Subprime Governance: Agency Costs in Vertically Integrated Banks

Transcription

Subprime Governance: Agency Costs in Vertically Integrated Banks and the 2008
Mortgage Crisis
Claudine Gartenberg*
NYU Stern School of Business
Lamar Pierce+
Washington University in St. Louis
May 6, 2015
This study examines how corporate governance altered the relationship between vertical
integration and performance in the mortgage industry prior to the 2008 crisis. Prior research
has argued that vertical integration of mortgage origination and securitization aligned
divisional incentives and improved lending quality. We show that vertical integration only
improved loan performance in those banks with strong corporate governance. Detailed
analysis of governance characteristics suggests that this effect is primarily explained by external
monitoring by institutional investors. We interpret these findings as suggesting that the
additional control afforded by vertical integration can, in the hands of poorly monitored
managers, offset gains from aligned divisional incentives. These findings support the view that
the effect of vertical integration depends on the specific characteristics and capabilities of
firms.
*NYU Stern School of Business, Management and Organizations, Tisch 709, 617-378-8710, [email protected]
+Olin Business School, Washington University in St. Louis
1. Introduction
In this paper we provide evidence that the relationship between vertical integration and
performance is contingent on the corporate governance of the integrated firm. We argue that the
aligned incentives of business units under vertical integration do not resolve moral hazard problems
if weak governance allows managers to profit from information distortion. We support this argument
using data from one of the most important market failures in recent history—the 2008 American
housing crisis. Although recent research in corporate finance has represented the vertical integration
of mortgage origination and securitization as having improved lending performance by aligning
internal divisional incentives (Purnanandam, 2011; Demiroglu and James, 2012), some of the
industry’s largest failures were vertically integrated firms. Washington Mutual, Countrywide, and New
Century Financial all pursued deliberate strategies to gain ownership and control over the vertical
chain and all collapsed spectacularly. These notable failures suggest that the relationship between
vertical integration and performance is more complex than its representation in recent studies.
How might this be? Although extensive theory in organizational economics and strategy
explains how vertical integration (or hierarchy) can indeed improve coordination between units by
reducing contracting hazards (Arrow, 1973; Williamson, 1985; Nahapiet and Ghoshal, 1998;
Williamson, 1999), related work argues that hierarchy can also generate coordination costs (Zhou,
2011; Rawley, 2010) as well as perverse incentives for managers to misrepresent, distort, and withhold
information for their own interests (Williamson, 1985; Eccles and White, 1988; Shleifer and Vishny,
1997; Osterloh and Frey, 2000; Nickerson and Zenger, 2004; Bidwell, 2012; Pierce, 2012). The
implication of these theories is that, although vertical integration may create the potential for improved
performance, this potential is only realized when firms possess the internal incentives and managerial
oversight necessary to implement internal transactions and activities with minimal distortion. As
Williamson (1985) repeatedly argues, when high-powered managerial incentives exist with the firm, as
2
is common in the banking industry, the imperfect monitoring and intervention of top managers and
owners is frequently insufficient to restrain this distortionary behavior.
We find evidence consistent with this argument by examining a panel of mortgage lenders
from 2000 to 2007. We first analyze individual mortgages issued in the 100 most active zip codes for
new home construction, measuring the relative distortion of underwriting standards by each lender
during this period. We construct a firm-year measure of lending quality by the incremental likelihood
that a mortgage defaults if it is originated by that firm, controlling for observable risk and borrower
and contract characteristics associated with that loan. If a firm chooses to subsidize its securitization
unit by lowering lending standards, those lowered standards will be captured by this metric.
We first show that, on average, vertically integrated lenders write higher quality loans than
nonintegrated firms. This average effect, however, masks significant differences in the default
likelihood between integrated firms of strong and weak governance. Specifically, integrated firms with
strong governance behave largely as prior empirical studies have found: mortgage default likelihood
decreases as the degree of vertical integration increases (Demiroglu and James, 2012). However, for
firms with weak governance, this relationship fails to hold: default likelihood does not fall with vertical
integration (and in some specifications actually increases). Our results suggest that the advantages of
vertical integration are offset by a weak governance structure.
How might this distortion occur in practice? Within integrated firms, managers have discretion
on both the underwriting quality and on the downstream disposition of loans into securitization pools,
which were subject to less market due diligence than individual loans sold by nonintegrated firms. 1
While managers of all lenders may have had incentives to increase volume through reduced mortgage
quality, managers of vertically integrated firms had a unique moral hazard problem. These managers
Testimony before the Financial Crisis Inquiry Commission, official transcript, Sept 23, 2010,
http://fcic.gov/hearings/pdfs/2010-0923-transcript.pdf, (hereafter FCIC), pages 169, 178.
1
3
controlled both the (hard-to-observe) upstream lending quality and the (also hard-to-observe) terms
by which these loans were transferred between units, as well as the (easy-to-observe and richly
rewarded) downstream securitization volumes. It is this moral hazard problem to which we attribute
the relatively worse observed lending quality by the poorly-governed, integrated firms.
We then demonstrate that, of the detailed governance mechanisms, monitoring by external
investors, particularly those with long-term perspectives, played the largest role in restraining excess
bank risk. The ratio, absolute number, and diffusion of institutional investors is associated with lower
default risk. Also, shareholders that are regulated banks or insurance companies—traditionally the
more conservative owners—are associated with lower default likelihood, relative to owners that are
hedge funds or other investment-oriented entities. In fact, firms with low levels of ownership by banks
or relatedly high levels of ownership by investment companies actually have a positive association
between vertical integration and default likelihood. That is, as vertical integration increases, lending
quality actually decreases, suggesting that these firms increased volume by actively lowering lending
standards. In contrast, we find no clear results from executive compensation or board composition—
two other common components of corporate governance.
This paper contributes to the empirical literature on firm scope and performance by providing
evidence that the integration-performance link is not universally positive in markets where information
accuracy is critical, as prior research has argued (Nickerson and Zenger, 2004; Lafontaine and Slade,
2007). In this sense, it contributes to a growing literature that argues that the firm boundary predictions
of transaction cost economics crucially interact with firm heterogeneity and capabilities (Jacobides and
Winter, 2005; Bidwell, 2010; 2012; Argyres and Zenger, 2012; Argyres et al., 2012). This paper build
on prior studies of horizontal integration (Gartenberg, 2014) and vertical disintegration (Jacobides,
2005) in the mortgage banking industry, but unlike those studies, explicitly examines how
4
heterogeneity within vertically-integrated firms can directly impact the efficiency of that organizational
structure.
This paper also contributes to a deep literature on the importance of corporate governance in
firm strategy and performance (e.g., Jensen and Zajac, 2004; Hambrick et al., 2008; Castañer and
Kavadis, 2013). This literature argues that the incentives embedded in executive compensation, as well
as the expertise (Castanias and Helfat, 1991; Westphal and Fredrickson, 2001; Feldman and
Montgomery, 2015), independence (Jensen and Meckling, 1976; Boyd, 1994; Westphal and Zajac,
1998), and motivation (Hambrick and Jackson, 2000) of the board of directors, all can shape a firm’s
strategic direction and performance. Similarly, the involvement of institutional shareholders is argued
to shape strategy and performance through both improved information and incentives for monitoring
(Shleifer and Vishny, 1986; Schnatterly et al., 2008). Our paper suggests that corporate governance
represents a persistent and heterogeneous firm attribute that directly influences the appropriate
boundary of the firm.
Finally, our paper contributes to the growing literature on the 2008 housing crisis (e.g., Shiller,
2008; Mayer et al., 2009; Shin, 2009). While research has shown that the housing crisis was preceded
by a large deterioration in mortgage quality (Dell’Ariccia et al., 2009), it has generally focused on the
market level, abstracting away from the firm (Demyanyk and Van Hemert, 2011), or focused on the
degree of “skin in the game” as the main determinant of quality differences between lenders
(Purnanandam, 2011; Demiroglu and James, 2012). Our paper, together with Gartenberg (2014),
highlights other organizational factors that have a first order effect on lending differences between
firms. Furthermore, these factors – corporate governance (in our case) and internal capital markets (in
Gartenberg, 2014)– are sufficiently general that it is plausible that they influence behavior of firms
beyond the mortgage industry and this specific time period.
5
2. Theoretical Background
2.1 Vertical Integration and Information Transfer
Both the economics and strategy literatures explicitly address how vertical integration can impact
performance in information-intensive industries such as financial services. Within economics, the
industrial organization literature has traditionally focused on market power benefits (Hart and Tirole,
1990; Lafontaine and Slade, 2007), while organizational economics has focused on contracting benefits
that reduce opportunism and holdup (Williamson, 1985; Grossman and Hart, 1986; Hart and Moore,
1990) and moral hazard (Holmstrom and Milgrom, 1991). Empirical evidence of reduced moral hazard
spans a variety of settings, including retail (Lafontaine and Shaw, 2005), life sciences (Azoulay, 2004)
and trucking (Baker and Hubbard, 2004; Nickerson and Silverman, 2002). In each setting, the
downstream party is responsible for two tasks, one of which is more easily observable and contractible.
To prevent this party from diverting attention away from the less contractible activity, the upstream
party integrates and employs weaker incentives to balance. Related work in strategy argues that
opportunism is less likely within integrated firms, thus facilitating accurate and efficient information
sharing (Nickerson and Zenger, 2004), particularly when two activities are complementary and must
be coordinated (Novak and Stern, 2009).
Yet vertically-integrated firms only enjoy improved information sharing if they can overcome
the opportunism common in hierarchical forms. Within economics, the actor making the integration
decision is implicitly assumed to be the principal or, at minimum, a manager whose interests are fully
aligned with those of the principal. However, research suggests that this may not always be the case.
Managers within firms may choose to distort or selectively transmit asymmetric information when
their incentives are not aligned with the organization (Milgrom and Roberts, 1990; Foss, 2003). These
incentives may be explicit in compensation systems or may be embedded in career concerns (Bradach
and Eccles, 1989) or social and emotional ties (Osterloh and Frey, 2000). Pierce (2012), for example,
6
finds evidence consistent with manipulation by managers of car manufacturers, who are able to
subsidize market share and earnings through distorted estimates of depreciation in lease contracts.
The primarily implication of these studies is that the impact of vertical integration on the performance
of information-intensive firms critically depends on the ability of the firm to control agency costs that
might distort or inhibit internal information transfer.
Williamson (1985) makes a similar argument that hierarchy fails to reduce opportunism when
managers within the firm retain high-powered incentives. Such incentives lead to what he calls
`accounting contrivances’ (1985, p. 138), where managers distort information on both transfer prices
and cost. The owners or top managers of firms are unable to resolve such problems because the costs
of monitoring are unavoidable. The bounded rationality of owners and top managers makes accurate,
timely, and complete intervention impossible, a problem that grows with the size and complexity of
the firm. In our case, we are agnostic whether these “contrivances” originate at the CEO level or
within the firm. Our basic intuition is that integration increases the opportunities for contrivances
overall, and that these opportunities are less restrained in firms with weak governance.
2.2 Corporate Governance and Managerial Agency Problems
The ability of a firm to control internal agency problems such as information distortion can
be represented by the quality of its corporate governance. The firm’s control over designing the
incentives of managers and the structure for monitoring manager conduct can restrain behavior that
is costly to shareholders. This monitoring can occur both through boards of directors and by external
(often institutional) shareholders.
The firm’s owners can directly influence the degree of managerial monitoring, but their
effectiveness depends on their ability and incentives to engage in this monitoring. Research in finance
has long argued that ownership concentration is an important factor in reducing agency problems,
since highly diversified ownership across diffuse stockholders generates weak incentives to focus
7
monitoring efforts on any one firm (Shleifer and Vishny, 1986; Admati et al., 1994) and because
