Do Investment Banks Matter for M&A
Returns?
Jack Bao
Fisher College of Business, Ohio State University
Alex Edmans
Wharton School, University of Pennsylvania and NBER
We document a significant investment bank fixed effect in the announcement returns of
M&A deals. The interquartile range of bank fixed effects is 1.26%, compared with a full-
sample average return of 0.72%. The results remain significant after controlling for the
component of returns attributable to the acquirer. Our findings suggest that investment
banks matter for M&A outcomes, and contrast earlier studies that show no positive link
between various measures of advisor quality and M&A returns. Differences in average re-
turns across banks are also persistent over time and predictable from prior performance.
Clients do not chase past returns, which may explain why persistence exists in M&A per-
formance while it is absent in mutual funds. (
JEL G24, G34)
Mergers and acquisitions (M&A) are among the most critical decisions a CEO
can make. Successful mergers can create substantial synergies, while mis-
guided acquisitions can lead to misallocation of companies to parents unable
to reap their full potential. In addition to these large effects on shareholder
value, a bad acquisition also increases the CEO��s risk of being fired (Lehn
and Zhao 2006). A prominent example is the departure of Carly Fiorina from
Hewlett Packard, which was widely attributed to her acquisition of Compaq.
The quality of M&A transactions is also of great importance to the economy
as a whole. The total value of M&A announced by a U.S. acquirer in 2007 was
$2.1 trillion, around 15% of GDP.
We thank an anonymous referee, the Editor (Alexander Ljungqvist), Franklin Allen, Alexander Dyck, Florian
Ederer, Vincent Glode, Itay Goldstein, Todd Gormley, Dirk Hackbarth, Kewei Hou, Jeff Jaffe, Dirk Jenter,
Adam Kolasinski, Anna Kovner, Gilberto Loureiro, Qingzhong Ma, Stew Myers, Micah Officer, David
Pedersen, Nagpurnanand Prabhala, Jun Qian, Michael Roberts, Rob Stambaugh, Ren�� Stulz, Torben Voetmann,
Mike Weisbach, Fei Xie, and seminar participants at the 2008 WFA meetings, 2008 FIRS meetings, 2007
EFA meetings, BYU, Exeter, MIT Sloan, New York Fed, Toronto, Vanderbilt, Virginia, and Warwick for
valued input. Special thanks to Jerry Hoberg, Wei Jiang, and Luke Taylor for extensive insightful com-
ments. We are grateful to Michael Roberts for generously providing Dealscan-Compustat data, and to
Cong Wang for data that were used in an earlier draft of this project. This article was previously circu-
lated under the titles ��Do Investment Banks Have Skill? Performance Persistence of M&A Advisors�� and
��How Should Acquirers Select Advisors? Persistence in Investment Bank Performance.�� Send correspon-
dence to Alex Edmans, 3620 Locust Walk, Philadelphia, PA 19104; telephone: (215) 746-0498. E-mail:
aedmans@wharton.upenn.edu.
c The Author 2011. Published by Oxford University Press on behalf of The Society for Financial Studies.
All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com.
doi:10.1093/rfs/hhr014
Advance Access publication May 6, 2011
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
Since CEOs make M&A decisions rarely, they typically lack experience
and seek counsel from investment banks. The
skilled-advice hypothesis is that
banks help clients identify synergistic targets and negotiate favorable terms.
If banks indeed provide valuable advice, it is reasonable to expect that the
highest-quality advisors lead to the best outcomes. However, existing research
generally fails to find such a relationship. Bowers and Miller (1990) and Michel,
Shaked, and Lee (1991) measure an advisor��s quality by the prestige of its
name and find no link with acquirer returns; Rau (2000) uses market share to
measure quality and documents a negative relationship. Servaes and Zenner
(1996) find no benefit of hiring any advisor at all, compared with executing the
deal in-house.1 These findings instead appear to support the
passive-execution
hypothesis, that banks do not provide true advice but are simply ��execution
houses,�� which undertake deals as instructed by the client. If true, such a con-
clusion has troubling implications. The investment banking industry, which
consumes a significant proportion of an economy��s talented human capital,
is mainly a deadweight loss to society. CEOs�� inexperience in M&A is not
mitigated by hiring an advisor, which may explain why so many acquisitions
destroy value.
This article reaches a different conclusion. Prior studies investigate the im-
portance of investment banks for M&A outcomes by hypothesizing a measure
of advisor quality, such as market share or name prestige, and correlating it
with this measure of quality. Such studies will find significant results only if
their chosen measures are truly accurate proxies of ability. We instead em-
ploy a fixed-effects analysis. This is a broader approach that examines whether
banks exhibit differential deal returns in the first place, without having to spec-
ify a measure of advisor quality with which any differential will be correlated.
Indeed, we find significant bank fixed effects to a deal��s three-day cumula-
tive abnormal return (CAR). Studying all banks that advised on at least ten
deals over 1980–2007 and controlling for time effects, the difference between
the 25th- and 75th-percentile banks is 1.26%. This difference is economically
meaningful applied to the mean bidder size of $10 billion and compared with
the mean CAR of 0.72%. An
F-test that bank fixed effects are equal is rejected
at the 1% significance level. Our results support the skilled-advice hypothesis
and contrast prior findings that banks have little effect on M&A outcomes, as
predicted by the passive-execution hypothesis.
Returns analyses have also been used to evaluate skill in mutual funds, hedge
funds, and security analysts. Our setting shares two challenges also faced by
studies of stock-picking ability. The first is performance attribution: Returns
1 To our knowledge, only Kale, Kini, and Ryan (2003) find gains to employing market-leading advisors. They
study 324 contested takeovers of public targets, and find that large banks are more likely to withdraw when the
price becomes too high. By contrast, both we and Rau (2000) find a negative link between market share and
performance across all M&A transactions (over 15,000 in our sample), of which approximately one-third are
public. One reason may be that the incentives to act in the client��s interest are far stronger in public situations,
where ��honest�� advice to withdraw from a deal is widely observed.
2287
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
are not purely the responsibility of the financial intermediary. In an invest-
ment setting, returns also depend on the portfolio��s factor loadings and re-
alized factor outcomes. Since investment skill depends on how a portfolio
performs over the long run, investment studies typically investigate long-
horizon returns. Therefore, the results are highly contingent on the benchmark
asset-pricing model used (Fama 1998). Benchmark adjustment is less of an
issue here, since performance can be measured by short-horizon announcement
returns: In an efficient market, they capture the full value impact of an acqui-
sition. Our setting faces a different performance attribution challenge—CAR
may be the responsibility of either the bank or the client. A bank may be asso-
ciated with positive (negative) CARs if it is systematically mandated by high-
quality (empire-building) clients. Many prior studies (e.g., Bowers and Miller
1990; Michel, Shaked, and Lee 1991; Rau 2000; Hunter and Jagtiani 2003) do
not tackle performance attribution and assume CAR results entirely from the
bank. Others control for deal characteristics (e.g., Servaes and Zenner 1996;
Kale, Kini, and Ryan 2003) but acknowledge that this solution may go too far
the other way, since deal characteristics are often the advisor��s responsibility.2
We control for the component of CAR that can be explained by
acquirer char-
acteristics that proxy for the likelihood that the client is empire building (such
as free cash flow and various governance measures, as used by Masulis, Wang,
and Xie 2007) and high quality (such as stock and operating performance and
Tobin��s
Q). We also add acquirer fixed effects to proxy for time-invariant un-
observable measures of quality. Even after these controls, the bank fixed effects
remain statistically and economically significant, with an interquartile range of
1.23%.
A second shared challenge is that average returns depend on not only ability,
but also scale. Maksimovic and Phillips (2002) argue that conglomerate firms��
lower productivity arises because they are able to accept all projects with posi-
tive NPV, including those with modestly positive NPV, whereas single-segment
firms with financing constraints can pursue only those with high NPV. Applied
to our setting, this
limited-capacity hypothesis posits that banks differ not in
ability, but in capacity to accept mandates. Small banks can work only on the
highest-return deals; large banks also accept mandates with small (but posi-
tive) value and thus exhibit lower average returns. We refute this hypothesis by
showing that the banks with the lowest average returns are not the most fre-
quent advisors—the correlation between bank fixed effects and market share
is an insignificant 0.03 (
p = 0.72). Indeed, the most frequent advisors occupy
the middle of the fixed-effects distribution.
While addressing the limited-capacity hypothesis, the finding that the most
positive and most negative fixed effects are associated with infrequent advisors
2 For example, Servaes and Zenner (1996) caveat their conclusion by acknowledging that ��it is not certain that
the [deal characteristics] affecting investment banking choice are exogenous. For example, it is possible that
investment banks influence the form of payment or the decision to pursue the acquisition.��
2288
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
may give rise to two quite different concerns. First, these extreme fixed effects
may result from measurement error arising from few observations, rather than
a ��true�� dispersion of advisor quality. Second, it may be that there is little
dispersion among the major banks that matter more for the M&A market—
i.e., the result is not generalizable to the most important advisors. Note that
infrequent advisors will not drive the statistical significance of our results: If
many banks have imprecisely measured fixed effects, the
F-test will have lit-
tle power. To address concerns that they drive our economic magnitudes, the
interquartile range reported earlier is based on fixed effects weighted by the
inverse of their standard errors. Moreover, to further investigate whether our
results hold among frequent advisors alone, we then test the equality of bank
fixed effects on a subsample of the largest banks that advise on at least 84
deals over the 28-year sample period (i.e., 3 per year). Even though the num-
ber of banks falls markedly from 143 to 42, the results stay significant with an
interquartile range of 0.74% (controlling for acquirer characteristics and ac-
quirer fixed effects) and the
F-test of equality is rejected at the 1% level. Thus,
there are meaningful differences even among the most active advisors.
Having documented that banks are associated with different CARs over the
entire period, we then ask whether these differences are predictable based on
historical data, and thus can be used by clients to guide their selection of advi-
sors. The fixed effect implies a persistent component to a bank��s CAR and thus
provides us with
a priori motivation to predict future returns using past returns,
rather than the market share and prestige measures previously studied. Indeed,
we find that performance is persistent: The top quintile of banks based on CAR
over the past two years outperforms the bottom quintile by 0.94% over the next
two years (significant at 1%). Persistence remains after removing the compo-
nent of CAR that can be attributed to client characteristics. Inconsistent with
the limited-capacity hypothesis, the low CARs of the bottom-quintile banks
do not arise from executing mildly positive transactions, but from executing
double the proportion of value-destructive deals as the top quintile. Regression
analyses yield similar results: A bank��s average returns are significantly linked
to its past average returns, even when controlling for market share. When inter-
acting past returns with market share, the interaction term is insignificant, but
past returns alone remain significant, suggesting that the effect of past returns
is similar for both frequent and infrequent advisors.
