Tempted by scope? Homebuilder mortgage affiliates,
lending quality and the housing crisis
Harvard Business School
*** Preliminary and incomplete – Please do not cite without permission ***
This paper investigates a potential cost of corporate scope: the introduction of moral hazard into an
organization. I examine evidence for moral hazard within homebuilders that engaged in mortgage
lending during the housing boom, specifically, whether managers excessively lowered underwriting
standards to sell more homes. I use a differences-in-differences approach with detailed mortgage data
from 1999 to 2009 to identify this hazard. Contrary to predictions, I find that builders maintained higher
standards than financial firms. Builders’ lending standards also deteriorated more slowly prior to the
mortgage crisis and, as a result, were most restrained relative to other lenders during 2005 and 2006,
when industry-wide standards were lowest. This restraint can be partly explained by the mortgage units’
limited capacity to hold loans, which led them to originate mortgages that could be easily sold to third
parties. This finding provides evidence against the popular perspective that a pure “originate to
distribute” lending model led to poor underwriting. The relatively worse performance of banks and large,
diversified financial institutions suggests that moral hazard may be more prevalent within firms with more
opaque operations rather than those subjected to market discipline. Overall, the study shows that incentive
conflicts caused by engaging in multiple, closely-related activities can be countered, at least in some
cases, by market discipline and appropriate organizational design.
First draft: April 8, 2010
This draft: December 6, 2010
JEL Codes: D21, D23, G21, L22, L25
Key Terms: Organization, incentives, corporate scope, mortgages, securitization
* Harvard Business School, Morgan Hall T81, Soldiers Field, Boston, MA 02163. Phone: 617-378-8710. Email:
firstname.lastname@example.org. I am grateful for data and assistance from the many generous and knowledgeable people at First
American CoreLogic, particularly Sam Khater and Trish Murray. I also thank Pierre Azoulay, Mihir Desai, Robert Gibbons,
Rebecca Henderson, Victoria Ivashina, Venkat Kuppuswamy, Ed Morrison, Tomasz Piskorski, Jan Rivkin, George Serafeim,
Catherine Thomas, Peter Tufano, Paul Willen, Kristin Wilson and Julie Wulf for their advice and very helpful comments. This
paper benefits from comments from participants at CCC at University of Michigan, the Federal Reserve Bank of Boston Urban
and Real Estate Economics Seminar and the research seminar series at CoreLogic.
By rushing into the mortgage business big-time, homebuilders helped fuel the housing crisis…Providing
loans to financially marginal buyers was one way [homebuilders] tried to prop up their financial
performance, says S&P's Smith. "You're trying to support earnings at high levels, so it's conceivable that
greed gets into people's minds."
- “Bonfire Of The Builders,” Businessweek cover
article, August 13, 2007
Can corporate scope lead to costly incentive conflicts within firms? Research in strategy emphasizes
competitive advantage that arises from leveraging intangible resources across related areas (Wernerfelt
1984; Peteraf 1993; Langlois and Foss 1999). Research on organizations, on the other hand, points out
the potential for incentive costs as the breadth of related, incompletely-contractible activities increases
(March and Simon 1958, Holmstrom and Milgrom 1991). A simple integration of these two literatures
suggests that accepting a certain level of incentive conflicts may be a consequence of creating competitive
advantage. This paper explores this proposition empirically by studying the lending practices of large
U.S. homebuilders with in-house mortgage operations. I use detailed mortgage data to investigate
evidence of incentive conflicts within builders.
During the 2000s, large builders sold nearly 30% of new homes in the U.S. and financed up to
70-80% of these homes.1 It has been argued that this combination of selling and financing homes created
the potential for a “hidden action” problem – or moral hazard – within builders, particularly during the
recent housing cycle. This hazard was said to manifest itself by managers’ excessively lowering lending
standards to inflate home sales.
Research on moral hazard has shown that managers can be tempted in this manner if they are
evaluated and compensated primarily on one activity (in this case, home sales). The temptation can
increase if the unrewarded activity (mortgage lending) complements the main activity (Holmstrom and
Milgrom 1991) or if the external environment changes to make subjective measures, such as reputation,
Source: “Bonfire of the Builders,” Businessweek, August 13, 2007 and author’s estimates from 10-K filings and
relatively less effective (Baker, Gibbons and Murphy 1994). Industry conditions during this period may
have created such an environment. An analysis of four representative large builders reveals a 98%
correlation between annual revenues of the homebuilding segment and annualized share price, suggesting
the centrality of home sales to the firms’ market valuation. In contrast, the mortgage segments
contributed only 2% to firm revenue,2,3 but supported home sales during this period: capture rates (the
percent of mortgage-financed homes sold by builders with mortgages funded by their in-house lending
units) averaged 64% and increased throughout the sample period. Furthermore, whereas home sales were
immediately observed, mortgages took months and years to default. These facts suggest that lending
quality distortions would be difficult to monitor directly and could have a large impact on home sales. In
addition, the dramatic market conditions between 2003 and 2006 suggest that the rewards to inflating
home sales during this period were greater than during prior years.
A simple comparison of aggregate mortgage performance appears to support the existence of lax
lending practices. Many of the geographies in which builders were most active – such as Phoenix, Las
Vegas and Miami – were also most affected by the mortgage crisis and subsequent housing collapse. In
the aftermath of the collapse, builder-originated mortgages defaulted at roughly 130-150% of the national
average and national new homes sales fell to historic lows by 2010.4 Multiple shareholder and consumer
lawsuits accused managers of manipulating lending standards and inflating home values, harming both
consumers and the firms themselves by causing more mortgage defaults, inventory write-downs and home
foreclosures.5 A rigorous empirical investigation, however, has not been done.
This setting has several attractive features for a study of scope-induced incentive conflicts: a set
of generally comparable firms, proprietary but available transaction-level data, and the establishment of
The firms also derived approximately 95% of profits from selling homes and less than 5% from funding
mortgages. Source: 10-K filings of Pulte, Lennar, KB Home and DR Horton.
Source: Author’s estimates from 10-K filings and mortgage database, understated in Figure 1b (see Figure 1b
Some cases include: In Re: Beazer Homes USA, Inc . Securities Litigation, Master File No: 1:07-cv-725-CC;
Mark Zachary v. Countrywide Home Loans Inc. and Countrywide KB Home Loans; Sodalin Kaing v. Pulte Homes,
in-house lenders occurring well before 1999. This last feature enables the events in the housing industry
from 1999 to 2009 to be treated as unanticipated at the time of the integration decision.
I construct a new and detailed dataset of 212,058 mortgages originated by all market participants,
both builders and financial firms, in the 100 zip codes that experienced the most new home construction
from January 1999 to mid-2009. I look for evidence that builders lowered lending standards more than
financial firms (both banks and non-depository mortgage lenders) during the boom (1999-2006), when
unprecedented availability of mortgage credit led to widespread deterioration in lending quality
(Dell’Ariccia, Igan and Laeven 2009; Demyanuk and Van Hemert 2009), and after the boom (2007-
2009), when builders were caught with excess inventory during the worst housing market on record.6
I find that, contrary to evidence of moral hazard, builder-originated mortgages perform
significantly better than mortgages from other lenders after controlling for geographic and
macroeconomic factors.7 Further, lending standards deteriorated less for builders than for other lenders
during the housing expansion; consequently, builders exercised their greatest relative restraint in 2005 and
2006, when industry-wide standards were lowest. This finding contradicts the simple proposition, stated
above, that incentive conflicts are a necessary cost of internalizing related activities.
To explain these counterintuitive results, I supplement this analysis with interviews from multiple
sources. Interviewees include builder executives and mid-level managers, as well as account executives
at depository banks, appraisers, industry consultants and secondary mortgage market investors. I propose
that builders, precisely because of their related scope, implemented effective organizational structures that
countered internal pressure to distort lending – acknowledged as the temptation to “buy deep.” Two
scope-related factors drove these organizational choices. First, builders assumed significant inventory
risk between the time when prospective homebuyers engaged in the sales process and when the home
closed, a risk not assumed by pure financial lenders. To reduce the number of buyers exiting during this
process, managers included a customer satisfaction component in the employee compensation plans,
As measured by housing starts. Source: U.S. Census.
These results are consistent with a concurrent study of builder lending quality by Agarwal et al (2010), which
focuses on the subprime market across all geographies.
measured after the sale had closed. This component – reported by some builders to be as significant as
the component linked to lending volumes – effectively blunted incentives to fund greater volumes of
riskier mortgages. Second, because both land acquisition and home construction were capital-intensive
activities, managers limited the funding available to their mortgage units used to hold mortgages prior to
selling them on the secondary market. This limited capacity led these units to focus on mortgages that
could be easily sold and less likely to be put back to them in case of early default or underwriting defects.
I find evidence that this channel increased lending quality.
I also investigate a primary market channel in which builders lent more safely because they
internalized the costs of foreclosure externalities within their subdivisions in the form of lost sales and
diminished pricing power. I do not find supporting evidence for this explanation.
This study faces three identification challenges. First, unobservable differences may drive
consumers to choose between builders and financial lenders. I address this challenge in two ways. First,
to create a common pool of potential borrowers, I restrict the sample to the 100 high growth zip codes –
the geographies dominated by new home construction – and include only conventional, purchase
mortgages of newly constructed, single-family detached homes. Second, I supplement the baseline
specification with a detailed matched sample analysis. I compare default rates between builder and non-
builder mortgages that are pair-wise matched within sale quarters, zip codes or census tracts, and sets of
mortgage and borrower characteristics. I use a variety of matching algorithms and variable specifications
to ensure that the results are robust. Figure 1 shows a representative census tract within which mortgages
were matched. This figure illustrates that, within housing developments, mortgages were supplied by a
mix of in-house and outside lenders, and that no unobserved separation in housing characteristics or
geography drives the results.
<< INSERT FIGURE 1 ABOUT HERE >>
The second challenge is to identify lending quality that reflects managerial moral hazard rather
than optimal firm choices. Builders’ profit functions differ from financial lenders due to both the
potential for tying – cross-subsidization between the two units – and other factors affecting the relative
costs of the firms. As such, a direct comparison of loans originated by builders to those by other lenders
is insufficient. Instead, I use a differences-in-differences approach and measure changes in relative
lending quality between builders and other lenders over time. I treat the years 1999-2002 – prior to the
greatest market growth and reported lending abuses – as the baseline period. Controlling for time-varying
variables that affect efficient relative quality levels, such as margins on home sales, I identify excessive
lending deterioration by comparing relative lending in later periods to this baseline period.
The third challenge is to identify the firms’ limited capacity to hold loans as a channel in
determining builders’ lending behavior. Because I do not observe the relative capacity of a firm to hold
loans, I create a proxy for this capacity as the degree to which each firm priced loans on “hard”
information. The increasing trend of pricing loans using hard information, particularly the borrower’s
FICO8 (credit) score, has been tied to the development of the secondary market, since “soft” information
cannot be verified by third party investors (Rajan, Seru and Vig, 2009). I find a significant negative
association between this proxy and a measure of firm-level lending quality, suggesting that firms with
limited capacities to hold loans maintained higher lending standards. Although direct causality cannot be
formally established without an instrument, I run several tests to establish the validity of this proxy as a
measure of funding constraints and corroborate this interpretation during interviews with builder
executives. This finding suggests that scope-induced moral hazard may have been more pronounced
within firms with more funding options. These funds enabled firms to hold loans for longer periods prior
to sale and, therefore, relaxed the constraint of originating loans with a known third-party buyer.9
FICO stands for “Fair Isaac Corporation,” the firm which created and manages the consumer credit scoring model,
the most widely used model in the world, according to FICO: http://www.fico.com/en/Company/Pages/about.aspx,
accessed June 14, 2010.
This interpretation was also supported by testimony by Keith Johnson before the Financial Crisis Inquiry
Commission on Sept 23, 2010 in which he discussed investment banks repeatedly submitting defecting loans to
This study makes two contributions to research on firm scope and competitive advantage. First, it
introduces the following tradeoff associated with internalizing related scope within firms: the firm may
benefit from leveraging resources across multiple activities; however, these activities may also impose
incentive conflicts on the organization. The costs of these conflicts must be considered when assessing
the appropriate breadth of activities within a firm. Second, the study suggests that the simplest
interpretation of this tradeoff – that internalizing related, incompletely-contractible activities necessarily
leads to incentive conflicts – is incomplete. The combination of organizational design and market forces
may create observable performance measures for otherwise unobservable actions. In the case of builders,
the requirement to dispose of mortgages quickly and thoroughly on the secondary market, combined with
effective implementation of incentives and reporting structures, countered temptations to engage in risky
lending (at least, relative to financial lenders). This latter finding suggests that scope-induced moral
hazard may be more pronounced in firms with more opaque operations.
This paper also makes two contributions to the large and growing literature on the causes and
consequences of the recent housing crisis (e.g., Shiller 2008; Mayer, Pence and Sherlund 2009; Foote et al
2008). First, it provides evidence contrary to the perspective that the “originate to distribute” lending
model, on its own, led to poor underwriting practices during this period (Purnanandam 2010). A high-
securitization environment may have indeed contributed to poor systemic underwriting (Keys et al 2010a;
Rajan, Seru and Vig 2009) and exacerbated the foreclosure crisis (Piskorski, Seru and Vig 2010). Within
this environment, however, I find that firms with limited capacity to hold loans before selling them to
third parties were more disciplined than firms with more funding options. Second, it shows that the
systemic deterioration in lending quality over the course of the boom (Dell’Ariccia, Igan and Laeven
2009) was not evenly shared by originators. In fact, some of the firms which displayed the greatest
deterioration – depository banks – were also the most regulated, a finding similar to Keys et al (2009).
mortgage securitization pools until they were accepted, http://fcic.gov/hearings/pdfs/2010-0923-transcript.pdf, page
The rest of this paper is organized as follows: Section I discusses corporate scope conflicts and
background on the 2007-2010 housing crisis. Section II investigates evidence for incentive conflicts
within builder with mortgage affiliates. Section III examines explanations for how these conflicts were
countered. Section IV discusses the implications of the study findings and concludes.
