Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
-
Upload
joebloggsscribd -
Category
Documents
-
view
225 -
download
0
Transcript of Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
1/32
ALLEN N. BERGER
ROBERT DEYOUNG
Technological Progress and the Geographic
Expansion of the Banking Industry
We test some predictions about the effects of technological progress ongeographic expansion using data on banks in U.S. multibank holding compa-nies over 1985-98. Specifically, we test whether over time (1) parentalcontrol over affiliate banks has increased and (2) the agency costs ofdistance from the parent have decreased. The data suggest that bankingorganizations control over affiliates has been increasing over time and thatthe agency costs of distance have decreased somewhat over time. Thefindings are consistent with the hypothesis that technological progress hasfacilitated the geographic expansion of the banking industry.
JEL codes: G21, G28, G34, L11Keywords: banks, efficiency, mergers, productivity, technological progress.
Over the past few decades, the banking industry has been
in a constant process of geographic expansion, both within nations and across
nations. At one time, nearly all customers were served by locally based institutions. In
contrast, it is now much more likely that the bank or branch providing services is
owned by an organization headquartered a substantial distance away, perhaps in
another state, region, or nation. To illustrate, between 1985 and 1998 the distance
between the largest bank and the other affiliate banks in U.S. multibank holding
The opinions expressed do not necessarily reflect those of the Federal Reserve Board, the ChicagoReserve Bank, or their staffs. The authors thank Bruno Biais, Bill Brainard, Neil Esho, Mark Flannery,Dan Gropper, Chris James, Loretta Mester, Steven Ongena, Marco Pagano, Fabio Panetta, Tony Saunders,Rick Sullivan, John Wolken, Oved Yosha, two anonymous referees, and other participants at the JFISymposium on Corporate Finance with Blurring Boundaries Between Banks and Financial Markets, theYale University Conference on The Future of American Banking: Historical, Theoretical, and EmpiricalPerspectives, and the Federal Reserve Bank of Chicagos Bank Structure and Competition Conferencefor helpful comments, and Nate Miller and Phil Ostromogolsky for outstanding research assistance.
Allen N. Berger is a Senior Economist of the Board of Governors of the Federal ReserveSystem and Wharton Financial Institutions Center (E-mail: abergerfrb.gov). RobertDeYoung is Associate Director, Research Branch, at the Federal Deposit Insurance Corpora-tion (E-mail: rdeyoungfdic.gov).
Received March 31, 2003; and accepted in revised form December 7, 2004.
Journal of Money, Credit, and Banking,Vol. 38, No. 6 (September 2006)Copyright 2006 by The Ohio State University
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
2/32
1484 : MONEY, CREDIT, AND BANKING
companies (MBHCs) increased by over 50% on average, from 123.35 to 188.91
miles, as many MBHCs acquired banks in other states and regions.The role of deregulation in the geographic expansion of this industry is wellunderstood. In the United States, a series of deregulations in the 1980s and early
1990s removed restrictions on intrastate and interstate banking, culminating with theRiegle-Neal Interstate Banking and Branching Efficiency Act of 1994, whichpermitted interstate branching in almost all states as of June 1997. In the EuropeanUnion, the set of actions known as the Single Market Programmeespecially thesingle license provision of the Second Banking Co-ordination Directive of 1989essentially allowed banking organizations to expand continent-wide. In LatinAmerica, the transition nations of Eastern Europe, and other regions, explicit and
implicit regulatory barriers to foreign entry have fallen, allowing banking organiza-
tions headquartered in other nations to gain significant market shares.The role of technological progress in facilitating geographic expansion is less
well understood. In any industry, there are potential diseconomies to geographicexpansion in the form of agency costs associated with monitoring junior managers ina distant locale. Improvements in information processing and telecommunicationsmay lessen these agency costs by improving the ability of senior managers locatedat the organizations headquarters to monitor and communicate with staff at distantsubsidiaries. In the banking industry, technologies such as ATM networks andtransactional Internet websites allow banks to interact efficiently with customersover long distances. Advances in financial technologies also facilitate long-distance
interfaces with customers. Greater useof quantitative methods in applied finance, suchas credit scoring, may allow banks to extend credit without geographic proximityto the borrower by hardening their credit information (Stein 2002). Similarly, thenew products of financial engineering, such as derivative contracts, may allow banksto unbundle, repackage, or hedge risks at low cost without respect to the distancefrom the counterparty. These financial innovations may allow senior managers tomonitor decisions made by loan officers and managers at distant affiliate banks moreeasily, and evaluate and manage the contributions of individual affiliate banks tothe organizations overall returns and risk more efficiently.