The existence of persistence in M&A advice contrasts the lack of persistence
in mutual funds. Berk and Green (2004) show that even if mutual funds possess
skill, this does not translate into persistence if investor flows respond to past
performance and there are diminishing returns to scale. We find that a bank��s
market share is independent of its past CAR (also found by Rau 2000). This
lack of performance chasing can potentially reconcile why persistence exists
in M&A advice but not mutual fund performance (e.g., Carhart 1997). Instead,
we find that mandate awards are highly correlated with past market share, even
though market share negatively predicts future performance.
2289
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
The selection of high-market-share rather than high-CAR advisors has two
quite different interpretations. First, it may be efficient, if clients build up
relationship-specific capital with particular banks and thus rationally retain
the same advisor irrespective of past performance; market share has predictive
power, as it measures a bank��s existing relationships. We show that retaining
a past advisor is associated with worse future performance, particularly if the
bank advised on negative-CAR transactions in the past, which does not support
the view that relationship-specific capital improves future M&A performance.
However, clients may derive other services from banking relationships, such
as lending and underwriting, which rationally induces them to retain their ex-
isting bank (Yasuda 2005; Ljungqvist, Marston, and Wilhelm 2006, 2009). Al-
ternatively, it may be inefficient if clients are not locked into relationships but
voluntarily choose advisors based on market share, under the misperception
that it predicts superior returns. Such behavior is consistent with real-life prac-
tices: Market-share league tables are widely publicized by both the media and
the banks themselves, and so both academics and practitioners have come to
use them as a measure of expertise.
Ertugurul and Krishnan (2010) also study the existence of ability in invest-
ment banking. They focus on individual bankers who switch advisors, rather
than on banks themselves.3 Another difference is that in addition to identifying
a fixed effect in the full sample, we also investigate persistence and thus the
predictability of future outcomes using past performance. Jaffe, Pedersen, and
Voetmann (2009) demonstrate persistence in M&A performance at the client
level. Mikhail, Walther, and Willis (2004) and Hoberg (2007) document persis-
tence in two other banking services, security analysis and equity underwriting.
This article proceeds as follows. Section 1 discusses the potential sources
of differential M&A returns across banks, and Section 2 describes the data.
Section 3 documents significant advisor fixed effects to M&A returns and
shows that average returns are predictable using past performance. Section 4
shows that mandate awards are correlated not with past performance but with
market share. Section 5 concludes.
1. Motivation: Why Might Banks Be Associated with Differential M&A
Returns?
This section discusses a theoretical framework for why there may be a bank
fixed effect to M&A returns, i.e., why certain banks may be systematically
3 We study banks rather than individual bankers for two reasons. First, a transaction typically leverages resources
across the entire bank (e.g., a debt-financed acquisition of a German chemicals target by a UK pharmaceuticals
firm may involve the M&A and debt product groups and the pharmaceuticals, chemicals, UK, and Germany
coverage teams). Second, it is difficult to know which particular banker worked on a certain deal. A bank��s
chemicals team consists of several bankers, many of whom will not be involved in the deal. If a bank��s skill
hinges on particular star bankers (who often move between firms) rather than the whole organization, we should
find weak bank fixed effects and bank-level persistence.
2290
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
associated with high- or low-return deals. To understand the possible sources
of correlation, we first outline the role that advisors play in M&A deals. Their
actual level of involvement can vary significantly across transactions, and falls
under three broad categories.
In a
bank-initiated deal, the advisor is involved in both selecting the deal
and negotiating terms, and thus is responsible for the entire CAR. In a
stan-
dard client-initiated deal, the client proposes the transaction but lacks the
ability to identify good deals and so relies on the bank to advise whether to
pursue it. Since the bank can reject a bad deal, it is again responsible for deal
selection in addition to negotiation, and thus the entire CAR. Not all banks
will reject the deal, but this failure to reject is for reasons within their re-
sponsibility. Some lack the ability to identify bad deals; others know that a
deal will destroy value but undertake it to maximize their own fee income
rather than pursuing the client��s interests. A bank cannot blame low CARs on
having to work on bad deals, since it controls its deal flow—just as a lender
cannot blame losses on poor credit quality, since it controls the loans it chooses
to write.
In a
fixated client deal, the acquirer has already decided on the target and
does not seek advice on its appropriateness; it uses the bank simply to exe-
cute the transaction on the best terms possible. This may occur in two cases.
First, the client may be skilled in identifying targets and does not need the
bank��s input. At the other extreme, the client is empire building or hubristic and
wishes to pursue a bad deal even if the bank cautions otherwise. By accepting
the mandate, the bank may be adding value compared with the counterfactual
of the client pursuing the acquisition with a rival advisor. The bank is not re-
sponsible for the component of CAR that can be attributed to the acquirer��s
skill or hubris, only the orthogonal component.
Given banks�� varying levels of involvement, systematic differences in aver-
age returns may stem from three sources. The
skilled-advice hypothesis is that
certain advisors possess ability, either in proposing targets (for bank-initiated
deals) or in negotiating terms (for all deals). Alternatively, variation may stem
from systematically turning away bad deals. This requires ability to identify
such deals, plus trustworthiness to turn them down. We use the term ��skilled
advice�� to include these three qualities of deal identification, transaction ne-
gotiation, and trustworthiness. Thus, even a bank with high-quality employees
may be classified as low-ability if its business model is to accept every man-
date, regardless of whether it will create client value.
The
passive-execution hypothesis is that banks lack ability in target selec-
tion or deal negotiation. Instead, variation in returns arises because the bank is
systematically mandated by skilled (empire-building) clients.4 In reality, banks
4 Fernando, Gatchev, and Spindt (2005) construct an assignment model where high-quality banks are systemati-
cally mandated by high-quality clients.
2291
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
exert substantial effort in pitching deals to clients: They allocate the majority of
bankers to client coverage (rather than deal execution) groups, which are pri-
marily responsible for pitching. Therefore, it seems unlikely that fixated client
deals are sufficiently prevalent to explain differences in average returns. How-
ever, since it is impossible to observe which party initiates a transaction and
provide direct statistics on this prevalence, to be conservative we also report
results controlling for the component of returns attributable to acquirer char-
acteristics. The passive-execution hypothesis would also be supported if bank
fixed effects are not significant in the first place.
The
limited-capacity hypothesis posits that banks differ not in ability, but
in their capacity to accept mandates. A bank may exhibit a high average CAR
because it can work on only the highest value–creating deals, whereas low
CAR may arise if a bank has the capacity to execute also mildly good deals. We
evaluate this hypothesis by investigating whether a bank��s low average CAR
stems from advising on deals with modest value, or value-destructive deals.5
2. Performance Metrics, Data, and Descriptive Statistics
We use
Thomson Financial��s Securities Data Company (SDC) data for merg-
ers announced between January 1980 and December 2007. Since deals that
involve a change of control are most likely to affect acquirer returns, we re-
tain only transactions categorized as ��Merger,�� ��Acquisition,�� ��Acquisition of
Assets,�� or ��Acquisition of Majority Interest�� and drop all deals for which the
acquirer��s initial stake exceeded 50% or its final stake was below 50%. We
also drop transactions for which the acquirer had no stock returns in the Center
for Research in Security Prices (CRSP) database or the deal value was below
$1 million (as in Rau 2000). Our final sample contains 15,344 deals.
Our value-creation measure is the acquirer��s (−1, +1) CAR over the CRSP
value-weighted index, which we winsorize at 1% and 99%.6 Stock returns are
the relevant performance measure, as they represent the change in shareholder
wealth, capitalizing all of the future effects of an acquisition; they are thus used
in the vast majority of investment-banking studies (e.g., Bowers and Miller
1990; Servaes and Zenner 1996; Rau 2000). While CAR refers to one specific
deal,
RET is the average CAR to all deals advised by a bank in a
j-year period.
5 The limited-capacity hypothesis is less likely for investment banks than corporations or mutual funds. Even small
banks are able to work on very large transactions—for example, the boutique Gleacher employs 50 staff and ad-
vised on Bank of Scotland��s $40 billion merger with Halifax, AT&T��s $22 billion sale to SBC Communications,
and MFS Communications�� $14 billion merger with WorldCom.
6 We also obtain beta model returns from Eventus and find similar results. The correlation between beta model
returns and returns above the CRSP value-weighted (VW) index is 98.5%. Since the beta model cannot be
calculated for several acquirers, we use returns above the CRSP VW index. In addition, Hackbarth and Morellec
(2007) show that betas change substantially upon a merger, and so a beta calculated based on historical data
is likely to be misleading. We use the CRSP VW index as a benchmark, because Rau and Vermaelen (1998)
document biases when using size and book-to-market adjusted CARs.
2292
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
To be included in the analysis, a bank must have announced at least 2
j deals
within the period.7
Some papers attribute the entire CAR to the bank, which constitutes an over-
attribution in fixated client deals. Others remove the component explained by
deal characteristics, but this leads to an underattribution, since deal charac-
teristics may be chosen by the advisor, either directly by initiating the deal
or indirectly by accepting a client-proposed mandate. We control for acquirer
characteristics that proxy for client quality or empire building, since they are
outside a bank��s control, taking its client base as given. Note that banks may
have some control over their client base: If a bank advises a standard client to
abandon a bad deal, it does not enter its client base. Therefore, controlling for
acquirer characteristics is conservative.
A number of our characteristics are related to governance. Masulis, Wang,
and Xie (2007) find that governance mechanisms are significantly related to
acquirer returns. Their primary measure is the Gompers, Ishii, and Metrick
(2003) index. Unfortunately, it is not suitable here because it is available only
from 1990 and we require a long time series to test for persistence; more-
over, it is available for only a subset of firms in a given year. We
therefore include other governance mechanisms studied by Masulis, Wang,
and Xie (2007): institutional ownership, leverage, and product market com-
petition (measured by the Herfindahl index and the industry��s median ratio
of selling expense to sales). The second group of characteristics are proxies
for acquirer quality, also from Masulis, Wang, and Xie (2007): Tobin��s
Q,
stock price runup, and operating performance. We also use the other bidder
characteristics studied by Masulis, Wang, and Xie (2007): free cash flow (which
may facilitate empire building) and size (which Moeller, Schlingemann, and
Stulz 2004 show is negatively correlated with returns). Since omitted acquirer
characteristics may overattribute CAR to the bank, we add additional controls
over and above those featured in prior literature. We include inside ownership
from Compact Disclosure, to measure management��s alignment with share-
holders. Where missing, we impute it using firm sales and age.8 To proxy
for empire-building intent, we include the number of acquirer Standard
Industrial Classification (SIC) codes and a dummy for whether it made an
acquisition in the previous five years. Finally, we include dummies for the
bidder��s Fama-French industry.9 Full variable definitions are given in Table 1.