I. Literature and background
A. Corporate scope conflicts
Firms with related scope are generally viewed favorably (Teece 1980; Wernerfelt 1984; Teece et al 1994),
and perform well compared to firms with unrelated scope (Montgomery and Wernerfelt 1988; Lang and
Stulz 1994; Berger and Ofek 1995; Villalonga 2004). Internalizing related scope has been shown to
confer many advantages to firms, including the benefits of shared knowledge (Chandler 1992; Kogut and
Zander 1992; Henderson and Cockburn 1996), knowledge production (Nickerson and Zenger 2004),
coordination (Williamson 1975, Chandler 1990), and shared competence and non-divisible assets across
businesses (Teece 1980; Teece 1982). In addition, strategy research has also shown that mutually
reinforcing activities, or tight fit, can be a key source of competitive advantage (c.f., Siggelkow and
Scope, however, can impose costs on firms. Aside from a large literature on firm inertia (e.g.,
Leonard-Barton 1992; Siggelkow 2001; Henderson, Greenstein and Bresnahan 2010), studies on the costs
of related (non-vertical) scope and tight fit have been limited. Organizations research, within both
management and economics, has long recognized the existence of incentive conflicts within large
organizations (Simon 1947). Two seminal theoretical contributions on the tradeoffs associated with scope
– transactions cost economics (Williamson 1975, 1985) and property rights literature (Grossman and Hart
1986; Hart and Moore 1990) – have focused on the costs of these conflicts. This work, however, has
implicitly emphasized vertical exchange in which buyers and suppliers transact in an environment of
opportunism and incompletely contractible relationships. It is unclear whether the underlying
assumptions of vertical exchange, as well as the associated empirical findings (Shelanski and Klein 1995;
Lafontaine and Slade 2007), generalize outside of a vertical setting. While Teece and co-authors have
extended transactions cost economics to horizontal scope (c.f., Teece et al 1994), this work has placed
more emphasis on the benefits of coherence across related areas rather than incentive conflicts potentially
introduced with the integration.
The internal capital markets literature within corporate finance has studied one potential channel
through which corporate scope imposes costs on organizations: rent-seeking by division managers during
the capital allocation process (Scharfstein and Stein 2000). This conflict, while an important cost to
diversified firms, does not directly address the costs of unprofitable choices by firm-level managers as a
consequence of horizontally-related and unevenly monitored scope, the focus of this paper.
These costs may not be inconsequential. Since 2000, firms – and managers within those firms –
in a variety of industries have been suspected of incentive conflicts induced by related activities.
Examples include accounting firms with auditing and consulting services,10 investment banks with equity
research,11 energy companies with commodity trading units12 and, most recently, investment banks with
proprietary trading and brokerage functions.13 Yet, comparatively little attention has been paid to these
potentially costly conflicts within strategic management and organizational economics.
Following Kaplan and Henderson (2005), one approach to studying these costs is to adapt models
of incentives from organizational economics. A series of work by Holmstrom and Milgrom (1991, 1994;
Holmstrom 1996) and Baker, Gibbons and Murphy (1994) describes moral hazard problems within firms.
This hazard can arise when firms internalize activities that require employees to perform multiple,
unequally-monitored tasks (in the case of Holmstrom and Milgrom) or tasks with imperfect objective
measures of performance (in the case of Baker, Gibbons and Murphy). Firm owners must deploy
The Sarbanes-Oxley Act of 2002, Section 201 banned auditing firms from providing certain consulting services;
however, empirical studies have generally found little evidence of worse auditing as non-audit fees increases
(Ashbaugh, LaFond and Mayhew 2003; Kinney, Palmrose and Scholz 2004; Romano 2005)
“Spitzer Raises the Heat on Citigroup,” Businessweek, October 2, 2002,
http://www.businessweek.com/bwdaily/dnflash/oct2002/nf2002102_2153.htm, accessed June 14, 2010.
“Caution: Energy Trading; Today the Power Industry Fears the Risk”, Washington Post, July 26, 2002; “El Paso
Planning to Leave Energy Trading Business,” New York Times, Nov 9, 2002; “Energy Trader Admits Faking
Transactions,” New York Times, May 14, 2002, “BP to Pay $303 Million to End U.S. Propane Probe,” Bloomberg,
Oct 23, 2007.
“Clients Worried about Goldman’s Dueling Goals,” New York Times, May 19, 2010.
complex incentives and organizational design to counter this moral hazard, including flatter compensation
and more subjective evaluations of success than one would observe in market transactions. This view has
been tested empirically to investigate outsourcing choices by Azoulay (2004) and Baker and Hubbard
(2003). To my knowledge, however, this work has not been applied to understanding how corporate
scope distorts behavior within firms.
In summary, this study represents an effort to unify two perspectives of the consequences of
internalizing related scope with firms – the competitive advantage view that internalizing these activities
is a source of the firm advantage and the organizational view that it can create costly incentive-related
distortions within a firm.
B. Industry structure
The housing boom began in the 1990s and rapidly accelerated from 2002 to 2006. It was followed by the
collapse of the housing sector, the largest decline in home prices since World War I (Shiller 2000), and
the most severe financial crisis since the Great Depression (Mian and Sufi 2009). Because of the
complexity of the institutional setting, I provide a brief description of the industry and timeline of the
The residential housing market can be divided into resales of existing homes and sales of new
construction. Because of structural differences between resale and new construction markets, I focus on
new construction in this study.14 Since World War II, the majority of new homes have been constructed
within residential subdivisions (Duany, Plater-Zyberk and Speck 2000). These subdivisions consist of
large land lots purchased by developers or builders who then construct planned communities that can
include thousands of homes. These subdivisions may require multiple years or decades to sell
According to interviewees, the resale market has underperformed the new construction market since at least the
end of World War II and, therefore, excluding resales is conservative. Consistent with this assertion, when I include
resales in the analysis, the results strengthen. Agarwal et al (2010) include resales in their study and find similar
completely, and so builders assume substantial land and home inventory risk with these large land
Mortgage markets, described in detail in Jacobides (2004), include both the primary market to
supply loans to homebuyers and the secondary securitization market, in which the loans are sold by
originators to securitizers (or transferred to a different division within the same firm) and pooled into
trusts that issue and sell mortgage-backed securities.15 These securities can then be sold to investors such
as pension funds, hedge funds and sovereign wealth funds (Coval, Jurek and Stafford 2009).
Mortgage originators can be categorized into four types that correspond to the scope of the parent
firms and are used in later analyses: builders’ financing affiliates, depository banks (including both
commercial banks such as Bank of America and thrifts such as Washington Mutual), non-depository
standalone mortgage banks (such as Countrywide16) and mortgage banks that are affiliated with larger
firms (including the lending arms of AIG, GE Capital, Lehman Brothers, Merrill Lynch and Bear
Mortgages can be classified as “conforming” and “non-conforming” loans. Conforming
mortgages adhere to the underwriting criteria of the two major government-sponsored enterprises
(hereafter GSEs) chartered with purchasing mortgages, FNMA (“Fannie Mae”) and FHLMC (“Freddie
Mac”).18 The GSEs agree to purchase conforming loans and then issue mortgage-backed securities that
include a GSE guarantee against mortgage defaults. In contrast, non-conforming loans, which include
loans to prime borrowers above the jumbo threshold, to borrowers without complete loan documentation
or to subprime borrowers who do not meet the credit standards to qualify for a conforming loan, have no
government support and are considered riskier than GSE-backed loans.19 These non-conforming
Jacobides (2004) provides a detailed discussion of the evolution and specialization of the mortgage banking
Prior to its sale to Bank of America.
Many of these lenders were known to specialize in subprime lending, and many are no longer in operation.
Conforming mortgages are made to prime borrowers (generally defined as borrowers with FICO (credit) scores
over 620), under a certain dollar amount (the “jumbo loan” limit) with certain loan attributes, such as full loan
documentation, loan-to-value limits, and no negative amortization or interest-only provisions.
Including Alt-A, mortgages with risk levels considered between prime and subprime.
mortgages are generally pooled into private-label (non-GSE) mortgage-backed securities by financial
institutions and sold to third party investors. See Appendix A.1 for a brief discussion of secondary
B.1 Industry timeline
The first half of the 2000s was marked by several trends in the U.S. market: 100% home price
appreciation between 2000 and 2006,20 increased monthly housing starts from 1.5 million in mid-2000 to
2.3 million in early 2006, and increased mortgage originations from $1 trillion in 2000 to $2.7 trillion in
2006.21 Concurrent with these trends was an increase in securitization, combined with an increasing
reliance on pricing loans based on “hard information,” particularly a consumer’s FICO score (Rajan, Seru
and Vig 2009).
Beginning in 2005 and 2006, national home prices and sales both leveled off. Forced repurchases
of loans started to increase in early 2005 as loan quality deteriorated.22 In the first half of 2007, the
private-label (non-GSE) securities market effectively shut down, leading to the failure of numerous
mortgage lenders.23 By January 2010, national home prices had fallen 30% from their peak, including
55%, 50%, and 46% in Las Vegas, Phoenix, and Miami, respectively.24 By March 2010, 24% of all U.S.
residential properties with mortgages were worth less than the mortgages on the properties, including
70% of all properties in Nevada, 51% in Arizona and 48% in Florida.25 Annualized new home sales were
measured at a record low of 267,000 in May 2010 versus a peak of 1.4 million in July 2005.26
II. Incentive conflicts within builders with mortgage affiliates
Based on both the Case-Shiller Composite 20 index and the LoanPerformance House Price Index, source:
www.calculatedriskblog.com, accessed June 23, 2010.
Source: Census Bureau.
New Century Financial Corporation Examiner’s Report, Case No. 07-10416.
The Mortgage Lender Implode-o-Meter lists 384 lender failures since late 2006, http://ml-implode.com/, accessed
June 14, 2010.
Source: S&P/Case-Shiller via www.calculatedriskblog.com accessed May 21, 2010.
Source: First American CoreLogic negative equity report.
Source: Census Bureau.
This section examines evidence of incentive conflicts within builders with mortgage affiliates. It has
three subsections: the first subsection briefly describes relevant institutional details about these firms and
the second presents allegations of distorted competition from field interviews and public reports of
regulators and other stakeholders. The final subsection presents an empirical analysis of lending and
home price patterns to determine the presence of these conflicts.
A. Institutional details
Builders established mortgage units beginning in the 1960s through the 1990s. By 1999, nearly all the
national and large regional builders had either internal units or joint ventures with financial institutions.
These units typically only funded mortgages associated with homes constructed by the parent. In
interviews, builder executives reported that the rationale for establishing in-house units was the desire for
an integrated purchasing experience of the homebuyer and control over the service standards of the
mortgage loan officers.
During this period, builders disposed of their loans on the secondary market through sales to
large financial institutions. Prior to these sales, mortgages were funded with revolving credit lines with
financial institutions known as “warehouse facilities.” These warehouse facilities were typically small
relative to home sales. For example, Pulte Homes – the largest U.S. homebuilder – reported a $955
million credit facility in 2006. With annual home sales of $14 billion, this credit facility allowed for
mortgages to be held for 16 to 30 days prior to sale.27
In 1974, concern about inadequate consumer protection led to passage of the Real Estate
Settlement Procedures Act (RESPA), to which these mortgage affiliates were subject. This act regulated
consumer disclosures of affiliated business relationships – such as between mortgage affiliates and their
parents – in order to reduce kickbacks between entities that obscure actual costs to consumers.
The calculation uses data reported in Pulte 10K for 2006: ($955 million credit facility *85% utilization)/($14
billion * 90% capture rate * 80% loan-to-value) = 29 days. Pulte reported an average of 47% utilization – common
in the industry to allow for seasonal fluctuations - which allowed mortgages to be held for 16 days before sale.
B. Allegations of distorted competition
During the spring of 2010, I conducted a series of interviews with industry practitioners in various roles,
including builders, appraisers, bankers, securitizers and industry consultants, to gather qualitative
evidence of lending practices and internal conflicts. In interviews with banks managers and industry
consultants, a general view emerged that builders with financing affiliates were beset by internal conflicts.
According to one industry consultant, “I saw some builders under pressure to take on inordinate risk [in
2006]. I literally watched CEOs and senior executives put extraordinary pressure on mortgage guys,
extraordinary, over the top, even firing them [to sell homes].” Another interviewee stated, “The dirty
secret is that they have come to depend on [these affiliates] for sales. Some of the worst lending I’ve seen
has come out of those affiliates.”
In addition to these interviews, I surveyed public information on builders, including press articles,
court dockets and Department of Housing and Urban Development (HUD) reports. In February 2010,
American Banker published an article critical of these mortgage affiliates, noting that three of the largest
national builders ranked among the 50 worst performing lenders of high-volume originations for the
Federal Housing Administration (FHA).28 The article stated:
“One of the less-examined cracks in the [housing market] foundations…is the lending
done in recent years by home builders' financing arms…Part of the problem, observers
believe, is that being an in-house lender at a company whose main business is selling
houses makes for loose underwriting. ‘Owner financing just creates a conflict,’ said
David Lykken, the president of the Austin consulting firm Mortgage Banking Solutions.”
Beginning in 2007, various government agencies launched investigations into builder lending
practices.29 KB Home and Beazer Homes, two of the largest builders, both exited their lending businesses
and paid substantial fines as a result of these investigations.
Jeff Horwitz and Kate Berry, “Another Source of Woe in FHA: Builder-Run Lenders,” American Banker,
February 1, 2010, Vol.175, No.16. Note that this article focused on FHA loans, partly because HUD regulates these
loans and makes FHA lender data publicly available; however, the statements can be interpreted to apply more
broadly to conventional loans.
See, for example, HUD Audit Reports 2009-LA-1018, 2006-LA-1014, 2006-LA-1001, as well as SEC and
Department of Justice investigations into Beazer Homes.
On June 3, 2010, HUD announced that it was soliciting general public comment on whether
builders’ mortgage affiliates violated RESPA by engaging in various business, including inflating home
values, lowering underwriting standards, and causing borrowers to have negative equity in their homes.30
This announcement followed announcements in 2009 of investigations of builders’ mortgage affiliates by
the Attorneys General of both Nevada and Arizona – two states that were most affected by the housing
These accusations could reflect either efficient competition or incentive conflicts with the firms.
Reflecting the latter allegation, a shareholder lawsuit filed against Beazer in June 2008 stated:
“Critically, Beazer Homes negated Beazer Mortgage's ability to make informed lending
decisions based on applicants' qualifications and simply required Beazer Mortgage to find
a way to make loans happen, even where the applicants were not qualified for the loans
they received...[Confidential witness-1] stated that from his observations at Beazer
Mortgage, ‘just about anything would be done to qualify someone for a loan…’”32
This shareholder lawsuit against firm managers captured a broader view that these efforts to inflate sales
were harmful to the firms.
C. Analysis of builder incentive conflicts
This section tests whether internalizing mortgage lending within builders introduced incentive conflicts.
I begin by comparing the relative quality of builder and non-builder loans. If builders lowered
underwriting quality of the lending unit in order to enable more sales, then:
H1: The quality of builder-originated loans will be worse than the quality of loans
originated by other lenders during the sample period (1999-2009).