In this study, we examine data on commercial banks in U.S. MBHCs over 1985-98, and test whether these data are consistent with some predictions about the effectsof technological progress. We assess changes over time in the ability of managersat the lead or parent bank in the MBHC to control the performance of theirnonlead affiliates, as well as changes over time in the agency costs associatedwith the distance between the parents and affiliates. While previous studies haveexamined these managerial control and geographic distance phenomena on a cross-sectional basis, this study breaks new ground by testing whether technologicalinnovation over time has improved the ability of banking companies to control theiroperations and/or mitigate the agency costs associated with distance. We study U.S.MBHCs over 1985-98 because the lions share of geographic expansion by U.S. banksduring this time period used the MBHC framework and because these organiza-
tions significantly increased their use of new information processing, telecommuni-cations, and financial technologies over this interval.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
3/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1485
We define control as the ability of the organizations senior managers to export
their managerial skills, policies, and procedures to their affiliate banks. Since wecannot directly observe how banks are managed, we proxy for control by measuring
the extent to which the efficiency rank of a nonlead bank affiliate varies with the
efficiency rank of the lead bank in the same MBHC. Efficiency ranks are described
in detail below. We measure the control derivative NONLEADRANKLEADRANKt, where NONLEADRANK is the efficiency rank of a nonlead bankaffiliate in a MBHC, LEADRANK is the efficiency rank of the lead bank in the
same MBHC, and |tindicates evaluation at period t. This derivative is expected to
be positive and to lie between 0 (no control) and 1 (very good control). We estimate
the control derivative in a multiple regression framework, and it represents the aver-
age degree to which multibank organizations in year tare able to control their affiliates.
Our hypothesis that technological progress has improved the control of banking
organizations yields the prediction thatNONLEADRANKLEADRANKtshouldbe increasing with t.
Our maintained assumptions are that the senior managers of the organization are
located at the lead bank (the largest bank in the MBHC) and that LEADRANK is
a good proxy for the skills, policies, and procedures available to manage the entire
organization. Importantly, both well-managed and poorly managed organizations
can have a high degree of control. Efficient senior managers with substantial control
may transfer best-practices techniques to junior managers at their affiliate banks,
whereas inefficient senior managers with significant control may transfer worst
practices to affiliates.
