Friends With Money

Journal of Financial Economics 103 (2012) 169–188
Contents lists available at SciVerse ScienceDirect
Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec
Friends with money$

Joseph Engelberg a,1, Pengjie Gao b,2, Christopher A. Parsons a,n
a
Rady School of Management, University of California at San Diego, La Jolla, CA, USA
b
Mendoza College of Business, University of Notre Dame, USA
a r t i c l e i n f o abstract
Article history: When banks and firms are connected through interpersonal linkages – such as their
Received 20 October 2010 respective management having attended college or previously worked together –
Received in revised form interest rates are markedly reduced, comparable with single shifts in credit ratings.
13 January 2011
These rate concessions do not appear to reflect sweetheart deals. Subsequent firm
Accepted 30 January 2011
performance, such as future credit ratings or stock returns, improves following a
Available online 23 August 2011
connected deal, suggesting that social networks lead to either better information flow
JEL classification: or better monitoring.
G21 & 2011 Published by Elsevier B.V.
G32
Keywords:
Asymmetric information
Bank lending
Cost of debt
Social connections
Lending outcomes
1. Introduction violations of the Modigliani and Miller (1958) perfect

capital market assumptions. Because reliance on external
Stein (2003) characterizes information and agency finance depends to a large extent on these frictions,
problems as the ‘‘most pervasive and important’’ technologies that ameliorate their effects have important
implications for a firm’s financing cost, capital structure
and investment policy. In this paper, we study whether
$
We thank Aydoğan Altı, Robert Barro, Dan Hamermesh, Jay Hartzell, personal relationships between the respective employees
Naveen Khanna, Tim Loughran, Todd Milbourn (discussant), Carola of borrowers and lenders represent such a mechanism.
Schenone, Paul Schultz, Clifford Smith, Jeremy Stein, Sheridan Titman, The expected effect of personal relationships in credit
and seminar participants at Michigan State University, Southern Metho-
dist University, University of Houston, University of Rochester, Univer-
markets is not obvious. On the one hand, a lender personally
sity of Virginia, University of Washington, and the 2010 Texas Finance beholden to a borrower could overlook its flaws, thereby
Festival. We wish to thank Jacqueline Higgins and Shoshana Zysberg at putting his or her own shareholders’ capital at undue risk.
Management Diagnostic Limited for assistance with the BoardEx data- On the other hand, such relationships could catalyze infor-
base and Jing Zhang at Moody’s-KMV for assistance with the Expected
mation flow or reduce monitoring costs, placing the con-
Default Frequencies (EDFTM) and EDF Implied Spreads (EIS) database.
Xian Cai and Mei Zhao provided superb research assistance. We also nected bank at an advantage relative to competing lenders.
thank William Schwert (the editor). Here, both parties stand to benefit: Banks make better
n
Corresponding author. Tel.: þ1 919 265 9246. lending decisions and, assuming the associated savings are
E-mail addresses: joseph_engelberg@kenan- shared, firms reduce their costs of capital.
flagler.unc.edu (J. Engelberg), [email protected] (P. Gao),
chris_parsons@kenan-flagler.unc.edu (C.A. Parsons).
The goals of this paper are twofold. First, we aim to
1
Tel.: þ1 919 962 6889. establish a causal link between borrower-lender personal
2
Tel.: þ1 574 631 8048. relationships and lending market outcomes. Second, we
0304-405X/$ - see front matter & 2011 Published by Elsevier B.V.

doi:10.1016/j.jfineco.2011.08.003
170 J. Engelberg et al. / Journal of Financial Economics 103 (2012) 169–188
explore whether such relationships lead banks to make Although reverse causality is eliminated by how we
choices that harm their own shareholders or whether form the connection variables, one might suspect that
they improve their capital allocation decisions. firm-bank personal connections could be correlated with
To address these questions, we assemble a data set of unobserved firm or lender attributes and that these
roughly 20 thousand commercial loans made to US attributes drive the results we observe. For example, firms
companies between 2000 and 2007. The set of borrowers with large management teams, on average, share more
involves more than five thousand public firms; the set of personal ties with any lender, not just the ones with
lenders, more than 19 hundred commercial banks (the whom they conduct business. Likewise, bigger or more
majority of which are private). Next, from BoardEx we active banks also (mechanically) have more average con-
obtain a list of common organizations in which each of nections and could also have cost advantages relative to
the 65 thousand unique directors and executives in our smaller lenders, allowing interest rates to be affected.
universe of firms and banks could have fostered personal Because many firms borrow from multiple syndicates
relationships. This list tells us, for instance, if the over our eight-year sample period, and because the most
president of Wachovia Bank and the chief executive active banks lend to multiple firms, we can estimate the
officer (CEO) of Pepsi Co. attended college together, or if model with fixed effects for both borrowers and active
they overlapped in their first job after graduate school. lenders. This specification is important because it applies
The main question: Do personal relationships such as to the specific endogeneity concerns discussed above as
these influence lending terms? well as to any argument that relies on systematic differ-
Establishing a causal relation requires a careful ences between connected versus unconnected firms, or
account of the endogeneity of personal relationships. between connected versus unconnected banks. Because
A serious concern is reverse causality, whereby lending we are explicitly controlling for both time-invariant bank
interactions lead to the formation of social relationships. and firm heterogeneity through fixed effects for each
To illustrate, suppose a banker provides financing to a group, and for time-varying risk through observable
firm at below market rates and is subsequently invited to control variables (e.g., credit ratings), identification comes
join the board of the CEO’s favorite charity, or perhaps from the differences-in-differences implied by the con-
even the board of the borrowing firm itself. Such an struction of a connection variable that is formed at the
example is typical of several that could generate correla- firm-bank level. It thus follows that any alternative
tion between lending terms and firm-bank personal interpretation must either appeal to dynamic perfor-
relationships, but not for causal reasons. mance changes within firms or to an omitted variable
Perhaps the most significant advantage of our data is that also operates at the firm-bank level.
that they allow us to infer relationships whose formation One such possibility is geographical proximity. If banks
predates, by several years or decades, the lending transac- close to their borrowers have information or monitoring
tions we analyze. If Pepsi borrows from a Wachovia-led advantages, and if personal connections are a function of
syndicate in 2004, we take as exogenous that their respec- distance, then the results we find could simply reflect the
tive top executives could have both received a masters of effects of local information networks (e.g., Coval and
business administration from Stanford University in 1974 or Moskowitz, 2001; Brickley, Linck, and Smith, 2003) in
both worked for Xerox in 1982. Such a long lag between commercial lending. As in Degryse and Ongena (2005), we
relationship formation and lending transactions removes find higher rates when borrowers and lenders are located
reverse causality concerns by construction. within the same city, but the effect of personal connec-
In pooled cross-sectional regressions of interest rates tions remains strong.
charged by syndicates, we find that the presence of at A second possibility is that personal connections are
least one preexisting, personal relationship between the simply picking up the familiar result that lending terms
firm and lender removed by at least 5 years relative to the can change when a firm and bank do repeated business
date of the lending transaction markedly reduces borrow- with each other.3 Empirical tests also reject this possibi-
ing costs. For firms with very good credit (A or better), the lity and provide some insight into previous findings. First,
effect is only 8 basis points (bps); because spreads are the impact of personal connections holds strongly both
bound at zero, the effect for highly rated firms cannot be for a firm’s historical banking partners and for banks with
large. However, the effect climbs steadily as credit quality which it has no prior lending experience. This finding
deteriorates. Firms with ratings in the BBB–B range can underscores that, in relationship banking, it appears to be
expect interest rate concessions of about 20 bps. The access to specific people that makes the difference, not
magnitude more than doubles again for firms rated even familiarity with a firm’s physical assets. Perhaps more
worse or that lack a rating altogether (45–50 bps). One important, the effect of past banking relationships is
might expect the result to strengthen not only because substantially weakened when personal relationships are
default risk increases borrowing costs, but also because added to the regression. This raises the possibility that
adverse selection and monitoring problems are most banking relationships could themselves stem from
severe for these firms. In pooled models controlling for a
variety of firm, industry, loan, and macroeconomic char-
acteristics, we observe similar magnitudes, averaging
between 15 and 20 bps across all credit categories, or 3
See Peterson and Rajan (1994), Berger and Udell (1995), Degryse
about 10% of the average charged spread. For comparison, and Van Cayseele (2000), Bharath, Dahiya, Saunders, and Srinivasan
the average spread between A and AA ratings is 16 bps. (2011), and Schenone (2010).
J. Engelberg et al. / Journal of Financial Economics 103 (2012) 169–188 171
personal relationships, suggesting an original determi- relationships. Although perhaps the most intuitive expla-
nant of one’s financing partners. nation is that personal relationships improve information
When we look at other lending terms besides interest flow and lending efficiency, an important caveat is that
rates, we find no evidence that creditors personally personal relationships could induce inefficiency in lend-
connected to their borrowers seek to protect themselves ing decisions, but nevertheless predict positive subse-
in other ways, such as loaning smaller amounts or using quent performance. To see how, imagine a banker who
more covenants to restrict the firm’s behavior. In fact, the is afraid of appearing corrupt and so requires personally
opposite pattern emerges. With the same set of controls connected firms to meet a higher standard compared with
employed in the spread regressions (e.g., size and prior more anonymous borrowers. Here, even if personally
activity of syndicate banks, firm characteristics, macro- connected bankers are no more informed than their
economic controls, etc.), we find that personally con- unconnected counterparts, they would nevertheless
nected syndicates lend somewhat more on average. (assuming all bankers have information the market does
Moreover, covenants are less likely to be included in deals not) be associated with superior deals ex post. Generally,
between personally connected firms and syndicate banks we are not able to definitively distinguish between these
and, when they are used, are fewer in number. stories, and ultimately, this limits what our study can say
The second half of the paper considers the normative about the impact of personal relationships on lending
implication of these results. Taking as given that firm- efficiency.
bank personal connections alter the terms of lending in A number of papers, many in international contexts,
the firm’s favor, we ask whether these appear to be good have explored whether lending decisions improve or
or bad decisions. Although the source of our banking data worsen when firms and banks are linked in some way
[Dealscan of Loan Pricing Corporation (LPC)] does not that compromises the latter’s objectivity. Generally, the
provide data on specific loan performance, we gain insight evidence suggests that such situations lead to wealth
by examining the evolution of each borrower’s credit transfers from lenders to borrowers, a perhaps unsurpris-
