Anand 2006

THE JOURNAL OF INDUSTRIAL ECONOMICS 0022-1821
Volume LIV June 2006 No. 2
RELATIONSHIPS, COMPETITION AND THE STRUCTURE

OF INVESTMENT BANKING MARKETS
Bharat N. Anandw
Alexander Galetovicz
It is well known that competition can destroy incentives to invest in firm-

specific relationships. This paper examines how the tension between
relationships and competition is resolved in the investment banking
market, which for decades has been characterized by both relationships
and competition. The model studies the impact on relationships of four
different dimensions of competition: non-exclusive relationships,
competition from arm’s-length intermediaries, non-price competition,
and endogenous entry. The analysis shows how market equilibrium
adjusts so that relationships are sustained in the face of such
competition. Banks are shown to establish relationships without either
local or aggregate monopoly power. The model rationalizes two distinct
empirical regularities of market structure: the invariance of market
concentration to market size; and a pyramidal market structure with an
oligopoly comprising similar-sized players at the top and a large number
of small banks at the bottom. The analysis may also shed light on the
industrial organization of other professional service industries.
I. INTRODUCTION
THE INDUSTRIAL ORGANIZATION OF PROFESSIONAL SERVICE MARKETS has been

largely ignored. A central feature of these markets is the existence of client-
firm relationships that are costly to establish and difficult to appropriate.1
But, professional service firms also compete vigorously in many dimensions.
This raises an intriguing question: how are the incentives to invest in

We thank two anonymous referees, an editor, colleagues and seminar participants at
various universities, the Econometric Society Meetings (Buenos Aires) and the National
Bureau of Economic Research (Summer Institute meetings) for helpful comments. Anand
gratefully acknowledges the financial support of the Division of Research at Harvard Business
School. Galetovic gratefully acknowledges the financial support of Fondecyt (project
C1970340), Fundación Andes, the Hewlett Foundation and the Mellon Foundation.
wHarvard University, Soldiers Field Road, Boston, Massachusetts, 02163 U.S.A.

e-mail: banand@hbs.edu
zUniversidad de los Andes, Av. San Carlos de Apoquindo 2000, Santiago, Chile.
e-mail: agaletovic@uandes.cl
1
One reason is that relationships are often embodied in human capital that can move
between firms, taking these relationships with them.
r Blackwell Publishing Ltd. 2006, 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA.
151
152 BHARAT N. ANAND AND ALEXANDER GALETOVIC
relationships preserved in the face of competition? Indeed, a key insight from

other relationship-based markets like commercial banking is that competi-
tion should weaken incentives to invest in relationships, and that some
market power is needed to preserve these investments (see, for example,
Petersen and Rajan [1994]). This paper examines how the tension between
relationships and competition is resolved in a particular professional service
industry: investment banking. The analysis may shed light on several
outstanding puzzles about this and related industries.
Our focus on the U.S. investment banking market is motivated by the fact
that there is a vast literature establishing that, for many decades, it has been
characterized by both relationships and competition.2 Competition occurs
along four dimensions. First, there are a large number of banks (more than
1,000). Second, each firm is also involved with many different banks which
offer similar services, so there is no exclusive dealing. Third, regulation does
not ‘contaminate’ the observed market structure and there is free entry and
exit. Fourth, industry accounts typically argue that there is brutal nonprice
competition.
Some observers have also pointed out, however, that competition in prices
is soft. Firms rarely choose underwriters through competitive bidding, and
investment banks cooperate on each deal via syndicates. Moreover, prices
are unusually stable, non-negotiable and, importantly, do not depend on
firm- or deal-specific characteristics: Matthews ([1994] p.161) notes that
spreads on high-quality, long-term corporate bonds have been 7/8% of
capital raised for several decades. Similar practices are observed in Britain,
where, for decades, underwriting fees have been equal to 1.25% of capital
raised.3 And, recently, Chen and Ritter [2000] documented that 90% of
initial-public-offerings deals between 1995 and 1998 that raised between $20
and $80 million had gross spreads of exactly 7%.4
2
Relationships between investment banks and corporations have always been important in
the U.S. investment banking market, and the sunk costs incurred by investment banks in
establishing and maintaining each relationship are large; see Nanda and Warther [1998]
Wilhelm [1999] and Wilhelm and Downing [2001].
3
See ‘Some Old Peculiar Practices in the City of London,’ The Economist, February 18th,
1995.
4
Rolfe and Troob ([2000] p.103) argue in a colorful recent account of investment banking
that spreads have stayed high ‘. . . because there has always been an unspoken agreement
among the bankers that when it comes to underwritings they won’t compete on price. The
spreads are sacrosanct. He who cuts spreads will himself become an outcast. . .The community
of investment banks has always been small enough so that if one bank were to break ranks on
the pricing issue, the others would quickly join forces and squash the offender . . . Every banker
knows that the pricing issue is a slippery slope best avoided because once the price cutting
begins, there’s no telling where it will end.’ See also ‘Overcharging Underwriters,’ The
Economist (June 27, 1998), where it is noted that ‘. . .studies in both countries suggest issuing
companies are overcharged, and that they are stung for more in America.’ Similar attributions
to bankers can be found elsewhere, as noted by Chen and Ritter ([2000] p. 1,106). For an
empirical analysis of the IPO market see Hansen [2001].
r Blackwell Publishing Ltd. 2006.
RELATIONSHIPS AND COMPETITION IN INVESTMENT BANKING 153
C8 Ratio Invol
1 11
1986
.8
1950
1960 1980
1955 1975
volume (in logs)

1965 1970 10
.6
C8 Ratio
.4
9
.2
0 8
1950 1986
year
Figure 1
Concentration and Volume in Underwriting
Source: Hayes, Spence, and Marks [1983], Table 1, and Eccles and Crane [1988], table 5.4.
‘C8-Ratio’ is the share of total volume of securities underwritten in any given year by the top eight
investment banks. Full credit is given to lead manager. ‘Volume’ is the logarithm of total volume of
securities underwritten in any given year (volume data is in real terms). Volume increased
seventeen-fold between 1950 and 1986.
These facts raise interesting questions. First, how does competition

impact relationships? Second, how do the various forms of competition
interact? For example, how is soft price competition sustained with a large
number of banks, and entry; and, does non-price competition dissipate
rents? Third, how can these facts on firm conduct be reconciled with those on
market structure? Specifically, the structure of investment banking is both
remarkably stable over time, and displays the following salient features:
Concentration is unrelated to market size: Market size has grown

significantly over time but concentration has not changed. Consider, for
example, Figure 1, which plots the C8 ratio (left vertical axis) and volume
underwritten (right vertical axis, in logs) between 1950 and 1986. Market
size grew about twenty fold, but concentration barely changed (and even
slightly increased). As another example, Figure 2, which plots the same
concentration ratio for mergers and acquisitions between 1987 and 1998,
shows the same regularity.
A two-layered pyramidal structure: with an oligopoly of similar-sized
players at the top, few mid-sized banks, and a large number of small
banks. Since the late nineteenth century, a group of between six and ten
banks with similar market shares (the so-called ‘bulge-bracket’ banks)
sumtop8v Intotval
1 14.2
.8
volume (in logs)

1987 1988 1991 1998
.6 1992
1990
C8 Ratio
1989 1993 1995 1996 1997

1994
.4
.2
0 11.5
1987 1998
year
Figure 2
Concentration and Volume in Mergers and Acquisitions
Source: Author’s processing of data from Securities Data Company. ‘C8-Ratio’ is the share of total
deal value of mergers and acquisitions brokered by the top eight investment banks in any given
year. Full credit is given to the acquiror’s lead bank. The sample of M&A deals is restricted to those
made by firms that do at least three such deals in the 12-year period 1987–1998. Maximum and
minimum volume over this time period differ by a factor of eight.
has consistently underwritten more than 70% of securities issued.

According to Hayes et al. ([1988] p.17): ‘As at virtually every stage of its
evolution, investment banking exhibited an oligopolistic industry
structure that was roughly pyramidal in shape, with a handful of
powerful firms at the apex.’
To explain these regularities, we start from the observation that
relationships are characterized by what Eccles and Crane [1988] call the
‘loose linkage’ between relationship costs and deal revenues: each
investment bank incurs costs to set up and nurture a relationship, but does
not directly charge for it nor receive any contractual assurance that it will be
selected by the firm on its deals. Instead, an investment bank collects fees
only when it does a deal.5 But then how can investment banks be assured that
the fees charged ex-post will cover the ex-ante cost of relationships?
We study this question with a model where infinitely-lived investment
banks sell services, which we call ‘deals,’ to firms. A firm is characterized by
its deal size v. Banks can employ one of two technologies to ‘do deals’: an
arm’s-length technology or a relationship technology. All banks can do
arm’s-length deals, but to use the relationship technology, a bank has to sink
5
By ‘deal’ we mean, for example, a security flotation, a merger or an acquisition.
$
βν
λc ν
kR
νβ ν Volume
-
Figure 3
The cost of doing a deal with alternative intermediation technologies.
a one-time entry cost, E. We call banks who invested in the relationship