We define agency costs of distance as the additional expenses or lost revenues
that arise as senior managers try to monitor and control local managers from a
greater distance. The general effect of these agency costs is expected to be negative,
as local managers operating at greater distance have more freedom to pursue their
own objectives, resulting in lower efficiency ranks. These agency costs may also
incorporate a potentially offsetting senior management effect: operating at a greater
distance from very inefficient senior managers may enhance efficiency if it interferes
with the transfer of worst practices. Thus, the agency costs of distance may depend on
the efficiency of senior management, with more costs involved with increased
distance from good senior management and less costs, or even gains in affiliateefficiency, associated with increased distance from bad senior management. Since
we cannot directly observe agency costs, we proxy (inversely) for the agency costs
of distance by measuring the extent to which the efficiency ranks of nonlead affiliate
banks vary with the distance from their lead banks. We measure the distance
derivativeNONLEADRANKlnDISTANCEt,where lnDISTANCE is the natu-ral log of the distance in miles between the affiliate bank and its lead bank. For
any value of t, this derivative is expected to be negative for most banks, as
greater distance implies more problems in aligning the incentives of local manag-
ers with those of the organization. However, when senior management is very
inefficient, the senior management effect may overwhelm the general agency costeffect, yielding a net positive distance derivative. We investigate this possibility
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
4/32
1486 : MONEY, CREDIT, AND BANKING
below. The distance derivative is estimated in the same multiple regression frame-
work as the control derivative. Our hypothesis that technological progress hasreduced the agency costs of distance yields the general prediction that
NONLEADRANKlnDISTANCEt, should be increasing with t.We acknowledge that factors other than technological change also affect the
control of parent organizations and the agency costs of distance. For this reason,
we include variables in our empirical analysis to account for changes in competi-
tion, regulation, and other environmental conditions that may potentially influence
parental control and the agency costs of distance, and we conduct numerous tests
to ensure that our results are not explained by sample selection bias, specification
choices, bank size class effects, choice of time period, and so forth. Moreover, as
in almost all studies of technological progress, it is difficult to deterministically link
performance improvements to technological change because technology is difficult
to specify, its effects are difficult to parameterize, and its adoption may in some
cases be endogenous to firm and industry performance. Nonetheless, previous
research suggests that banking firms are one of the best places to test for the effects of
technological improvement. Banks have embraced substantial advances in both
physical and financial technologies during the past two decades, and the broader
industry category of which banking is a part, Depository and Nondepository Financial
Institutions, is the most information technology-intensive industry in the United
States (Triplett and Bosworth 2002, Table 2). As well, when specific banking techno-
logies have been studied, such as improvements in the processing of electronic
payments, very large productivity gains have been directly linked to technological
changes (e.g., Hancock, Humphrey, and Wilcox 1999).1
1. REVIEW OF RELATED RESEARCH LITERATURE
To our knowledge, no prior studies have directly examined whether and by how
much technological change over time has improved the ability of senior bank
managers to control the performance of their affiliates, branches, or other business
units. As well, we are unaware of previous research investigating whether and by
how much technological advances have reduced the agency costs associated withthe distance from these geographically remote units. However, there are prior studies
of related topics, including the degree to which bank productivity has improved over
time, the degree to which bank-borrower distance has increased over time, the effects
of organizational form on bank performance, and the cross-sectional effects of control
and distance on bank efficiency. We briefly review some of this related research.
Because the banking industry is relatively information-intensive, it may have
been among the first industries to take meaningful advantage of the benefits of
information processing and telecommunicationstechnological advances that could
improve parental control over their affiliate banks and reduce the agency costs of
1. See Berger (2003) for a general review of technological progress in banking.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
5/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1487
distance. Data from the Bureau of Labor and Statistics, which are based on a
simple weighted measure of transactions per employee hour, are consistent withthis possibility (Furlong 2001). Studies using cost productivity or linear programming
methods to measure productivity for U.S. banks in the 1990s often found either
productivity declines or only very slight improvements (e.g., Wheelock and Wilson,
1999, Stiroh, 2000, Berger and Mester, 2003). However, the latter of these studies
found increasing profit productivity even while cost productivity declined, suggesting
substantial revenue-based productivity improvements. This finding suggests that
technological progress made banks more profit efficient by allowing them to offer
a wider array of costlier, but higher quality services that generated increased revenues
in excess of the higher costs.2
Recent research has also found that banks have been increasing the distances atwhich they make small business loans over time (e.g., Petersen and Rajan, 2002,
Hannan, 2003) and that this trend has been facilitated in part by the adoption of
the small business credit scoring technology, which is associated with increased
out-of-market lending (Frame, Padhi, and Woosley 2004). These findings are consis-
tent with the hypothesis that technology may have increased parental control and
reduced the agency costs of distance by making it easier to monitor loan officers
and other personnel at greater distances. Of course, these findings may have other
causes as well. For example, one study found a link between the increased integration
of state and regional economies and the geographic expansion of banks (Morgan,
Rime, and Strahan 2003), suggesting that a more geographically integrated economyeases the management tasks of MBHCs.