Federal Register, Vol. 75, No. 106, June 3, 2010, Proposed Rules, http://edocket.access.gpo.gov/2010/pdf/2010-
13350.pdf, page 3.
http://www.azcentral.com/business/articles/2010/03/26/20100326pulte-homebuilder-ariz.html, accessed August 2,
In Re: Beazer Homes USA Inc. Securities Litigation, Case 1:07-cv-00725-CC.
The absolute comparison in H1 between builders and other lenders does not differentiate between
efficient cross-subsidization and incentive conflicts.33 To distinguish between these two factors, I run a
differences-in-differences analysis comparing changes in the lending quality of builders relative to other
lenders over time. I consider 1999-2002 to be a baseline period, during which time builders’ lending was
least distorted by incentive conflicts. I then compare relative lending in later periods to this baseline
period, controlling for factors that may impact optimal relative lending. The choice of 1999-2002 as a
baseline period is based on two observations. First, employees’ ability to manipulate lending standards
was lower because the private-label securities market was much smaller compared to 2003-2006 (Keys et
al 2010b); therefore, there was less demand for non-conforming loans. Second, this period better
reflected the housing conditions of the prior decade in contrast to the environment of irrational home price
expectations and reported lending abuses of the later boom years. As such, the likelihood that subjective
performance measures were violated was lower during 1999-2002 than during later periods. This
assumption follows from literature on repeated interaction and relational contracts (Bull 1987; Baker,
Gibbons and Murphy 2002).
The second assumption is that the optimal relative lending thresholds between types of lenders
should only change if the profitability of homes changes; otherwise, any factors that affect lending
standards - such as interest rates, GSE guidelines or secondary market demand for mortgages - should
affect both builders and other lenders equally. I use home price appreciation as a proxy for home
profitability. If incentive conflicts do not exist within builders’ organizations, the relative lending quality
between builders and other lenders should not change over time, controlling for home prices. Conversely,
if incentive conflicts do exist within builders, then:
Optimally worse lending quality has been noted in the auto lending industry by Barron, Chong and Staten (2008)
and Pierce (2010) who both find that auto finance captives perform worse than non-captive lenders, and by Carey,
Post and Sharpe (1998) in the corporate lending markets, who find that finance companies serve observably riskier
customers and attribute this segmentation to reputational or regulatory factors.
H2: Lending quality of builders relative to other lenders will deteriorate from the
baseline period (1999-2002) to later periods, controlling for home price changes.
C.2 Data and econometric model
I construct a comprehensive database of mortgages in the geographies in which these potential incentive
conflicts would be expected to be most pronounced: the top 100 zip codes as measured by new home
construction over the last decade.34 I define loan quality as the likelihood of mortgage default conditional
on the observable set of attributes of a loan. I compare mortgages issued by builders to those issued by
other lenders to obtain the relative lending quality of builders. To ensure comparability, I include only
conventional, purchase mortgages of newly constructed, single-family detached homes. I assume that this
restriction produces a similar population of prospective homebuyers for all potential lenders. As such,
any differences in measured lending quality would reflect either screening choices by lenders based on
soft information or treatment effects wherein lenders place borrowers into safer loans.
To test H1, I use a Cox hazard model to calculate the relative hazard rate of defaults of mortgages
originated by builders versus other lenders, controlling for borrower risk and contract characteristics.35 I
verify this Cox model with a series of matched analyses that use a range of variables and matching
algorithms. To test H2, I repeat the hazard analysis of H1, but divide the sample into four time periods to
look for deterioration of builder lending over time.
I employ the following model in the hazard analysis:
CoreLogic provided the list of these 100 zip codes.
The hazard rate of default is defined as:
Pr(t t T t | T t )
h(t ) lim
t 0 t
Or, in words, the probability of failure within an interval t conditional on having survived until time t , divided
by the duration t as t approaches 0.
h j (t | xi ) h j 0 (t ) exp( j j 0 NonBuilderIndi j1Macroi
j 2 Lenderi j 3 Risk j 4Contracti j , yearYearFEi j , zip ZipFEi i )
Where h(t | xi ) is the hazard of default of loan i and h0 (t ) is the baseline hazard. NonBuilderInd is an
indicator variable equal to one if the lender is not a builder (throughout the analysis, the builders are the
omitted lender type unless otherwise specified), and captures systematic differences in underwriting
quality between builders and other lenders, as well as any remaining risk unobservable to the
econometrician.36 Macro is a vector of macroeconomic controls which includes the constructed home
price index, Lender is a vector of lender controls, Risk is a vector of standard hard risk metrics on the
borrower (e.g., FICO, combined loan-to-value ratio), and Contract is a vector of mortgage
characteristics, including product type (Fixed, Adjustable, etc.) and options (Interest Only period,
Negative amortization, etc). YearFE and ZipFE represent year and zip code fixed effects. See
Appendix A.2 for a more detailed definition of these variables.
The subscript j represents the time period of each specification. In tests of H1, j 1 and refers
to 1999-2009. In test of H2, j ranges from 1 to 4, each referring to the four times periods analyzed:
baseline period (1999-2002), mid-boom (2003-2004), late-boom (2005-2006), and post-boom (2007-
j 0 are the main variables of interest. To test H1, if builders have lower lending standards than
other lenders conditional on observable risk and contract characteristics, j 0 – the coefficients on the non-
builder indicator – should be less than 1, indicating that mortgages from other lenders have lower hazard
of default than builder mortgages. If H2 is supported, j 0 should decrease in later boom periods relative
to the baseline period, indicating that builder loans became relatively more hazardous as time progressed.
This remaining risk can include both “hard risk” which is verifiable to the investors and “soft risk” which is only
observable to the originator, such as the loan officer’s subjective assessment of the buyer’s future employment
prospects (see Rajan et al 2009, Berger et al 2005) for discussions of hard and soft information in lending)
This result would support the hypothesis that incentive conflicts within builders led to lower underwriting
quality as conditions became more favorable for these conflicts.
C.3 Data and sample selection
C.3.1 Data sources
The primary sources for the study consist of county public record filings and the Loan Performance
Servicer database, which were merged for this study in cooperation with First American CoreLogic, the
data provider. The county public records database is an extensive database which consists of the
complete public record filings, including all liens related to purchase and refinancing mortgages, and
notices filed of defaults and foreclosures. The Loan Performance Servicing database consists of mortgage
servicer-provided information which covers approximately 80% of all mortgages in the United States,
according to CoreLogic.37 The merge produces a dataset that links the lender and mortgage
characteristics only available through county records (such as identity of the lender, combined first and
second lien loan-to-value at time of origination, and whether a notice of default has been filed against the
mortgage) to borrower and product characteristics available in the proprietary Servicing database (FICO,
loan prices and other contract terms). I supplement these two sources with data on the lender, hand-
collected or merged from Compustat (for public firms), and macroeconomic data from the Census
Bureau, the Bureau of Labor Statistics, Freddie Mac, and other public sources.
The dependent variable in the study is an indicator whether a notice of mortgage default was filed with
the county, together with the date of filing. The macroeconomic variables include the tract level median
income in 2000, obtained from the Census Bureau, the Freddie Mac 30-year fixed rate mortgage rate, and
state-level unemployment figures obtained from the Bureau of Labor Statistics. Four standard risk
Not to be confused with the Loan Performance Securities database, which covers 95% of the subprime market. To
my knowledge, this is first time First American has released the Servicer database, as well as allowed these two
sources to be merged for academic use.
metrics are included: combined loan-to-value ratio,38 FICO score, a flag indicating that either incomplete
or no loan documentation was submitted at time of origination, and debt-to-income ratio.39 Seven
variables characterizing the mortgage contract were included, again from the Servicer database: four
indicators on the mortgage rate type (fixed rate mortgage, adjustable rate mortgage, hybrid fixed-
adjustable mortgage, and balloon), and three indicators for contract terms (interest only period,
prepayment penalty clause and negative amortization provisions).
The variables describing the lender types and characteristics were hand-collected, merged from
Compustat or generated from the mortgage data. Because of the long tail of mortgage lenders in the
database, I rank-ordered the lenders by the number of loans each contributed to the dataset, and hand-
coded lenders until 80% of the loans in the database were coded. The remaining lenders were allocated to
a “small miscellaneous” category.40 Of the 251 hand-coded lenders, I first identified the parent firms of
these lenders, if any, using a combination of Capital IQ, Thompson and press archive searches and
accounting for changes of control via mergers and acquisitions. If the parent firm was publicly-traded in
the U.S., firm performance data was merged from Compustat, including total assets, returns on assets and
SIC code. For private firms and lenders with no identifiable parent, the SIC code was identified using
Capital IQ or press mentions. These firms were assigned to four categories: builders (SIC codes 1520,
1531), depository banks (SIC codes 6020, 6021, 6022, 6035), standalone mortgage banks (SIC codes
6162, 6163), and affiliated mortgage banks (SIC codes 4213, 6141, 6159, 6172, 6211, 6311, 6331, 6531,
6798, 7200). Many of the lenders in this last category were known to specialize in subprime lending, and
many are no longer in operation. The final categories included 34 builders, 75 depository banks, 121
This combined LTV measure was calculated from the public record data and corrects for the “silent seconds”
issue in other studies, in which piggyback mortgages at time of origination are not detectable in the data (and why
first-lien LTV is often censored at 80%).
I use the “back-end” debt-to-income, a measure of the total monthly liabilities of the borrower, including the debt
on the subject property, divided by the total monthly income of the borrower(s).
The smallest ten lenders that were coded each contributed an average of 38 loans, while the largest ten lenders
contributed an average of 26,170 loans; therefore, the incremental contribution of each additional lender was
minimal. Additionally, upon inspection, most of the lenders in the long tail are small private mortgage funders,
brokers, or specialized lending vehicles about whom little public information is available. Many of these entities
likely table-funded (immediately sold) their loans, and hence did not take on any credit risk.
standalone mortgage companies, 40 large affiliated mortgage companies, and 2,333 small miscellaneous
firms. For the builders, 10 of the 34 were joint ventures with banks. Their builder classification was
retained, but a joint venture flag was added to all specifications. Joint venture-funded loans comprise
12% of all builder-financed loans in the sample.
C.3.3 Sample selection and descriptive statistics
The sample selection is outlined in Table 1 Panel A. I begin with all public deed filings for the top 100
zip codes as ranked by new home construction from 1999 to 2009. The initial deed filings were screened
for new construction purchase mortgages only,41 which provides an initial sample of 779,315 mortgages
originated between January, 1999 and September, 2009. All mortgages designated as corporate-owned,42
interfamily transfers or private party sales, and all condominiums, townhouses and other miscellaneous
property types were then excluded, as well as mortgages with missing or duplicate data,43 removing
184,925 records. I also excluded FHA and VA mortgages.44,45 After this exclusion, a sample of 476,672
purchase mortgages from the county filings remained. This sample was submitted to First American to
merge with the Loan Performance Servicing database. The match rate between the two databases was
44%, providing a final sample of 212,058 mortgages.46
<< INSERT TABLE 1 ABOUT HERE >>
As a robustness check, in unreported results, I included the resale market, which strengthens the results.
Fraud regarding concealing corporate ownership was rampant during this period and particularly in these
geographies, and so corporate ownership was likely underreported in our sample. Consequently, actual corporate
owners may be in the final sample, along with individually-owned investment properties.
Many of the records with missing data, upon inspection, are early versions of later, completed deed filings, and
hence also duplicate records.
FHA stands for the Federal Housing Authority and VA stands for the Veterans Administration, both government
agencies. These mortgages were excluded for two reasons: first, FHA/VA mortgages are governed by separate
regulations and hence behave differently from conventional mortgages: they are generally made to first-time or low-
income borrowers, with a very low initial down payment, and carry insurance against a default, in exchange for a
monthly premium. Also, most studies, unless specifically focusing on the FHA, exclude these mortgages, and
hence, I also exclude them for comparability purposes.
In unreported results, I run many of the same analyses in this study against a subsample of only FHA and VA
mortgages and find no material or statistical differences between lenders, which likely reflects the stricter
regulations within this market during this period.
According to CoreLogic, the typical match rate is about 60%; however, this rate was somewhat lower because of
the number of new homes involved where tax identification information is missing from one of the databases.
Overall, an analysis of the merged sample reveals that the data is generally unbiased for this study
or biased against the findings, although it is not fully representative of the submitted dataset. A
discussion of the potential biases introduced through the merge process is included in Appendix A.3.
Figure 2 shows geographic characteristics of the sample. Figure 2a shows the relative mortgage
activity and locations of the zip codes included in the sample. The zip codes are primarily located within
California, Arizona, Nevada, Texas and Florida, with the highest concentrations of construction within
Las Vegas, Nevada and Phoenix, Arizona. Figure 2b displays the same zip codes, but now encoded by
the concentration of mortgages financed by builders. Both Las Vegas and Phoenix have among highest
proportion of mortgages financed by builders (42% for both cities). Figure 2c shows the proportion of
mortgages to default within each area. Las Vegas, southern California and Florida show the worst overall
performance, while Texas is among the best, with less than 6% rate of mortgage default within most zip
<< INSERT FIGURE 2 ABOUT HERE >>
Figure 2d plots the home price index specifically constructed using this sample of homes. This
index was calculated using resales of homes obtained from the public records. The figure shows that the
regions within the sample experienced home price changes of similar magnitude to the aggregate national
figures. Between 1999 and the peak in 2006, home prices increased by 80%, followed by a decline of
30% between late 2006 and 2009. Figure 2e plots annual number of mortgages issued by lender types
over the sample period. Builders show the largest share, increasing mortgage activity from 319 loans in
1999 to 15,864 loans in 2006. Depository banks and standalone mortgage banks have the next largest
shares and growth, issuing 10,163 and 9,848 loans in 2006, respectively. Builders also show the largest
share growth, shown in Figure 2f, doubling from 19% in 1999 to 38% in 2006.
Table 2 shows the descriptive statistics on the merged mortgage database (refer to Appendix A.2
for variable definitions). Builder-financed homes tend to cost less than homes financed by other types
($281,602 versus $289,669 for other lenders). The buyers are also less risky on observable
characteristics, with lower combined LTV (80.64% versus 83.20% for other lenders) and higher FICO
scores (723 versus 706 for other lenders), although “low or no documentation status” – whether the loan
documentation was only partial or missing at time of sale (so called “Liar Loans”) is higher for builders
than for their peers (56.62% versus 52.11% for other lenders). The final risk measure, debt to income
ratio – traditionally a noisy measure – is indistinguishable from the other lender types. Builders also
offer, on average, safer product types and contract characteristics than their peers, with a higher rate of
fixed rate mortgages (76.53% versus 65.42%) and lower rates of mortgages with negative amortization or
pre-payment penalty clauses than their peers.
