Full text of Working Paper (Federal Reserve Bank of Cleveland) : Mortgage Companies and Regulatory Arbitrage, Working Paper 12-20R : Working Paper 2012-20R4

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

w o r k i n g
p

12 20R

Mortgage Companies and
Regulatory Arbitrage
Yuliya Demyanyk and Elena Loutskina

FEDERAL RESERVE BANK OF CLEVELAND

Working papers of the Federal Reserve Bank of Cleveland are preliminary materials circulated to
stimulate discussion and critical comment on research in progress. They may not have been subject to the
formal editorial review accorded official Federal Reserve Bank of Cleveland publications. The views stated
herein are those of the authors and are not necessarily those of the Federal Reserve Bank of Cleveland or of
the Board of Governors of the Federal Reserve System.
Working papers are available on the Cleveland Fed’s website at:

www.clevelandfed.org/research.

Working Paper 12-20R

October 2015*

Mortgage Companies and Regulatory Arbitrage
Yuliya Demyanyk and Elena Loutskina

Mortgage companies (MCs) do not fall under the strict regulatory regime of
depository institutions.We empirically show that this gap resulted in regulatory
arbitrage and allowed bank holding companies (BHCs) to circumvent consumer
compliance regulations, mitigate capital requirements, and reduce exposure to
loan-related losses. Compared to bank subsidiaries, MC subsidiaries of BHCs
originated riskier mortgages to borrowers with lower credit scores, lower incomes, higher loan-to-income ratios, and higher default rates. Our results imply
that precrisis regulations had the capacity to mitigate the deterioration of lending
standards if consistently applied and enforced for all types of intermediaries.

Keywords: Banking Regulation, Regulatory Arbitrage, Shadow Banking, Lending Standards, Mortgage, Foreclosure, Bank, Crisis.

JEL Codes: G21, G28, D12.

*First version: September 2012. Earlier revisions November 2012, May 2013,
and April 2014.
Yuliya Demyanyk is at the Federal Reserve Bank of Cleveland, and she can be
reached at ydemyanyk@clev.frb.org. Elena Loutskina is at the University of
Virginia, and she can be reached at eloutskina@virginia.edu. The authors thank
Jennifer Bray, Mark Dobbins, Tom Fitzpatrick, Mark Greenlee, Chris Henderson,
Paul Kaboth, and James Thomson for sharing their knowledge about supervision
and regulation of banks and mortgage companies. They thank Sumit Agarwal,
Scott Frame, John Krainer, Marc Lipson, Atif Mian, Amit Seru, Philip Strahan,
Daniel Streitz, the participants of American Economic Association Meetings, the
Federal Reserve System Committee on Financial Structure and Regulation Meetings, Financial Intermediation Research Society conference, and the seminar
participants at the University of Virginia, the Federal Reserve Bank of Cleveland,
Georgetown University, Hong Kong University of Science and Technology, University of Hong Kong, and Chinese University of Hong Kong for useful discussions and comments.

Introduction

The collapse of the housing market in 2007 initiated an economic downturn with a profound impact
on the world economy. The securitization market, “shadow banking,” and inadequate regulation are
widely blamed for the deterioration of lending standards and, ultimately, the crisis.1 While we know
about regulatory inconsistencies (Agarwal, Lucca, Seru, and Trebbi (2012)), there is little direct
empirical evidence on how regulation aﬀects the behavior of “shadow banking” credit intermediaries
(Keys, Mukherjee, Seru, and Vig (2009) and Acharya, Schnabl, and Suarez (2011)). This paper
shows that a regulatory gap between depository credit intermediaries (banks) and nondepository
credit intermediaries (mortgage companies—MCs) altered the behavior of even regulated lenders
and contributed to the deterioration of underwriting standards in the mortgage market. Our results
suggest that regulations prior to 2007 had the capacity to mitigate the deterioration of precrisis
lending standards, but only if applied and enforced similarly across all lenders.
Depository institutions are subject to safety and soundness regulations, deposit insurance requirements, and consumer compliance regulations, among others. These regulations are designed
to curb banks’ risk-taking behavior stemming from underpriced deposit insurance.2 The seminal
banking literature, however, questions the necessity of regulating credit intermediaries that originate loans to distribute (OTD). Origination of risky loans requires soft information production, and
thus should be dominated by banks which are better ex ante screeners (Leland and Pyle (1977),
Boyd and Prescott (1986)) and more eﬃcient ex post monitors (Diamond (1984)). OTD lenders,
such as MCs, should predominantly originate loans based on hard information that they can pass
to the secondary market investors (Rajan, Seru, and Vig (2009), Loutskina and Strahan (2011),
and Gorton and Pennacchi (1995)). By extension, MCs should be “market-regulated” to maintain
high underwriting standards.
MCs are eﬀectively market-regulated. They fall under a signiﬁcantly smaller set of regulations as compared to banks. Even MC subsidiaries of heavily regulated bank holding companies
(BHCs) enjoy weak regulations and lax enforcement of them. To isolate and evaluate the impact of
1

See, e.g., Mian and Suﬁ (2009), Keys, Mukherjee, Seru, and Vig (2010), Mian and Suﬁ (2010), and Purnanandam
(2011) for the evidence of secondary loan market impact on the deteriorating lending standards. Agarwal, Ambrose,
and Yildirim (2010), Demyanyk and Van Hemert (2011), and Mayer and Pence (2008) document the signiﬁcant role
of subprime lending in the 2007 crisis.
2
Flannery (2007) argues that “left to themselves, banks would accept too large a default probability, so supervisors
design constraints to increase bank safety.” The regulations should be binding to ensure their eﬀectiveness.

the regulatory gap, we examine the behavior of BHCs that lend through both aﬃliated depository
institutions (ADIs) and aﬃliated MCs (AMCs). The within-BHC analysis allows us to nonparametrically control for parent-speciﬁc heterogeneity such as access to securitization markets, economies
of scale or scope in loan sales, loan inventory management, risk aversion, etc. The core assumption
underlying our empirical strategy is that, absent the regulatory diﬀerences, all the loans originated
and securitized through AMCs could have been originated and securitized through ADIs.3
Using the within-BHC strategy, we empirically evaluate the implications of three diﬀerences
in the regulation of AMCs and ADIs. First, as detailed in Section 2 of the paper, the safety and
soundness regulations require banks to hold capital even for loans they are planning to sell. In
contrast, MCs have no explicit capital requirements as they do not fall under banking regulations.
BHCs can also avoid consolidating AMCs for capital requirement purposes. As a result, by lending
through their AMCs, BHCs can conserve their capital.
Second, ADIs have to recognize loan impairments as soon as they occur; the performance
of their loan portfolios aﬀects the parent BHCs’ capital requirements, loan-loss provisions, and
the price of deposit insurance. AMCs, on the other hand, are guided by Generally Accepted
Accounting Principles (GAAP) and have a lot of ﬂexibility in recognizing losses—they can “sit” on
nonperforming loans in the expectation of working them out or selling them to a special “scratchand-dent desk” entity. Moreover, AMCs are structured as limited liability entities, thus keeping
a BHC’ exposure to an AMC’s lending activities limited to its equity investment. As a result, a
parent BHC does not have to provision for or recognize losses from AMC loan portfolios to the full
extent.
Third, while banks are subject to consumer compliance regulations that make origination of
inferior-quality loans costly, these regulations are not enforced for MCs. Thus, in general, MCs can
originate lower-quality noncompliant loans that banks cannot, and in particular, BHCs can have
inferior underwriting standards at their AMCs.4
3
To better understand our empirical strategy, consider a world with a regulatory regime that is uniform across
MCs and DIs. In this world, the loans originated by AMCs have to be fully consolidated, the capital provisioned for,
checked for consumer compliance, etc. In other words, AMC loans would put the same strain on a BHC’s capital
requirements or risk-management needs as ADI loans. Banning regulatory diﬀerences, we thus, question the need for
BHCs to establish separate legal entities such as MCs.
4
Appendix A presents a story of HSBC Finance, formerly well-known subprime lender, Household Finance Corporation. Acquired by HSBC Group in 2003, HSBC Finance ramped up its subprime lending to become number one
player in this market segment in 2006. The case succinctly illustrates the implications of regulatory gap explored in
this paper.

In this paper, we empirically evaluate whether lending through AMCs allowed BHCs to mitigate
capital requirements, limit exposure to mortgage-related losses, and engage in riskier lending. In
our ﬁrst set of tests we ﬁnd that undercapitalized BHCs are more likely to establish an AMC or
to increase lending through existing AMCs. These relationships cease to exist during the period of
regulatory uncertainty when capital arbitrage was almost eliminated by regulators. The evidence
suggests that by lending through AMCs as opposed to lending through ADIs, BHCs can circumvent
their capital requirements.
We then document that BHC mortgage-related losses are positively related to the extent of
OTD lending by ADIs, but unrelated to the extent of OTD lending by AMCs. To show this, we
exploit the 2007 shutdown of the secondary loan market that led to signiﬁcant amounts of already
originated-to-distribute and rapidly deteriorating loans accumulating on lenders’ balance sheets
(Purnanandam (2011)). Our results are consistent with BHCs being able to avoid loan losses from
their AMC loan portfolios.
Using a ZIP code–level representative borrower framework, we show that AMC borrowers tend
to have lower credit scores, higher loan-to-income ratios, and lower incomes as compared to ADI
borrowers.5 We further conﬁrm the ZIP code–level evidence using a sample of about 1.4 million
individual loans originated in 2005-2006 by ADIs and AMCs of BHCs. Using the loan-level data,
we also ﬁnd that AMCs tend to originate riskier types of loans such as adjustable-rate and interestonly mortgages. Furthermore, AMC borrowers are 5% more likely than ADI borrowers to become
delinquent or default on their loans within two years of origination. This 5% default diﬀerential
is economically signiﬁcant given the average delinquency and default rate of 9% in our loan-level
sample. These ZIP code-level and loan-level results are consistent with precrisis AMC lending
standards being inferior to those of ADIs. We oﬀer a set of robustness tests to illustrate that our
conclusions cannot be explained by (i) diﬀerences in the extent to which AMCs and ADIs participate
in the securitization market (GSE or privately securitized loans, subprime loans, etc.); (ii) ADIs
serving a wider set of consumer ﬁnance market segments as compared to AMCs; (iii) potential
diﬀerences in AMC and ADI considerations regarding a parent BHC’s reputation.
Finally, we illustrate that our results are not speciﬁc to the sample of MCs that are aﬃliated with
5

Similarly, Carey, Post, and Sharpe (1998) document that ﬁnance companies tend to lend to riskier and more
leveraged corporate borrowers.

BHCs. Speciﬁcally, we show that a one-standard-deviation increase in 2005-2006 MC market share
(AMCs and independent MCs—IMCs—combined) is associated with a 1.4% (1.2%) increase in ZIP
code–level mortgage delinquency (foreclosure) rates. The economic eﬀect is similar in magnitude
to that of subprime lending.
Our paper sheds light on a number of important issues that played a role in the 2007 credit
crisis: the role of the “shadow banking” system, the inadequate regulation thereof (see, e.g., Gorton
and Metrick (2010)), and associated risk-taking in the ﬁnancial sector. First, we contribute to the
literature on the behavior of the largest shadow banking intermediaries. MCs have consistently
originated about half of the mortgages in the U.S. economy since 1992 (Figure 1). In 2006, they
originated $1.35 trillion in mortgages, relative to the $1.12 trillion originated by banks. MCs continued to originate more than 30% of U.S. mortgages through 2010. In this line of research, Keys,
Mukherjee, Seru, and Vig (2009) is the closest study to ours. It documents that heavily regulated banks securitized lower-quality subprime loans than weakly regulated MCs. We complement
these ﬁndings by oﬀering an economic rationale behind the origination, rather than securitization,
decisions of these intermediaries.
Second, by documenting that BHCs change their organizational structure to accommodate risky
lending strategies, we add to the literature linking a bank’s organizational form and its function.
While prior studies mainly focused on the eﬀect of corporate structure on an intermediary’s ability
to produce and exploit soft information (Berger, Miller, Petersen, Rajan, and Stein (2005)), we
evaluate how regulatory diﬀerences and associated heterogeneity across aﬃliates allow BHCs to
compartmentalize risk within an organization in pursuit of riskier lending strategies. Our results
suggest that this reallocation of risk within an organization contributed to risk build-up in the
economy. We thus complement studies exploring how interconnectedness and risk sharing across
intermediaries contribute to systemic risk (Acharya, Pedersen, Philippon, and Richardson (2010);
Adrian and Brunnermeier (2008) and Acharya, Iyer, and Sundaram (2015)).
Finally, our study contributes to the literature exploring regulatory design and its consequences.
The international banking literature documents variation in regulation across countries and links
it to banks’ risk-taking behavior (Barth, Caprio, and Levine (2004); Ongena, Popov, and Udell
(Forthcoming)). U.S. studies acknowledge the existence of inconsistent regulatory enforcement
and oversight (Kane (2000) and Calomiris (2006)) and evaluate their implications (Rosen (2003)).
4

Agarwal, Lucca, Seru, and Trebbi (2012) document discrepancies in federal and state regulators’
enforcement actions in supervising the exact same banks. We identify a regulatory gap between
MCs and banks and document its consequences for both ﬁnancial institutions and consumers.
Understanding the economic rationale for such ﬁnancial intermediaries’ behavior is a ﬁrst step to
eﬃcient regulatory design.