Other distance-related research has found that expanded geographic reach has
had generally favorable effects on bank performance. Some studies found that larger,
more geographically integrated institutions tend to have better risk-expected return
frontiers (e.g., Hughes, Lang, Mester, and Moon, 1996, Demsetz and Strahan, 1997).3
Others found that banking organization mergers and acquisitions (M&As) raise
profit efficiency in a way consistent with the benefits of improved geographic
diversification, although M&As may not have much effect on cost efficiency (e.g.,
Akhavein, Berger, and Humphrey, 1997, Berger, 1998). These studies generally
found that the pre-merger gap in efficiency between the acquirer and target hadrelatively little effect on the change in efficiency surrounding the M&A, suggesting
that the ability of acquirers to export their skills, policies, and procedures to targets
may be limited. In contrast, studies of international banking expansion tend to find
that foreign affiliates operate less efficiently than domestic banks in developed nations,
particularly in the United States (e.g., DeYoung and Nolle, 1996, Berger, DeYoung,
2. Studies using 1980s data often found productivity declines during the early part of the decadeassociated with adjustments to the deregulation of deposit rates (e.g., Berger and Humphrey 1992,Humphrey and Pulley 1997). Many of these studies also found that the industry had adjusted to the newregime by the late 1980s.
3. A study of simulated mergers among small U.S. banks suggested that such mergers may alsogenerate risk reductions, but that the key risk-reducing benefits for these banks stem from increasedsize, not greater geographic integration (Emmons, Gilbert, and Yeager 2004).
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
6/32
1488 : MONEY, CREDIT, AND BANKING
Genay, and Udell, 2000), although cross-border expansion may involve many other
costs due to social, economic, and legal differences across countries.A number of cross-sectional studies examined the effects of organizational form
and ownership on bank efficiency, with mixed results. Some have found that affiliate
banks in MBHCs are more efficient than nonaffiliated independent banks (e.g.,
Spong, Sullivan, and DeYoung, 1995, Mester, 1996), while others have found that
MBHCs are less efficient (e.g., Grabowski, Rangan, and Rezvanian 1993). Studies
of the efficiency of individual branches of large branch-banking organizations found
significant dispersion, consistent with relatively weak organizational control over
individual branches (e.g., Sherman and Ladino, 1995, Berger, Leusner, and Mingo,
1997). Finally, a study of banking offices in Texas finds that rural banking offices
are more likely than urban banking offices to be locally owned; the authors conclude
that the activities in which rural banks engage are more difficult to manage from a
distance and hence require decision-making authority at the local level (Brickley,
Linck, and Smith 2003).
To our knowledge, only two previous studies have directly explored the concepts
of control and agency costs of distance in banking. A study of 715 commercial
bank affiliates in five U.S. states in 1973 and 1974 found that the ability of MBHCs
to control the performance of their affiliates varies both across MBHCs and
across affiliates within MBHCs (Rose and Scott 1984). A more recent cross-
sectional analysis of U.S. banks over 1993-98 found that parent organizations exercise
significant control over the efficiency of their affiliates, although this control tends
to dissipate rapidly with the distance to the affiliate (Berger and DeYoung 2001).
These studies were purely cross-sectional in nature, and hence did not test for
intertemporal changes in control or agency costs of distance.
The current study extends this literature by testing whether the ability of banking
companies to control their affiliates has improved over time, presumably due to
access to new information processing, telecommunications, and financial technolog-
ies. We perform these tests for U.S. commercial bank holding companies from 1985
to 1998, a period during which (1) most large banking companies were still organized
as MBHCs with affiliates located at various distances from headquarters, and (2)
the implementation of new technologies potentially provided senior managers with
better tools for controlling the behavior of those affiliates. Thus, we test not onlywhether technological progress has enhanced the potential for managerial control,
but also whether technological advances have mitigated the agency costs of distance.
Neither of these questions has been tested in the extant literature.