<< INSERT TABLE 2 ABOUT HERE >>
Figure 3 shows Kaplan-Meier failure estimates of mortgage default for builder and non-builder loans.
These failure estimates correspond to the cumulative probability that a mortgage will default as a function
of the elapsed months following origination, unadjusted for any loan characteristics. Figure 3a shows the
cumulative failure estimate for all years (1999-2009) while Figure 3b shows the failure estimates by time
period. These graphs shows that builder mortgages have consistently lower failure rates than non-builder
mortgages and that these differences become particularly pronounced during the later time periods.
<< INSERT FIGURE 3 ABOUT HERE >>
D.1 Choice of mortgage characteristics over time
One indication of lending quality is the lenders’ choices of borrower and mortgage characteristics
over time. Table 3 shows these choices for builders and non-builders from 2002-2006, the years
immediately preceding the housing bust. The dependent variables are seven of the key risk indicators of
mortgages. The coefficients on the year effects in the table show that these indicators consistently
became riskier during this period, with systemic increases in combined loan-to-value (CLTV) ratios, low
documentation (low-doc) loans, and negative amortization, interest only and prepayment mortgage
provisions, and decreases in the proportion of fixed interest rate loans. Throughout this period, in
contrast, builders underwrote loans with lower overall CLTV, higher FICO scores and lower risk in all
other categories except for the proportion of low documentation status loans. This metric likely reflects
builders’ desire for the faster loan processing associated with low-doc loans. Overall, this table shows
that builders systematically selected safer borrowers and loan categories than the market during the period
preceding the crisis.
D.2 Relative hazard
Table 4 reports the results of the pooled hazard tests of H1. Column (1) shows the unconditional relative
hazard between non-builders and builders for the years 1999 to 2009. This specification shows that, on
an unconditional basis, the hazard of loans originated by builders is statistically indistinguishable from
other lenders (the hazard rate of 1.2878 is indistinguishable from 1). This specification allows for all
choices by lenders to vary, including geography, macroeconomic conditions, and years in which to offer
loans, as well as choice of borrower risk and contract characteristics. Column (2) includes lender and
macroeconomic controls and fixed effects for years and zip codes. In this specification, non-builder
lenders have a significantly higher hazard of defaults relative to the builders: holding geographies, years,
lender attributes and macroeconomic conditions constant, the hazard of default of non-builder loans
increases to 3.0646 times higher than builder loans and is significant at the 1% level. The difference in
the results between Columns (1) and (2) indicates that builders offered mortgages in riskier geographies,
years and macroeconomic conditions; however, controlling for those choices, their hazard of lending is
lower than other lenders.
<< INSERT TABLE 4 ABOUT HERE >>
Column (3) adds the observable risk characteristics into the specification. Column (4) includes
risk and contract terms and, therefore, models the full specification in equation (1). In both Columns (3)
and (4), the coefficient on the non-builder indicator attenuates, indicating that builder lenders targeted
observably safer borrowers and contract terms relative to non-builders. However, after controlling for
market segmentation choices, the hazard of non-builder lenders in Column (4) remains 1.3924 times
greater than builders and significant at the 1% level.
The direction and magnitude of the coefficients on other variables within Table 3 shows that
other factors are also associated with mortgage default. Mortgages by publicly traded firms have a 30%
higher hazard rate, while mortgages issued by builders’ joint venture partners have a lower hazard rate
than builders’ in-house units. Among the most explanatory variables are home prices: a 35% drop in
home prices is associated with a 102% increase in the hazard of default. Also significant is the combined
loan to value ratio: a one standard deviation increase from the mean of 82.3 to 96.3% is associated with a
175% increase in relative hazard. Finally, as expected, FICO score strongly explains mortgage defaults,
as expected: a one standard deviation decrease in FICO score (63 points) increases relative hazard by
To validate the hazard analysis, I run a matched analysis using a variety of matching variables
and algorithms. I run three types of matching algorithms: propensity scoring as implemented by Leuven
and Sianesi, (2003), coarsened exact matching (CEM) (Iacus, King and Porro, 2009) and a nearest
neighbor match (Abadie et al 2004). Within each algorithm, I increase the number of variables on which
to perform the match. Table 5 shows the results of this analysis. Panel A shows a difference of 2.96% in
the rate of mortgage default between builder and non-builder loans in an unmatched baseline. Because
census tracts were stripped from the merged database to protect consumer anonymity, I first run the
analysis on the unmerged (deeds) database to match within census tract and sale quarter (Panel B).
Census tracts represent a smaller geographic area than zip codes, with a mean of 9.3 census tracts per zip
code in this sample, and therefore allow more uniform geography within which to match. The average
treatment effect is consistently negative and significant and, in most cases, larger than the unmatched
baseline. For example, using a nearest neighbor match within census tract, sale quarter, mortgage
amounts and product types produces an estimated 3.63% lower default rate by builders than non-builder
lenders. I then run the analysis on the merged database in order to add in FICO score and other product-
level variables from the Loan Performance data (Panel C). In these specifications, I match within zip
code and sale quarter. Again, the results are consistently negative and generally larger than the
unmatched comparison. For example, the average estimated effect using CEM and the most extensive set
of controls (Column 15) is a 3.22% lower default rate by builders, significant at the 1% level. This
analysis confirms the results from the hazard analysis: that builder loans perform significantly better than
non-builder loans within this sample.
<< INSERT TABLE 5 ABOUT HERE >>
Overall, Tables 4 and 5 show that builders constructed homes and provided mortgages in riskier
areas than other lenders. Within these areas, however, they targeted safer borrowers and offered safer
mortgage contracts. In addition, after controlling for these choices, they also appeared to underwrite
higher quality loans. This underwriting could be due either to additional screening on soft information or
a treatment effect in which lenders place borrowers into safer loans. In unreported regressions, both
factors appear to be present. These results conclusively reject H1 that either profit-maximizing cross-
subsidization or incentive conflicts led to relatively laxer lending practices to place unqualified buyers
into homes they could not afford.
Table 6 presents the results of H2, the differences-in-differences analysis of whether moral hazard
led to faster deterioration in lending standards within builders relative to the rest of the industry. This
analysis tests the central research question of this paper: whether tightly-integrated corporate scope
introduces incentive conflicts into firms. For this test, I run the same specifications as in Columns (2)-(4)
in Table 3, but now subdivided by time period. For space reasons, only 0 , the coefficient on the non-
builder indicator, is reported. The table shows the results of the three basic models over four main time
periods during the housing cycle: the baseline years (1999-2002), the middle boom (2003-2004), the late
boom (2005-2006) and the post boom (2007-2009). The results from the baseline period are shown in the
top panel. On an unconditional basis, the hazard of other lenders is greater than builders at 2.0063 and
significant at the 1% level (indicating that other lenders generally targeted a riskier segment); however,
controlling for observable risk and contract terms, the relative hazard for other lenders is not statistically
different from the builders. These results suggest that, during the baseline period, the quality of lending
between builders and non-builders was essentially the same.
<< INSERT TABLE 6 ABOUT HERE >>
Across the time periods, the relative hazard increases in most specifications until the post-boom
period, in which the relative hazard of non-builder lenders decreases but still remains above the baseline
period. Notably, the late-boom period (2005-2006), in which lending standards have been shown to be
the least strict in the industry (Demyanyk and Van Hemert 2009), is the period in which the builders
appear to show the greatest restraint relative to other lenders, in terms of both point estimates and
statistical significance. The hazard coefficients are all larger than the baseline year and strongly
significant in all three specifications, indicating that relative default hazard increases over time for non-
builders. The post-boom period (2007-2009) is perhaps the most interesting result – in this time period,
the builders presumably had the most incentive to dispose of excess home inventory, and yet the relative
standards appear to remain slightly more restrained than during the baseline period. Overall, these results
reject H2 and, therefore, do not provide evidence for the existence of moral hazard within builders during
these periods, at least relative to financial lenders.
III. How did builders avoid scope-related conflicts?
The remainder of this paper focuses on explanations for how builders’ lending practices may have been
disciplined beyond those of other lenders during this period. I first discuss builder organizational choices
that restrained lending practices and I then analyze two market mechanisms that may have countered
moral hazard within the firms: a secondary market and a primary market channel.
A. Organizational choices
Builder interviewees stated that they had established employee incentives and organizational structures to
reduce any employee temptation to “buy deep” at the expense of the firm. According to the interviewees,
loan officers at builders were compensated with lower-powered incentives than their counterparts in pure
financial institutions, particularly the wholesale divisions of larger depository or mortgage banks. Most
builders discussed implementing incentive contracts that generally included at least a 50% base salary,
with bonuses based on a combination of production volume and post-sales customer satisfaction surveys.
In contrast, within wholesale divisions, loans were supplied to financial firms through mortgage brokers –
who were compensated entirely on volume and pricing.47 Interviewees also discussed their greater ability
to monitor mortgage quality within builders than within other institutions. Because of the small size of
their warehouse facilities and the builders’ stated intention not to hold loans, lending practices that
resulted in hard-to-sell loans or forced repurchases would be easy to detect.
Interviewees reported a variety of organizational structures. Divisional heads of the mortgage
units universally did not report to the managers responsible for home sales; however, beyond this
relatively straightforward choice, there was no common organizational structure. In some firms, the loan
officers were centralized in call centers far from the home sales, and in other firms, loan officers were co-
located with the regional sales teams. In one case, the builder appointed a team to each subdivision to
work on-site. This team included a sales manager and a loan officer working together in adjacent offices.
B. Secondary mortgage market channel
Builders cite the limited capacity of their mortgage units to hold loans prior to sale into the secondary
market as restraining lending practices. I develop and test hypotheses for the presence of this secondary
This statement is anecdotal and could not be independently verified for this study.
The ideal test of this channel would use a source of exogenous variation in firms’ access to
funding to identify a relation between secondary market reliance and lending standards. However, no
such instrument could be identified for this study. Therefore, I develop two predictions that, if supported,
would not be easily explained by alternative channels.
For the first prediction, I exploit the heterogeneity of the other lenders and apply the converse of
the builder logic. Lenders with more opaque loan portfolios and easier access to funding beyond
constrained warehouse facilities should be more susceptible to internal conflicts and deteriorate their
lending standards more during this period. I divide the other classifiable lenders into depository banks,
standalone mortgage banks and affiliated mortgage banks. Depository banks both have access to
diversified, insured deposit funds and maintain “held-to-maturity” (more opaque) loan portfolios.
Affiliated mortgage banks similarly have access to parental support, often of large investment banks.
Standalone mortgage banks have neither of those funding alternatives. They rely on limited warehouse
facilities and on the secondary market for funding, similar to builders. Therefore, if reliance on the
secondary mortgage markets disciplined originators, then:
H3: Compared to builders, the relative quality of loans originated by depository banks
and affiliated mortgage banks will deteriorate more than loans originated by standalone
mortgage banks during 2003-2006 relative to the baseline period of 1999-2002.48
I also test the relation between lending quality and loan pricing at the firm level. Rajan, Seru and
Vig (2009) show that both securitizers and investors of mortgage-backed securities relied more on hard
For a test of H3 to provide consistent evidence of the secondary market hypothesis, it needs to demonstrate that
variation in access to funding is the relevant source of differences in the lending practices between these
organizational types. The other primary difference that has been noted is regulatory regime (Keys et al 2009).
Depository banks are regulated by various government agencies and insured by the Federal Deposit Insurance
Corporation, while the other organization types are much more lightly regulated. However, as noted by Keys et al
(2009), absent a moral hazard problem within the banks themselves, greater regulation would lead to the opposite
prediction as H3; namely, that bank underwriting quality should be higher than other firm types.
(verifiable) than soft (unverifiable) information about the borrower.49 If a lender originates a mortgage
with the intent to sell it easily on the secondary market, that lender should set the interest rate of the loan
based more on the hard information.50 If these lenders were also more constrained in their capacity to
hold loans, and therefore acted to minimize the risk of defective loans being put back to them, then:
H4: Firms that price loans more on hard information will have higher lending quality,
conditional on the observable characteristics of the loan.
B.1 Results of secondary mortgage market channel tests
The specifications used to test H3 are similar to those used to test H2 above, as described in equation (1)
over the four time periods that represent the baseline (1999-2002) through post-boom (2007-2009) years.
In this test, I replace the non-builder indicator variable of H2 with indicators representing the four non-
builder lender categories, and compare the relative changes over time in the coefficients on depository
banks and affiliated mortgage banks to standalone mortgage banks.
The results are shown in Table 7, with only the coefficients on the lender indicators presented.
Panel A presents the results of the pooled analysis across all years, 1999-2009. Panel B shows the direct
test of H3 by breaking out the relative hazard of the lender categories by time period. The results support
H3, showing that, relative to builders, both depository banks and affiliated mortgage banks deteriorated
Examples of hard information include FICO scores, loan-to-value and debt-to-income ratios, and loan
documentation, while unverifiable soft information might include a loan officer’s assessment of an applicant’s
credibility, her future employment prospects and intuition about her likelihood of defaulting.
Note that, while the prices at which the mortgages were sold on the secondary market are not available, the rates
charged to the consumer are critical to the valuation of the mortgage, and therefore should be subject to this
verifiability restriction and valid to use in this context.
lending more than standalone mortgage banks by 2005-2006, relative to the baseline period of 1999-
<< INSERT TABLE 7 ABOUT HERE >>
To test H4, I create firm-level measures for reliance on hard information and lending quality. I
use a subsample of non-conforming loans for this analysis since conforming loans, by definition, can be
immediately sold to the GSEs.52 I run this analysis for the late boom period, 2005 to 2006, when the
largest differences between lenders were observed. Because of the requirement for sufficient numbers of
mortgages and mortgage defaults to calculate both the R-squared and the hazard by lender, only the 172
most active firms could be used for this analysis.
For the hard information measure, I run the following model once per firm using ordinary least
squares and extract the adjusted R-squared:
pricei j j1Macroi
j 2 Lenderi j 3 Risk j 4Contracti yearYearFEi zip ZipFEi i
Following Rajan, Seru and Vig (2009), I use the initial rate of the loan for the dependent variable.53 For
the firm-level lending quality, I replace the non-builder indicator with firm fixed effects in equation (1)
and extract the coefficient on each firm. This coefficient captures the “firm effect” on the overall hazard
of loan default.
I begin by testing the validity of pricing R-squared as a measure of the intent to sell loans quickly
to investors. The alternative interpretation is that this measure reflects simpler firm operations: firms
Standalone mortgage banks show worse behavior in the post-boom period (2007-2009); however, the environment
during this period was significantly different and many of these lenders may have been positioning themselves for
bankruptcy or to be sold.