smaller shareholders will tend to free-ride off larger ones. These predictions have been largely
supported by empirical work linking financial performance and other outcomes with large
shareholders (Bethel et al., 1998; Bertrand and Mullainathan, 2001; Schnatterly et al., 2008). Similarly,
institutional ownership, particularly among those with long-term investment perspectives, is thought
to improve governance (Gorton and Kahl, 1999; Gillan and Starks, 2007), although evidence on its
effectiveness is mixed.
Monitoring is also thought to be related to the size and composition of the corporate board,
although theoretical and empirical literatures do not agree on the exact nature of this relationship.
Board independence, for example, is often argued to be important for effective monitoring because
of reduced conflicts of interests (Weisbach, 1988). The evidence, however, is mixed. Although
Hermalin and Weisbach, (2001) conclude that independent boards do appear to implement better
policies, they find no consensus in the literature that this leads to better financial performance, possibly
because independent boards possess less firm-specific knowledge (Feldman and Montgomery, 2015)
or are less involved with the firm (Westphal, 1999). Similar mixed evidence exists on whether board
size relates to the quality of monitoring. Hermalin and Weisbach (2001) conclude that small boards
appear to make better decisions, but these decisions do not appear to improve firm performance.
Executive compensation is often argued by scholars in finance and strategy to be an important
tool for the board to align manager interests with the interests of shareholders. However, others argue
that compensation largely reflects management’s power in capturing the board of directors (Boyd,
1994; Westphal and Zajac, 1995; Bebchuck and Fried 2004; Garvey and Milbourn, 2006; Gopalan et
al., 2010), and has complex interactions with many other characteristics of the firm (Finkelstein and
Hambrick, 1989, Zajac and Westphal, 1994). Consequently, the empirical literature on the link between
executive compensation and performance is inconsistent (Murphy, 1999).
8
2.3 Corporate Governance and Vertical Integration
Despite extensive literature on both vertical integration and governance, few empirical studies
have examined their interaction and, particularly, how it affects firm behavior. This shortcoming is
important because corporate governance represents the type of distinct and persistent firm
heterogeneity that is argued to help determine the efficiency of vertical integration decisions (Jacobides
and Winter, 2005; 2012; Argyres and Zenger, 2012; Argyres et al., 2014). Although Novak and Stern
(2009) and Pierce (2012) argue that a lack of managerial monitoring, or poor corporate governance,
generates poor performance in vertically integrated firms in the automotive industry, neither observe
the variation in governance necessary to test this proposition.
Despite this shortage of empirical work, a substantial theoretical body suggests that vertical
integration should create unique challenges that might elevate the need for corporate governance,
particularly in information-intensive industries such as financial services. Information sharing between
firm divisions is rife with opportunity for distortion, even when incentives are aligned among divisions
(Eccles and White, 1988). Managerial incentives might not be perfectly aligned with those of their
divisions (Williamson, 1985; Foss, 2003), particularly when political power and long-term career
concerns might motivate influence activities (Bradach and Eccles, 1989; Milgrom and Roberts, 1990;
Wulf, 2002; Argyres and Mui, 2007). Agency behavior may occur even in the presence of aligned
incentives, since managers may be motivated by emotional or social factors (Osterloh and Frey, 2000).
In the absence of strong corporate governance, where management is both effectively
monitored and appropriately incentivized, we therefore expect the information sharing gains from
vertical integration to dissipate. Incentive alignment and better information transmission at the
division level may be impeded by lack of oversight of managers and perverse incentives for them to
pursue their own goals and agendas. We therefore expect that while vertical integration may indeed
improve performance outcomes in information-intensive industries such as banking, it does so only
9
when corporate governance structures are strong. Figure 1 represents the predicted differences in
information quality gains from vertical integration between firms with strong and weak governance.
Hypothesis 1: Information sharing gains from vertical integration will be greater in firms with strong corporate
governance.
<<< INSERT FIGURE 1 HERE >>>
3. Empirical Setting
We explore vertical integration and governance in the context of the mortgage industry. We
define vertical integration as the combination of mortgage origination and securitization within a single
parent firm. Mortgage origination, the “upstream” function, is the process of creating and
underwriting individual mortgages. Mortgage securitization, the “downstream” function, is the process
of pooling together individual mortgages and issuing mortgage-backed securities (MBS)—financial
instruments that are backed by the underlying mortgage pool.
Prior to 2000, securitization by private industry participants (so-called "private-label
securitization") was relatively uncommon. Much of the mortgage industry was disintegrated due to
gains from specialization and the standardization of mortgage information that facilitated coordination
(Jacobides, 2005). Originators typically either held the loans for their lifespan or sold them to
government-sponsored entities 2 or to deposit-taking banks. With the widespread expansion of credit
in 2000s, private-label securitization grew rapidly (Mian and Sufi, 2009). Rising home prices,
expectations of continued price appreciation, and low unemployment led to few mortgage defaults
that – combined with favorable credit ratings by ratings agencies – in turn resulted in laxer screening
by MBS investors than purchases of whole loans. 3 The additional demand for mortgage-backed
5 We use HighG instead of G itself in our main specification (and similarly with other governance variables) to ease
interpretation of coefficients. We also obtain statistically and economically similar results using raw governance measures
5 We use HighG instead of G itself in our main specification (and similarly with other governance variables) to ease
10
securities also led to systemically deteriorating lending standards throughout the industry until the
housing bubble burst in 2008 (Demyanyk and Van Hemert, 2011).
In this environment, vertical integration by industry participants occurred in both directions –
firms that were primarily mortgage originators engaged in securitization (e.g., Countrywide Financial
and New Century Financial) and firms that were primarily commercial financial institutions with
securitization operations expanded into mortgage origination (e.g., Goldman Sachs and Bear Stearns).
This setting provided conditions that encouraged managers to inflate securitization revenues by
engaging in increasingly risky lending. First, the high demand for MBS made securitization operations
very profitable for firms. Second, rising home prices led to a prolonged period of low mortgage
defaults. These low defaults made it hard to monitor deteriorating lending standards externally because
the negative outcomes of lending decisions were largely deferred until housing prices ceased to
appreciate in mid-2006. Third, the retention of residual cash flows (the so-called “equity tranche”) by
securitizing firms encouraged less thorough screening by investors. 4
With lax screening by MBS investors and low default rates masking poor lending standards,
we argue that poorly-monitored managers lowered lending standards to increase securitization volume
and short-term profits. These reduced lending standards could result in two ways. First, lenders could
target a risky customer segment, such as consumers with low credit scores or income levels. Targeting
a riskier segment is not equivalent, however, to lax lending or poor operating performance: segmenting
customers by risk is a valid strategy if it is priced appropriately, particularly when the firm possesses
specific capabilities for managing high-risk segments. Second, lenders could reduce the quality of
underwriting (screening and matching consumers with appropriate financial products), conditional on
customer risk segment. Lower underwriting quality could involve fraudulent applications, such as
We use HighG instead of G itself in our main specification (and similarly with other governance variables) to ease
5
11
misstatement of income or employment status. Alternatively, it could result from less effort in
evaluating “soft” information about a consumer, such as the quality of future employment and
earnings potential, or general trustworthiness and character. This soft information, important for all
risk segments, is an important determinant of mortgage default. We label this second form of lending
standards as “underwriting hazard” in this paper.
4. Data and Methods
4.1 Empirical Strategy
At a high level, our empirical strategy is to first replicate the findings from prior research that
“skin in the game” is correlated with lending quality and then demonstrate that this result is contingent
on the governance of the firm. To implement this strategy, we first to construct a panel that includes
firm-year measures of mortgage default likelihood and other mortgage-related controls. We then
merge this panel with our vertical integration measure—the amount of securitization performed by a
firm in a given year—and measures of firm governance from the finance and economics literatures.
Our choice to construct a firm-year panel for analysis, rather than using individual loans as the unit of
analysis, is based on several important and related factors. First, individual loans in our data do not
represent independent firm choices, but instead collectively represent a firm’s underlying underwriting
(and securitization) strategy at a given time. Because our independent variables of interest, governance
and vertical integration, are firm-level constructs, we were concerned that a loan-level analysis would
overstate the number of independent observations in our dataset. Second, because there is variation
in firm portfolio size in our dataset, we were concerned that loan-level data would overweight the
importance of large lenders in testing our firm-level hypotheses.
We will first demonstrate that our firm-level panel produces results consistent with the recent
paper by Demiroglu and James (2012) when employing their specification. This step is important for
12
establishing that our data is substantively similar to prior work that demonstrates a positive
relationship between vertical integration and risk-taking. Once this prior result is replicated, we then
explore how this relationship is moderated by governance. We show that governance is a critical firmlevel moderator and that its inclusion in our specification dramatically changes the relationship
between vertical integration and performance.
We then examine the role of shareholder composition by substituting our governance indices
with various characteristics of institutional owners. Finally, we relate vertical integration and
governance to whether the firm was still operating at the end of 2010. While the results of this last
analysis are suggestive at best, they are at least consistent with the implicit link between default
likelihood and overall risk to the firm.
4.2 Lending quality, vertical integration and governance
Appendix B describes the detailed methods by which we calculate default likelihood as a firmyear measure of the likelihood that a loan underwritten by a bank defaults, conditional on the
observable loan attributes. This approach is similar to risk-adjusted performance measures used in the
health economics literature (e.g., Huckman and Pisano, 2006), where hospital or surgeon performance
is calculated by estimating mortality or morbidity conditional on observable patient characteristics
known to increase risk. This approach has also been used to estimate emissions testing fraud based
on suspiciously high pass rates (Pierce and Snyder, 2008; Bennett et al., 2013).
Our second stage is a regression with default likelihood as the dependent variable.
Governance, vertical integration and their interaction are the independent variables of interest,
together with relevant controls. The main specification is as follows:
𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐷𝐷𝑖𝑖𝑖𝑖 = 𝛽𝛽0 + 𝛽𝛽1 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑆𝑆𝑖𝑖𝑖𝑖 + 𝛽𝛽2 𝐻𝐻𝐻𝐻𝐻𝐻ℎ𝐺𝐺𝑖𝑖𝑖𝑖 + 𝛽𝛽3 𝐻𝐻𝐻𝐻𝐻𝐻ℎ𝐺𝐺𝑖𝑖𝑖𝑖 𝐿𝐿𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑆𝑆𝑖𝑖𝑖𝑖 + 𝑋𝑋𝑖𝑖𝑖𝑖′ 𝛿𝛿 + 𝜖𝜖𝑖𝑖𝑖𝑖
(1)
Where FirmDL is the firm-level default likelihood (represented as δ in equation 1 above) for firm i in
year t, defined in the section below, that serves as the performance measure for the upstream lending
13
units. LogMBS is the log of the residential mortgage-backed securities issued by the firm during year t.
HighG is an indicator that the firm has an above-median G-index measure (“G”), where the managerial
“dictatorship” level increases with G (Gompers, Ishii and Metrick 2003) (see Appendix C). 5 For
robustness, we collected two additional governance indices: the Entrenchment Index (EI) (Bebchuk
et al., 2009) and the Anti-Takeover Index (ATI) (Cremers and Nair, 2005). X is a vector of firm-level
controls, including both general measures for all firms and financial information from Compustat for
public firms (See Table 1 for a list of these controls).
4.3 Data and variable construction
The data are constructed from several primary sources. The mortgage data used to calculate
default likelihood come from merging county public records with a national mortgage servicer
database through the cooperation of CoreLogic. The securitization data were obtained from
Thompson SDC. The G-index, Entrenchment Index and Anti-Takeover measures were obtained
from the Investor Responsibility Research Center via WRDS. These main data sources were
supplemented with firm data from Compustat for public firms. Firm age, merger and survival data
(whether the firms were in operation by the end of 2010) were hand-collected from Capital IQ and
other public sources. Macroeconomic data for the calculation of firm default likelihood were obtained