Mortgage Companies and Financial Regulation

Historically, ﬁnancial regulation in the U.S. was implemented through two types of regulators: the
“institutional regulators” and the “functional regulators.” The institutional regulators oversaw the
activities of depository institutions and ensured their compliance with safety and soundness regulations and consumer compliance regulations, among others.6 The functional regulators oversaw
all entities engaged in respective functional activities. For example, the Securities and Exchange
Commission (SEC) enforced the U.S. securities laws.7
Despite the fact that MCs held a 50% market share since the early 1990s (Figure 1), these
ﬁnancial institutions were largely ignored by the fragmented U.S. regulatory system for two reasons:
(i) MCs did not hold deposits and, hence, did not need a charter from an institutional regulator,
and (ii) the MCs’ activities did not fall under the domains of the functional regulators. The only
authorities that had a legal standing to oversee MCs were the Federal Trade Commission and State
Attorney Generals. Both had the ability to bring punitive actions against MCs, but only if they
observed unfair and deceptive practices evidenced by a pattern of customer complaints (Engel and
McCoy (2011)). Before the 2007 crisis, this “repeat complaint-oriented supervision” had little power
to systematically aﬀect MC practices, thus leaving MCs essentially free of a regulatory burden.
One might reasonably argue that AMCs of BHCs should have fallen under the oversight of the
BHCs’ regulators. After all, they were subsidiaries that, as we now know, presented a signiﬁcant
threat to their parents, and ultimately to their depositors as well. Below, we discuss the history
and philosophy of the banking regulation that allowed AMCs to fall between the cracks of ﬁnancial
6

Diﬀerent types of the depository institutions are supervised by Oﬃce of the Comptroller of the Currency, Federal
Deposit Insurance Corporation (FDIC), the Federal Reserve System, and the Oﬃce of Thrift Supervision.
7
The most prominent functional regulators are the SEC, established in 1934 to enforce the U.S. securities law; the
Commodity Futures Trading Commission, established in 1974 to oversee futures and options markets; and the state
insurance regulators.

regulations.8

Safety and Soundness Regulation
Since the Glass-Steagall Act of 1934, U.S. depository institutions (banks, thrifts, credit unions,
and savings and loans—collectively labeled “banks” in this section) were closely supervised to
protect depositors. Underpriced deposit insurance created incentives for banks to “accept too large
of a default probability” (Flannery (2007)). Safety and soundness regulations were designed to
curb such risk-taking behavior and protect the federal safety net of the FDIC. Supervision was,
and still is, done through periodic site visits and examination of banks’ Reports of Condition and
Income (Call Reports). Initially, the safety and soundness supervision aimed simply at avoiding
bank failures. Over time, the supervision expanded to evaluation of banks’ risks, such as ﬁnancial,
interest rate, credit, liquidity, and others. Regulators’ assessment of these risks ultimately formed
bank risk ratings (the CAMELS ratings).9
Under the Bank Holding Company Act of 1956, the safety and soundness standards only applied
to individual banks rather than BHCs. Concerns that BHCs could rapidly shift resources at the
expense of some depository subsidiaries led the Federal Reserve to advocate the “source of strength”
doctrine under which a BHC must assist its troubled depository aﬃliates before the failure is
imminent. The Financial Institutions Reform Recovery and Enforcement Act of 1989 (FIRREA)
codiﬁed this doctrine, while the Federal Deposit Insurance Improvement Act of 1991 extended it by
requiring a parent BHC to take corrective action when an insured depository subsidiary failed to
meet required levels of capitalization (“prompt corrective action provisions”).10 The sole guiding
principle of this “umbrella supervision” was to protect depositors and the federal safety net of the
FDIC (Lee (2012) and Greenlee (2008)). The legislation remained silent regarding the safety and
soundness standards for non-depository subsidiaries of BHCs.
The Gramm–Leach–Bliley Act (GLBA) of 1999 codiﬁed the lack of regulatory oversight over
8

The discussion below is partially based on our extensive interviews with bank supervisors and examiners. We
do not provide a detailed overview of U.S. banking regulation, but rather discuss unique features of it relevant to
this study. Agarwal, Lucca, Seru, and Trebbi (2012) present a detailed discussion of the fragmented and inconsistent
nature of the U.S. regulatory system.
9
Rosen (2003) describes the safety and soundness regulations and CAMELS ratings in detail.
10
Ashcraft (2008) discusses in detail the economic implications of the doctrine and associated legislature. This
study documents a signiﬁcant improvement in the aﬃliate banks’ safety and soundness upon the passage of this
reform.

the MCs. GLBA dismantled the legal barriers to conglomeration in the ﬁnancial services industry
and expanded the span of allowable nonbank activities for BHCs. Concerned with inability of bank
examiners to understand the new lines of business, GLBA deferred the examination and oversight
of nonbank subsidiaries to their respective functional regulators.11 The problem was that MCs had
never been assigned a functional regulator. The BHC regulators were granted visitorial powers,
advisory in nature, to evaluate nonbank subsidiaries, but only if the latter constituted a serious
risk to the ﬁnancial safety and soundness of a depository institution. The risk status was to be
determined by the respective functional regulators, which did not exist for MCs.12
One major implication of this regulatory gap was MCs having no explicitly imposed or enforced
capital requirements. Supervisors routinely require banks and BHCs to meet minimum capital
requirements. Even loans originated by banks with the sole purpose of distribution are subject to
sizable capital requirements.13,14 BHCs were able to avoid capital provisioning for loans held on
AMCs’ balance sheets. For example, it was suﬃcient for a BHC to hold a 79% or less equity stake in
an MC subsidiary to avoid consolidating it for capital requirement purposes.15 Anecdotal evidence
also suggests that even fully owned and controlled AMCs were not always capital provisioned for.
Regulators were content with not allocating capital for loans held (i) for sale; (ii) oﬀ banks’ balance
sheets; (iii) in limited liability nondepository subsidiaries.16 Thus, AMCs could take over loan
origination from ADIs and help their depository aﬃliates and parent BHCs to conserve capital.
The ﬁrst part of our analysis explores whether less-capitalized BHCs were more likely to establish
AMCs.
Legislators were aware of the risks nonbank subsidiaries presented for their depository aﬃliates.
They established formal ﬁnancial barriers between bank and nonbank aﬃliates. Sections 23A and
11

For example, the Federal Reserve System: Purposes and Functions (2005) explicitly states that “nonbank subsidiaries ... are supervised and regulated by their appropriate functional regulators.”
12
See Mandanis Schooner (2002) and Engel and McCoy (2011).
13
For example, consider a bank that annually originates OTD mortgages in the amount of 50% of its total assets
(75th percentile in our sample). Under an optimistic 90-day holding period for these loans (Engel and McCoy (2011)),
this bank would have to hold a rolling inventory of mortgages amounting to 12.5% of its total assets (50% × 90/360).
With 8% risk weight of mortgages in the capital requirements, these “temporary” holdings would lead to 1% higher
capital requirements, which is signiﬁcant compared to the average bank equity capital (Table 2).
14
Note that our arguments are valid irrespective of whether OTD loans are fully sold or are swapped for highly
rated tranches of RMBS securities (Isil, Nadauld, and Stulz (2012)). Both mechanisms provide capital relief to a
parent BHC, but require time to implement.
15
BHCs can still recognize proﬁts from nonconsolidated AMCs via dividend payments.
16
We have heard this claim repeatedly in our numerous interviews with bank regulators, examiners, and legal
scholars.

23B of the Glass-Steagall Act explicitly prohibited any transactions between bank and nonbank
aﬃliates that could adversely aﬀect the health of a depository institution. GLBA took one step
further and required BHCs to conduct any nonbank activities “in nonbank subsidiaries that are
separately incorporated, separately capitalized, and insulated by ‘ﬁrewalls’ from their aﬃliated
banks” (see, e.g., Gouvin (2002) and Wilmarth (2002)). While the “source of strength” doctrine
and the “prompt corrective action provisions” forced BHCs to carry almost full responsibility for
their troubled depository subsidiaries, the regulators explicitly prohibited BHCs from rescuing
troubled nonbank subsidiaries. As a result, BHCs’ exposure to the limited liability AMCs was
indeed limited to the equity investment. While these regulations protected depositors, they did not
eliminate BHCs’ ability to pursue risky activities. BHCs merely had to do so through their AMCs.
BHCs could also avoid recognizing losses from AMCs’ operations. The mandatory Call Reports
guidelines require banks to recognize loan delinquencies and defaults as soon as they occur and to the
full amount of underlying loans, irrespective of the collateral values.17 MCs, being nondepository
institutions, only had to adhere to GAAP loss-reporting requirements. Speciﬁcally, MCs had to
recognize loan impairment when “it was probable that a loss has been incurred.”18 The MCs’ losses
were also recorded net of expected collateral values. As a result, MCs enjoyed a lot of ﬂexibility
in recognizing loan-related losses. AMCs, for example, had an option to delay recognizing losses
in the expectation of working loans out or selling them in the secondary market through so-called
“scratch-and-dent desks” entities.
The ability to avoid loan-loss recognition was extremely beneﬁcial for BHCs. AMCs’ loan losses,
if recognized by a parent, would be charged against BHC equity capital and, hence, aﬀect BHCs’
cost of funds (Stein (1998)). Loan impairments, such as 30-day-past-due status, increase the costs
of loan-loss provisioning and the price of the deposit insurance. In the second set of our empirical
analysis, we evaluate whether BHCs were able to avoid incurring losses from AMCs’ loan portfolios.

Consumer Compliance
Consumer compliance regulations, an extension of the safety and soundness regulations, are the
second signiﬁcant part of the banks’ regulatory oversight. Consumer compliance regulations started
17

See FFIEC’s Instructions of Preparation of Consolidated Report of Condition and Income, 2001.
See, e.g., Financial Accounting Standards Board (FASB) Accounting Standards Codiﬁcation (ASC) 310-10 and
FASB ASC 450-20.
18

with Congress passing the Truth in Lending Act in 1968. The Act requires disclosure statements
to be given to consumers about their mortgage loans. The Act was amended in 1994 with the
Home Ownership and Equity Protection Act (HOEPA) to curb subprime lending, which had started
gaining popularity at the time (Riegle–Neal Community Development and Regulatory Improvement
Act of 1994).19 One part of HOEPA requirements was legally binding for all intermediaries, but
applicable only to a small portion of the subprime market. Another part of HOEPA was more
guidance than law, and aimed to broadly prevent “unfair and deceptive lending practices.” Despite
enforcing both parts of the consumer compliance regulation for depository institutions, the Federal
Reserve under Greenspan never enacted the second part of the regulation on a broad scale (Engel
and McCoy (2011)).
Consumer compliance regulations require bank examiners to evaluate lending standards and
procedures established by depository institutions, the training of lending oﬃcers, and the adherence
to established lending practices. Supervisors go as far as evaluating a small random sample of all
originated (not only retained) loans to verify compliance. They ensure that, for example, banks
verify borrowers’ ability to repay their loans and do not approve loan applications based solely on
expectations of house price appreciation. Noncompliant banks are subject to ﬁnancial penalties
and a prolonged examination of a wider subset of originated loans. Between 2003 and 2007 the
Federal Reserve took at least 22 enforcement actions against the state member banks for persistent
violation of consumer compliance. All these actions included requirement that borrowers repayment
source and ability to repay had to be consistently documented in all future loans.
MCs were subject only to a subset of consumer compliance regulations, which were rarely, if
ever, enforced.20 Similar to examinations of AMCs for safety and soundness, BHC supervisors had
no legal standing to evaluate AMCs for consumer compliance. The January 1998 letter from the
19
The HOEPA regulation was made ineﬀective by proactive regulation avoidance and exemption lobbying done by
both depository and nondepository lenders (Engel and McCoy (2011)). Even the state HOEPA laws had minimal to
no impact on the subprime market in general and behavior of subprime lenders in particular (Bostic, Engel, McCoy,
Pennington-Cross, and Wachter (2008)). A 2001 Federal Reserve System study shows that only 5% of subprime loans
fell under HOEPA regulation. Consistently, in the unreported set of tests, we ﬁnd no diﬀerence in either mortgage
company activity or its impact on mortgage defaults when we compare states with strong mini-HOEPA laws to those
with weak/no mini-HOEPA laws. In this analysis, we exploit the index of mini-HOEPA laws restrictiveness compiled
by Bostic, Engel, McCoy, Pennington-Cross, and Wachter (2008).
20
Only one action was undertaken against an MC aﬃliate of a BHC. In 2001 CitiFinancial faced $70 million civil
money penalty for questionable lending practices in 2001. This enforcement action was motivated by Citigroups
aggressive acquisition strategy in the commercial banking domain and was considered a slap on a wrist given the
volume of CitiFinancial and Citigroup mortgage originations.

Board of Governors of the Federal Reserve System clearly states:21
“... the Board adopted a policy that the Federal Reserve will (1) not routinely conduct
consumer compliance examinations of nonbank subsidiaries of bank holding companies,
and (2) not investigate consumer complaints relating to these subsidiaries. This action
formalizes a policy regarding examinations that has been System practice all along. In
addition, with regard to complaint investigations, the action establishes a policy in an
area where the Reserve Banks have exercised their own discretion in the past. All consumer complaints against such entities received by the Federal Reserve System will now
be referred to the Federal Trade Commission and not be investigated by Reserve Banks.”
No enforcement of the consumer compliance regulations for AMCs relaxed parent BHCs’ regulatory constraints and expanded the latter’s ability to take risks. They merely had to pursue
this risky, noncompliant lending through AMCs. Our ﬁnal set of empirical tests evaluates the
risk-taking behavior of ADIs and AMCs within a BHC by comparing the ex ante and ex post loan
characteristics of two types of aﬃliates.
All told, compared to banks, MCs were subject to a smaller set of regulations that were rarely, if
ever, enforced. In this paper, we evaluate whether this regulatory gap aﬀected the behavior of even
heavily regulated BCHs and motivated them to (a) originate more loans through AMCs to mitigate
capital requirements and (b) serve riskier marginal borrowers through AMCs to circumvent the
consumer compliance regulations and avoid loan-related losses. The regulatory gap allowed BHCs
to pursue risker investment activities through AMCs, thus contributing to overall economic risk.22

Data and Sample Selection

In this study, we combine data from several sources. In the ﬁrst part of the paper, we establish
the existence of regulatory arbitrage. Here, we evaluate the BHCs’ ﬁnancial conditions from the
Call Reports. In the second part of the paper, we evaluate the lending standards of banks and
21

See http://www.federalreserve.gov/boarddocs/caletters/1998/9801/caltr9801.htm.
The lack of safety and soundness regulation in general and no capital requirements in particular contributed to
the risk-shifting behavior of, for example, Countrywide, which, despite the obvious deterioration of loan quality at
the beginning of the mortgage crisis in early 2007, did not shut down its subprime arm until well into the crisis:
November 2007. Similarly, Landier, Sraer, and Thesmar (2011) document that New Century Financial Corporation
reacted to adverse economic conditions by aggressively investing in riskier (“interest only”) loans.
22

MCs. The challenge of this analysis lies in the lack of readily available data that capture the loan
originator, the origination decision, and borrower characteristics at the loan level. We use the Home
Mortgage Disclosure Act (HMDA) data that cover a comprehensive set of loans (about 87% of all
loans originated) and lenders (99% of the sector based on total assets) in the U.S. economy, but lack
core loan-risk characteristics such as a borrower’s credit score and the value of a property. Other
data sources, e.g., data from credit bureaus, provide us with a rich set of borrower characteristics,
but lack information about loan originators.
In this study, we adopt two approaches to resolve this issue. First, we use the ZIP code–level
representative borrower approach from Mian and Suﬁ (2009). Eﬀectively, we compare representative
borrowers to whom a BHC lends mostly through its ADIs to representative borrowers to whom the
BHC lends mostly through its AMCs (BHC-ZIP-year level of observation). This approach allows
us to draw conclusions based on a very broad sample of U.S. borrowers, lenders, and markets.
However, it is based on average, rather than individual, loan and borrower characteristics and only
allows us to exploit across ZIP code variation in borrower characteristics. Second, we conﬁrm our
representative borrower results using the loan-level data. We merge HMDA, Call Reports, and
loan-level databases to build a loan-level sample for which lender types, borrower characteristics,
and loan performance are jointly available. This sample allows us to exploit the within–ZIP code
heterogeneity across borrowers. Below, we describe in detail the sample formation procedures and
merging criteria for all of the utilized data sets.