technology, relationship banks; banks who only can do arm’s length deals are
termed fringe banks.
The cost of implementing an arm’s-length deal increases linearly in deal
size (i.e., a deal of size v costs bv), as seen in Figure 3. By contrast, each bank-
firm relationship involves a cost R which is independent of v and is sunk by
the time the deal is done and fees are collected. A second feature of
relationships is that they are non-verifiable, so that banks cannot directly
charge for the sunk costs R. Instead, they can only charge fees when they do a
dealFi.e., there is loose linkage.6 Third, relationships are not excludable:
banks can free-ride on the information gathered through others’ relation-
ships, so that the deals for a firm that has relationships can be implemented
at a very low cost by other relationship banks who did not sink R.7 Last, each
firm establishes k 4 1 relationships.
The first set of results is about competition between relationship and
fringe banks. Does it erode the incentives to establish relationships? The key
result is that scale economies inherent to relationships provide relationship
banks with a cost advantage when providing services to large firms, because
6
The assumption that relationships are nonverifiable distinguishes this setting from the large
body of work that studies investment incentives, incomplete contracting and the hold-up
problem (see Hart [1995]). In those settings, ex-post hold up can be moderated by specific ex-
ante allocation of decision rights over the assets: examples include the allocation of ownership
rights over the asset as in Grossman and Hart [1986] and Hart and Moore [1990], or the
contractual right of one party to block the use of the asset in a transaction with a third party, as
in Segal and Whinston [2000].
7
Relationship banks can free ride on each other’s efforts by copying their financial products,
poaching competitors’ employees, or by getting information on rivals’ ideas from client firms
themselves. Section 2.1 discusses these sources of free-riding in more detail.
R o bv for firms whose v is large enough. Competition from fringe banks will
not threaten relationships because it cannot: any fee that makes profits for a
fringe bank also makes ex-post profits for a relationship bank. Hence
relationship banks can and do charge lower fees in equilibrium. It follows
that any firm, regardless of deal volume, would prefer to do its deals with
relationship banks. However relationship banks would make losses for firms
with small deal volumes. Thus, all firms want relationships, whereas
relationship banks ration smaller firms out. In other words, the market will
be vertically segmented: relationship banks will serve large firms and fringe
banks will serve small firms. This separation is similar to the common
distinction in practice between ‘bulge bracket’ banks, which serve
predominantly large corporations, and the rest, which serve smaller firms.
The second set of results characterizes competition among relationship
banks. To recoup the investments in relationships made each period, price
competition must be soft in equilibrium, which is sustainable with an
implicit contract between relationship banks not to undercut. This implicit
contract yields an oligopoly such that relationship banks have similar
market shares and concentration cannot fall below a given bound. In
addition, we show that this lower bound on concentration does not fall as
market size increases. Last, because the implicit contract serves an efficiency
role, it is robust to entry despite yielding rents to banks in equilibrium.8,9
Imperfect competition can be traced to the non-verifiability and non-
excludability of relationships. We show that if banks could charge R directly,
the relationship segment could be perfectly competitive and there would be
no tension between relationships and competition despite scale economies
from relationships. The reason is that scale economies exist only at the level
of each bank-firm relationship, i.e., at the ‘local’ level. On the other hand,
aggregate relationship costs increase (linearly) with the number of relation-
ships that the intermediary establishes. We also show that nonexcludability
of relationships is necessary to explain why the investment banking market is
an oligopoly. If, as a counterfactual, relationships were nonverifiable but
excludable, implicit contracts would still be needed to soften price
competition, but would not impose any restrictions on aggregate market
structure.
8
The study of self-enforcing norms in the relationship segment is methodologically related to
Dutta and Madhavan [1997] who study implicit collusion in broker-dealer markets. The
collusive equilibrium in that model rationalizes a striking series of practises which have been
empirically documented. Unlike that model, the implicit contract here is not a purely
facilitating device, but supports rents that are necessary for a relationship segment to exist.
Consequently, a self-enforcing norm is sustainable in equilibrium even with free entry.
9
Pichler and Wilhelm [2001] obtain a similar efficiency result in their analysis of investment
banking syndicates. They study how the relationships and the potential for free-riding on
information-gathering, shapes syndicates. Membership stability across deals creates entry
barriers that provide banks with quasi-rents, which stimulate investments in information
production.
The comparative static exercises in section 3 distinguish changes in the
number of banks from a change in the intensity of price competition.10 The
general message from these exercises is that when competition gets more
intense, the endogenous adjustment in fees or in the number of banks undoes
their deleterious effects on the incentives to establish relationships. In
addition, vertical segmentation suggests caution when assessing how
concentrated or competitive the investment banking industry is. Specifically,
market segmentation implies that looking at the industry as a single market
will unduly deflate traditional measures of concentration (e.g., the inverse of
a Herfindahl index). Section 4 discusses additional welfare implications.
Our paper is related to a large literature on relationships in financial
markets. Petersen and Rajan [1995] were the first to point out that imperfect
competition is necessary to maintain relationships.11 That conclusion still
holds here, but the analysis shows why neither local monopoly power nor
aggregate monopoly is necessary to establish relationships. Aggregate
monopoly power is not necessary because relationships imply local scale
economies, not aggregate ones. Local monopoly power is not necessary
because endogenous entry and exit together with soft price competition can
undo the deleterious effect of multiple relationships. This ‘possibility result’
also contrasts with previous studies which have generally posited the need
for either exclusive relationships (see, for example, the discussion in Hellwig
[1991]) or aggregate market power (following Petersen and Rajan [1995]).
Free-riding problems in information production by investment banks and
the facilitating role of market structure are also studied in Benveniste et al.
[2002]. Market power allows banks to bundle IPOs in the same industry, that
are sold to a similar investor pool. This in turn allows the ‘smoothing’ of
underpricing across a wave of IPOs. The burden of compensating investors
for costly information production is then shared more equitably between
pioneers and followers. Empirical support for the theory is provided in
Benveniste et al. [2003]. The analysis in that pair of papers takes the existence
of market power in investment banking as given, whereas this study
examines where market power comes from.
This paper is also related to Anand and Galetovic [2000], which presents a
model to explain why intermediaries may finance the production of assets
10
This distinction follows Sutton [1991] and others who note that in models with endogenous
entry the number of economic actors is an uninformative measure of competition.
11
See also Mayer [1988] and Hellwig [1991] for conceptual tratments of this issue. There is a
large theoretical literature that explores the benefits and costs of relationships; for surveys of
this literature see Berger [1999], Boot [2000] and Ongena and Smith [2000]. For example,
Berglöf and von Thadden [1994], Boot and Thakor [1994], Chemmanour and Fulghieri [1994],
Diamond [1991], Rajan [1992] and von Thadden [1995] all model the benefits of long-term
bilateral relationships. Several papers in the literature have also studied the cost of exclusive
relationships that come from the exploitation of market power when banks can hold up firms
(e.g., Greenbaum et al. [1989], Rajan [1992], Sharpe [1990] and von Thadden [1998]).
over which they cannot establish property rights (such as information and
certain types of human capital). That paper shows that investment in non-
excludable assets requires soft price competition, which may be the
equilibrium of a repeated game between intermediaries. By contrast, the
focus in this paper is the tension between relationships and competition. While
soft price competition is a feature of investment banking, most observers
describe this industry as brutally competitive in almost all other dimensions.
In this model we study these other dimensionsFfirms have multiple
relationships, there is a large fringe of banks that do not establish relationships
in equilibrium, and relationship banks can exert sales effortsFand
endogenously derive market structure. We show why relationships are not
deleteriously impacted by price competition from fringe banks (because of
vertical segmentation), nor by entry and non-price competition (because
market structure endogenously adjusts to preserve investment incentives).
Finally, our paper is related to Boot and Thakor [2000], who study how
commercial banks that make relationships are affected by competition from
banks that do not (‘arm’s-length banks’).12 Both the characteristics and the
consequences of bank-client relationships are different in their model from
the one studied here. The reason for these differences is, quite simply, that
the source of relationships in the two markets are different. We discuss in
more detail these differences between commercial and investment banking,
and its consequences for how the tension between relationships and
competition is solved, in section 4.2.
The rest of the paper is organized as follows. Section 2 documents the
importance of relationships in investment banking, motivates the formal
model and shows the well known tension between competition and
relationships. Section 3 describes how the tension between relationships
and competition is solved, and presents the results of the model. Section 4
discusses several extensions and robustness checks. Section 5 discusses other
industries where relationships are important, and their similarities and
differences from investment banking. Section 6 concludes.
II. THE MODEL
II(i). Motivation
II(i)(i). Relationships in Investment Banking Relationships between
investment banks and firms provide banks with access to firm-specific
information that can be used to structure deals or price securities. Crane and
12
Yafeh and Yosha [2001] have also recently studied how competition from arm’s length
loans affect relationship lending. Their focus, however, is not on the canonical intertemporal
problem caused by sunk relationship investments, but on intratemporal competition between
the arm’s-length and relationship segments and the strategic use of relationships as an entry-
deterrence device.
Eccles ([1993] p. 136) note that ‘access and information exchange are the key
elements in the definition of relationships’ between investment banks and
client firms. Consequently, suppliers with relationships often have ‘preferred
vendor status’ because without such information they would, for example,
be ‘making virtually random blue-book pitches with little chance of hitting
the target.’13 Similarly, in a recent survey, Boot ([2000] p. 7) points out that
even the task of underwriting public issues involves absorbing credit
and placement risk which may be ‘facilitated by the proprietary information
and multiple interactions that are the hallmark of relationship banking.’
And, Wilhelm and Downing [2001] note that while changes in information
technology might commoditize those investment banking services that have
to do mostly with the storage and dissemination of information to investors,
the information needed for corporate advisory services still rests largely on
bank-firm relationships.
Relationships are well-documented for the U.S. market.14 Until about 25
years ago, the rule in the industry was that a corporation would establish a
relationship with only one investment bank.15 While relationships have
varied in strength over time, they still remain important today.16 In a recent
Institutional Investor survey of 1,600 chief financial officers of firms made in
August, 2001, 44% of those who prefer ‘specialized institutions’ for their
different needs, and 64% of those who prefer one-stop banks, stated that
their primary reason for choosing a bank was ‘prior relationships’ with it.
Moreover, evidence on firms’ choices of investment banks points indirectly
to the strength of relationships as well. For example, Baker [1990] examined
ties between investment banks and corporations with market value of more
than $50 million between 1981 and 1985. He reports that the 1,091
corporations that made two or more deals during this period used three lead
banks on average (these firms made eight deals on average). All but nine
granted more than 50% of their business to their top three banks and, on
average, 59% of the business was allocated to the top bank. Similarly, Eccles
and Crane ([1988] ch.4) report that among the 500 most active corporations
in the market between 1984 and 1986, 55.6% used predominantly one bank
to float their securities, and the rest maintained relationships with only a few
banks. They did not find any corporation selecting underwriters on a deal-
by-deal basis. James [1992] finds that in the first common stock security
13
Crane and Eccles ([1993] pp. 131–136).
14
See Wihelm and Downing ([2001]) for an overview.
15
Eccles and Crane ([1988] pp. 53 and 54) attribute the end of exclusive relationships to the
increasing sophistication of capital markets. Specifically, the capital markets offer more
product variety and, as firms have themselves become financially more sophisticated, they now
encourage solicitations from more than one bank.
16
See Nanda and Warther [1998] for an analysis of the trends in the strength of underwriting
relationships. Crane and Eccles ([1993] p.132) note that relationships were even more
important in the early 1990’s than they were in the previous decade.
offering after an initial public offering (IPO), 72% of firms choose the same
lead bank as before; for debt offerings, 65% of issuers do not switch banks.
And, Krigman et al. [2001] show that 69% of firms that made an IPO
between 1993 and 1995 and a secondary equity offering within three years of
the IPO, chose the same lead underwriter in both transactions.
It has been argued that relationships may subject the firm to a hold up
from the intermediary with whom it has a relationship. As we will see now,
however, the opposite seems to be more relevant in the case of investment
banks: firms may find it too easy to switch investment banks once they have
established the relationship.17
II(i)(ii). The Technology of Relationships Three characteristics of

relationships have been extensively documented. First, firms and investment
banks interact constantly in the course of a relationship, but the bank is paid
only when a deal is made. Eccles and Crane [1988] call this the ‘loose linkage
between costs and fees.’ It implies that investment banks recoup the costs
incurred to set up and nurture a relationship only if selected to do a deal. As
Eccles and Crane ([1988] pp. 39–40) point out, one reason for loose linkage is
that it is difficult for business firms to evaluate the quality of the advice
provided, unless deals are done. In other words, it is not merely an industry
practise that can be changed, but a technological feature of relationships
which stems from non-verifiability.18
Second, banks incur substantial costs to establish and nurture a
relationship and these are sunk by the time they compete for doing the
deal for a given firm. As Crane and Eccles ([1993] p. 142) point out, ‘. . . the
strategy of investment banks [is] to incur substantial costs in delivering
valueFin the form of advice, special studies, and market informationFas a
way of creating obligations that are hopefully converted into transaction
fees in the future.’ Moreover, a significant fraction of this sunk cost is
incurred by the investment bank because most of the exchange of
information takes place through direct interaction with the bank’s staff
person (often referred to as a ‘relationship manager’).
Third, investment banks often cannot establish property rights because
information is not excludable.19 To begin, because most of the exchange of
17
Ongena and Smith [2001] study the duration of firm-bank relationships in Norway and
find that firms are more likely to leave a given bank as the relationship matures, thus suggesting
that firms do not get locked into relationships in commercial banking either.
18
Indeed, because banks ‘are willing to incur current costs in the hope of getting future fees,
[t]his gives the customer an opportunity to receive services that he or she may never have to pay
for’ (Crane and Eccles [1993] p. 143). The extreme case of loose linkage is the ‘analysis’ function
of investment banks, where banks earn most of their commisions from investors who trade the
firm’s security.
19
A good or service is excludable if the owner can prevent others from using it at a very low
cost.
information takes place through direct interaction between the firm and the
investment bank’s staff person, the relationship-specific knowledge often
walks with employees when they are hired away.20 As an example, Deutsche
Bank built a global investment bank in a year (Deutsche Morgan Grenfell)
by hiring away staff en masse from other major banks.21 At the same time,
Eccles and Crane ([1988] pages 61–62) note that banks often fear that firms
will take their ideas to be implemented by rival banks for less money.
Copying is quite easy: Tufano [1989] notes that most product innovations by
banks are copied by rivals within a day of introduction.22 And, as is the case
in commercial banking, most firms have more than one relationship.23
II(ii). Model Description

There are many identical and risk-neutral investment banks and a
continuum of firms of measure f which wants to do deals. Each firm is
described by its deal volume v which is uniformly distributed in the interval
½0; v, with
Rv density function gðvÞ ¼ f 1v and corresponding cdf
1
GðvÞ ¼ 0 f vdu ¼ f vv. Thus f vv is the measure of firms whose deal volume
is at most v.
Banks can implement deals using either a relationship or an arm’s-length
technology. All banks can do arm’s-length deals, but to use the relationship
technology, a bank must sink an entry cost E once. Banks who invest in this
technology will be called ‘relationship banks.’ Banks who don’t will be called
‘fringe banks.’24
Firms that do deals with a relationship bank establish k 4 1 relationships
(i.e., relationships are not exclusive). Following Eccles and Crane [1988] we
call this the firm’s group of ‘k core banks.’ For simplicity, k is assumed to be
exogenous and the same for all firms (in section 3.5 we discuss why this
assumption is not restrictive). Note that we assume away any differentiation
among core relationship banksFany of the k banks is as good as any other
20
See Anand and Galetovic [2000].
21
There are other examples as well: Wilhelm and Downing [2001] note that when ‘Bruce
Wasserstein and Joseph Perella walked away from First Boston’s top-ranked M&A business in
1988 . . . virtually overnight, First Boston fell from the ranks of serious contenders for new
M&A business’ (pages 68–69).
22
Tufano [1989] estimates the costs of designing a security, including product development,
marketing and legal expenses, to be between $0.5 million and $5 million. These products cannot
be patented and all details become publicly available once the offering is filed with the SEC. For
models of product innovation in investment banking with weak property rights, see
Bhattacharya and Nanda [2000] and Persons and Warther [1997].
23
See Eccles and Crane [1988]. Moreover, in their survey on relationships in commercial
banking, Ongena and Smith [2000] conclude that multiple relationships are a common feature
of nearly all countries for which evidence has been collected.
24
Explicitly modeling the entry cost E also allows us to endogenize the number of
relationship banks m later on.
to do the deals of the firm. Figure 3 shows the investment banking market
from the perspective of a firm with relationships.
Each time a bank establishes a relationship with a given firm, it incurs a
sunk cost R which is independent of volume v. Hence, there are scale
economies at the level of each relationship (or local level). Banks incur no
sunk cost when they do an arm’s-length deal, but such a transaction imposes
a transaction cost on firms. The magnitude of the transaction cost depends
on (i) whether the firm has a group of k core banks; and (ii) the type of bank it
transacts with. Specifically:
Assumption 2.1. When the firm does not have any relationships, then
implementing the deal with any bank imposes a transaction cost bv on the
firm.
Assumption 2.2. When the firm has a group of k core banks, then
implementing a deal with a non-core relationship bank imposes a
transaction cost av, with a 2 ½0; RvÞ (that is, a is ‘small’). On the other hand,
a deal implemented by a fringe bank imposes a transaction cost bv, with
a < Rv < b < 1.
Assumption 2.1 says that firms pay a transaction cost which grows with v
when they do an arm’s length deal. One reason is that without the knowledge
that is gathered in a relationship it is less likely that the right deal structure
will be chosen. Thus mistakes are more likely (see Eccles and Crane [1988]
for an elaborate account) and their costs should be roughly proportional to
the size of the deal. At the same time, this assumption implies that the
relationship technology is efficient for firms with large enough deal volume.
It would be hard to justify the contrary: if arm’s length technologies were
always more efficient, then relationships would not be observed in the first
place.25 Last, because we assume that the arm’s length technology does not
exhibit economies of scale at the firm level (larger deals are more costly),
there is no loss of generality in assuming that banks incur no cost when using
itFcompetition would ensure that firms pay any cost incurred by banks in
equilibrium.
Assumption 2.2 says that relationships create an externality that reduces
the cost of doing a deal with a non-core relationship bank. Moreover, the
benefit is substantial because a is ‘small.’ This captures the idea, as we saw
25
Ljungqvist et al. [2001] examine 2,143 IPOs by issuers in 65 countries outside the United
States between 1992 and 1999. They find that firms selling their securities to U.S. investors
through U.S. banks typically charge higher direct fees, but including underpricing they are
considerably cheaper. They argue (p.25) that ‘. . .the more sophisticated capital markets
outside the U.S. have only recently begun to develop. . .the relationships that link key
intermediaries in the venture capital and the private equity markets to the primary and
secondary markets.’
above, that information is not excludable because senior bankers can switch
or the firm can approach another bank to implement the ideas suggested by a
relationship bank.
The assumption that a < Rv < b says that while non-core banks can free-ride
on relationship investments, fringe banks cannot. This asymmetry captures
the idea that banks that employ the relationship technology are
differentiated from banks that do not. Recall that to establish and exploit
the advantages of relationships requires two types of costs: the entry cost E of
setting up a ‘relationship infrastructure,’ and the cost R of gathering firm-
specific information through a relationship with it. While this latter
information can be used to better structure deals and offer value-added
services that cater to the firm’s needs, implementing these deals and services
requires the additional infrastructure costs E. In practice, this ‘infrastruc-
ture’ may involve investments in a broad product line, experience with
particular product specialties, and a large retail distribution networkFall of
which allow banks to more efficiently implement the complex deals that
relationships invoke.26 Differences between non-core banks (that entered
the relationship segment but do not have a relationship with a particular
firm) and fringe banks are thus embodied in E, making it cheaper for non-
core banks to free-ride on relationships by core banks than it would be for
fringe banks to do so.27
Last, it is also assumed that each time a relationship or a fringe bank i does
a deal (but only then), it charges a fee that is a proportion li 2 ½0; 1Þ of the
dollar value of the deal. Hence, total fees charged by bank i to a firm that
generates a deal of size v are liv. For simplicity, we assume that each
relationship bank charges the same proportional fee to all the firms in its core
group with whom it does a deal, regardless of v.28 Below we extend the
analysis to consider non-linear fee schedules and show that this does not
change our conclusions.
26
Eccles and Crane ([1988] pages 100–108) describe in detail the different investments that
banks with relationships make. For example, they note the banks’ ‘concern about the
consequences to customer relationships of not having a full product line’; the ‘experience
expertise in the many product specialties required to serve the needs of these customers’; the fact
that the ‘availability of a large retail network is undoubtedly a strength when soliciting certain
kinds of customers, such as companies with household names that issue securities;’ and that
‘those with a full product line, such as all six of the special bracket firms, believe that there are
strong interdependencies across the various financial markets and that they can serve some
customers better by being able to meet all their needs.’
27
In practice, bulge-bracket banks are similar multiproduct firms which specialize in doing
the deals of large corporations (see, for example, Hayes et al. [1983]). Thus, it is not difficult for
Goldman Sachs to execute a deal that has been designed by Merrill Lynch. By contrast, fringe
banks tend to be small boutiques which do not have the infrastructure which is typical of bulge-
bracket banks.
28
In practice, there is evidence that smaller deals tend to pay higher fees as a proportion of
deal size (see Ritter [1987] and Lee et al. [1996]).
Now in each period, the timing of the game is as follows:

1. Each firm establishes a relationship with k investment banks, at
which point banks incur the sunk relationship cost R.
2. Investment banks offer a fee which is a fraction l of the dollar value
of the deal.
3. The firm chooses an investment bank
4. Deals are implemented, fees paid and the game ends.
II(iii). The Well-Known Tension in a One-Period Model

We can now examine the well-known tension between relationships and
competition in the one-period game. Assume that there are m 4 k
relationship banks (below we endogenize m). The equilibrium of this game
is straightforward.29 Non-exclusive relationships imply that any relation-
ship bank that is a member of the firm’s core group can do the deal at zero
cost after the relationship cost R has been sunk. Hence, in a one-period game
every bank undercuts, and Bertrand competition drives the equilibrium fee
to zero.
It is important to note that even if relationships were exclusive (k 5 1),
competition from non-core banks would drive fees to a, which is not enough
to recover R because a < RvFthis is nonexcludability. Loose linkage, in turn,
implies that investment banks cannot charge for establishing relationships.
Anticipating all this, no investment bank will establish a relationship in the
first place.
In this setting, relationships can emerge only if competition is imperfect.
This is the well-known tension between relationships and competition. In
what follows, we characterize this imperfectly competitive equilibrium
market structure when relationship banks compete with each other and with
fringe banks.
III. RELATIONSHIPS AND COMPETITION
III(i). The Repeated Game

Banks will invest in relationships only if they anticipate that they will not be
undercut. Among the mechanisms that can restrain price competition are
regulations, frictions like informational monopolies, contracts and self-
enforcing norms.30 Neither of the first three seem very relevant in restraining
price competition in the investment banking market. On the other hand, as
mentioned earlier, many accounts of the industry suggest that price
competition is restrained by informal unwritten rules. This suggests that a
29
The formal proof is in Appendix A.
30
See Aoki and Dinç [1997].
repeated-game is appropriate to study relationships and competition. Thus
assume:
Assumption 3.1. (i) Banks are infinitely lived and by sinking E they can enter
into the relationship at the beginning of any period thus remaining a
relationship bank forever.
(ii) Relationship banks play the one-period game an infinite number of
times. k1
(iii) Banks discount
future with discount factor d 2 k ; 1 .
the
1
(iv) E < 1d f r bV r
k R , with f as defined below.
Assumptions (iii) and (iv) ensure that an implicit contract that sustains
relationships exists: assumption (iii) ensures that the implicit contract can be
sustained with k core banks, and assumption (iv) says the sunk entry cost can
be recovered when the fee is at its maximum, b. In addition, we make the
following simplifying assumption.
Assumption 3.2. Each generation of firms lives for only one period.
Assumption 3.2 is made for simplicity. It may seem surprising at first
because in practice relationships tend to last for years and information can
be reused to some extent (see, for example, James [1992]). But, even there, the
key point is that banks constantly interact with the firmFso that established
relationships must be nurtured over timeFwhereas deals occur only at
discrete moments. In a more elaborate model, time would be continuous,
relationship expenditures by banks would be made every instant and
depreciate over time and deals would randomly arrive at discrete moments.
Thus, at each moment, relationship banks would have a stock of sunk
relationship investments that they would lose if they engaged in a price war.
Our one period assumption captures this with a much simpler structure.31
Now as is well known from the theory of repeated games (see Fudenberg
and Maskin [1986]), multiple equilibria are endemic. As Sutton [1991, 1998]
has shown in a different context, however, it can still be very useful to
characterize the bounds on the variables under study that obtain in
equilibrium. Following the approach of Sutton, we will characterize the
bounds on fees, market shares, and concentration that define the range over
which cooperative equilibria exist. To do so, we consider equilibria with the
strongest feasible punishments. Specifically, any deviation by an inter-
mediary is assumed to destroy the implicit contract forever.32 Moreover, we
study an equilibrium where entry into the relationship segment is
31
A different but related interpretation is that R is sunk in the sense that it does not affect
short-run price competition, as in Sutton [1991].
32
In section 3.5, we explain why this assumption is exactly what is needed to obtain lower
bounds on concentration and fees.
accomodated as long as it is profitable.33 Because it is tedious to specify

strategy combinations that yield a subgame perfect equilibrium, this is
relegated to the appendix. Here we only characterize the equilibrium path.
III(ii). Four Key Conditions

This section describes four conditions that must hold in an equilibrium with
relationships. Assume that a symmetric equilibrium with relationships exists
where all relationship banks charge fee lc (where the superscript ‘c’ stands
for ‘core bank’).34
Competition by fringe banks. The first condition says that the fee charged
by a relationship bank cannot exceed b, that is
ð3:1Þ lc)b:
Because the arm’s length technology exhibits constant returns to scale, and
all transactions costs are borne by firms, competition ensures that the fee
charged by fringe banks, call it lf, will equal zero in equilibrium. But a firm
characterized by volume v which does its deals with a fringe bank would
incur a transaction cost bv, from which condition (3.1) follows. (Notice that
even if banks incurred the transaction cost bv, firms would be charged this
full cost in equilibrium, hence the assumption on whether banks or firms
bear the arm’s-length cost is not central.)
Relationships and deal size. The second condition says that banks will not
establish relationships with firms that generate low volumes of deals. To see
this note that since each firm’s core group contains k relationship banks, each
bank in that group wins a given deal with probability k1. Therefore, banks will
establish relationships only with those firms with volume v such that
1 c
ð3:2Þ l v R*0;
k
from which a lower bound v ¼ kR lc follows. Sunk set up costs introduce scale
economies at the level of each deal. Since these set up costs are incurred by
relationship banks that cannot charge for them directly, banks will choose not
to establish relationships with firms that generate a low volume of deals. Note
that since lc)b (from 3.1), it follows that v*vb kR b:
Conditions (3.1) and (3.2) are depicted in Figure 3, which plots deal
volume on the horizontal axis and the total cost of doing a deal of a given
volume on the vertical axis. Given lc, firms whose total deal volume does not
exceed v*vb are rationed out by relationship banks. On the other hand, note
that in equilibrium any firm such that v*v will get k relationships, because
banks make profits in expected value.
33
That is, we rule out equilibrium threats such as ‘if you enter, we will no longer cooperate.’
34
Strategy combinations that ensure this is an equilibrium are derived in Appendix B. In
section 3.5 it is shown that results do not change if banks charge nonlinear fees.
For future reference it is useful to note
that the measure of firms
vv
establishing relationships equals f v f r ; and the average size of a
vþv
deal done by relationship banks is V ¼ 2 .
Relationships and imperfect competition. Consider a relationship bank
who has already sunk this period’s relationship costs and now has to decide
whether to undercut rivals. The bank will compare the one-time gains of
undercutting to increase its market share today from Zi to 1; against the cost
of losing the long run gains from cooperation from the next period on.
We start by computing the value of the implicit contract. Bank i will
compete for deals with k 1 other banks in each core group of which i is a
member. Thus bank i will make deals of value Vk on average. Each firm will
c
pay l kV in fees on average and total costs will be R cper firm, regardless of the
number of deals done. Hence, profits per firm are l kV R on average. If bank
r
Pma fraction Z35i of all f firms that establish relationships
i has relationships with
(with Zi 2 ½0; 1 and j¼1 Zj ¼ k), its long-run profits of cooperation are
c
1 r lV
f Z R :
1d i k
The payoff from undercutting is obtained as follows. If the bank
undercuts
c inperiod t, it will get the one-time gains of undercutting on top of
f r Zi l kV R . The gains from undercutting are obtained as follows. By
c
setting lci slightly below lc, bank i can get an additional lc V l kV ¼

1 k1 lc V on average from firms with whom it has a relationship.
Moreover, by setting lnc i (where the superscript ‘nc’ stands for ‘non-core
bank’) slightly below lc a, bank i can win deals from the remaining ð1
Zi Þf r firms with whom it does not have a relationship, thus obtaining slightly
less than ðlc aÞV per firm. If undercutting destroys the implicit contract
forever, the unilateral deviator gets
c
r lV r 1 c c
f Zi R þ f Zi 1 l þ ð1 Zi Þðl aÞ V:
k k
Therefore, after some trivial algebra, the implicit contract condition reads:
c
d r lV r 1 c c
ð3:3Þ f Z R *f Zi 1 l þ ð1 Zi Þðl aÞ V:
1d i k k
35
Note that Zi is not a market share. Bank i may have a relationship with all firms and yet not
be a monopoly, since each firm has relationships with k banks. There is a direct relation between
Zi and i’s market share, however. If on average banks get a fraction k1 of deals made by
P firms with
whom they have a relationship, bank i will make a fraction mi Zki of all deals, with m j¼1 mj ¼ 1.
Thus, mi is bank’s i market share.
For future reference it is useful to note that when all banks establish
relationships with the same number of firms, Zi ¼ mk . Then, condition (3.3)
can be rewritten as

d f r lc V fr
ð3:4Þ k R * ½ðm 1Þlc ðm kÞaV:
1dm k m
The implicit contract condition (3.3) is, of course, equivalent to the standard
collusion condition in oligopoly. Nevertheless, here ‘collusion’ or ‘coopera-
tion’ is necessary for the existence of a market with relationships, unlike in
standard oligopolies where it is merely a collusive device. In other words,
here collusion serves an efficiency role. For this reason, henceforth we will
use the more neutral term ‘implicit contract’ when we refer to condition (3.3)
and use ‘cooperation’ most of the time.
Entry into the relationship segment and sustainability. The fourth condition
says that relationship banks must make enough profits in present value to
pay the entry cost E. Thus
c
1 lV
ð3:5Þ f r Zi R *E:
1d k
Note that this is a standard sustainability condition: the permanent flow

of profits if relationships are sustained must be enough to pay the sunk costs
of entering the relationship segment.
III(iii). Investment Banking Structure

We are now ready to show that the four conditions rationalize observed
market structure. In particular, the market separates into two segments
which do not compete with each other: in one, relationship banks serve large
firms; in another, fringe banks serve small firms. Conditions (3.1) and (3.2)
jointly explain vertical segmentation. Conditions (3.4) and (3.5), in turn,
explain market structure, competition and entry into the relationship
segment. Last, we discuss how observed market structure rules out
alternative explanations.
Vertical segmentation. The claim is that vertical segmentation implies that
fringe banks do not effectively compete with relationship banks. The
argument proceeds in two parts. The first part of the argument says that low-
volume firms are rationed out of relationships by relationship banks. That is,
these firms would like to establish a relationship but relationship banks will
not do so with them. To see this, note that condition (3.1) implies that the
maximum that can be charged by relationship or arm’s length banks for
doing a firm’s deals is bv. Since each firm establishes k relationships,
relationship banks will not establish a relationship with a firm whose volume
bv
is less than vb kR
b , because in that case k < R. But, on the other hand, since
lc)b, these firms would like to establish a relationship. It follows that low-
volume firms must be rationed out from relationships by banks.
The second part of the argument says that relationship banks are not
threatened by competition from fringe banks. This is because of scale
economies inherent in the relationship technology. Since bv > kR for v > vb ,
it follows that relationship banks will have a cost advantage which grows
with volume; this is seen clearly in Figure 3. Thus, arm’s-length banks cannot
compete for the business of large-volume firms because they are inherently
more costly (and they must be, otherwise there would be no value to
relationships).
In conclusion, low-volume firms would like to be served by relationship
banks, but will be rejected by them. On the other hand, the relevant banking
market for high-volume firms is the relationship segment. The cost
advantage of relationship banks in serving these firms is large enough
that fringe banks are not meaningful competitors, despite the fact
that relationship banks make rents. This analysis yields a central result
on why relationships are sheltered from competition from arm’s-length
banks:
Result 3.3. (Vertical segmentation). There are two different markets:
fringe banks serve low-volume firms and a few large relationship banks serve
large-volume firms. Fringe banks do not compete with relationship banks.
Result 3.3 stems from the assumption that firms differ in their deal
volume. Therefore, relative to arm’s length lending, relationship lending
(which uses a high sunk cost–low marginal cost technology) is more
advantageous the larger the volume of deals. This difference protects
relationship lending from arm’s length competition at the margin.36
The relationship segment. We now study the relationship segment. The
implicit contract conditions (3.3) and (3.4) impose restrictions on competi-
tion and market structure. The first result indicates that the relationship
segment is an oligopoly:
Proposition 3.4. Relationships will be established only if there are few

relationship banks with similar market shares.
Proof. Fix the equilibrium number of banks, m. Then:

ð3:6Þ Z)Zi)k ðm 1ÞZ;
36
This result contrasts with Boot and Thakor [2000], who find that more intense competition
from arm’s length lenders (similar to a lower b here) reduces relationship lending by
commercial banks at the margin. The reason is that: (a) firms differ in quality, but the size of
each loan is the same for all; and, (b) although relationship lending adds value, the increment in
value is smaller for higher quality firms. Consequently, when the cost of arm’s length lending
falls, firms switch at the margin from bank to capital market lending.
k
C
c
0 δ kR β
1-(1-δ)k V
Figure 4
L is the set of pairs (lc, m) such that the no undercutting condition (3.3) is satisfied. m
is the
number of relationship banks, and lc is the fee charged by these banks in an equilibrium with
relationships. lc cannot exceed b because firms would switch to fringe banks charging lf 5 0. Locus
CC traces the maximum number of relationship banks, m, for any given admissible fee lc; or,
c
conversely, the lower bound on the fee, l , for any admissible number of relationship banks.
where
ðlc aÞV
Z¼
d lc V
< 1:
ðlc aÞV þ 1d 1 c
k R 1k l V
Since mi Zki , condition (3.6) also imposes a lower and upper bound on
market shares:
m)mi)1 ðm 1Þm m:
It is straightforward to see that the upper bound on mi must be less than 1.