I define non-conforming loans as loans with any of the following features: FICO score less than 620, less than full
loan documentation, mortgage amounts above the jumbo threshold or interest-only, prepayment or negative
The results are essentially unchanged when I rerun this analysis only with ARMs and use the margin charged over
the floating index once the teaser period expires.
choose to have simple lending procedures and, as a consequence, find their loans simpler to securitize.
To rule out this alternative, I examine the association between lagged firm performance and pricing R-
squared. If pricing R-squared reflects intent to sell loans quickly, then, as firm performance decreases,
firms become more constrained and pricing R-squared should increase. To test this mechanism, I use a
first-differences approach and relate changes in lagged return on assets, return on equity and profit margin
to changes in pricing R-squared between the 2003-2004 period and 2005-2006 periods. Table 8 Panel A
shows the results of this test. Because only public firms with sufficient loans in both 2003-2004 and
2005-2006 could be used, 48 firms qualified for this test. The test confirms the negative association
between lagged changes in all three measures of firm performance and changes in pricing R-squared,
supporting the interpretation that this measure reflects firms’ intent to offload mortgages.54
<< INSERT TABLE 8 ABOUT HERE >>
To test H4, Figure 4a plots the firm-level default hazard versus adjusted R-squared of equation
(2), with each firm observation weighted by the number of loans. The figures show a clear negative
relation between the R-squared and the hazard, supporting H4 that lending quality is associated with
increased reliance on the secondary market.55 Figure 4b repeats the plot of hazard versus adjusted R-
squared, but now distinguishes between lender types. The figure shows that builders generally have
higher R-squared and lower hazard during 2005-2006, as do standalone mortgage banks, while depository
banks and affiliated mortgage banks are more dispersed, and dominate the low R-squared, high hazard
region of the plot. Figure 4c shows the sample result by overlaying the plot with the cumulative
distribution function of the adjusted R-squared values by lender type. Again, this result shows that
builders and standalone mortgage banks price loans based more on hard information and that these firms
have lower firm-level hazards of default. Using the causal interpretation that pricing R-squared reflects
In unreported regressions, this negative association is also maintained in subsamples of other firm types,
Note that the confidence interval of the linear fit does not account for the estimation error in both the hazard and
secondary market reliance established in Table 8 Panel A, Figure 4c can be understood as illustrating that,
as reliance on the secondary market increased, underwriting quality increased.
<< INSERT FIGURE 4 ABOUT HERE >>
Table 8 Panel B presents these findings in a regression format. In this table, each observation
represents one firm, and observations are weighted by number of loans. Column (1) shows the relation
between firm type and pricing R-squared. The result shows that depository banks and affiliated mortgage
banks had lower R-squared than builders. Columns (2)-(6) shows the relation between pricing R-squared
and the estimated firm-level hazard. Column (2) includes only firm type fixed effects and shows that
depository banks and affiliated mortgage banks had higher firm-level hazard than builders. Column (3)
shows the negative relation between pricing R-squared and hazard. These results show an economic
effect: a one standard deviation increase in adjusted R-squared is associated with a 9% decrease in the
firm-level hazard of default. Column (4) includes both pricing R-squared and fixed effects for lender
type. The coefficients on the firm types attenuate compare to (2), indicating that the pricing R-squared
explains a portion of the lender-type differences. For example, the coefficient on Depository Banks
decreases from 0.1210 in Column (2) to 0.0396 in Column (4). These results support H4 that pricing R-
squared is negatively associated with the firm-level hazard, and that builders and standalone mortgage
banks price loans more on hard information than depository banks and affiliated mortgage banks.
Interestingly, Column (4) also shows that not all differences between lender types are explained
by the R-squared of the firms, as the fixed effects by lender type are generally significant in both
specifications. This finding suggests that other factors also drive differences between firm types.
One potential concern from this test is that the negative association between pricing R-squared
and hazard is driven by the inclusion of builders in the sample who may – for some outside reason – price
loans more on hard information. Columns (5) and (6) address this concern by running similar tests as (3)
and (4), but omitting builders from the regressions. Both regressions show a negative and significant
Overall, the results support H3 and H4 and the assertion that limited capacity to hold loans affects
C. Primary mortgage market channel
The second explanation for higher lending quality is that builders lent more safely because they
internalized the costs of foreclosures within their subdivisions, assuming the inventory and pricing risk
for future sales if a property foreclosed in their neighborhood. The cost of foreclosure externalities has
been shown to be quite large: Campbell, Giglio and Pathak (2009) show that a foreclosure lowers the
price of homes within a quarter-mile radius of the property, and estimate a 1% reduction on the prices of
homes within 0.05 miles. Financial lenders, on the other hand, were not pre-committed to subdivisions
and, therefore, did not internalize foreclosure costs in the same way.
As in the secondary market test, there is no strong instrument that exogenously assigns varying
levels of subdivision construction between builders. Therefore, I again develop two second-order
predictions to provide supporting evidence, using Campbell, Giglio and Pathak’s (2009) result that
proximate foreclosures have costly consequences on surrounding homes.
First, within subdivisions, foreclosures of homes sold earlier in the lifecycle should be costlier to
builders than homes sold later because they affect demand and prices on a greater number of future
homes. Therefore, all else equal, builders should be relatively more careful with earlier sales than later
sales. If foreclosure prevention disciplined builder lending behavior, then:
H5: Within a subdivision, builder-financed loans for homes sold early in the lifecycle will
perform better than loans for later homes, all else equal, while the performance of non-
builder financed loans will exhibit no such pattern.
Second, as the density of homes within a subdivision increases, the cost of foreclosures should
correspondingly increase. If foreclosure prevention disciplined builder lending behavior, then:
H6: The more homes sold by a builder within a subdivision, the better builder-originated
loans will perform, all else equal, while non-builder originated loans will exhibit no such
C.1 Results of costs of foreclosure externalities tests
To test H5, I use census tracts as a proxy for subdivisions, which cannot be observed. Within this
sample, each zip code contains a median level of 9 census tracts with 1,453 mortgages per census tract
and 90 mortgages per census tract-lender pair. For this analysis, I create a subsample of mortgages that
includes only homes constructed by builders with lending units. This subsample allows a direct
comparison of homes constructed by the same firm, some of which are financed by this firm and others
by outside lenders. I construct a measure of the order of homes constructed by each builder in a given
census tract over the sample period, normalized from 0 (earliest) to 1 (latest).56 I then create indicators
reflecting whether a mortgage was in the earliest or latest quartile of mortgages in that census tract, and
then interact those variables with an indicator of whether the mortgage was funded by the builder. If H5
is supported, the coefficient of the builder interacted with the earliest quartile should be less than one
(safer) and the coefficient of the builder interacted with the latest quartile should be greater than one
(riskier). The results of this test are show in Table 9 Column (1). Neither coefficient is significant. I also
conduct a robustness check to see if the composition of homebuyers changes over the subdivisions’
lifecycle. If the pool of homebuyers consistently increases in quality as subdivisions mature, the effect in
H5 may be moderated. I replace hazard of default with FICO as the dependent variable and run an
For each builder-census tract combination, I sort each mortgage from the earliest to the latest mortgage funded,
and then divide the order of each mortgage by the sum of all mortgages over the entire sample period.
ordinary least squares model. The results are presented in Column (2). If the pool of buyers becomes
systematically safer as subdivisions age, then the coefficient on the first (last) quartile of building order
should be less (greater) than zero, which it is not. The combined results of Columns (1) and (2) provide
no systematic evidence that builders lend to safer consumers earlier in the lifespan of a subdivision.
<< INSERT TABLE 9 ABOUT HERE >>
To test H6, I use the total number of homes in each census tract, rather than the ordering of
homes, and create two separate measures of building density. For the first measure, I use the full sample
of all mortgages, and calculate the log of the sum all sales for each lender within a census tract. This sum
is then interacted with an indicator variable equal to one if the mortgage was financed by a builder. If H6
is supported, the coefficient on this latter variable should be less than one; that is, as more homes are sold
in a given census tract, the potential negative externalities of foreclosures increase, and therefore, the
hazard of builder loans will be differentially less than other lenders.57 The results of this test are shown in
Column (3), and again, the hypothesis is not supported.
The final test of H6 is performed on a subsample of mortgages that include only homes
constructed (but not necessarily financed) by builders with lending units. A second measure of building
density is defined as the log of the sum of all sales of home constructed by each builder, regardless of
who financed the homes. As in the previous test, this variable is interacted with an indicator equal to one
is the mortgage was funded by a builder. If H6 is supported, the coefficient on that variable should be
significant and less than one. The results for this test are in Column (4) and do not support this
There is a question whether the endogeneity of the building density measure affects these results. The
endogeneity argument is as follows: firms choose to construct more in census tracts that are unobservably superior
and therefore attract unobservably higher quality borrowers. As a result, builders can lend more risky mortgages to
these homebuyers to profit from additional quality and therefore we should observe no differences in lending
between regions. This story, however, should not affect the prediction in H6 for the following reason: it affects
builders and other lenders equally, and does not contradict the result that every marginal default in denser
subdivisions should be costlier to builders than defaults in less dense areas, but not other lenders. Therefore, in
equilibrium, we should observe a difference in the lending patterns between builders and other lenders as a function
of the density of subdivisions.
The results of these four tests shown in Table 8 do not support the hypotheses that foreclosure
externalities restrained builders during this period.
IV. Discussion and conclusion
This study attempts to unify two views of the role of related corporate scope – that this scope can be a
source of competitive advantage for a firm and that it may introduce incentive conflicts inside
organizations. Homebuilders with mortgage units provide an interesting context to integrate these views.
On the one hand, mortgage units evidently provided benefits to builders in their ability to manage
customer relationships and risk associated with selling homes. During interviews, builder executives
consistently stated that the same levels of service and sales predictability were not feasible with outside
lenders or with joint venture arrangements. These assertions were supported by firm actions: except when
mandated to divest by law, every large builder had established an in-house lender by 1999, the beginning
of the sample period. On the other hand, extreme pressure to sell homes and maintain high earnings
levels during the housing bubble provided at least the appearance of moral hazard within these firms,
even to industry insiders. During that period, managers were suspected of lowering lending standards to
boost home sales and profits to the detriment of both consumers and the firms themselves.
This paper begins by testing the simple integration of the competitive advantage and incentives
views: that internalizing beneficial complementary activities can result in incentive costs. However, I
find no evidence of moral hazard within builders. In fact, I find that builders’ lending behavior was more
restrained than their competitors. Furthermore, while lending standards systemically deteriorated from
1999 to 2006, builder lending standards deteriorated less than their peers.
These results can be explained by a combination of organizational choices made by builders that
restrained their actions. I find evidence that builders may have implemented flatter incentives that their
financial competitors in order to manage housing inventory risk, a risk faced only by builders. I also find
evidence that the limited capacity of the builders’ in-house lenders to hold loans disciplined their
behavior. Within these firms, available capital was allocated to purchasing land and constructing homes.
This limited capital provided to mortgage operations led to commitments to originate loans that could be
sold easily to others with a lower likelihood of forced repurchases due to defective underwriting. I
provide corroborating evidence for this explanation by showing that the lenders with the most funding
resources – depository banks and affiliated mortgage lenders – showed greater lending deterioration
during the period than either builders or standalone mortgage banks. I also show that the lender’s
capacity to hold loans – proxied by the degree to which the firm priced mortgages based on hard
information – is positively associated with the lending standards of a firm. These results suggest that,
conditional on operating in the high securitization environment pre-2007, firms with limited capacity to
hold loans were more disciplined than firms with concurrent access to other sources of funding. This
finding contradicts the conventional wisdom that firms that depended on selling mortgages on the
secondary market (the so-called “originate and sell” or “originate to distribute” model) engaged in worse
lending practices than firms with capacity to hold loans.
One limitation of this study is the absence of systematic internal firm data. Interviews for this
study indicated that managers employed a wide variation of practices and it is unclear which practices
were most significant in addressing incentive conflicts. Future research could address this limitation by
gathering information on internal incentive systems and organizational structures within builders and,
ideally, other lenders. Another potentially fruitful area is whether the deterioration in lending standards
noted within some financial firms is attributable to scope-related incentive conflicts within those firms,
the opacity of these firms, or some interaction between these two factors.
This research has several implications for studies of corporate scope. First, it suggests a potential
organizational tradeoff associated with internalizing activities that appear ex ante to be complementary:
that these activities may introduce incentive conflicts and other influence activities into the firm. This
tradeoff may become especially pronounced if the external environment shifts. As managers expand the
horizontal scope of their firms’ activities, they must consider the appropriate organizational structures to
reduce these costs without forgoing the benefits of the tight integration.
This view alone, however, is incomplete. It ignores the role that organizational structure and the
market can play in restraining costly incentive conflicts – the second implication of this study. In this
context, the discipline provided by having very limited capacity to hold defective loans appears to have
restrained builders, particularly during the worst period of industry lending practices. This finding has a
larger interpretation that could be tested in future research: that the degree to which incentive conflicts
impose costs on firms is related to the opacity of the firm operations. Costly actions may be more
constrained if the negative consequences are readily observed and measured against an external
benchmark. On the other hand, firms with harder-to-measure operations – because, for example, their
outcomes are not observed for many years or they are sufficiently specialized such that comparisons are
infeasible – may be more susceptible to these conflicts.
This study also has several implications for firm managers. It highlights the need to weigh the
benefits of internalizing related scope with potential costs from introducing conflicts. It also suggests that
designing an organization that facilitates immediate market feedback and transparent performance
measures may assist managers in counteracting these negative consequences. Conversely, managers of
opaque firms must be particularly vigilant for these conflicts. For policymakers, transparent firms that
may appear conflicted to outside observers may be less distorted than seemingly unconflicted, but more
Abadie, Alberto, David Drukker, Jane L. Herr and Guido W. Imbens, 2004, Implementing matching
estimators for average treatment effects in Stata, Stata Journal, 4(3): 290-311.
Agarwal, Sumit, Gene Amromin, Anna Paulson and Sriram Villupuram, 2010, Homebuilders, affiliated
financing arms and the mortgage crisis, Extended Abstract.
Ashbaugh, Hollis, Ryan LaFond and Brian Mayhew, 2003, Do nonaudit services compromise auditor
independence? Further Evidence, The Accounting Review, Vol. 78, No. 3.
Azoulay, Pierre, 2004, Capturing knowledge within and across firm boundaries, American Economic
Revie w, Vol. 94, No. 5.
Baker, George, Robert Gibbons and Kevin Murphy, 1994, Subjective performance measures in optimal
incentive contracts, The Quarterly Journal of Economics, Vol. 109, No. 4.
Baker, George, Robert Gibbons and Kevin Murphy, 2002, Relational contracts and the theory of the firm,
The Quarterly Journal of Economics, Vol. 117, No. 1.