3.1

Loan Origination and Lender Identity

We obtain information on loan originations from a comprehensive sample of mortgage applications collected by the Federal Reserve under provisions of the HMDA.23 The data cover the loan
originations by 99% of depository and nondepository ﬁnancial institutions. The HMDA Loan Application Registry provides us with the mortgage origination date, the identity of the lender, the
dollar amount of the loan, the borrower income, and whether the lender retained the loan or sold
it to a third party. We only consider loans that have valid information for these attributes. We
23

HMDA was passed into law by Congress in 1975 and expanded in 1988, with the purpose of informing the public
and the regulators about whether or not ﬁnancial institutions adequately serve local credit needs. Since the onset
of the crisis, the data has been widely used in the academic research. See, e.g., Loutskina and Strahan (2009) for
detailed data description.

drop subsidized loans (e.g., those sponsored by the Veteran Administration), home equity loans,
and construction loans. We also require the census tract provided by the HMDA to map to ZIP
code and Core Based Statistical Area (CBSA) identiﬁers. This set of restrictions leaves us with
96.65 million loans originated between 1999 and 2006, out of which 54.5 million were originated by
mortgage companies and 36 million were originated by BHCs with MC aﬃliates.
We augment HMDA data with the “HMDA Lender File” compiled by Robert Avery from the
Board of Governors of the Federal Reserve System. The lender ﬁle provides a variety of identifying
information for all lenders who have ever ﬁled a HMDA report. It allows us to identify depository
and nondepository lending institutions and match them to their respective parent BHCs. We
exclude about 1.5% of the lenders (0.001% of loans) where we cannot clearly identify if the originator
was a depository or nondepository institution. The resulting sample covers 15,280 distinct lenders
of which 10,904 are depository institutions.
Table 1 Panel A provides summary statistics of HMDA loan and borrower characteristics for four
types of lenders: independent depository institutions (IDIs), independent MCs (IMCs), aﬃliated
depository institutions (ADIs), and aﬃliated MCs (AMCs). Consistent with our premises, we
observe that MC borrowers are characterized by higher loan-to-income ratios and lower absolute
and relative incomes as compared to bank borrowers.24 MCs securitize about 75 percent of loans.
Banks, on the other hand, still originate the dominant share of loans to hold and securitize only 35
to 40 percent of mortgages, primarily through the GSEs. As compared to banks, AMCs and IMCs
show higher engagement in subprime lending.
The exact deﬁnition of a subprime loan is elusive and varies from study to study.25 The common
element across all deﬁnitions of a subprime loan is a high default risk. In line with this common
thread we exploit the HMDA requirement as of 2004 to report yield spread on all loans with
interest rates exceeding the prime rate by three (ﬁve) percentage points for ﬁrst-lien (subordinatelien) loans. We classify all loans with excess yield spread as subprime. This subprime proxy is loan
24

We have conducted (i) the t-tests for the diﬀerence in mean characteristics of MCs’ and bank aﬃliates’ loans
(columns (3) and (4)), and (ii) the Kolmogorov-Smirnov tests to access diﬀerences in distributions of these variables.
The resulting statistics uniformly indicate signiﬁcant diﬀerences in distributions with p-values below 1%. We do not
report the results of these tests for brevity.
25
The term “subprime” can be used to describe certain characteristics of the borrower, lender, or a type of security
that the loan can become a part of. The Board of Governors of the Federal Reserve System, OCC, FDIC, and OTS
deﬁne all borrowers with a FICO credit score less than 660 as subprime. The U.S. Department of Housing and Urban
Development uses HMDA data and interviews lenders to identify subprime lenders among them. There are, however,
some subprime lenders making prime loans and some prime lenders originating subprime loans.

speciﬁc and is available for years 2004-2006.26 For robustness, we also proxy the subprime activity
in a given ZIP code by the fraction of nonagency securitized subprime mortgages: the number
of subprime and Alt-A loans originated in each year and ZIP code normalized by a number of
loans originated in the same geography (reported by HMDA). The data come from the CoreLogic
LoanPerformance Securities (LP) database, which contains information about over 90% of all U.S.
nonagency securitized mortgages (some 20 million loans in the subprime and Alt-A segments).

3.2

Characteristics of Bank Holding Companies

The “HMDA Lender File” allows us to map the HMDA lending institutions to parent BHC data
provided by the Call Reports. All FDIC-insured commercial banks and BHCs are required to ﬁle
Call Reports with the regulators on a quarterly basis. These reports contain detailed information
on BHCs’ income statement, balance sheet, and some oﬀ-balance-sheet activities.
We obtain the following BHC ﬁnancial information from these reports: BHCs’ size, proﬁtability,
share of mortgages in the loan portfolio, liquidity position, equity capital, and loan performance
measures. We ensure that the changing reporting requirements are reﬂected in our calculations
and that our measures are consistent over time. While the majority of the variables are standard
to the banking literature, we would like to describe in detail the on-balance-sheet measures of
mortgage defaults: (i) net chargeoﬀs (net of recoveries) on one-to-four-family residential mortgages,
and (ii) nonperforming mortgages in this category, i.e., mortgage loans that are 90 days past
due or delinquent. Both variables are normalized by the volume of one-to-four-family residential
mortgages on BHCs’ balance sheets. The net mortgage chargeoﬀs have an immediate impact on
bank proﬁtability. However, a bank facing an overwhelming number of potential foreclosures might
recognize chargeoﬀs gradually. The charge-oﬀs measure of mortgage quality could to some extent
be subject a bank’s discretion (Agarwal, Amromin, Ben-David, and Dinc (2013)). Therefore, we
also analyze the nonperforming mortgages that are free from this bias and provide a more direct
on-balance-sheet measure of borrowers’ default rates.
Table 2 reports summary statistics for the sample of BHCs mapped to HMDA data. It shows
that only 10% of BHCs had MC aﬃliates, and those BHCs were on average 16 times larger than
26

The HMDA “high-cost lending” measure of subprime activity is highly correlated with the measure based on the
“subprime lender” identiﬁer—reported by the U.S. Department of Housing and Urban Development. The correlation
is 0.89.

BHCs lending only through their ADIs. The radically diﬀerent securitization rate and share of
deposits in total assets suggests that BHCs with MC subsidiaries preferred (or found it easier) to
grow by funding themselves via the securitization market rather than by raising deposits. BHCs
that were more active in the secondary loan market held smaller liquid fund balances.

3.3

Representative Borrower Characteristics and Default Rates

HMDA data provide a number of loan and lender characteristics, but lack some characteristics that
might be considered crucial in the loan-risk assessment, such as a borrower credit score and house
price appreciation (Demyanyk and Van Hemert (2011)). To ﬁll this gap, we obtain borrower characteristics from the anonymized Equifax data provided by the Federal Reserve Bank of New York’s
Consumer Credit Panel. These data contain a wide range of consumer credit–related information
for a random 5% of almost all individuals who have a social security number and a credit report in
the U.S. (about 12 million consumers). The credit score reported by Equifax allows us to construct
a “Low Credit Score” measure at the ZIP code–year level. It equals the share of consumers in a
ZIP code with a credit score below 660.27
Using Equifax data, we also construct two measures of loan performance at the ZIP code level.
First, we construct a cumulative foreclosure rate over the 2007-2008 period deﬁned as the number
of individuals who had at least one home in foreclosure within the past 24 months as of December
2008, normalized by a total number of individuals with mortgages as of December 2008. There
were, on average, 3.1% (3.5%) of borrowers in foreclosure during 2007-2008 (2008-2009).28 Since
the event of foreclosure is subject to lenders’ and/or servicers’ discretion, we also utilize a quarterly
ZIP code–level delinquency rate equal to a share of mortgage borrowers who were more than 30 days
past due on their loans or in “severe derogatory” state as deﬁned by Equifax. We normalize this
measure by a total number of individuals with mortgages in the same ZIP code–quarter. Given that
borrowers can be classiﬁed as delinquent on their loans in multiple quarters, we average the quarterly
delinquency rates across eight quarters of 2007-2008 to capture a representative delinquency. The
27
The range of possible credit score values in Equifax data is 280–850. Equifax uses the 660 cutoﬀ point in identifying
borrowers with “subprime” scores. For details, see http://news.equifax.com/index.php?s=18010&item=96773. Our
results are robust to using a continuous measure of credit scores.
28
For comparison, the Mortgage Bankers Association reports the average rate of homes in foreclosure between 2007
and 2008 (2008 and 2009) to be 3.2% (3.6%). We report the results based on 2007-2008 foreclosures and delinquencies.
The results based on 2008-2009 data are qualitatively and quantitatively similar.

delinquencies-based measure is free from the lenders’ discretion and provides a more unbiased proxy
of borrowers’ quality. In 2007-2008, on average, 5.5% of mortgage borrowers were delinquent on
their loans in any given quarter.
Finally, we augment our representative borrower characteristics with the ZIP code–level house
price indices (HPIs) from CoreLogic. These indices are calculated using a weighted repeat sales
methodology and are normalized to 100 as of January 2000. We use the county-level HPIs where
ZIP code HPIs are not available. Our results are virtually unchanged if we restrict our sample to
only those ZIP codes for which the HPIs are not missing. Table 1 Panel A reports the summary
statistics for the variables obtained from Equifax and CoreLogic. Combined, HMDA, Equifax, and
CoreLogic provide us with the sample that covers about 49% of the ZIP codes and represents 81%
of the U.S. population.

3.4

Individual Loan-Level Data

In our ﬁnal set of tests, we conﬁrm the ZIP code representative borrower results using the loan-level
analysis. Similar to Agarwal and Evanoﬀ (2013), we merge HMDA with two credit registries: Lender
Processing Services, Applied Analytics (LPS) and LP data. LPS includes loan-level information
collected from residential mortgage servicers. As of July 2008, LPS includes loans from nine of the
top ten servicers, and represents around two-thirds of the mortgage market in the U.S. Prior to
2005, however, the LPS coverage and the set of available loan characteristics is limited. Therefore,
we restrict our analysis to loans originated in the 2005-2006 period, wherein the loan-level data
oﬀer widest coverage both geographically and in terms of available loan characteristics. LPS data
consistently underrepresent the riskier types of mortgages, such as subprime. We supplement it with
LP data that contain information on securitized subprime loans. Combined, LP and LPS provide
us with the largest mortgage credit registry available. We map LPS-LP loan registry to HMDA
originations at the loan-level based on the origination date, ZIP code of the property location, loan
amount, loan type, loan purpose, occupancy type, and lien. We exclude from our analysis all loans
that do not have a unique combination of the above characteristics in both HMDA and LPS-LP
data and implement a sequential mapping procedure to eliminate any double counting of loans
due to the overlapping nature of LPS and LP data. We are able to match 1.4 million loans out
of 8.9 million loans originated by BHC subsidiaries in the 2005-2006 period (HMDA). The sample
15

contains loans from 31% of the ZIP codes that house 59% of the U.S. population. To adhere to
the contract terms of the data providers, we discard all HMDA lender identifying information and
only retain lender types and anonymized identiﬁers of BHCs.
Panel B of Table 1 reports the summary statistics for LP-LPS-HMDA merged data. As compared to the entire loan population reported in HMDA, the LP-LPS-HMDA loans are on average
larger, and granted to people with higher income levels and higher loan-to-income ratios. The
privately securitized and subprime loans are overrepresented, while the GSE securitized and reﬁnancing loans are underrepresented in LP-LPS-HMDA data relative to HMDA sample. Admittedly,
this sample is not a random draw from the U.S. mortgages pool and might be subject to some (unobserved) selection. Despite this fact, we believe that this sample provides a unique opportunity
to evaluate our hypotheses.
The univariate comparison of columns (3) and (4) suggests that, as compared to borrowers
of ADIs, borrowers obtaining loans from AMCs have lower incomes, higher loan-to-income ratios,
lower FICO scores, and higher loan-to-value ratios. MCs originate riskier types of loans: interestonly loans, adjustable-rate mortgages, and loans with prepayment penalty. AMC borrowers are
2.8% more likely to become delinquent within two years of loan origination, which is economically
signiﬁcant when compared to the average delinquency rate of 9%. Similarly, AMC borrowers
are more likely to end up in foreclosure as compared to ADI borrowers. Similar, and yet more
striking, diﬀerences are observed when one compares IMC and IDI borrower characteristics. Overall,
the univariate comparison of AMC and ADI loan characteristics provides support to our lending
standards–related hypothesis.

4
4.1

Empirical Tests and Results
Capital Arbitrage

Our ﬁrst hypothesis posits that by lending through AMCs as opposed to ADIs, a BHC can conserve
its capital. While ADIs need to hold capital for all originated loans on their books, even for those
they plan to sell in the secondary market, AMCs do not have to capital provision for their loans. A
BHC can also avoid consolidating its AMCs for capital requirement purposes. Thus, lending done
by a BHC via its AMCs as opposed to its ADIs could allow a BHC to extend its mortgage business
16

without a mandatory capital expansion.
The relationship between a decision to establish (or increase lending through) an AMC subsidiary and the amount of a BHC’s regulatory capital is hard to establish. BHCs engage in regulatory arbitrage exactly to increase or maintain their regulatory capital ratios (e.g., Tier 1 capital
ratio). As BHCs increase the amount of AMC lending, their regulatory capital ratio is not expected
to change or is even expected to increase. One way to solve the problem is to use the BHC balance
sheet equity capital ratio (ratio of book value of equity to total assets) instead of the regulatory
capital. BHCs engage in regulatory capital arbitrage to maintain or even reduce their risk-based
assets while increasing the total assets. Acharya, Schnabl, and Suarez (2011) argue that this increase in total assets relative to risk-weighted assets is captured in the equity capital but not in the
regulatory capital. In line with this argument, we adopt the following empirical design:

AMC Lending i,t = Y eart + BHCi + β1 Equity Capital i,t−1 +
β2 BHC Controls i,t−1 + εi,t ,

(1)

where the core variable of interest Equity Capital i,t−1 is measured as percentage of total assets. Our
objective is to explore whether less capitalized BHCs are more likely to establish AMCs (or increase
lending through their existing AMCs). We control for other measures of ﬁnancial constraints: the
liquidity position, cost of deposits, share of deposit ﬁnancing, and bank size. We further control
for AMC and ADI loan portfolio structures to capture the heterogeneity across these two types
of intermediaries due to specialization. Two sets of respective ﬁxed eﬀects absorb any BHC- or
time-speciﬁc unobserved heterogeneity. Thus, our results are based on the within-BHC variation
in the variables of interest. The annual data spans from 1992 through 2006 and covers 217 BHCs
that had an AMC in at least one year.
First, we evaluate the extensive margin and explore whether less capitalized BHCs are more
likely to establish an AMC. Panel A of Table 3 reports the results of the linear probability model
where AMC Lending i,t is a dummy variable that equals one if a BHC i lends through an AMC
subsidiary in year t and zero otherwise.29 To avoid perfect collinearity between the dependent
29

