unemployment rate 1.958 -0.624** 1.161 0.242 (1.215) (0.246) (0.710) (0.218) Year FE Yes Yes Yes Yes County FE Yes Yes Yes Yes Observations 7,178 7,178 7,178 7,178 R-squared 0.398 0.403 0.370 0.301 51

Table 11: Nonbank concentration and mortgage-related complaints This table presents the results from the regression of the complaint ratio on nonbank market concentration. The dependent variable in the first-stage regression is Nonbank HHI (i.e., the market concentration within nonbanks), which is the summation of the squares of the market share of each nonbank within the nonbank mortgage market in the county. The dependent variable in the second-stage regressions is Complaint ratio, which is the number of mortgage-related complaints filed against nonbanks in a year scaled by all mortgages originated by nonbanks in the previous five years. The dependent variable in column (2) is calculated using all mortgage-related complaints; the dependent variable in column (3) is calculated using complaints about post-origination mortgage servicing. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. First stage Second stage Second stage (1) (2) (3) All complaints Complaints about servicing Nonbank market share 0.166*** (0.050) Nonbank HHI -3.090* -3.128* (1.814) (1.841) log(Income per capita) 0.124*** -0.171 0.031 (0.042) (0.436) (0.438) Unemployment rate -0.237 2.246 2.152 (0.206) (1.570) (1.535) Year FE Yes Yes Yes County FE Yes Yes Yes Observations 7,174 7,174 7,174 F-stat 11.20 52

Table 12: Nonbanks’ national market share and mortgage-related complaints This table presents the results from the regression of the complaint ratio on the national market share of the nonbanks that originate loans in the county. The dependent variable in thefirst-stageregressionisNational market share, whichisthesummationofnationalmarket share of all nonbanks that originate loans in the county. Each nonbank’s national market share is calculated as the amount of mortgages originated by the nonbank in the previous year scaled by the total amount of mortgages originated by all lenders in the nation in the previous year. The dependent variable in the second-stage regressions is Complaint ratio, which is the number of mortgage-related complaints filed against nonbanks in a year scaled by all mortgages originated by nonbanks in the previous five years. The dependent variable in column (2) is calculated using all mortgage-related complaints; the dependent variable in column (3) is calculated using complaints about post-origination mortgage servicing. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. First stage Second stage Second stage (1) (2) (3) All complaints Complaints about servicing Nonbank market share 0.069*** (0.007) National market share -7.381** -7.463** (3.718) (3.785) log(Income per capita) 0.082*** 0.026 0.230 (0.016) (0.459) (0.467) Unemployment rate -0.278*** 0.770 0.661 (0.053) (1.781) (1.717) Year FE Yes Yes Yes County FE Yes Yes Yes Observations 7,178 7,178 7,178 F-stat 84.90 53

Table 13: Nonbanks’ demand for technology-related skills: Evidence from job posting data This table presents the OLS estimates from the regression of the aggregate number of job postings that require technology-related skills by nonbanks in a county on nonbanks’ market share in the county. The dependent variable is log(1+number of job postings with technologyrelated skills), which is the log of the total number of job postings that require technologyrelated skills by nonbanks in a county-year. Column (5) excludes Quicken Loans and Guaranteed Rate Inc., two of the largest fintech nonbanks. Income per capita is from BEA. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. log(1+number of job postings with technology-related skills) (1) (2) (3) (4) (5) Nonbank market share 1.889*** 1.513*** 0.813** 0.907** 1.133*** (0.251) (0.294) (0.400) (0.398) (0.387) log(Income per capita) 1.284** 1.264** (0.505) (0.504) Year FE No Yes Yes Yes Yes County FE No No Yes Yes Yes Observations 3,227 3,227 3,227 3,227 3,183 R-squared 0.060 0.108 0.697 0.698 0.672 54