rating subsequent to initiating a bank deal. Although not ing conclusion given the (often extraordinary) conflicts of
specifically related to a given transaction, these summary interest imposed on the lending bank.5 Our study is
statistics measure a firm’s ability to meet its outstanding related to the extent that personal relationships also
debt obligations, part of which includes the bank transac- present an opportunity for a bank to have more intimate
tions we analyze.4 knowledge of a borrower. However, the lack of incentive
We consistently find that after borrowing, the credit conflicts is an important difference and likely contributes
ratings of personally connected firms improve compared to why we find such a positive effect of personal connec-
with their unconnected counterpart borrowers. As a tions on lending decisions. In addition, the exogeneity of
typical example, of the 1,290 BB-rated firms that initiate relationship formation allows for a causal interpretation
syndicated bank deals with at least one connected bank, often made difficult in other settings.
63% maintain the same credit rating in the years imme- Finally, our study contributes to a growing literature
diately following, 22% improve, and 15% worsen. that explores the impact of personal networks on business
In contrast, the comparable distribution for the 1,880 and investment decisions. See Cohen, Frazzini, and Malloy
BB-rated firms completing deals with unconnected banks (2008) for evidence that personal connections enhance
is 64%, 11%, and 25%, respectively. Remarkably, such a information flow among investment professionals,
pattern holds across every credit rating category (AAA/AA, Schmidt (2008) for evidence that information about
A, BBB, etc.), as well as for alternative measures of risk mergers travels across personal networks, and Fracassi
[e.g., Moody’s Expected Default Frequency (EDF) and (2008) for evidence that social relationships among
Moody’s EDF Implied Spread (EIS)]. executives and board members influence investment
Analysis of subsequent stock returns indicates even policy.
stronger results and confirms that such performance We organize the paper as follows. In the next section,
improvements were not foreseen by the market. Pooled we describe the lending and connections data, and we
timeseries cross-sectional regressions of characteristic outline our empirical strategies. We begin our formal
risk-adjusted stock returns (following Daniel, Grinblatt, analysis in Section 3, where we explore the extent to
Titman, and Wermers, 1997) indicates 1-, 2-, and 3-year which firm-bank connections influence lending terms
excess returns of 3%, 10%, and 17%, respectively. In other including interest rates, covenants, and loan amounts.
words, firms completing deals to connected banks experi- Section 4 is dedicated to answering the question of
ence substantially higher stock returns than those bor- whether or not personal connections are associated with
rowing from unconnected syndicates. Fama and MacBeth better or worse future firm performance. We consider
(1973) regressions indicate similar effects. robustness and some extensions to our basic results in
At a minimum, their superior ex post performance Section 5, and then we conclude in Section 6.
indicates that personally connected deals are fundamen-
tally different – namely better – than those lacking such
5
Domestic studies include Kroszner and Strahan (2001) and Güner,
Malmendier and Tate (2008). Rajan and Zingales (1998) and
4
Because credit ratings pertain to a firm’s public debt, analyzing Charumilind, Kali and Wiwattanakantang (2006), Morck and
credit rating changes represents a conservative way of measuring Nakamura (1999) and Hoshi, Kashyap and Scharfstein (1990), Laeven
changes in the likelihood of default on more senior claims, such as (2001), and La Porta, Lopez-de-Silanes, and Zamarripa (2003) study
syndicated bank loans. connected lending in Asia, Japan, Russia, and Mexico, respectively.
2. Data and identification strategy that all the connections we analyze are ultimately
‘‘social’’). Although social connections could have a causal
Management Diagnostic Limited (MDL) is a data pur- influence on lending behavior, BoardEx does not list the
veyor that collects biographical information on executives start and end dates for many of them, e.g., we cannot in all
and board members of public companies. Its main pro- cases tell how long a CFO has served on the board of the
duct, BoardEx, reports work histories, educational back- Bronx Zoo, only that he is currently serving (see also
grounds, and current participation in social organizations Schmidt, 2008; Fracassi and Tate, 2009). In such cases, we
for CEOs, chief financial officers (CFOs), other executives, are not able to tell whether this seat came after a banking
and current directors. BoardEx has been used to examine transaction with another Bronx Zoo director, or vice versa.
the role of social networks in a variety of corporate For this reason, we ignore social connections entirely in
finance settings (e.g., Schmidt, 2008; Cohen, Frazzini and our main analysis. What we lose in statistical power,
Malloy, 2008; Fracassi and Tate, 2009). however, we gain in the ability to make precise, causal
We supplement BoardEx with biographical informa- inferences insofar as personal connections influence lend-
tion on personnel from a large number of public and ing outcomes.
private commercial banks, made generously available In Panel B of Table 1, we list summary statistics for all
after a custom data request to MDL. The union of these three possible types of connections: school, third-party
data results in a universe of 5,057 public US firms, 1,924 past professional, and social. The connection measures are
commercial banks, and 65,074 individuals (either direc- calculated at the syndicate level; for example, the mean
tors or executives at their respective institutions).6 From value of Third-Party Past Professional Connections is 1.28,
these we infer interpersonal linkages between bankers indicating that executives or directors of the typical
and borrowers. borrower share roughly two past jobs (since removed by
Interpersonal relationships are endogenous, a recogni- 5 years or more) with executives or directors at any of the
tion that plays an important role in how we construct our syndicate banks. Past School Connections are far less
network variables. In particular, because we intend to common (mean 0.26), in part because of the time restric-
explain corporate lending behavior with pre-existing tion we impose: Two individuals must have attended the
personal connections between lenders and borrowers, it same educational institution (e.g., Harvard Business
is crucial that we eliminate reverse causation, e.g., a School), but no more than 2 years apart (e.g., respective
commercial banker undercutting her competition by a graduation years of 1991 and 1992 would count as a
few basis points, expecting to be rewarded with a seat on connection, but 1991 and 1993 would not). As seen, social
the borrower’s board. connections are the most common. Throughout our main
Instead, we wish to identify examples in which social analysis, we neglect entirely the effects of social connec-
connections are plainly exogenous to the lending deals we tions, which could be subject to reverse causation in the
analyze. Consequently, we focus on two specific types of context of lending deals.
connections that meet this criterion: (1) school connec- The remainder of the connection-formation process,
tions, formed when two people graduate from the same however, is more subjective, and no theory exists to
educational institution within 2 years of one another (e.g., provide guidance. For example, one might expect a firm’s
Stanford Class of 1984 or 1985), and (2) third-party past connections to the lead bank in a syndicate to be most
professional connections, formed when two people over- valuable for information flow or monitoring, so that
lap through either a common past job (e.g., both having perhaps we should consider only these. However, the fact
worked for IBM in their first job after graduation) or past that syndicate members often conduct multiple deals
board membership (e.g., both having served on Coca together, and rotate the identity of the lead bank across
Cola’s board). Third-party past professional connections deals (e.g., Cai, 2010) suggests that personal connections
must predate the lending deal by more than 5 years and to seemingly peripheral participant banks might be simi-
cannot involve either the borrowing firm or lending larly valued.7 A second consideration is measurement
institution in any way. This requirement ensures error. Clearly, many connections we assign as such are
that connections inferred between a banker at bank X not (most people do not know every member of their
and manager at firm Y are formed at a distant place graduating class, let alone keep in touch with them years
(say, at firm Z or during college) and time (at least 5 later), which attenuates any marginal effects we observe.
years ago). The specification we report balances our intuition between
As a practical matter, this eliminates most of the which connections we think have the potential for
connections we can infer, including those that arise from information transmission and errors-in-variables bias.8
current common participation in social organizations
such as charities, volunteer groups, and museum boards. 7
Cai (2010) studies the syndicated lending market from 2004 to
To distinguish them from their school and third-party
2006 and finds that 77% of lead arrangers also participate in syndicates
past professional analogs, we refer to these as social in which they are not the lead and that ‘‘it is a common practice for
connections (admitting a slight abuse of language given lenders to maintain stable relationships with certain other lenders and
rotate their roles between leading and participating within the group’’.
See also Lin and Paravisini (2010) for evidence of reputation concerns
among syndicate members.
6 8
About 16 hundred of the banks in our data set are private, making In unreported results, we experiment with a number of alternative
this the first study to consider the impact of networks and information definitions for connections, including considering only those between
flow involving nonpublic firms. the borrower and lead bank in the syndicate, requiring connections to be
Table 1
Summary statistics.
The table reports summary statistics for several variables used in the paper. Panel A reports data on syndicated loans, extracted from the Dealscan
database. Variables shown are the Dollar Value of Each Syndicated Loan in millions of dollars, the Total Number of Covenants, the All-in Drawn Spreads in
basis points, the Number of Lenders, and the Number of Local Banks. A lender is considered local if its headquarters is within 100 km of the headquarters of
the borrower. Panel B reports summary statistics for our personal connections variables. Past School Connections is calculated by summing all instances in
which a director or executive of the borrower and a director or executive of the lender attended the same educational institution and matriculated within
2 years of one another. Third-Party Past Professional Connections is formed similarly, but with a common past employer. All professional connections are at
least 5 years removed from the date of any banking activity and do not include instances in which the connection was made at the lending bank or the
borrowing firm. With Current Social Connections, we sum all instances in which a director or executive of the borrower and a director or executive of the
lender have active roles in a common social organization, e.g., serving on the board of United Way. Deal in the Past 1–3 Years Indicator, Deal in the Past 4–6
Years Indicator, and Deal in the Past 7 Years or Earlier Indicator take values of one if the current borrower borrowed from one or more members of the
current syndicate in the most recent three years, the 3 years before that, or the three before that, respectively. Panel C reports the summary statistics for
several borrower fundamentals, including one-year lagged total assets in millions of dollars (Total Assets), profitability as of the most recent fiscal year-
end prior to the loan origination (Profitability), tangible assets normalized by the lagged total assets (Tangibility), market-to-book ratio (M/B), capital
expenditures normalized by lagged total assets (Capital Expenditures/Total Assets), Expected Default Frequency at the end of the month prior to the loan
origination (Expected Default Frequency), and EDF Implied Spread at the end of the month prior to loan origination (EDF Implied Spread).
Mean Median Standard deviation 10th Percentile 90th Percentile
Panel A: Deal characteristics