The choice of the linear probability model, as opposed to probit model, is motivated by the incidental parameter problem associated with the probit/logit ﬁxed eﬀect regressions (Lancaster (2000)). In the unreported set of
robustness tests, we implement the ﬁxed-eﬀect logit regression analysis with the Chamberlain (1980) adjustment that

variable and the BHC ﬁxed eﬀects, we only consider BHCs with at least one change of status
between lending and not lending through an AMC. We ﬁnd that AMCs are more likely to be
established by BHCs that have a lower equity capital ratio. A one-standard-deviation decrease in
Equity Capital i,t−1 (2.78%) leads to about ﬁve percentage points of increase in the likelihood of
establishing an AMC.
Second, we evaluate whether the lower level of equity capital is associated with an increase in
the extent of AMC lending. In Panel B of Table 3, AMC Lending i,t is the dollar volume of a BHC
lending through AMCs normalized by the total assets of the parent BHC. The dependent variable,
thus, captures any change in AMC originations, including de novo establishments or closings of
AMCs by the same BHC. In this part of the analysis, we aim to isolate the intensive margin and
only consider BHC-year observations with above-zero AMC lending. The results indicate that the
equity capital has a material impact on the extent of the AMC lending. A one-standard-deviation
decrease in equity capital is associated with an increase in AMC originations equivalent to 3% of a
BHC’s total assets.30
The results in both Panels are consistent with less capitalized BHCs being more likely to lend
through their AMCs. One can argue, however, that the results could be due to an omitted variable
bias. For example, rapidly growing and, hence, less capitalized BHCs might ﬁnd it more eﬃcient
to expand lending through AMCs that face almost no geographic entry barriers. To eliminate such
explanations, we exploit the regulatory uncertainty stemming from Enron’s collapse in 2001.
In early 2002, the regulators oﬃcially called FASB to reexamine the consolidation practices for
“variable interest entities” (VIE). The industry considered the regulatory change a real possibility,
which is evident from the asset-backed commercial paper market stalling in response to this announcement (Acharia and Schnabl (2010)). In 2003, the FASB issued guidance number 46, which
recommended consolidation of a wide range of VIEs. If the guidance were to be followed by a
regulatory change, BHCs would have had to consolidate AMC activities irrespective of the equity
stake held in aﬃliates or the degree of control over them. Such a ruling would have had a dramatic
impact on BHCs capital requirements. Pursuing AMC lending when it could be consolidated to
produces consistent estimates. We ﬁnd the results to be qualitatively similar.
30
Our conclusions are consistent with Acharya, Schnabl, and Suarez (2011), who ﬁnd that capital arbitrage was
a core consideration behind banks establishing asset-backed commercial paper conduits, another shadow banking
vehicle.

BHCs’ balance sheets at any moment might have proved to be extremely costly for BHCs. In July
2004, however, U.S. regulators oﬃcially backed down and the ﬁnal regulatory change was minor.
In columns (3) and (6) of Table 3, we exploit this regulatory uncertainty as an exogenous shock
to the ability of BHCs to save on equity capital by redistributing lending away from ADIs and to
AMCs. We evaluate whether the coeﬃcient of the core variable of interest is rendered insigniﬁcant
during the 2002-2004 period. Speciﬁcally, we interact Equity Capital i,t−1 with P ost2002 (P ost2005 )
dummy equal to one for years after and including 2002 (2005) and zero otherwise. The results show
a statistically and economically insigniﬁcant link between the BHCs’ capital and their willingness to
establish (or extend the activities through) AMCs during the period of regulatory uncertainty. Upon
the resolution of said uncertainty, in 2005-2006, the relationship between BHCs’ equity capital and
AMC activities is somewhat reestablished. While we cannot fully eliminate the omitted variable bias
concerns, the regulatory uncertainty results impose signiﬁcant additional hurdles on the alternative
economic mechanisms potentially driving our results.

4.2

Limited Loss Exposure

In this section, we evaluate whether the limited liability structure of AMCs and their GAAPdriven loan-loss and loan-impairment recognition practices allowed BHCs to limit recognition of
AMC loan-losses. The challenge in establishing this relationship lies in the lack of the balance sheet
data for MCs or their loan portfolio performance. MCs originated the majority of their loans to
distribute and, hence, passed the credit risk to the secondary market, which further inhibits our
ability to capture loan losses. To compare whether BHCs recognized AMC and ADI loan losses
diﬀerently, we exploit the exogenous freeze of the mortgage-backed securities market in mid-2007.
Purnanandam (2011) documents that when the secondary market came under pressure, banks were
stuck with large quantities of rapidly deteriorating loans they previously originated to distribute.
Consequently, banks that were more active in the OTD business model before the crisis experienced
higher loan-related losses during and after the crisis.
We augment the Purnanandam (2011) identiﬁcation strategy to evaluate whether the onbalance-sheet BHC losses were diﬀerent if “to-be-securitized” loans were originated by ADIs or

AMCs of a BHC. We estimate the following diﬀerence-in-diﬀerence regression equation:

On-balance-sheet Loan Losses it = BHCi + Quartert +
β1 P ostQ1, 2007 × Total OTD 2006
+ β2 P ostQ1, 2007 × OTD through AMC 2006
+
i
i
+γ1 P ostQ1, 2007 × Loan Portfolio Controls 2006
+ γ2 Bank Controls it + εit ,
i

(2)

where On-balance-sheet Loan Losses it for BHC i in quarter t is either the net chargeoﬀs (Panel A of
Table 4) or the nonperforming mortgages, i.e., those that are past due 90 days or more (Panel B).
Both measures are normalized by the balance sheet volume of one-to-four-family mortgages as of
the beginning of the quarter. Furthermore, in case BHCs don’t recognize losses from AMC loan
portfolios as mortgage-related losses but rather as losses from investment activities, we analyze the
quarterly net income of BHCs (Panel C). P ostQ1, 2007 is a dummy variable that equals one after
the disruption of the secondary market (i.e., after the ﬁrst quarter of 2007) and zero otherwise.
The sample spans from the last quarter of 2005 to the second quarter of 2008. Total OTD 2006
i
reﬂects the extent of a BHC OTD activity before the crisis as reported in HMDA. Speciﬁcally, it is
equal to the volume of loans originated and subsequently securitized by all BHC aﬃliates in 2006,
normalized by the balance sheet volume of one-to-four-family mortgages as of the end of 2005.
Similarly, OTD through AMC 2006
is the 2006 volume of AMCs’ loans originated and securitized,
i
similarly normalized.
The coeﬃcient β1 captures the impact of aggregate BHC OTD activities on loan losses after
the freeze of the secondary market, while β2 captures the incremental eﬀect of AMC OTD activity.
Following Purnanandam (2011), we expect β1 to be positive. If the loan losses incurred by ADIs
and AMCs are recognized similarly, we expect the coeﬃcient β2 to be zero. The existence of the
regulatory arbitrage, however, should lead to negative β2 that fully or partially oﬀsets the positive
eﬀect of β1 .
In the regression, we control for a variety of bank ﬁnancial conditions. BHC-speciﬁc ﬁxed eﬀects
allow to control for a host of BHC-speciﬁc, time-invariant factors (e.g., the risk preferences of a
BHC’s management team). Time-speciﬁc ﬁxed eﬀects capture the impact of economy-wide factors
on mortgage performance. Since the inferences could be aﬀected by heterogeneity across ADIs and
AMCs loan portfolios (see Table 1) we control for the following 2006 loan portfolio characteristics
20

interacted with P ostQ1, 2007 dummy: percent of jumbo loans, percent of reﬁnancing loans, loanto-income ratio, loan size, borrower income, percent of minority borrowers, and average borrower
income relative to area income. These characteristics are estimated separately for AMCs and ADIs,
as well as for loans securitized and loans retained, thus creating four groups of control variables
not reported in the table for brevity.
Table 4 reports the results: BHC mortgage-related losses are positively related to the extent
of OTD lending by ADIs but unrelated to the extent of OTD lending by AMCs. Consistent with
Purnanandam (2011), we ﬁnd that BHCs with a higher extent of OTD activity experience higher
mortgage-related losses after Q1 of 2007. In contrast, higher OTD origination by BHCs through
AMCs subsidiaries is not associated with higher BHC-level losses. The coeﬃcient on AMC lending
(β2 ) almost entirely oﬀsets the adverse eﬀect of aggregate OTD activity (β1 ) on performance of a
BHC’s mortgage portfolio. The evidence is consistent with BHCs being able to avoid recognizing
AMC loan losses.
One can argue, however, that despite loan portfolio controls, the results can be attributed to
ADI loans being of a lower quality than AMC loans. We conﬁrm that this is not the case via two
empirical tests. In Section 4.3, we explicitly compare the loan characteristics of ADIs and AMCs.
In Table 4, we augment the above empirical setup and evaluate whether BHCs recognized ADI and
AMC subprime-related losses similarly. Speciﬁcally, we interact the P ostQ1, 2007 dummy with
the extent of subprime lending by respective subsidiaries. Both variables are normalized by the
balance-sheet volume of one-to-four family mortgages as of the end of 2005. It is hard to argue that
ADI and AMC subprime loans were signiﬁcantly diﬀerent in quality. Columns (3) and (6) show that
a high extent of BHC subprime originations in 2006 is associated with signiﬁcantly inferior BHC
loan performance post Q1 of 2007. The eﬀect, however, is almost entirely negated if all subprime
loans are originated through AMCs. Overall, the results lead us to conclude that BHCs had the
ability not to recognize AMC loan losses.31
31
In an additional untabulated set of robustness tests, we ensure that BHCs do not merely delay loan-loss recognition
from AMC activities. We evaluate the cumulative performance of BHCs over diﬀerent horizons post-Q2 of 2007 and
ﬁnd results similar to those reported in Table 4. Furthermore, we verify that our results cannot be attributed to
AMCs and ADIs’ mortgage market specialization (Table 1). We decompose BHCs’ (AMCs’) OTD activity into GSE,
private securitization of subprime loans, and other private securitization, and ﬁnd the results quantitatively and
qualitatively similar.

4.3

AMC and ADI Lending Standards

We have presented evidence consistent with the regulatory gap between ADIs and AMCs allowing
BHCs to circumvent their capital requirements and limit exposure to AMC loan losses. These
factors create incentives for BHCs not only to originate more loans though AMCs, but also to
originate inferior-quality loans through them. The absence of consumer compliance enforcement
for MCs makes this risk-shifting strategy viable. In this section, we empirically compare the ex
ante and ex post risk characteristics of ADI and AMC loans. First, we utilize a representative ZIP
code–level borrower framework to evaluate the ex ante diﬀerences in the ADI and AMC lending
standards. Second, to ensure the robustness of our results, we analyze a sample of about 1.4 million individual loans. Finally, we quantify the contribution of aggregate MC lending to borrowers’
defaults using a broad sample of ZIP code–level delinquency and foreclosure rates.

Within-BHC Lending Standards: Representative Borrower Analysis
We analyze the lending standards of ADIs and AMCs within a BHC using a representative ZIP
code–level borrower framework (BHC-ZIP-year level of analysis) similar to Mian and Suﬁ (2009).
Our core objective is to isolate the eﬀect of the diﬀerent regulatory environments of AMCs and
ADIs. We achieve it via two dimensions of sample selection. First, as discussed earlier, we only
consider within-BHC lending of ADIs and AMCs. With parent BHCs being ultimate pooling agents
and ensuring equal ability to securitize loans across all of its subsidiaries, our within-BHC analysis
compares the lending standards of ADIs and AMCs that could have securitized the same loans.
Second, to eliminate the eﬀect of potential market-entry frictions, we exclude from the analysis
markets where a BHC lends only through an ADI or only through an AMC. Speciﬁcally, by only
considering CBSAs served by both AMCs and ADIs in a given year-BHC, we eliminate potential
alternative explanations stemming from diﬀerent barriers to entry, strategic expansion goals, and
conglomeration activity. Eﬀectively, we ensure, through sample selection, that ADIs and AMCs of
a BHC had the ability to originate and securitize the same loans.

We start with the following regression model:

Share of AMC Lending izt = β1 Low Credit Score zt + β2 Loan-to-Income Ratio izt +
β3 Relative Borrower Income izt + Market-Year ct + BHCi +
γControlsizt + εizt ,

(3)

where Share of AMC Lending izt is a ratio of AMC lending to total mortgage lending within BHC
i, ZIP code z, and year t. The core variables of interest are the three measures of borrower
quality: (i) Low Credit Scorezt is a fraction of individuals with a credit score below 660 residing
in a ZIP code, (ii) Loan-to-Income Ratioizt is an average loan-to-income ratio of a BHC-speciﬁc
representative borrower, and (iii) Relative Borrower Incomeizt is an average BHC borrower’s income
relative to median area income as reported by HUD. Note that we do not aim to establish causality.
We are agnostic as to whether the presence of AMCs provides riskier borrowers access to capital or
whether the presence of the riskier borrowers attracts an AMC to a given geography. Both could
stem from the regulatory gap.
We include three types of additional control variables. We control for the ultimate originator’s
(the BHC’s) ﬁnancial conditions: liquidity, size, cost of funds, equity capital, and mortgage and
C&I loan specialization. We also control for the lagged annual house price appreciation. The
CBSA–year ﬁxed eﬀects, Market-Year ct , ensure that our results are not driven by diﬀerences in
local economic conditions (e.g., income growth). BHCi ﬁxed eﬀects absorb the BHC-speciﬁc timeinvariant heterogeneity (e.g., risk preferences). Standard errors are clustered at the CBSA-year
level. The availability of Equifax data restricts us to evaluating the lending standards post 1999.
To avoid the contaminating eﬀect of the crisis on the lending behavior of ﬁnancial institutions, we
end our sample in 2006.
Columns (1) and (2) of Table 5 report our baseline results consistent with AMCs having inferior precrisis lending standards as compared to ADIs of the same BHC. Speciﬁcally, we ﬁnd that
BHCs lend disproportionately more through their AMCs in ZIP codes characterized by borrowers
with lower credit scores, higher loan-to-income ratios, and lower relative incomes. These results
are consistent with the univariate comparison of ADI and AMC loan characteristics presented in
Table 1.
23