Baker, George and Tom Hubbard, 2003, Make versus buy in trucking: asset ownership, job design, and
information, American Economic Review, Vo. 93, No. 3.
Baker, Malcolm and Jeffrey Wurgler, 2007, Investor sentiment in the stock market, The Journal of
Economic Perspectives, Vol. 21, Issue 2.
Barron, John, Andrew Chong and Michael Staten, 2008, The emergence of captive finance companies and
risk segmentation of the consumer loan market: theory and evidence, Journal of Money, Credit and
Banking, Vol. 40, Issue 1.
Berger, Philip and Eli Ofek, 1995, Diversification's effect on firm value, Journal of Financial Economics,
Vol 37, Issue 1.
Bradley, D.J., Jordan, B.D., Ritter, J.R., 2008. Analyst behavior following IPOs: the ‘bubble period’
evidence. Review of Financial Studies, Vol. 21.
Bresnahan, Timothy; Shane Greenstein and Rebecca Henderson 2009, Schumpeterian competition within
computing markets and organizational diseconomies of scope, working paper.
Bull, Clive, 1987, The existence of self-reinforcing implicit contracts, The Quarterly Journal of
Economics, Vol. 102, No. 1.
Campbell, John, Stefano Giglio and Parag Pathak, 2009, Forced sales and home prices, NBER working
Carey, Mark, Mitch Post and Steven Sharpe, 1998, Does corporate lending by banks and finance
companies differ? Evidence on specialization in private debt contracting, The Journal of Finance, Vol. 53,
Chandler, Alfred, 1990, Scale and Scope: The Dynamics of Industrial Capitalism, Belknap Press of
Harvard University Press.
Chandler, Alfred, 1992, Organizational capabilities and the economic history of the industrial enterprise,
The Journal of Economic Perspectives, Vol. 6, Issue 3.
Coval, Joshua, Jakub Jurek and Erik Stafford, 2009, The economics of structured finance, Journal of
Economic Perspectives, Vol. 23, No 1.
Cowen, Amanda , Boris Groysberg, Boris and Paul Healy, 2006, Which types of analyst firms are more
optimistic?, Journal of Accounting and Economics, Vol. 41.
Dell’Ariccia, Giovanni, Igan Deniz and Luc Laeven, 2008, Credit booms and lending standards: Evidence
from the subprime mortgage market, working paper.
Demyanyk, Yuliya and Otto Van Hemert, 2009, Understanding the subprime mortgage crisis, Review of
Financial Studies, advanced access.
Duany, Andres, Elizabeth Plater-Zyberk and Jeff Speck, 2001, Suburban Nation: The Risk of Sprawl and
the Decline of the American Dream, North Point Press.
Foote, Christopher, Kristopher Gerardi, Lorenz Goette and Paul Willen, 2008a, Just the facts: An initial
analysis of subprime’s role in the housing crisis, Journal of Housing Economics, Vol 17, Issue 4.
Gibbons, Robert, 2010, Handbook of Organizational Economics, Forthcoming.
Grossman, Sanford and Oliver Hart, 1986, The costs and benefits of ownership: a theory of vertical and
lateral integration, Journal of Political Economy, Vol 94.
Hart, Oliver and John Moore, 1990, Property rights and the nature of the firm, Journal of Political
Economy, Vol 98.
Henderson, Rebecca and Kim Clark, 1990, Architectural innovation: The reconfiguration of existing
product technologies and the failure of established firms, Administrative Science Quarterly, Vol. 35, No.
Henderson, Rebecca and Ian Cockburn, 1996, Scale, scope and spillovers: the determinants of research
productivity in drug discovery, Rand Journal of Economics, Vol. 27.
Holmstrom, Bengt 1999, The firm as a subeconomy, The Journal of Law, Economics, and Organization,
Vol 15, 74-102.
Holmstrom, Bengt and Paul Milgrom, 1994, The firm as an incentive system, American Economic
Review, Vol. 84, No. 4.
Holmstrom, Bengt and Paul Milgrom, 1991, Multitask principal-agent analyses: Incentive contracts, asset
ownership, and job design, Journal of Law, Economics and Organization, Vol. 7, special issue.
Iacus, Stefano, Gary King and Giuseppe Porro, 2009, Causal inference without balance checking:
Coarsened exact matching, working paper.
Jacobides, Michael, 2004, Industry change through vertical disintegration: how and why markets emerged
in mortgage banking, Academy of Management Journal, Vol. 48, No. 3.
Keys, Benjamin, Tanmoy Mukherjee, Amit Seru and Vikrant Vig, 2009, Financial regulation and
securitization: evidence from subprime loans, Journal of Monetary Economics, Vol. 56 Issue 5.
Keys, Benjamin, Tanmoy Mukherjee, Amit Seru and Vikrant Vig, 2010a, Did securitization lead to lax
screening: Evidence from subprime loans, Quarterly Journal of Economics, Vol 125, No 1.
Keys, Benjamin, Tanmoy Mukherjee, Amit Seru and Vikrant Vig, 2010b, 620 FICO, Take II:
securitization and screening in the subprime mortgage market, Working Paper.
Kinney, William R. JR, Zoe-Vonna Palmrose and Susan Scholz, 2004, “Auditor independence, non-
audit services, and restatements: was the U.S. government right?” Journal of Accounting Research Vol.
42, No 3.
Kogut, Bruce and Udo Zander, 1992, Knowledge of the firm, combinative capabilities, and the replication
of technology, Organization Science, Vol. 3, No. 3.
Lafontaine, Francine and Margaret Slade, 2007, Vertical integration and firm boundaries: the evidence,
Journal of Economic Literature, Vol 45, No 3.
Lang, Larry and Rene Stulz, 1994, Tobin’s Q, corporate diversification, and firm performance, Journal of
Political Economy, Vol 102, No 6.
Langlois, Richard and Nicolai Foss, 1999, Capabilities and governance: The rebirth of
production in the theory of economic organization, Kyklos Vol. 52: 201-18.
Leonard-Barton , Dorothy, 1992, Core capabilities and core rigidities: A paradox in managing new
product development, Strategic Management Journal, Vol. 13, SI.
Leuven, Edwin and Barbara Sianesi, 2003, PSMATCH2: Stata module to perform full Mahalanobis and
propensity score matching, common support graphing, and covariate imbalance testing.
Macher, Jeffrey and Barak Richman, 2008, “Transaction cost economics: An assessment of empirical
research in the social sciences.” Business and Politics 10: 1-63.
March, James and Herbert Simon, 1958, Organizations, USA: Wiley.
Mayer, Christopher, Karen Pence and Shane Sherlund, 2009, The rise in mortgage defaults, Journal of
Economic Perspectives, Vol 23, No 1, 27-50.
Mian, Atif and Amir Sufi, 2009, The consequences of mortgage credit expansion: Evidence from the U.S.
mortgage default crisis, Quarterly Journal of Economics, Vol. 124, No 4.
Mian, Atif and Amir Sufi, 2010, House prices, home equity-based borrowing, and the U.S. household
leverage crisis, American Economic Review, revise and resubmit.
Montgomery , Cynthia and Birger Wernerfelt, 1988, Diversification, Ricardian rents, and Tobin's q,
The RAND Journal of Economics, Vol. 19, No. 4.
Nickerson, Jack and Todd Zenger, 2004, A knowledge-based theory of the firm – The problem-solving
perspective, Organization Science, Vol. 15, No 6.
Peteraf, Margaret, 1993, The cornerstones of competitive advantage: A resource-based view, Strategic
Management Journal, Vol. 14, No. 3.
Pierce, Lamar, 2010, Organizational structure and the limits of knowledge exploitation: evidence from
consumer automobile leasing, Working Paper.
Piskorski, Seru and Vig, 2010, Securitization and distressed loan renegotiation: Evidence from the
subprime mortgage crisis, Journal of Financial Economics, Vol 97, Issue 3.
Porter, Michael, 1996, What is strategy? Harvard Business Review, Vol. 74, Issue 6.
Purnanandam, Amiyatosh, 2010, Originate-to-distribute model and the subprime mortgage crisis,
Rajan, Uday, Amit Seru and Vikrant Vig, 2008, The failure of models that predict failure: distance,
incentives and defaults, Chicago GSB Research Paper No. 08-19.
Scharfstein, David and Jeremy Stein, 2000, The dark side of internal capital markets: Divisional rent-
seeking and inefficient investment, Journal of Finance, Vol. 55, No. 6.
Shelanski, Howard and Peter G. Klein, 1995, Empirical research in transaction cost economics: A Review
and Assessment, Journal of Law, Economics, & Organization, Vol. 11, No. 2.
Shiller, Robert, 2008. The Subprime Solution: How Today's Global Financial Crisis Happened, and What
to Do about It, Princeton University Press.
Shiller, Robert, 2000. Irrational Exuberance, Princeton University Press.
Siggelkow, Nicolaj 2001, Change in the presence of fit: The rise, the fall, and the renaissance of Liz
Claiborne, Academy of Management Journal, Vol. 44, Issue 4.
Siggelkow, Nicolaj, 2002, Evolution toward fit, Adminstrative Science Quarterly, Vol. 47.
Siggelkow, Nicolaj and Michael Porter, 2008, Contextuality within activity systems and sustainability of
competitive advantage, Academy of Management Perspective, Vol. 22, Issue 2.
Simon, Herbert, 1991 (1947), Administrative Behavior, New York: The Free Press.
Strahan, Philip, 2008, Bank structure and lending: What we do and do not know, Handbook of Financial
Intermediation and Banking, Amsterdam: Elsevier.
Teece, David, 1980, Economies of scope and the scope of the enterprise, Journal of Economic Behavior
and Organization, 223-247.
Teece, David J, Richard Rumelt, Giovanni Dosi and Sidney Winter, 1994, Understanding corporate
coherence: Theory and evidence, Journal of Economic Behavior & Organization, Vol. 23 Issue 1.
Teece, David, Gary Pisano and Amy Shuen, 1997, Dynamic capabilities and strategic management,
Strategic Management Journal, Vol. 18, Issue 7.
Tripsas, Mary, 1997, Surviving radical technological change through dynamic capability: evidence from
the typesetter industry, Industrial and Corporate Change, Vol. 6, No. 2.
Villalonga, Belen, 2004, Diversification discount or premium? New evidence from the Business
Information Tracking Series, The Journal of Finance, Vol. 59, No. 2.
Wernerfelt, Birger, 1984, A resource-based view of the firm, Strategic Management Journal, Vol 5, 171-
Williamson, Oliver, 1975, Markets and hierarchies, analysis and antitrust implications, New York:
Williamson, Oliver, 1985, The Economic Institutions of Capitalism, The Free Press.
Appendix A.1 Description of secondary market transactions
Mortgages were resold on the secondary market using two types of transactions: whole loan sales and
securitizations. In a whole loan sale, loans are completely transferred from one party to another. In
securitizations, cash flow rights are transferred, with the riskiest, residual cash flows (the “equity tranche”)
typically retained by the securitizing entity. Industry insiders reported that due diligence was greater for whole
loan transactions than for securitizations because securitizing firms bore the residual default risk for the loans.58
Builders exclusively engaged in whole sales to financial institutions during this period.59 In whole loan sales,
originators periodically assembled a pool of loans and either offered the pool to a specific investor or put it up for
bid to a set of investors. Loan pool documentation provided key “hard” information about the borrowers and
mortgage terms, including FICO score, loan to value ratio and whether full documentation was provided. Full
loan packages would also be provided to prospective investors upon request.
Investors had three levers to ensure loan quality. First, they could “kick out” loans from the pool which
they deemed poor quality.60 Originators would then attempt to re-offer kicked out loans to other buyers; however,
often these loans would take longer to place and be sold at a discount. Second, they could “put back” loans,
forcing originators to repurchase the loans at any time post-sale if any representations and warranties in the marter
loan purchase agreements were found to be violated. The most common reps and warranties covered situations in
which borrowers defaulted within the first three payments, fraud or appraisal issues during the origination process
were found or the originator’s declared lending standards had been violated.61 Like kicked-out loans, these
repurchased loans could potentially be resold – after a period of time and often at a loss rate of approximately
25% according to interviewees. Originators also had to compensate the buyers for lost interest income, as well as
pay interest for the additional utilization of the warehouse lines. Lastly, investors could offer lower prices for the
loans. While no industry-wide data is available, interviewees reported that investor due diligence began
increasing in early 2005 as home price appreciation began to moderate, and then dramatically increased in 2006
and 2007 as home prices leveled and defaults rose.62
New Century Financial Corporation Examiner’s Report, Case No. 07-10416, pg 135.
Source: 10-K filings, interviews and Thompson SDC MBS listings.
For example, New Century Financial Corporation experienced a 14.95% kick-out rate by December 2006 prior to
bankruptcy in early 2007, op cit, pg 161.
One interviewee estimated that, by 2007, repurchases accounted for 5 to 10% of the notional value of mortgages, a rate that
had been increasing since early 2005.
The bankruptcy examiner for New Century Financial Corporation reported that, in 2005, 20-30% of the loan files in New
Century pools would be fully examined by investors. By 2006, 30-35% of loan files were examined and by late 2006, some
investors examined every loan file. Page 163.
Appendix A.2 Variable definitions
Variable Definition Source
Notice of default A notice filed in the county public record that the borrower has officially Public records
defaulted on the mortgage. This notice is considered the first step in foreclosure
Sale amt ($) Sale price appearing on the recording document Public records
Mtg amt ($) Amount of the mortgage, first-lien only Public records
Initial interest rate Original or initial interest rate as of the loan's first payment date Servicing db
Margin Basis points added to the ARM index to determine the coupon rate of the ARM Servicing db
CLTV Ratio of the first and second-lien amount to the sale amount of the house at the Public records
time of purchase
FICO FICO (credit) score of borrower at time of loan origination Servicing db
Low or no documentation Indicator equal to one if loan documentation was not complete Servicing db
Back-end DTI The total monthly liabilities of the borrower, including the debt on the subject Servicing db
property, divided by the total monthly income of the borrower(s)
Adjustable rate mortgage Indicator equal to one if mortgage rate is determined by a spread over a floating Servicing db
Fixed rate mortgage Indicator equal to one if mortgage rate is determined by a fixed coupon rate Servicing db
Hybrid mortgage Indicator equal to one if mortgage rate is fixed for an initial period and then floats Servicing db
for the remainder of the loan
Balloon mortgage Indicator equal to one if mortgage balance does not fully amortize over the term Servicing db
of the note, thus leaving a lump sum due at maturity
Negative amortization Indicator equal to one if mortgage has a provision that allows monthly payments Servicing db
below the interest coverage amount, rolling the difference into the principal
Interest only Indicator equal to one if mortgage has provision that allows monthly payments to Servicing db
cover only interest owed, rather than interest plus amortization of the loan
Prepayment penalty Indicator equal to one if mortgage has provision that assesses a penalty to the Servicing db
borrower if a loan is prepaid
Log tract 2000 med inc Median income in 2000 of households within a census tract Census
Freddie 30 yr rate Interest rate for 30-year fixed mortgages published by Freddie Mac Freddie Mac
Regional home prices A home price index at the county level constructed from the mortgages in this Public records
State-level Quarterly state unemployment rates Bureau Labor Statistics
Log annual number loans Log of the number of loans originated by each lender within this dataset Public records
Lender independent Indicator equal to one if lender does not have a parent company Researcher
Parent public Indicator equal to one if parent company is publicly traded Compustat
Parent log assets Log of the parent assets if parent company is publicly traded Compustat
Lender JV with bank Indicator equal to one if builder’s lending unit is a joint venture with a financial Various
Appendix A.3 Comparison of the full and merged datasets
Overall, the merged sample is not fully representative of the submitted dataset; however, an analysis of the
differences reveals that the data is generally unbiased for this study or biased against the findings. The coverage
of merged sample is geographically representative; however, subprime loans are under-represented. The overall
default rate, therefore, of the merged sample is lower than the full sample.