Dollar value of each syndicated loan (millions of dollars) 656.00 225.00 1,670.00 25.00 2,500.00
Total number of covenants 3.14 3.00 3.39 0.00 9.00
All-in Drawn Spreads (basis points) 206.48 187.50 146.95 40.00 375.00
Number of lenders 7.50 5.00 8.42 1.00 17.00
Number of local banks 1.79 1.00 2.79 0.00 5.00
Panel B: Connection measures

Past school connections per syndicated loan 0.26 0.00 0.87 0.00 1.00
Third-party past professional connections per syndicated loan 1.28 0.00 4.15 0.00 4.00
Current social connections per syndicated loan 2.17 0.00 6.12 0.00 6.00
Deal in past 1–3 years indicator 0.16 0.00 0.36 0.00 1.00
Deal in past 4–6 years indicator 0.15 0.00 0.36 0.00 1.00
Deal in past 7 years or earlier indicator 0.10 0.00 0.31 0.00 1.00
Panel C: Firm characteristics

Total Assets (millions of dollars) 13044.20 1217.82 65290.59 87.55 18954.20
Profitability 0.38 0.13 32.65 0.02 0.27
Tangibility 0.58 0.46 6.79 0.08 0.91
M/B 1.81 1.34 2.83 0.95 2.93
Capital Expenditure/Total Assets 0.08 0.04 0.28 0.00 0.15
Expected Default Frequency (percent) 2.65 0.44 5.26 0.03 8.62
EDF Implied Spread (percent) 323.18 117.38 540.64 21.30 888.68
However, we proceed with the acknowledgment that revolving line of credit.10 For each facility, Dealscan lists a
superior specifications could provide even more informa- number of relevant firm and borrower characteristics
tive estimates. including the amount loaned (or available as a line of
Our analysis involves bank loans made to publicly credit), the identity of the firm and participant banks, the
traded companies within the US, the majority of which stated purpose of the loan, information about covenants,
are syndicated between multiple banks that share lending interest rate, maturity, and presence or absence of secur-
risk. See Sufi (2007) for a detailed discussion of the itized collateral. Our main variables of interest are the rate
syndicated loan market. The source for these data is charged (the all-in drawn spread), covenant variables, and
Dealscan, a proprietary product from Loan Pricing Cor- deal size, which we analyze as functions of the preexisting
poration. This is by now a standard data source, and personal connections between personnel at firm and syndi-
because a number of other papers provide excellent cate banks. However, we employ the majority of the other
descriptions of its features, we refer the reader interested available variables as controls. In Panel A of Table 1, we list a
in more detail than we provide to these.9 number of relevant summary statistics. Because these are
The unit of observation in Dealscan is a ‘‘credit facility,’’ standard, we omit their discussion.
which can be either a loan with a specific maturity or a A considerable part of our analysis concerns the ex
post performance of borrowers after initiating a syndi-
cated loan, specifically as it relates to firm-bank personal
(footnote continued) connections. Ideally, we would examine how individual
formed via multiple channels (e.g., requiring two individuals to have loans perform, but because such data are generally not
overlapped in school and shared a common past employer), and defining
connections only for the firm’s CEO and CFO, rather than for the largest
10
possible set of executives and directors made available by BoardEx. All About 20% of our observations correspond to separate facilities
results hold for each of these alternative specifications. within a lending package. We consider each such facility a separate
9
For recent examples, see Bharath, Dahiya, Saunders and Srinivasan observation (e.g., as does Bharath, Sunder, and Sunder, 2008), but note
(2007) and Qian and Strahan (2007). nearly identical results if aggregated to the package level.
available, we examine various firm-level proxies instead. among 5,721 deals (29%). In such cases, the average
Two of these are very familiar: changes in public credit (median) credit spread is 127 (88 bps). In the remaining
ratings and risk-adjusted stock returns, the former from 13,833 cases, the average spread is considerably higher,
Dealscan (Compustat also lists these) and the latter with an average (mean) of 239 (225) basis points.
from the Center for Research in Security Prices (CRSP). However, in a regression that controls for other determi-
Our distribution of credit ratings (not reported) is nants of credit risk, this difference settles to approximately
standard, with a modal value (BB) just below the invest- 28 basis points (Table 2, Column 1). As seen, this is compar-
ment-grade threshold. Hovakimian, Kayhan, and Titman’s able to shifts in credit ratings. For example, an improvement
comprehensive study of credit rating targets (2009, of two rating categories from A to AAA decreases borrowing
Table 1) finds a nearly identical distribution (see Table 1). costs by 174–144¼30 bps, whereas a single upgrade from
Shown also in Table 1 (Panel C) are summary statistics BBB to A reduces the interest rate by 144–102¼42 bps.12
for two proprietary credit risk measures made available to An important set of controls is the indicators for previous
us from Moody’s KMV: Expected Default Frequencies banking, but not personal, relations between the borrower
(EDFs) and EDF Implied Spread (EIS). 11 These provide and syndicate banks. Theories of financial intermediation
alternative ways of measuring changes in default risk have been advanced to predict both positive and negative
subsequent to a syndicated loan deal and, relative to effects on spreads for repeated firm-bank interactions. Boot
ratings, offer broader and timelier coverage. The first is and Thakor (1994) argue that when reusable information is
a numerical analog to a firm’s credit rating, and the generated in the process of originating a bank loan, sub-
second is a synthetic spread based upon the firm’s sequent spreads are lower because, effectively, the fixed
Expected Default Frequency. Importantly, EDF Implied costs of information production can be spread out over a
Spread is intended to predict spreads on bonds, not on larger number of transactions. However, banking relations
senior bank debt. Thus, EDF Implied Spreads and All-in can create or exacerbate hold-up problems (Hart and Moore,
Drawn Spreads on bank debt are not directly comparable. 1994), increasing the lender’s market power.13 Whether
spreads decline over the course of a banking relationship is,
3. Personal connections and lending terms thus, an empirical question recently taken up by Bharath,
Dahiya, Saunders, and Srinivasan (2011), who find that
We begin our analysis with a simple question: do repeated transactions are generally associated with reduced
lenders personally connected to their borrowers cut them borrowing costs.
better deals? We focus primarily on three terms easily Following these authors, we include dummy variables
available from Dealscan: credit spreads, deal size, and for whether the borrower has transacted with at least one of
restrictive covenants. the syndicate members in the last 3 years (t-3 through
present), in the previous 3 years (t-6–t-4), or even further
3.1. Credit spreads back (t-9–t-7). Confirming the findings of Bharath, Dahiya,
Saunders, and Srinivasan (2011), Column 1 indicates that
Unless a firm can issue riskless debt, the interest rate it previous banking relations are associated with lower
pays will include a spread, usually quoted in basis points spreads and, intuitively, that this declines as the relationship
above LIBOR (London Interbank Offered Rate) or 10-year becomes stale. However, even the largest banking relation-
US Treasury yields. Dealscan employs the former bench- ship indicator has a magnitude (13 bps) less than half that
mark. In our sample of syndicated bank deals, the average of the firm-bank personal relationship indicator.14
(median) spread is 206 (188) bps, indicating that if banks Also included is the number of lenders in the syndicate
can borrow from other banks at 5%, then, over the same (Number of Lenders), as well as the number of aggregate deals
horizon, the average (median) firm can borrow at a completed by the syndicate members in the previous year
statutory rate of 7.06% (6.88%). (Number of Loans Offered by Syndicate Prior Year). With these
To get a sense of the relation between spreads and
connected lending, we focus first on simple, univariate
comparisons. We are able to construct firm-syndicate 12
The notable increase in spreads between BB and BBB ratings
personal relationship measures for almost 20 thousand corresponds to the investment-grade threshold. Several important
deals, although this number is trimmed substantially in investor groups are restricted from holding non-investment-grade debt
regressions that require data availability for the large securities, which can include corporate bonds and syndicated loans (a
ruling by the US Treasury Department in 1936; Financial Institutions
number of firm and industry characteristics we employ. Reform, Recovery, and Enforcement Act (FIRREA) of 1989). See Kisgen
For the time being, we consider this larger set, but keep in and Strahan (2009) for a summary of the historical development of
mind that we are not controlling for other important regulations on credit ratings for bond market participants.
13
determinants of interest rates. Of the 19,554 deals However, analysis in DeAngelo (1981) suggests that more concen-
trated relationships (auditors rather than lenders in her study) could not
matched with our connections database, at least one
increase market power in the way described. The reason is that a larger
school or third-party past professional connection portfolio increases the incentive for the auditor to service any given client,
between the borrowing firm and a syndicate bank exists which works in the client’s favor. In the current context, this argument
would imply an additional reason that spreads should decline over the
course of a banking relationship, similar to the fixed cost argument.
11 14
See Bohn and Crosby (2003) and Agrawal, Arora and Bohn (2004) Alternatively, we have split the sample into two groups: those in
for an overview of the methods behind the estimation of Expected which the firm has conducted a prior deal with a current syndicate
Default Frequency and EDF Implied Spread. See Dvorak (2008) for partner, and those in which it has not. The effect of personal connections
discussion of the adoption of these credit risk measures in practice. of credit spreads is nearly identical in both groups.
Table 2
Firm-bank personal connections and interest rates.
The table relates the firm’s borrowing cost, measured as its All-in Drawn Spread, to borrower and lender personal connections. Key control variables include a
set of lender characteristics, loan characteristics, and macroeconomic conditions at time of loan origination. The Connected Indicator takes a value of one if there
exists at least one school connection or third-party past professional connection between the borrower and any syndicate bank. Deal in the Past 1–3 Years
Indicator, Deal in the Past 4–6 Years Indicator, and Deal in the Past 7 Years or Earlier Indicator take a value of one if the current borrower borrowed from one or
more members of the current syndicate in the most recent three years, the 3 years before that, or the three before that, respectively. The set of loan
characteristic control variables include the logarithm of time until maturity [i.e., the tenor length in months; Log(Maturity)] and the Number of Lenders in the
loan syndicate. The set of syndicate characteristic control variables include the total number of syndicated loan transactions conducted by the participating
banks in the prior year (Number of Loans Offered by Syndicate Prior Year), and the number of local banks in the syndicate (Local Bank Indicator), in which local is
defined as within 100 km of the headquarters of the borrower. The set of macro control variables include the levels and changes in default spread (the yield
spread between BAA and AAA corporate bond indices), the level of and changes in term spreads (the yield spread between ten-year Treasury and three-month
Treasury), and the most recent monthly returns of the Standard & Poor’s 500 index. Seniority Fixed Effect indicates whether the loan is explicitly secured,
whether it is unsecured, or whether this information is missing in Dealscan. Year, industry, and firm fixed effects are conventionally defined. We use Fama and
French 30-industry classifications to define industry dummy variables. Column 1 examines all loans; Columns 2, 3 and 4 examine high (A, AA, and AAA),
medium (B, BB, and BBB), and low rating (CCC and below) loans, respectively, and Column 5 examines loans of firms lacking public credit ratings. Robust
standard errors clustered by firm are in parentheses. n, nn, and nnn represent significance at the 10%, 5% and 1% level, respectively.
Dependent variable: All-in Drawn Spreads
All loans (1) High rating loans (2) Medium rating loans (3) Low rating loans (4) Unrated Loans
(5)
Connected Indicator 27.68nnn 8.452nn 20.12nnn 51.11nn 46.52nnn