One can argue that despite the fact that we conduct our analysis within-BHC, the results
could still be attributed to other diﬀerences across ADIs and AMCs in addition to the regulatory
gap. Table 1 documents signiﬁcant heterogeneity across ADI and AMC lending and securitization
specialization.
First, our results could be due to the higher extent of AMCs’ participation in mortgage securitization in general, and private securitization in particular. Mian and Suﬁ (2009) show that lending
standards deteriorated due to oversupply of funds from the securitization market. Since AMCs sell
a larger share of their loans, they should be more inﬂuenced by the precrisis boom in the respective
funding supply as compared to ADIs. To alleviate this concern, we conduct our analysis within the
pool of privately securitized (non-GSE) loans (column (5)). We intentionally exclude loans securitized through the GSEs, as these loans are subject to much higher and, most importantly, uniform
lending standards—including high credit score and no less than 20% downpayment. The results
indicate that this heterogeneity across ADIs and AMCs cannot fully explain our baseline evidence
that BHCs lend disproportionately more through their AMCs in ZIP codes with inferior-quality
borrowers.
Second, to account for diﬀerent rate of participation of ADIs and AMCs in the subprime market,
we control for the involvement of BHCs in subprime originations in the respective ZIP codes (column
(3)). Furthermore, when we build our dependent variable using BHC lending only in the prime
segment of the mortgage market (column (4)), we still observe observe evidence consistent with the
underwriting standards of AMCs being inferior to those of ADIs.
The third potential explanation for our results stems from MCs only originating mortgages,
while banks provide a wide set of consumer ﬁnancial services: savings accounts, small-business loans,
money market accounts, etc. ADIs can be motivated to actively steer better-quality borrowers to
their branches, since these borrowers have the capacity to consume other ﬁnancial services. ADIs
can either oﬀer more attractive deal terms on loans as compared to those oﬀered by AMCs, or
have AMC managers redirecting better-quality customers to ADI branches.32 To ensure that our
baseline results are not driven by such customer-steering behavior, we analyze the diﬀerence in
lending standards between AMCs and ADIs in the geographies (CBSAs) where a parent BHC does
32

Agarwal and Evanoﬀ (2013) explore another aspect of lenders’ steering behavior. They analyze loans that were
denied by one BHC subsidiary but subsequently approved by another subsidiary.

not have physical branches and ADIs’ ability to cross-sell services is limited at best. About 60%
of the BHC-ZIP-year observations in our data come from such geographies. Column (6) of Table 5
suggests that AMCs have inferior lending standards as compared to those of ADIs, even in the
areas where BHCs do not have a physical branch presence.
Forth, our results could also be attributed to ADIs carrying a parent BHC’s name and, thus,
being fully responsible for shaping the latter’s reputation (Carey, Post, and Sharpe (1998)). The
richness of our data allows us to evaluate the validity of the reputation-based economic mechanism
driving the wedge in the lending standards. Speciﬁcally, we compare the lending behavior of
ADIs to that of AMCs with names consonant to the parent BHC names (e.g., Bank of America
Mortgage). We operate under the plausible assumption that such AMCs are as responsible for
shaping the reputation of the parent BHC as ADIs. Column (7) reports the results of our baseline
analysis after we exclude the “generic-name” AMCs from the sample. The empirical evidence does
not support the reputation concerns being a primary explanation for the diﬀerence in mortgagelending standards between AMCs and ADIs.
Finally, in the set of robustness tests (not tabulated), we further verify that our results are not
driven by the other market segmentation between AMCs and ADIs. We ﬁnd a similar wedge in
the lending standards if we limit our sample to (i) conventional loans, (ii) ﬁrst liens, (iii) subprime
loans, etc. All our results are consistent with AMCs lending to inferior quality borrowers as compared to those of ADIs.

Within-BHC Lending Standards: Loan-Level Evidence
In this subsection, we conﬁrm the ZIP code representative borrower evidence using a sample of
1.4 million individual loans originated in 2005-2006 from a matched LPS-LP-HMDA-Call Report
sample. Admittedly, this sample is not necessarily random and might not be fully representative of
the U.S. economy. In designing and evaluating our tests, we are cognizant of this issue. Nonetheless,
the ability to jointly control for a wide set of lender, borrower, and loan characteristics allows us
to eﬀectively eliminate a number of potential omitted variable driven explanations. The loanlevel within-BHC analysis also addresses some reservations associated with representative borrower
framework and aims to solidify the notion that AMCs’ lending standards are inferior to those of
ADIs.
25

First, we analyze whether AMCs are associated with riskier borrowers and riskier loan types at
origination. For this purpose we estimate the following regression equation:

AMC Loan i = γ1 Borrower Characteristics i + γ2 Loan Characteristics i +
γ3 Controlsizt + Market-Year ct + BHCj + εi ,

(4)

where AMC Loan i is a dummy variable equal to one if a loan i was originated by an AMC and zero
if it was originated by an ADI. The CBSA–year ﬁxed eﬀects, Market-Year ct , ensure that our results
are not driven by diﬀerences in local economic conditions. BHC ﬁxed eﬀects absorb time-invariant
heterogeneity across BHCs. As in the earlier analysis, we do not aim to establish causality but
rather evaluate whether AMCs are associated with lower-quality loans. Table 6 presents the results
with the coeﬃcients multiplied by a hundred for tractability. Columns (1) and (2) report the results
of the analysis without restricting the sample to CBSA-BHC-years where both types of aﬃliates
operated. We go back to our original sample selection restriction in column (3).
The results suggest that AMCs are more likely to originate loan to borrowers with lower credit
scores, lower incomes, and higher loan-to-value ratios. We further document that AMCs are more
likely to originate loans of riskier types, such as adjustable-rate mortgages and interest-only mortgages. The results persist even if we restrict our consideration only to privately securitized loans
(column (4)).
Next we compare performance of ADI and AMC loans within two years of origination. Since
the mortgage defaults were disguised precrisis by house price appreciation (Demyanyk and Van
Hemert (2011)), we only consider the performance of loans originated in 2005 and 2006. Table 7
reports the result of the following regression analysis:

Mortgage Default i = β0 AMC Loan i + β1 Borrower Controls i + β2 Loan Controls i +
β3 Lender Controls k + β4 House Price Appreciation z,t +
Market-Year ct + BHCk + εz ,

(5)

where Mortgage Default i is a dummy variable equal to one if the loan was in severe delinquency (60

days or more past due), foreclosure, or real estate owned within two years of origination.33 CBSAyear ﬁxed eﬀects control for vintage year of the loan, the impact of local economic conditions,
and legal environment on a borrower propensity to default. We control for the ZIP code–level
house price appreciation over two years since loan origination. The coeﬃcients are multiplied by a
hundred for tractability.
We ﬁnd that mortgages originated by AMCs are on average 5.6% more likely to end up in
default within two years of origination. Given that the average default rate in LPS-LP-HMDA
matched sample is 9%, the eﬀect is economically signiﬁcant. Since we explicitly control for a vast
set of borrower, loan, and lender characteristics, these results cannot be explained by AMCs and
ADIs occupying diﬀerent segments of the mortgage market. The analysis supports our argument
that the regulatory inconsistencies allow BHCs to originate riskier loans.
Finally, the loan-level data allow us to evaluate the heterogeneity in risk-taking behavior across
BHCs. We explore it along four BHC characteristics: (i) equity capital, (ii) size, (iii) extent of
securitization activity (normalized by total assets), and (iv) extent of MC lending activity (share
of originations). By interacting the AMC Loan dummy with these BHC characteristics (columns
(4) through (7) of Table 7) we can infer the types of BHCs that more aggressively pursued riskier
lending strategies through their AMCs.
First, we ﬁnd that larger and less capitalized BHCs are associated with lower-quality loans
through their AMCs (columns (4) and (5)). The capital-related evidence supports the notion that
the need to conserve capital motivated BHCs to pursue mortgage lending through AMCs. Second,
BHCs that originated a larger share of OTD mortgages were less likely to pursue inferior lending
strategies through AMCs (column (6)). This evidence is consistent with Agarwal, Chang, and Yavas
(2012), who ﬁnd that portfolio loans were more likely to default than securitized loans.Finally, we
observe that BHCs with a higher share of loans originated though AMC subsidiaries are characterized by inferior-quality loans (column (7)). This evidence is consistent with AMCs originating
risky loans that ADIs could not originate due to regulatory constraints.

Aggregate Eﬀect of MC Lending: Representative Borrower Default Analysis
33

In the unreported set of tests, we conducted the delinquency and foreclosure analysis separately. We ﬁnd the
results quantitatively and qualitatively similar to those reported in Table 7.

We have shown evidence consistent with AMCs lending to ex ante lower-quality borrowers that
are more likely to default ex post. One can argue, however, that the within-BHC results are based
on about 30% of ﬁnancial intermediaries engaged in mortgage lending and, thus, are not necessarily
representative. In this section, we consider a broad sample of ZIP code–speciﬁc borrower defaults to
establish the generality of our conclusions. Speciﬁcally, we deviate from the within-BHC framework
and evaluate whether IMCs and AMCs lending is associated with higher ZIP code–level delinquency
and foreclosure rates. This analysis allows us to quantify the economic eﬀect of unregulated MC
lending on borrowers’ defaults.
Similar to our previous approach, we evaluate whether higher mortgage default and delinquency
rates in the 2007-2008 period are associated with higher MC lending in the 2005-2006 period:

Mortgage Defaults z,Post-Crisis = CBSAj + β1 MC Share z,Precrisis + β2 Controlsz,Precrisis + εz ,

(6)

where Mortgage Defaults z in a ZIP code z is proxied by two variables: (i) the average quarterly
delinquency rate over the 2007-2008 period; and (ii) the foreclosure rate measured as a fraction of
the mortgage holders in a given ZIP code who were in foreclosure in any period of time between
2007 and 2008. The delinquency rate is a better proxy of borrower quality since it is free of a
lender/servicer/holder contribution to the decision to foreclose on a property. MC Share z,Precrisis is
a share of loans originated by aﬃliated and independent MCs over the 2005-2006 period. We control
for the borrower characteristics precrisis and the house price appreciation in 2007-2008. CBSA ﬁxed
eﬀects accommodate the impact of a (sometimes rapidly) deteriorating local economy. Since, in
this analysis, we deviate from the within-BHC identiﬁcation framework, in the untabulated set of
tests, we ensure that the aggregate MC lending behavior is consistent with the results presented in
Table 5—namely, MCs in general lend to lend in ZIP codes with lower-quality borrowers.
Table 8 reports the results of the delinquency and foreclosure analysis, respectively. The coefﬁcients are multiplied by a hundred for tractability. Columns (1) and (4) document that precrisis
MC lending is associated with higher default rates postcrisis. The coeﬃcient on MC Share z is
positive and both economically and statistically signiﬁcant. A one-standard-deviation increase in
the share of MC lending (12%) is associated with to about 1.3% increase in delinquency rate and
1.2% increase in foreclosure rate. These numbers, correspondingly, represent about a ﬁfth of the

average Q4 2008 delinquency rate (5.5%) and about a third of the average 2007-2008 foreclosure
rate (3.22%). For comparison, a one-standard-deviation increase in subprime lending is associated
with increase in delinquency rates of about 0.96% and foreclosure rates of 1.1%. The impact of MC
activity on mortgage defaults is, thus, similar in magnitude to that of subprime.
Our results cannot be explained by MCs being dominant players in the subprime market.
The results persist after we control for the extent of mortgage securitization in a given ZIP code
(columns (1)and (4)). Even the share of MC activity in the prime market (as deﬁned in HMDA)
is adversely correlated with mortgage defaults (columns (2) and (5)). Admittedly, the aggregate
impact of MC activity could be entirely due to the presence of large IMCs such as New Century
Financial. In columns (3) and (6) we separate the activities of IMCs and AMCs into two diﬀerent
regressors. The results are consistent with AMCs being as responsible for the inferior lending standards as IMCs.

Concluding Remarks

We examine two important issues that played a role in the 2007 credit crisis: the rise of “shadow
banking” and the inadequate regulation thereof. We bring to light the regulatory gap between
banks and MCs that relaxed the regulatory constraints even for heavily regulated BHCs. Lending
through AMCs allowed BHCs to mitigate regulatory capital requirements, avoid recognizing costly
loan losses, and pursue risky lending while still adhering to banking regulations. Combined, our
empirical evidence suggests that the pre-Dodd-Frank regulatory standards were not as inadequate as
they are perceived to be. The inconsistent coverage and enforcement of these regulations, however,
eroded their eﬀectiveness and contributed to the deterioration of lending standards.
Our study contributes to the literature examining the economic determinants of shadow ﬁnancial intermediaries’ behavior. Despite being the largest shadow banking intermediaries, mortgage
companies were largely perceived as inconsequential given the pass-through nature of the intermediation they had been providing. We now know that nondepository ﬁnancial intermediaries
play an important role in shaping access to ﬁnance by consumers and ﬁrms and aﬀect individual
agents’ and overall economic risks (Gorton (2010) and Adrian and Ashcraft (2012)). This study

contributes to the understanding of the shadow intermediaries’ behavior, which is important not
only from the perspective of MCs’ contribution to the 2007 crisis, but also from the perspective of
the mortgage market’s future development. MCs continue to hold 30% market share and, hence,
have a signiﬁcant role in shaping Americans’ access to the dream of home ownership.
Given the magnitude of the market shadow banks serve, one cannot underestimate the importance of the ongoing debate on the supervisory and regulatory design of the ﬁnancial sector.
Surprisingly, few voices in this debate address the lack of uniform enforcement of existing regulations and the elimination of regulatory arbitrage. Most new reform initiatives are geared toward
regulating a wider set of ﬁnancial-institution activities, as opposed to consistently enforcing existing regulations across all intermediaries. While we do not argue that banking supervision and
regulations were optimal before the crisis in 2007, or that the inconsistent regulations were solely
responsible for the deterioration of lending standards—the securitization market’s demand for loans
signiﬁcantly contributed to this phenomenon—our results do suggest that regulatory design that is
“institutional” in nature is ﬂawed. It ignores ﬁnancial institutions’ ability to circumvent regulations
and engage in risky investments via innovative corporate structures and instruments.
BHCs continue to innovate and adapt their corporate structure in response to market conditions and the regulatory environment. In 1995 only 35% of BHC subsidiaries were nondepository
institutions. In 2010 this number reached almost 60%. The increasingly complex organizational
structure reﬂects changes in risk allocation both within and across BHCs. Studies have brought
to our attention how the increase in interconnectedness across intermediaries and the associated
(idiosyncratic) risk sharing are contributing to systemic risk (Acharya, Iyer, and Sundaram (2015),
Brunnermeier, Dong, and Palia (2010)). Our paper suggests that the allocation of risk within a
BHC might also contribute to the build-up of systematic risk. Financial intermediaries might utilize
limited liability pass-through entities to compartmentalize risk and capture the upside of the highrisk segments of the market, while limiting their downside exposure. Preliminary analysis of CDS
spreads and systemic risk measures (see Online Appendix) indicate that, while lending through an
MC aﬃliate mitigates the mortgage loss exposure of a parent BHC, it is associated with higher
CDS spreads and systemic risk measures after Q2 2007.
Such risk allocation might lead to diﬀerent (potentially inferior) risk-assessment and riskmanagement procedures within a BHC. After all, function does follow organizational form (Berger,
30

Miller, Petersen, Rajan, and Stein (2005)). With BHCs continuing to shift risky investments to
specialized nondepository aﬃliates, we need to better understand not only how such behavior affects access to credit by consumers and businesses, but also the resulting changes in the aggregate
level of risk in the economy.