from Freddie Mac, US Census Bureau and the Bureau of Labor Statistics. The sample construction is
described in more detail in Appendix C.
4.3.1 Descriptive statistics: Table 1 contains descriptive statistics of the data. We can see that
approximately 33% of the firm-year observations are firms that issued MBS during that year, with the
principal averaging a total of $18,893 million. The mean diversification index is 0.4690 with a wide
variation across lenders. Lastly, we were able to gather firm failure data for 162 of the 170 firms in the
We use HighG instead of G itself in our main specification (and similarly with other governance variables) to ease
5
14
panel and find that 46% of the firms were still operating by the end of 2010. Panel B shows statistics
for the mortgage-level dataset used to construct lending quality and loan-related controls. The average
home price is $316,848, higher than the national average, reflecting the Alt-A subsample which
includes so-called “jumbo-loans” (loans above the conforming loan limits required for a government
guarantee). The combined loan-to-value is 86%, above the standard “rule of thumb” of 80%, which
reflects the period’s high household leverage. The prevalence of loan clauses such as prepayment
penalties and interest-only and negative amortization provisions, are also representative of the risky
loan structures common during this period. Similarly, most of the loans (77%) are classified as “low
or no documentation,” another indication of the laxer underwriting. The mean decline in home prices
in the sample from the peak of the cycle to the trough in 2009 is 21.7 percent, with a maximum of 39.3% in Nevada (primarily Las Vegas) and a minimum of +0.28% (essentially flat) in North and
South Carolina.
<<< INSERT TABLE 1 HERE >>>
Appendix Table A1 shows the variable correlations. The vertical integration measure is negatively
correlated to default likelihood, supporting the prior result that “skin in the game” does matter,
although the magnitude is small and statistically insignificant. The three governance indices (G, ATI
and EI) are all positively correlated with default likelihood (and statistically significant for two of the
three measures) and negatively correlated to whether the firm was still in operation in 2010.
5. Empirical Analysis
5.1 Sample Validation
Given prior results by Demiroglu and James (2012) demonstrating a positive relationship
between vertical integration and lending performance, we first show that our different data sample
produces similar results using their models. To do so, we reproduce their logistic regression model
15
where the unit of analysis is the individual mortgage and the dependent variable is an indicator of
mortgage default. Our independent variable of interest is the level of vertical integration, or the logged
volume of mortgage backed securities in the year of loan origination, controlling for firm assets. As in
their analysis, we initially restrict our sample to Alt-A mortgages originated between 2006 and 2007
and calculate default likelihood using a similar set of control variables. These control variables include
some borrower risk measures (FICO (credit) score, loan-to-value ratio), mortgage characteristics
(indicators for floating interest rate, low- or no-documentation, negative amortization and prepayment
penalty provisions) and a measure of local home price decline. Their controls do not include other
measures that we subsequently include in our main analysis, such as additional borrower risk measures
(loan interest rate, debt-to-income ratio) and more detailed mortgage characteristics (indicators for
floating, hybrid and balloon provisions, interest-only pricing, multiple payment options, new
construction) and more detailed geographic and macroeconomic controls (census tract median
income, state indicators, Freddie rates and Federal Reserve funds rate). These latter controls are strong
predictors of default and commonly used to assess mortgage risk by both underwriters and third-party
loan purchasers; however, they are often only available in proprietary and anonymized form. To our
knowledge, our study and Gartenberg (2014) are unique in having access to this full set of controls
matched to originator identity.
Column 1 of Table 2 presents the replication results using the DJ time period, mortgage
classification (Alt-A), control variable set, and standard error treatment. Following their approach, we
cluster our errors at the MSA (metropolitan area) level. The coefficient on vertical integration is
qualitatively similar to their estimate, with a negative and statistically significant relationship between
vertical integration and default likelihood.
16
Columns 2 – 6 reproduce and then extend this analysis at the panel level, with firm-level default
likelihood replacing loan default as the dependent variable. 6 For our dependent variable in Columns
1-3, we use a default likelihood calculated in a first stage that includes only the control variables used
in DJ. As with Column 1, Column 2 restricts the panel to 2006-07, and we find a negative (albeit
insignificant) coefficient. Column 3 expands the panel to include 2000-2007 and the coefficient is now
negative and strongly significant, corresponding to the results in DJ. Column 4 replaces robust
standard errors with more conservative block bootstrapping at the lender level, which treats the error
terms within a lender as correlated. 7 The results remain significant. These models provide confidence
that our panel yields substantively similar results to those used by DJ. Column 5 repeats Column 3,
replacing the dependent variable with default likelihood calculated in the first stage using our full set
of controls. The coefficient reduces substantially in magnitude and is no longer significant. Column 6
adds the controls in the second stage and the coefficient becomes positive and remains insignificant.
The difference between the Columns 1-4 and Columns 5-6 underscores the importance of controlling
for observable risk in the analysis. It also suggests that skin-in-the-game may have led to targeting of
safer populations, rather than more diligent underwriting practices.
5.2 Governance, vertical integration and loan quality
Our replication analysis shows that, although our sample produces similar results to prior
work, the results are also sensitive to omitted firm- and loan-level variables. Although we cannot
extrapolate how our additional control variables would impact analyses using these other samples, our
results suggest that the relationship between vertical integration and underwriting performance is not
uniformly positive. We next examine the role of governance and show that it is a critical determinant
of how vertical integration relates to loan quality.
6
7
Appendix Table A2 replicates this analysis at the loan level.
Throughout the analysis, we block bootstrap our standard errors (by lender) with 800 repetitions.
17
We begin the analysis by showing a simple scatter plot of our data in Figure 2a. The x-axis
plots our vertical integration measure, the log of MBS issued by a firm in a given year. The y-axis plots
the default likelihood. The diamond markers refer to firms with strong governance (G values below
median), while the circle markers refer to firms with weak governance (G values at or above median).
Each marker is weighted by the number of mortgages issued in the mortgage database. Clustered on
the left of the plot are the non-integrated firms (that did not issue MBS), while the remainder of the
plot includes the integrated firms that issued MBS. Figure 2b repeats the plot, replacing the scatter
plot with linear fits of both the High G and Low G firms. Two results are apparent from these plots.
First, High G firms appear to have higher default likelihood than Low G firms. Second, the
relationship between vertical integration and default likelihood appears to be fundamentally different
for high G and Low G firms, as is evident from the different slopes in Figure 2b. This second result
is a preliminary version of one of the main findings of the paper.
Figure 3 provides a related visual depiction of our analysis. We divide the data into four
subsamples – integrated and non-integrated for both high-G and low-G firms - and plot the kernel
densities of default likelihood from our first-stage regression. The figure strongly suggests a role of
governance in vertically-integrated firms. For well governed low-G firms, the distribution of default
likelihood for vertically-integrated firms is substantially to the left of the non-integrated firms, which
is consistent with vertical integration providing performance benefits when governance is strong.
However, for high-G firms with weak governance, we see no apparent difference between the
distributions of integrated- and non-integrated firms.
Figures 2 and 3 together suggest that governance plays a major role in defining the relationship
between vertical integration and default likelihood.
18
Table 3 provides multivariate results for how governance moderates the effect of vertical
integration on default likelihood. 8 For this table and all subsequent analyses, we include only the
subsets of the panel for which the governance measures are available. 9 Column 1 is the baseline
regression, containing the full set of first stage controls to calculate default likelihood. Column 2
further adds in a substantial set of firm-level control variables detailed in Table 1. In both models,
vertical integration continues to have a negative relationship with risk, but only for well-governed
firms with low G scores. The High G Indicator*Log(MBS) interaction is positive and significant and
shows the divergence in the behavior of high and low-G firms. For low-G firms, underwriting hazard
decreases as the levels of securitization increases. This effect is economically significant: a one standard
deviation increase in log securitization results in 35% of a standard deviation decrease in mortgage
default likelihood. In contrast, for high-G firms (firms with poorly monitored managers), default
likelihood appears to be unaffected by vertical integration, as the sum of the main effect and
interaction term are not statistically different from zero (Wald: p = 0.4427).
In columns 3-6, we repeat our base and fully-controlled models using the Entrenchment Index
and Anti-takeover Index and finds nearly identical results. Collectively, these models show our results
to be robust to different measures of overall firm governance. 10
For space purposes and readability, we display only the coefficient estimates for the main independent variables of
interest, and suppress the estimates for the control variables. Appendix Tables A3 and A4 reproduce Tables 2 and 3 with
the controls displayed, as do all the Appendix Tables.
9 In unreported tables, we also conduct an alternative analysis on the complete panel in which we include dummies for
missing governance variables. We obtain substantively identical results.
10
One observation from our regressions is that the coefficients on our governance indices in all six models are negative.
This negative sign should be interpreted as the marginal effect of weak governance on the default hazard of nonintegrated firms. This result is consistent with data in Figures 2a and 3 showing lower average hazard for firms with high
G-values. Although this seems inconsistent with theory on the role of governance in determining risk, we note that the
parameter is only marginally significant in two of the six models in Table 3 and insignificant in the other four. Although
we can only speculate on this imprecise coefficient, one possibility is simply that, because non-integrated mortgage
originators have no skin in the game (Demiroglu and James, 2012), there may have been perceived to be little financial
cost to shareholders from excessive risk in their portfolios. Regardless, the role of governance in vertically-integrated
banks, which is the focus of this paper is clear—the relationship between skin-in-the-game and quality is undermined by
weak governance.
8
19
5.2.1 Robustness
Several questions arise from this analysis. First, do well-governed and poorly-governed
integrated firms differ on other dimensions that could drive our results? Appendix Table A5 shows
the balance between these firms in size, age and industry: Well-governed integrated firms are larger,
issue more loans and are likelier to be depository banks than poorly-governed integrated firms. We
therefore replicate the analysis of Table 3 on a matched sample of firms and report our results in Table
A6. We performed a stringent match, dropping 43% of observations of integrated firms in order to
select a subsample in which well-governed and poorly-governed firms matched on observables (see
last two columns of Table A5 for the balance of the matched sample). Table A6 shows that our results,
while somewhat attenuated, are largely replicated on the matched sample.
A second question is whether the log of mortgage-backed securities issued is the best vertical
integration measure in our context. By controlling for total firm assets, this measure captures the
relative scale of mortgage securitization within a firm’s operations, which we consider to be a
definition of vertical integration that aligns with our proposed mechanism of complementarities
between governance quality and firm boundaries. An alternative approach is to normalize
securitization activity by either firm size or, more directly, by origination activity. We define three
additional measures of vertical integration: i) a 0/1 dummy whether the firm issued any MBS, ii) MBS
divided by the number of loans issued by the firm in our dataset and iii) MBS divided by firm assets. 11
Importantly, the correlations between these various measures and our main measure of vertical
integration run between 0.18 and 0.49, showing that they are meaningfully different from each other.
Our results are robust to this alternative approach. Tables A7-A9 replicate Table 3 using these
measures and show that the results remain, if not strengthen.
11
We use logs for the latter two measures in order to produce less skewed distributions.
20
5.3 Specific Governance Mechanisms
Given the relationship between aggregate measures of governance and vertical integration, we
next examine underlying mechanisms that may drive our results. We divide these mechanisms into
two categories. The first is external governance, in the form of the concentration and characteristics
of the firm’s institutional shareholders. The second is internal governance, including board
composition and structure of executive incentives. One of the challenges for this analysis, in common