(2.720) (3.373) (3.366) (2.088) (5.808)
AAA credit rating 173.8nnn 6.951
(8.628) (6.381)
AA credit rating 161.2nnn
(8.000)
A credit rating 144.1nnn 8.272
(6.006) (5.458)
BBB credit rating 102.3nnn 110.0nnn
(5.467) (5.384)
BB credit rating 44.24nnn 43.75nnn
(5.192) (4.666)
B credit rating 3.582
(5.053)
CCC credit rating 35.98nnn
(4.664)
CC credit rating 15.55 35.31
(12.75) (25.14)
C credit rating 1.563 36.96
(37.17) (37.78)
Log(Maturity) 1.564 1.463 0.0325 32.18 2.340
(5.596) (3.012) (8.268) (51.12) (10.16)
Deal in past 1–3 years indicator 13.30nnn 0.651 8.215nn 11.27 19.37nnn
(2.853) (4.089) (3.648) (20.11) (5.568)
Deal in past 4–6 years indicator 7.361nn 3.399 9.694nnn 14.13 0.378
(2.967) (4.947) (3.266) (24.91) (7.024)
Deal in past 7 years or earlier indicator 6.773nn 3.151 4.279 40.18 12.75n
(3.043) (2.503) (3.983) (26.98) (6.845)
Number of lenders 0.207 0.231 0.0587 0.191 0.319
(0.164) (0.192) (0.178) (0.742) (0.447)
Number of loans offered by syndicate prior year 0.0210nnn 0.00483nnn 0.0177nnn 0.0322nnn 0.0305nnn
(0.00111) (0.00179) (0.00129) (0.0102) (0.00242)
Local bank indicator 0.535 0.298 0.252 0.965 1.207
(0.470) (0.475) (0.535) (3.838) (1.302)
Macroeconomic controls Yes Yes Yes Yes Yes
Seniority fixed effect Yes Yes Yes Yes Yes
Year fixed effect Yes Yes Yes Yes Yes
Number of observations 17,428 1,705 8,666 359 6,698
Adjusted R2 0.431 0.368 0.448 0.250 0.230
variables, we wish to model any size effects that could lead does not appear significant, whereas more active banks
larger or more active syndicate banks to charge different charge somewhat lower spreads.
rates of their borrowers.15 As seen, the number of lenders In addition, we collect for each borrower and syndicate
bank their zip codes and, when available, calculate the
distance between their respective headquarters. If located
15
We also estimate each of our models with indicators for indivi-
less than 100 km apart, the Local Bank Indicator takes a
dual banks, with little change in the results. See Section 5 for these and value of one and zero otherwise. We include this variable
other issues related to robustness. for two reasons. The first is that if information collection
or monitoring costs depend on proximity, then we want these results. We do note, however, that, as pointed out
to account for these cost differences in our regressions. by Faulkender and Petersen (2006), the decision to secure
The second is that the main variables of interest, those a public debt rating is endogenous and is correlated with
relating to personal connections, could be highly corre- the firm’s information environment. Specifically, firms
lated with the proximity between a bank and borrower. with sensitive information could find the increased dis-
To make sure that firm-bank personal connections are not closure requirements of public debt issues undesirable,
simply picking up common location, we model the latter and thus private debt issues are more attractive. In such
explicitly. The Local Bank Indicator has a positive relation situations, personal connections that confer trust are
with spreads, which is similar to the findings in Degryse likely to be of particular value.
and Ongena (2005). The effect is small and statistically A potential criticism of the results of Table 2 is that
insignificant in Table 2 but is significant in other specifi- although we have controlled for the probability of default
cations (e.g., Table 3). Nevertheless, the inclusion of the with credit ratings, we have not accounted for differential
Local Bank Indicator has little effect on the relation recoveries given default. Because recoveries depend on
between personal connections and spreads. industry and firm characteristics (for evidence, see
Finally, we include a number of macroeconomic con- Altman and Kishore, 1996; Acharya, Bharath, and
trols. Following Fama and French (1989) and Collin- Srinivasan, 2007), we include in Table 3 a number of
Dufresne, Goldstein, and Martin (2001), we include the firm- and industry-specific control variables likely to
following five variables: Level of Term Spread (the differ- affect asset recoveries in liquidation. Along with dummies
ence between ten-year Treasury yield and three-month for each of the Fama and French (1989) 30 industry
Treasury yield), One-Year Change in Term Spread, Default classifications, we also include each firm’s lagged total
Spread (the difference between the Moody’s Baa corpo- assets [in logarithms; Log (Total Assts)]), market-to-book
rate bond index yield and the Moody’s Aaa corporate ratio (M/B), capital expenditures scaled by assets (Capital
bond index yield), One-Year Change in Default Spread, Expenditure/Total Assets), percentage of assets that are
and One-Year Value-Weighted Return on the S&P 500 tangible (Tangible/Total Assets), profitability earnings
index. Generally, none of these provides any significant before interest, taxes, depreciation, and amortization
explanatory power. We also include year dummies, the (EBITDA, scaled by assets; Profitability), and capital asset
logarithm of the loan or credit line’s maturity (in months), pricing model beta (CAPM Beta). If creditors account
and indicators for whether or not the facility is secured for the expected correlation of default losses with the
with collateral. aggregate market (Ross, 1985; Almeida and Philippon,
The second through fifth columns break up this regres- 2007), we should expect a positive coefficient on the
sion by credit rating groups. Deals in which the bor- latter.
rower’s credit rating is A or better (A, AA, or AAA) are Requiring data availability for all of these variables
shown in Column 2, which indicates that, on average, substantially reduces the size of our sample, to just
personally connected deals are perceived by syndicates as over 11 thousand observations. Summary risk measures
being less risky. The point estimate on the personal are so important for predicting spreads, but because
connections indicator is 8 bps, which, although small so many firms are not publicly rated, in Table 3 we
in an absolute sense, is almost 20% of the average spread account for default risk with Moody’s KMV Expected
for this group (mean 43 bps). Default Frequency for which we have more extensive
The same analysis is repeated for credit rating groups coverage. We group firms into deciles of EDF and then
BBB-B and CCC-C, respectively, in subsequent columns. include dummies for nine of these in the regressions.
Results from the BBB-B group indicate substantial varia- Including the numerical value of EDF makes almost no
tion in credit quality, with spreads ranging 110 basis difference.
points on average between categories. Moreover, the The first column of Table 3 shows the results. Although
effect of firm-bank personal relationships is over twice the coefficient on the personal connections indicator
as strong, leading to an average reduction in the spread of (Connected Indicator) drops somewhat, it remains highly
20 bps with personal relationships present. The fourth significant, both statistically (Po0.001) and economically
column contains only 359 observations, but because the ( 18 bps).16 As before, this coefficient becomes more
magnitude on the relationship indicator is so high negative for firms with worse credit ratings, although, to
( 51 bps), it nevertheless yields a statistically significant save space, we do not repeat this disaggregation. Most of
estimate for this sample. Perhaps the most immediate the firm-level variables either are, or border on being,
takeaway from Table 2 is that personal relationships are a statistically significant, with size, profitability, and mar-
robust determinant of borrowing costs, but mostly for ket-to-book having the most predictive power.
firms with poor credit. For making causal inferences, it is important that
The final column shows the results for the roughly 45% personal connections are not simply capturing other firm
of firms lacking a public credit rating at the time the attributes that could affect borrowing costs. In the second
syndicated deal is initiated. Interestingly, the effect of column, we include firm fixed effects and, thus, hold the
personal relationships for these unrated firms is similar to
those observed for low credit rating firms (particularly 16
The reduction in magnitude on the firm-bank personal connec-
those with CCC credit or worse), with a magnitude of tions indicator coefficient is primarily due to the changing of the sample
47 bps. Because we know relatively little about the (firms without Compustat data are more likely to be young, small,
credit characteristics of these firms, we do not emphasize growth firms), not to the addition of new control variables.
Table 3
Firm-bank connections and loan spreads.
This table relates the firm’s borrowing cost to borrower and lender personal connections. Key control variables include a set of borrower financial
fundamentals, lender characteristics, loan characteristics, and macroeconomic conditions at time of loan origination. The Connected Indicator takes a
value of one if there exists at least one school connection or third-party past professional connection between the borrower and any syndicate bank. The
logarithm of this variable is self-explanatory. The dependent variables in columns 1 and 2 are the All-in Drawn Spreads reported by Dealscan. The
dependent variables in columns 3 and 4 are the logarithm of the All-in Drawn Spreads. The set of borrower fundamental control variables include the
capital asset pricing model (CAPM Beta) estimated using the past 3 years of monthly returns with a minimum of 18 monthly observations, logarithm of
total assets [Log(Total Assets)], market-to-book ratio (M/B), capital expenditures normalized by lagged total assets (Capital Expenditure/Total Assets),
tangible assets normalized by the lagged total assets (Tangibility/Total Assets), and profitability as of the most recent fiscal year end prior to the loan
origination (Profitability). The set of loan characteristic control variables include the logarithm of time until maturity [i.e., the tenor length in months;
Log(Maturity)], and the Number of Lenders in the loan syndicate. The set of syndicate characteristic control variables include the total number of
syndicated loan transactions conducted by participating banks in the prior year (Number of Loans Offered by Syndicate Prior Year), and the number of local
banks in the syndicate (Local Bank Indicator), in which local is defined as within 100 km of the headquarters of the borrower. The set of macro control
variables include the levels and changes in default spread (the yield spread difference between BAA and AAA corporate bond indices), the level of and
changes in term spread (the yield spread difference between ten-year Treasury and three-month Treasury), and the most recent monthly returns of the
Standard & Poor’s 500. Seniority Fixed Effect indicates whether the loan is explicitly secured, whether it is unsecured, or whether this information is
missing in Dealscan. EDF Decile Fixed Effect pertains to the set of dummy variables that take a value of one if the borrower’s monthly Expected Default
Frequency (EDF) value at time of loan origination falls into one of the ten EDF deciles. Year, industry, and firm fixed effects are conventionally defined. We
use Fama and French 30-industry classifications to define industry dummy variables. Robust standard errors clustered by firm are in parentheses. n, nn,
and nnn represent significance at the 10%, 5% and 1% level, respectively.
Dependent variable: All-in Drawn Spreads Dependent variable: log(All-in Drawn Spreads)
(1) (2) (3) (4)
nnn nnn
Connected Indicator 17.77 17.29
(3.431) (3.704)
Log (1 þNumber of Connections) 0.134nnn 0.0483nnn
(0.0146) (0.0118)
CAPM Beta 0.470 0.169 0.0186nn 0.00545
(1.611) (1.899) (0.00797) (0.00832)
Log(Total Assets) 4.338nnn 13.36nn 0.0687nnn 0.0548nn
(1.621) (5.415) (0.0120) (0.0221)
M/B 1.517nn 3.704nn 0.0199nnn 0.0231nn
(0.682) (1.704) (0.00637) (0.0100)
Capital Expenditure/Total Assets 1.448 29.89n 0.0110 0.162nn
(15.65) (17.87) (0.0756) (0.0678)
Tangibility/Total Assets 6.505 1.063 0.0290 0.0213
(4.387) (5.087) (0.0233) (0.0251)
Profitability 31.05nnn 75.12nnn 0.136nn 0.388nnn
(8.608) (17.01) (0.0602) (0.0750)
Log(Maturity) 11.04nn 5.328 0.112nnn 0.0184
(5.156) (4.918) (0.0280) (0.0235)
Deal in past 1–3 years indicator 3.446 1.088 0.0199 0.000936
(3.344) (3.313) (0.0204) (0.0189)
Deal in past 4–6 years indicator 3.087 0.510 0.0151 0.0204
(3.158) (3.141) (0.0192) (0.0162)
Deal in past 7 years or earlier indicator 8.134nn 9.701nnn 0.0202 0.0469nn
(3.412) (3.363) (0.0231) (0.0200)
Number of loans offered by syndicate prior year 0.0175nnn 0.0121nnn 8.82e-05nnn 7.46e-05nnn
(0.00150) (0.00142) (9.63e-06) (7.71e-06)
Local bank indicator 2.149nnn 2.548nnn 0.00935nn 0.0128nnn
(0.618) (0.679) (0.00393) (0.00330)
Number of lenders 0.200 0.845nnn 0.00213 0.00393nn
(0.300) (0.289) (0.00221) (0.00157)
Macroeconomic controls Yes Yes Yes Yes
Seniority fixed effect Yes Yes Yes Yes
EDF decile fixed effect Yes Yes Yes Yes
Year fixed effect Yes Yes Yes Yes
Industry fixed effect Yes No Yes No
Firm fixed effect No Yes No Yes
Number of observations 11,003 11,003 11,003 11,003
Adjusted R2 0.504 0.745 0.615 0.860
borrower constant but vary the lending syndicate. This probability of and losses given default in Column 1, latent
procedure admits only the set of firms that complete at firm characteristics play an important role in lenders’ risk
least one deal with a connected syndicate and at least one assessments. Nonetheless, the inclusion of firm dummies
with a nonconnected syndicate. leaves the personal connections indicator nearly
The marked increase in R2 (from 0.50 to 0.75) makes unchanged. Holding the borrowing firm constant, Column
clear that, despite our attempts to control for the 2 indicates that the presence of at least one school or
third-party past professional connection reduces the certain types of provisions are more or less common in
charged interest rate by 17 basis points (Po0.001). connected deals. For about one-third of the deals, no
The third and fourth columns of Table 3 show the covenant is listed in Dealscan; for the remaining two-
results when we model the personal connection—credit thirds, the average number of covenants is 4.7, with a
spread relation with logarithms. Comparing Columns 1 standard deviation of 3.1.
and 3, we see that a logarithmic specification not only Table 4, Panel A, presents the results of analyzing loan
provides a substantially better fit (R2 ¼0.615), but also covenants as a function of our personal connections
strengthens the statistical significance of firm-bank per- variables. We employ the same set of firm, loan, bank,
sonal connections. The coefficient on the log of connec- and macroeconomic controls as in Table 3. In the first two
tions indicates that by doubling the number of personal columns, the dependent variable is discrete, taking a
connections between a firm and its syndicate partners, value of one if any covenants are listed by Dealscan and
the firm pays a spread over 13% less. On average, this zero otherwise. The marginal effects shown in these
means that 1.5 additional connections (the mean of this columns indicates only suggestive evidence for the indi-
variable) are associated with a spread reduction of cator connections variable (Column 2), but a stronger
approximately 179 0.134¼24 bps. The final column result for the more continuous connections variable
shows that although including firm fixed effects substan- (Column 1). By doubling the number of personal connec-
tially decreases the magnitude of the spread-connection tions, the probability of covenants being included
elasticity (point estimate of 0.048), it remains highly decreases by 2.3%, a result significant at the 1% level.
significant (P o0.001). In unreported results, we find that this result – like all
Before proceeding, we briefly note that the nonlinear others in the paper – is considerably stronger for firms
relation between spreads and firm-bank personal connec- with poor credit ratings.
tions indicated in the log–log specification is confirmed in For robustness, shown in the next columns are results
a number of unreported specifications (e.g., quadratic, from linear regressions, in which the dependent variable
nonparametric regressions). Regardless of the empirical is the number of covenants included (possibly zero). We
model, we consistently find that the value of each con- conduct this exercise to allow firm fixed effects. As in the
nection diminishes as the aggregate number of firm-bank previous columns, the logarithmic specification indicates
connections within the syndicate increases. Given that a negative relation between firm-bank personal connec-
spreads are bound from below at zero, this result might tions and covenants. The discrete specification for the full
not be particularly surprising. However, this constraint sample does not.
binds for only firms of the highest credit quality, and as
we have already seen, these are exceptional cases. 3.3. Deal size
3.2. Covenants The results so far indicate that firm-bank personal