sknab lanoitidart tsniaga stnialpmoc detaler-egagtrom dna erahs tekram ’sknabnoN :41 elbaT morf ))8(-)6( dna )4(-)2( snmuloc( setamitse VI fo segats-dnoces eht dna ))5( dna )1( snmuloc( SLO eht stneserp elbat sihT snmuloC .ytnuoc eht ni erahs tekram ’sknabnon no sknab lanoitidart fo oitar tnialpmoc level-ytnuoc eht fo noisserger eht .gnicivres egagtrom noitanigiro-tsop tuoba stnialpmoc edulcni )8(-)5( snmuloc ;stnialpmoc detaler-egagtrom lla edulcni )4(-)1( serahs tekram ’sknabnon ni egnahc eht ;doirep elpmas ruo gnirud segnahc ytnuoc-nihtiw era selbairav lla ,sesylana VI eht nI gol eht yb dna ))7(-)6( dna )3(-)2( snmuloc( stset sserts ot tcejbus sknab ot erusopxe etna xe s’ytnuoc eht yb detnemurtsni si ni elbairav tnedneped ehT .))8( dna )4( snmuloc( level etats eht ta deriuqer dnob yrtsudni egagtrom fo tnuoma rallod eht fo ni sknab lanoitidart tsniaga delfi stnialpmoc detaler-egagtrom fo rebmun eht si hcihw ,oitar tnialpmoC si )5( dna )1( snmuloc snmuloc ni elbairav tnedneped ehT .sraey evfi suoiverp eht ni sknab lanoitidart yb detanigiro segagtrom lla yb delacs raey a ;sknab lla sedulcni )8( ,)6( ,)5( dna ,)4( ,)2( ,)1( snmuloc ni elpmas ehT .soitar tnialpmoc ni egnahc ehT si )8(-)6( dna )4(-)2( ;AEB morf si atipac rep emocnI .stset sserts rof detceles reven era taht sknab sedulcni )7( dna )3( snmuloc ni elpmas eht dradnatS .1.0 < p∗ dna ,50.0 < p ∗ ∗ ,10.0 < p∗ ∗ ∗ .1A elbaT ni denfied era selbairav llA .SLB morf si etar tnemyolpmenU .level ytnuoc eht ta deretsulc era srorre gnicivres tuoba stnialpmoC stnialpmoc llA yteruS ssertS SLO yteruS ssertS SLO VI dnoB VI tseT VI dnoB VI tseT )8( )7( )6( )5( )4( )3( )2( )1( llA detsetnU llA llA llA detsetnU llA llA sknaB sknaB sknaB sknaB sknaB sknaB sknaB sknaB *282.0- *962.0erahs tekram knabnoN )051.0( )851.0( ***195.2- **243.0- ***848.1- ***117.2- **974.0- ***589.1erahs tekram knabnoN∆ )429.0( )751.0( )983.0( )289.0( )681.0( )014.0( *042.0- **782.0- )atipac rep emocnI(gol )521.0( )431.0( ***970.1- **971.0- ***486.0- ***181.1- **912.0- ***987.0- )atipac rep emocnI(gol∆ )443.0( )670.0( )881.0( )663.0( )090.0( )891.0( ***198.3 ***498.3 etar tnemyolpmenU )975.0( )706.0( 955.0 982.0- *892.1 940.0- 424.0- 738.0 etar tnemyolpmenU∆ )941.1( )103.0( )647.0( )322.1( )753.0( )787.0( oN oN oN seY oN oN oN seY EF raeY oN oN oN seY oN oN oN seY EF ytnuoC 055 527 527 871,7 055 527 527 871,7 snoitavresbO 55

Table 15: Borrower income of traditional banks This table presents the OLS estimates from the regression of borrower income of traditional banks on nonbanks’ market share in the county. The dependent variable in column (1) is Average Income, which is the mean value of incomes of traditional banks’ borrowers in the county; the dependent variable in column (2) is Income dispersion, which is the interquartile range for the incomes of traditional banks’ borrowers in the county by Average Income. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Average income Income dispersion (1) (2) Nonbank market share 0.089*** 0.027 (0.031) (0.026) Year FE Yes Yes County FE Yes Yes Observations 7,397 7,397 R-squared 0.93 0.40 56

Table 16: Nonbanks’ market share and mortgage-related complaints conditional on minority populations This table presents the OLS estimates from the regression of the county-level complaint ratio on nonbanks’ market share in the county, conditional on the percentage of minority populations. The dependent variable is Complaint ratio, which is the number of mortgage-related complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year. Columns (1)-(2) include all mortgage-related complaints; columns (3)-(4) include complaints about post-origination mortgage servicing. Hispanic population and Non-white population are from the 2010 Census files; Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio (1) (2) (3) (4) All complaints Complaints about servicing Nonbank market share 0.006 0.008 -0.007 0.004 (0.020) (0.018) (0.019) (0.017) Nonbank market share × -0.078*** -0.064*** Hispanic population (0.021) (0.020) Nonbank market share × -0.090*** -0.090*** Non-white population (0.020) (0.020) log(Income per capita) -0.055*** -0.069*** -0.046** -0.060*** (0.021) (0.021) (0.020) (0.020) Unemployment rate 0.239*** 0.250*** 0.269*** 0.272*** (0.084) (0.083) (0.080) (0.079) Year FE Yes Yes Yes Yes County FE Yes Yes Yes Yes Observations 7,178 7,178 7,178 7,178 R-squared 0.570 0.570 0.557 0.559 57