2. METHODOLOGY AND DATA
We test for intertemporal changes in parental control and the agency costs of
distance by analyzing the effects of lead bank efficiency rank and distance to the
lead bank on the efficiency ranks of nonlead banks in U.S. MBHCs between 1985and 1998. The U.S. data over this 14-year period provide an excellent opportunity
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
7/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1489
for analyzing technological change, parental control, and distance effects. The U.S.
market is economically large and geographically vast, and for most of this timeperiod U.S. banking companies were generally required to use the MBHC framework
to operate in multiple states. Because this organizational framework requires each
affiliate to prepare a complete set of financial statements, we can observe the
performance of each affiliate bank (lead and nonlead) separately, which allows us
to estimate and test the changes in our control and distance derivatives. Obtaining
clear measures of efficiency, parental control, geographic distance, and agency
costs of distance are essential to our ultimate objective, which is to test whether
technological progress has significantly affected the ability of banking companies to
control their operations and/or mitigate the agency costs associated with distance.
We run our analyses separately by size of bank because of the different products,
markets, and technologies for large and small banks. Large banks tend to produce
more transactions-driven loans for large, wholesale customers based on hard
quantitative information, such as certified audited financial statements. In contrast,
small banks tend to specialize in relationship lending to small, retail customers
based on soft information culled from close contacts over time of the loan
officer with the firm, its owner, and its local community (e.g., Stein, 2002, Berger,
Miller, Petersen, Rajan, and Stein, 2005). Technological progress may affect the
control and distance derivatives differently for large and small banks. Advances in
information processing, telecommunications, and financial technologies are likelyto be better adapted to processing, transmitting, and analyzing the hard information
used by large banks than the soft information used by small banks, so there may
be a greater increase over time in the control of large nonlead affiliates. The newer
technologies often allow for the electronic transmission of hard information with
little or no effect of distance, which may also result in a larger increase in control
and a greater reduction in agency costs associated with longer distances for large
banks. However, any of these effects could alternatively be greater for small affiliates
than for large affiliates. It is possible that the technological improvements during the
sample period may have allowed MBHCs to apply some of the hard-information
techniques to small bank affiliates for the first time, in effect hardening their credit
information. Thus, there may be a greater increase in control and/or a larger
reduction in agency costs of distance over time for small bank affiliates than for large
affiliates as MBHCs bring small business credit scoring and other hard-information
techniques to small affiliates.
For these reasons, we separate the nonlead banks in MBHCs into two samples
based on size. Our main sample includes annual observations of nonlead banks with
gross total assets (GTA) in excess of $100 million (real 1998 dollars) in that year,
and our small bank sample includes observations of nonlead banks in which GTA
is less than or equal to $100 million (real 1998 dollars). We exclude from bothsamples all observations of banks that are less than 5 years old because prior
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
8/32
1490 : MONEY, CREDIT, AND BANKING
research found that very young banks do not approach the efficiency of older small
banks for several years.
4
As noted above, we assume that LEADRANK is a good measure of the managerial
ability of the senior staff of the MBHC and that these senior managers are located at
the lead bank. The first assumption allows us to interpretNONLEADRANK/LEADRANK as a good proxy for control or how closely the performance of non-lead managers conform to the performance of their senior managers at the lead bank.
The second assumption allows us to interpret NONLEADRANK/lnDISTANCE asa good proxy for the agency costs of distance, or the losses due to the extra
difficulties of senior managers in monitoring or controlling the performance of local
managers from a greater distance. We believe these assumptions to be reasonable
the senior management of large MBHCs also usually directly manages the largest
bank in the organization.
2.1 Regression Specification and Tests
Our analysis is primarily based on panel regressions of the efficiency rank of
nonlead banks in MBHCs (NONLEADRANK) on the efficiency rank of the lead
bank (LEADRANK) and the log of the distance to the lead bank (lnDISTANCE).