connections lead to less stringent lending terms, and that
Interest rates are but one mechanism by which syndi- firms with the worst credit (for which adverse selection
cate banks can protect themselves ex ante from the risk of and managerial incentive problems are likely the greatest)
having financed a poor project or from ex post risk benefit the most. Here, we consider whether the effects
shifting by management. The state-dependent transfer we find apply only to small loans, or whether they
of control rights via covenants is another. Here, we generalize to larger stakes.
explore whether personally connected lenders substitute In Table 4, Panel B, we consider as the dependent
interest rate concessions for tighter or more restrictive variable the natural logarithm of the deal size, or tranche
covenants that constrain the firm’s behavior. amount. All columns employ the same set of control
Essentially, covenants are provisions in a debt contract variables used in previous tables including firm size (lag
that specify technical default. Even if a firm has not of total assets, volatility, Fama and French 30 industry
missed an interest or principal payment, violation of a classification, etc.).
covenant shifts control rights to the lender(s), requiring Estimates in the first and second columns suggest that
the borrower, for example, to accelerate principal repay- increasing the number of firm-bank personal connections
ment or post additional collateral. Covenants are discre- increases the amount lent. The log–log specification
tionary features in credit agreements and often pertain to indicates an elasticity of roughly 3.5 (Column 1). The
operating performance or debt coverage ratios. A number discrete specification (Column 2) shows that compared
of recent papers have investigated the role of covenants with deals lacking personal connections, syndicated deals
insofar as they relate to creditor intervention (Chava among personally connected members are over 13%
and Roberts, 2008; Nini, Smith, and Sufi, 2009; Roberts larger, translating to roughly $45 million on average. In
and Sufi, 2009a), renegotiation (Roberts and Sufi, 2009b), the final two columns of Panel B with firm fixed effects,
and the sales of syndicates loans (Drucker and Puri, 2009; both specifications indicate a strong, positive relation.
Gupta, Singh, and Zebedee, 2008). Compared with the specification in Column 1, the inclu-
We take a reduced form approach and simply sum the sion of firm fixed effects slightly strengthens the result.
number of covenants (if any) listed for each credit facility. The elasticity is 0.076, indicating that 1.5 additional
Besides that reflected by their prevalence, our analysis connections increase the average loan balance by
ignores any information reflected in the covenants them- more than $40 million. The discrete model shown in the
selves, e.g., whether they are strict or slack, or whether final column indicates a slightly strengthened effect for
Table 4
Firm-bank connections, loan covenants and loan sizes.
Panel A relates the number of covenant restrictions of the loan to borrower and lender personal connections. Panel B considers as the dependent
variable the natural logarithm of the loan amount (dollars). Control variables include a set of borrower financial fundamentals, lender characteristics,
loan characteristics, and macroeconomic conditions at time of loan origination. The Connected Indicator takes a value of one if there exists at least
one school connection or third-party past professional connection between the borrower and any syndicate bank. The logarithm of this variable
is self-explanatory. The same set of firm, loan, lender, industry, and macro controls in Table 3 are employed here. The dependent variable in columns 1
and 2 is a dummy variable that takes a value of one if the firm has any covenants listed in Dealscan; the dependent variable in Columns 3 and 4 is
Number of Covenants. Robust standard errors clustered by firm are in parentheses. n, nn, and nnn represent significance at the 10%, 5% and 1% levels,
respectively.
Panel A: Firm-bank connections, loan covenants
Dependent variable: Covenant Indicator Dependent variable: Number of Covenants
(1) (2) (3) (4)
Connected Indicator 0.0124 0.0715

(0.0146) (0.112)
nnn n
Log (1 þ Number of Connections) 0.0226 0.112
(0.00829) (0.0634)
Firm characteristics controls Yes Yes Yes Yes
Loan characteristics controls Yes Yes Yes Yes
Bank characteristics controls Yes Yes Yes Yes
Industry fixed effect Yes Yes No No
Firm fixed effect No No Yes Yes
Pseudo (Columns 1 and 2) or Adjusted (Columns 3 and 4) R2 0.378 0.377 0.678 0.678
Panel B: Firm-bank connections and loan sizes
log(Tranche Amount) log(Tranche Amount) log(Tranche Amount) log(Tranche Amount)
(1) (2) (3) (4)
Connected Indicator 0.134nnn 0.147nnn

(0.0318) (0.0347)
Log (1 þ Number of Connections) 0.0352n 0.107nnn
(0.0205) (0.0208)
Industry fixed effect Yes Yes No No
Firm fixed effect No No Yes Yes
Adjusted R2 0.652 0.653 0.812 0.812
personal connections on loan balances, compared with problems or improving the bank’s ability to monitor and
the model without firm effects. alleviate borrower’s moral hazard incentives – then the
concessions shown in Tables 2–4 could be warranted. The
4. Ex-post performance ideal test would be to compare default rates between
connected and unconnected syndicates. Unfortunately,
The results of Section 3 indicate that firm-bank perso- Dealscan does not provide data on the performance of
nal connections shift lending terms in the borrower’s individual loans, and because the secondary market for
favor but are silent with respect to the reasons why. such securities is extremely illiquid, examining prices is
Holding risk constant, more lenient terms would result in not feasible. Absent performance data on specific loans,
a wealth transfer from the bank to the firm’s shareholders. we examine various firm-level performance metrics that,
However, if firm-bank connections alter the risk profile of while noisy, nevertheless provide information about the
the borrower – either by mitigating adverse selection firm’s ability to service its debt obligations: credit ratings,
Fig. 1. Credit ratings (CRs) evolution for connected and unconnected firms. This figure shows the evolution of long-term public debt ratings. Ratings
for firms that complete loans with personally connected banking syndicates (at least one school or third-party past professional connection) are shown in
gray; those of their counterparts borrowing from nonconnected syndicates are shown in black. Initial credit ratings are those as of the loan’s start date.
Final ratings correspond to those as of July 2009. Firms with initial ratings above AA or below CCC are omitted due to a small number of observations.
Expected Default Frequencies, EDF Implied Spreads, and the sample by about 20%, relative to that analyzed in
stock returns. All of these are benchmarked to the date of Section 3.17
the syndicated deal and tracked forward. In Fig. 1, we compare the evolution of future credit
ratings following personally connected deals (gray bars),
to that following unconnected deals (black bars). Initial
4.1. Future credit ratings credit ratings are shown in each panel, starting with
rating category AA. Final ratings are those as of July
If a firm’s fundamentals deteriorate after securing a 2009. The striking differences between the black and gray
loan or line of credit, this should be captured by changes bars in Fig. 1 underscore the importance of personal
in its future credit ratings. Dealscan provides, for every connections as an ex ante indicator of deal quality. As
firm with publicly rated debt, the long-term rating at seen, the credit ratings of connected (unconnected) firms
the time the syndicated deal is initiated. From Moody’s tend to drift upward (downward) or remain the same.
(and then cross-checked with Compustat), we obtain Without exception, this pattern holds for every initial
each borrower’s future credit rating 12, 24, and 36 rating category, a remarkable finding given that we are
months subsequent to the deal of interest. In addition, analyzing changes in ratings, not levels. The probability of
we collect ratings as of July 2009, the date the data were being downgraded following a connected deal, by rating
assembled. category is AAA: 4.7%, AA: 5.8%, A: 9.7%, BBB: 6.2%, BB:
Before proceeding, we note one important change to 14.4%, B: 5.0%, and CCC: 0%. The comparable list for firms
the sample. In Section 3, the unit of observation was the that borrow from unconnected syndicates is AAA: 10%,
individual credit facility, which occasionally included AA: 44.2%, A: 15.6%, BBB: 10.5%, BB: 23.6%, B: 7.0%, and
multiple tranches within a loan package defined by firm, CCC: 0%. The mirror pattern is seen for upgrades.
syndicate group, and origination date. In other words, a Table 5 puts these univariate patterns in a regression
syndicate might for example simultaneously provide a framework. The first, second, and third pairs of columns,
$500 million line of credit at 7% and a subordinated $300 respectively, track credit ratings changes at the 12-, 24-,
million line of credit at 8%. Following Bharath, Sunder, and and 36-month interval after the initiation of a syndicated
Sunder (2008), we treated these as independent observa- bank deal. In each case, the dependent variable is a
tions in our previous analysis. However, while the fact discrete indicator Credit Rating Downgrade, taking a value
that loan characteristics vary across tranches justifies of one if the firm is subsequently downgraded (e.g., BBB to
their inclusion in the previous application, this is clearly BB or below) and zero otherwise.18
inappropriate when examining firm-level performance.
Even if a firm borrows against multiple lines of credit 17
The results in Section 3 are nearly identical in each specification.
within the same loan package, this clearly constitutes 18
Considering separately upgrades, downgrades, and no change in a
only one independent observation for the firm’s ex post single ordered probit specification yields similar qualitative predictions,
performance. Collapsing at the package level reduces but for ease of interpretation, we use the binary specification and
Table 5
Firm-bank connections and future credit rating downgrades.
The table reports the marginal effects of the borrower and lender personal connections on future credit rating changes at different horizons. The same
standard set of firm, loan, industry, and macro controls in Table 3 are employed here. The dependent variables are indicators for whether the firm
experienced a downgrade in its long-term Standard & Poor credit rating over various horizons after completing a syndicated loan. The initial credit rating
is the borrower’s credit rating when the syndicated deal was completed. Marginal effects from Probit regressions are shown. n, nn, and nnn represent
significance at the 10%, 5%, and 1% level, respectively.
Credit rating downgrade: future 12 Credit rating downgrade: future 24 Credit rating downgrade: future 36
months months months
(1) (2) (3) (4) (5) (6)
nnn nnn nnn

Connected Indicator 0.0226 0.0561 0.0724
(0.00753) (0.0113) (0.0145)
Log (1 þ Number of Connections) 0.0104nnn 0.0142nn 0.0186nn
(0.00400) (0.00586) (0.00755)
Firm characteristics controls Yes Yes Yes Yes Yes Yes
Bank characteristics controls Yes Yes Yes Yes Yes Yes
Macroeconomic controls Yes Yes Yes Yes Yes Yes
EDF decile fixed effect Yes Yes Yes Yes Yes Yes
Year fixed effect Yes Yes Yes Yes Yes Yes
Industry fixed effect Yes Yes Yes Yes Yes Yes
Number of observations 5,758 5,758 5,154 5,154 4,255 4,255
Pseudo R2 0.089 0.089 0.106 0.101 0.122 0.117
As seen in Columns 1, 3, and 5, the presence of at least EDF in either direction would move the corresponding
one personal connection has a dramatic effect on the credit rating approximately one-half a rating category.
future trajectory of credit rating changes. With each The logarithmic specification for connections is somewhat
passing year, connected firms are about 2.5% less likely weaker from a statistical significance perspective. How-
to be downgraded than their unconnected counterpart ever, all the point estimates are negative, and the final
borrowers. By the third year, the effect is over 7% and is column is significant at the 5% level.
significant at far better than the 1% level. In the second, A similar picture emerges in Panel B, where each firm’s
fourth, and sixth columns, the logarithmic specification future EIS is modeled as a function of firm-bank personal
also significantly predicts downgrades, and more so at connections, along with the usual set of control variables.
longer horizons. The first column indicates that even controlling for the
firm’s initial EIS, the presence of personal connections to
4.2. EDFs and EISs syndicate members reduces its future, expected borrow-
ing cost by 49 basis points 12 months in advance. By 24
The preceding exercise is possible only for firms with months, the expected reduction is 77 basis points, in the
public debt ratings. Here, we gain roughly three thousand neighborhood of being upgraded from junk (BB or worse)
observations by regressing future EDFs (Panel A) and EISs to investment grade (BBB or worse). At 3 years, the
(Panel B), both firm-level credit risk estimates provided marginal effect is 80 basis points. As in the EDF regres-
by Moody’s, on the firm-bank personal connections vari- sions, the log specification (Columns 2, 4, and 6) is not as
ables used in our previous tests. Although we include the strong but paints largely the same picture.
same set of firm and industry characteristics as in pre- Because EIS is designed to measure spreads for public
vious regressions, the key control variable is the value of debt, the magnitudes observed in Table 6 are substantially
either EDF or EIS when the loan originates. higher than what is observed in Tables 2 and 3. Bank debt
Comparing Table 6, Panel A, Columns 1 (12 months), 3 is almost always written senior to bonds, a priority
(24 months), and 5 (36 months), the presence of firm- structure that inherently places the latter at higher risk.
bank personal connections remains an important predic- We present the EIS results to emphasize exactly this
tor of Expected Default Frequency over each window. distinction. Table 2 already shows that the impact of
As in Table 5, the effect pronounces at longer horizons. personal connections on borrowing costs is decreasing
For example, in the 36-month period shown in Column 5, in default probability. If the same dynamics apply to more
firm-bank personal connections are associated with junior claims (e.g., bond placements with institutional
almost a three-fourths unit decrease in EDF. To put this investors), then the magnitudes we find for bank loans are
in perspective, the average firm has an EDF of 2.71, which likely a lower bound on the more general effects in debt
would correspond roughly to a BB rating. A unit shift of markets.
4.3. Stock returns