7 Where a deal has multiple advisors, the deal is credited to each advisor separately. This is consistent with how
SDC constructs market-share league tables.
8 Specifically, we winsorize sales at 1% and 99% and regress inside ownership on sales and age. We then use the
coefficients to predict inside ownership for the firms where it is missing. The
R2 of the first-stage regression is
13%.
9 We use acquirer industry fixed effects rather than running the analysis for each industry separately (i.e., studying
the fixed effect of a particular bank-industry group) because very few banks undertake at least 2
j transactions
within a given industry in
j years, the minimum required to calculate an accurate
RET measure.
2293
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
Table 1
Definition of variables used in the analyses
Panel A: Used in the calculation of residuals for CAR
Variable
Definition
RUNUP
Log stock return for the acquirer from −210 to −11.
Q
Log of Tobin��s Q.
Tobin��s Q = Market value of assets / Total assets (#6).
Market value of common stock = Common shares outstanding (#25) * Price
(#199).
Market value of assets = Book value of assets (#6) + Market value of common
stock – Book value of common stock (#60) – Balance sheet deferred taxes (#74).
LEVERAGE
LEVERAGE = Book debt / (Total assets (#6) – Book equity + Market equity).
Book equity = Total assets (#6) – Total liabilities (#181) – Preferred stock (#10) +
Deferred taxes (#35, if available)
Substitute Redemption value of preferred stock (#56) if Preferred stock is missing.
Book debt = Total assets (#6) – Book equity.
Market equity = Common shares outstanding (#25) * Price (#199).
FCF
FCF = Free cash flow / Total assets (#6).
Free cash flow = Operating income before depreciation (#13) – Interest expense
(#15) – Income taxes (#16) +
Deferred taxes and investment tax credit (#35 -
#35 from previous year) – Preferred dividends (#19) – Common dividends (#21).
SIZE
Log of Total assets (#6).
HERFINDAHL
��
i ( firm salesi(#12)
industry sales )2
, where industries are defined by the Fama-French 49
industries.
SELLEXP
SELLEXP = median selling expenses (#189) over Sales (#12) for industry.
INST
Fraction of outstanding common shares owned by institutions from Thomson
Financial 13f filings.
OPPERF
Firm operating performance minus the industry median in the past year.
Operating performance = Operating income before depreciation (#13) / 0.5
(Total assets + last year��s total assets (#6)).
INSIDER
Insider ownership as a fraction of total shares outstanding, from Compact Disclo-
sure. Where missing, we impute it using Sales (#12) and firm age (from CRSP).
ACQSIC
Log of number of acquirer SIC codes.
REPEAT ACQUIRER
Dummy variable that equals 1 if the acquirer announced an acquisition in the
previous 5 years.
Panel B: Constructed for direct use in quintile analysis and regressions
Variable
Definition
RET
Average CAR (3-day cumulative abnormal return) for deals advised by an invest-
ment bank over a given number of years.
RETRES
Average residual from a regression of CAR on acquirer characteristics defined in
Panel A.
SHARE
Market share by value of acquirer-advised deals for an investment bank over a
calendar year.
Where applicable, we include the Compustat item number in the description.
All variables are calculated for the fiscal year ending the year before deal
announcement.10
10 Our regression of CAR on characteristics is run on the entire sample with year fixed effects. Using a rolling
window would cause data from the early period of the sample to be dropped and would also produce less precise
estimates. Full-sample regressions are thus often used in asset pricing. We are not assuming that clients use past
data to estimate the characteristics parameters for themselves when choosing banks. Instead, we posit that clients
already have in mind a model of the effect of acquirer characteristics on returns, which they use to isolate the
portion of CAR that is outside the bank��s control. As econometricians, we are attempting to estimate this model,
for which we require the full sample.
2294
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
Table 2
Regression of CAR on acquirer characteristics
CAR
RUNUP
−0.0018
(1.07)
Q
−0.0055
(2.89)∗∗∗
LEVERAGE
0.0185
(4.05)∗∗∗
FCF
−0.0569
(7.45)∗∗∗
SIZE
−0.0033
(8.57)∗∗∗
HERFINDAHL
0.0008
(0.06)
SELLEXP
−0.0433
(2.22)∗∗
INST
−0.0097
(4.06)∗∗∗
OPPERF
0.0475
(5.72)∗∗∗
INSIDER
0.0110
(2.08)∗∗
ACQSIC
−0.0018
(1.92)∗
REPEAT ACQUIRER
−0.0028
(1.89)∗
Year FE
Yes
Acquirer Industry FE
Yes
# obs
12,622
R2 (%)
4.17
* significant at 10%; ** significant at 5%; *** significant at 1%.
CAR is the return in excess of the CRSP value-weighted index over a (–1, +1) window relative to the announce-
ment date. The regressors are described in Table 1. The sample period is 1980–2007, and
t-statistics are in
parentheses.
We use CARRES to denote the residual CAR after controlling for acquirer
characteristics, and define
RETRES as the average CARRES over a
j-year
period. The regression results are shown in Table 2. Most coefficients are of
the expected sign: Returns are increasing in leverage, operating performance,
and inside ownership, and decreasing in free cash flow and the number of
SIC codes. The
R2 of 4% is commensurate with Masulis, Wang, and Xie��s
(2007)
R2 of 5%. Their
R2 is marginally higher because they include deal
characteristics, which are not appropriate here, since they are under the bank��s
control.
Since the bank is responsible for raw CAR in all but fixated client deals,
it constitutes our core measure. As with any investment decision, an M&A
transaction should be undertaken if the NPV, irrespective of project charac-
teristics, exceeds zero. A bank cannot justify a negative-NPV transaction by
arguing that other clients with, say, the same number of SIC codes undertook
even worse deals if it had the option to turn away the deal in the first place.
2295
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
3. Return Differences across Investment Banks
3.1 Full-sample fixed effects
Most prior research on advisor ability attributes a deal��s CAR entirely to the
bank and studies the association between average CAR and a variable hypoth-
esized to proxy for bank quality, such as market share or reputation (measured
by the prestige of the bank��s name). Such analyses will find significant results
only if ability is correlated with their chosen measures of advisor quality. Thus,
the absence of a link with market share or reputation need not imply that banks
do not matter for M&A outcomes.
We therefore take a broader approach. Rather than hypothesizing a measure
of bank ability, we investigate whether banks exhibit differential CARs in the
first place by estimating the bank fixed-effect component of a deal��s returns.
We regress CAR on bank fixed effects while controlling for time fixed effects,
since market enthusiasm for M&A may have varied over time. We then add
acquirer characteristics to proxy for observable measures of quality or em-
pire building. Finally, we add acquirer fixed effects to proxy for time-invariant
unobservable measures of advisor quality. If two banks merge (e.g., Deutsche
Bank buys Bankers Trust), we construct one fixed effect for the target (Bankers
Trust) and a separate fixed effect for the acquirer (Deutsche Bank) both before
and after the merger.
The results are shown in Table 3 for the 143 banks that advised on at least
ten deals over 1980–2007 or were acquired by a bank that advised on at least
ten deals. Panel A finds that, in all specifications, the fixed effects are highly
significantly different from each other (
p < 0.01). Panel B demonstrates the
economic significance of these differences. The difference between the 25th-
and 75th-percentile banks is 0.9–1.3%,11 compared with the average CAR of
0.72% and the mean bidder size of $10 billion. To our knowledge, these results
constitute the first large-scale evidence that certain banks are systematically as-
sociated with superior M&A returns, and contradict prior findings that advisor
quality has no positive effect on M&A outcomes.
While supportive of the skilled-advice hypothesis, differential returns could
also be consistent with the limited-capacity hypothesis, if the banks with the
highest fixed effects advise on the fewest transactions. To investigate this,
Figure 1 presents a scatter plot of a bank��s fixed effect against its market share
(by number of deals). Inconsistent with the limited-capacity hypothesis, the
most frequent advisors occupy the middle, rather than the lower end, of the
distribution. There is a hump-shaped relationship between market share and
fixed effects, and the correlation between the two variables is an insignificant
−0.05 to 0.05.
11 Following Bertrand and Schoar (2003), we weight each fixed effect by the inverse of its standard error to account
for estimation error. Without this weighting, the interquartile range of fixed effects is 1.7–2.2%.
2296
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
Table 3
Bank fixed effects to a deal��s CAR
Panel A: Investment Bank Fixed Effects
Controls
Bank FE
F-test
Obs
Adj
R2 (%)
(1)
Time FE
1.65 (0.0000, 142)
15,344
0.97
(2)
Acq chars, time FE
1.44 (0.0006, 141)
12,622
3.98
(3)
Acq chars, acq FE, time FE
1.47 (0.0003, 138)
10,430
3.27
Panel B: Distribution of Bank Fixed Effects
Std Dev
25th
75th
Interquartile Range
(1)
1.37%
–0.91%
0.35%
1.26%
(2)
1.39%
–0.53%
0.41%
0.94%
(3)
2.00%
–0.65%
0.58%
1.23%
Panel C: Fixed Effects for Active Banks
Bank FE
F-test
Std Dev
25th
75th
Interquartile Range
(1)
1.69 (0.0038, 41)
0.64%
–0.85%
0.21%
1.06%
(2)
1.41 (0.0431, 41)
0.66%
–0.45%
0.28%
0.74%
(3)
1.60 (0.0092, 41)
0.86%
–0.47%
0.27%
0.74%
Panel D: Investment Bank Fixed Effects with Filters
Bank FE
F-test
Banks FE
F-test
Controls
(all banks)
(active banks)
Obs
Adj
R2(%)
(1)
Time FE
1.60 (0.0000, 142)
1.68 (0.0043, 41)
14,955
0.98
(2)
Acq chars, time FE
1.45 (0.0004, 141)
1.45 (0.0305, 41)
12,347
3.62
(3)
Acq chars, acq FE, time FE
1.48 (0.0002, 138)
1.70 (0.0036, 41)
10,235
3.37
Panel E: Distribution of Bank Fixed Effects with Filters
Banks
Std Dev
25th
75th
Interquartile Range
(1)
All
1.36%
–0.88%
0.34%
1.22%
(1)
Active banks
0.64%
–0.80%
0.31%
1.11%
(2)
All
1.40%
–0.74%
0.39%
1.12%
(2)
Active banks
0.67%
–0.45%
0.17%
0.61%
(3)
All
2.00%
–0.73%
0.51%
1.23%
(3)
Active banks
0.88%
–0.53%
0.25%
0.78%
Panel A reports
F-tests for the equality of bank fixed effects from a regression of (–1, +1) CAR on bank fixed
effects and listed controls. Acquirer characteristics are defined in Table 1. Panel B reports the distribution of
bank fixed effects, weighted by the inverse of the standard errors of the fixed effects. Panel C reports analysis
on fixed effects for banks that announced at least 84 deals over the full sample. Panels D and E repeat these
analyses filtering out acquirers with stock prices below $1 and zero or missing trading volume on either the
day after announcement or both two and three days before announcement.