with many governance studies, is that governance choices are both endogenously determined by firm
characteristics and co-determined with each other. Similar to other recent studies (Erkens et al., 2012;
Beltratti and Stulz, 2012), we are cautious about drawing causal conclusions about these results; rather,
we view them as valuable correlations that suggest how shareholder composition can be used as
indicators to predict management actions. Appendix Table A10 presents correlations between the
governance variables used in this analysis. Consistent with the notion that governance components
are co-determined, CEO share ownership is negatively correlated to board size, CEO and CFO share
price and volatility sensitivities are also strongly correlated to overall institutional ownership and
composition, as are board size and independence. Since we cannot disentangle these factors, the
following results must be interpreted in that context.
5.3.1 External Shareholder Composition
We focus on measures of institutional ownership, examining i) the ratio of institutional to noninstitutional investors, ii) the number and iii) concentration (HHI) of institutional investors and finally,
iv) the types of firms that make up the investor base. We repeat our panel-level analysis using these
measures for our governance variables and regressing default likelihood on Log(MBS), the specific
governance measure, and their interaction. As with Table 3, we repeat these regressions both with and
without controls.
21
Table 4 presents regression results on the role of institutional ownership on vertical integration
and default likelihood. Columns 1 and 2 examine the institutional ownership ratio, calculated as ratio
of outstanding shares owned by institutional owners to total outstanding shares, while Columns 3 and
4 tests the number of institutional investors. Columns 5 and 6 investigate IO concentration, measured
as the Herfindahl index of all institutional investors. For firms with high levels and counts of
institutional ownership, as well as low concentration, vertical integration is associated with low default
risk. However, for other firms, the interaction term is sufficiently large to counteract the main effect
of vertical integration.
Table 5 presents results for bank and insurance (columns 1 and 2) and investment company
(columns 3 and 4) ownership. Results are again similar to the figures. Vertically-integrated firms with
high bank and insurance ownership have lower default risk, while those with investment company
ownership have higher default risk. These results are consistent with the belief that while bank and
insurance companies are more conservative, investment companies tend to be more aggressive (Del
Guercio, 1996; Falkenstein, 1996).
<<< INSERT TABLES 4 AND 5 HERE >>>
We propose the following interpretation of the findings: it appears that external monitoring
has a strong and fairly straightforward moderating effect on integration. Concentrated institutional
owners – particularly conservative owners such as banks and insurance companies – are associated
with a negative relationship between integration and default likelihood. We cannot disentangle
whether these shareholders actually monitored lending behavior more effectively or if they instead
selected higher quality firms as investing targets. Plausibly, both of these factors – treatment and
matching – were present in this context.
5.3.2 Internal Governance and Incentives
22
In Appendix Tables A11 and A12, we report tests of the moderating effects of CEO and CFO
compensation and board size and composition. Altogether, we find some suggestive correlations but
no consistent moderating role for these measures of executive incentives and board monitoring, with
the possible exception of CEO shareholdings. We caution, however, that our null results cannot
determine that these features played no role in mortgage lending quality, given the endogeneity of
both internal and external governance characteristics of firms. We can say, conservatively, that these
attributes are not as predictive as external shareholder composition. Appendix D contains a more
detailed discussion of these results.
5.4 Governance, vertical integration and firm failure
In our final analysis, we investigate the link between governance, vertical integration and firm
failure. We interpret our results here cautiously since many factors contribute to the failure of these
lenders during the study period. However, we include it as one piece of evidence that higher default
likelihood was not a successful strategy, at least as measured by ex post firm survival.
To perform this analysis, we replace default likelihood as the dependent variable with a 0/1
indicator that the lender was still operating by the end of 2010 - the case for 46% of the sample’s firms
– and then collapse the observations into a cross-sectional dataset where we demean all control
variables from 2003 onward. We do this latter step since the dependent variable varies at the firm, and
not at the firm-year, level. The results of our analysis are shown in Table 6. Columns 1 and 2 show
the results of a logit specification that relates whether the firm was still in operation at the end of 2010
to the default likelihood used as the dependent variable in earlier analyses. We also include an indicator
whether the firm received government support through the emergency TARP funding plan, which
significantly improved firms’ likelihood to survive. We show a strong negative correlation between
default likelihood and firm failure, providing evidence that firms that engaged in worse lending, as we
measured it, were also significantly likelier to fail. Columns 3-4 replace default likelihood with
23
governance, vertical integration, and their interaction. Interestingly, we show that only the interaction
terms are significant; that is, firm failure is predicted only by the combination of weaker governance
and vertical integration, and not by either term alone. Again there may be alternative explanations for
this observed correlation, but it supports the notion that managers of weaker-governed firms used
control over broader scope to engage in value-destructive behaviors.
6. Empirical Challenges
Our interpretation of the paper’s results raises several important questions. First, can we
reasonably interpret higher default likelihood as worse performance by firms? One might argue that
managers simply took on more risk that was, ex ante, optimal for the firm. In this alternative
interpretation, the fact that these firms failed at greater frequency by the end of 2010 is evidence that,
ex post, their choice of riskier lending did not pay off, and not that vertical integration enabled worse
behavior by managers. For this to be true, we would have to believe that integrated firms with weak
governance had different ex ante optimal risk thresholds than integrated firms with strong governance.
Most problematic for our interpretation would be differences that affect the capital costs of these
firms (and hence their optimal risk thresholds), such as geographic concentration, asset mix or firm
size. While our controls largely account for these differences, we do not have the data to rule out this
possibility fully. However, we believe that this alternative is unlikely for several reasons. First, our data
supports reduced risk-taking for weak governance firms: these firms are smaller on average and less
geographically diversified than their strong governance counterparts. While we do not have a measure
for asset diversification, if we assume it to be correlated with geographic diversification (larger banks
operate in more states and simultaneously have wider product lines), then these weak governance
firms would also have less diversified asset bases than strong governance firms. Finally, other factors
equal, it is not clear why capital costs should be lower as governance decreases. If anything, capital
24
providers should be more wary of dealing with these firms, who should respond by decreasing the
overall risk that they assume.
More generally, a significant empirical challenge with our study is how to draw causal
inferences from a cross-sectional research design. Integration and governance decisions are both
endogenous. Absent an outside shock or appropriate instrument, systematic differences between
integrated and non-integrated firms may simultaneously drive the integration decision, governance,
and performance differences. Alternately, governance might drive both the integration decision (e.g.,
Amihud and Lev, 1981; Castañer and Kavadis, 2013) and performance. Addressing endogeneity is a
widespread challenge in studying both governance and vertical integration and one for which we have
no solution in our current data. Although some prior work has used an instrumental variable approach
to attempt to address this common challenge, commonly used instrumental variables such as firm
choices in prior years are unlikely to satisfy the exclusion restriction. Ultimately, we can only speculate
on causality and address this challenge by examining the plausibility of alternative interpretations of
our results.
Aside from differences in optimal lending thresholds discussed above, a second alternative
that could explain our main interaction result is that low quality CEOs encouraged worse underwriting
and high levels of securitization, but these two choices were independent of each other. We do not
see this explanation as inconsistent with our story. Low quality governance enables the persistence of
low quality CEOs, who are partly defined by their choices to make profit-reducing decisions for their
own benefit. It is certainly feasible that the selection and removal of low quality CEO’s is one of the
key mechanisms through which governance moderates the integration/performance relationship.
Another challenge related to the cross-sectional design is the interpretation of results using
specific governance attributes. External shareholder composition, board attributes and executive
compensation are endogenously and jointly determined. In this case, we believe that documenting the
25
correlational results is a contribution: the moderating effect of specific governance attributes on
vertical integration has not been demonstrated or discussed in prior research and raises multiple
avenues for future research. More challenging is how to interpret differences between the governance
attributes. The moderating effect of external shareholders appears strong while less so for executive
compensation. However, it is not possible to conclude with confidence that external shareholder
composition is the primary moderator of vertical integration.
Appendix E also considers three additional empirical challenges that we omit from this general
discussion because of space limitations: 1) whether we can distinguish between rent-seeking behavior
by CEOs and behavioral explanations such as overconfidence (Malmendier and Tate, 2005; Galasso
and Simcoe, 2011) or simple myopia, 2) sample size considerations, given the number of firms in our
panel for which governance data were available and 3) our definition of vertical integration and
alternatives.
7. Conclusion
This study shows that the relationship between integration and performance is strongly
contingent on governance quality. We find that the combination of vertical integration and strong
governance is indeed associated with better firm performance, as measured by the likelihood of
mortgage default. Conversely, the presence of weak governance appears to eliminate any gains from
integration. This result is broadly consistent with Williamson’s (1985) argument that hierarchy is not
universally successful in mitigating hazards, but rather that the inclusion of high-powered incentives
within the firm combined with constraints on oversight and selective intervention yield accounting
manipulation and distorted transfer prices. More importantly it supports the recent argument by
strategy scholars that the incentive and coordination gains from vertical integration are not
independent of other firm characteristics such as resources or capabilities (Mayer and Argyres, 2005;
26
Argyres and Zenger, 2012; Argyres et al., 2012; Jacobides and Winter, 2005; 2012). Although we are
wary of designating corporate governance as a capability, it certainly represents a heterogeneous and
persistent resource that improves firm performance. In that sense, our results seem to validate the
importance of examining the intersection between multiple theoretical approaches—in this case,
agency theory, transaction cost economics, and the resource-based theory.
Our study provides an interesting complement to earlier work by Jacobides (2005), whose
important qualitative study of mortgage banking a decade earlier focused on the disintegration of the
industry due to gains from information standardization. His work emphasized the importance of the
coordination `frictions’ presented by Coase (1937) in determining firm structure in this industry. Our
work, while acknowledging this importance, suggests that the transaction cost of opportunism
emphasized by Williamson (1985) is an equally if not more powerful factor in the efficiency of vertical
integration. One significant insight from Jacobides (2005) is that the vertical disintegration decisions
of many of the firms in his study (including Countrywide) are far from the careful strategic alignment
decisions predicted by theory. Instead, they were driven by short-term profit and market share
considerations. Our study suggests that, regardless of the cost efficiencies of information
standardization, the hazards of information distortion emphasized by transaction cost economics and
agency theory were ignored by a set of myopic, weakly-governed integrated firms.
Our results suggest that external monitoring is an important governance mechanism for
constraining managerial agency problems. Lenders whose investors have long time horizons or the
best market information (e.g., other banks) have the lowest default risk. We note that, in an equity
market with record participation, there may be sufficient capital from aggressive or less-informed
sources to support (in the short-run) lenders whose managers engaged in the securitization of
underpriced poor-quality mortgages. In contrast, boards monitoring and executive compensation do
not consistently or strongly moderate the association between firm boundaries and performance,
27
which may reflect board capture or the inability of the board and top executives to understand
complex internal and market dynamics.