(footnote continued)
present marginal effects. For firms completing deals in the latter part of
our sample, not enough time has passed for their future credit ratings to
The three dependent variables considered so far –
be analyzed (e.g., a firm borrowing in December 2007 does not, at the future credit ratings, Expected Default Frequency, and EDF
time of writing, have a 36 months ahead rating). Implied Spread – are all explicitly designed to evaluate the
Table 6
Connections and alterative measures of future credit risk.
The table relates future Expected Default Frequency (EDF, Panel A) and EDF Implied Spread (EIS, Panel B) to borrower and lender past connections, a set of
borrower fundamentals, lender characteristics, and macroeconomic conditions at the time of loan origination. The set of control variables is the same as
those reported in Table 3. The Number of Connections describes the sum of current school connections and third-party past professional connections. The
reference date is when the syndicated deal is initiated. Robust standard errors clustered by firm are in parentheses. n, nn, and nnn represent significance at
the 10%, 5%, and 1% level, respectively.
Panel A: Connections and Firm’s Future Expected Default Frequencies (EDF)
Dependent variable: Expected Default Frequency
EDF 12 months-ahead EDF 24 months-ahead EDF 36 months-ahead
nnn nnn
Connected Indicator 0.427 0.741 0.734nnn
(0.104) (0.177) (0.211)
Log (1þ Number of Connections) 0.140nn 0.215n 0.311nn
(0.0599) (0.124) (0.135)
Current EDF 0.636nnn 0.637nnn 0.366nnn 0.368nnn 0.228nn 0.229nn
(0.0659) (0.0659) (0.0784) (0.0784) (0.0898) (0.0897)
Adjusted R2 0.527 0.526 0.293 0.291 0.215 0.213
Panel B: Connections and firm’s future EDF Implied Spreads
Dependent variable: EDF Implied Spread
EIS 12 months-ahead EIS 24 months-ahead EIS 36 months-ahead
nnn nnn
(11.43) (19.49) (24.36)
Log (1þ Number of Connections) 18.96nnn 22.24 34.97nn
(6.577) (14.20) (15.49)
Current EDF Implied Spreads (EIS) 0.525nnn 0.527nnn 0.357nnn 0.359nnn 0.203nnn 0.203nnn
(0.0572) (0.0572) (0.0612) (0.0612) (0.0631) (0.0629)
Adjusted R2 0.519 0.518 0.333 0.332 0.256 0.254
firm’s ability to service its debt obligations. Stock returns control variables. However, in Panel A, the dependent
are also useful in this regard and importantly, are immune variable is each stock’s size, book-to-market ratio, and
from the criticism that credit rating changes are serially price momentum characteristic-adjusted return, follow-
correlated or are predictable from other information not ing Daniel, Grinblatt, Titman, and Wermers (1997). Essen-
captured in our regressions. It is important to note, tially, this approach adjusts individual stock returns by
however, that tests of firm-bank connectivity for future subtracting the returns from a portfolio with similar size,
stock returns are joint tests. They test whether the book-to-market ratio, and price momentum. As before,
information in connected or unconnected deals is value- we allow borrower-syndicate personal connections to
relevant for equity prices and whether the market enter in both a discrete and logarithmic specification.
impounds this information immediately into prices.19 The first two columns of Panel A indicate that, over a
Table 7 contains three panels. Compared with Table 6, one-year window, only suggestive evidence exists that
each panel considers the same horizons, sample, and stock returns of connected borrowers are higher than
those of their unconnected counterparts. Both point
19
estimates are positive, but the standard errors are rela-
In an efficient market in which all personal connections were
publicly available at the time of the loan, we would not expect
tively large by comparison. In the third and fourth
predictability for stock returns following connected or unconnected columns, we see stronger evidence that returns are pre-
deals. dictable from a firm’s personal connectedness to its
Table 7
Connections and future stock returns.
The table relates future stock returns of the borrower to borrower and lender personal connections, a set of borrower financial fundamentals, lender
characteristics, and macroeconomic conditions at time of loan origination. The dependent variable in both panels is the cumulative Daniel, Grinblatt,
Titmann and Wermers (1997) characteristic-adjusted returns 12, 24, and 36 months after loan origination. The set of control is the same as those
reported in Table 3. The Number of Connections describes the sum of current school connections and third-party past professional connections. The
reference date is when the syndicated deal is initiated. Panel A shows the results of time-series cross-sectional regressions and Panel B shows the results
of (monthly) Fama and MacBeth regressions. Robust standard errors clustered by firm are reported in Panel A and Fama and MacBeth standard errors are
reported in Panel B. n, nn, and nnn represent significance at the 10%, 5%, and 1% level, respectively.
Panel A: Connections and firm’s future cumulative returns, time-series cross-sectional regressions
Dependent variable: Return at Different Horizons
12-months ahead 24-months ahead 36-months ahead
Connected Indicator 0.0344n 0.106nnn 0.170nnn

(0.0189) (0.0312) (0.0430)
Log (1 þ Number of Connections) 0.0164n 0.0499nnn 0.0743nnn
(0.00886) (0.0149) (0.0206)
Adjusted R2 0.025 0.025 0.037 0.036 0.051 0.049
Panel B: Connections and firm’s future cumulative returns, Fama and MacBeth regressions
Dependent variable: Return at Different Horizons
12-months ahead 24-months ahead 36-months ahead
nn nnn
(0.0239) (0.0302) (0.0519)
Log (1 þ Number of Connections) 0.0280nnn 0.0823nnn 0.1168nnn
(0.0104) (0.0152) (0.0248)
Average Cross-Sectional R2 0.076 0.067 0.055 0.051 0.059 0.050
syndicate members. The log specification indicates that against the personal connections variables. For example,
doubling the number of personal connections increases in July 2005, we regress the 12-, 24-, or 36-month future,
the firm’s risk-adjusted stock returns by almost 5% characteristic-adjusted returns of every firm that bor-
(Po0.001). The discrete specification effectively com- rowed in that month. By running such a regression month
pares connected versus unconnected deals and indicates by month, we eliminate by construction cross-sectional
a two-year, risk-adjusted difference of over 10%. The final correlation. The averaged coefficients on the connections
two columns show that, at the three-year horizon (we use variables are shown in Panel B and, in every case,
the most recent stock price if 3 years have not passed), strengthen relative to those seen in Panel A.
connected borrowers perform 17% better than borrowers We also experiment with calendar time portfolios that
not personally connected to their syndicates (Po0.001). involve long positions in connected borrowers and short
Annualized, this corresponds to a risk-adjusted (excess) positions in unconnected ones. Because we have such a
return of 5.6%. short time span, the number of monthly observations
One potential concern is that the results in Panel A afforded by such an approach is small (around one
could be picking up common, date-specific factors that hundred). In unreported results, we find trading profits
influence returns. Although we have little reason to on par with the results observed in Panels A and B. Long-
believe that such time effects would be systematically short portfolios average between 20 and 30 basis points
related to personal connections, Panel B presents the per month and regardless of the holding period (12, 24, or
results of Fama and MacBeth monthly regressions. Here, 36 months), yield positive trading profits in more than
we consider each month as a separate family of observa- half the months. However, even the best of these yields
tions and regress future risk-adjusted stock returns only a t-statistic in the 1.8 range, bordering on statistical
significance, but relatively impressive for such a limited first two columns of Table 8. As before, we include both
number of monthly observations. the discrete (Column 1) and logarithmic (Column 2)
The evidence in this section speaks to the reason that specifications. In both columns, we see that all three
more lenient terms are awarded to personally connected varieties are negatively related to credit spreads, with
firms. One the one hand, bankers could gain value from the Social Connections Indicator having the largest point
cutting their friends good deals (i.e., on terms not justified estimate ( 13 bps versus 9 for the other types).
by the firm’s fundamentals or future prospects) and could Given the strong result for social connections, it is
therefore be willing to finance such private benefits with tempting to formulate causal explanations for the impact
their own shareholders’ money. On the other hand, of social connections on spreads similar to that for the
personal relationships could reduce monitoring costs or other types of connections. One could argue that because
information asymmetries, often cited as reasons that common social organizations provide a natural venue for
institutional lending might exist at all (e.g., Bernanke, relationships to persist into the future (school and third-
1983). party past professional connections have no comparable
We find no evidence that the favorable lending terms venue), that they would be particularly costly to damage.
extended to personally connected firms stem from agency In connected deals in which such valuable social relation-
problems on the part of bankers. Whether measured by ships are effectively pledged as collateral, we might
future stock returns or credit ratings, firms perform better expect larger marginal effects on credit spreads. While
after completing a deal with a personally connected consistent with the evidence in Table 8, so, too, is the
syndicate, suggesting that instead of facilitating poor possibility for banking transactions to influence – instead
deals, firm-bank connections appear to reduce the risk of being influenced by – the social connections we
faced by member banks. None of the evidence herein can observe. Without a way to distinguish between the two,
tell us whether personal connections allow syndicates to we interpret the effects of social connections as merely
choose better deals ex ante, or whether they allow suggestive evidence in support of the other connection
syndicate banks to monitor their borrowers more effi- variables.
ciently. While interesting, the distinction between
adverse selection and moral hazard is secondary to 5.2. Syndicate features
whether connected deals are better or worse, to which
the evidence in this section does speak. The majority of our control variables, like most studies
of capital structure, are defined at the firm level. Partly,
5. Robustness and other considerations this is because detailed data on financing’s supply side are
comparatively scarce. In situations in which frictions are
5.1. Connection types low and capital providers are relatively homogenous (e.g.,
bond markets), we would perhaps not expect lender-
Because we wish to make causal inferences between specific attributes to play an important role. This is less
personal relationships and lending behavior, we consider applicable to bank financing, where the ability to screen
connections formed only at third-party venues (school or and monitor borrowers could differ considerably between
other firms or banks not involved in the deal analyzed) banks. To the extent that such differences are correlated
and at least 5 years prior to the deal of interest. The time with our connection measures, the coefficients we report
restriction is imposed to rule out any reverse causality, could be biased.
such as membership in social organizations being a Perhaps the most obvious possibility is that larger or
reward for a favorable banking deal. Practically, this more active banks have scale economies that allow them
means that we ignore the majority of the possible con- to undercut their smaller counterparts. Moreover, because
nections we can infer. Connections exist not only from larger banks have more employees and directors, the
common schooling institutions or past workplaces, but expected number of personal connections with any bor-
also from active roles in common social organizations, rower is larger.20 In the third column of Table 8, we
e.g., think tanks (Council on Foreign Relations), charities exclude from consideration any deal in which any of the
(Saint Agnus Foundation), nonprofit organizations five most active banks was a participant. As seen, this
(National Urban League), and philanthropies (Boston restriction has an enormous impact on the number of
Science Museum). Including such connections confers a observations (11,003 in Table 3 versus 3,948 in Table 8,
marked increase in statistical power. Through sheer size, Column 3), reflecting the ubiquity of the most active
connections formed within the universe of social organi- commercial banks. Nonetheless, even when the largest
zations far outnumber those formed via common school- banks are absent, the effect of firm-bank personal con-
ing institutions and third-party workplaces. However, nections survives. The coefficient reported in Table 8
without being able to identify the specific dates when (0.13) is nearly identical to the full sample (0.12), and
such social relationships are formed (and thus leaving remains highly significant (P o0.001). Similar magnitudes
them vulnerable to the reverse causality critique), we are observed if the sample is cut even further, but as the
cannot defend their inclusion in our main analysis.
With this caveat in mind, we break up our existing 20
We have already addressed this possibility in some detail pre-
connection measure into its components (Third-Party Past viously, having controlled for the number of lenders in the syndicate as
Professional Connection Indicator and School Connection well as the aggregate lending activity of its member banks in all
Indicator) and to it add Social Connection Indicator in the regressions.
Table 8
Loan spreads and alternative definitions of connections.
The table relates All-in Drawn Spreads to borrower and lender personal connections, a set of controls for borrower fundamentals, lender characteristics,
loan characteristics, and macroeconomic conditions at time of loan origination, as well as a set of specified fixed effects. In column 1, the dependent
variable is numerical All-in Drawn Spreads; in columns 2 to 4, the dependent variable is its natural logarithm. In Column 3, we exclude all observations
involving busy syndicates, those that ranked in the Top 5 in terms of loan volume the previous year. In Column 4, we aggregate all observations but
include indicator variables for every bank in the Top 20 ranked by previous year deal volume. Robust standard errors clustered by firm are in parentheses.
n nn
, , and nnn represent significance at the 10%, 5%, and 1% level, respectively.
Dependent Variable:
Dependent Variable: Log(All-in Drawn Spreads)
All-in Drawn Spreads
(1) (2) (3) (4)
nnn
Past School Connections 9.152
(2.988)
Third-Party Past Professional Connections 8.723nn
(3.415)
Current Social Connections 13.92nnn
(3.411)
Log (1 þ Number of School Connections) 0.0699nn
(0.0295)
Log (1 þ Number of Professional Connections) 0.128nnn
(0.0161)
Log (1 þ Number of Social Connections) 0.0410nnn
(0.0144)
Log (1 þ Number of Connections) 0.126nnn 0.128nnn
(0.0363) (0.0140)