At the same time, the lower bound on the market share of any given bank
implies that the maximum number of banks, m, that is consistent with
relationships is given by m1 ¼ kZ. Thus, in equilibrium relationship banks must
be ‘few.’ &
Proposition 3.4 implies that relationship banks must have a similar
number of relationships. On the one hand, if one becomes too small and
establishes relationships with few firms, then cheating becomes profitable.
On the other hand, if one relationship bank becomes too large, then there
will be too few relationships left for the other relationship banks, which
would like to deviate from the implicit contract. Notice that the lower bound
on market shares also imposes an upper bound on the number of relationship
banks, thus a lower bound on concentration. Figure 4 plots L, the set of
pairs of lc and m such that condition (3.4) holds. It is seen that the upper
bound m is increasing with the equilibrium fee lc.
Note that we have deduced an upper bound on the number of firms that
comes from an implicit contract condition, not a standard entry condition
like (3.5). Of course, in equilibrium the number of banks in the relationship
segment must also be small enough to pay the entry cost E. Thus the second
result:
Result 3.5. In a subgame perfect equilibrium with symmetric market
shares where entry is accommodated as long as it is profitable the number of
relationship banks is at most equal to minfm; mzp g, with mzp defined by
c
1 fr lV
ð3:7Þ k R E:
1 d mzp k
(To simplify the notation, we henceforth assume that m and mzp are
integers.) Equation (3.7) is a standard zero-profit condition, but it does not
necessarily determine the number of relationship banks. The result says that
< mzp then there exist equilibria where relationship banks make profits
if m
net of entry costs, and yet there can be no further entry because cooperation
(or collusion) would no longer be sustainable.
Note that E 4 0, however small but strictly positive, is needed to ensure
that entry stops when m 5 mzp. If E 5 0, banks would be indifferent between
remaining outside or entering an industry whose only equilibrium is such
that no relationships are ever established. Thus, m 5 mzp as E ! 0, but it
would be indeterminate if E 5 0.
Two important implications. Proposition 3.4 and Result 3.5 yield two
important implications which we can use to check the model against
observed market structure. The first is that one can rule out that the observed
oligopolistic investment banking structure is a consequence of standard
scale economies from exogenous sunk costs. To see why, note that mzp falls
with market size f. This is a standard result in models with exogenous sunk
costs: as the market grows in size concentration falls in equilibrium
(see Mas Collel et al. [1995] ch. 12) for a rigorous proof of this assertion).
By contrast, it can be seen from condition (3.3) that m is independent of
market size f r.
Why is that so? Note that both sides of condition (3.3) are multiplied by f r,
the number of firms that establish relationships. Thus, when the size of the
market increases both the gains from cooperation and the gains from
cheating increase in the same proportion. Hence, any combination (lc,m)
satisfying condition (3.3) in the smaller market will also satisfy it in the larger
market. It follows that concentration should not vary with market sizeFthe
regularity depicted in Figures 1 and 2.
The second implication concerns the role of nonexcludability of
relationships (i.e., a must be much smaller than lc) in explaining observed
market structure. To see this, assume, to the contrary, no externalities

between core and non-core banks, so that lc 5 a. Then (assuming equal
market shares for simplicity) the implicit contract condition (3.3) reduces to

fr d fr 1
ð3:8Þ ðlc V kRÞ* 1 lc V:
m1d m k
and the sustainability condition is
fr 1
ðlc V kRÞ*E:
m1d
As before, the implicit contract condition is not affected by market size fr as
it appears on both sides. But now an increase in the number of relationships
no longer makes cooperation harder, because both the gains from
cooperating and cheating fall as the number of firms increases. For this
reason, a relationship bank with a small market share is no longer a threat to
cooperation: if you are small and establish few relationships, then you can
gain little market share by undercutting. By contrast, when relationships are
not excludable, the gains from cheating increase as market share falls
because the unilateral cheater grabs the whole market. Because the implicit
contract condition no longer restricts m when relationships are excludable,
only the sustainability condition matters; but then concentration falls as
market size increases.
In other words, the regularities depicted in Figures 1 and 2 restrict possible
explanations of market structure quite tightly. Worth noting is that these
two regularities are similar to those obtained in markets with endogenous
aggregate sunk costs (Sutton [1991]). Here, however, the exogenous sunk
cost R is incurred only at the local level. As a result, the aggregate technology
exhibits constant returns to scale. Hence, sunk costs are naturally ‘escalated’
when the size of the market and the number of firms increases.
III(iv). Relationships and the Intensity of Competition

So far we have studied how market equilibrium embodies three faces of
competition: multiple relationships, entry, and arm’s-length deals. We now
examine the effect of changes in the intensity of competition: a fall in a, a
firm’s cost of switching from a core to a non-core bank, and an increase in the
number of relationships k. Since we have shown that fringe and relationship
banks serve different segments, we can ignore fringe banks in this analysis
and examine how the upper bound on m and the lower bound on lc Fi.e. the
set L in Figure 4Fvaries with changes in exogenous parameters. This
‘bounds’ approach follows Sutton [1991].
Consider a fall in a. Because relationships are not excludable, both core
and non-core banks have a temptation to undercut. If a falls the gains from
undercutting increase because non-core banks need to discount their fees by
m
C′
k
C
c
0 δ kR β
1-(1-δ)k V
Figure 5
The effect on the no-undercutting locus CC of a smaller switching cost from a core bank to a
non-core relationship bank
Locus CC 0 traces the effect of a smaller switching cost on the maximum number of relationship
for any given admissible fee lc. The shaded region indicates the set of pairs (lc, m) such
banks, m,
that the noundercutting condition (4.1) is satisfied after a firm’s cost to switch from a core to a non-
core bank falls.
a smaller amount to compete with core banks. Then, the implicit contract
becomes less attractive unless fees increase. To confirm this intuition, totally
differentiate (3.4) and rearrange to obtain:
dlc ðm kÞa
¼ d
;
da 1d ðm 1Þ ð1 eV;lc Þ lac ðm kÞeV;lc
where eV;lc is the elasticity of average volume to a change in the fee lc (see
Appendix C for the details of the derivation). This derivative is negative as
long as the no-undercutting locus CC derived from (3.3) is upward sloping.
Hence, when a falls, locus CC shifts leftward (see Figure 5). Similarly,
mk
dm
¼ > 0:
da lc a
Result 3.6. When switching costs from core to non-core banks fall, fees tend
to be higher for a given number of banks m. Conversely, concentration
increases for a given fee lc .
Moreover, if a falls from a 0 to a00 o a 0 then Lða00 Þ Lða0 Þ. Therefore, the
implicit contract is harder to sustain when competition from non-core banks
is more intense.
Result 3.6 appears counterintuitive because lower switching costs for

firms (or easier free-riding) are thought to decrease market power of banks.
That effect, while true, ignores the equilibrium consequences. The reduced
profit from each relationship is counteracted by exit of relationship banks,
and therefore higher aggregate market shares for those that stay. In other
words, increased competition for deals is offset by a decrease in competition
for relationships.
Consider now what happens when k increases. For a variety of reasons,
firms have tended over the last two decades to increase the number of
investment banks with which they have relationships. Eccles and Crane
([1988] ch. 4) term this a shift from a ‘dominant bank model’ to a ‘core group
model.’ Firms may increase the number of relationships because of ‘increased
information flow and ideas from multiple relationships,’37 an increase in
underwriting and corporate restructuring business that can be allocated
amongst more banks, or a desire to increase competition among relationship
banks. Some observers question whether relationships can survive this trend.
Do multiple relationships weaken the incentives to establish them?
Assume that k increases, so that firms establish relationships with more
banks. To study the equilibrium effects on fees, substitute lc into condition
(3.4), let it hold as an identity, and totally differentiate with respect to lc and
k (see Appendix C for the details of the derivation). Rearranging yields
dlc ½ðm 1Þlc ðm kÞa2kv d cv

þ aV þ 1d l 2k
ð3:9Þ ¼ d a
;
dk 1d ðm 1Þ ð1 eV;l Þ þ lðm kÞeV;l V
v
where we used that @V @k ¼ 2k. This derivative is positive as long as a is
sufficiently small (again, see Appendix C), which implies:
Result 3.7. For a given number of banks, m, fees tend to be higher when
firms establish more relationships.
Multiple relationships are often thought to toughen price competition.
Result 3.7 runs counter to this intuition. While the analysis confirms that the
effect of an increase in k is to reduce the probability of winning a deal, and
c
therefore the net margin per firm to each bank, l kV R, it shows that the
gains of unilaterally undercutting also increase. Thus, given m, each bank
wants to establish relationships only if fees increase. Similarly,
v cv

dm ½ðm 1Þlc ðm kÞ2k þ aV 1d d
l 2k R
¼ ;
dk ðlc aÞV
which is negative. An increase in k therefore reduces L, the set of pairs (l, m)

that can be sustained in equilibrium. Multiple relationships reduce firm-
37
Eccles and Crane ([1988] p.78)
specific rents to banks since revenues per firm fall while relationship costs do
not. The increased profits from which to recoup relationship costs can then
be created by increasing prices or inducing exit.
An increase in k will, for a given lc, not just affect the concentration of
banks in the relationship segment of the market, but the size of the
relationship segment itself. To see this, note that the lower bound v on firm
volume, kR/lc, increases in k. This will both reduce the number of firms that
establish relationships with banks, f r, and the aggregate volume of deals
vþv
intermediated by relationship banks, f r 2 . Thus, the effect of firms
establishing more relationships is to increase concentration of relationship
banks on the one hand, while increasing the size of the market served by the
competitive fringe on the other.38 This apparent increase in both
competition and concentration might explain why the effects of such
changes often appears puzzling to observers.
To summarize, both these results stress that one cannot analyze the impact
of a change in market conditions at the local firm-bank level. Market
equilibrium involves adjustments at the aggregate levelFin fees or market
concentrationFto preserve the incentives to incur the sunk costs of
relationships.
Result 3.8. An increase in the intensity of competition need not destroy

relationships as long as the implicit contract remains sustainable.
Result 3.8 illustrates a general lesson: the deleterious effects on relation-
ships of changes in the intensity of one type of competition may be partially
undone by changes of market structure and the intensity of other types of
competition.
III(v). Robustness and Extensions

In this section we examine the robustness of the results to several modeling
assumptions and explore the consequences of other extensions.
III(v)(i). Which Assumptions Drive the Results? Non-verifiability of

relationships. To examine the role of loose linkage for market structure,
assume the reverse: that is, banks can directly charge a firm for a
relationship. Then banks would compete and charge R for establishing a
relationship and a firm with volume v would establish a relationship only if
kR)bv. Moreover, a firm with v 5 vb would be indifferent between either
type of bank, not rationed by relationship banks. Hence:
38
Boot and Thakor [2000] find a somewhat similar result in the context of commercial bank
relationships. They argue that the effect of increased interbank competition on relationship
lending by commercial banks includes both a negative absolute effect on volume of loans lent
through relationships but a positive relative (substitution) effect on the capacity devoted by
banks to relationship lending.
Result 3.9 (Relationships and perfect competition). If relationships were

verifiable and investment banks could charge directly for them, then
relationships could be sustained in a perfectly competitive market.
Result 3.9 implies that imperfect competition is a consequence of

relationships being nonverifiable, not a consequence of the economies of
scale exhibited by the relationship technology.
Game-theoretic issues. We have used a grim trigger strategy to sustain the
implicit contractFi.e., one deviation destroys the equilibrium foreverF
which turns out to be the strongest feasible punishment in this case.39 One
may wonder whether the bounds on m and lc we derived remain valid for
weaker punishments.
Consider an alternative strategy that yields to intermediaries f r Zi vp > 0 in
present value in the punishment phase (i.e., the payoff during a punishment
is larger if the relationship segment fr is larger, or if the bank has a larger
market share). In that case, the implicit-contract condition would read
c
d r lV r 1 c c
f Z R *f Zi ð1 Þl þ ð1 Zi Þðl aÞ V
ð3:10Þ 1d i k k
þ df r Zi vp ;
which implies, after some algebra, that the bounds are now determined by
the equation

lc V kR ð1 dÞvp 1 d 1 w lc a
¼ ð1 Þ þ ðm 1Þ :
lc V d k lc
The weaker the punishment, the harder it is to fulfill the implicit contract
condition (3.10). To restore incentives to cooperate, the gains from cheating
must fall relative to the gains from cooperating. Hence, either the maximum
number of intermediaries, call it mw, must be smaller than m or lc must
increaseFthat is, minimum concentration and the maximum fee are higher
with a weaker punishment. Therefore, the set of pairs (m, l) in L with the
strongest feasible punishment contains the set Lw with punishment f r Zi vp .
One may also be concerned that banks may renegotiate out of the
strongest feasible punishment. Nevertheless, the bounds we derived are still
the same, because we have shown that any set of pairs (m, l) sustainable with
weaker punishments will be a subset of the set of pairs (m, l) in LFif
anything, fees and concentration will be higher with a renegotiation-proof
implicit contract.
39
It is the strongest feasible punishment because relationship banks can not be forced to get a
payoff of less than zero.
Nonlinear fees. Linear fees are not necessary to any of our results. First,
rationing follows from non-verifiable relationships, because relationship
banks cannot charge more than bv to any firm, regardless of how this sum is
collectedFwhether through a linear fee, a two-part tariff or a more complex
nonlinear schedule. Similarly, the result that high-volume firms are served
only by relationship banks follows purely from bv 4 kR. Indeed, this
condition allows, for example, not just a linear fee such that bv*lc v*kR,
but nonlinear schedules as well, call them F(v), such that F kðvÞ*R. Again,
market separation follows from non-verifiable relationships, a problem that
cannot be solved by charging non-linear fees.
Of course, whether relationship banks charge linear or nonlinear fees may
affect v, the cutoff volume below which relationship banks ration firms. As
seen, if a linear fee lc is charged, then v ¼ kR
lc . Thus, the exact value of v only
affects the relative sizes of the relationship and arm’s-length segments, not
the result that banks in these segments effectively do not compete.
Second, condition (3.3) could be substituted by any fee schedule F ðvÞ, and
then it would read "

d r Ev*v ½FðvÞ r 1
f Z R *f Zi 1 Ev*v ½FðvÞ
1d i k k
ð3:11Þ #
þ ð1 Zi ÞEv*v ½Fðv; aÞ ;
where Ev*v ½F ðvÞ is the expected fee income generated from a firm that
established a relationship and Ev*v ½F ðv; aÞ)Ev*v ½F ðvÞ is the expected
income generated from a firm that is poached after undercutting. In other
words, rents sustain the implicit contract in equilibrium and these need not
come from linear fees.
Are fees observable in practice? The implicit contract condition (3.3)
assumes that relationship banks observe the fees that have been offered by
rivals and act upon them in the next period when they make their decision
whether to continue cooperating or punish. But because deals between firms
and investment banks contain complex transactions, it might seem
unrealistic to assume that fees are observable.
Nevertheless, pricing in this industry is quite simple since fees are a
proportion of the dollar amount of the deal and companies like Securities
Data Company record the fees of every deal made and sell the information.
Moreover, in practice underwriting is done through syndicates where top
banks regularly meet.40 A deviator would need to eschew syndicates, which
is observable. More important from the point of view of the modeling, fees
40
For example, Eccles and Crane ([1988] p. 94) report that of the 6,327 domestic security
issues led by one of the six top banks between 1984 and 1986, 60.4% were comanaged by
another top six bank.
need not be observable to define a strategy that punishes deviators.