Vertical integration influences performance through a multitude of mechanisms, many of
which interact with other elements of organizational design such as incentives, monitoring,
competition, and regulation. Our study can only provide a body of descriptive evidence on how one
such element, governance, changes the integration-performance relationship, and thereby illustrate
potential oversights by prior literature. It cannot, however, answer questions of causality nor isolate
internal organizational mechanisms. We believe that the gross economic importance of our setting
elevates the relevance of our correlational evidence, but we encourage scholars with settings with
better internal organizational data or exogenous shocks to governance (or vertical integration) to
explore many of the issues that we cannot address here. Data that can meet all such conditions are
rare, however. Although a few papers might exploit an exogenous shock to organizations, such as
Norway’s regulated inclusion of female board members (Ahern and Dittmar, 2012; Kogut et al., 2013)
or a regulatory shock to scope (Natividad and Rawley, 2012), identifying organizational data on other
dimensions and understanding the source of variation in those data, remains a challenge.
References
Admati AR, Pfleiderer P, Zechner J. 1994. Large shareholder activism, risk sharing, and financial market
equilibrium. Journal of Political Economy 102(6): 1097-1130.
Ahern K, Dittmar A. 2012. The changing of the boards: The impact on firm valuation of mandated female
board representation. Quarterly Journal of Economics 127(1): 137-197.
Amihud A, Lev B. 1981. Risk reduction as a managerial motive for conglomerate mergers. Bell Journal of
Economics 12(2): 605-617,
Argyres N, Mui V. 2007. Rules of engagement, credibility, and the political economy of organizational dissent.
Strategic Organization 5: 107-154.
Argyres Nicholas S, Felin Teppo, Foss Nicolai, Zenger Todd. 2012. Organizational economics of capability
and heterogeneity. Organization Science 23(5): 1213-1226.
Argyres Nicholas S, Zenger Todd. 2012. Capabilities, transaction costs, and firm boundaries. Organization
Science 23(6): 1643-1657.
Azoulay P. 2004. Capturing knowledge within and across firm boundaries American Economic Review 94(5):
1591-1612.
28
Baker G, Hubbard T. 2004. Contractibility and asset ownership: On-board computers and governance in U.S.
Trucking. Quarterly Journal of Economics 119(4): 1443-1479.
Balachandran S, Kogut B, Harnal H. 2010. The probability of default, excessive risk, and executive
compensation: A study of financial services firms from 1995 to 2008. Columbia Business School Research
Paper.
Bebchuk L, Cohen A, Ferrell A. 2009. What matters in corporate governance?. Review of Financial Studies 22(2):
783-827.
Bebchuck L, Fried J. 2004. Pay Without Performance: The Unfulfilled Promise of Executive Compensation. Cambridge,
MA: Harvard University Press.
Beltratti A, Stulz RM. 2012. The credit crisis around the globe: Why did some banks perform better?. Journal
of Financial Economics 105(1): 1-17.
Bennett VM, Pierce L, Snyder JA, Toffel MW. 2013. Customer-driven misconduct: How competition
corrupts business practices. Management Science 59(8): 1725-1742.
Berle A, Means G. 1932. The Modern Corporation and Private Property. (New York: Harcourt, Brace & World).
Bertrand M, Mullainathan S. 2001. Are CEOs rewarded for luck? The ones without principals are. Quarterly
Journal of Economics 116(3): 901-932.
Bethel JE, Liebeskind JP, Opler T. 1998. Block share purchases and corporate performance. The Journal of
Finance 53(2): 605-634.
Bidwell M. 2010. Problems deciding: how the structure of make-or-buy decisions leads to transaction
misalignment. Organization Science 21(2): 362-379.
Bidwell, M. 2012. Politics and firm boundaries: How organizational structure, group interests, and resources
affect outsourcing. Organization Science 23(6): 1622-1642.
Boyd Brian K. 1995. "CEO duality and firm performance: A contingency model."Strategic Management
Journal 16.4: 301-312.
Bradach JL, Eccles RG. 1989. Price, Authority, and Trust: From Ideal Types to Plural Forms. Annual Review of
Sociology. 15: 97-118.
Castañer X, Kavadis N. 2013. Does good governance prevent bad strategy? A study of corporate governance,
financial diversification and value creation by French corporations. Strategic Management Journal 34(7):
863-876.
Castanias, Richard P, Helfat Constance E. 1991. "Managerial resources and rents." Journal of management 17.1:
155-171.
Coase Ronald H. 1937. "The nature of the firm." Economica 4.16: 386-405.
Cremers KJ, Nair VB. 2005. Governance mechanisms and equity prices. The Journal of Finance 60(6): 28592894.
Del Guercio D. 1996. The distorting effect of the prudent-man laws on institutional equity
investments. Journal of Financial Economics 40(1), 31-62.
Dell’Ariccia G, Igan D, Laeven L. 2012. Credit booms and lending standards: Evidence from the subprime
mortgage market. Journal of Money, Credit and Banking 44(2‐3): 367-384.
Demiroglu C, James C. 2012. How important is having skin in the game? Originator-sponsor affiliation and
losses on mortgage-backed securities. Review of Financial Studies 25(11): 3217-3258.
Demyanyk Y, Van Hemert O. 2011. Understanding the subprime mortgage crisis. Review of Financial Studies
24(6): 1848-1880.
Eccles RG, White HC. 1988. Price and authority in inter-profit center transactions. The American Journal of
Sociology. 94(s1):
Erkens DH, Hung M, Matos P. 2012. Corporate governance in the 2007–2008 financial crisis: Evidence from
financial institutions worldwide. Journal of Corporate Finance 18(2): 389-411.
29
Fahlenbrach R, Stulz RM. 2011. Bank CEO incentives and the credit crisis. Journal of Financial Economics 99(1):
11-26.
Falkenstein EG. 1996. Preferences for stock characteristics as revealed by mutual fund portfolio holdings. The
Journal of Finance 51(1): 111-135.
Fama EF. 1980. Agency problems and the theory of the firm. The Journal of Political Economy 88(2): 288-307.
Feldman ER, Montgomery CA. 2015. Are incentives without expertise sufficient? Evidence from Fortune
500 firms. Strategic Management Journal Forthcoming.
Foss N. 2003. Selective intervention and internal hybrids: Interpreting and Learning from the rise and decline
of the Oticon Spaghetti Organization. Organization Science 14: 331-349.
Galasso, A, Simcoe T. 2011. CEO overconfidence and innovation. Management Science 57(8): 1469-1484.
Gartenberg C. 2014. Do parents matter? Effects of lender affiliation through the mortgage boom and bust.
Management Science 60(11): 2776-2793.
Garvey GT, Milbourn T. 2006. Asymmetric benchmarking in compensation: Executives are rewarded for
good luck but not penalized for bad. Journal of Financial Economics 82(1): 197-225.
Gillan SL, Starks LT. 2007. The evolution of shareholder activism in the United States. Journal of Applied
Corporate Finance 19(1): 55-73.
Gompers P, Ishii J, Metrick A. 2003. Corporate governance and equity prices. Quarterly Journal of Economics
118(1): 107-155.
Gopalan R, Milbourn T, Song F. 2010. Strategic flexibility and the optimality of pay for sector performance.
Review of Financial Studies 23(5): 2060-2098.
Gorton G, Kahl M. 1999. Blockholder identity, equity ownership structures, and hostile takeovers. National
Bureau of Economic Research Working Paper No. w7123.
Grossman S, Hart O. 1986. The costs and benefits of ownership: a theory of vertical and lateral integration.
Journal of Political Economy 94: 691-719.
Hambrick DC, Jackson EM. 2000. Outside directors with a stake: The linchpin in improving governance.
California Management Review, 42(4): 108-127.
Hambrick Donald C, Werder Axel V, Zajac Edward J. 2008. "New directions in corporate governance
research." Organization Science 19.3: 381-385.
Hart O, Moore J. 1990. Property rights and the nature of the firm. Journal of Political Economy 98: 1119-1158.
Hart O, Tirole J. 1990. Vertical integration and market foreclosure. Brookings Papers on Economic Activity:
Microeconomics.
Hermalin BE, Weisbach MS. 2001. Boards of directors as an endogenously determined institution: A survey
of the economic literature. National Bureau of Economic Research Working Paper No. w8161.
Holmstrom B, Milgrom P. 1991. Multitask principal-agent analyses: Incentive contracts, asset ownership, and
job design. Journal of Law, Economics and Organization 7: 24-52.
Huckman R, Pisano G. 2006. The firm specificity of individual performance: Evidence from cardiac
surgery. Management Science 52(4): 473-488.
Jacobides, Michael G. 2005. "Industry change through vertical disintegration: How and why markets emerged
in mortgage banking." Academy of Management Journal 48.3: 465-498.
Jacobides, Michael G, Winter Sidney G. 2005. "The co‐evolution of capabilities and transaction costs:
explaining the institutional structure of production."Strategic Management Journal 26.5: 395-413.
Jacobides, Michael G, Winter Sidney G, Kassberger Stefan M. 2012. "The dynamics of wealth, profit, and
sustainable advantage." Strategic Management Journal 33.12: 1384-1410.
Jensen M, Ruback R. 1983. The market for corporate control: the scientific evidence. Journal of Financial
Economics 11(1): 5-50.
30
Jensen, Michael, Zajac Edward J. 2004. "Corporate elites and corporate strategy: How demographic
preferences and structural position shape the scope of the firm." Strategic Management Journal 25(1):
507-524.
Kogut B, Colomer J, Belinky M. 2014. Structural equality at the top of the corporation: Mandated quotas for
women directors. Strategic Management Journal 35(6): 891-902.
Lafontaine F, Shaw K. 2005. Targeting managerial control: evidence from franchising, RAND Journal of
Economics 36(1): 131-150.
Lafontaine F, Slade M. 2007. Vertical integration and firm boundaries: the evidence. Journal of Economic
Literature 45(3): 629-685.
Malmendier U, Tate G. 2005. CEO overconfidence and corporate investment. The Journal of Finance 60(6):
2661-2700.
Mayer C, Pence K, Sherlund SM. 2009. The rise in mortgage defaults. The Journal of Economic Perspectives 23(1):
27-50.
Mayer KJ, Nickerson JA. 2005. Antecedents and performance implications of contracting for knowledge
workers: evidence from information technology services. Organization Science 16(3): 225-243.
McDonald M, Khanna P, Westphal James D. 2008. Getting them to think outside the circle: Corporate
governance, CEOs' external advice networks, and firm performance. Academy of Management
Journal 51(3): 453-475.
Mian A, Sufi A. 2009. The consequences of mortgage credit expansion: Evidence from the U.S. mortgage
default crisis. Quarterly Journal of Economics 124(4): 1449-1496.
Milgrom P, Roberts J. 1990. Bargaining costs, influence costs, and the organization of economic activity. In
Alt JE, Shepsle KA (eds.), Perspectives on Positive Political Economy Cambridge University Press: 57-89.
Murphy KJ. 1999. Executive compensation. Handbook of Labor Economics 3: 2485-2563.
Nahapiet J, Ghoshal S. 1998. Social capital, intellectual capital, and the organizational advantage. Academy of
Management Review 23(2): 242-266.
Natividad G, Rawley E. 2012. Firm focus, routines, and performance. Unpublished working paper. Columbia
Business School.
Nickerson J, Silverman B. 2003. Why aren’t all truck drivers owner-operators? Asset ownership and the
employment relationship in interstate for-hire trucking. Journal of Economics and Management Strategy
12(1): 91-118.
Nickerson JA, Zenger TR. 2004. A knowledge-based theory of the firm—The problem-solving perspective.
Organization Science 15(6): 617-632.
Novak S, Stern S. 2009. Complementarity among vertical integration decisions: evidence from automobile
product development. Management Science 55(2): 311-332.
Osterloh M, Frey B. 2000. Motivation, knowledge transfer, and organizational forms. Organization Science 11(5)
538-550.
Pierce L. 2012. Organizational structure and the limits of knowledge sharing: Incentive conflict and agency in
car leasing. Management Science 58(6): 1106-1121.
Purnanandam A. 2011. Originate-to-distribute model and the subprime mortgage crisis. Review of Financial
Studies 24(6): 1881-1915.
Rawley E. 2010. Diversification, coordination costs, and organizational rigidity: evidence from
microdata. Strategic Management Journal 31(8): 873-891.
Roberts J. 2004. The Modern Firm: Organizational Design for Performance and Growth, Oxford University Press,
USA.
Schnatterly K, Shaw K, Jennings W. 2008. Information advantages of large institutional owners. Strategic
Management Journal 29(2): 219-227.
31
Shiller RJ. 2008. The Subprime Solution: How Today's Global Financial Crisis Happened, and What to Do About It.
Princeton University Press.
Shin HS. 2009. Reflections on Northern Rock: The bank run that heralded the global financial crisis. Journal of
Economic Perspectives 23(1): 101-120.
Shleifer A, Vishny R. 1989. Management entrenchment: The case of manager-specific investments. Journal of
Financial Economics 25(1): 123-139.
Shleifer A, Vishny R. 1997. A survey of corporate governance. Journal of Finance 52(2): 737-783.
Tadelis S. 2002. Complexity, flexibility and the make-or-buy decision. American Economic Review 92(2): 433-437.
Weisbach MS. 1988. Outside directors and CEO turnover. Journal of Financial Economics 20: 431-460.
Westphal JD. 1999. Collaboration in the boardroom: Behavioral and performance consequences of CEOboard social ties. Academy of Management Journal 42(1): 7-24.
Westphal, J, Fredrickson J. 2008. Who directs strategic change? Director experience, the selection of new
CEOs, and change in corporate strategy. Strategic Management Journal 22(12): 1113-1137.
Westphal J, Zajac E. 1995. Who shall govern? CEO/board power, demographic similarity, and new director
selection. Administrative Science Quarterly 40(1): 60-83.
Williamson O. 1985. The Economic Institutions of Capitalism, The Free Press.
Williamson O. 1988. Corporate finance and corporate governance. Journal of Finance 43: 567-91.
Williamson O. 1999. Strategy research: Governance and competence perspectives. Strategic Management Journal.
20(12): 1087-1108.
Wulf J. 2002. Internal capital markets and firm-level compensation incentives for division managers. Journal of
Labor Economics 20(2): 219-262.
Zajac Edward J, Westphal James D. 1994. "The costs and benefits of managerial incentives and monitoring in
large US corporations: When is more not better?" Strategic Management Journal 15(S1): 121-142.
32
Figure 1: Predicted Relationship Between Vertical Integration and Shared Information
Quality Under Different Corporate Governance Strengths
Figure 2: Corporate Governance and the Relationship Between Vertical Integration and
Default Likelihood
Figure 2A
33
-2
-1
Default Likelihood
0
1
2