Seniority Fixed Effect Yes Yes Yes Yes

EDF Decile Fixed Effect Yes Yes Yes Yes
Industry Fixed Effect Yes Yes Yes Yes
Top 20 Bank Fixed Effect No N0 No Yes
Adjusted R2 0.506 0.622 0.457 0.639
number of observations decreases, so, too, does the ability believe that this measurement error is systematically
to make statistical inferences. related to unobserved, genuine connections, the esti-
The fourth column again considers the full sample but mated coefficients are biased to zero, implying lower
includes fixed effects for each of the 20 most active banks, bounds on any true relationship.
defined by the number of deals in the previous year (84% of
our observations include at least one of these banks).
Notably, their inclusion increases the explanatory power 5.4. The Higher bar hypothesis
increases almost 2 percentage points, indicating the pre-
sence of lender-specific attributes on credit spreads. How- We know from Tables 5 to 7 that connected borrowers
ever, the effect of bank-firm personal connections remains perform better ex post and from Tables 2 to 4 that this
virtually unchanged compared with the previous column or superiority is, at least partially, reflected by better deal
to Table 3, indicating an elasticity of slightly over 0.12 terms. However, this evidence alone does not necessarily
(Po0.001). Other unreported robustness checks include a tell us anything about efficiency, i.e., whether personal
larger number of fixed effects, or interacting previous years’ connections improve lending decisions. It could be the case
activity with firm-bank personal connections, none of which that personal relationships harm efficiency, but in such a
has a meaningful effect on the variable of interest. way that still generates the empirical patterns we observe.
To appreciate this possibility in more detail, consider the
5.3. Measurement error following simple model.21 There are two types of banks,
those personally connected (C) to their borrowers and those
All analysis involves proxies for personal connections not (N). Moreover, four types of firms request debt financing:
between firms and lenders. Never do we observe these 1, 2A, 2B, and 3. Type 1 and 3, good and bad firms,
relationships directly. Thus, when we include one’s school respectively, are easy to evaluate. Type 1 firms are always
classmates or past coworkers in a regression of lending
terms or ex post performance, we have certainly intro- 21
We thank Jeremy Stein for first suggesting the higher bar
duced errors-in-variables. Because we have no reason to alternative and the simple model that illustrates its intuition.
awarded credit (by either C or N), and type 3 firms are always 6. Conclusions
denied. By contrast, types 2A and 2B are harder to evaluate.
Perhaps their information is softer or their managerial quality A number of theories credit the very existence of banks
difficult to evaluate. In any case, 2A firms are positive net with screening or monitoring advantages relative to more
present value (NPV), and 2B firm are negative NPV. disperse creditors. Yet, what exactly is it about banks, and
Now consider what would happen if personal relation- some more than others, that confers them special ability to
ships solved the adverse selection problem faced by poten- manage such difficult borrowers? A banker’s answer to this
tial lenders. C banks, by virtue of their personal connections, question likely involves the word ‘‘relationship.’’ This paper
could distinguish between 2A (good) firms and 2B (bad) studies a specific kind of relationship: personal relationships
ones, and they would lend only to the former. N banks, between employees at firms and their lenders.
lacking the information required to make this distinction, We ask two related questions: (1) Do personal rela-
would either always or never lend, depending on the tionships lead to more favorable financing terms? (2) If so,
parameter values. In either case, it is easy to see how the are these decisions justified by ex post performance?
performance of a C bank’s borrowers could exceed that of an With detailed data on roughly 20 thousand syndicated
N bank: Good firms are always awarded credit, and bad loans by more than five thousand public US firms and
firms are always denied.22 This is consistent with the almost two thousand commercial banks, we find that the
evidence in the paper that connected borrowers perform answer to both questions is ‘‘yes.’’ Compared with syndi-
better ex post and are awarded better deal terms ex ante. cated deals in which the firm’s management (or directors)
Now consider an alternative possibility: Connections is not personally connected to any syndicate bank,
impart no special information, meaning that, like N banks, connected ones are associated with substantially lower
C banks cannot distinguish between types 2A and 2B interest rates, fewer covenants, and larger loan amounts.
firms. However, what if C banks are wary of the percep- The interest rate concessions depend on the borrower’s
tion of corruption and, consequently are reluctant to lend risk, with higher risk firms awarded larger rate reduc-
to firms that require a subjective evaluation—namely, all tions. Furthermore, after initiating a deal with a person-
type 2 firms? In this case, C banks loan only, or mostly, to ally connected syndicate, firms improve their credit
type 1 firms, which are objectively creditworthy. Assum- ratings and enjoy substantially higher stock returns. Thus,
ing that N banks still find it profitable to loan to type 2 the concessions in lending terms in connected situations
firms, and that the typical type 2 borrower is inferior to appear justified by ex post performance.
the typical type 1 borrower, the ex post performance of It is difficult to posit a plausible, noncausal interpreta-
C’s borrowers exceed that of N’s borrowers. tion for the role played by firm-bank personal connections
This is obviously not a formal model, but it does in the commercial loan market. By focusing exclusively on
illustrate how both an information story (where connec- personal relationships formed several years prior to the
tions improve lending decisions) and a higher bar story banking deals we analyze and at different venues from
(where they do not) are difficult to distinguish based on ex the borrower or lender, we exclude the possibility that
post performance alone. In the former, private information personal relationships are a product of existing or antici-
is generated equally to all potential lenders, but connected pated banking relationships.
borrowers simply use it differently. In the latter, personal Taken together, the evidence here identifies personal
connections change the information set of lenders. Clearly, relationships as a technology that allows banks to excel in
teasing out whether connections create private information problems situations, in which a borrower’s creditworthi-
or whether they effectively parse deals by levels of existing ness is difficult to evaluate or when active monitoring is
private information is difficult. Predictions for ex post required (Diamond, 1984, 1991; Fama, 1985). Examples of
performance outcomes are similar. this technology at work are microcredit groups such as
The higher bar story, in its simplest form, represents a the Grameen Bank of Bangladesh. There, borrowers are
friction for borrowers, with no offsetting benefit. Thus, if screened and monitored by members of their social circle,
there is a cost of applying for a loan, then any uncertainty which allows credit to be provided even in the absence of
about clearing the bar means that firm have an incentive collateral (Besley and Coate, 1995; Woolcock, 1998;
to avoid personally connected banks. In an extreme case, Yunus, 1993). In this market, personal relationships create
we should see only (or mostly) arm’s-length transactions value by implicitly monetizing social capital, making
and, regardless of how we define social connections in the tangible the information and reciprocity afforded mem-
current context, this is counterfactual. Unreported results bers of a social network. The evidence in this paper
indicate that personal relationships greatly increase the suggests that such a model can also act at the corporate
probability of a deal occurring, implying either that firms level. How firm-bank personal relationships alter lending
are more likely to approach cozy borrowers or are more terms over the life of a loan, such as following covenant
likely to be approved once they do apply (maybe both). violations or in renegotiation, we leave to future work.
Unlike the higher bar story, this is exactly what the
information story would predict.
22
Appendix A
If N firms find it profitable to lend to any type 2 firm, then C banks
differ only by winnowing out 2B firms. If N firms stay away from type 2
firms altogether, then the same result requires that the typical 2A firm Variable definitions and constructions are given in
be more attractive than the typical type 1 firm. Table A1.
Table A1
Variable definitions and construction.
Variable name Variable definitions and constructions Source of data
All-in Drawn Spreads All-in drawn spreads of each tranche Dealscan