Among builder types, affiliated mortgage banks are the most under-represented in the final sample
(3.81% of the final sample versus 6.46% of the full sample), followed by small miscellaneous (11.55% versus
12.64%) and standalone mortgage banks (21.01% versus 21.72%). This under-representation should only be an
issue if it biases the sample in the same direction as the findings. For both small miscellaneous and standalone
mortgage banks, the mean default rate appears to bias the sample against the findings (mean default rates of
9.20% and 13.65% in the matched sample, versus 10.59% and 14.76% in the full sample). These default rates
compare to 7.86% versus 8.79% rates for builders in the merged and full samples, respectively. Compared to
depository banks, standalone mortgage banks, and small miscellaneous, therefore, builders’ defaults rates are less
skewed downwards by the merge, which biases against the results found in this study. Affiliated mortgage banks,
however, have a mean default rate of 21.76% in the final sample versus 18.36% in the full sample. Results using
this lender type, therefore, should be interpreted with caution. Note, however, that the lending quality of these
firms was substantially worse than builders in both the full and merged samples.
The other differences between the samples are either minor or also biased against the findings. For
example, while the matched sample is overall safer than the full sample – the combined loan to value is somewhat
lower and the mean census tract income level is slightly higher – the overall builder variables are less affected by
the merge than for other lenders, again biasing against the findings. Finally, in unreported results, I run the same
hazard models on the full and matched samples using control variables available in both samples and obtain
indistinguishable results. Therefore, I conclude that, while the match is not a random sample of the full public
record data, it does not introduce biases that would drive the results, with the possible exception of the relative
hazard of builders versus affiliated mortgage lenders.
Figure 1: Representative census tract, showing a mix of mortgages originated by builders and outside lenders. This tract contains three developments, two of which are owned by builders
with in-house lenders.
Figure 2a: Mortgage activity in top zip codes for new construction, 1999-2009. This figure displays the counties with the top 100 zip codes as ranked by new home construction from 1999
to 2009. Each county is coded by the average amount of new home mortgages underwritten in the zip codes within these counties during the sample period.
Figure 2b: Percent of purchase mortgages financed by builders, 1999-2009. This figure displays the counties with the top 100 zip codes as ranked by new home construction from 1999 to
2009. Each county is coded by the average percent of mortgages underwritten by builders in the zip codes within these counties during the sample period.
Figure 2c: Percent of mortgages to default, 1999-2009. This figure displays the counties with the top 100 zip codes as ranked by new home construction from 1999 to 2009. Each county is
coded by the average percent of mortgages that had defaulted by December 2009 in the zip codes within these counties during the sample period.
Figure 2d: Change in home prices in sample zip codes.
HPI Constructed from Home Sales in Sample
1999 2001 2003 2005 2007 2009
Figure 2e: Number of new home purchase mortgages underwritten by each lender type by year.
New home purchase mortgages
2000 2005 2010
Builders Depository banks
Standalone mtg banks Affiliated mtg banks
Small misc lenders
Figure 2f: Market share of new home purchase mortgages underwritten by each lender type by year.
2000 2005 2010
Builders Depository banks
Standalone mtg banks Affiliated mtg banks
Small misc lenders
Figure 3a: Kaplan-Meier estimates of mortgage default rates, 1999-2009. This graph shows the cumulative estimated probability of loan default by lender type as a function of the number
of months since loan origination. 95% confidence intervals not adjusted for within-lender correlations.
Estimated cumulative default rate, 1999-2009
0 50 100 150
Months since origination
Figure 3b: Kaplan-Meier failure estimates by time period. This graph shows the cumulative estimated probability of loan default by lender type as a function of the number of months
since loan origination, subdivided by time periods in which loans were originated. 95% confidence intervals not adjusted for within-lender correlations.
Estimated cumulative default rate, 1999-2002 Estimated cumulative default rate, 2003-2004
0 50 100 150 0 20 40 60 80
Months since origination Months since origination
Estimated cumulative default rate, 2005-2006 Estimated cumulative default rate, 2007-2009
0 20 40 60 0 10 20 30 40
Months since origination Months since origination
Figure 4: Test of secondary market hypothesis. This figure shows three plots of the firm-level hazard of default versus pricing R-squared,
a proxy for the capacity of firms to hold mortgages post-origination. See text for a more detailed description of this proxy. The y-axes
correspond to firm-level hazard rates, calculated from the coefficients on the firm fixed effects in a pooled hazard regression on all non-
conforming mortgages from 2005 to 2006. The x-axes plot the pricing R-squareds from firm-by-firm regressions which relate the prices of
the mortgages to the hard information characteristics of the borrower and the contract terms for non-conforming mortgages (equation (2) in
the text). Each observation corresponds to one firm and is weighted by the number of loans by the firm in the dataset. Each graph includes
a linear fit, also weighted by number of loans.
Figure 4a: Plot of hazard versus R-squared
Hazard versus Pricing R-squared, 2005 to 2006
0 .2 .4 .6 .8 1
Figure 4b: Plot of hazard versus R-squared, by firm type
0 .2 .4 .6 .8 1
Builders Depository banks
Standalone mtg banks Affiliated mtg banks
Figure 4c: Plot of hazard versus R-squared, with R-squared cumulative distributions
R-squared Cumulative Distribution, 2005 to 2006
0 .2 .4 .6 .8 1
Builders Depository banks
Standalone mtg banks Affiliated mtg banks
Table 1: Sample selection. This table describes the sample selection process and differences between the full and merged samples. Panel
A shows the excluded data from the initial sample of all new home construction purchase mortgages and the match rate between the public
record data and the Loan Performance Servicing database. Panel B show the differences between the full sample and the merged sample
constructed from public records.
A: Sample selection
New home construction purchase mortgages 779,315
less condos and townhouses 99,073
less missing data 71,219
less corporate owners and interfamily and private party sales 14,633
less FHA/VA 117,718
Total purchase mortgages 476,672
Matched with Loan Performance Servicing (44% match) 212,058
B: Comparison between full and merged samples
Full sample Unmerged Merged Difference Sig
Mean Mean Mean (Merged-Full)
Builder (%) 31.92 29.44 35.02 3.10 ***
Depository bank (%) 27.25 26.17 28.60 1.35 ***
Standalone mtg bank (%) 21.72 22.29 21.01 -0.71 ***
Affiliated mtg bank (%) 6.46 8.58 3.81 -2.65 ***
Small misc (%) 12.64 13.52 11.55 -1.09 ***
Notice of default (%) 10.68 11.40 9.78 -0.90 ***
Mortgage amount ($) 217,575 216,561 218,842 -1,267 ***
Combined LTV (%) 82.94 83.45 82.30 0.64 ***
Census tract median income 61,895 60,676 63,359 1,464 ***
Notice of default by lender type
Builder (%) 8.79 8.43 7.86 -0.93 ***
Depository bank (%) 9.77 8.87 7.89 -1.88 ***
Standalone mtg bank (%) 14.76 15.60 13.65 -1.11 ***
Affiliated mtg bank (%) 18.36 17.15 21.76 3.4 ***
Small misc (%) 10.59 11.54 9.20 -1.39 ***
Table 2: Descriptive statistics. This table provides the mean values of key variables within the sample data. See Appendix A.2 for
Builders All non- Depository Standalone Affiliated Small misc
builders banks mtg banks mtg banks
n=74,262 n= 137,796 n=60,657 n=44,550 n=8,087 n=24,502
Mortgage amounts and prices
Sale amt ($) 281,602 289,669 296,932 288,060 280,088 277,779
Mtg amt ($) 213,353 221,800 226,681 220,912 215,453 213,428
Initial interest rate (%) 6.10 6.19 6.04 6.15 7.19 6.30
Reset spread (%) 2.56 3.25 2.78 3.14 5.25 3.32
Borrower observable risk
CLTV (%) 80.64 83.20 82.98 83.35 86.27 82.45
FICO 723 706 710.9 707.8 656.3 708
Low or no documentation (%) 56.62 52.11 36.01 67.99 49.65 51.85
Back-end DTI 18.66 18.70 15.66 25.24 15.14 15.97
ADJ (%) 11.65 19.38 20.98 17.27 27.96 16.42
Fix (%) 76.53 65.42 67.19 63.66 32.56 75.09
Hyb (%) 11.43 11.81 8.94 17.49 20.08 5.86
Bal (%) 0.39 3.39 2.9 1.58 19.4 2.63
Negative amortization (%) 5.44 9.66 7.77 14.76 1.6 9.22
Interest only (%) 21.96 22.10 17.65 27.35 34.95 16.1
Prepayment penalty (%) 5.45 19.29 13.59 22.04 50.89 17.65
Notice of default (%) 7.86 10.80 7.89 13.65 21.76 9.2
Log annual number loans 2,833 2,156 1,766 2,994 440 111
Lender independent (%) 0 71.00 59.03 98.63 3.98 100
Parent public (%) 92.01 72.41 96.54 38.7 86.84 0
Lender JV with bank (%) 11.56
Table 3: Differences in borrowers and mortgage terms between builders and non-builders, 2002-2006. This table shows the changes in
builder and non-builder borrower and mortgage characteristics during the housing boom years. All year variables represent mortgage
origination year. See Appendix A.2 for variable definitions. Standard errors clustered by origination year. Standard errors in (), ***, **, *
significant at 1, 5 and 10% level.
OLS Logit (marginal effects)
Low Fixed Negative Prepayment
CLTV FICO documentation interest rate amortization Interest only penalties
(1) (2) (3) (4) (5) (6) (7)
Builder*2002 -0.0227*** 20.6799*** -0.2469*** 0.2819*** -0.0523 -0.1797***
(0.0003) (0.1436) (0.0029) (0.0025) (0.0013) (0.0032)
Builder*2003 -0.0346*** 19.5911*** 0.0195* 0.2238*** -0.2101*** -0.2445*** -0.1829***
(0.0002) (0.1327) (0.0021) (0.0016) (0.0023) (0.0013) (0.0019)
Builder*2004 -0.0393*** 22.1257*** 0.0105 0.1702*** -0.1503*** -0.0659*** -0.2330***
(0.0003) (0.1899) (0.0020) (0.0017) (0.0010) (0.0017) (0.0024)
Builder*2005 -0.0387*** 21.0664*** 0.0633*** 0.1651*** -0.0368*** -0.0404*** -0.2727***
(0.0008) (0.1618) (0.0032) (0.0043) (0.0021) (0.0013) (0.0031)
Builder*2006 -0.0301*** 12.6969*** 0.0729*** 0.1377*** -0.0653*** -0.0411*** -0.1947***
(0.00003) (0.2516) (0.0033) (0.0043) (0.0019) (0.0015) (0.0031)
2003 0.0084*** 3.3903*** -0.1071*** -0.0898*** -0.0359*** 0.2904*** -0.0274***
(0.0004) (0.1390) (0.0008) (0.0017) (0.0013) (0.0015) (0.0012)
2004 0.0071*** 1.2608** -0.0511*** -0.2403*** 0.0085 0.5965*** -0.0051*
(0.0007) (1.2608) (0.0018) (0.0038) (0.0024) (0.0032) (0.0027)
2005 0.0067*** 2.4610*** 0.0178*** -0.2487*** 0.0587*** 0.6800*** 0.0174***
(0.0012) (0.3788) (0.0044) (0.0025) (0.0030) (0.0041) (0.0023)
2006 0.0105*** 4.3104*** 0.0425*** -0.1867*** 0.0471*** 0.7289*** -0.0123**
(0.0010) (0.3094) (0.0029) (0.0025) (0.0019) (0.0039) (0.0021)
Constant 0.7746*** 696.2810***
State fixed effects Y Y Y Y Y Y Y
Observations 143,369 119,466 83,323 143,378 51,173 77,823 127,540
R-squared 0.093 0.023 0.043 0.112 0.092 0.180 0.080
Table 4: Hazard of default of mortgages, 1999-2009. This table compares the hazard of default of mortgages issued by the non-builder
lenders from 1999-2009 relative to builders (omitted), controlling for borrower risk and mortgage characteristics. In Column (1), the
relative hazard is unconditional. Column (2) adds zip code and origination year fixed effects, lender and macro characteristics. Column (3)
adds hard risk metrics, and Column (4) adds mortgage characteristics. See Appendix A.2 for variable definitions. All errors are clustered by
lender. Standard errors in (), ***, **, * significant at 1, 5 and 10% level.