Figure 2B
0
4
8
12
Ln(MBS+1)
High G
High G
Low G
Low G
Each observation represents one firm-year observation. The horizontal axis measures the log of mortgages issued by the firm in a
given year. The vertical axis represents a measure of loan performance, the firm-specific likelihood that a mortgage will default
conditional on the observable characteristics of the mortgage. 0 represents the market average, while observations below 0 represent
higher loan performance (lower default likelihood), and above 1 is lower loan performance (higher default likelihood).
0
Kernel density
.2
.4
.6
Figure 3: Corporate Governance and the Relationship Between Vertical Integration and
Default Likelihood
0
2
4
6
Default Likelihood (odds ratio)
Non-integrated, Low G
Integrated, Low G
8
Non-integrated, High G
Integrated, High G
Density distribution of default likelihood by governance and integration. “Low G” includes the firms with governance index at or
below the median level, where higher values represent worse governance. “High G” firms include firms that are above the median
level.
34
Table 1: Descriptive statistics
Panel A: Firm-level descriptive characteristics
Firm-year
obs
Default likelihood
Mean
Standard
deviation
Source
608
0.2532
1.0907
First stage estimate
% firm-yr obs issuing MBSB
Amount MBS issued
G index
608
217
203
0.3339
18,893
9.1626
26,826
2.5867
Anti-takeover index (ATI)
Entrenchment Index (EI)
203
203
2.2463
2.2266
0.7502
1.4065
Thomson SDC
Thomson SDC
IRCC via Andrew Metrick
Bebchuk, Cohen and Ferrell
(2009)
Cremers and Nair (2005)
% firm-yr obs from public firms
Age of firm (years)
608
576
0.5033
51.81
0.5004
55.74
Compustat
Capital IQ and public sources
Number annual loans in mortgage db
Diversification index
608
603
588
0.4690
1179
0.3007
County deeds
County deeds
Total assets (public firms only) ($000)
% Commercial bank
291
608
339,317
31.09
502,357
Compustat
Compustat and Capital IQ
% Mortgage lenders
Large financial institutions
608
608
48.36
20.56
% Operating in 2010 (firm-level obs)
162
46.30
Public sources
Panel B: Loan-level descriptive characteristics
316,848
Standard
deviation
197,816
Mortgage amount($)
105,780
243,172
142,473
County deeds
Combined LTV
105,780
0.8619
0.1370
County deeds
Origination FICO
105,780
715
50
Servicer database
89.595
24.0291
20.5534
Servicer database
New construction
105,780
0.5017
0.5000
County deeds
Adjustable
105,780
0.4729
0.4993
County deeds
Fixed rate
105,780
0.2492
0.4325
County deeds
Other interest type
105,780
0.2779
0.4480
County deeds
Initial interest rate
105,681
6.0648
1.6133
Servicer database
Prepayment indicator
Interest-only
97,742
65,333
0.3058
0.3077
0.4607
0.4616
Servicer database
Servicer database
Negative amortization
46,829
0.1188
0.3236
Servicer database
Sale amount ($)
Debt-to-income
Low or no doc
Number of
loans
105,780
Mean
Source
County deeds
77,199
0.7719
0.4196
Servicer database
Med census tract income (2000)
104,433
64,559
21,084
US Census Bureau
Peak-to-trough change in home
prices
Freddie rate
105,760
-0.2169
0.2370
County deeds (calculated)
105,780
6.1628
0.4493
Fed funds rate
Notice of default issued
105,780
105,780
3.4644
0.1556
1.6268
0.3625
Federal Reserve Bank
County deeds
.This table provides summary statistics of key variables within the sample data. Panel A provides firm-level summary statistics,
describing the panel data used in the main analysis in the paper, while Panel B provides mortgage-level summary statistics, describing
the mortgage database used to generate the underwriting hazard measures.
35
Table 2: Average Relationship Between Vertical Integration and Firm-Level Underwriting Risk
Time Range
Dependent Variable:
Log(MBS Total)
First stage controls
Second stage controls
Year FE
Adjusted R-squared
Error clusters:
Observations
(1)
2006-2007
(2)
2006-2007
(3)
2000-2007
(4)
2000-2007
(5)
2000-2007
(6)
2000-2007
Loan
Default
Default
Likelihood
Default
Likelihood
Default
Likelihood
Default
Likelihood
Default
Likelihood
-0.0123***
(0.0039)
DJ
-Included
0.229
MSA
41,932
-0.0086
(0.0108)
DJ
-Included
0.041
-230
-0.0303***
(0.0073)
DJ
-Included
0.165
-608
-0.0303***
(0.0071)
DJ
-Included
0.165
Lender
608
-0.0078
(0.0096)
Full
-Included
0.114
Lender
608
0.0091
(0.0131)
Full
Included
Included
0.166
Lender
607
Note: Column (1) analyzes loan default at the loan level, while Columns (2)-(6) analyze default likelihood at the firm-year level. Columns (1) and (2) use a similar
time frame to Demiroglu and James (2012), while the other columns use our longer period. Column (1) clusters standard errors at the county (FIPS) level, which
parallels the Demiroglu and James MSA approach. Columns (4)-(6) cluster at the lender level, which generally increases standard error size. For a list and significance
of the DJ and full controls use in the first stage to calculated Default Likelihood, refer to Appendix Table A2. Standard errors in parentheses, calculated by blockbootstrapping by lender. * significant at the 10% confidence level, ** significant at the 5% confidence level, *** significant at the 1% confidence level.
36
Table 3: How Governance Influences the Vertical Integration and Underwriting Risk Relationship
Dependent variable:
Default likelihood
Index:
Log(MBS Total)
High G Index
High GI X Log(MBS)
(1)
(2)
(3)
(4)
(5)
(6)
G
-0.0880***
(0.0242)
-0.2757
(0.2333)
0.1086***
(0.0304)
G
-0.0790***
(0.0282)
-0.4762*
(0.2614)
0.0984***
(0.0364)
EI
-0.1147***
(0.0276)
EI
-0.0887***
(0.0289)
ATI
-0.0905***
(0.0178)
ATI
-0.0855***
(0.0248)
-0.3662
(0.2544)
0.1265***
(0.0316)
-0.5094*
(0.2721)
0.1040***
(0.0353)
-0.1075
(0.2316)
0.1211***
(0.0294)
Full
-Included
0.260
Lenders
203
-0.3742
(0.2885)
0.1217***
(0.0367)
Full
Included
Included
0.319
Lenders
203
High E Index
High EI X Log(MBS)
High Antitakeover Index
High ATI X Log(MBS)
Year FE
Adjusted R-squared
Error Clusters
Observations
Full
-Included
0.186
Lenders
203
Full
Included
Included
0.294
Lenders
203
Full
-Included
0.209
Lenders
203
Full
Included
Included
0.300
Lenders
203
Note: High GI, High EI and High ATI defined as 0/1 indicators equal to 1 if the underlying governance index (G, Entrenchment
and Anti-Takeover Index, respectively) is greater than the mean value in the dataset. Standard errors in parentheses, calculated by
block-bootstrapping by lender. * significant at the 10% confidence level, ** significant at the 5% confidence level, *** significant
at the 1% confidence level
37
Table 4: Institutional Ownership Models
(1)
Dependent variable:
Default likelihood
Log(MBS Total)
High IO Ratio
High IO Ratio X Log(MBS)
(2)
(3)
IO Ratio
0.0203
0.0352
(0.0228)
(0.0236)
0.5137***
0.4261**
(0.1929)
(0.2139)
-0.0722**
-0.0673**
(0.0282)
(0.0313)
IO Number
0.0441**
0.0522**
(0.0193)
(0.0230)
0.0856
(0.1916)
-0.0831***
(0.0251)
High IO Number
High IO Number X Log(MBS)
(4)
High IO HHI X Log(MBS)
Full
-Included
284
Lender
0.136
Full
Included
Included
284
Lender
0.205
Full
-Included
284
Lender
0.165
(6)
IO HHI
-0.0326*
-0.0256
(0.0187)
(0.0200)
0.3666
(0.2375)
-0.0890***
(0.0291)
High IO HHI
Year FE
Adjusted R-squared
Error clusters
Observations
(5)
Full
Included
Included
284
Lender
0.217
0.0342
(0.2014)
0.0717***
(0.0255)
Full
-Included
284
Lender
0.168
-0.0903
(0.2117)
0.0640**
(0.0281)
Full
Included
Included
284
Lender
0.213
Note: IO Ratio refers to the ratio of shares owned by institutional owners to total shares. IO Number is the number of institutional
owners. And IO HHI measures the concentration of institutional ownership (as a Herfindahl measure). High IO Ratio, High
Number and High HHI defined as 0/1 indicators equal to 1 if the underlying institutional ownership measure is greater than the
mean value in the dataset. Standard errors in parentheses, calculated by block-bootstrapping by lender. * significant at the 10%
confidence level, ** significant at the 5% confidence level, *** significant at the 1% confidence level.
38
Table 5: Institutional Composition
(1)
Dependent variable:
Default likelihood
Log(MBS Total)
High Bank-Ins Ratio
High Bank-Ins Ratio X Log(MBS)
(2)
Bank and Insurance
0.0443**
0.0599***
(0.0175)
(0.0206)
0.3742*
0.5500**
(0.1935)
(0.2140)
-0.1267*** -0.1256***
(0.0256)
(0.0265)
High Invest Co
High Invest Co X Log(MBS)
Year FE
Adjusted R-squared
Error clusters
Observations
Full
-Included
0.211
Lenders
284
Full
Included
Included
0.258
Lenders
284
(3)
(4)
Investment Companies
-0.0660***
-0.0525***
(0.0192)
(0.0201)
-0.0980
(0.2044)
0.1023***
(0.0261)
Full
-Included
0.212
Lenders
284
-0.2073
(0.2396)
0.1020***
(0.0274)
Full
Included
Included
0.255
Lenders
284
Note: Bank and Insurance refers to the percent of institutional owners that are depository banks or insurance companies, traditionally
conservative, regulated owners. Investment Companies refers to the percent of institutional owners that are hedge funds, family offices
or other investment vehicles that are traditionally more aggressive owners. High Bank-Ins Ratio and High Invest Co are defined as 0/1
indicators equal to 1 if the underlying institutional composition measure is greater than the mean value in the dataset. Standard
errors in parentheses, calculated by block-bootstrapping by lender. * significant at the 10% confidence level, ** significant at the
5% confidence level, *** significant at the 1% confidence level
39
Table 6: Firm Failure, Governance and Vertical Integration
(1)
Dependent variable:
Firm in operation by end of 2010
Default Likelihood
(2)
All firms
-0.8609*** -0.8772***
(0.2248)
(0.2514)
Log(MBS Total)
High G Index
High G Index X Log(MBS)
Received TARP funds
First Stage Controls
Second Stage Controls
Pseudo R-squared
Observations
1.1021**
(0.5021)
Full
-0.124
154
2.0105***
(0.7634)
Full
Included
0.266
152
(3)
(4)
Firms with G data
0.0923
(0.0928)
2.0352
(1.5281)
-0.6325**
(0.2926)
2.9684**
(1.3317)
Full
-0.407
42
0.2439
(0.1833)
1.0900
(1.4954)
-0.6257**
(0.2616)
3.3038*
(1.8220)
Full
Included
0.504
42
Note: This analysis in this table uses a firm-level cross-sectional dataset constructed from the firm-year panel. The variables in this
cross-sectional data were calculated as the averages across the 2003-2007 years of the panel. Each model is a logit specification
with the dependent variable Firm in Operation By End of 2010. Robust standard errors in parentheses. * significant at the 10%
confidence level, ** significant at the 5% confidence level, *** significant at the 1% confidence level
40
Appendix A: Tables and Figures
Figure A1: Institutional Ownership and the Relationship Between Vertical Integration and Default
Likelihood
0
.1
Kernel density
.2
.3
.4
Figure 4a:Institutional Ownership
0
2
4
6
8
Non-integrated, Low inst investors
Non-integrated, High inst investors
Integrated, Low inst investors
Integrated, High inst investors
0
.1
Kernel density
.2
.3
.4
Figure 4b: Number of Institutional Owners
0
2
4
6
8
Non-integrated, Low num inst invest
Non-integrated, High num inst invest
Integrated, Low num inst invest
Integrated, High num inst invest
Figure 4c: Institutional Ownership Concentration (HHI)
.4
Kernel density
.2
.3
.1
0
0
2
4
6
8
Non-integrated, Low inst invest HHI
Non-integrated, High inst invest HHI
Integrated, Low inst invest HHI
Integrated, High inst invest HHI
Figure A2: Types of Owners and the Relationship Between Vertical Integration and Default
Likelihood
0
Kernel density
.2
.4
.6
Figure 5a: Bank and Insurance
0
2
4
6
8
Non-integrated, Low bank/ins ownership
Non-integrated, High bank/ins ownership
Integrated, Low bank/ins ownership
Non-integrated, High bank/ins ownership
0
Kernel density
.2
.4
.6
Figure 5b: Investment Professionals
0
2
6
4
8
Non-integrated, Below median
Non-integrated, High invest ownership
Integrated, Low invest ownership
Integrated, High invest ownership
Table A1: Correlations
1
2
3
4
5
1
Default likelihood
1
2
Log of MBS
-0.03
1
3
G index
0.21*
0.01
1
4
Anti-takeover
0.10
0.11
0.63*
1
0.23*
-0.17*
0.78*
0.59*
1
6
7
8
9
10
11
12
13
14
index
5
Entrenchment
index
6
Parent public
0.03
0.18*
0.09
.
.
1
7
Log firm age
-0.02
0.38*
-0.07
-0.04
-0.23*
0.56*
1
8
Log number loans
-0.16*
0.31*
-0.11
-0.05
-0.15*
0.32*
0.33*
1
9
Diversif. index
-0.08*
-0.17*
0.17*
0.11
0.29*
-0.45*
-0.39*
-0.45*
1
10
Log assets
-0.02
0.38*
-0.10
-0.17*
-0.40*
0.92*
0.67*
0.37*
-0.45*
1
11
Depository banks
-0.11*
-0.02
0.16*
-0.04
-0.13
0.51*
0.53*
0.30*
-0.21*
0.56*
1
12
Mortgage lender
-0.02
0.10
-0.15*
-0.24*
0.05
-0.79*
-0.56*
-0.32*
0.44*
-0.76*
-0.65*
1
13
Financial inst
0.15*
-0.10
-0.06
0.23*
0.11
0.39*
0.09*
0.05
-0.30*
0.30*
-0.34*
-0.49*
1
14
Operating in 2010
-0.35*
0.01
-0.20*
-0.08
-0.21*
-0.05
0.04
-0.10*
0.34*
-0.00
0.11*
0.05
-0.19*
Standard errors * significant at 5% level.
1
Table A2: Average Relationship Between Vertical Integration and Individual Loan Default
with Added Control Variables
Sample:
Time Range:
Dependent variable:
Log(MBS)
Low or no doc
Origination FICO
Combined LTV
Adjustable
Negative Amortization
Peak-to-Trough Chg Home Prices
Prepayment Indicator
Debt-to-Income
(1)
(2)
Individual Loans
2006-2007
Loan Default
-0.0123***
(0.0039)
0.6395***
(0.0653)
-0.0075***
(0.0004)
5.7315***
(0.6369)
0.1403**
(0.0697)
0.0700
(0.1090)
-3.8647***
(1.1112)
0.5056***
(0.0872)
Individual Loans
2000-2007
Loan Default
-0.0154***
(0.0035)
0.5900***
(0.0470)
-0.0078***
(0.0005)
5.1403***
(0.4695)
-0.0523
(0.0552)
0.3342***
(0.0898)
-3.1972***
(0.9584)
0.4483***
(0.0745)
Hybrid Mortgage
Balloon or Other Mortgage
Interest Only Mortgage
Initial Interest Rate
New construction flg
Log mortgage amt
Multi-payment Option ARM
Log 2000 Census Tract Median Income
Freddie Mac Interest Rate
Fed Funds Rate
Constant
Year FE
State Fixed Effects
Controls:
Error Clusters:
-2.0747***
(0.7418)
Included
-DJ
MSA
-1.4661**
(0.6405)
Included
-DJ
MSA
(3)
Individual
Loans
2000-2007
Loan Default
-0.0070***
(0.0025)
0.4733***
(0.0491)
-0.0074***
(0.0005)
4.5514***
(0.2983)
0.1991***
(0.0510)
1.1942***
(0.0922)
-1.0769
(0.8080)
0.2667***
(0.0454)
-0.0023***
(0.0008)
0.1025
(0.0802)
0.6496***
(0.0643)
0.2199***
(0.0599)
0.1699***
(0.0256)
-0.0648
(0.0477)
0.4987***
(0.0511)
-0.2913***
(0.0953)
-0.5327***
(0.1185)
-0.0837
(0.0587)
0.2379***
(0.0583)
-2.7178
(1.8811)
Included
Included
DJ+Ours
MSA
(4)
Individual
Loans
2000-2007
Loan Default
-0.0070
(0.0051)
0.4733***
(0.1192)
-0.0074***
(0.0004)
4.5514***
(0.1716)
0.1991**
(0.0929)
1.1942***
(0.1599)
-1.0769***
(0.1552)
0.2667***
(0.0797)
-0.0023
(0.0018)
0.1025
(0.1150)
0.6496***
(0.0944)
0.2199***
(0.0707)
0.1699***
(0.0332)
-0.0648***
(0.0216)
0.4987***
(0.0319)
-0.2913***
(0.0885)
-0.5327***
(0.0887)
-0.0837*
(0.0461)
0.2379***
(0.0522)
-2.7178**
(1.3190)
Included
Included
DJ+Ours
Lender
Observations
Pseudo R-squared
41,932
0.229
105,748
0.228
104,396
0.262
104,396
0.262