Appendix Table A1: Variable descriptions Variable name Definition and source Number of complaints The number of mortgage-related complaints in a county filed with the Consumer Financial Protection Bureau (CFPB). Number of outstanding mortgages (1) The number of federal tax filings with mortgage interest payments in a county in the previous year from theIRS,or(2)thetotalnumberofmortgagesoriginated in a county during the last five years from the Home Mortgage Disclosure Act (HMDA) dataset. Complaint ratio The number of complaints divided by the number of outstanding mortgages. Nonbank Lenders are classified into traditional and nonbanks by using the variable ”FORE” in the Avery file. Lenders are defined as nonbanks if FORE is ”IMB”, and as traditional banks if FORE is one of the followings: FSB, SAL, SSB, CPB, BHC, NAT, NMB, SMB. Nonbank market share Thedollaramountofmortgagesoriginatedbynonbanks in a county in the previous year divided by the total dollar amount of mortgages originated by all lenders in the county in the previous year. All data used in this computation are from the HMDA dataset. Income per capita County-level average income per capita from the Bureau of Economic Analysis (BEA). Unemployment rate County-level unemployment rate from the Bureau of Labor Statistics (BLS). 58

Variable name Definition and source Stress test exposure The percentage of the county’s mortgage market that is served by banks subject to stress tests right before the inception of stress tests. Stresstestexposure = c,t (cid:80) mortgage ( i∈stresstestbanks i,c,t ) (6) (cid:80) mortgage i∈alllenders i,c,t Mortgage market HHI The concentration of the mortgage market in a county. (cid:88) HHI = w2 , c,t i,c,t i∈alllenders (7) mortgage i,c,t wherew = ( ) i,c,t (cid:80) mortgage i∈alllenders i,c,t Nonbank HHI The concentration of mortgage market among nonbanks. (cid:88) HHI = w2 , Nonbanks,c,t i,c,t i∈nonbanks (8) mortgage i,c,t wherew = ( ) i,c,t (cid:80) mortgage i∈nonbanks i,c,t National market share Theamountofmortgagesoriginatedbynonbanksinthe previous year scaled by the total amount of mortgages originated in the nation in the previous year. 59

Variable name Definition and source Average income (Borrowers of nonbanks) Borrowers’ average income for nonbanks in a county. Averageincome = nonbanks,c,t (cid:80) income ( b∈nonbankborrowers b,c,t ) (9) #of nonbankborrowersincountycinyeart Income dispersion (Borrowers of nonbanks) The interquartile range of the income distribution for the nonbanks’ borrowers divided by Averageincome . nonbanks,c,t Average income (Borrowers of traditional banks) Borrowers’ average income for traditional banks in a county. Averageincome = banks,c,t (cid:80) income ( b∈bankborrowers b,c,t ) (10) #of bankborrowersincountycinyeart Income dispersion (Borrowers of traditional banks) The interquartile range of the income distribution for the traditional banks’ borrowers divided by Averageincome . banks,c,t Hispanic population The percentage of the Hispanic population in a county from the 2010 Census files. Non-white population The percentage of the non-white population in a county from the 2010 Census files. 60

Table A2: Nonbanks’ market share and mortgage-related complaints: Using an alternative measure of outstanding mortgages in computing the complaint ratio This table presents the OLS estimates from the regression of the county-level complaint ratio onnonbanks’marketshareinthecounty. ThedependentvariableisComplaint ratio,whichis the number of mortgage-related complaints filed in a year scaled by all mortgages originated in the previous five years. Column (1) includes all mortgage-related complaints; column (2) includes complaints about post-origination mortgage servicing; column (3) includes complaints about mortgage originations. Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio (1) (2) (3) All complaints Complaints about Complaints about servicing originations Nonbank market share -0.295*** -0.309*** 0.014 (0.106) (0.103) (0.029) log(Income per capita) -0.348*** -0.291*** -0.058** (0.117) (0.111) (0.026) Unemployment rate 3.035*** 3.033*** 0.002 (0.522) (0.512) (0.117) Year FE Yes Yes Yes County FE Yes Yes Yes Observations 7,178 7,178 7,178 R-squared 0.638 0.623 0.390 61