F-statistics,
p-values, and numbers
of constraints are listed. The number of constraints equals the number of bank fixed effects estimated in the
regression minus one. In specification (3), the adjusted
R2 for this specification is after removing acquirers that
undertake only one transaction, and then demeaning by acquirer before regressing on acquirer characteristics,
time dummies, and bank dummies. The sample period is 1980–2007.
While Figure 1 provides evidence against the limited-capacity hypothesis,
it may raise a different concern: Perhaps the significant results of Table 3 are
driven by infrequent advisors whose fixed effects are noisily measured, or who
are relatively unimportant for the M&A market as a whole. To address this
concern, the interquartile ranges of fixed effects that we reported above are
after weighting each fixed effect by the inverse of its standard error. Moreover,
we further investigate this hypothesis by testing for equality of fixed effects
focusing only on the largest banks that advised on at least 84 transactions over
2297
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
Figure 1
Plots of estimated bank fixed effects against number of deals announced in the full sample
The fixed effects are estimated from regressions of the (–1, +1) CAR of deals on bank fixed effects and control
variables. Plot (1) includes time fixed effects as controls, (2) includes acquirer characteristics and time fixed
effects, and (3) includes acquirer characteristics, acquirer fixed effects, and time fixed effects.
the sample period, i.e., 3 per year. 42 banks meet this criterion. Panel C shows
that the fixed effects remain jointly statistically significant at the 1% level,
and the interquartile range is an economically meaningful 0.74%, even when
continuing to control for acquirer characteristics and acquirer fixed effects.
Another measurement concern is that while CAR measures the full value
impact of a deal in an efficient market, it may understate it if part is incor-
porated into prices before or after announcement. The former will occur if
the deal leaks out early, bringing the measured returns of both good and bad
2298
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
deals toward zero. The latter will occur if investors do not notice certain effects
of the transaction until later and this failure is not unbiased (i.e., they fail to
notice more good than bad effects, or vice versa). While long-run returns would
capture a greater proportion of the transaction��s impact, they would also in-
corporate many other corporate events (e.g., dividend and earnings changes
not due to the acquisition) and hence suffer from a high noise-to-signal ratio.
Moreover, errors resulting from failure to use the ��true�� benchmark model of
stock returns are compounded over long horizons (Fama 1998).
In the classical ��errors-in-variables�� problem, where measurement error is
symmetric (i.e., the average error is zero) and similar across observations,
mismeasurement simply attenuates the results. Our setting differs from the
standard problem in two ways. First, mismeasurement arising from leakage or
delayed reaction is asymmetric: Positive (negative) true returns are associated
with negative (positive) errors. If the mean return were zero, positive and nega-
tive true returns would be equally likely, and so the average error would also be
zero—thus, the results would again be attenuated. By contrast, our mean CAR
is positive, and so the average error is negative, biasing reported returns toward
zero and thus below the mean. This would not be a problem if mismeasurement
were similar across banks, since it would reduce measured
RET evenly across
the sample. However, our second difference is that mismeasurement may be
more serious for certain banks. For example, small banks may advise small
clients, whose deals are less closely followed by M&A arbitrageurs or the
media, leading to less leakage and thus higher measured CAR. Thus, a low
RET fixed effect may result from measurement error rather than underperfor-
mance.12 We address the mismeasurement explanation in a number of ways.
First, it is reasonable to assume that transactions with measured CARs exceed-
ing 10% in absolute value did not suffer from attenuation. The remaining 87%
of deals are the subset for which attenuation may be present. The mean CAR
for this subset is −0.016%, very close to zero. Hence, any attenuation is in-
deed toward the mean, as in the classic errors-in-variables setting, and leads to
our results being understated.13 Second, in Panels D and E we filter out low-
price and low-volume acquirers and show that the results are barely affected.14
12 For example, assume that banks A and B both execute deals with true value creation of 1.5%, 1.0%, and −1.0%
(i.e., a positive mean). Bank A��s CARs fully capture the value, and so its
RET is 0.5%. Bank B��s CARs capture
only half of the value due to leakage or underreaction, and so its CARs will be 0.75%, 0.5%, and −0.5%, yielding
a
RET of 0.25%.
13 Continuing the earlier example, assume that banks A and B both execute deals with true value creation of 12%,
1%, and −1%. Bank A��s CARs fully capture the value, and bank B��s CARs for the final two deals are halved.
Since the deals that exhibit attenuation have a zero average return, both banks will have the same
RET.
14 Specifically, we keep only cases where the acquirer��s price is at least $1 before the acquisition and where there
is trading volume either two or three days before the announcement and trading volume on the day after the
announcement. These filters are similar in spirit to Diether, Lee, and Werner (2009), who use an initial $1 and
zero volume screen. Our setting differs from the anomalies literature (e.g., Diether, Lee, and Werner 2009) in
that we require only that CAR be a signal of value creation and do not require that it be tradable; thus, the M&A
literature does not typically employ such filters. Further, CRSP reports the midpoint of bid and ask if a closing
price is not available.
2299
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
For example, the interquartile range is 0.78%, even when focusing on the
42 largest banks and adding all controls. Third, we later show (in Panel D
of Table 5) that the low average returns of low-
RET banks stem from their
undertaking a high proportion of value-destructive deals, rather than deals with
small but positive measured value.15
To demonstrate the results on an individual bank level, Table 4 provides
summary statistics on the fifteen largest banks by number of deals. There is
significant variation in the average returns to each bank, which range from
−0.12% (UBS) to 1.47% (Bank of America). Controlling for acquirer charac-
teristics sometimes has a marked impact on banks�� performance measures: For
example, Goldman Sachs has the second lowest fixed effect (−0.93%) when
controlling only for year dummies, but it rises to an above-average 0.17% when
adding acquirer characteristics and acquirer fixed effects.
3.2 Persistence in announcement returns
While significant bank fixed effects suggest that advisors matter for M&A
returns, the results of Table 3 are not actionable by clients in their selection
Table 4
Summary statistics for the top 15 investment banks by number of announced deals from 1980–2007
Number Market Share
Bank FE
Bank FE
Bank FE
Investment Bank
of Deals
by Value
RET
RETRES
(1)
(2)
(3)
Goldman Sachs
1,126
23.64%
0.07%
0.13%
–0.93%
0.13%
0.17%
Morgan Stanley
1,108
14.91%
0.09%
0.02%
–0.91%
–0.03%
–0.31%
Merrill Lynch
997
16.91%
0.18%
–0.13%
–0.85%
–0.22%
–0.47%
CSFB
828
10.85%
0.38%
–0.41%
–0.60%
–0.45%
–0.70%
SSB/Citigroup
821
16.74%
0.51%
–0.00%
–0.38%
–0.08%
0.24%
Lehman
803
8.08%
0.58%
0.15%
–0.35%
0.07%
0.09%
JP Morgan Chase
500
8.22%
0.28%
0.02%
–0.48%
0.02%
0.27%
Lazard
427
7.74%
0.46%
0.27%
–0.47%
0.28%
–0.24%
DLJ
407
3.02%
0.68%
–0.64%
–0.45%
–0.75%
–1.12%
Bear Stearns
374
4.43%
0.98%
0.23%
–0.04%
0.14%
0.05%
UBS Warburg
367
7.32%
–0.12%
–0.45%
–0.98%
–0.45%
–0.41%
Bank of America
299
4.21%
1.47%
0.67%
0.64%
0.65%
0.58%
Salomon (pre-merger)
284
1.98%
0.66%
–0.23%
–0.56%
–0.33%
–0.62%
JP Morgan (pre-merger)
267
3.41%
0.62%
0.43%
–0.46%
0.41%
0.21%
Deutsche Bank
244
4.34%
1.04%
0.82%
0.26%
0.88%
0.08%
Avg over entire sample
15,344
0.72%
0.00%
The averages provided in the last row include deals for all investment banks in the sample. RET is a bank��s
average (–1, +1) CAR; RETRES is a bank��s average CARRES, the acquirer characteristic unexplained return
(residuals from regressing CAR on acquirer characteristics). The three final columns display bank fixed effects as
estimated in Table 3. Specification (1) includes time fixed effects; specification (2) includes time fixed effects and
acquirer characteristics; specification (3) includes time and acquirer fixed effects, and acquirer characteristics.
15 A further hypothesis is that banks differ not in skill, but in the fees that they charge: Low
RET banks may be
adding the same value as their rivals, but charging higher fees. We are unable to calculate ��pre-fee�� CARs, as
fees are available for only 2,046 deals within our sample. However, among the deals that do have fee data, we
find that the correlation between acquirer returns and fees (scaled by market cap) is a slightly
positive 0.013,
inconsistent with the hypothesis that low
RET results from high fees.
2300
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
decisions, since they are based on the full 28-year sample. We therefore an-
alyze whether clients can predict positive future returns based on historical
data. The existence of a bank fixed effect implies a persistent component to a
bank��s average CAR and thus provides
a priori motivation for predicting future
returns using an advisor��s past returns, rather than the market share and reputa-
tion measures previously studied. This
a priori motivation mitigates potential
concerns of data-mining for advisor characteristics with predictive power. We
calculate persistence in advisor performance in a similar manner to Jegadeesh
and Titman (1993) for stocks and Carhart (1997) for mutual funds. At the
start of each year, we sort banks into quintiles based on
RET for the past
j calendar years, where
j = {1, 2, 3}. Next, for each quintile, we calculate
RET for all banks within the quintile over the next
k calendar years, where
k = {1, 2, 3}. We report the difference in
RET between the top (Q5) and bottom
(Q1) quintiles.16,17
Table 5 illustrates the results. Panel A documents significant persistence in
raw CAR in eight out of the nine time horizons. For example, when
j =
k = 2,
the difference between Q1 and Q5 is 0.94%, significant at the 1% level. This
result need not imply differential ability if fixated client deals comprise a sub-
stantial proportion of all transactions, and so we next control for acquirer char-
acteristics.18 Panel B illustrates persistence in the component attributable to
advisors (
RETRES). Thus, the persistence in
RET does not arise because banks
are systematically mandated by fixated acquirers, consistent with the skilled-
advice hypothesis.