References
Acharia, V., and P. Schnabl (2010): “Do Global Banks Spread Global Imbalances? The Case of AssetBacked Commercial Paper during the Financial Crisis of 2007-09,” IMF Economic Review, 58, 37–73.
Acharya, V., L. H. Pedersen, T. Philippon, and M. Richardson (2010): “Measuring Systemic Risk,”
Working Paper.
Acharya, V. V., A. M. Iyer, and R. K. Sundaram (2015): “Risk-Sharing and the Creation of Systemic
Risk,” Working Paper.
Acharya, V. V., P. Schnabl, and G. Suarez (2011): “Securitization Without Risk Transfer,” Forthcoming in Journal of Financial Economics.
Adrian, T., and A. B. Ashcraft (2012): “Shadow Banking Regulation,” Federal Reserve Bank of New
York, Staﬀ Report No. 559.
Adrian, T., and M. Brunnermeier (2008): “CoVar,” NBER Working Paper No. 17454.
Agarwal, S., B. W. Ambrose, and Y. Yildirim (2010): “The Subprime Virus: Theory and Evidence,”
Working Paper.
Agarwal, S., G. Amromin, I. Ben-David, and S. Dinc (2013): “The Politics of Mortgage Foreclosures,”
Working Paper.
Agarwal, S., Y. Chang, and A. Yavas (2012): “Securitization and Adverse Selection in Mortgage
Lending,” Journal of Financial Economics, 105(3), 640–660.
Agarwal, S., and D. D. Evanoff (2013): “Loan Product Steering in Mortgage Markets,” Working Paper.
Agarwal, S., D. Lucca, A. Seru, and F. Trebbi (2012): “Inconsistent Regulators: Evidence from
Banking,” NBER Working Paper No. 17736.
Ashcraft, A. B. (2008): “Are Bank Holding Companies a Source of Strength to Their Banking Subsidiaries?,” Journal of Money, Credit and Banking, 40(3), 273–294.
Barth, J., G. Caprio, and R. Levine (2004): “Bank Regulation and Supervision: What Works Best?,”
Journal of Financial Intermediation, 13, 205–248.
Berger, A. N., N. H. Miller, M. A. Petersen, R. G. Rajan, and J. C. Stein (2005): “Does
Function Follow Organizational Form? Evidence from the Lending Practices of Large and Small Banks,”
Journal of Financian Economics, 76(2), 237269.
Bostic, R., K. Engel, P. McCoy, A. Pennington-Cross, and S. Wachter (2008): “State and Local
Anti-Predatory Lending Laws: The Eﬀect of Legal Enforcement Mechanisms,” Journal of Economics and
Business, 60(1-2), 47–66.
Boyd, J., and E. Prescott (1986): “Financial Intermediary-Coalitions,” Journal of Economic Theory,
38(2), 211–232.
Brunnermeier, M. K., G. Dong, and D. Palia (2010): “Banks’ Non-Interest Income and Systemic
Risk,” Working Paper.
Calomiris, C. W. (2006): “The Regulatory Record of the Greenspan Fed,” American Economic Review
Papers and Proceedings, 96(2), 170–173.
Carey, M., M. Post, and S. Sharpe (1998): “Does Corporate Lending by Banks and Finance Companies
Diﬀer? Evidence on Specialization in Private Debt Contracting,” Journal of Finance, 53(3), 845–878.
Chamberlain, G. (1980): “Analysis of Covariance with Qualitative Data,” Review of Economic Studies,
47, 225–238.
Demyanyk, Y., and O. Van Hemert (2011): “Understanding the Subprime Mortgage Crisis,” Review of
Financial Studies, 24(6), 1848–1880.

Diamond, D. (1984): “Financial Intermediation and Delegated Monitoring,” Review of Economic Studies,
51, 393–414.
Engel, K., and P. McCoy (2011): The Subprime Virus: Reckless Credit, Regulatory Failure, and Next
Steps. Oxford University Press, Inc.
Flannery, M. J. (2007): “Supervising Bank Safety and Soundness: Some Open Issues,” Economic Review,
Federal Reserve Bank of Atlanta, Q1–2, 83–100.
Gorton, G., and A. Metrick (2010): “Regulating the Shadow Banking System,” Brookings Papers on
Economic Activity, 41(2), 261–312.
Gorton, G., and G. Pennacchi (1995): “Banks and Loan Sales: Marketing Nonmarketable Assets,”
Journal of Monetary Economics, 35, 389–411.
Gorton, G. B. (2010): Slapped by the Invisible Hand: The Panic of 2007. Oxford University Press, USA.
Gouvin, E. J. (2002): “Financial Hoding Company Liability After Gramm-Leach-Bliley,” in Financial
Modernization After Gramm-Leach-Bliley, ed. by P. A. McCoy, pp. 37–64. LexisNexis.
Greenlee, M. (2008): “Historical Review of “Umbrella Supervision” by the Board of Governors of the
Federal Reserve System,” Review of Banking and Financial Law, 27, 407–459.
Isil, E., T. Nadauld, and R. Stulz (2012): “Why Did Holdings of Highly-Rated Securitization Tranches
Diﬀer So Much Across Banks?,” Working Paper.
Kane, E. J. (2000): “Ethical Foundations of Financial Regulation,” Journal of Financial Services Research,
12(1), 51–74.
Keys, B., T. Mukherjee, A. Seru, and V. Vig (2009): “Financial Regulation and Securitization:
Evidence from Subprime Loans,” Journal of Monetary Economics, 56, 700–720.
(2010): “Did Securitization Lead to Lax Screening? Evidence From Subprime Loans,” Quarterly
Journal of Economics, 125.
Lancaster, T. (2000): “The incidental parameter problem since 1948,” Journal of Econometrics, 95,
391–413.
Landier, A., D. Sraer, and D. Thesmar (2011): “The Risk-Shifting Hypothesis: Evidence from Subprime Originations,” Working Paper.
Lee, P. L. (2012): “The Source-of-Strength Doctrine: Reverted and Revisisted,” The Banking Law Journal,
129(10), 771–801.
Leland, H., and D. Pyle (1977): “Informational Asymmetries, Financial Structure, and Financial Intermediation,” Journal of Finance, 32(2), 371–387.
Loutskina, E., and P. Strahan (2009): “Securitization and the Declining Impact of Bank Finance on
Loan Supply: Evidence from Mortgage Originations,” Journal of Finance, 64, 861–889.
(2011): “Informed and Uninformed Investment in Housing: The Downside of Diversiﬁcation,”
Review of Financial Studies, 24, 1447–1480.
Mandanis Schooner, H. (2002): “Functional Regulation: The Securitization of Banking Law,” in Financial Modernization After Gramm-Leach-Bliley, ed. by P. A. McCoy, pp. 189–198. LexisNexis.
Mayer, C., and K. Pence (2008): “Subprime Mortgages: What, Where, and to Whom?,” in In Housing
Markets and the Economy: Risk, Regulation, and Policy: Essays in honor of Karl E. Case., ed. by E. L.
Glaeser, and J. M. Quigley. Cambridge: Lincoln Institute of Land Policy.
Mian, A., and A. Sufi (2009): “The Consequences of Mortgage Credit Expansion: Evidence from the U.S.
Mortgage Default Crisis,” Quarterly Journal of Economics, 124(4), 1449–1496.
(2010): “The Great Recession: Lessons from Microeconomic Data,” American Economic Review,
100(2), 51–56.

Ongena, S., A. Popov, and G. F. Udell (Forthcoming): “When the Cats Away the Mice Will Play:
Does Regulation At Home Aﬀect Bank Risk Taking Abroad?,” Journal of Financial Economics.
Purnanandam, A. (2011): “Originate-to-Distribute Model and the Subprime Mortgage Crisis,” Review of
Financial Studies, 24(6), 1881–1915.
Rajan, U., A. Seru, and V. Vig (2009): “The Failure of Models that Predict Failure: Distance, Incentives
and Defaults,” Working Paper.
Rosen, R. J. (2003): “Is Three a Crowd? Competition Among Regulators in Banking,” Journal of Money
Credit and Banking, 35(6), 967–998.
Stein, J. (1998): “An Adverse-Selection Model of Bank Asset and Liability Management with Implications
for the Transmission of Monetary Policy,” RAND Journal of Economics, 29, 466–486.
Wilmarth, A. E. J. (2002): “How Should We Respond to the Growing Risks of Financial Conglomerates?,”
in Financial Modernization After Gramm-Leach-Bliley, ed. by P. A. McCoy, pp. 65–134. LexisNexis.

Figure 1: Lending Volumes by Lending Institution Type
The ﬁgure shows lending volumes in $ billions for independent banks (IDIs), BHC bank subsidiaries (ADIs), BHC nonbank
mortgage subsidiaries (AMCs), and independent mortgage companies (IMCs). Panel B depicts the market shares of respective
ﬁnancial intermediaries. Data source: HMDA.

Panel A: Lending Volume, $ Billions
IDIs
ADIs
AMCs
IMCs

3000
2500
2000
1500
1000
500
0
1992

1994

1996

1998

2000 2002
Year

2004

2006

2008

2010

2006

2008

2010

Panel B: Market Shares, %
100
90
80
70
60
50
40
30
IDIs
ADIs
AMCs
IMCs

20
10
0
1992

1994

1996

1998

2000 2002
Year

2004

Table 1: Mortgage Lender: Summary Statistics.
This table reports summary statistics for four types of mortgage lenders: (i) independent depository institutions, IDIs; (ii) independent mortgage companies, IMCs; (iii) depository institutions aﬃliated with MC, ADIs; and (iv) MC subsidiaries of BHCs,
AMCs. Panel A is based on HMDA data for years 1999 through 2006 and reports summary statistics for ZIP code–level variables
from Equifax and CoreLogic data sets. Panel B reports summary statistics for the matched LP-LPS-HMDA data that covers
loans originated in the 2005-2006 period.

IDIs
Panel A: ZIP Code Representative Borrower, 1999-2006
Number of Loans (millions)
30.20
Loan Volume ($ billions)
4,962
Average Loan ($1,000)
164.20
Average Borrower Income ($1,000)
97.70
High-yield (Subprime) Loans (%, post 2004)
13.21
Securitized through GSEs (%)
28.08
Privately Securitized (%)
16.86
Reﬁnancing (%)
63.40
Jumbo(%)
8.41
Average Loan-to-Income Ratio
1.99
Borrower Income to Area Income
1.88
House Price Appreciation (t-1, %)
10.39
Fraction of Mortgage Holders with Low FICO Score (%)
26.59
Share of Mortgages in Foreclosure, 2007-2008 (%)
3.00
Share of Delinquent Mortgages, 2007-2008 (%)
5.46

IMCs

ADIs

AMCs

31.10
5,256
169.10
91.00
39.11
18.55
58.15
54.87
7.41
2.44
1.71
10.00
28.72
4.11
6.76

11.86
1,851
155.90
100.30
15.87
21.95
9.72
58.55
8.27
1.82
1.91
7.16
26.56
3.04
5.51

23.48
4,055
172.90
94.50
18.95
53.93
19.18
59.86
8.36
2.17
1.78
9.55
27.33
3.20
5.77

9.98
2.06
0.41
197.30
96.30
60.09
5.54
47.62
41.03
2.38
1.52
12.76
46.94
655.28
111.63
18.42
39.88
32.95
7.80
40.55
81.50
22.25
11.45

4.33
0.79
0.18
224.60
111.00
17.85
37.11
21.42
27.75
2.38
1.73
10.01
18.94
717.10
76.37
15.59
13.94
14.51
6.67
35.50
90.59
7.72
3.88

4.59
0.58
0.12
214.30
106.50
26.27
26.94
39.10
29.77
2.39
1.68
11.71
23.57
695.82
82.69
17.63
22.57
26.03
6.78
40.11
89.97
10.57
5.13

Panel B: Loan Level Data, 2005-2006
Total Number of Loans, HMDA (millions)
Number of Loans, Matched Sample (millions)
Loan Volume ($ billions)
Average Loan ($1,000)
Average Borrower Income ($1,000)
High-yield (Subprime) Loans (%, post 2004)
Securitized through GSEs (%)
Privately Securitized (%)
Reﬁnancing (%)
Average Loan-to-Income Ratio
Borrower Income to Area Income
House Price Appreciation (t-1, %)
Fraction of Borrowers with Score Below 660 (%)
Average FICO Score
Loan to Value Ratio (%)
Interest Only (%)
Prepayment Penalty (%)
Adjustable-Rate Mortgages (%)
Initial Interest Rate (%)
Full Documentation (%)
First Lien (%)
Delinquency Rate, 2 Years of Origination (%)
Foreclosure Rate, 2 Years of Origination (%)

7.01
1.13
0.29
255.90
114.00
17.02
27.44
33.21
46.50
2.55
1.77
13.53
19.55
708.76
68.07
16.14
25.11
32.21
6.00
45.23
89.15
9.61
4.14

37
8.83

36.37
12.50

C&I Loans (% of Loan Portfolio)

0.50

Mortgages (% of Loan Portfolio)

Letters of Credit (% of Total Assets)

64.81

Equity Capital (% of Total Assets)

Net Income (% of Total Assets)

1.64

Cost of Deposits (%)

80.47

Deposits (% of Total Assets)

0.09
26.34

Mortgage Chargeoﬀs (% of Mortgages)

Liquidity (% of Total Assets)

0.61

29.26

1.83

10.62

5.21

8.16

13.29

1.21

67.22

3.15

0.99

9.72

12.44

1.56

1.23

32.14

18.38

27.78

53.04

79,398

Mean StDev

Full sample

Nonperforming Loans (% of Mortgages)

Securitization Rate (%)

AMC Originations (% of Total Assets)

Mortgage Originations (% of Total Assets)

Total Assets, $Billion

Number of Observations

*** p<0.01, ** p<0.05, * p<0.1.

12.47

36.46

0.43

64.84

8.94

1.64

81.10

26.67

0.09

0.60

26.32

–

9.04

2.23

8.20

13.24

1.00

68.06

3.17

0.99

9.25

12.55

1.63

1.25

30.87

–

19.41

24.50

72,738

StDev

12.89

35.39

1.29

64.48

8.67

1.61

73.88

22.72

0.12

0.71

61.32

21.83

27.87

7.66

13.74

2.37

57.30

2.92

0.97

11.86

10.49

0.39

1.01

28.08

59.91

69.00

37.77 160.73

6,660

Mean StDev

MC Aﬃliates

MC Aﬃliates
Mean

BHCs with

BHCs without

The sample spans from 1992 through 2006. Absolute values of t-statistics are reported in parentheses.

t-stat

–

(3.90)

(7.75)

(0.48)

(7.34)

(2.18)

0.42***

–1.08***

(4.31)

(6.15)

0.86*** (29.23)

–0.36

–0.28***

–0.03**

–7.22*** (48.38)

–3.95*** (28.92)

0.03***

0.10***

35.00*** (96.52)

–

18.83*** (22.19)

35.55*** (18.03)

–

Diﬀerence

characteristics are from Call Reports. The volume of mortgages lending and the extent of the securitization activities are from HMDA. The unit of observation is BHC-year.