Table A3: Nonbanks’ market share and mortgage-related complaints: Using the number of loans instead of the dollar amount of loans in computing nonbanks’ market share This table presents the OLS estimates from the regression of the county-level complaint ratio on nonbanks’ market share in the county. The dependent variable is Complaint ratio, which is the number of mortgage-related complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year. Column (1) includes all mortgage-related complaints; column (2) includes complaints about postorigination mortgage servicing; column (3) includes complaints about mortgage originations. Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio (1) (2) (3) All complaints Complaints about Complaints about servicing originations Nonbank market share -0.051*** -0.054*** 0.003 (0.016) (0.016) (0.005) log(Income per capita) -0.053** -0.043** -0.010* (0.021) (0.020) (0.005) Unemployment rate 0.282*** 0.306*** -0.023 (0.086) (0.082) (0.027) Year FE Yes Yes Yes County FE Yes Yes Yes Observations 7,178 7,178 7,178 R-squared 0.569 0.557 0.353 62

Table A4: Nonbanks’ market share and mortgage-related complaints: Excluding Quicken Loans and Top 3 nonbanks This table presents the OLS estimates from the regression of the county-level complaint ratio onnonbanks’marketshareinthecounty. ThedependentvariableisComplaint ratio,whichis the number of mortgage-related complaints filed in a year scaled by all mortgages originated in the previous five years. Columns (1)-(3) exclude complaints filed against Quicken Loans; columns (5)-(6) exclude complaints filed against the top 3 nonbanks (i.e., Quicken Loans, United Shore Financial Service, Guaranteed Rate Inc.). The main independent variable is Nonbank market share, which is the amount of mortgages originated by nonbanks in a county scaled by the total amount of mortgages originated by all lenders in the county in the previous five years, excluding loans originated by Quicken Loans (columns (1)-(3)) or excluding loans originated by the top 3 nonbanks (i.e., Quicken Loans, United Shore Financial Service, Guaranteed Rate Inc) (columns (5)-(6)). Columns (1) and (4) include all mortgage-related complaints; columns (2) and (5) include complaints about post-origination mortgage servicing; columns (3) and (6) include complaints about mortgage originations Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio Excluding Quicken Excluding Top 3 (1) (2) (3) (4) (5) (6) All Servicing Originations All Servicing Originations Nonbank market share -0.272** -0.279*** 0.007 -0.281** -0.285*** 0.004 (0.110) (0.107) (0.028) (0.111) (0.107) (0.028) log(Income per capita) -0.346*** -0.289*** -0.057** -0.357*** -0.298*** -0.059** (0.117) (0.111) (0.026) (0.118) (0.112) (0.026) Unemployment rate 3.003*** 3.017*** -0.008 2.958*** 2.979*** -0.015 (0.524) (0.515) (0.116) (0.522) (0.514) (0.117) Year FE Yes No No Yes No No County FE Yes No No Yes No No Observations 7,169 7,169 7,169 7,167 7,167 7,167 R-squared 0.639 0.624 0.390 0.640 0.625 0.390 63

Table A5: Nonbanks’ market share and mortgage-related complaints: Post-2012 This table presents the OLS estimates from the regression of the county-level complaint ratio on nonbanks’ market share in the county. The dependent variable is Complaint ratio, which is the number of mortgage-related complaints filed in a year scaled by the number of federal tax filings with mortgage interest payments in the county in the previous year. Column (1) includes all mortgage-related complaints; column (2) includes complaints about postorigination mortgage servicing; column (3) includes complaints about mortgage originations. Income per capita is from BEA; Unemployment rate is from BLS. All variables are defined in Table A1. ∗∗∗p < 0.01, ∗∗p < 0.05, and ∗p < 0.1. Standard errors are clustered at the county level. Complaint ratio (1) (2) (3) All complaints Complaints about Complaints about servicing originations Nonbank market share -0.056*** -0.061*** 0.006 (0.020) (0.020) (0.007) log(Income per capita) -0.057** -0.045* -0.011 (0.026) (0.025) (0.007) Unemployment rate 0.337*** 0.355*** -0.019 (0.117) (0.112) (0.032) Year FE Yes Yes Yes County FE Yes Yes Yes Observations 6,219 6,219 6,219 R-squared 0.569 0.557 0.369 64