The above panels calculate
RET and
RETRES using an equally weighted
average of a bank��s CARs. Equal weighting is appropriate if one believes that
each transaction is a separate measure of the bank��s ability and thus should be
considered equally; it is also consistent with the main analysis on bank fixed
effects, which treats each deal equally. Indeed, Mikhail, Walther, and Willis
(2004) and Hoberg (2007), who study performance persistence in two other
investment-banking services (analyst recommendations and equity underwrit-
ing), also use equal weighting. However, weighting CARs by the size of the
transaction may be appropriate if one believes that banks put greater effort into
larger deals and so they are a more accurate measure of ability, or larger deals
16 As future returns are overlapping, we correct for mechanical autocovariance by using Newey-West standard
errors for panel data.
17 To illustrate our treatment of bank mergers, we continue with the example of the Deutsche Bank (DB) and
Bankers Trust (BT) merger in June 1999. Consider a regression of two-year
RET on past two-year
RET. For
any observations where
RET ends in 1998 or earlier, DB and BT enter separately, and both
RET and
RET are
calculated on a stand alone basis. For any observations where
RET ends in 1999 or later, we drop the two stand-
alone observations and create one combined observation. Specifically,
RET for 1998–1999 includes all deals
advised by either DB, BT, or the merged entity during this period. To be consistent, the past
RET measure will
also include all deals advised by either bank or the merged entity in 1996–1997.
18 Since the persistence analysis uses short windows of one to three years, we do not include acquirer fixed effects,
because identifying purely on repeat acquirers who switch advisors would mean that very few deals are used to
calculate
RETRES and lead to substantial noise.
2301
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
Table 5
Persistence in a bank��s average returns and source of low returns
Panel A: Persistence in Raw Returns
Future RET Measured Over
Quintiles Measured Over
1yr
2yrs
3yrs
1yr RET
Q1
0.92%
0.66%
0.69%
Q5
1.49%
1.48%
1.44%
Q5-Q1
0.57%
0.82%
0.76%
(1.32)
(2.73)∗∗∗
(2.85)∗∗∗
2yrs RET
Q1
0.52%
0.61%
0.74%
Q5
1.47%
1.55%
1.36%
Q5-Q1
0.95%
0.94%
0.63%
(2.58)∗∗
(3.07)∗∗∗
(1.99)∗∗
3yrs RET
Q1
0.59%
0.57%
0.60%
Q5
1.68%
1.73%
1.54%
Q5-Q1
1.09%
1.16%
0.94%
(2.74)∗∗∗
(3.55)∗∗∗
(2.91)∗∗∗
Panel B: Persistence in Residual Returns
Future RETRES Measured Over
Quintiles Measured Over
1yr
2yrs
3yrs
1yr RETRES
Q5-Q1
0.78%
0.80%
0.85%
(1.71)∗
(2.51)∗∗
(3.32)∗∗∗
2yrs RETRES
Q5-Q1
1.35%
0.92%
0.55%
(2.86)∗∗∗
(2.85)∗∗∗
(1.88)∗
3yrs RETRES
Q5-Q1
0.75%
0.55%
0.37%
(1.76)∗
(1.82)∗
(1.29)
Panel C: Persistence in Returns (Transaction value-weighted measures)
Quintiles
Future RET Measured Over
Quintiles
Future RETRES Measured Over
Measured Over
1yr
2yrs
3yrs
Measured Over
1yr
2yrs
3yrs
1yr RET
1yr RETRES
Q5-Q1
0.62%
1.01%
1.00%
Q5-Q1
0.95%
1.31%
0.82%
(1.07)
(2.01)∗∗
(2.22)∗∗
(1.63)
(2.93)∗∗∗
(1.95)∗
2yrs RET
2yrs RETRES
Q5-Q1
0.98%
0.98%
0.81%
Q5-Q1
0.38%
0.65%
0.23%
(1.79)∗
(1.88)∗
(1.52)
(0.69)
(1.26)
(0.43)
3yrs RET
3yrs RETRES
Q5-Q1
0.45%
1.36%
1.07%
Q5-Q1
0.62%
0.75%
0.50%
(0.84)
(2.47)∗∗
(2.01)∗∗
(1.08)
(1.27)
(0.80)
Panel D: Percentage of Positive CAR deals, by RET Quintile
RET Measured Over
1yr
2yrs
3yrs
Q1
26.51%
34.90%
37.56%
Q5
71.64%
67.41%
64.67%
Q5-Q1
45.13%
32.51%
27.11%
(25.12)∗∗∗
(22.07)∗∗∗
(19.26)∗∗∗
(
continued)
2302
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
Table 5
Continued
Panel E: Persistence by Size of Advisor
Large Advisors
Small Advisors
Terciles
Future RET Measured Over
Halves
Future RET Measured Over
Measured Over
1yr
2yrs
3yrs
Measured Over
1yr
2yrs
3yrs
1yr RET
1yr RET
T3-T1
0.40%
0.48%
0.41%
H2-H1
1.07%
0.80%
1.24%
(1.50)
(2.53)∗∗
(2.29)∗∗
(1.62)
(1.62)
(2.79)∗∗∗
2yrs RET
2yrs RET
T3-T1
0.45%
0.43%
0.33%
H2-H1
1.69%
1.04%
0.96%
(1.82)∗
(2.23)∗∗
(1.58)
(3.12)∗∗∗
(2.50)∗∗
(2.48)∗∗
3yrs RET
3yrs RET
T3-T1
0.48%
0.41%
0.37%
H2-H1
1.17%
1.16%
1.02%
(1.77)∗
(1.97)∗∗
(1.73)∗
(2.26)∗∗
(2.56)∗∗
(2.46)∗∗
* significant at 10%; ** significant at 5%; *** significant at 1%.
Panel A sorts banks into quintiles based on their
RET (average (–1, +1) CAR) over the past
j calendar years,
where
j = {1,2,3}. To be included in the analysis, a bank must have announced at least 2
j deals over the relevant
period. Q1 represents the banks with the lowest past
RET, Q5 the highest. For each quintile, we then calculate
the average CAR to future acquisitions announced by the banks in that quintile over the next
k calendar years,
where
k = {1,2,3}. Panel B repeats the analysis for
RETRES (average of CARRES, the acquirer characteristic
unexplained return). Panel C examines
RET and
RETRES calculated using transaction-value weighted measures.
Panel D studies the correlation between
RET and the contemporaneous percentage of positive CAR deals. Panel
E subdivides the sample into the top 2/3 of banks by number of deals in each sorting period (Large Advisors)
and the bottom 1/3 (Small Advisors); these banks are sorted each period into terciles and halves, respectively.
The sample period is 1981–2007. Newey-West
t-statistics are in parentheses.
are more complex and thus a truer test of advisor quality. Panel C repeats the
analysis under size weighting. The results for our core performance measure,
RET, are now stronger in terms of economic significance in eight of the nine
cells and are at least 10% significant in six. The results for
RETRES remain
positive in all cells but are significant in only two.
While suggestive of persistence, the above results admit other interpreta-
tions. A notable feature of Panel A is that the average returns are positive even
for Q1. Thus, it is consistent with both the limited capacity and mismeasurement
hypotheses: Q1��s low
RET arises not due to bad deals, but due to deals with
modest measured returns—either because they actually generate small value,
or because they actually generate large value that is undermeasured. We ad-
dress both interpretations by calculating the ��success ratio�� of each bank: the
percentage of deals that have a positive CAR. The correlation between one-
year
RET and bank success ratio is 0.68. Panel D illustrates that Q5 has ap-
proximately double the success ratio of Q1, 65–72% compared with 27–38%.
Therefore, inconsistent with these hypotheses, Q1��s low
RET stems from a
high proportion of value-destructive deals, rather than from deals with small
but positive measured returns. Additional suggestive evidence against the
limited-capacity hypothesis is in Figure 1, which shows no correlation between
a bank��s
RET and its market share.
While Table 5 is a bank-level analysis, which considers each bank equally,
another approach is a deal-level analysis. The Online Appendix shows that the
2303
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
results remain robust—deals where the advisor was in the top quintile based on
two-year prior performance outperform the bottom quintile by 0.91%, signifi-
cant at the 1% level. This persistence continues to hold after controlling for ac-
quirer characteristics. The Online Appendix also considers other performance
measures than CAR. It documents persistence in a bank��s completion ratio and
speed of completion and shows that these are weakly positively correlated with
RET.
As in Section 3.1, we also subdivide the analysis into the largest and smallest
banks. Note that such a division significantly reduces power by lowering our
sample size. Other persistence analyses in corporate finance typically study
accounting variables (e.g., Bertrand and Schoar 2003; Lemmon, Roberts, and
Zender 2008), which are directly under the manager��s control. Moreover, many
accounting variables are naturally persistent: For example, Lemmon et al. study
the level of leverage, which is naturally persistent, as any changes to leverage
are with respect to last period��s leverage as a starting point. Thus, far fewer ob-
servations are needed to achieve power. By contrast, M&A returns start from a
��clean slate�� each time—the starting point is zero, rather than the return of the
last M&A deal. Moreover, M&A returns are notoriously noisy because they
reflect the market��s reaction rather than a variable controlled by management;
indeed, previous studies of M&A returns (e.g., Masulis, Wang, and Xie 2007)
find low
R2 values, implying that M&A returns are difficult to explain. Thus,
stratifying the sample will markedly reduce the statistical power of our tests, as
well as lower the cross-sectional variation within each subsample. Neverthe-
less, we perform such a stratification to investigate the source of the significant
results in Table 5. We define ��large banks�� as those in the top two-thirds by
number of deals in the period over which past performance is measured, and
��small banks�� as those in the lowest third. We divide banks into terciles of past
performance in the large-bank subsample and halves within the small-bank
subsample, rather than quintiles, to obtain approximately the same number of
banks in each quantile as before (since 2/3 * 5 is close to 3 and 1/3 * 5 is close
to 2). While the number of banks remains similar within each quantile, moving
from quintiles to terciles or halves reduces the variation between the top and
bottom quantiles.