This table reports summary statistics for the sample of BHCs with and without MC subsidiaries. The sample contains BHC-year observations that map to HMDA. The ﬁnancial

Table 2: Bank Holding Company: Summary Statistics.

Table 3: Rationale for Establishing MC Subsidiaries.
Panel A reports results of the following linear probability model:
AMC Subsidiary i,t

Y eart + BHCi + β1 Equity Capital i,t−1 + β2 BHC Controls i,t−1 + εi,t ,

where AMC Subsidiary i,t is a dummy variable that equals one if a BHC i has an MC subsidiary in year t and zero otherwise.
Only BHCs with at least one change of status between lending and not-lending through an AMC are included in the sample.
The coeﬃcients reﬂect the marginal eﬀects of the respective variables.
Panel B reports the results of the following linear regression model:
AMC Lending i,t = Y eart + BHCi + β1 Equity Capital i,t−1 + β2 BHC Controls i,t−1 + εi,t ,
where AMC Lending i,t , is the volume of a BHC lending through an MC subsidiary normalized by total assets. Only BHC-year
observations with above zero lending through AMCs are included in the analysis.
Equity Capital i,t is the ratio of book value of equity to total assets at the beginning of the year. D2002 (D2005 ) is a dummy
variable equal to one for years after and including 2002 (2005). The set of control variables includes Liquidity (Securities/Total
Assets), Cost of Deposits, Deposits/Total Assets, BHCs’ Size (log of total assets) and loan portfolio structure (Mortgages/Total
Assets, and C &I Loans/Assets). All speciﬁcations include year and BHC ﬁxed eﬀects. Standard errors are clustered at BHC
level. Absolute values of t-statistics are reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Panel A: Establishing MC Subsidiary

Panel B: Extent of MC Lending

–1.91***
(2.92)

–1.41**
(1.99)

–3.57***
(4.49)

–1.18***
(2.61)

–0.68**
(2.40)

–1.40**
(2.57)

Equity Capitalt−1 × D2002

–

4.45***
(5.15)

–

0.99
(1.64)

Equity Capitalt−1 × D2005

–

–1.06*
(1.66)

–

–1.01*
(1.76)

Liquidityt−1

–

–0.63***
(2.70)

–0.52**
(2.24)

–

–0.08
(0.52)

–0.04
(0.26)

Cost of Depositst−1

–

3.38
(1.19)

3.38
(1.20)

–

–1.45
(0.77)

–1.32
(0.70)

Log(Assets)t−1

–

0.21***
(5.83)

0.24***
(6.58)

–

0.025
(0.97)

0.03
(1.33)

Deposits/Assetst−1

–

–0.19
(0.80)

–0.13
(0.57)

–

–0.89***
(5.79)

–0.85***
(5.53)

Mortgages/Assets

t−1

–

–0.01
(0.03)

0.03
(0.15)

–

0.30*
(1.91)

0.32**
(2.04)

C&I Loans/Assetst−1

–

–0.19
(0.54)

–0.20
(0.56)

–

–0.22
(0.94)

–0.21
(0.92)

Yes
Yes

2,576

2,170

3,028

2,557

0.30

0.35

0.37

0.58

0.54

Equity Capitalt−1

Year Fixed Eﬀects
BHC Fixed Eﬀects
Number of Observations
R

=
β2 P ostQ1, 2007 × OTD through AMC 2006
+ γ1 P ostQ1, 2007 × Loan Portfolio Controls 2006
+ γ2 BHC Controls it + εit ,
i
i

BHCi + Quartert + β1 P ostQ1, 2007 × Total OTD 2006
+
i

0.44*
(1.84)
5,076

C&I Loans/Assets

Number of Observations

0.49

0.25*
(1.87)

Liquidity

4.88**
(2.02)

Cost of Deposits

0.43
(0.45)

Equity Capital

–0.18***
(3.32)

log(Assets)

–
–

AMC Subprime × P ost Q1 2007
–0.30***
(2.61)

–
–

Subprime × P ost Q1 2007

Originations × P ost Q1 2007

–
–

0.06***
(2.64)

OTD through AMC × P ost Q1 2007

Total OTD × P ost Q1 2007

0.64

5,013

0.18
(0.81)

0.25
(1.40)

3.67**
(2.42)

0.60
(0.86)

–0.14**
(2.12)

–0.29
(1.60)

–
–

–0.17***
(2.72)

0.13***
(2.86)

0.73

5,013

0.15
(0.80)

0.22
(1.37)

2.40**
(1.97)

0.70
(1.04)

–0.12*
(1.96)

–0.17***
(2.64)

–0.63***
(7.07)

0.60***
(8.46)

0.01
(0.77)

–0.01
(0.35)

Panel A: Net Mortgage
Chargeoﬀs

0.58

5,076

0.27
(0.40)

–0.35
(1.15)

4.14
(1.34)

–2.68
(1.41)

0.057
(0.41)

0.08
(0.29)

–
–

0.05*
(1.87)

0.60

5,013

0.07
(0.10)

–0.35
(0.85)

3.15
(1.00)

–2.51
(1.30)

0.16
(0.81)

0.15
(0.72)

–
–

–0.20***
(3.06)

0.12***
(2.65)

0.61

5,013

0.07
(0.09)

–0.39
(0.96)

1.90
(0.64)

–2.42
(1.23)

0.18
(0.87)

0.26**
(2.11)

–0.76***
(7.62)

0.61***
(5.40)

0.002
(0.06)

–0.01
(0.31)

Panel B: Nonperforming
Mortgages

clustered at the BHC-level. Absolute values of z-statistics are reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

0.91

5,076

–0.39***
(2.58)

–0.30**
(2.57)

0.02
(0.03)

0.44
(0.92)

0.03
(0.82)

0.04
(0.79)

–
–

–0.004
(0.38)

0.92

5,013

–0.31
(1.29)

–0.26
(1.41)

0.602
(0.50)

0.38
(0.51)

0.04
(0.90)

0.04
(0.51)

–
–

0.05***
(2.84)

–0.02**
(2.37)

0.92

5,013

–0.30
(1.30)

–0.25
(1.40)

0.96
(0.79)

0.36
(0.47)

0.04
(0.84)

0.01
(0.11)

0.22***
(4.53)

–0.17***
(5.00)

-0.01
(0.84)

0.01
(1.18)

Panel C: Net Income

Deposits/Total Assets, Size, Mortgages/Total Assets, and C &I Loans/Assets. BHCi denotes BHC ﬁxed eﬀects and Quartert are quarter ﬁxed eﬀects. The standard errors are

mortgages originated in 2006 normalized by the beginning-of-the-year mortgages on BHC balance sheet. BHC Controls are Liquidity (Securities/Total Assets), Cost of Deposits,

) is the total (AMC originated) volume of subprime
(AMC Subprime 2006
2006 normalized by the beginning-of-the-year mortgages on the BHC’s balance sheet. Subprime 2006
i
i

by the beginning-of-the-year mortgages on BHC balance sheet. Similarly, OTD through AMC 2006
is the volume of OTD mortgages originated through AMC of a BHC in
i

dummy variable set to zero for quarters before and including 2007Q1, and one after 2007Q1. Total OTD 2006
is the volume of OTD mortgages originated in 2006 normalized
i

the mortgages scaled by the outstanding mortgages at the beginning of the quarter (Panel B), or net income normalized by the total assets (Panel C). P ostQ1, 2007 is a

On-balance-sheet Loan Losses it , is either the mortgage chargeoﬀs of BHC i during quarter t scaled by the outstanding mortgages at the beginning of the quarter (Panel A),

On-balance-sheet Loan Losses it

This table reports the results from the following regressions:

Table 4: BHC Mortgage-related Losses and MC Activity.

Table 5: Aﬃliated MCs and Banks Lending Standards: Representative Borrower Analysis.
This table reports the results from the following regressions:
Share of AMC Lending izt

β1 Low Credit Score zt + β2 Loan-to-Income Ratio izt + β3 Relative Borrower Income izt +
Market-Year ct + BHCi + γBHC Controls izt−1 + εizt ,

where Share of AMC Lending izt is a ratio of AMC lending to total mortgage lending within BHC i, ZIP code z and year t.
Low Credit Scorezt is a fraction of people with credit scores below 660 residing in a given ZIP code. Loan-to-Income Ratioizt
is a loan-to-income ratio of a BHC-ZIP-code-year representative borrower. Relative Borrower Incomeizt is a borrower’s income
relative to median CBSA income. BHC Controls are Liquidity, Cost of Deposits, Deposits/Total Assets, Size, Mortgages/Total
Assets, and C &I Loans/Assets. Market-Year ct are CBSA-year ﬁxed eﬀects and BHCi are BHC ﬁxed eﬀects. Columns (1)
through (3) present the analysis for the full sample of BHC loans. The results reported in column (4) are based on the sample of
only prime BHC loans, results in column (5) are based on the privately securitized BHC loans, results in column (6) are based
on loans originated in geographies where BHCs do not have branches, and the results in column (7) are based on a sample where
AMCs have names consonant with their parent BHCs. The unit of observation is BHC-ZIP-code-year. The sample contains
36.3 million loans originated between 1999 and 2006. Standard errors are clustered at the CBSA-year level. Absolute values of
t-statistics are reported in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Full
Sample

Prime
Loans

Securitized No-Branch Same Name
Loans
Loans
MC Loans

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Low Credit Score

0.05***
(3.01)

0.04***
(2.74)

0.03**
(2.28)

0.01*
(1.71)

0.03**
(2.28)

0.07***
(8.66)

0.03**
(2.02)

Loan-to-Income Ratio

0.01***
(8.12)

0.02***
(7.69)

0.02***
(7.71)

0.02***
(9.78)

0.02***
(7.71)

0.01***
(7.33)

0.02***
(7.50)

Borrower/Area Income

–0.001*** –0.003*** -0.003***
(3.29)
(8.12)
(8.09)

–0.002***
(4.75)

–0.003***
(8.09)

–0.002***
(4.19)

–0.003***
(7.52)

Share of Minority

–0.04***
(5.38)

–0.02***
(4.07)

–0.02***
(4.13)

–0.02***
(5.33)

–0.02***
(4.13)

–0.01*
(1.83)

–0.02***
(3.67)

House Price Appreciation, t-1

–0.39***
(7.45)

–0.35***
(8.59)

–0.35***
(8.60)

–0.03
(1.44)

–0.35***
(8.60)

–0.25***
(5.35)

–0.34***
(8.18)

Subprime Lending (LP)

–

–0.02*
(1.84)

–0.08***
(7.67)

–0.02*
(1.84)

0.08***
(9.00)

–0.02
(1.38)

Securities/Assets

–

–0.01
(1.15)

–0.01
(1.34)

–0.33***
(5.22)

–0.01
(1.34)

0.03***
(4.13)

–0.08***
(4.48)

Log(Assets)

–

2.40***
(6.33)

2.42***
(6.39)

–0.38***
(21.76)

2.42***
(6.39)

3.87***
(10.51)

1.91***
(4.00)

Cost of Deposits

–

–0.21***
(4.21)

–0.22***
(4.34)

–5.48***
(9.68)

–0.22***
(4.34)

–0.08**
(2.00)

–0.08
(0.86)

Deposits/Assets

–

–0.63***
(6.67)

–0.64***
(6.87)

–0.66***
(19.29)

–0.64***
(6.87)

0.15*
(1.80)

–0.46***
(5.37)

Capital/Assets

–

–2.99***
(18.61)

–2.99***
(18.57)

6.15***
(17.84)

–2.99***
(18.57)

–2.26***
(9.32)

–4.00***
(7.20)

Mortgages/Assets

–

–1.58***
(34.83)

–1.59***
(35.39)

–0.60***
(5.59)

–1.59***
(35.39)

–1.15***
(23.13)

–1.41***
(26.45)

C&I Loans/Assets

–

–0.32**
(2.49)

–0.32**
(2.50)

–0.13
(0.83)

–0.32**
(2.50)

–0.40***
(3.82)

–0.53***
(3.16)

BHC Fixed Eﬀects

Yes

CBSA× Year Fixed Eﬀects

Yes

Number of Observations

1,492,948

1,137,775

372,698

1,137,775

684,353

1,111,298

0.45

0.47

0.75

0.47

0.48

0.46

Adjusted R

Table 6: Aﬃliated MCs and Aﬃliated Banks Lending Standards: Loan-level Analysis.
This table reports the results of the following regression analysis:
AMC Loan i

γ1 Borrower Characteristics i + γ2 Loan Characteristics i +
γ3 Controlsizt + Market-Year ct + BHCj + εi ,

where AMC Loan i is a dummy variable that equals one if a loan was originated by an AMC and zero otherwise. BHC controls
include Liquidity (Securities/Total Assets), Cost of Deposits, Deposits/Total Assets, Size (log of total assets), Mortgages/Total
Assets, and C &I Loans/Assets. We also control for House Price Appreciation, (t-1): the ZIP code house price appreciation over
one year prior to loan origination. We include CBSA-year and BHC-speciﬁc ﬁxed eﬀects. Standard errors are clustered at BHC
level. Columns (1) and (2) are based on the entire sample of loans. Column (3) contains the sample to loans originated in CBSAs
where both bank and MC aﬃliates lend in a given year. Column (4) is based on the sample of privately securitized mortgages
only. The sample contains loans originated over 2005-2006 period. The coeﬃcients are multiplied by 100 for tractability.
Absolute values of t-statistics are reported in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

(1)

(2)

(3)

(4)

Low Credit Score

2.56***
(11.62)

2.27***
(10.34)

3.67***
(9.94)

3.74***
(12.18)

Loan-to-Value Ratio

0.001**
(2.05)

0.001***
(2.65)

0.001***
(4.66)

0.001***
(6.77)

–0.05
(0.45)

–0.02
(0.15)

–0.26*
(1.88)

–0.68***
(3.66)

House Price Appreciation, (t-1)

7.51***
(7.79)

2.46***
(2.61)

5.15***
(3.09)

13.34***
(5.41)

Initial Interest Rate

–0.57***
(4.93)

–0.29**
(2.24)

–0.54***
(2.95)

1.68***
(5.93)

Adjustable Rate Mortgage

0.78***
(2.73)

0.69**
(2.46)

0.94**
(2.17)

2.94***
(5.77)

Interest Only Mortgage

1.30***
(8.32)

1.25***
(8.33)

1.58***
(5.90)

-0.74
(1.57)

Full Documentation

2.42***
(6.13)

2.37***
(5.96)

4.07***
(7.64)

2.33***
(6.09)