Note: Standard errors in parentheses. Column 1 uses a similar time frame to Demiroglu and James (2012), while the other columns
use our longer panel. Columns 1-3 cluster standard errors at the county level, which parallels the Demiroglu and James MSA approach.
Column 4 clusters at the lender level, which generally increases standard errors and reflects the approach we use for the rest of our
analysis. * significant at the 10% confidence level, ** significant at the 5% confidence level, *** significant at the 1% confidence level.
Table A3: Table 2 with second stage controls displayed
Time Range
Dependent Variable:
Log(MBS Total)
Parent is public
(1)
2006-2007
Default
Likelihood
-0.0086
(0.0108)
(2)
2000-2007
Default
Likelihood
-0.0303***
(0.0073)
(3)
2000-2007
Default
Likelihood
-0.0303***
(0.0071)
(4)
2000-2007
Default
Likelihood
-0.0078
(0.0096)
0.3882***
(0.0913)
DJ
1.2708***
(0.2150)
DJ
1.2708***
(0.2205)
DJ
1.1820***
(0.2288)
Full
Log Firm Age
Log Number Loans
Diversification Index
Log Firm Assets
Mortgage Lender
Financial Institution
Constant
Second stage
controls
Year FE
Adjusted R-squared
Error clusters:
Observations
(5)
2000-2007
Default
Likelihood
0.0091
(0.0131)
-0.0363
(0.2194)
-0.0562
(0.0663)
-0.1351***
(0.0408)
-0.3816**
(0.1804)
-0.0516
(0.0365)
0.1661
(0.1346)
0.3805***
(0.1253)
1.6870***
(0.4457)
Full
--
--
--
--
Included
Included
0.041
-230
Included
0.165
-608
Included
0.165
Lender
608
Included
0.114
Lender
608
Included
0.166
Lender
607
Table A4: Table 3 with second stage controls displayed
Index:
Dependent variable:
Log(MBS Total)
High G Index
High GI X Log(MBS)
High E Index
(1)
G
(2)
G
-0.0880***
(0.0242)
-0.2757
(0.2333)
0.1086***
(0.0304)
-0.0790***
(0.0282)
-0.4762*
(0.2614)
0.0984***
(0.0364)
High EI X Log(MBS)
(3)
EI
(4)
EI
Default likelihood
-0.1147***
-0.0887***
(0.0276)
(0.0289)
-0.3662
(0.2544)
0.1265***
(0.0316)
-0.5094*
(0.2721)
0.1040***
(0.0353)
1.2537***
(0.2988)
Full
-Included
0.209
Lenders
203
0.1657*
(0.0988)
0.0168
(0.0713)
0.3919
(0.3765)
-0.1748***
(0.0481)
0.1850
(0.2405)
0.7026***
(0.2134)
0.7067
(1.0813)
Full
Included
Included
0.300
Lenders
203
High ATI X Log(MBS)
Log Firm Age
Log Number Loans
Diversification Index
Log Firm Assets
Mortgage Lender
Constant
Year FE
Adjusted R-squared
Error Clusters
Observations
1.1676***
(0.2727)
Full
-Included
0.186
Lenders
203
0.1336
(0.0973)
0.0563
(0.0723)
0.4997
(0.3920)
-0.1840***
(0.0500)
0.4252
(0.2679)
0.7461***
(0.2140)
0.0833
(1.0618)
Full
Included
Included
0.294
Lenders
203
(5)
ATI
(6)
ATI
-0.0905***
(0.0178)
-0.0855***
(0.0248)
-0.1075
(0.2316)
0.1211***
(0.0294)
-0.3742
(0.2885)
0.1217***
(0.0367)
0.1612*
(0.0956)
0.0636
(0.0747)
0.4480
(0.3931)
-0.1801***
(0.0585)
0.5553**
(0.2736)
0.5142**
(0.2114)
0.8918
(1.0887)
Full
Included
Included
0.319
Lenders
203
1.1315***
(0.2400)
Full
-Included
0.260
Lenders
203
Table A5: Descriptive Statistics by Integration and Governance
Full Sample
Matched Sample
Non -
Integrated
Integrated
Integrated –
Integrated -
Integrated –
integrated
– All
-Low G
High G
Low G
High G
% firm-yr obs from public firms
0.3012
0.9064
1
1
1
1
Age of firm (years)
33.03
86.84
121.78
86.63
107.67
91.39
Number annual loans in mortgage db
264
1,236
2,079
776
1,023
958
0.5580
0.2912
0.3025
0.3238
0.3330
0.3232
Total assets (public firms only) ($000)
177,377
442,121
548,028
270,129
373,724
342,910
% Commercial bank
21.73
49.75
69.23
51.67
61.54
66.67
% Mortgage lenders
65.19
14.78
17.95
6.67
12.82
5.12
Large financial institutions
13.09
35.47
12.82
41.67
25.64
28.21
% Operating in 2010 (firm-level obs)
53.39
30.43
50.00
26.32
0.3846
0.1794
Table A6: Matched Analysis
Index:
Dependent variable:
Log(MBS Total)
High G Index
High GI X Log(MBS)
High E Index
(1)
G
(2)
G
-0.0361
(0.0321)
-0.2016
(0.2510)
0.0392
(0.0378)
-0.0403
(0.0310)
-0.3390
(0.2829)
0.0642
(0.0410)
High EI X Log(MBS)
(3)
EI
(4)
EI
Default likelihood
-0.0609
-0.0538
(0.0379)
(0.0354)
-0.2967
(0.2696)
0.0708*
(0.0426)
-0.4222
(0.3081)
0.0744*
(0.0423)
1.4403***
(0.3199)
Full
-Included
0.111
Lenders
143
0.2436**
(0.1112)
0.0575
(0.0952)
0.8678*
(0.4714)
-0.1970***
(0.0742)
0.4772
(0.4321)
0.9102***
(0.2166)
-0.0392
(1.3687)
Full
Included
Included
0.215
Lenders
143
High ATI X Log(MBS)
Log Firm Age
Log Number Loans
Log Firm Assets
Mortgage Lender
Constant
Year FE
Adjusted R-squared
Error Clusters
Observations
1.3661***
(0.3532)
Full
-Included
0.095
Lenders
143
0.2280**
(0.1062)
0.0623
(0.0938)
0.8767*
(0.4636)
-0.2086***
(0.0782)
0.5682
(0.4252)
0.9540***
(0.2289)
-0.2505
(1.4057)
Full
Included
Included
0.208
Lenders
143
(5)
ATI
(6)
ATI
-0.0598**
(0.0251)
-0.0470*
(0.0275)
0.0129
(0.2346)
0.0775**
(0.0347)
-0.1047
(0.2736)
0.0702*
(0.0377)
0.2327**
(0.1079)
0.0654
(0.0992)
0.8649*
(0.4687)
-0.1710**
(0.0872)
0.6967
(0.4239)
0.7812***
(0.2158)
0.1571
(1.3733)
Full
Included
Included
0.222
Lenders
143
1.3640***
(0.2882)
Full
-Included
0.159
Lenders
143
Table A7: Alternative measures of vertical integration – MBS Indicator
Index:
Dependent variable:
MBS Indicator (0/1)
High G Index
High GI X MBS Indicator
High E Index
(1)
G
(2)
G
-0.8319***
(0.2452)
-0.1825
(0.2434)
0.8612***
(0.3008)
-0.7167***
(0.2769)
-0.3399
(0.2697)
0.6850*
(0.3567)
High EI X MBS Indicator
(3)
EI
(4)
EI
Default likelihood
-1.0527*** -0.8262***
(0.2628)
(0.2778)
-0.2533
(0.2603)
1.0322***
(0.3060)
-0.3846
(0.2718)
0.7896**
(0.3463)
1.2739***
(0.3050)
Full
-Included
0.196
Lenders
203
0.1656*
(0.0983)
0.0345
(0.0700)
0.4099
(0.3842)
-0.1672***
(0.0468)
0.2171
(0.2265)
0.7398***
(0.2069)
0.6977
(1.0991)
Full
Included
Included
0.293
Lenders
203
High ATI X MBS Indicator
Log Firm Age
Log Number Loans
Log Firm Assets
Mortgage Lender
Constant
Year FE
Adjusted R-squared
Error Clusters
Observations
1.2091***
(0.2783)
Full
-Included
0.175
Lenders
203
0.1364
(0.0970)
0.0596
(0.0688)
0.5198
(0.4011)
-0.1749***
(0.0490)
0.3754
(0.2529)
0.7835***
(0.2080)
0.2892
(1.0710)
Full
Included
Included
0.286
Lenders
203
(5)
ATI
(6)
ATI
-0.7483***
(0.1747)
-0.6037***
(0.2175)
0.0707
(0.2371)
0.8489***
(0.2904)
-0.1010
(0.2873)
0.6280*
(0.3624)
0.1176
(0.0955)
0.0578
(0.0708)
0.4990
(0.3989)
-0.1549***
(0.0543)
0.4352*
(0.2574)
0.6334***
(0.2119)
0.5532
(1.0942)
Full
Included
Included
0.291
Lenders
203
1.1074***
(0.2475)
Full
-Included
0.231
Lenders
203
Table A8: Alternative measures of vertical integration – normalized by loans issued
Index:
Dependent variable:
Log MBS / Num Loans Issued
High G Index
High GI X (Log MBS / Num Loans Issued)
High E Index
(1)
G
(2)
G
-0.4057***
(0.1562)
-0.1091
(0.2332)
0.5614***
(0.1948)
-0.4179**
(0.1625)
-0.4399*
(0.2599)
0.5554**
(0.2226)
High EI X (Log MBS / Num Loans Issued)
(3)
EI
(4)
EI
Default likelihood
-0.4949*** -0.4873***
(0.1650)
(0.1660)
-0.1744
(0.2508)
0.6363***
(0.1952)
-0.5114*
(0.2763)
0.6351***
(0.2118)
1.0242***
(0.3064)
Full
-Included
0.185
Lenders
203
0.1767*
(0.0978)
-0.0330
(0.0690)
0.3781
(0.3840)
-0.1753***
(0.0469)
0.2835
(0.2327)
0.6666***
(0.2169)
0.8799
(1.0755)
Full
Included
Included
0.304
Lenders
203
High ATI X (Log MBS / Num Loans Issued)
Log Firm Age
Log Number Loans
Log Firm Assets
Mortgage Lender
Constant
Year FE
Adjusted R-squared
Error Clusters
Observations
0.9916***
(0.2872)
Full
-Included
0.170
Lenders
203
0.1413
(0.0959)
-0.0078
(0.0667)
0.5017
(0.4016)
-0.1775***
(0.0490)
0.4173*
(0.2499)
0.7165***
(0.2181)
0.4596
(1.0723)
Full
Included
Included
0.292
Lenders
203
(5)
ATI
(6)
ATI
-0.4564***
(0.1182)
-0.4415***
(0.1393)
-0.0091
(0.2277)
0.6660***
(0.1825)
-0.3400
(0.2950)
0.6934***
(0.2382)
0.1570
(0.0968)
-0.0072
(0.0673)
0.4479
(0.4117)
-0.1742***
(0.0581)
0.5368**
(0.2589)
0.4874**
(0.2196)
0.9220
(1.0749)
Full
Included
Included
0.315
Lenders
203
0.9975***
(0.2530)
Full
-Included
0.241
Lenders
203
Table A9: Alternative measures of vertical integration – Normalized by firm assets
Index:
Dependent variable:
Log MBS/Assets
High G Index
High GI X (Log MBS/Assets)
(1)
G
-0.6867**
(0.3098)
-0.0486
(0.2559)
1.0057**
(0.3923)
(2)
G
-0.7535**
(0.3386)
-0.3235
(0.2672)
0.8704**
(0.4229)
High E Index
(3)
EI
Default likelihood
-1.3069*** -1.0686***
(0.3558)
(0.3670)
-0.3114
(0.2652)
1.5469***
(0.4052)
High EI X( Log MBS/Assets)
(4)
EI
High ATI X (Log MBS/Assets)
Log Number Loans
Log Firm Assets
Mortgage Lender
Constant
Year FE
Adjusted R-squared
Error Clusters
Observations
0.9271***
(0.2876)
Full
-Included
0.158
Lenders
203
0.1169
(0.1001)
0.0531
(0.0744)
0.4804
(0.4188)
-0.1868***
(0.0547)
0.3886
(0.2573)
0.7683***
(0.1968)
2.9277***
(1.0882)
Full
Included
Included
0.281
Lenders
203
1.1476***
(0.2968)
Full
-Included
0.193
Lenders
203
(6)
ATI
-0.8576***
(0.2320)
-0.8575***
(0.2739)
0.0139
(0.2296)
1.2496***
(0.3495)
-0.2164
(0.2465)
1.1256***
(0.3862)
0.1394
(0.0959)
0.0569
(0.0765)
0.4721
(0.4052)
-0.1715***
(0.0561)
0.5654**
(0.2657)
0.5666***
(0.2157)
2.9638***
(1.1043)
Full
Included
Included
0.299
Lenders
203
-0.4685*
(0.2708)
1.2280***
(0.4468)
Log Firm Age
(5)
ATI
0.1657
(0.1029)
0.0136
(0.0755)
0.3858
(0.4004)
-0.1794***
(0.0533)
0.1676
(0.2401)
0.6924***
(0.2022)
2.9839***
(1.0929)
Full
Included
Included
0.296
Lenders
203
1.0045***
(0.2515)
Full
-Included
0.226
Lenders
203
Table A10: Governance Variable Correlations
1
2
3
4
5
6
7
8
9
10
11
12
1
Board Independent
1
2
Board Size
-0.08
1
3
CEO Share Own
-0.28*
-0.27*
1
4
CFO Share Own
-0.08
-0.25*
0.31*
1
5
Log CEO Price Sen
0.05
-0.00
.013
-0.02
1
6
Log CFO Price Sen
0.35*
0.10
-0.27*
0.11
0.64*
1
7
Log CEO Vol Sen
0.30*
0.04
-0.32*
-0.18*
0.63*
0.62*
1
8
Log CFO Vol Sen
0.38*
0.22*
-0.46*
-0.14
0.59*
0.82*
0.81*
1
9
IO Ratio
0.20*
-0.33*
0.01
0.05
0.25*
0.10
0.20*
0.27*
1
10
IO Number
0.06
0.38*
-0.21*
-0.33
0.43*
0.48*
0.62*
0.66*
0.26*
1
11
IO HHI
-0.24*
0.10
-0.08
0.18*
-0.21*
-0.21*
-0.46*
-0.37*
-0.59*
-0.39*
1
12
Bank and Insurance
0.27*
0.29*
-0.09
-0.18*
-0.02
-0.02
0.25*
0.34*
-0.20*
0.27*
0.03
1
13
Investment Owners
-0.19*
-0.27*
0.08
0.12
0.05
0.05
-0.20*
-0.35*
0.16*
-0.31*
0.04
-0.94*
13
1
Table A11: CEO Compensation
(1)
Dependent variable:
Default likelihood
Log(MBS Total)
High CEO Ownership
High CEO X Log(MBS)
(2)
Share ownership
0.0514**
0.0774***
(0.0251)
(0.0283)
0.0758
0.4287*
(0.2076)
(0.2571)
-0.0902*** -0.1113***
(0.0296)
(0.0344)
(3)
Price sensitivity
-0.0408*
-0.0423*
(0.0243)
(0.0228)
-0.0904
(0.2533)
0.0005
(0.0340)
High CEO Price Sensitivity
High CEO X Log(MBS)
(4)
High CEO X Log(MBS)
Full
-Included
0.179
Lender
287
Full
Included
Included
0.252
Lender
287
Full
-Included
0.141
Lender
184
(6)
Volatility Sensitivity
0.0119
-0.0237
(0.0270)
(0.0262)
-0.1535
(0.2601)
0.0390
(0.0334)
High CEO Vol. Sensitivity
Year FE
Adjusted R-squared
Error clusters
Observations
(5)
Full
Included
Included
0.293
Lender
184
0.4493
(0.2843)
-0.0982**
(0.0392)
Full
-Included
0.190
Lender
175
0.6815**
(0.3092)
-0.0528
(0.0378)
Full
Included
Included
0.331
Lender
175
Table A12: CFO Compensation
(1)
Dependent variable:
Default likelihood
Log(MBS Total)
High CFO Ownership
High CFO X Log(MBS)
(2)
Share ownership
-0.0840*
-0.0535
(0.0453)
(0.0438)
-0.1611
-0.2639
(0.3712)
(0.3904)
0.0544
0.0472
(0.0515)
(0.0499)
(3)
Price sensitivity
0.0097
0.0156
(0.0278)
(0.0308)
0.2008
(0.2559)
-0.0770**
(0.0369)
High CFO Price Sensitivity
High CFO X Log(MBS)
(4)
High CFO X Log(MBS)
Full
-Included
150
Lender
0.123
Full
Included
Included
150
Lender
0.280
Full
-Included
131
Lender
0.117
(6)
Volatility Sensitivity
0.0468
0.0486
(0.0308)
(0.0362)
0.3078
(0.2638)
-0.0649
(0.0404)
High CFO Vol. Sensitivity
Year FE
Adjusted R-squared
Error clusters
Observations
(5)
Full
Included
Included
131
Lender
0.309
0.0735
(0.2921)
-0.1093***
(0.0383)
Full
-Included
122
Lender
0.196
0.0399
(0.2957)
-0.0985**
(0.0417)
Full
Included
Included
122
Lender
0.351
Table A13: Board Composition
(1)
Dependent variable:
Default likelihood
Log(MBS Total)
High Board Independence
High BI X Log(MBS)
(2)
Board Independence
-0.0842*** -0.0582**
(0.0223)
(0.0238)
-0.2671
-0.2155
(0.2189)
(0.2440)
0.0814***
0.0579*
(0.0309)
(0.0311)
High Board Size
High BS X Log(MBS)
Year FE
Adjusted R-squared
Error clusters
Observations
Full
-Included
0.158
Lender
206
Full
Included
Included
0.295
Lender
206
(3)
(4)
Board Size
-0.0002
0.0023
(0.0209)
(0.0255)
-0.2428
(0.2109)
-0.0675**
(0.0289)
Full
-Included
0.240
Lender
206
0.1708
(0.2514)
-0.0500
(0.0315)
Full
Included
Included
0.290
Lender
206
Table A14: List of Lenders with Governance Data
ABN-AMRO Mortgage (sold to Citigroup)
American International Group
American Mortgage Network
American Priority Mortgage
Astoria Financial Corp
BancMortgage
Banco Popular
Bank of America
Bank of North Georgia
Bank One
BNC Mortgage
CIT Group
CitiMortgage
Colonial Bank
Commonwealth United
Countrywide Financial Corp
Downey Financial Corp
E-loan Inc
Equifirst Corp
E-trade Mortgage Corp
Finam LLC
First Franklin Financial Corp
First Horizon National Corp
Flagstar Bancorp
Golden West Financial Corp
Greenpoint Mortgage
Home123 Corp (New Century Financial)
Impac Mortgage
IndyMac Bancorp
Irwin Financial Corp
JP Mortgage Chase
Lehman Brothers Bank
Long Beach Mortgage (Washington Mutual)
M&T Bank
National City Corp
New Century Financial Corp
Option One Mortgage (H&R Block)
Ownit Mortgage
Pinnacle Financial
Popular Inc
Principal Residential Mortgage
Regions Financial Corp
Residential Community Mortgage (Wells Fargo)
Southtrust Mortgage
Sterling Bankshares
Sun America Mortgage Corp
Suntrust Banks Inc
U S Bancorp
Wachovia Corp
Washington Mutual Inc
Webster Financial Corp
Wells Fargo Bank
WMC Mortgage
Appendix B: Estimating firm lending quality
From the initial sample of county deeds, a total sample of 105,780 “Alt-A” purchase mortgages
from the 100 zip codes issued between 2000 and 2007 were identified for use in calculating default
likelihood, our firm-level measure of lending quality. We restricted the sample to Alt-A mortgages
both to correspond to prior research and also because lenders had more underwriting discretion for
Alt-A mortgages than for the so-called “conforming” mortgages, which were guaranteed by one of
the two large government housing agencies and therefore subject to stricter lending guidelines.
To calculate lending quality, we estimate the likelihood that a loan underwritten by a bank
defaults, given the observable attributes of that loan. This approach is similar to risk-adjusted
performance measures used in the medical and health economics literature (e.g., Huckman and Pisano
2006), where hospital or surgeon performance is calculated by estimating mortality or morbidity after
controlling for observable patient characteristics known to increase risk. This approach has also been
used to estimate misconduct, where incentivized agents have private information that influences
outcomes but is unobservable to principals or other parties (Pierce and Snyder 2008; Bennett et al.
2013).