Dep var: Relative hazard of Unconditional Geo controls Geo, risk Geo, risk and contract
mortgage default controls controls
(1) (2) (3) (4)
All non-builder lenders 1.2878 3.0646*** 1.6835*** 1.3833***
(0.2036) (0.4059) (0.1391) (0.1025)
Log mortgage amt 1.7254*** 1.6604***
Borrower observable risk
CLTV 80.5867*** 56.0620***
Origination FICO 0.9937*** 0.9939***
Low or no doc flag 1.6240*** 1.5302***
Back-end DTI 1.0008 1.0005
Adjustable rate mtg 1.4032***
Hybrid arm-fixed mtg 1.3156***
Balloon and other mtg 1.9441***
Negam flag 1.2394***
IO flag 1.2485***
Prepay flag 1.1205**
Log tract 2000 med inc 0.8760* 0.8559** 0.8532***
(0.0613) (0.0522) (0.0504)
Freddie 30 yr rate 1.0914** 1.1711*** 1.1593***
(0.0381) (0.0414) (0.0401)
Regional home prices 0.1010*** 0.1203*** 0.1336***
(0.0099) (0.0110) (0.0119)
State-level unemployment 1.0659*** 1.0785*** 1.0873***
(0.0136) (0.0131) (0.0134)
Log annual number loans 1.0249 0.9897 0.9691
(0.0312) (0.0212) (0.0188)
Lender is independent 0.7005*** 0.8518* 0.9328
(0.0953) (0.0824) (0.0803)
Public parent flag 1.9756*** 1.4377** 1.2965**
(0.4532) (0.2225) (0.1581)
Log parent assets 0.9358*** 0.9703** 0.9815*
(0.0181) (0.0133) (0.0108)
Lender is bank JV 1.3434** 0.8959** 0.8341***
(0.1662) (0.0443) (0.0440)
Zip code fixed effects N Y Y Y
Origination year fixed effects N Y Y Y
Observations 212,058 209,316 209,301 209,301
Pseudo R-squared 0.001 0.071 0.092 0.095
Table 5: Matched analysis as alternative specification. This table uses three different matching techniques as alternatives to the hazard analysis used in Table 4. These tests compare the
rates of default for mortgages issued by builders (treatment group) and non-builder lenders (control group) from 1999-2009. Panel A shows the unconditional comparison of the rates of
default between groups. Panel B uses the full sample (pre-LP merge) to match mortgages within census tracts and vintage year-quarters. Panel C uses the merged sample to match
mortgages within zip code and vintage year-quarters. All other matched variables are specified for each column. For performance reasons, a 10% subsample was used for all analyses.
Within Panels B and C, the estimated average treatment effect on the treated (ATT) is displayed first, followed by the average treatment effect on combined sample (ATE). Columns (1)-(3)
and (10)-(12) use propensity score matching implemented by Leuven and Sianesi, (2003) without replacement and a 25% standard deviation caliper. Columns (4)-(6) and (13)-(15) use
coarsened exact matching (Iacus, King and Porro, 2009) with census tracts or zip codes and origination year-quarters forced into exact matches and continuous variables (Sale amounts,
CLTV, product types, other controls) coarsened. Columns (7)-(9) and (16)-(18) use a nearest neighbor match (Abadie et al 2004) with the census tracts, zip codes and vintage years forced
into exact matches. “Risk controls” in Columns (11) and (14) refer to borrower risk controls used in Table 4 and product controls in (12) and (15) refer to product type and contract term
controls in Table 4. Standard errors in (), ***, **, * significant at 1, 5 and 10% level.
Matched analysis: effect of builder origination on mortgage defaults
A: Unmatched baseline Builder mortgages Non-builder mortgages Difference SE
Rate of mortgages that defaulted: 0.0786 0.1083 -0.0296*** (0.0014)
Propensity score match Coarsened exact match Nearest neighbor match
B: Within census tract and vintage year-quarter pairs (10% subsample of full sample)
Mtg amount, Mtg amount, Mtg amount,
Mtg amount, CLTV, Mtg amount, CLTV, Mtg amount, CLTV, product
Other matched variables: CLTV product type CLTV product type Mtg amount CLTV type
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Avg treatment on treated (ATT) -0.0534*** -0.0373*** -0.02490*** -0.0610*** -0.0441*** -0.0395*** -0.0584*** -0.0457*** -0.0363***
(0.0039) (0.0040) (0.0039) (0.0036) (0.0053) (0.0055) (0.0045) (0.0042) (0.0043)
Avg treatment on full sample -0.0307† -0.0292† -0.0261† -0.0424*** -0.0350*** -0.0340***
(0.0039) (0.0037) (0.0037)
Matched builder mtgs (ATT) 13,078 12,531 12,358 13,695 7,496 6,842 10,629 8,992 8,218
Total builder mtgs 14,180 14,177 14,177 15,163 15,163 15,163 15,163 15,163 15,163
C: Within zip code and vintage year-quarter pairs (10% subsample of merged sample)
Mtg amount, Mtg amount, Mtg amount,
Mtg amount, risk, product Mtg amount, risk, product Mtg amount, CLTV, FICO,
Other matched variables: risk controls controls risk controls controls Mtg amount CLTV, FICO product type
(10) (11) (12) (13) (14) (15) (16) (17) (18)
Avg treatment on treated (ATT) -0.0574*** -0.0271*** -0.0242*** -0.0690*** -0.0391*** -0.0322*** -0.0627*** -0.0411*** -0.0487***
(0.0052) (0.0051) (0.0051) (0.0047) (0.0069) (0.0074) (0.0060) (0.0058) (0.0060)
Avg treatment on full sample -0.0371† -0.0325† -0.0338† -0.0542*** -0.0411*** -0.0451***
(0.0051) (0.0049) (0.0049)
Matched builder mtgs (ATT) 6,740 6,525 6,279 7,372 3,715 2,825 4,084 2,792 2,599
Total builder mtgs 7,426 7,426 7,426 7,426 7,426 7,426 7,426 7,426 7,426
standard errors unavailable
Table 6: Hazard of default of mortgages by period in the housing cycle. This table compares the hazard of default of mortgages issued by
the four non-builder lenders in the early, late and post-boom periods of the housing cycle, relative to builder (omitted type), borrower risk
and mortgage characteristics. Within each period, the hazard of default is calculated with just fixed effects and lender and macro controls
(see Table 4 Column (2) for specification), conditional on risk (Table 4 Column (3)), and conditional on risk and contract terms (Table 4
Column (4)). See Appendix A.2 for variable definitions. All errors are clustered by lender. ***, **, * significant at 1, 5 and 10% level.
Baseline (1999-2002) All non-builder Pseudo R2
(N= 29,428) Lenders
Geo controls 2.0063*** 0.054
Geo, risk controls 1.3678 0.080
Geo, risk and contract controls 1.0728 0. 096
Mid boom (2003-2004)
Geo controls 3.5386*** 0.052
Geo, risk controls 1.7421*** 0.075
Geo, risk and contract controls 1.0862 0. 083
Late boom (2005-2006)
Geo controls 3.2559*** 0.047
Geo, risk controls 1.7319*** 0.069
Geo, risk and contract controls 1.4289*** 0.072
Post boom (2007-2009)
Geo controls 2.3403*** 0.058
Geo, risk controls 1.5015*** 0.082
Geo, risk and contract controls 1.3882** 0.084
Table 7: Hazard of default of mortgages by lender type. This table compares the hazard of default of mortgages issued by the four non-
builder lenders in the early, late and post-boom periods of the housing cycle, relative to builder (omitted type), borrower risk and mortgage
characteristics. Panel A shows the pooled analysis from 1999 to 2009, while Panel B divides mortgages by period within the housing
cycle. Within each period, the hazard of default is calculated with just fixed effects and lender and macro controls (see Table 4 Column (2)
for specification), conditional on risk (Table 4 Column (3)), and conditional on risk and contract terms (Table 4 Column (4)). See
Appendix A.2 for variable definitions. All errors are clustered by lender. ***, **, * significant at 1, 5 and 10% level.
A: Pooled analysis (1999-2009)
All years (1999-2009) Depository Standalone Affiliated mtg Small misc Pseudo R2
(N=209,299) banks mtg banks banks
Unconditional 0.9538 1.6222*** 2.5232*** 1.0748 0.004
(0.1444) (0.2971) (0.4383) (0.1505)
Geo controls 2.0714*** 2.8856*** 3.6762*** 2.3717*** 0.072
(0.3524) (0.5731) (0.4687) (0.4587)
Geo, risk controls 1.4918*** 1.5234*** 1.8844*** 1.5096** 0.092
(0.1683) (0.2165) (0.1382) (0.2190)
Geo, risk and contract controls 1.3020*** 1.2442* 1.5427*** 1.3720** 0. 095
(0.1186) (0.1491) (0.1038) (0.1834)
B: By time period
Baseline (1999-2002) Depository Standalone Affiliated mtg Small misc Pseudo R2
(N=29,428) banks mtg banks banks
Geo controls 1.6108 2.4435** 2.6521*** 1.6828 0.055
(0.5291) (0.8495) (0.7045) (0.7349)
Geo, risk controls 1.2865 1.6251 1.3875 1.3166 0.080
(0.3798) (0.4995) (0.2923) (0.5319)
Geo, risk and contract controls 1.0196 1.1929 1.0975 1.0631 0. 096
(0.2626) (0.3029) (0.2103) (0.4027)
Mid boom (2003-2004)
Geo controls 2.2603** 3.6033*** 3.9219*** 3.2447** 0.052
(0.9009) (1.1255) (1.1088) (1.4951)
Geo, risk controls 1.4542 1.6411** 1.9638*** 1.5374 0.075
(0.3965) (0.3790) (0.3871) (0.5719)
Geo, risk and contract controls 0.8513 1.0042 1.2411 0.8845 0. 083
(0.1862) (0.1724) (0.2080) (0.3023)
Late boom (2005-2006)
Geo controls 2.0401*** 2.7965*** 3.8401*** 2.4112*** 0.048
(0.3628) (0.6314) (0.5813) (0.5231)
Geo, risk controls 1.5452*** 1.4790*** 1.9443*** 1.5185*** 0.070
(0.1834) (0.2240) (0.1620) (0.2408)
Geo, risk and contract controls 1.3770*** 1.2120 1.6044*** 1.4109** 0.072
(0.1216) (0.1541) (0.1148) (0.1889)
Post boom (2007-2009)
Geo controls 1.9278*** 2.6851*** 2.6222*** 1.9104** 0.059
(0.3385) (0.6799) (0.5730) (0.5529)
Geo, risk controls 1.3126 1.5150* 1.4656* 1.5770 0.079
(0.2631) (0.3393) (0.3208) (0.6342)
Geo, risk and contract controls 1.3344* 1.4962** 1.3421 1.6485 0.084
(0.2212) (0.3026) (0.2843) (0.6123)
Table 8: Test of secondary market hypothesis. This table presents the results of the tests to validate the secondary market channel. Panel
A shows the results of the tests for the validity of pricing R-squared as a proxy for limitations on firms’ capacity to hold loans. Each
specification relates lagged changes in firm performance to changes in the pricing R-squared proxy, using a first differences approach.
Panel B relates the pricing R-squared proxy to firm-level hazard of default. The analysis is a cross-sectional ordinary least squares
regression in which each observation represents one firm. The dependent variable is the hazard coefficient on the firm fixed effect from a
pooled hazard regression on all non-conforming mortgages originated during 2005-2006. All regressions weight each observation by the
number of loans in the database from the lender. Standard errors do not account for estimation errors in pricing R-squared and firm-level
hazard rate. Standard errors in (), ***, **, * significant at 1, 5 and 10% level.
A: Relation between lagged firm performance and pricing R-squared (differenced)
Dependent variable: (Pricing R-squared)
Performance measure: Return on assets Return on equity Profit margin
Builders only All firms Builders All firms Builders All firms
(1) (2) (3) (4) (5) (6)
(lagged performance measure) -3.1469*** -0.5245*** -0.5630*** -0.0786*** -5.7775*** -0.2745***
(0.0802) (0.0639) (0.0319) (0.0144) (0.1758) (0.0304)
Log(lagged firm size) 0.1043*** 0.0509*** 0.0840*** 0.0505*** 0.1153*** 0.0497***
(0.0017) (0.0009) (0.0017) (0.0009) (0.0019) (0.0009)
Constant -0.6736*** -0.2608*** -0.5453*** -0.2667*** -0.7276*** -0.2578***
(0.0152) (0.0085) (0.0157) (0.0084) (0.0158) (0.0085)
Firm category fixed effects N Y N Y N Y
Observations 15 48 15 48 15 48
Adjusted R-squared 0.204 0.360 0.145 0.359 0.183 0.360
B: Relation between pricing R-squared and firm-level hazard
Dependent variable: Pricing R- Firm-level hazard
All firms All firms All firms All firms Excluding Excluding
(1) (2) (3) (4) (5) (6)
Pricing R-squared (2005-06) -0.3884*** -0.2300*** -0.3685*** -0.2223**
(0.0102) (0.0119) (0.0120) (0.1024)
Depository banks -0.1415*** 0.1210*** 0.0396***
(0.0017) (0.0032) (0.0033)
Standalone mortgage banks -0.0085*** 0.0415*** -0.0174*** -0.1221***
(0.0012) (0.0025) (0.0026) (0.0424)
Affiliated mortgage banks -0.2503*** 0.3516*** 0.2489*** 0.0401
(0.0033) (0.0060) (0.0068) (0.0552)
Constant 0.6607*** 0.9320*** 1.2413*** 1.1252*** 1.2362*** 1.2019***
(0.0007) (0.0014) (0.0067) (0.0080) (0.0078) (0.0625)
Observations 172 172 172 172 141 141
Adjusted R-squared 0.217 0.071 0.045 0.084 0.043 0.041
Table 9: Test of foreclosure prevention hypothesis. This table displays the results of tests for evidence that builders’ behavior is motivated
by the desire to prevent foreclosures in their subdivisions. The variable “First quartile home order” and “Last quartile home order” in
Columns (1) and (2) are indicator variables which are equal to 1 if the mortgage is issued in the first or last quartile of the order in which
homes were constructed within a census tract. “Log number of homes by lender” in Column (3) is the log of the sum of all mortgages
funded by a lender in a given census tracts over the whole sample period. “Log number of homes constructed by builder” in Column (4) is
the log of the sum of all homes sold in a given census tract for each builder in our sample. Columns (1)-(3) include all new construction
purchase mortgages from 1999 to 2009, while Column (4) includes only homes constructed (but not necessarily financed) by builders in
our sample. All controls are the same as displayed in Table 3 Column (4). All errors are clustered by lender. See Appendix A.2 for variable
definitions. ***, **, * significant at 1, 5 and 10% level.
Expected Hazard by Fico by Hazard by Hazard by log
magnitude building building log number number homes
for hazard order order homes – builder-sold
models quartiles quartiles only
(1) (2) (3) (4)
Builder 0.6805*** 26.4418*** 0.7163*** 0.7129**
(0.0495) (6.1694) (0.0862) (0.1162)
First quartile home order 0.9521 1.0091
Last quartile home order 1.0070 -0.1811
Builder*First quartile home order <1 1.1148** -1.6549
Builder*Last quartile home order >1 0.9859 0.6727
Log number homes by lender 0.9337***
Builder*log number homes by <1 1.0195
Log number homes constructed by 0.9822
Builder*Log number homes <1 0.9945
constructed by builder
Risk controls Y Y Y Y
Contract term controls Y Y Y Y
Lender controls Y Y Y Y
Zip code fixed effects Y Y Y Y
Vintage year fixed effects Y Y Y Y
Observations 103,518 103,518 209,301 103,518
Pseudo R-squared 0.099 0.218 0.095 0.099