First Lien

–5.15***
(9.00)

–5.14***
(9.44)

–8.67***
(10.09)

–20.12***
(19.40)

Prepayment Penalty

–8.49***
(9.52)

–8.33*** –11.66***
(9.50)
(10.39)

–6.59***
(4.09)

log (Borrower Income)

BHC Fixed Eﬀects

Yes

CBSA × Year Fixed Eﬀects

Yes

BHC Controls Included

Yes

Joint Bank-MC in CBSA Restriction

Yes

924,721

396,479

0.80

Number of Observations
Adjusted R

1,373,098 1,373,098
0.86

0.86

Table 7: Mortgage Defaults: Loan-level Analysis.
This table reports the results of the following regression analysis:
Mortgage Default i

β0 AMC Loan i + β1 Borrower Controls i + β2 Loan Controls i + β3 BHC Controls i +
β4 House Price Appreciation z,t + Market-Year ct + BHCk + εz ,

where Mortgage Default i is a dummy variable that equals one if a loan was 60 or more days past due, in severe delinquency, in
foreclosure, or real-estate owned within two years since origination, and zero otherwise. AMC Loan i is a dummy variable that
equals one if a loan was originated by an aﬃliated MC. House Price Appreciation, (t+2) is the ZIP code house price appreciation
over two years after a loan was originated. BHC Securitization Activity is the share of securitized loans among those originated
by all BHC aﬃliates in a given year. BHC controls include Liquidity (Securities/Total Assets), Cost of Deposits, Deposits/Total
Assets, Size (log(Total Assets)), Mortgages/Total Assets, and C &I Loans/Assets. All speciﬁcations include CBSA-Year and
BHC-speciﬁc ﬁxed eﬀects. The sample contain loans originated over 2005-2006 period. Standard errors are clustered at BHC
level. In column (3) we change the sample and only consider loans originated in CBSAs where both bank and MC aﬃliates lend
in a given year. The coeﬃcients are multiplied by 100 for tractability. Absolute values of t-statistics are reported in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

MC Dummy

(1)

(2)

(3)

(4)

(5)

(6)

(7)

5.62***
(14.45)

5.74***
(14.13)

5.58***
(13.94)

-43.25***
(7.99)

11.95***
(8.88)

13.88***
(12.10)

–2.65***
(6.08)

MC Dummy × Size

2.50***
(9.40)

MC Dummy × Equity Capital

–74.27***
(4.38)

MC × BHC Secur. Activity

–11.54***
(7.50)

MC × BHC Share MC Lending

14.02***
(19.95)

Low Credit Score

11.42***
(45.07)

11.32***
(42.96)

11.50***
(39.08)

11.29***
(42.87)

11.30***
(42.98)

11.32***
(42.80)

11.22***
(42.97)

Loan-to-Value Ratio

0.001**
(2.23)

0.001**
(2.55)

0.001**
(3.79)

0.001**
(2.58)

0.001**
(2.52)

0.001***
(2.64)

1.80**
(2.42)

Loan-to-Income Ratio

0.79***
(8.65)

0.78***
(8.66)

0.65***
(7.21)

0.78***
(8.54)

0.78***
(8.64)

0.78***
(8.62)

0.78***
(8.65)

–19.83***
(15.85)

–21.55***
(16.14)

–21.55***
(16.12)

–21.39***
(15.73)

-21.57***
(-16.11)

House Price Appreciation, (t+2)
log (Borrower Income)

–23.49*** –21.51***
(16.85)
(16.06)
0.13
(0.61)

0.05
(0.26)

–0.15
(0.75)

0.01
(0.06)

0.05
(0.24)

0.04
(0.18)

0.03
(0.16)

Initial Interest Rate

3.19***
(26.71)

3.12***
(24.72)

3.28***
(21.89)

3.17***
(24.86)

3.12***
(24.80)

3.16***
(24.99)

3.11***
(25.27)

Adjustable-Rate Mortgage

3.94***
(18.61)

4.02***
(19.18)

4.99***
(20.82)

3.93***
(18.27)

4.01***
(19.09)

4.00***
(18.75)

3.94***
(19.18)

Prepayment Penalty

5.72***
(9.96)

5.90***
(10.58)

5.53***
(10.63)

6.01***
(10.64)

5.95***
(10.66)

5.91***
(10.52)

6.04***
(11.17)

Interest-Only Mortgage

2.37***
(16.29)

2.54***
(17.17)

2.53***
(16.48)

2.47***
(16.85)

2.54***
(17.01)

2.54***
(17.13)

2.45***
(16.93)

Full Documentation

–2.68***
(12.76)

–3.21***
(13.25)

–2.24***
(5.99)

–3.26***
(13.38)

–3.23***
(13.19)

–3.22***
(13.30)

-3.28***
(-13.49)

First Lien

–5.52***
(9.86)

–5.83***
(10.73)

–6.13***
(9.94)

–6.01***
(11.06)

–5.80***
(10.69)

–6.03***
(11.29)

-5.69***
(-10.73)

Securitized through GSEs

–1.57***
(11.17)

–1.51***
(11.21)

–1.03***
(5.44)

–1.71***
(12.66)

–1.53***
(11.45)

–1.60***
(12.02)

-1.66***
(-12.28)

Privately Securitized

–3.46***
(14.97)

–3.16***
(13.22)

–1.94***
(4.55)

–3.22***
(13.15)

–3.15***
(13.32)

–3.15***
(12.84)

-3.08***
(-13.41)

Yes
Yes
No
No

Yes
Yes
Yes
No

Yes
Yes
Yes
Yes

Yes
Yes
Yes
No

1,373,098

924,721

1,373,098

0.17

0.18

0.17

BHC Fixed-Eﬀects
CBSA × Year Fixed-Eﬀects
BHC Controls Included
Within the same BHC-CBSA-Year
Number of Observations
R2

Table 8: Impact of MC Activity on Delinquencies and Foreclosures.
This table reports the results from the following regressions:
Default Rate z = CBSAj + β1 MC Share z,Pre-Crisis + β2 Controlsz + εz ,
where Default Rate z is either Delinquency Rate z , a fraction of mortgage holders in a ZIP code that were delinquent on their
loans in any period of time between January 2007 and December 2008, or Foreclosure Rate z , a fraction of mortgage holders
in a ZIP code that were in foreclosure in any period of time between January 2007 and December 2008. MC Share z,Pre-Crisis
is a share of loans originated by aﬃliated and independent MCs over the 2005-2006 period. The set of controls include (i)
the borrower characteristics measured based on a 2005-2006 pool of ZIP code borrowers; (ii) house price appreciation over
the 2007-2008 period; and (iii) extend of subprime originations and securitization activity in a ZIP code in 2005-2006. All
speciﬁcations include CBSA ﬁxed eﬀects. Standard errors are clustered at CBSA level. The coeﬃcients are multiplied by 100
for tractability. Absolute values of t-statistics are reported in parentheses.
*** p<0.01, ** p<0.05, * p<0.1

Delinquency
Measured over 2005–2006

Foreclosure

(1)

(2)

(3)

(4)

(5)

(6)

12.27***
(7.82)

–
–

9.59***
(4.69)

–
–

Prime MC Share

–
–

3.53***
(3.38)

–
–

3.74***
(2.66)

–
–

Aﬃliated MC Share

–
–

12.38***
(4.93)

–
–

7.76*
(1.93)

Independent MC Share

–
–

12.26***
(7.51)

–
–

9.68***
(4.70)

18.16***
(14.16)

8.35***
(9.51)

18.16***
(14.33)

Loan-to-Income Ratio

0.13
(0.40)

0.75*
(1.82)

0.13
(0.41)

–0.07
(0.20)

0.24
(0.69)

–0.08
(0.24)

Borrower/Area Income

–0.10
(1.23)

0.27***
(2.92)

–0.10
(1.21)

–0.02
(0.22)

0.15
(1.25)

–0.01
(0.07)

1.57***
(3.48)

0.18
(0.53)

1.57***
(3.49)

–2.24
(1.03)

–1.71
(1.16)

–2.24
(1.04)

0.37
(0.17)

0.59
(0.28)

0.31
(0.14)

4.03***
(6.15)

–
–

4.02***
(6.17)

1.91***
(3.35)

–
–

1.94***
(3.50)

–
–

17.99***
(22.15)

–
–

9.75***
(7.93)

–
–

–1.03
(0.62)

2.56
(1.60)

–1.03
(0.64)

–1.00
(0.41)

1.41
(0.53)

–0.94
(0.39)

Yes

Number of Observations

6,190

0.78

0.82

0.78

0.68

0.69

0.68

MC Share

Low Credit Score

Share of Minority
House Price Appreciation 2007-2008
Subprime Lending (LP)
Subprime Lending (HMDA)
Share Securitized
CBSA Fixed Eﬀects

9.02*** 4.18*** 9.04***
(8.74)
(3.79)
(8.60)

2.94*** 2.28*** 2.93***
(5.78)
(5.71)
(5.68)

Appendix

HSBC Finance
“With the beneﬁt of hindsight the Group wishes it hadn’t made the investment,”
Mike Georghegan, CEO of HSBC Holdings
HSBC Finance Corporation started its corporate life in 1878 under the name of Household Finance
Corporation of Minneapolis.34 It was built on small personal loans and served mostly smaller
consumers who were potentially priced out of mainstream ﬁnancial services. When the subprime
loan market started growing at the turn of the century, Household Finance quickly became a big
player in this market niche. By 2002 Household Finance already had a slightly shady reputation by
lending to customers not served by banks and engaging in aggressive collection practices. So the
2002 decision of the British banking giant, HSBC Group, to acquire Household Finance was a bit
of a surprise to the market. The New York Times noted that The merger established that lending
to subprime customers was a respectable business.35
The new nondepository subsidiary of HSBC North America Bank Holding Company, a core
US subsidiary of HSBC Group, was quickly renamed HSBC Finance. Under the new name, the
company started to rapidly expand HSBCs consumer-ﬁnance oﬀerings to subprime borrowers. By
2006, HSBC Finance was on top of the mortgage-ﬁnance charts with almost 8% market share in the
subprime market. The chief executives of the conglomerate described HSBC Finance as “a dream
portfolio” envied by others in the industry.
Under the HSBC umbrella, the company restructured into two mortgage divisions: a wholesale
channel, Mortgage Services, and a retail channel, Consumer Lending. Mortgage Services aggressively expanded into stated-income loans. By 2006, it is estimated that HSBC Finance held $11.8
billion in stated-income loans on its books. The machine did not stop at the ﬁrst signs of the mortgage markets weakness. In 2007, Mortgage Services originated another $7.8 billion of stated-income
loans.
Consumer Lending did not fall behind. HSBC Finance described the business model of their
retail arm as “focused on subprime customers who rely on drawing cash against the equity of their
34
35

http://www.fundinguniverse.com/company-histories/household-international-inc-history/
“The Deal that Fueled Subprime,” New York Times, March 2009.

homes to help meet their cash needs.” In May 2007, HSBC Finance securitized a pool of loans in
which almost a quarter of borrowers owed more on their houses than the homes were worth and
more than 50% of the borrowers had a track record of not paying their bills.36
Considering the lending philosophy of HSBC Finance, it is not surprising that the business
imploded during the ﬁnancial crisis. What is surprising is that the executives of the company
continued to push their lending agenda well into late 2007 and even early 2008. For example,
HSBC decided to stop originating stated-income loans only in December 2007, well after federal
regulators warned about the hidden dangers of these loans.
HSBC Finance reported drastic losses in 2007 and 2008. But it was only in March 2009 that
HSBC executives ﬁnally threw in the towel. The announcement came 23 months after New Century
Financial Corporation, the second-largest subprime lender, had ﬁled for bankruptcy. It came well
after HSBC Finance executives knew about the 2007 pretax loss of $4.9 billion from HSBC Finance
operations, compared with a proﬁt of $1.4 billion in 2006.37 It came after HSBC Finance was named
a core defendant in at least 5 class action lawsuits accusing the company of predatory lending.
After announcing the defeat, the company cut 6,100 jobs and closed almost 800 of its consumer
lending outlets. The executives stated that they “... believe that this is increasingly likely given
that Household Finance is eﬀectively no longer a going concern and that market conditions in the
United States continue to deteriorate...”38
What the major attraction was for HSBC to acquire Household Finance Corporation is debatable. What is clear is that at the time of acquisition, HSBC was already operating 417 depository
branches in New York (mostly), California, Connecticut, Delaware, Washington, D.C., Florida,
Maryland, New Jersey, Pennsylvania, Virginia, and Washington.39 In 2003, the HSBC North
America BHC held $26 billion in family mortgages on its balance sheet, which already represented
a signiﬁcant presence in the mortgage market.40 HSBC depository subsidiaries had little presence
in the high-yield or alternative-products segments of the market.41 One might argue that the BHC
wanted to expand to a high-return segment of the consumer ﬁnance market, but not by expanding
36

HSBC Home Equity Loan Trust (USA) 2007-2. Engel and McCoy(2011).
HSBC Finance Corporation, 2008 Annual Report.
38
“HSBC to Raise $17.7 Billion as Subprime Cuts Proﬁt,” Bloomberg, March 2009.”
39
Summary of deposits report, FDIC.
40
BHC Call Reports, Schedule 9Y-C for HSBC North America Corp., 2006-2009
41
HMDA data. Engel and McCoy (2011) conﬁrm this inference.
37

lending through its existing infrastructure but rather through a separate nondepository entity.
It is also clear that regulators did nothing to stop the HSBC Finance mortgage machine. Despite
HSBC Finance being the largest originator in the subprime sector, one cannot ﬁnd any enforcement
action by regulators against HSBC Finance before the 2007 crisis. There is also no evidence of
mortgage oversight by regulators over HSBC Finance, an aﬃliate of the BHC, even in the second
half of 2007, when it was obvious that HSBC Finance’s origination practices were predatory in
nature.
Finally, the ﬁnancial reports of HSBC Finance and its immediate parent, HSBC North America
Corporation, leave the question about consolidation practices open. In 2003 (2004), HSBC Finance
reported $49 billion ($61 billion) of real-estate secured loans on its balance sheet.42 Yet HSBC North
America reported only $26 billion ($46 billion) of mortgages in their annual report.43 Similarly,
if one tries to trace whether the $4.9 billion loss incurred by HSBC Finance aﬀected the parent
BHC, HSBC North America, one might ﬁnd it diﬃcult. It is hard to reconcile this enormous loss
of the aﬃliate with the much smaller $2.93 billion loss reported by the parent in 2007. Given that
HSBC North America reported $2.65 billion in proﬁt in 2006, it is unlikely that the same company
increased its proﬁts in 2007 suﬃciently enough to oﬀset the $4.9 billion loss of the subsidiary.

42
43

HSBC Finance Corp 10K Annual Report, 2006.
BHC Call Reports, Schedule 9Y-C for HSBC North America Corp., 2006-2009.