Capital Expenditure/Total Assets Capital Expense (t)/Total Assets (t-1) Compustat
CAPM Beta Beta estimate from the capital asset pricing model (CAPM), using the past 36 CRSP
months of monthly returns, with a minimum of 18 months of return data
Change of Default Spread Change of default spread between current month and prior month Federal Reserve
Change of Term Spread Change of term spreads between current month and prior month Federal Reserve
Characteristics-Adjusted Return, 12 Cumulative Daniels, Grinblatt, Titmann, and Wermers (1997; DGTW) CRSP, /Compustat
Months Ahead characteristic-adjusted return 12 months ahead, beginning at the month
immediately after the deal
Characteristics-Adjusted Return, 24 Cumulative DGTW characteristic-adjusted return 24 months ahead, CRSP, Compustat
Months Ahead beginning at the month immediately after the deal
Characteristics-Adjusted Return, 36 Cumulative DGTW characteristic-adjusted return 36 months ahead, CRSP, Compustat
Months Ahead beginning at the month immediately after the deal
Deal in Past 1–3 Years Indicator Dummy variable that takes the value of one if the firm borrows from a bank Dealscan
in the syndicate during the prior three years
Deal in Past 4–6 Years Indicator Dummy variable that takes the value of one if the firm borrows from a bank Dealscan
in the syndicate during the four to 6 years before the current year
Deal in Past 7 Years or Earlier Dummy variable that takes the value of one if the firm borrows from a bank Dealscan
Indicator in the syndicate more than 6 years before the current year
Default Spread Difference between Moody’s BAA corporate bond index yield and Moody’s Federal Reserve
AAA corporate bond index yield
EDF Decile Fixed Effect Dummy variable that equals one if the EDF value falls into one of the ten EDF Moody’s KMV
deciles, in which EDF deciles are defined over the cross-sectional EDF values
within the month
EDF Implied Spread (EIS) Product of the estimated Expected Default Frequency and the estimated Moody’s KMV
expected loss given default (LGD)
Expected Default Frequency (EDF) Computed and calibrated to actual default events by Moody’s KMV (see Moody’s KMV
Crosbie and Bohn, 2003, for details)
Idiosyncratic Volatility Residual standard deviation of the estimate from the capital asset pricing CRSP
model using the past 36 months of monthly returns, with a minimum of 18
months of return data
Industry Fixed Effect Industry fixed effect, where the industry classification is defined by Fama and CRSP
French (1997) 30-industry classifications
Level of Term Spread Difference between the ten-year Treasury yield and the three-month Federal Reserve
Treasury yield
Local Bank Indicator Dummy variable that takes the value of one if a syndicate member bank is Hand-collected
located within 100 km of the borrower’s headquarters and zero otherwise
Log(Maturity) Logarithm of tenor length Dealscan
Log(Total Assets) Logarithm of Total Assets (AT) at (t) Compustat
M/B Market value of equity/book value of equity CRSP, Compustat
Number of Lenders Number of lenders within each syndicate Dealscan
Number of Loans Offered by Total number of nonoverlapping loans offered by syndicate members during Dealscan
Syndicate Prior Year the prior year
Profitability Operating Income Before Depreciation (t)/Total Assets (t-1) Compustat
Return [t-1, 0] Cumulative past 12-month raw return CRSP
Return [t-3, t-2] Cumulative past 36-month raw return excluding the most recent 12-month CRSP
return
Seniority Fixed Effect Dummy variable that takes the value of one if the loan is a senior loan, and Dealscan
zero otherwise
Tangibility/Total Assets [Property, Plant, and Equipment (PP&E) þ Inventory] (t)/Total Assets (t-1) Compustat
References Berger, A., Udell, G., 1995. Relationship lending and lines of credit in
small firm finance. Journal of Business 68, 351–381.
Acharya, V., Bharath, S., Srinivasan, A., 2007. Does industry-wide distress Bernanke, B., 1983. Nonmonetary effects of the financial crisis in propaga-
affect defaulted firms? Evidence from creditor recoveries. Journal of tion of the Great Depression. American Economic Review 73, 257–276.
Financial Economics 85 (3), 787–821. Besley, T., Coate, S., 1995. Group lending, repayment incentives and
Agrawal, D., Arora, N., Bohn, J., 2004. Parsimony in practice: an EDF- social collateral. Journal of Development Economics 46, 1–18.
based model of credit spreads. Technical Paper, Moody’s KMV. Bharath, S., Dahiya, S., Saunders, A., Srinivasan, A., 2011. Lending
Almeida, H., Philippon, T., 2007. The risk-adjusted cost of financial relationships and loan contract terms. Review of Financial Studies
distress. Journal of Finance 62 (6), 2557–2586. 24 (4), 1141–1203.
Altman, E., Kishore, V., 1996. Almost everything you wanted to know Bharath, S., Sunder, J., Sunder, S., 2008. Accounting quality and debt
about recoveries on defaulted bonds (Financial Analysts Journal contracting. Accounting Review 83, 1–28.
November/December, 57–64). In: Barnhill, T., Maxwell, W. (Eds.), Bharath, S., Dahiya, S., Saunders, A., Srinivasan, A., 2007. So what do I
High Yield Bonds: Market Structure Portfolio Management and get? The bank’s view of lending relationships. Journal of Financial
Credit ModelsMcGraw-Hill, New York, pp. 1999. Economics 85, 368–419.
Bohn, J., Crosby, P., 2003. Modeling default risk. Technical Paper, Hart, O., Moore, J., 1994. A theory of debt based on the inalienability of
Moody’s KMV. human capital. Quarterly Journal of Economics 109, 841–879.
Boot, A., Thakor, A., 1994. Moral hazard and secured lending in an Hoshi, T., Kashyap, A., Scharfstein, D., 1990. The role of banks in reducing
infinitely repeated credit market game. International Economic the costs of financial distress in Japan. Journal of Financial Econom-
Review 35 (4), 899–920. ics 27 (1), 67–88.
Brickley, J., Linck, J., Smith, C., 2003. Boundaries of the firm: evidence from Hovakimian, A., Kayhan, A., Titman, S., 2009. Credit rating targets.
the banking industry. Journal of Financial Economics 70, 351–383. Working Paper, Baruch College and University of Texas, Austin,
Cai, J., 2010. Competition or collaboration? The reciprocity effect in loan unpublished.
syndication. Working Paper, Federal Reserve Bank of Cleveland, Kisgen, D., Strahan, P., 2009. Do regulation on credit ratings affect a
unpublished. firm’s cost of capital? Working Paper 14890, National Bureau of
Chava, S., Roberts, M., 2008. How does financing impact investment? The Economic Research, Boston, MA, unpublished.
role of debt covenants. Journal of Finance 63 (5), 2085–2121. Kroszner, R., Strahan, P., 2001. Bankers on boards: monitoring, conflicts
Charumilind, C., Kali, R., Wiwattanakantang, Y., 2006. Connected lend- of interest, and lender liability. Journal of Financial Economics 62 (3),
ing: Thailand before the financial crisis. Journal of Business 79, 415–452.
181–218. La Porta, R., Lopez-de-Silanes, F., Zamarripa, G., 2003. Related lending.
Collin-Dufresne, P., Goldstein, R., Martin, J., 2001. The determinants of Quarterly Journal of Economics 118 (1), 231–268.
credit spread changes. Journal of Finance 56, 2177–2207. Laeven, L., 2001. Insider lending and bank ownership: the case of Russia.
Cohen, L., Frazzini, A., Malloy, C., 2008. The small world of investing: Journal of Comparative Economics 29 (2), 207–229.
board connections and mutual fund returns. Journal of Political Lin, H., Paravisini, D., 2010. What’s bank reputation worth? The effect of
Economy 116 (5), 951–979. fraud on financial contracts and investment. Working Paper, Colum-
Coval, J., Moskowitz, T., 2001. The geography of investment: informed bia University, unpublished.
trading and asset prices. Journal of Political Economy 109, 811–841. Modigliani, F., Miller, M., 1958. The cost of capital, corporation finance and
Daniel, K., Grinblatt, M., Titman, S., Wermers, R., 1997. Measuring the theory of investment. American Economic Review 48 (3), 261–297.
mutual fund performance with characteristic based benchmarks. Morck, R., Nakamura, M., 1999. Banks and corporate control in Japan.
Journal of Finance 52 (3), 1035–1058. Journal of Finance 54 (1), 319–339.
DeAngelo, L., 1981. Auditor size and audit quality. Journal of Accounting Nini, G., Smith, D., Sufi, A., 2009. Creditor control rights and firm
and Economics 1, 183–199. investment policy. Journal of Financial Economics 92 (3), 400–420.
Degryse, H., Ongena, S., 2005. Distance, lending relationships, and Petersen, M., Rajan, R., 1994. The benefits of lending relationships:
competition. Journal of Finance 60, 231–266. evidence from small business data. Journal of Finance 49 (1), 3–37.
Degryse, H., Van Cayseele, P., 2000. Relationship lending within a bank- Qian, J., Strahan, P., 2007. How laws and institutions shape financial
based system: evidence from European small business data. Journal contracts: the case of bank Loans. Journal of Finance 62 (6),
of Financial Intermediation 9 (1), 90–109. 2803–2834.
Diamond, D., 1984. Financial intermediation and delegated monitoring. Rajan, R., Zingales, L., 1998. Which capitalism? Lessons from the East
Review of Economic Studies 51, 393–414. Asian crisis. Journal of Applied Corporate Finance 11 (3), 1–11.
Diamond, D., 1991. Monitoring and reputation: the choice between bank Roberts, M., Sufi, A., 2009a. Control rights and capital structure: an
loans and directly placed debt. Journal of Political Economy 99 (4), empirical investigation. Journal of Finance 64 (4), 1657–1695.
689–721. Roberts, M., Sufi, A., 2009b. Renegotiation of financial contracts: evi-
Drucker, S., Puri, M., 2009. On loan sales, loan contracting, and lending dence from private credit agreements. Journal of Financial Econom-
relationships. Review of Financial Studies 22 (7), 2835–2872. ics 93, 159–184.
Dvorak, B., 2008. A brief history of active credit portfolio management. Ross, S., 1985. Debt and taxes and uncertainty. Journal of Finance 40,
Technical Paper, Moody’s KMV. 637–658.
Faulkender, M., Petersen, M., 2006. Does the source of capital affect Schmidt, B., 2008. Costs and benefits of ‘‘friendly’’ boards during mergers
capital structure? Review of Financial Studies 19 (1), 45–79 and acquisitions. Working Paper, University of Southern California,
Fama, E., 1985. What’s different about banks? Journal of Monetary unpublished.
Economics 15, 29–39 Schenone, C., 2010. Lending relationships and information rents: do
Fama, E., French, K., 1989. Business conditions and expected returns on banks exploit their information advantage? Review of Financial
stocks and bonds. Journal of Financial Economics 25, 23–49. Studies 23 (3), 1149–1199
Fama, E., MacBeth, J., 1973. Risk, return and equilibrium: empirical tests. Stein, J., 2003. Agency, information and corporate investment. In: Con-
Journal of Political Economy 81, 607–636. stantinides, G., Harris, M., Stulz, R. (Eds.), Handbook of the Economics of
Fracassi, C., 2008. Corporate finance policies and social networks. Work- Finance, vol. 1. North-Holland, Amsterdam, pp. 111–165.
ing Paper, University of California, Los Angeles, unpublished. Sufi, A., 2007. Information asymmetry and financing arrangements:
Fracassi, C., Tate, G., 2009. External networking and internal firm evidence from syndicated loans. Journal of Finance 62, 629–668.
governance. Working Paper, University of California, Los Angeles, Woolcock, M., 1998. Social capital and economic development: toward a
unpublished. theoretical synthesis and policy framework. Theory and Society 27,
Güner, B., Malmendier, U., Tate, G., 2008. Financial expertise of directors. 151–208.
Journal of Financial Economics 89 (1), 20–43. Yunus, M., 1993. Reflections on the essentials of Grameen system. In:
Gupta, A., Singh, A., Zebedee, A., 2008. Liquidity in the pricing of Conference Report and Conference Papers. Going Forward, The Peer
syndicated loans. Journal of Financial Markets 11, 339–376. Group Lending Exchange. Hot Springs, Arkansas; Toronto, Canada.

Friends With Money

Uploaded by

Copyright:

Available Formats

Friends With Money

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Friends With Money

Uploaded by

Copyright:

Available Formats

Journal of Financial Economics 103 (2012) 169–188

Contents lists available at SciVerse ScienceDirect

Journal of Financial Economics

Friends with money$

1. Introduction violations of the Modigliani and Miller (1958) perfect

0304-405X/$ - see front matter & 2011 Published by Elsevier B.V.

Mean Median Standard deviation 10th Percentile 90th Percentile

Panel A: Deal characteristics

Panel B: Connection measures

Panel C: Firm characteristics

Dependent variable: All-in Drawn Spreads

Connected Indicator 27.68nnn 8.452nn 20.12nnn 51.11nn 46.52nnn

(1) (2) (3) (4)

3.2. Covenants The results so far indicate that ﬁrm-bank personal

Panel A: Firm-bank connections, loan covenants

Dependent variable: Covenant Indicator Dependent variable: Number of Covenants

(1) (2) (3) (4)

Connected Indicator 0.0124 0.0715

Panel B: Firm-bank connections and loan sizes

log(Tranche Amount) log(Tranche Amount) log(Tranche Amount) log(Tranche Amount)

(1) (2) (3) (4)

Connected Indicator 0.134nnn 0.147nnn

(1) (2) (3) (4) (5) (6)

nnn nnn nnn

4.3. Stock returns

Panel A: Connections and Firm’s Future Expected Default Frequencies (EDF)

Dependent variable: Expected Default Frequency

EDF 12 months-ahead EDF 24 months-ahead EDF 36 months-ahead

Panel B: Connections and ﬁrm’s future EDF Implied Spreads

Dependent variable: EDF Implied Spread

EIS 12 months-ahead EIS 24 months-ahead EIS 36 months-ahead

Dependent variable: Return at Different Horizons

12-months ahead 24-months ahead 36-months ahead

Connected Indicator 0.0344n 0.106nnn 0.170nnn

Dependent variable: Return at Different Horizons

12-months ahead 24-months ahead 36-months ahead

(1) (2) (3) (4)

Firm characteristics controls Yes Yes Yes Yes

Seniority Fixed Effect Yes Yes Yes Yes

Variable name Variable deﬁnitions and constructions Source of data

All-in Drawn Spreads All-in drawn spreads of each tranche Dealscan

You might also like