We allow the control and distance effects to vary over time by including a time
variable (t) in these regressions. A number of additional exogenous variables
are included to account for other factors that may affect the efficiency of the nonlead
bank. These regressions take the form:
NONLEADRANKit
1*LEADRANKit 2*lnDISTANCEit 3*LEADRANKit*lnDISTANCEit
1*t 1/22*t2 1*t*LEADRANKit 2*t*lnDISTANCEit
3*t*LEADRANKit*lnDISTANCEit 1/21*t2*LEADRANKit
1/22*t2*lnDISTANCEit 1/23*t2*LEADRANKit*lnDISTANCEit (1)
4*MSAit 5*HERFit 6*lnBKASSit 7*1
/2lnBKASS2
it
8*lnHCASSit 9*1/2lnHCASS2it 10*BKMERGEit 11*HCMERGEit 12*MNPLit
13*NUMAFFSit 14*1/2NUMAFFS2it 15*UNITBit 16*LIMITBit 17*INTERSTATEit 18*ACCESSit 19-68*STATE DUMMIESi it,
4. DeYoung and Hasan (1998) found that the average new bank chartered between 1980 and 1994
needed nine years to become as profit efficient as the average mature small bank, although most of thisperformance shortfallhad been erasedby thefifth year. DeYoung (2003) found that de novo banks charteredafter 1994 tended to mature more quickly than those from the pre-1994 period.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
9/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1491
where i indexes the nonlead affiliate bank and t indexes the time period, t
1,...,14 for the years 1985-98. We use OLS to estimate Equation (1) separately andindependently for the main sample and the small bank sample. Within each of these
samples, we also estimate Equation (1) for all of the observations and for a subset
that includes only survivor banks that continued to exist through the end of the
sample period.
The key variables in Equation (1) are NONLEADRANK, LEADRANK, and
lnDISTANCE. NONLEADRANK is the profit efficiency rank of the nonlead bank
affiliate. LEADRANK is the profit efficiency rank of the lead bank, defined as the
largest bank in the MBHC. lnDISTANCE is the natural log of the distance in miles
as the crow flies between the cities or towns in which the lead and nonlead
banks are located (one mile added to DISTANCE before logging). The natural log
form of lnDISTANCE allows for the likelihood that the travel time or cost per
mile is decreasing in distance, i.e., time and cost economies of scale in distance.
The database from which the distances are calculated matches the latitude and
longitude over more than 19,000 different U.S. locations, nearly all cities, towns,
and counties with over 5000 inhabitants.5
We include the interaction term LEADRANK*lnDISTANCE in Equation (1) to
account for the likelihood that parental control diminishes with distance or that
the agency costs of distance may be greater when the lead bank is more efficiently
managed (the senior management effect discussed above). We use a quadratic speci-
fication for the time variable t, specifying both tand 1/2t2, and we interact both of
these terms with the other important regressors, LEADRANK, lnDISTANCE, and
LEADRANK*lnDISTANCE. This specification smoothes out year-to-year fluctua-
tions in the control derivative and the distance derivative over time, and allows
these derivatives to follow nonlinear time paths.
We expect the control derivative to be positive for all values oftand to be between
0 (no control) and 1 (very good control). We test the following null hypothesis:
NONLEADRANKLEADRANKtmean 1 3*lnDISTANCE
(1 3*lnDISTANCE)*t (1/21 1/23*lnDISTANCE)*t2 0 , (2)
where Equation (2) is evaluated for different values of lnDISTANCE. Once again
we note that profit efficiency and parental control are independent concepts: efficient
senior managers with substantial control may transfer best practices to junior manag-
ers at their affiliates, while inefficient senior managers with substantial control may
transfer worst practices to affiliate banks.