The left side of Panel E of Table 5 gives the
RET results for large banks.
Seven of the nine cells are significant, which suggests that persistence does ex-
ist among the large-bank subsample alone. The results are slightly weaker than
in Table 5, consistent with reduced power. The right side of Panel E demon-
strates the findings for small banks. The results remain significant in seven of
the nine cells, even though the sample size has fallen by two-thirds. One of
the motivations for removing small banks is that their
RET may be measured
with significant error given the small number of deals. Thus, the variations in
bank fixed effects in Table 3 may reflect noise rather than true differences in
performance. However, while noise may explain cross-sectional variation at a
given point in time, it is unlikely to account for time-series persistence—noise
2304
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
implies that returns might be measured to be very high in one period, and then
very low in an adjacent period. By contrast, we find evidence of persistence
for even infrequent advisors, suggesting that their return differences reflect
true variation in performance rather than noise.
3.3 Regression analysis
The analysis of Section 3.2 studies differences in average performance between
the top and bottom quantiles. To ensure that the results are not driven purely
by banks at the extremes (i.e., the best and worst banks alone), we conduct a
similar analysis using regressions. In addition to using all banks in the sample,
regressions also allow us to control for other potential predictors of past re-
turns. In particular, Rau (2000) finds that
RET is negatively related to a bank��s
past market share; he does not examine the effect of past returns. We wish to
examine whether the explanatory power of past
RET remains when controlling
for market share. We run the following pooled regression:
RETt,
t = ��
t + ��
RRETt−
j,
t−1 + ��
SSHAREt−
j,
t−1,
(1)
where
SHAREt−
j,
t−1 is the market share over the past
j calendar years, by
dollar value of deals (using number of deals leads to similar results). Since
we have shown that bank fixed effects are significant, and our regressors may
not capture the full fixed effects, the residuals for deals advised by the same
bank might be correlated. We therefore cluster standard errors by bank. The
results are illustrated in Panel A of Table 6. The regressions replicate the posi-
tive association between future
RET and past
RET documented in the quintile
analysis, for all values of
j. Market share is significantly negatively related
to future returns in all three specifications in which it is included. The Online
Appendix shows similar results in a deal-level analysis.
We are again interested in whether our results apply to the M&A advisory
market in general or are driven by small acquirers. Panel B adds an inter-
action term between
RET and
SHARE. The interaction term is insignificant
(
t-statistics all below 0.5), suggesting that persistence is not weaker for larger
banks. Moreover, the coefficients on
RET and
SHARE alone retain their signif-
icant coefficients in all specifications. While stratifying the quintile analysis of
Section 3.2 significantly reduces sample size, a regression analysis allows us to
use an interaction term to study how the results vary across banks of different
sizes without such a reduction. In addition, while the coefficient on
RET gives
the significance of the marginal effect if
SHARE = 0, we also include the co-
efficient if
SHARE = 5%, i.e., for a frequent advisor with a 5% market share.
From Table 4, a bank with a 5% market share would be among the top ten
advisors. The table demonstrates that the marginal effect of past
RET remains
significant even for frequent advisors.
Another interesting question is whether ability differences have decreased
over time. Morrison and Wilhelm (2007, 2008) show that investment-banking
2305
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
Table 6
Determinants of RET
Panel A: Determinants of Bank-Level RET
1yr
1yr
2yrs
2yrs
3yrs
3yrs
Past
j years
RET
0.0958
0.0908
0.1974
0.1876
0.1789
0.1619
(2.22)∗∗
(2.09)∗∗
(3.38)∗∗∗
(3.21)∗∗∗
(2.82)∗∗∗
(2.56)∗∗
SHARE
−0.0346
−0.0351
−0.0416
(3.17)∗∗∗
(3.00)∗∗∗
(3.59)∗∗∗
# obs
993
993
947
947
897
897
R2 (%)
6.16
6.51
6.22
6.56
5.06
5.56
Panel B: Determinants of Bank-Level RET with RET-SHARE Interaction
1yr
1yr
2yrs
2yrs
3yrs
3yrs
Past
j years
RET
0.0968
0.0919
0.2044
0.1920
0.1786
0.1566
(2.03)∗∗
(1.93)∗
(3.25)∗∗∗
(3.05)∗∗∗
(2.66)∗∗∗
(2.33)∗∗
SHARE
−0.0346
−0.0347
−0.0421
(3.16)∗∗∗
(2.95)∗∗∗
(3.66)∗∗∗
RET X
−0.0893
−0.0945
−0.6364
−0.3889
0.0280
0.4964
SHARE
(0.11)
(0.11)
(0.46)
(0.27)
(0.02)
(0.35)
# obs
993
993
947
947
897
897
R2 (%)
6.16
6.52
6.24
6.57
5.06
5.57
RET (at
0.0924
0.0871
0.1726
0.1725
0.1800
0.1814
5% SHARE)
(2.12)∗∗
(1.92)∗
(2.40)∗∗
(2.34)∗∗
(2.31)∗∗
(2.30)∗∗
Panel C: Determinants of Bank-Level RET with First-Half Interaction
1yr
1yr
2yrs
2yrs
3yrs
3yrs
Past
j years
RET X
0.0406
0.0290
0.0998
0.0806
0.1222
0.0921
FIRST HALF
(0.43)
(0.31)
(0.81)
(0.64)
(0.88)
(0.64)
RET
0.0826
0.0805
0.1666
0.1618
0.1404
0.1310
(1.49)
(1.46)
(2.19)∗∗
(2.14)∗∗
(1.87)∗
(1.76)∗
SHARE X
−0.0256
−0.0219
−0.0326
FIRST HALF
(1.05)
(0.77)
(1.16)
SHARE
−0.0244
−0.0265
−0.0304
(2.01)∗∗
(2.36)∗∗
(2.58)∗∗
# obs
993
993
947
947
897
897
R2 (%)
6.19
6.58
6.33
6.67
5.19
5.72
* significant at 10%; ** significant at 5%; *** significant at 1%.
The dependent variable is
RET, a bank��s average (–1, +1) CAR across all deals announced by the bank over a
single calendar year. The explanatory variable
RET is the average CAR over the past j calendar years, where
j =
{1,2,3}.
SHARE is the bank��s market share, by value of deals, over the past
j calendar years. To be included in the
regression, a bank must have announced at least 2
j deals over the relevant period. In Panel B, an interaction term
between
RET and
SHARE is included. The marginal effect of RET at a
SHARE of 5% is also included. In Panel
C,
FIRSTHALF equals 1 if the dependent variable is in 1994 or earlier, and 0 otherwise. The data are pooled
across all banks, and regressions are estimated using year fixed effects, clustering standard errors by bank. The
sample period is 1981–2007, and
t-statistics are in parentheses.
skills have become commoditized over time, in part due to rapid increases
in computer power and the rise in general skills resulting from MBA de-
grees. If true, and if our results are driven primarily by the early part of our
sample, our findings become less interesting because we are documenting a
phenomenon that is no longer in existence. We investigate this hypothesis in
2306
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
Panel C of Table 6 by adding an additional regressor,
RET*FIRSTHALF, where
FIRSTHALF is a dummy variable that equals 1 if the dependent variable is in
the first half of our performance sample, 1981–1994. This is an appropriate
cutoff, as Morrison and Wilhelm (2008, Figure 3) document a marked increase
in computer power around 1993–1994. The interaction term is positive but
insignificant in all six specifications. In the presence of the interaction term,
the coefficient on
RET now measures persistence using the second half of the
dataset only. Importantly, even though we have less power, past
RET remains
significant in all specifications for
j �� 2, suggesting that the ability differences
we document remain significant today.
4. Do Clients Chase Performance?
Even if a financial intermediary possesses superior ability (our fixed-effects re-
sults of Table 3), Berk and Green (2004) show that this need not translate into
performance persistence (our quantile and regression results, of Tables 5–6)
if two necessary conditions hold: Clients chase past returns, and there are di-
minishing returns to scale. Indeed, our persistence results for M&A advice
contrast the lack of persistence in mutual funds (e.g., Carhart 1997). Since past
RET positively predicts future
RET, it would seem logical for clients to select
banks based on past performance, in which case persistence will disappear if
there are diminishing returns to scale. For ability to translate into persistence,
it is sufficient for one of the necessary conditions present in Berk and Green
(2004) to be violated. We investigate whether clients respond to past perfor-
mance (the first condition) because it is much cleaner to measure than returns
to scale.
Table 7 studies the effect of past performance on a bank��s future market
share. Since bank/client relationships take time to develop, large banks are
likely to have persistently high market shares irrespective of past performance.
We therefore include either a bank fixed effect or the bank��s past market share
as a regressor. Standard errors are clustered at the bank level.
Strikingly, even though
RET is a positive predictor of future performance,
the first six columns of Table 7 show that it is an insignificant determinant
of market share (consistent with Rau 2000). By contrast, even though market
share negatively predicts performance, it is strongly significantly related to
future share. The last three columns of Table 7 repeat the analysis adding an
interaction between
RET and
SHARE that is generally insignificant, suggesting
that the results are similar across all sizes of banks.19 This lack of performance
chasing is a sufficient condition for ability to translate into persistence and
thus underpins the results of Section 3. The strong performance-chasing by
19 The coefficient on the interaction term in the one-year analysis is significant at the 10% level, but the marginal
effect of
RET at a 5% market share is not significant.
2307
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
T
able
7
Determinants
of
mark
et
share
1yr
2yrs
3yrs
1yr
2yr
3yrs
1yr
2yrs
3yrs
Constant
0
.0099
0
.0068
0
.0059
0
.0098
0
.0068
0
.0059
(4
.55
)∗∗∗
(4
.08
)∗∗∗
(4
.64
)∗∗∗
(4
.59
)∗∗∗
(4
.07
)∗∗∗
(4
.65
)∗∗∗
RET
0
.0007
−
0
.0259
−
0
.0499
−
0
.0025
−
0
.0370
−
0
.0357
−
0
.0313
−
0
.0486
−
0
.0518
(0
.03
)
(0
.46
)
(0
.63
)
(0
.14
)
(1
.14
)
(1
.09
)
(1
.49
)
(1
.74
)∗
(1
.52
)
SHARE
0
.6805
0
.8118
0
.8489
0
.6830
0
.8114
0
.8480
(13
.09
)∗∗∗
(17
.67
)∗∗∗
(28
.08
)∗∗∗
(13
.26
)∗∗∗
(17
.70
)∗∗∗
(28
.82
)∗∗∗
RET
X
3
.1504
1
.0993
1
.5095
SHARE
(1
.80
)∗
(0
.30
)
(0
.42
)
Bank
FE
Y
es
Y
es
Y
es
No
No
No
No
No
No
#
obs
1,168
1,079
1,017
1,168
1,079
1,017
1,168
1,079
1,017
R
2
(%)
59
.72
60
.69
61
.14
44
.53
54
.83
58
.23
44
.81
54
.85
58
.24
RET
(at
0
.1262
0
.0063
0
.0237
5%
SHARE)
(1
.60
)
(0
.04
)
(0
.15
)
*
significant
at
10%;
**
significant
at
5%;
***
significant
at
1%.