For example, banks could condition their actions on their market shares,
and enter into a punishment phase when they fall below a certain
threshold.41
III(v)(ii). Other Extensions
Endogenous number of relationships. We have assumed that k is
exogenous. But, one might expect from equation (3.9) that by lowering k,
a firm could lower the fee that it pays. As we explain below, there are several
reasons why this ‘endogeneity’ can be ignored in the analysis.
First, notice that equation (3.9) is a market-level relationship, not a firm
level one. Hence, firms are price takersFthey take the equilibrium market
fee lc as given when optimizing k. Second, there still could be an endogeneity
concern when firms are price takers, if the level of fee lc would affect the
optimal number of relationships chosen by a firm (in that case, k and lc
would be simultaneously determined in equilibrium). There are obviously
many factors that should affect a firm’s optimal choice of k (for example,
access to a wider array of ideas and approaches, and the cost of time and
attention of management). But, there is no reason that optimal k must vary
with fees. To see why, suppose that a change in some other parameter alters
the equilibrium fee lc. Would this cause firms to change the number of
relationships that they establish? No. The reason is that the cost and benefit
of an extra relationship by a firm is not affected by fees, since these are paid
only when a deal is done.
Last, consider a more general case where, in equilibrium, banks offer a fee
schedule such that lower fees are charged to a firm that established fewer
relationships (the fee schedule could be linear or non-linear in k). In this case,
a firm will optimize the number of relationships since fewer relationships can
reduce the costs of doing any deal. Our market level analysis would not be
affected, however. To see why, note that then a firm with volume v that
chooses k relationships would generate, say, F ðv; kÞ dollars in fees, and the
bank would make her deal with probability k1. Then only firms such that
F ðv;kÞ
k *R would get deals in equilibrium, and the implicit contract condition
would be

d r ½F ðv; kÞ
f Z Ek ;v*vðkÞ R
1d i k

r 1
*f Zi Ek;v*vðkÞ 1 ½F ðv; kÞ þ ð1 Zi ÞEk;v*vðkÞ ½F ðv; k; aÞ :
k
41
For example, in Green and Porter’s [1984] classic model of collusion firms choose
unobservable quantities but condition their strategies on observable prices.
As with fees that are nonlinear in v, v, the cutoff volume below which
relationship banks ration firms, may be affected by variation in k. But the
exact value of v only affects the relative sizes of the relationship and arm’s-
length segments. The essential point is that endogenizing k does not change
either the result on vertical segmentation nor the predictions on market
structure.
Nonprice competition. Investment banks compete in various non-
price dimensions. They incur sales expenditures, advertise, provide ‘free’
advice on other financial and investment matters. Indeed, it is often
claimed that banks price some services below their cost in order to get access
to clients. One might ask how nonprice competition would alter the
previous results. In particular, is it the case that nonprice competition
might dissipate the rents that banks get in equilibrium, thus undermining
the incentive to establish relationships? In the appendix we extend the
model to include nonprice competition among banks. Call E the amount
spent in sales efforts per firm by each bank in a symmetric equilibrium.
In that case (as is shown in the appendix), the implicit contract condition
would read
c
d r k l V fr
f R E * ½ðm 1Þlc ðm kÞaV;
1d m k m
which is almost the same as before except that costs E are now taken into
account as well.
The central point is that nonprice competition does not change the
requirement that banks make excess profits in equilibrium in order to
establish relationships. In many models, ex-ante nonprice competition is a
mechanism to dissipate ex-post rents. This does not occur here because
relationships play an efficiency role (and require rents) and the gains from
cheating are independent of sales efforts. Thus, sales efforts do not do away
with the need for soft price competition, which is the source of rents in this
model. Moreover, it can be easily seen that the bounds on concentration and
fees are, if anything, tighterFconcentration and fees tend to be higher with
sales efforts.
Might sales effort competition be dissipative, and thus inefficient? It
certainly may be, for standard reasons. On the other hand, sales efforts may
also allow firms access to better ideas to do deals. Last, even though nonprice
competition is not a means to compete away rents, it does restrain bank
market power by imposing an upper bound on the fee lc that banks can
charge in equilibrium (see the appendix for details).
The multiproduct nature of investment banks. Bulge-bracket banks are
multiproduct firms, so that concentration in any one market may mask the
fact that leading banks differ across products. Nevertheless, the top, bulge
bracket banks tend to be the same in most product lines (see, for example,
Eccles and Crane [1988], Santomero and Babel [2001] p. 500). Moreover,
as we show in the appendix, as long as the economics of the
technology exhibits these characteristics in some segments of the investment
banking industry, then the implicit contract condition must hold across
products. In particular, this implies that multiproduct competition
cannot dissipate rentsFotherwise cooperation would no longer be self-
enforceable.
Welfare. A central implication of the analysis is that collusion among
relationship banks serves an efficiency role. Nevertheless, there might be
typical concerns that acompany market power, such as inefficient exclusion
or exploitation by intermediaries. Can one say anything about the overall
welfare consequences of relationships?
First, inefficient exclusion is unlikely. The reason is that in equilibrium,
small volume firms who want relationships with banks are rationed out of
the relationship segmentFwhen lc v kR < 0. If the fee lc increases, fewer
firms are rationed out of relationships and the size of the relationship
segment increases with lc. Hence, exclusion falls as the fee increases, for
exactly the same reason that in a market with pre-existing rationing (e.g., a
market with a price ceiling fixed below the equilibrium level), sales increase
when the price increases.
Second, the cost of market power (e.g., exploitation of firms by banks)
is unlikely to overturn the beneficial aspects of relationships. To
see why, notice that the appropriate welfare comparison is between a
market with no relationships and one with imperfect competition.
Now, because firms always have the choice of doing arm’s-length deals, it
seems reasonable to think, by revealed preference, that a market with
relationships and imperfect competition Pareto-dominates a market with no
relationships.
IV. DISCUSSION
IV(i). Off-Equilibrium Episodes

A useful question to ask is whether investment banks ever ‘undercut’ and
how such episodes resolve. As discussed, free-riding can take many forms:
price undercutting, human capital poaching, or copying of ideas. While
unilateral changes in price schedules have not been observed, there are some
accounts of unusual poaching behavior. Even though the movement of
bankers from one bank to another is fairly common, it is unlikely that these
are construed as ‘off-equilibrium’ actions. However, a few episodes do stand
out. Perhaps the most famous is Deutsche Bank’s attempt to gain status as a
‘bulge bracket’ bank in the early 1990’s by en masse hiring of staff from other
banksFthis series of hirings, including that of ‘the flamboyant technology
star Frank Quattrone,’ was described as having ‘brought the baseball term
‘free agent’ to Wall Street.’42 Several years later, ‘when UBS AG’s Warburg
Dillon Read lured veteran healthcare star Benjamin Lorello away from
Citigroup’s Salomon Smith Barney to orchestrate its global healthcare
investment banking practice, the headline made front-page news.’ 43
The reason that each of these cases was construed as a ‘poaching raid’ was
probably both because of the number of bankers that accompanied these
departures (in the Deutsche Bank case, it involved more than 200 senior
bankers) and because of the large increase in salaries. More interesting is
that, even in an industry where banker mobility is fairly common, the
Lorello hiring was described by one headhunter, as ‘an auction, completely
without regard for the rules.’44
Each of these episodes was followed by competitor retaliation: following
Deutsche Bank’s hiring spree, CS First Boston offered Quattrone and his
100-member high-tech group upto $1 billion to leave; one account noted that
‘rarely has a poaching raid aroused as much schadenfreude among top
investment bankers’45 as their subsequent defection. Similarly, when
Deutsche Bank poached numerous traders in over-the-counter derivatives
and fixed-income trading from Commerzbank in 1998, Commerzbank
retaliated two years later by hiring away many of Deutsche’s own traders,
‘including some who were originally poached from Commerzbank by
Deutsche and are now being rehired on what one London banker claimed
would be ‘pretty massive bonus promises’.46 In response, Deutsche Bank cut
off its lines of business with Commerzbank in the derivatives and fixed
income markets in a move that ‘both hoped was temporary.’47 One industry
account in 2000 noted that the net result of these raids and retaliations was
that, in its attempt to achieve bulge-bracket status, ‘Deutsche Bank in New
York (was still) not even near the starting gate.’48
IV(ii). Applicability to Other Industries

Can one apply this model to explain how relationships are sustained in other
industries? Here, we briefly discuss both the generalizability and specificity
of the investment banking problem.
IV(ii)(i). Professional Service Industries The analysis here may be

usefully extended to explore other professional service industries (e.g.,
42
‘More Equal,’ by Erica Copulsky, Investment Dealer’s Digest, May 10, 1999.
43
Ibid.
44
Ibid.
45
‘Quattrone and the U.S. Market,’ Financial News, July 6, 1998.
46
‘Deutsche Bank Cuts Ties with Rival over Poaching’, by Vincent Boland, William Lewis
and Tony Major, Financial Times, March 31, 2000.
47
Ibid.
48
‘The Clouds on CSFB’s Gleaming Horizon,’ Financial News, March 13, 2000.
consulting, law, advertising) that share characteristics with investment

banking. Consider, for example, advertising agencies. Agencies make
significant investments in client-firm relationships. And, these relationships
are often difficult to appropriate because they are embodied in human
capital that can move to other firms. Further, like investment banking, entry
is unregulated.
Both firm conduct and market structure bear close resemblance to that of
investment banking as well. On the one hand, agencies compete fiercely for
market share and in sales efforts to get relationships. On the other hand, price
competition was restrained: for a long time, advertising agencies would charge
clients flat 15% commissions that were independent of deal volume, and did
not vary with project characteristics. Moreover, the market is is dominated
by a few firms with mostly similar market shares;49 concentration is
largely invariant to market size50; and, market structure is ‘pyramidal’ and
vertically segmented, with the few large firms co-existing with hundreds of
boutiques that serve smaller clients51. A more thorough analysis of this
and other professional service markets that builds on the approach here may
be fruitful.
IV(ii)(ii). Commercial Banking It is well known that relationships are

important in commercial banking and it has long been debated whether
competition hurts them. Are relationships in commercial banking ‘pro-
tected’ by a similar industrial organization?
There are two significant differences which suggest not. For one,
commercial banks maintain a relationship only while they lend money.
Because loans are kept in the bank’s balance sheet, they generate a steady flow
of interest income while the relationship is ongoing. Thus, it is unlikely that
bank-firm relationships are characterized by loose linkage. For another, the
cost of switching to the equivalent of a non-core bank is comparably more
important in commercial banking. As Boot and Thakor ([2000] p. 683) point
out, commercial banks enjoy some degree of local market power because of
the illiquidity of each loan due to its information sensitivity.52 And condition
(3.8) shows that relationships impose no restriction on aggregate market
49
In 2004, the worldwide share of advertising revenues accruing to the top five firms was
74.1%, with their individual market shares being 18.3% (Interpublic), 18.2% (Omnicom),
16.5% (Publicis), 15.5% (WPP), and 5.6% (Havas). Data from Advertising Age, http://
www.adage.com/images/random/agrpt04_piechart.pdf
50
For example, the share of worldwide advertising revenue accruing to the top five
advertising groups was 61% in 1990 and 74% in 2004, a period during which market size grew
by over 20% (Advertising Age, various years).
51
Thus, for example, Collis [1992] describes the industry as ‘hourglass shaped, with the
mega-agencies at the top, a thin middle, and a profusion of specialist, boutique, and new
agencies at the bottom.’ See also King et al. [2003].
52
See also James [1987], Kang et al. [2000], Lummer and McConnell [1989], and Shockley
and Thakor [1997].
structure if banks without relationships (non-core banks) incur similar
lending costs. Not surprisingly, the pyramidal structure that characterizes
investment banking is not a feature of commercial banking structure.
V. CONCLUSION
The tension between relationships and competition is a long-standing one.