We calculate default likelihood by extracting the coefficients on firm indicators in an
unconditional logistic regression model performed on the individual mortgage data that controls for
publicly-observable risk through a standard set of loan characteristics. This model estimates
parameters associated with the likelihood of default of mortgages issued by firms in this study. The
first stage logit specification is as follows, with the coefficients on the firm fixed effects used in the
second stage analysis (Equation 2) represented by 𝛿𝛿𝑖𝑖𝑖𝑖 .
′
𝑃𝑃𝑃𝑃(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑)𝑗𝑗 = Λ(𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑡𝑡 ′𝑗𝑗 𝛽𝛽1𝑡𝑡 + 𝑟𝑟𝑟𝑟𝑟𝑟𝑘𝑘𝑗𝑗′ 𝛽𝛽2𝑡𝑡 + 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑜𝑜𝑗𝑗′ 𝛽𝛽3𝑡𝑡 + 𝑓𝑓𝑓𝑓𝑓𝑓𝑚𝑚𝑖𝑖𝑖𝑖
𝛿𝛿𝑖𝑖𝑖𝑖 + 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑒𝑒𝑗𝑗′ + 𝜖𝜖𝑗𝑗 )
(1)
𝑒𝑒 (𝑎𝑎)
Where Pr(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑)𝑗𝑗 is the default probability for loan j, Λ(a) = 1+𝑒𝑒 𝑎𝑎 is the logit function.
𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 is a vector of mortgage characteristics, including interest rate structure (fixed, adjustable or
hybrid) and indicators of whether various options are attached to the mortgage (interest only period,
negative amortization, pre-payment penalties). 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 is a vector of observable risk metrics on the
borrower (credit score, debt-to-income ratio, low- or no-documentation submitted, loan-to-value
ratio, initial interest rate on the loan). 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 is the vector of i firm indicators in year t, 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 is a
vector of macroeconomic controls (geography-specific home price index, census tract median income,
Freddie published mortgage rates, Fed Funds rate). Finally, 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 represents state fixed effects. We
run this model once per year to construct firm-year estimates of 𝛿𝛿, the default likelihood.
Appendix C: Construction of sample
Because of the amount of data stored in county public records, a comprehensive national dataset was
unrealistic to analyze. Furthermore, the need to link multiple public records to account for
downstream sales and refinancings and multiple liens on a single property made a nationwide random
sampling approach infeasible. To address these challenges jointly, we obtained the full set of county
records for the top 100 zip codes as ranked by new home construction from 1999 to 2009. This sample
provides geographic diversity (although limited to high-growth regions) while providing enough
records to estimate default likelihood.
Securitization data: For our vertical integration measure, we aggregated the principal amount of
all mortgage-backed securities issued in a given year, obtained by Thomson SDC, to the parent
company level. The aggregation included not only securities issued by the parent companies but also
those issued by affiliated trusts. For example, securitizations issued by CWABS 2007-8 and Long
Beach NIM Trust 2001-3, two entities named in the SDC data, were included in the total principal
amounts of Countrywide Financial and Washington Mutual, respectively. The aggregated data were
then matched to the county deed data manually using the names of the parent companies.
Between 2000 and 2007, 33% of firm-year observations included some level of securitization
activity, accounting for 69% of the loans issued in our mortgage database. The annual amounts of
MBS varied from a low of $12 million issued by Principal Residential Mortgage in 2002 to a high of
$186,951 million issued by Countrywide Financial in 2005 and a mean of $20,196 million across all
firm-years. The greatest activity occurred in 2006, with $943,691 million in private-label residential
mortgage-related securitization issuances by the firms in our dataset. 1 We define the degree of vertical
integration by the amount of securitizations issued by a firm in a given year, controlling for total firm
assets. We also obtain similar results when we use alternate definitions, including i) a dummy variable
1
According to Thompson SDC. Numbers may vary, depending on how principal is collected.
whether the firm issued MBS and ii) MBS normalized alternately by the total loans issued in our dataset
and iii) by firm assets (reported in the Appendix).
G-index and governance variables: For our primary measure of governance quality, we use the Gindex, as constructed by Gompers, Ishii and Metrick (2003). This measure is a summary of 24
governance rules compiled by the Investor Responsibility Research Center (IRRC) that determine the
degree to which shareholders are able to monitor and discipline managers. The rules measure antitakeover provisions; however, research using these data has found that the G measure (and antitakeover provisions in general), predict agency behavior on the part of CEOs (Gompers et al., 2003;
Shleifer and Vishny, 1989; Jensen and Ruback, 1983). Firms with high G values, dubbed
“Dictatorships,” are firms in which managers are able to operate with considerable discretion without
board monitoring or power to punish manager actions. On the other end of the spectrum, firms with
low G-values are dubbed “Democracies” and have managers who are considered to be more subjected
to a system of checks and balances enforced by shareholders. These data are only available for USbased publicly-traded companies. As such, the G value was available for 33% of the firm-year
observations in the sample that accounted for 66% of the mortgages issued in the sample.
In this study, in order to calculate meaningful interactions between the degree of managerial
monitoring and the level of vertical integration, we created an indicator variable equal to one if the Gindex of the firm is equal to or above the median value of 9 (on a scale of 1 to 16). The median value
of G was calculated on an unweighted basis, so firms with at- or above-median values of G (“high G
firms”) accounted for 29% of the loans of the firms with available values of G and 21% of all the
loans in the database. Firms with both high G values and securitization activity account for 10% of all
firm-year observations or 30% of firm-year observations for which the G value is available. Weighted
by the number of loans, these firms account for 10% of all loans issued in the database and 15% of
all loans issued by firms with G values.
For robustness, we collected two additional governance indices: the Entrenchment Index (EI)
(Bebchuk et al., 2009) and the Anti-Takeover Index (ATI) (Cremers and Nair, 2005). Both indices
were constructed from subsets of the 24 provisions provided by the IRRC, with the entrenchment
index (scaled from 0-6) based on six provisions deemed the least noisy measures of management
entrenchment, 2 and the anti-takeover index (scored 0-3) constructed from three provisions deemed
the most indicative of whether a firm is protected from takeovers. 3 We reverse-code the latter index,
since it was originally constructed so that higher scores indicate better governance, the opposite
convention of the G and EI indices.
Firm-level control variables: The control variables include both general measures for all firms and
financial information from Compustat for public firms. Hand-collected controls include the age of
the firm and indicators for industry of the parent firm (depository banks, mortgage lenders,
diversified financial institutions or homebuilders). Controls derived from the mortgage data include
the log of the total number of loans issued in that year by the parent firm and a measure of
geographic diversification of the lender. This measure was a calculation of geographic inequality
using a Gini-coefficient algorithm (where a measure close to 0 indicates a well-diversified lender and
a measure close to 1 indicates a lender with high levels of geographic concentration) and controls for
loan portfolio effects within lenders: lenders that are more diversified may be willing to take on
more loan-level risk if they have ex ante expectations that the housing conditions will not be
correlated across geographies. The financial controls for public firms from Compustat include the
log of firm assets, and the standard deviation of returns on assets for the past five years.
The six provisions are: staggered (classified) boards, poison pills, golden parachutes, supermajority required for
mergers, limit on charter amendments and limits on shareholder bylaw amendments.
3 The three provisions are: Staggered (classified) boards, preferred blank check, restrictions on calling special meetings or
on action through written consent.
2
Appendix D: Executive compensation and board monitoring
We find that, in firms with high levels of CEO shareholdings, vertically-integrated banks
appear to have lower risk, while firms with high levels of CFO shareholdings appear riskier. We see
no consistent moderating effect of CEO price or volatility sensitivities (as defined by Core and Guay
2002). We do not find any effects for CFO shareholdings or price sensitivity but do find a negative
moderating effect of CFO volatility sensitivity.
Overall, these mixed findings reflect the mixed findings in prior research. Executive
compensation in banks has been frequently blamed for the financial crisis, with critics arguing that
CEO’s had strong incentives to focus on short-term gains and high risk (Fahlenbrach and Stulz 2011).
Yet existing work has found little evidence that executive compensation played a major role in
explaining the financial crisis. Although executive share ownership is argued to reduce CEO agency
costs by aligning incentives with those of the shareholders (Murphy 1999), Fahlenbrach and Stulz
(2011) found no evidence that higher executive ownership in banks improved stock returns during
the financial crisis, and some evidence suggests that it reduced returns (Balachandran et al. 2010).
We also examine the moderating effect of board composition and size (Appendix Table A13) and
find no consistent patterns. Again, these results reflect the mixed theoretical predictions and
empirical results from other studies (Hermalin and Weisbach, 2001). For example, despite theory
arguing that board independence and small size improves oversight, Erkens and colleagues (2012)
linked independent boards at financial institutions to worse stock returns during the financial crisis
and found no link to board size, a result that may reflect the inherent conflict between reduced
expertise and improved incentives in independent boards (Feldman and Montgomery, 2015).
Appendix E: Other empirical challenges
It is also difficult in our results to distinguish between rent-seeking behavior by CEOs and
behavioral explanations such as overconfidence (Malmendier and Tate 2005; Malmendier et al. 2011)
or simple myopia. In this latter view, CEOs are behaviorally inclined to focus on short-term firm
performance and to ignore signals of the worsening housing market. Strong governance, then, forces
CEOs to take a longer view. We believe that that both myopic and agency behavior were likely driving
CEO choices in our setting. Myopia is consistent not only with our findings but also with prior
research that finds that CEOs lost substantial wealth during the housing downturn (Fahlenbrach and
Stulz 2011).
Several other empirical challenges are important to note. The first is our relative small sample
size. While the sample includes 88% of Alt-A mortgages across a geographic area that is broadly
representative of national home prices and default rates, ultimately it includes 170 firms, of which 53
have governance data and can be used in the primary analysis that estimates the interaction between
governance and vertical integration. Two concerns arise from the small sample. First, our small sample
provides limited statistical power for our analyses. Since we find significant results when testing our
main hypothesis, we believe that this concern is mostly alleviated in our setting. We do, however,
remain cautious in drawing inferences from our null findings, which may be null because there is
actually no effect or insufficient power, and also acknowledge the possibility of false positives in our
limited sample. The second concern is the external validity of the results. Given the nature of the
analysis and the data available, we are constrained to publicly-traded firms based in the United States
(the only lenders with standardized governance data) that had sufficient annual mortgage volume for
us to estimate lending quality. Our sample comprises most of the largest financial institutions and
mortgage lenders in the United States and covers 66% of the Alt-A loans in our database. Appendix
Table A14 includes a list of the 53 lenders in our sample with governance data and Appendix Table
A8 provides descriptive statistics by integration and governance subsamples. Conservatively, then, our
results apply to large public firms and are less conclusive about foreign and smaller private lenders.
Lastly, a question may also arise about our definition of vertical integration. We define
integration as the level of securitization by a firm that also originates mortgages, controlling for the
overall assets of the firm. We argue that this is the most appropriate definition in this context, since
it reasonably captures the ex-ante probability that a loan will be securitized internally at the time that
it is originated. 4 Another approach would be to define integration at the individual loan level and,
specifically, as whether each loan was securitized by the firm or sold. However, given that this
information is unknown at the time of underwriting, we do not believe that this definition is as
appropriate for our research question as the definition used in the study.
Firms still sold bulk loans to outside firms, but the proportions tended to be low (except for loans to government
sponsored agencies). See Levin (Washington Mutual), NCEN (New Century), and Countrywide Financial financial
statements for examples.
4

Subprime Governance: Agency Costs in Vertically Integrated Banks

Transcription

Similar documents

WHy CHOOsE FIrsT NaTIONaL BaNk MOrTgagE sErVICEs? Home Buying Counseling.

4. COMMUNITY DEVELOPMENT MARK PATCHETT MPCS

How to get control of IT Projects and Systems?

(IT) will play a critical role in supporting all areas of the Univ

Governance challenges in the world of infrastructure Part 2

PRBI at a Glance - PR Boutiques International

Why Partner with Northpoint

UNITED NATIONS NATIONS UNIES TJO Grade Level

Sample Hardship Letter

The Money Advice Service: Mortgage Paperwork Checklist