The
dependent
v
ariable
is
a
bank��s
mark
et
share,
by
v
alue
of
deals,
in
one
particular
year.
RET
is
the
bank��s
av
erage
(–1,
+1)
CAR
o
v
er
the
past
j
calendar
years,
where
j
={1,2,3
}.
SHARE
is
the
bank��s
mark
et
share,
by
v
alue
of
deals,
o
v
er
the
past
j
calendar
years.
The
data
are
pooled
across
all
banks,
and
regressions
are
estimated
using
bank
fix
ed
effects
and
clustering
standard
errors
at
the
bank
lev
el.
T
o
be
included
in
the
results,
a
bank
must
hav
e
announced
at
least
2
j
deals
o
v
er
the
period
used
to
estimate
RET
and
SHARE
.
The
sample
period
is
1981–2007,
and
t-statistics
are
in
parentheses.
2308
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
mutual fund investors, and the absence of such behavior by acquirers, may
thus reconcile the performance persistence in M&A with its absence in mutual
funds. Our results echo those of Kaplan and Schoar (2005), who document
persistence in private equity performance (also found by Axelson et al. 2010)
and that top-performing funds grow proportionally more slowly than poorly
performing funds, which explains why persistence can arise.
The significance of past market share, even though it negatively predicts fu-
ture performance, and irrelevance of past returns, even though it is a positive
predictor, may appear at first glance to be inefficient. However, applying the
Berk and Green (2004) framework to our setting, responsiveness to perfor-
mance requires not only learning about ability from past returns but also
competitive provision of M&A mandates by clients. Thus, lack of performance-
chasing can occur if either assumption is violated, and therefore has different
interpretations. Mandate awards may be noncompetitive if clients build up
relationship-specific capital with a particular bank, which can be leveraged
by continuing to use it for future deals. Thus, while past market share is a
predictor of future mandates, it may not be a
determinant (i.e., actively used
by acquirers in their selection decisions)—instead, its significance arises as it
proxies for the extent of existing relationships. Table 8 investigates the
relationship-specific capital hypothesis by studying repeat acquirers, who have
conducted at least one acquisition in the prior five years while being public.
Using a previously mandated advisor is associated with a lower CAR of 0.24
percentage points (
t-statistic of 1.74). If the advisor had generated a nega-
tive average CAR for that particular client in question, the CAR is 0.79 per-
centage points lower than using past advisors that generated positive CARs
Table 8
Relationship between a deal��s (–1, +1) CAR and the use of a past advisor
Panel A
(1)
(2)
(3)
(4)
Retained only
Retained an old
advisors with
advisor with
Did not retain
Retained an
positive past
negative past
and old advisor
old advisor
performance
performance
CAR
0.33%
0.09%
0.46%
−0.33%
t-statistic
(3.28)∗∗∗
(0.89)
(3.37)∗∗∗
(2.41)∗∗
# obs
3,793
4,291
2,263
2,028
Panel B
(1)–(2)
(1)–(3)
(1)–(4)
(3)–(4)
CAR
0.24%
−0.13%
0.66%
0.79%
t-statistic
(1.74)∗
(0.78)
(3.89)∗∗∗
(4.09)∗∗∗
* significant at 10%; ** significant at 5%; *** significant at 1%.
Deals in which the acquirer has made an acquisition in the last five years while public are considered. Panel A
divides deals into groups according to whether the acquirer retained an advisor from a past transaction. Panel B
examines the differences in average CAR between groups. The sample period is 1985–2007, and
t-statistics are
in parentheses.
2309
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
(
t-statistic of 4.09). These findings are consistent with the
RET persistence re-
sults of Section 3. However, even if using a past advisor is correlated with poor
future M&A performance, it is not inefficient if the bank provides many other
non-M&A services. For example, Yasuda (2005) and Ljungqvist, Marston, and
Wilhelm (2006, 2009) find that clients select underwriters based on lending re-
lationships or analyst coverage. Even if banks are providing no other services,
clients may be ��locked in�� to a past advisor because working on a previous deal
gives the advisor an information monopoly: Ljungqvist, Marston, and Wilhelm
(2006, 2009) find that past co- or lead underwriters are typically appointed as
future lead underwriters. Asker and Ljungqvist (2010) highlight another source
of lock-in: clients�� wish to avoid sharing banks with product market rivals.
Learning about ability from past returns will not occur if clients are unaware
of the persistence of
RET and mistakenly believe that market share is a good
measure of quality. Under this interpretation, market share is significant not
because clients are locked in and it proxies for existing relationships, but be-
cause clients actively select on it. Indeed, both the insignificance of
RET and
the significance of
SHARE are fully consistent with real-life practices in the
investment-banking industry, where league tables on market share are widely
publicized and used as proxies for expertise. Therefore, industry participants
have grown to take it for granted that market share equates with quality; sim-
ilarly, many academic studies, such as those by Rau (2000), Kale, Kini, and
Ryan (2003), and Hunter and Jagtiani (2003), use market share as their mea-
sure of quality. However, we show that it is actually negatively correlated with
performance. However, returns are less publicized, which hinders clients�� abil-
ity to use returns as an additional metric to learn about quality even if they are
not locked in.
Given the lack of performance-chasing and the importance of prior relation-
ships, it is logical for banks to accept even bad deals. Not only will the mandate
boost fee income today, but it will also create new relationships and thus the
ability to win future mandates. Even though accepting bad deals will depress
RET, this does not reduce future mandates, due to either lock-in or failure to
learn. Indeed, if certain banks are systematically nonselective and accept bad
deals, this would lead to the negative correlation between market share and
RET that we document.
As a preliminary investigation into the importance of lock-in for M&A, we
study the extent to which clients switch M&A advisors. During 1985–2007,
we find that on 21.4% of M&A deals where the acquirer had made at least
one acquisition in the past five years, the advisors also covered all of the
acquirer��s M&A deals over the past five years. On 50.1% of deals, the
advisors covered at least one deal over the past five years. Ljungqvist and
Wilhelm (2005) find that 64.1% of equity issuers used the same underwriter
for their initial public offering and first seasoned equity offering. Our fig-
ures are somewhat lower, tentatively suggesting that lock-in may be slightly
weaker in M&A. One potential explanation is that a significant amount of
2310
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
M&A advice is target specific, and thus knowledge built up during one trans-
action may be less applicable for future deals. Our findings are consistent with
Francis, Hasan, and Sun (2009), who find that fewer than 20% of deals are
conducted by exclusive advisors, which they interpret as evidence that exist-
ing relationships are less influential for advisor choice in M&A than in other
banking services.
Given the importance of other banking services, we also gather SDC is-
suance data to measure debt and equity underwriting relationships, and
Dealscan data to measure lending relationships.20 Since Dealscan data are
sparse until 1988, the lending numbers are calculated from 1993; for consis-
tency, we use the same timeframe for the issuance data. We find that on 9.5%
of M&A deals over 1993–2007, the advisors also covered all of the acquirer��s
issuance over the past five years; on 45.4% of deals, banks covered at least
one issue. For lending, the numbers drop to 6.7% and 15.8%, respectively.
While we do not have a benchmark for a formal test, these relatively low num-
bers suggest that clients use different advisors for their M&A business than
for their underwriting and borrowing decisions. Indeed, while there are plau-
sibly strong synergies between lending and bond underwriting (Yasuda 2005),
the connection between lending and M&A advice appears to be weaker. We
also attempt to study the selection decisions of clients who are not locked in
and thus have freedom on advisor choice. We take a subsample of clients that
engaged in no M&A deals, issuance, or borrowing with our 143 advisors in
the past five years, and thus have no existing relationships. We note that this
is not a perfect measure of freedom. First, it may include ��unfree�� clients: If
some of these acquirers intend to engage in issuance or borrowing in the future,
they may select an M&A advisor who provides these services. Second, it may
exclude ��free�� clients: Some clients may have engaged in issuance or borrow-
ing with an M&A advisor in the past but have no intention of doing so in the
foreseeable future. If lock-in, rather than failure to learn, is the reason for the
aggregate insignificance of
RET for market share in Table 7, then
RET should
be positively correlated with future mandates for ��free�� acquirers. By contrast,
Table 9 shows that
RET is negatively correlated with future mandates in most
specifications (though statistically insignificant) for this subsample. This sug-
gests that failure to learn is a cause of the insignificance of past performance.
However, we note that these results are only suggestive, due to the difficulty
of identifying ��free�� clients, mentioned earlier, and our small sample size: We
can identify ��free�� clients from only 1993 onward, and only 1,224 deals were
conducted by such acquirers.
If clients are indeed not fully locked in, our results suggest that they may
be able to improve their advisor-selection decisions by using information on
20 We thank Michael Roberts for providing us with the table to link Dealscan to Compustat, used in Chava and
Roberts (2008).
2311
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
Table 9
Logit regressions of advisor choice by acquirers without a previous M&A, issuance, or lending relation-
ship over the last five years with one of our 143 banks
1yr
1yr
2yrs
2yrs
3yrs
3yrs
Past j years
Constant
−4.4597
−4.5908
−4.4199
−4.5797
−4.4063
−4.5746
(168.92)∗∗∗
(144.50)∗∗∗
(154.13)∗∗∗
(125.75)∗∗∗
(141.84)∗∗∗
(114.91)∗∗∗
RET
−0.3508
0.0059
−2.2120
−1.3952
−2.5828
−1.1367
(0.37)
(0.01)
(1.56)
(0.90)
(1.47)
(0.59)
SHARE
4.0123
4.5560
4.5307
(8.26)∗∗∗
(8.40)∗∗∗
(8.07)∗∗∗
# obs
67,806
67,806
63,586
63,586
60,791
60,791
Pseudo
R2 (%)
0.00
0.56
0.02
0.68
0.02
0.66
* significant at 10%; ** significant at 5%; *** significant at 1%.
The analysis is at the acquirer-bank level, and the dependent variable equals 1 if the acquirer mandated that
particular bank. Banks that do not advise on M&A in the year of the deal are treated as unavailable and excluded
from the analysis.