This paper examines how this tension is resolved in the investment banking
market, that has for decades been characterized by both relationships and
competition.
Perhaps the central message of the paper is that competition need not kill
relationships. This insight emerges clearly when one embeds relationships in
an equilibrium analysis. The investment banking market offers an example
of how market equilibrium adjusts so that valuable relationships can still be
provided. Thus, for example, when competition gets more intense in some
dimensions (for example, firms increase the number of relationships that
they establish, or sales efforts pitched at establishing relationships increase),
the endogenous adjustment in fees or in the number of banks undo their
deleterious effects on the incentives to establish relationships.
The second insight about competition and relationships comes from the
technology of relationships. Specifically, price competition from inter-
mediaries doing arm’s length deals does not hurt relationships with large
firms, because arm’s length deals are inherently more costly for these firms.
Relationships are therefore sheltered from such competition by the scale
economies intrinsic to relationships.
Last, our results caution against examining the effects of different types of
competition in isolation. Investment banking shows that different ways of
competition interact. The technology of relationships imposes restrictions
on this interaction, and on the industrial organization of the industry. These
insights are probably useful to understand market structure in other
professional service industries where relationships are important.
APPENDIX
A. Equilibrium in the one-period game

We now rigorously describe the timing of actions, and then characterize the equilibrium
of the one-period game. The time line is as follows:
1. Each firm randomly contacts k relationship banks.
2. Each bank chooses those firms with which it wants to establish relationships,
and incur the corresponding sunk cost R. If k banks establish relationships
with a given firm, then that is the core group of banks of the firm.
3. Firms announce deals.
4. (Fee offers) Each relationship bank i simultaneously makes a price offer
lci 2 ½0; 1 [ f1g to all firms with whom it has a relationship, an offer lnc
i 2
½0; 1 [ f1g to all firms in a core group that does not include i and an offer
lnr
i 2 ½0; 1 [ f1g to firms that have no relationships (superscript ‘c’ stands
for ‘core’, superscript ‘nc’ for ‘non core’ and superscript ‘nr’ for ‘no
relationship’). These offers are expressed as a fraction of deal volume (thus,
they represent commissions or percentage fees). l 5 1 means that no offer
was made. Obviously lci ¼ 1 if bank i is in no core group and lnc
i ¼ 1 if it has
a relationship with every firm. Simultaneously, each fringe bank j makes a
price offer lfj 2 ½0; 1 [ f1g to all firms.
5. Each firm chooses the bank offering the lowest fee net of transaction costs. If
x 4 1 banks tie, then each bank wins the deal with probability x1.
6. Deals are implemented, fees paid and the game ends.
To define bank strategies let H be the set of possible histories right before banks make
fee offers. A strategy by a relationship bank i is a tuple ðRi ; Li Þ. Ri : ½0; v ! f0; Rg is a
function that indicates whether bank i will establish a relationship with those firms that
selected i to form part of the core group. Since firms are completely described by v, bank
i’s decision can be conditioned on firm type. Li ¼ ½lci ; lnc nr
i ; li is a three-dimensional
3
vector function Li : H ! ½½0; 1 [ f1g . In turn, a strategy by fringe bank j is a
function lfj : H ! ½0; 1 [ f1g. Proposition A.1 characterizes the set of subgame
perfect equilibria of this game. (See Appendix B for a strategy combination that is a
subgame perfect equilibrium of the one-period game.)
Proposition A.1. In any subgame perfect equilibrium, no relationships are established

and lf 5 0.
Proof. Suppose, by way of contradiction, that bank i establishes a relationship with a

firm. Suppose first that k 4 1. In any subgame with relationships, Bertrand
competition for deals between core banks drives lc to 0 in equilibrium. On the other
hand, if k 5 1, Bertrand competition with relationship banks drives lnc to a and hence
lci to a < Rv in equilibrium in any subgame where bank i is the firm’s only relationship
bank. Hence, in both cases bank i loses money if it becomes a relationship bank,
therefore lci ¼ 1. Finally, note that since there are no variable cost of doing deals,
Bertrand competition among fringe banks drives lf to 0 in equilibrium. &
Hence no relationships are established in equilibrium, only banks that do not
establish relationships are active, and each deal costs bv to firms. Moreover, since there
are no variable costs of doing deals, fees are driven to zero in equilibrium. For firms
with v such that b)kR v (i.e. firms that do small and infrequent deals) this equilibrium is
efficient. By contrast, firms with v such that b > kR v would want to establish k
relationships and compensate banks for the incurred sunk costs. This can be
summarized in the following result:
Result A.2. The equilibrium of the one-shot game is efficient for firms that do small
deals. It is inefficient for firms that do large deals.
A straightforward implication of Result A.2 is that low-volume firms will never
establish relationships. Moreover, Result A.2 shows that establishing relationships
is not efficient for every type of client. Firm volume will determine which technology is
efficient.

B. A subgame perfect equilibrium with no undercutting
In this appendix, we study a symmetric subgame perfect equilibrium where all
relationship banks charge the same fee lc 4 0 period after period, all fringe banks
compete and charge lf 5 0, relationships are profitable and there is free entry into the
relationship segment. The outcome of this equilibrium are the conditions examined in
the text.
We start by defining a strategy combination that is a subgame perfect equilibrium of
the one-period game.
Definition B.1. Call strategy combination P (for ‘punishment’) the following

combination of strategies
For all relationship banks i

1. Ri ðvÞ ¼ 0 for all v 2 ½0; v;
2.
8
>
> ð0; 0; 0Þ if i is member of at least one core group and not member
<
of at least another core group;
ðlci ; lnc nr
i ; li Þ ¼
>
> ð1; 0; 0Þ if i is member of no core group but at least one core group exists;
:
ð1; 1; 0Þ if no core groups are established:
For all fringe banks j, lfj ¼ 0.

Part (i) of the strategy of relationship banks implies that no bank establishes a
relationship. Part (ii) implies that i undercuts other relationship banks on all histories
such that i is in a core group. Finally, part (iii) implies that bank i always undercuts
when not in a core group. We are now ready to prove the following lemma:
Lemma B.2. Strategy combination P is a subgame perfect equilibrium of the one period
game.
Proof. Consider first histories where at least one firm establishes relationships with k
banks and forms its core group. According to P, for these histories we have to
distinguish three cases:
8
>
> ð0; 1; 0Þ if bank i is member of all core groups;
<
c nc nr ð0; 0; 0Þ if there is at least one core group where bank i is not a member
ðli ; li ; li Þ ¼
>
> but bank i is member of at least one core group;
:
ð1; 0; 0Þ if bank i is member of no core group:
In any of these three cases, any unilateral deviation by bank i setting lci > 0 or lnci >0
or lnr
i > 0 as the case may be will not increase its payoff, since it would get no deals.
Consider next histories where no firm forms a core group. Then, ðlci ; lnc nr
i ; li Þ ¼
ð1; 1; 0Þ according to P. Setting lnr i > 0 will not increase i’s payoff since it would get
no deals.
Last, setting Ri ðvÞ ¼ R for one or more v’s will not increase i’s payoff because
according to strategies no other firm establishes relationships. &
The following corollary follows from Proposition A.1 and Lemma B.2.

Corollary B.3. All banks receive a payoff equal to 0 in the one period game.
Thus, since playing strategy combination P forever is clearly a subgame perfect

equilibrium in the infinitely repeated game, it follows that it can be used to construct a
subgame perfect punishment. We now define ‘undercutting.’
Definition B.4. Let lc be the fee charged in an equilibrium with relationships. Then
there is undercutting in period t if minflci ðtÞ; lnc c
i ðtÞ þ ag < l for at least some i.
Note that lnr and lf are not part of the definition. We are assuming that neither
‘undercutting’ in the fringe segment, nor fringe banks setting fees such that lf þ b < lc ,
destroys cooperation.
We now specify a strategy combination such that cooperation is a subgame perfect
equilibrium. To do so, it is useful to assign each possible history of the game into one of
two disjoint sets.
Definition B.5. We say that the history of the game at period t is ‘cooperative’ if
(a) m)mðlc Þ;
(b) no undercutting has occurred so far. That is, for all t < t,
minflci ðtÞ; lnc c
i ðtÞ þ ag*l . Any other history is non-cooperative.
Notation B.6. We denote the state of the game at period t by ft. The state of the game
after a history with no undercutting is cooperative and is denoted by fc. Any other state
of the game is ‘non-cooperative’ and is denoted by fnc.
Note that this definition implies that the initial state of the game is cooperative. Next
we define some notation we need to define strategies:
Notation B.7. As in the text, Zi denotes the share of firms with v*v such that bank i is in
their core group. Furthermore, we denote by Z i the fraction of firms with v*v that
have a core group where relationship bank i is not a member.
Note that 1 Zi Zi is the fraction of firms with v*v who did not form a core
group. Hence, if all firms formed a core group then Zi ¼ 1 Zi . Furthermore, 1
Zi Zi ¼ 1 Zj Zj for all i, j. We can now define the symmetric strategy
combination C (for ‘cooperative’).
Definition B.8. Call strategy combination C the following combination of strategies:
Entry into the relationship segment (only fringe banks move):

1. If ft ¼ fc ,
– (i) enter if i ¼ m þ 1; . . . ; min½mzp ; m;

– (ii) remain a fringe bank if i > min½mzp ; m.
2. Otherwise do not enter the relationship segment.
1. (Establishing relationships)
– If ft ¼ fc then play
R for v*v;
Ri ðvÞ ¼
0 for v < v:
– Otherwise, play according to P.

2. (Fee offers)
– If ft ¼ fc and

d k lc V 1
ðB:1Þ fr R *max f r Zj 1 lc þ Zj ðlc aÞ V
1d m k j k
holds, then play
ðlc ; 1; 0Þ if Zi > 0;
ðlci ; lnc nr
i ; li Þ ¼ ð1; 1; 0Þ: if Zi ¼ 0:
– Otherwise play according to P.
For all fringe banks j play lfj ¼ 0.

Condition (B.1) says that bank i will not undercut in period t provided that continued
cooperation is more profitable than undercutting, given period’s t ex ante fee
offers. Lemma B.9 characterizes the outcome path induced by C:
Lemma B.9. Along the path induced by C

(i) all entry occurs in the first period;
(ii) all relationship banks i play ðlci ; lnc nr c
i ; li Þ ¼ ðl ; 1; 0Þ for all t;
(iii) all relationship banks have the same market share;
(iv) all fringe banks play lfj ¼ 0 for all t.
Proof. The proof is straightforward and we leave it to the reader. &
We now state and prove the main result of this appendix:
Proposition B.10. Let ðlc ; mÞ 2 L. Then, strategy combination C is a subgame perfect

equilibrium in the infinitely repeated game.
Proof. To prove this Proposition, we show that players’ strategies are optimal after any
history. Since this is a repeated game with bounded payoffs, it suffices to show that one-
step unilateral deviations from strategies are not profitable after any history (Hendon et
al. [1996]).
Now according to C histories can be classified into two groups, cooperative and non
cooperative.
Non-cooperative histories (ft 5 fnc):

(i) At the beginning of the period it is optimal for any fringe bank not enter the
relationship segment, which would leave long-run losses equal to E.
(ii) We know that when all other relationship banks are playing according to
P in the one-period game, it is optimal for relationship bank i to do the same.
Since all relationship banks will play according to P forever after, it is also
optimal for relationship bank i to play according to P in any period of the
repeated game.
(iii) Last, note that playing lfj ¼ 0 is optimal for fringe banks in the one-period
game, hence it is also optimal to play so in the repeated game.
Cooperative histories (ft 5 fc):
(i) It is optimal for fringe banks m þ 1; . . . ; min½mzp ; m to enter the relation-
ship segment, since according to strategies, there will be cooperation in the
future. For fringe banks min½mzp ; m; . . . it is optimal not to enter the
relationship segment, since further entry would switch the state to non-
cooperative, in which case long-run profits gross of entry cost E are zero; or else
raise m above mzp.
(ii) Next consider histories after which the state of the game is cooperative (f 5 fc)
and relationship banks must decide whether to establish relationships. Clearly a
relationship bank cannotgain bydeviating and setting RðvÞ ¼ 0 for firms such
c
that v*v (it would lose l kV R per firm in the current period according to
strategies) or by sinking the relationship cost with a firm such that v < v (since
such a firm will not c be successful
in establishing a core group according to
strategies because lkv R < 0). Hence setting RðvÞ ¼ R for v*v and RðvÞ ¼ 0
for v < v is optimal.
(iii) Now consider decision by relationship bank i if condition (B.1) does not hold.
Then all relationship banks play according to P, from which we know it is
optimal not to deviate.
(iv) Now consider histories after which the state of the game is cooperative (ft 5 fc),
relationship banks must make fee offers, Zi > 0 and condition (B.1) holds. Then
bank i can not gain by undercutting (as condition [B.1] implies). On the other
hand, if bank i would set lci > lc or lnc a
i 2 ðl1 þ a; 1Þ it would not get any further
nr
deals; and setting li > 0 would not get any further deals either. Thus, playing
ðlci ; lnc nr c
i ; li Þ ¼ ðl ; 1; 0Þ is optimal. Moreover, if Zi ¼ 0 but Zi > 0 it would not
gain deviating from setting ðlci ; lnc nr
i ; li Þ ¼ ð1; 1; 0Þ. Last, if Zi ¼ Zi ¼ 0
relationship bank i cannot gain by undercutting.
(v) Last, note that playing lfj ¼ 0 is optimal for fringe banks in the one-period
game, hence it is also optimal to play so in the repeated game. This completes the
proof. &
C. Comparative equilibria
In this appendix we obtain the comparative equilibria derivatives that are presented in
the text. All are obtained by totally differentiating the identity
c
d lV
k R ½ðm 1Þlc ðm kÞaV 0
1d k
which is derived from the no undercutting condition (3.4). Totally differentiating
@V lc
this identity with respect to lc, m, k and a, recalling that eV;l @lc , and simplifying
V
yields

d a
ðm 1Þ ð1 eV;lc Þ c ðm kÞeV;lc Vdlc ðlc aÞVdm
1d l

@V d @V
½ðm 1Þlc ðm kÞ þ aV lc R dk þ ðm kÞVda 0;
@k 1d @k
which can be rewritten as
ðC:1Þ Adlc Bdm Cdk þ Dda 0:
It will be useful to sign the coefficients in identity (C.1). Clearly B 4 0 (since lc 4 a)
and D 4 0 (since m 4 k). To sign C note first that ðm 1Þlc ðm kÞ ¼ k lc > 0.
R v
Moreover, since @V@k ¼ 2lc ¼ 2k > 0 it follows that
@V v
lc R ¼ lc R < 0
@k 2k
v
since lc k R ¼ 0 by the definition of v. It follows that C 4 0. Finally, noting that
v
eV;lc ¼ vþv, A an be rewritten as