RET is the bank��s average (–1, +1) CAR over the past
j calendar years.
SHARE is the bank��s
market share by value of deals over the past
j calendar years. Standard errors are clustered by deal. The sample
period is 1993–2007, and
t-statistics are in parentheses.
past returns to supplement market-share information.21 An increased focus on
returns may in turn dissuade banks from accepting value-destructive transac-
tions. The findings also have implications for the nature of contracts between
acquirers and advisors. McLaughlin (1990) finds that banks are paid for deal
completion but not value creation. He suggests that reputational concerns may
be sufficient to align banks with clients, since a bank that performs poorly
will not win future mandates. However, the insignificance of
RET implies that
banks�� implicit incentives are also low—if clients are locked in or do not learn
from
RET, banks can be less concerned with returns—and so explicit incen-
tives would be valuable. In a similar vein, clients frequently solicit fairness
opinions to verify that the terms are ��fair�� (Kisgen, Qian, and Song 2009). As
part of its mandate, an advisor should ensure that the client is undertaking only
favorable deals in the first place, and there should be no need for a separate
fairness opinion. The prevalence of such opinions is consistent with the view
that incentives to act in clients�� interests are insufficient, due to either lock-in
or failure to learn.
5. Conclusion
This article finds a significant investment-bank fixed effect in the announce-
ment returns to an acquisition. The positive association between certain banks
and high returns can be predicted by clients using past performance—a bank��s
21 Since shareholders are diversified, they are not concerned with idiosyncratic risk. Thus, a value-maximizing
manager should maximize expected CAR rather than a measure adjusted for the variance. However, managers
may care about the variance of CAR if they are concerned with their own undiversified human capital. We find
that a bank��s
RET is highly correlated with its ��Sharpe ratio�� of
RET divided by the standard deviation of CAR.
Over three-year non-overlapping windows beginning in 1981, we find the average Spearman rank correlation of
RET and the Sharpe ratio to be 0.965.
2312
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
returns are persistent. While most prior research attributes the entire CAR to
the advisor, we remove the component that can be explained by acquirer char-
acteristics; the orthogonal component remains persistent. These results suggest
that certain banks have ability in identifying acquisitions or negotiating terms,
or trustworthiness in turning down bad deals. They contrast prior findings that
bank quality, as measured by market share or reputation, has no positive ef-
fect on M&A outcomes, thus suggesting that banks do not matter. The lack of
performance-chasing by clients potentially explains why persistence exists in
M&A advice even though it is absent in mutual fund performance.
This article suggests a number of questions for future research. First, it may
be interesting to study
why clients do not chase performance and whether such
behavior is efficient—i.e., disentangle whether it results from rational lock-in
or inefficient failure to learn (e.g., due to an acquirer��s private benefits from
working with a prestigious bank). Agency variables such as governance may
explain advisor choice, just as they do for acquirer returns (Masulis, Wang,
and Xie 2007). Second, the low returns to ability appear puzzling. While su-
perior performance in equity underwriting is rewarded with higher future mar-
ket share (Dunbar 2000; Hoberg 2007), and superior returns to private equity
investing increases future fund flows (Chung et al. 2010), there seems to be
little reward for good M&A advice. Third, we have focused on acquirer returns,
because these are frequently negative, and so advisor selection is particularly
important for bidders to ensure value creation; in addition, far more bidders are
public than are targets. It would be fruitful to investigate whether target returns
also exhibit a bank fixed effect, and whether the banks that consistently create
value for bidders are also skilled at defense mandates.
References
Asker, J., and A. Ljungqvist. 2010. Competition and the Structure of Vertical Relationships in Capital Markets.
Journal of Political Economy 118:599–647.
Axelson, U., T. Jenkinson, P. Strömberg, and M. S. Weisbach. 2010. Borrow Cheap, Buy High? The Determi-
nants of Leverage and Pricing in Buyouts. Working Paper, London School of Economics.
Berk, J. B., and R. C. Green. 2004. Mutual Fund Flows and Performance in Rational Markets.
Journal of Political
Economy 112:1269–95.
Bertrand, M., and A. Schoar. 2003. Managing with Style: The Effect of Managers of Firm Policies.
Quarterly
Journal of Economics 118:1169–208.
Bowers, H., and R. Miller. 1990. Choice of Investment Banker and Shareholders�� Wealth of Firms Involved in
Acquisitions.
Financial Management 19:34–44.
Carhart, M. 1997. On Persistence in Mutual Fund Performance.
Journal of Finance 52:57–82.
Chava, S., and M. R. Roberts. 2008. How Does Financing Impact Investment? The Role of Debt Covenants.
Journal of Finance 63:2085–121.
Chung, J., B. Sensoy, L. H. Stern, and M. S. Weisbach. 2010. Pay for Performance from Future Fund Flows: The
Case of Private Equity. Working Paper, Ohio State University.
Diether, K. B., K. H. Lee, and I. M. Werner. 2009. Short-sale Strategies and Return Predictability.
Review of
Financial Studies 22:575–607.
2313
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
The Review of Financial Studies / v 24 n 7 2011
Dunbar, C. 2000. Factors Affecting Investment Bank Initial Public Offering Market Share.
Journal of Financial
Economics 55:3–41.
Ertugurul, M., and K. Krishnan. 2010. Advisor Skill and Acquisition Performance: Do Investment Bankers Make
a Difference? Working Paper, University of Massachusetts.
Fama, E. F. 1998. Market Efficiency, Long-term Returns, and Behavioral Finance.
Journal of Financial Eco-
nomics 49:283–306.
Fernando, C. S., V. A. Gatchev, and P. A. Spindt. 2005. Wanna Dance? How Firms and Underwriters Choose
Each Other.
Journal of Finance 60:2437–69.
Francis, B. B., I. Hasan, and X. Sun. 2009. Choice of Financial Advisors and Reasons to Switch: Evidence from
M&As. Working Paper, Rensselaer Polytechnic Institute.
Gompers, P. A., J. L. Ishii, and A. Metrick. 2003. Corporate Governance and Equity Prices.
Quarterly Journal
of Economics 118:107–55.
Hackbarth, D., and E. Morellec. 2007. Stock Returns in Mergers and Acquisitions.
Journal of Finance 63:
1213–52.
Hoberg, G. 2007. The Underwriter Persistence Phenomenon.
Journal of Finance 62:1169–206.
Hunter, W., and J. Jagtiani. 2003. An Analysis of Advisor Choice, Fees, and Effort in Mergers and Acquisitions.
Review of Financial Economics 12:65–81.
Jaffe, J., D. Pedersen, and T. Voetmann. 2009. Mergers and Persistence: A Test of Differential Skill in Corporate
Finance. Working Paper, University of Pennsylvania.
Jegadeesh, N., and S. Titman. 1993. Returns to Buying Winners and Selling Losers: Implications for Stock
Market Efficiency.
Journal of Finance 48:65–91.
Kale, J. R., O. Kini, and H. E. Ryan. 2003. Financial Advisors and Shareholder Wealth Gains in Corporate
Takeovers.
Journal of Financial and Quantitative Analysis 38:475–501.
Kaplan, S. N., and A. Schoar. 2005. Private Equity Performance: Returns, Persistence, and Capital Flows.
Jour-
nal of Finance 60:1791–823.
Kisgen, D., J. Qian, and W. Song. 2009. Are Fairness Opinions Fair? The Case of Mergers and Acquisitions.
Journal of Financial Economics 91:179–207.
Lehn, K., and M. Zhao. 2006. CEO Turnover after Acquisitions: Are Bad Bidders Fired?
Journal of Finance
61:1759–811.
Lemmon, M. L., M. R. Roberts, and J. F. Zender. 2008. Back to the Beginning: Persistence and the Cross-section
of Corporate Capital Structure.
Journal of Finance 63:1575–608.
Ljungqvist, A., F. Marston, and W. J. Wilhelm, Jr. 2006. Competing for Securities Underwriting Mandates:
Banking Relationships and Analyst Recommendations.
Journal of Finance 61:301–40.
. 2009. Scaling the Hierarchy: How and Why Investment Banks Compete for Syndicate Co-management
Appointments.
Review of Financial Studies 22:3977–4007.
Ljungqvist, A., and W. J. Wilhelm, Jr. 2005. Does Prospect Theory Explain IPO Market Behavior?
Journal of
Finance 60:1759–90.
Maksimovic, V., and G. M. Phillips. 2002. Do Conglomerate Firms Allocate Resources Inefficiently across
Industries? Theory and Evidence.
Journal of Finance 57:721–68.
Masulis, R., C. Wang, and F. Xie. 2007. Corporate Governance and Acquirer Returns.
Journal of Finance
62:1851–89.
McLaughlin, R. 1990. Investment-banking Contracts in Tender Offers: An Empirical Analysis.
Journal of
Financial Economics 28:209–32.
2314
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
Do Investment Banks Matter for M&A Returns?
Michel, A., I. Shaked, and Y.-T. Lee. 1991. An Evaluation of Investment Banker Acquisition Advice: The Share-
holders�� Perspective.
Financial Management 20:40–49.
Mikhail, M., B. Walther, and R. Willis. 2004. Do Security Analysts Exhibit Persistent Differences in Stock-
picking Ability?
Journal of Financial Economics 74:67–91.
Moeller, S. B., F. P. Schlingemann, and R. M. Stulz. 2004. Firm Size and the Gains from Acquisitions.
Journal
of Financial Economics 73:201–28.
Morrison, A. D., and W. J. Wilhelm, Jr. 2007.
Investment Banking: Institutions, Politics, and Law. Oxford, UK:
Oxford University Press.
. 2008. The Demise of Investment-banking Partnerships: Theory and Evidence.
Journal of Finance
63:311–50.
Rau, P. R. 2000. Investment Bank Market Share, Contingent Fee Payments, and the Performance of Acquiring
Firms.
Journal of Financial Economics 56:293–324.
Rau, P. R., and T. Vermaelen. 1998. Glamour, Value, and the Post-acquisition Performance of Acquiring Firms.
Journal of Financial Economics 49:223–53.
Servaes, H., and M. Zenner. 1996. The Role of Investment Banks in Acquisition.
Review of Financial Studies
9:787–815.
Yasuda, A. 2005. Do Bank Relationships Affect the Firm��s Underwriter Choice in the Corporate-bond Under-
writing Market?
Journal of Finance 60:1259–92.
2315
at University of Pennsylvania Library on June 26, 2011
rfs.oxfordjournals.org
Downloaded from