1 d a
ðm 1Þ v c ðm kÞv ;
vþv 1d l
whose sign is ambiguous but positive if a is sufficiently small. Now if A 4 0 then the
following result follows.
Proposition C.1. If A 4 0 then:

dm A
¼ > 0;
dlc B
dlc C
¼ > 0;
dk A
dlc D
¼ < 0;
da A
dm C
¼ < 0;
dk B
dm D
¼ > 0:
da B
Proof. By direct substitution. &
D. Nonprice competition
D.1. Nonprice competition I: analysis To study nonprice competition, it is
assumed that at the beginning of each period, each relationship bank spends a total
amount E i f r Ei in sales expenditure (or, ‘sales effort’) to contact firms. Sales efforts
result in contacts with firms according to the following assumptions:
Assumption D.1. (i) A relationship bank spends at least E ¼ f r E in sales expenditures

to contact firms.
(ii) Firms contact relationship banks who have spent the k’th highest amounts in
sales expenditures. If y > 1 banks spend the kth highest amount, then each bank
contacts a firm with probability 1y.
(iii) If fewer than k banks spend E or more, then firms contact no bank.
It will be useful to number relationship banks according to their sales expenditures
and adopt the following notational convention: E 1*E 2*. . .*E k*. . .*E m : That is,
relationship bank 1 spends the (weakly) largest amount in sales expenditures. Note that
according to this convention Ek is by definition the (possibly not unique) kth highest
sales expenditure.
The thrust of assumption D.1 is to make the marginal gains in relationships very
sensitive to sales expenditures at the margin. As will be seen below, this assumption is
not extreme. Specifically, one consequence of nonprice competition is to yield an upper
bound on lc. It turns out that any assumption that makes the marginal benefits to
relationships less sensitive to sales effort at the margin yields an upper bound on lc that
is smaller.
We now study a symmetric equilibrium where all relationship banks spend E*E
(strategy combinations that support this equilibrium are rigorously constructed
below). Ceteris paribus, sales efforts reduce the profits from cooperating on prices by e
every period.
Hence,
the long-run payoff from sticking to the implicit contract is
d r k lc V
1d f m k R E (recall that if all relationship banks spend the same amount in
sales effort, they obtain the same number of relationships). On the other hand, at the
time that a relationship bank decides to undercut, sales expenditures are sunk, just as
relationship costs R are. Hence, the gains from undercutting are not affected by sales
efforts, and the no-undercutting condition reads
c
d k lV fr
ðD:1Þ fr R E * ½ðm 1Þlc ðm kÞaV:
1d m k m
This condition is very similar to (3.4) except for term E on the left-hand-side.cCall LE
@l
the set of points in the space (lc,m) such that condition D.1 holds. Then @E > 0 and
@m E
@E < 0, so that L L. Thus:
Proposition D.2. When nonprice competition by banks increases, fees tend to be higher
for a given number of banks m. Conversely, the market tends to be more concentrated
for a given lc.
Proof. To prove this result, let condition (D.1) hold as an identity and then totally
differentiate with respect to lc, m, and E. This yields
d d
Adlc ðlc a þ EÞdm mdE ¼ 0;
1d 1d
where A is defined as in Appendix B. Setting dm 5 0, straightforward manipulations
yield
dlc d m
¼ > 0:
dE 1 d A
Similarly, setting dlc 5 0 and rearranging yields
dm d m
¼ < 0:
dE 1 d lc a þ 1d
d
This completes the proof. &
The intuition behind Proposition D.2 should be clear by now: sales efforts, and, more
generally, any sunk expenditures, reduce the gains from the implicit contract but not
those of cheating. Adhering to the price norm must then be made more attractive,
which is achieved either by exit (and increased concentration) or higher fees. The
invariance of the gains of cheating to sunk expenditures has another implication:
Result D.3. Relationship banks do not compete away rents with sales efforts.
In many models ex-ante nonprice competition is a mechanism to dissipate ex-post

rents. This does not occur here because the incentive to make relationship-investments
relies on rents. Thus, sales efforts do not do away with soft price competition, which is
the source of rents in this model.
Result D.3 has another interesting implication on cross-subsidies among different
lines of business for multiproduct banks. Suppose that one way to attract clients to do
‘high margin’ deals is to sell them other commodity services at fees below cost. This
cross subsidization should not dissipate rents in the high margin activities performed by
the bank. If that were the case, then relationship banks would want to unilaterally
deviate from price norms, thus undermining relationships.
There is one sense, however, in which nonprice competition restrains the market
power of banks. As is shown now, nonprice competition imposes an upper bound on
the fee that investment banks can charge in equilibrium. To see this, note that an
additional way for a relationship bank (say, bank 1) to deviate from equilibrium is by
escalating sales expenditures. In that case bank 1 establishes a relationship with every
firm and increases the proportion of firms in its portfolio of relationships, Z1, from mk to
1. If this occurs, so that bank 1 establishes a relationship with every firm and
E2 ¼ Em ¼ E, there are k 1 relationships left for m 1 banks for any given firm. Zi
then falls from mk to m1
k1
for i 5 2,3,. . .,m. Thus, if every bank sticks to lci ¼ lc ; and then
every bank, including bank 1, pays Ei 5 E from t þ 1 on, then bank 1 makes a one-time
gain of slightly less than
c c
r lV r k l V
f R E1 f R E
k m k
c
lV k
Ef r R ð1 Þ
k m
c
lV
fr R DZ1 > 0;
k
since relationship bank 1 sets E1 shade above E. But, of course, if this is so, then playing
E1 5 E forever cannot be the outcome of an equilibrium, since every bank would have
an incentive to unilaterally increase sales efforts every period. It follows that

relationship banks will not escalate sales expenditures only if that makes undercutting
profitable. This is so if (given m) lc is such that the no-undercutting condition (3.4) does
k1
not hold. Since Zi (for i 5 2,3,. . .,m) falls to m1 when bank 1 escalates its sales efforts,
c
continued cooperation is not profitable if l is such that
c
d r k l V fr
ðD:2Þ f R E < ½ðm rÞlc ðm krÞaV;
1d m k m
with r mkm1
k1
.
Figure 6 plots the right and left hand sides of condition (D.2) as a function of lc. As
can be seen, condition (D.2) holds for lc < lc , with lc Fderived straightforwardly from
condition (D.2) Fbeing equal to
d
ð1dÞkR ðm krÞaV
d
:
1d ðm rÞ V
It is straightforward to show that lc > lc (see below). Thus:
Result D.4. Competition for establishing relationships sets an upper bound on fees
charged by relationship banks.
The intuition behind Result D.4 is as follows. When lc is too high, adhering to the
implicit contract is very attractive and it is not profitable to undercut even if a unilateral
deviation in bank’s 1 sales efforts reduces everybody’s else’s market share. But this, of
course, cannot occur in equilibrium because then every relationship bank would like to
unilaterally increase sales effort. This determines the upper bound on lc. In other
words, fees that are too high make it profitable to escalate sales efforts.
Note that the upper bound on lc would be smaller had a unilateral increase in sales
effort yielded a smaller increase in the market share of bank 1. This follows because the
gains from undercutting (the left hand side in condition (D.2) would then be
correspondingly smaller. Therefore, there is no loss of generality in assuming that a
marginal increase in sales effort will enable the escalating relationship bank to grab all
relationships.
This role of nonprice competition in reducing rents is similar to that obtained in
standard models, where ex-ante nonprice competition can eliminate ex-post rents. The
difference is that here nonprice competition only restricts the size of such rents, and
does not eliminate them. The reason, again, is that rents are necessary to support
efficient relationship investments, a feature that is absent when rents are merely a
consequence of market power.
D.2. Nonprice competition II: a formal game. This subsection presents a formal game
that underpins the analysis in the previous subsection. To analyze nonprice competition
we replace the first stage of the one period game. Instead of firms randomly choosing k
relationship banks, we have relationship banks choosing sales effort Ei.
It is easy to show that in the one-period game relationship banks will not spend
anything in sales efforts and that no relationships will be established. Thus, as before,
the equilibrium in the one period game can be used as a subgame perfect punishment.
Profits RHS
LHS
0 λc λc
Figure 6
Non-price competition and the upper bound on fees
Locus LHS plots a relationship bank’s long-run profits from continued cooperation as a function
of the fee lc. Locus RHS plots a relationship bank’s one-time profits when unilaterally
undercutting after another relationship bank has unilaterally escalated sales efforts. The
intersection of LHS and RHS is the upper bound on the fee that can be charged in any cooperative
equilibrium.
Call again this subgame perfect punishment P. Next define ‘undercutting’ and
‘cooperative’ and ‘non-cooperative’ states exactly as in the previous section.53 Last, we
need one piece of additional notation to keep track of the fraction of firms that contact
relationship bank i in response of i’s sales effort:
Notation D.5. We denote by gi the fraction of firms with v*v that contact relationship
bank i after i has chosen Ei.
Recall that, by definition, E1*E2*. . .*Em : Hence, our assumptions imply that gi
is a function gi : Rm
þ ! ½0; 1 such that
8
>
> 0 if Ei < E or Ei < Ek ;
<1
if Ei ¼ Ek*E and y banks make the kth largest sales effort;
ðD:3Þ gi ðE1 ; :::; Ei ; :::; Em Þ¼ y1
>
> if E1 ¼ Em*E;
:m
1 if Ei > Ek and Ei*E:
Function
Pgi summarizes how banks sales efforts bring about contacts with firms.
Note that mi¼1 gi ¼ k if Ek*E. We can now define a strategy combination that is a
subgame perfect equilibrium in the game with sales effort.
Definition D.6. Call strategy combination C the following combination of strategies:

1. (Sales effort)
If ft ¼ fc then play Ei ¼ E*E;
Otherwise, play Ei ¼ 0.
53
Note that this implies that the state of the game is determined only by the pricing behavior
of relationship banks, and not by their sales efforts.
2. (Establishing relationships)
If ft ¼ fc and
c
d r k l V 1
ðD:4Þ f R E *max f r gj 1 lc þ ð1 gj Þðlc aÞ V
1d m k j k
then play
R for v*v;
Ri ðvÞ ¼
0 for v < v:
Otherwise, play Ri ðvÞ ¼ 0 for all v 2 ½v; v

3. (Fee offers)
If ft ¼ fc and
c
d k lV 1
ðD:5Þ fr R E *max f r Zj 1 lc þ Zj ðlc aÞ V
1d m k j k
then play
ðlc ; 1; 0Þ if Zi > 0;
ðlci ; lnc nr
i ; li Þ ¼ ð1; 1; 0Þ if Zi ¼ 0:
Otherwise play according to P.

For all fringe banks j play lfj ¼ 0.
Like in the previous section, condition (D.5) says that bank i will not undercut in
period t provided that continued cooperation is more profitable than undercutting. Note
that this no–undercutting condition is exactly the same as condition (B.1) in the previous
section, except for the fact that sales effort expenditures E are included in the left-hand
side of condition (D.5). Lemma B.9 characterizes the outcome path induced by C.
Lemma D.7. Along the path induced by C
(i) all relationship banks i select Ei ¼ E;

(ii) all relationship banks i play ðlci ; lnc nr c
i ; li Þ ¼ ðl ; 1; 0Þ for all t;
(iii) all relationship banks have the same market share;
(iv) all fringe banks play lfj ¼ 0 for all t.
Proof. The proof is straightforward and we leave it to the reader. &
We now state and prove the main result of this section:
Proposition D.8. Let ðlc ; mÞ 2 LE . Then strategy combination C is a subgame perfect

equilibrium in the infinitely repeated game with sales effort. Moreover,

d k

c R þ E m1 ðm krÞaV
lc )lc < l ¼ 1d m V d ;
m 1d ðm rÞ
with r mkm1
k1
.

Proof. To prove this proposition, we show that the players’ strategies for the repeated
game are optimal after any history. Again, since this is a repeated game with bounded
payoffs, it suffices to show that one-step unilateral deviations from strategies are not
profitable after any history.
Now as before, according to C histories can be classified in two groups, cooperative
and non-cooperative. Consider, then, histories after which the state of the game is non-
cooperative (ft ¼ fnc ). We know that when all other banks are playing according to P
in the one-period game, it is optimal for bank i to do the same. Since all banks will play
according to P forever after, it is also optimal for bank i to play according to P in any
period of the repeated game.
Now consider histories after which the state of the game is cooperative (ft 5 fc)
and relationship banks must make fee offers. Then, with the exception of the sales
effort E on the left-hand side of (D.5), the continuation game’s strategies look exactly
as in the game without sales effort. Hence, one-shot deviations from strategies are
unprofitable.
Next consider histories after which the state of the game is cooperative and
relationship banks must decide whether to establish relationships with firms. If
condition (D.5) holds, and all relationship banks conform to strategies, then gj ¼ Zj
(that is, all relationship banks establish relationships with all firms they contacted) and
1 gj ¼ Zj for all j. Hence condition (D.5) also holds and cooperation continues. A
unilateral deviation by relationship bank i not establishing relationships is therefore
unprofitable (see the proof of Proposition B.10). On the other hand, if condition (D.4)
does not hold, then no relationship bank establishes relationships, and so it is optimal
for i not to establish them either.
Next consider sales effort decisions when the state of the game is cooperative.
According to strategies all relationship banks play Ej 5 E. Hence, a unilateral deviation
is to play E1 6¼ E. If E1 > E then E1 > E2 ¼ . . . ¼ Em . It follows from (D.3) that g1 5 1
k1
and g2 ¼ . . . ¼ gm ¼ m1 . Such unilateral deviation is unprofitable if
c
d r k l V fr
ðD:6Þ f R E < ½ðm rÞlc ðm krÞaV;
1d m k m
otherwise it would pay to deviate to increase market share for one time. Now some
straightforward algebra shows that condition (C.4) holds if and only if lc < lc .
Last, consider playing Em o E. Then gm 5 0 and clearly condition (D.5) does not
hold, since it holds with equality with lc ¼ lc and minj gj ¼ m1
k1
. Hence, if relationship
bank m deviates selecting Em o E, then no relationships are established in that period
and profits are foregone. This completes the proof. &
REFERENCES
Allen, F. and Gale, D., 1995, ‘A Welfare Comparison of Intermediaries and

Financial Markets in Germany and the U.S.,’ European Economic Review, 39, pp.
179–209.
Allen, F. and Gale, D., 2000, Comparing Financial Systems (Cambridge: MIT Press).
Anand, B. and Galetovic, A., 2000, ‘Information, Non-excludability and Financial
Market Structure,’ Journal of Business, 73, pp. 357–402.
Anand, B. and Galetovic, A., 2001, ‘Investment Banking and Security Market
Development: Does Finance Follow Industry?’, working paper WP/01/90, Interna-
tional Monetary Fund. Downloadable in http://www.imf.org/external/pubs/ft/wp/

2001/wp0190.pdf.
Aoki, M. and Dinç, S., 1997, ‘Relational Financing as an Institution and its Viability
under Competition,’ CEPR publication 488.
Auerbach, J. and Hayes, S., 1986, Investment Banking and Diligence: What Price
Deregulation (Boston: Harvard Business School Press).
Baker, W., 1990, ‘Market Networks and Corporate Behavior,’ American Journal of
Sociology, 96, pp. 589–625.
Benveniste, L.; Busaba, W. and Wilhelm, W., 2002, ‘Information Externalities and the
Role of Underwriters in Primary Equity Markets,’ Journal of Financial Intermediation,
11, pp. 61–86.
Benveniste, L.; Ljungqvist, A.; Wilhelm, W. and Yu, X., 2003, ‘Evidence of Information
Spillovers in the Production of Investment Banking Services,’ Journal of Finance, 58,
pp. 577–608.
Berger, A., 1998, ‘The ‘Big Picture’ of Relationship Finance,’ in Blanton, J. L., Williams,
A. and Rhine, S-L. (eds.) Business Access to Capital and Credit (A Federal Research
System Research Conference).
Berglöf, E. and von Thadden, E. L., 1994, ‘Short-Term versus Long-Term Interests:
Capital Structure with Multiple Investors,’ Quarterly Journal of Economics, 109,
pp. 1055–1084.
Bhattacharyya, S. and Nanda, V., 2000, ‘Client Discretion, Switching Costs, and
Financial Innovation,’ Review of Financial Studies, 13, pp. 1101–1127.
Bliss, C. and Di Tella, R., 1997, ‘Does Competition Kill Corruption?,’ Journal of Political
Economy, 105, pp. 1001–1023.
Boot, A., 2000, ‘Relationship Banking: What Do We Know?,’ Journal of Financial
Intemediation, 9, pp. 7–25.
Boot, A. and Thakor, A., 1994, ‘Moral Hazard and Secured Lending in an Infinitely
Repeated Credit Market Game,’ International Economic Review, 35, pp. 899–920.
Boot, A. and Thakor, A., 2000, ‘Can Relationship Banking Survive Competition?,’
Journal of Finance, 55, pp. 679–713.
Carosso, V., 1970, Investment Banking in America (Cambridge, Massachusetts: Harvard
University Press).
Chemmanour, T. and Fulghieri, P., 1994, ‘Reputation, Renegotiation and the Choice
Between Bank Loans and Publicly Traded Debt,’ Review of Financial Studies, 7,
pp. 475–506.
Chan, Y.; Greenbaum, S. and Thakor, A., 1986, ‘Information Reusability, Competition,
and Bank Asset Quality,’ Journal of Banking and Finance, 10, pp. 243–253.
Chen, H. and Ritter, J., 2000, ‘The Seven Per Cent Solution,’ Journal of Finance, 55,
pp. 1105–1131.
Collis, D., 1992, ‘Saatchi & Saatchi Co.: PLC, Corporate Strategy.’ Harvard Business
School Case No. 792-056.
Crane, R. and Eccles, R., 1993, ‘Customer Relationships in the 1990s,’ in Hayes, S. (ed.),
Financial Services: Perspectives and Challenges (Boston: Harvard Business School
Press).
Diamond, D., 1991, ‘Monitoring and Reputation: The Choice Between Bank Loans and
Privately Placed Debt,’ Journal of Political Economy, 99, pp. 689–721.
Detragiache, E.; Garella, P. and Guiso, L., 2000, ‘Multiple versus Single Banking
Relationships: Theory and Evidence,’ Journal of Finance, 55, pp. 1133–1161.
Dunbar, C., 2000, ‘Factors Affecting Investment Bank Initial Public Offering Market
Share,’ Journal of Financial Economics, 55, pp. 3–41.
Dutta, P. and Madhavan, A., 1997, ‘Competition and Collusion in Dealer Markets,’

Eccles, R. and Crane, R., 1988, Doing Deals: Investment Banks at Work (Boston: Harvard
Business School Press).
Fudenberg, D. and Maskin, E., 1986, ‘The Folk Theorem in Repeated Games with
Discounting or with Incomplete Information,’ Econometrica, 54, pp. 533–554.
Green, E. and Porter, R., 1984, ‘Non-cooperative Collusion Under Imperfect Price
Information,’ Econometrica, 52, pp. 87–100.
Greenbaum, S.; Kanatas, G. and Venezia, I., 1989, ‘Equilibrium Loan Price
Under the Bank-Client Relationship,’ Journal of Banking and Finance, 13,
pp. 221–235.
Grossman, S. and Hart, O., 1986, ‘The Costs and Benefits of Ownership: A
Theory of Vertical and Lateral Integration,’ Journal of Political Economy, 94, pp.
691–719.
Hansen, R. S., 2001, ‘Do Investment Banks Compete in IPO?: The Advent of the ‘‘7%
Plus’’ Contract,’ Journal of Financial Economics, 59, pp. 313–346.
Hart, O., 1995, Firms, Contracts and Financial Structure (Oxford: Oxford University
Press).
Hart, O. and Moore, J., 1990, ‘Property Rights and the Nature of the Firm,’ Journal of
Political Economy, 98, pp. 1119–1158.
Hayes, S. and Hubbard, 1990, Investment Banking: A Tale of Three Cities (Boston:
Harvard Business School Press).
Hayes, S. and Regan, A., 1993, ‘Securities Underwriting and Investment Banking
Competition,’ in Hayes, S. (ed.), Financial Services: Perspectives and Challenges
(Boston: Harvard Business School Press).
Hayes, S.; Spence, M. and Marks, D., 1983, Competition in the Investment Banking
Industry (Cambridge: Harvard University Press).
Hellwig, M., 1991, ‘Banking, Financial Intermediation and Corporate Finance,’ in
Giovannini, A. and Mayer, C. (eds.), European Financial Inegration (New York:
Cambridge University Press).
James, C., 1987, ‘Some Evidence on the Uniqueness of Bank Loans,’ Journal of Financial
Economics, 19, pp. 217–235.
James, C., 1992, ‘Relationship-Specific Assets and the Pricing of Underwriter Services,’
Kang, J.; Shivdasani, A. and Yamada, T., 2000, ‘The Effect of Bank Relations on
Investment Decisions: An Investigation of Japanese Takeover Bids,’ Journal of
Finance, 55, pp. 2197–2218.
King, C.; Silk, A. J. and Ketelhohn, N., 2003, ‘Knowledge Spillovers and Growth in the
Disagglomeration of the U.S. Advertising Agency Industry,’ Journal of Economics &
Management Strategy, 12, 3, pp. 327–362.
Krigman, L.; Shaw, W. and Womack, K., 2001, ‘Why do Firms Switch Underwriters,’
Journal of Financial Economics.
Lee, I.; Lochhead, S.; Ritter, J. and Zhao, Q., 1996, ‘The Costs of Raising Capital,’
Journal of Financial Research, 19, pp. 59–74.
Ljungqvist, A.; Jenkinson, T. and Wilhelm, W., 2003, ‘Global Integration in Primary
Equity Markets: The Role of U.S. Banks and U.S. Investors,’ Review of Financial
Studies, 16, pp. 63–99.
Lummer, S. and McConnel, J., 1989, ‘Further Evidence on the Bank Lending Process and
the Capital Market Response to Bank Loan Agreements,’ Journal of Financial
Economics, 25, pp. 99–122.
Mas-Collell, A.; Whinston, M. and Green, J., 1995, Microeconomic Theory (New York:
Oxford University Press).
Matthews, J., 1994, Struggle and Survival on Wall Street (New York: Oxford University
Press).

Mayer, C., 1988, ‘New Issues in Corporate Finance,’ European Economic Review, 32,
pp. 1167–1189.
Moel, A., 2001, ‘The Role of American Depositary Receipts in the Development of
Emerging Markets,’ Economia, 2, pp. 209–273.
Nanda, V. and Warther, V., 1998, ‘The Price of Loyalty: An Empirical Analysis of
Underwriting, Relationships and Fees,’ manuscript, University of Chicago, Graduate
School of Business.
Ongena, S. and Smith, D., 2000, ‘Bank Relationships: A Review,’ in Zenios, S. A. and
Harker, P. (eds.), Performance of Financial Institutions (New York: Cambridge
University Press).
Ongena, S. and Smith, D., 2001, ‘The Duration of Bank Relationships,’ Journal of
Financial Economics, 61, pp. 449–475.
Persons, J. and Warther, V., 1997, ‘Boom and Bust Patterns in the Adoption of Financial
Innovations,’ Review of Financial Studies, 10, pp. 939–968.
Petersen, M. and Rajan, R., 1994, ‘The Benefits of Lending Relationships: Evidence from
Small Business Data,’ Journal of Finance, 49, pp. 3–37.
Petersen, M. and Rajan, R., 1995, ‘The Effect of Credit Market Competition on Lending
Relationship,’ Quarterly Journal of Economics, 110, pp. 407–443.
Pichler, P. and Wilhelm, W., 2001, ‘A Theory of the Syndicate: Form Follows Function,’
Rajan, R., 1992, ‘Insiders and Outsiders: The Choice between Informed and Arm’s-
Length Debt,’ Journal of Finance, 47, pp. 1367–1400.
Rajan, R., 1996, ‘The Entry of Commercial Banks into the Securities Business:
A Selective Survey of Theories and Evidence,’ in Saunders, A. and Walter,
I. (eds.), Universal Banking: Financial System Design Reconsidered (Homewood,
Illinois: Irwin).
Ritter, J., 1987, ‘The Costs of Going Public,’ Journal of Financial Economics, 19,
pp. 269–281.
Rolfe, J. and Troob, P., 2000, Monkey Business: Swinging Through the Wall Street Jungle
(New York: Warner Books).
Santomero, A. and Babel, D. F., 2001, Financial Markets: Instruments and Institutions,
(McGraw-Hill Irwin).
Segal, I. and Whinston, M., 2000, ‘Exclusive Contracts and Protection of Investments,’
Rand Journal of Economics, 31, pp. 603–633.
Sharpe, S., 1990, ‘Asymmetric Information, Bank Lending and Implicit Con-
tracts: A Stylized Model of Customer Relationships,’ Journal Finance, 45, pp.
1069–1087.
Shockley, R. and Thakor, A., 1997, ‘Bank Loan Commitments: Data, Theory and Tests,’
Journal of Money, Credit and Banking, 29, pp. 517–534.
Sutton, J., 1998, Technology and Market Structure: Theory and History (Cambridge,
Massachusetts: MIT Press).
Sutton, J., 1991, Sunk Costs and Market Structure: Price Competition, Advertising, and the
Evolution of Concentration (Cambridge, Massachusetts: MIT Press).
Tufano, P., 1989, ‘Financial Innovation and First-Mover Advantages,’ Journal of
Financial Economics, 25, pp. 213–240.
von Thadden, E., 1995, ‘Long-term Contracts, Short-term Investment and Reputation,’
Review of Economic Studies, 62, pp. 557–575.
von Thadden, E., 1998, ‘Asymmetric Information, Bank Lending, and Implicit
Contracts: The Winner’s Curse,’ working paper, University of Lausanne.
Wilhelm, W., 1999, ‘Internet Investment Banking: The Impact of Information
Technology on Relationship Banking,’ Journal of Applied Corporate Finance, 12, pp.
21–27.

Wilhelm, W. J. and Downing, J. D., 2001, Information Markets: What Businesses Can
Learn from Financial Innovation (Cambridge, Massachusetts: Harvard Business School
Press).
Yafeh, Y. and Yosha, O., 2001, ‘Industrial Organization of Financial Systems and
Strategic Use of Relationship Banking,’ European Finance Review, 5, pp. 63–78.

Anand 2006

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Anand 2006

Uploaded by

Document Information

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

Anand 2006

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF INDUSTRIAL ECONOMICS 0022-1821

Volume LIV June 2006 No. 2

RELATIONSHIPS, COMPETITION AND THE STRUCTURE

It is well known that competition can destroy incentives to invest in ﬁrm-

THE INDUSTRIAL ORGANIZATION OF PROFESSIONAL SERVICE MARKETS has been

wHarvard University, Soldiers Field Road, Boston, Massachusetts, 02163 U.S.A.

relationships preserved in the face of competition? Indeed, a key insight from

volume (in logs)

These facts raise interesting questions. First, how does competition

Concentration is unrelated to market size: Market size has grown

volume (in logs)

1989 1993 1995 1996 1997

has consistently underwritten more than 70% of securities issued.

a one-time entry cost, E. We call banks who invested in the relationship

II. THE MODEL

II(i)(ii). The Technology of Relationships Three characteristics of

II(ii). Model Description

Now in each period, the timing of the game is as follows:

II(iii). The Well-Known Tension in a One-Period Model

III. RELATIONSHIPS AND COMPETITION

III(i). The Repeated Game

accomodated as long as it is proﬁtable.33 Because it is tedious to specify

III(ii). Four Key Conditions

setting lci slightly below lc, bank i can get an additional lc V  l kV ¼

Note that this is a standard sustainability condition: the permanent ﬂow

III(iii). Investment Banking Structure

Proposition 3.4. Relationships will be established only if there are few

Proof. Fix the equilibrium number of banks, m. Then:

It is straightforward to see that the upper bound on mi must be less than 1.

market structure. To see this, assume, to the contrary, no externalities

III(iv). Relationships and the Intensity of Competition

Result 3.6 appears counterintuitive because lower switching costs for

dlc ½ðm  1Þlc  ðm  kÞa2kv d cv

which is negative. An increase in k therefore reduces L, the set of pairs (l, m)

Result 3.8. An increase in the intensity of competition need not destroy

III(v). Robustness and Extensions

III(v)(i). Which Assumptions Drive the Results? Non-veriﬁability of

Result 3.9 (Relationships and perfect competition). If relationships were

Result 3.9 implies that imperfect competition is a consequence of

need not be observable to deﬁne a strategy that punishes deviators.

IV(i). Off-Equilibrium Episodes

IV(ii). Applicability to Other Industries

IV(ii)(i). Professional Service Industries The analysis here may be

consulting, law, advertising) that share characteristics with investment

IV(ii)(ii). Commercial Banking It is well known that relationships are

The tension between relationships and competition is a long-standing one.

A. Equilibrium in the one-period game

Proposition A.1. In any subgame perfect equilibrium, no relationships are established

Proof. Suppose, by way of contradiction, that bank i establishes a relationship with a

r Blackwell Publishing Ltd. 2006.

Deﬁnition B.1. Call strategy combination P (for ‘punishment’) the following

For all relationship banks i

For all fringe banks j, lfj ¼ 0.

r Blackwell Publishing Ltd. 2006.

Thus, since playing strategy combination P forever is clearly a subgame perfect

Deﬁnition B.8. Call strategy combination C the following combination of strategies:

Entry into the relationship segment (only fringe banks move):

– (i) enter if i ¼ m þ 1; . . . ; min½mzp ; m;

setting lci slightly below lc, bank i can get an additional lc V l kV ¼

dlc ½ðm 1Þlc ðm kÞa2kv d cv