Our hypothesis that technological progress has improved the control of banking
organizations predicts that NONLEADRANKLEADRANKt should be in-creasing over time for any given value of distance. We test the null hypothesis of no
change in parental control over time in two different ways. First, we test the difference
in the control derivative between t 14 and t 1 (i.e., between 1998 and 1985):
5. We deleted a small number of observations for which the location could not be determined. Wewere able to match the location of over 99% of the lead banks and more than 98% of the nonlead banks.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
10/32
1492 : MONEY, CREDIT, AND BANKING
NONLEADRANKLEADRANKt14
NONLEADRANKLEADRANKt1 (1 3*lnDISTANCE)*(14 1)
(3) (1/21 1/23*lnDISTANCE)*(142 12) 0 ,
where Equation (3) is evaluated at the mean value of lnDISTANCE over the entire
time period. This procedure avoids confounding the measured effects of control for
a given distance with the effects of changes in distance over time. Second, we test
whether the control derivative is increasing in tat the mean of the data, using the
following second derivative:
2NONLEADRANKLEADRANKttmean (1 3*lnDISTANCE) 2*(1/21 1/23*lnDISTANCE)*t 0 . (4)
The distance derivative is an inverse measure of the agency costs of distance.
We expect the distance derivative to be negative in general for any value of t
(i.e., positive agency costs). However, as discussed above, this derivative may be
positive for very low levels of LEADRANK due to the potentially offsetting senior
management effect. We test the following null hypothesis:
NONLEADRANKlnDISTANCEtmean 2 3*LEADRANK (2 3*LEADRANK)*t
(1/22 1/23* LEADRANK)*t2
0 , (5)where Equation (5) is evaluated for various values of LEADRANK. Our hypothesis
that technological progress has reduced the agency costs of distance predicts that
NONLEADRANK/lnDISTANCE should be increasing over time for any givenvalue of lead bank efficiency.
We test the null hypothesis of no change in the agency cost of distance in two
different ways. First, we test the difference in the distance derivative between t
14 and t 1:
NONLEADRANKlnDISTANCEt14
NONLEADRANK
lnDISTANCEt1 (2 3*LEADRANK)*(14 1)
(1/22 1/23*LEADRANK)*(142 12) 0 , (6)where Equation (6) is evaluated at the mean value of LEADRANK over the entire
time period. Second, we test whether the distance derivative is decreasing in
absolute value withtat the means of the data, using the following second derivative:
2NONLEADRANKlnDISTANCEttmean
(2 3*LEADRANK) 2*(1/22 1/23* LEADRANK)*t 0 . (7)
To conserve degrees of freedom, only LEADRANK and lnDISTANCE are inter-acted with t in Equation (1). Hence, the coefficients on the remaining exogenous
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
11/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1493
variables may be interpreted as the average effect of each variable over time. The
market structure variables MSA and HERF are included to account for differencesin competition, demand, and other market conditions that may affect bank perfor-
mance. The variables lnBKASS and lnHCASS are included to account for the
influences of bank size and holding company size on affiliate bank efficiency.
The merger dummy variables BKMERGE and HCMERGE are included to help
account for short-term changes in bank performance associated with the consolida-
tion process. Themarket nonperformingloans-to-total loans ratio MNPL is intended to
capture the local economic conditions that are most relevant to bank performance.
The variable NUMAFFS is included to account for potential economies or disecono-
mies of scale in managing nonlead banks. NUMAFFS, lnBKASS, and lnHCASS
are included as both first- and second-order terms in Equation (1) to allow fornonlinear scale effects. The regulatory dummy variables UNITB, LIMITB, INTER-
STATE, and ACCESS attempt to capture the changes in state-by-state restrictions
on geographic expansion at various times during the sample period. The vector
STATE DUMMIES (counting the District of Columbia as a state, and excluding
California as the base case) helps account for additional differences in conditions
across states. Definitions and summary statistics for all of the variables (except
STATE DUMMIES) are included in Table 1.
Estimation of Equation (1) raises some econometric issues. First, the dependent
variable NONLEADRANK and the independent variable LEADRANK are gener-
ated using OLS techniques (described in detail below). The inclusion of a generated
regressor among the explanatory variables can in some circumstances affect the
reported standard errors; however, Pagan (1984) demonstrates that the coefficient
standard errors are consistent when the generated regressor is a residual from a
least squares regression, which is the case here with LEADRANK. The additional
noise in the generated dependent variable NONLEADRANK increases the regression
standard error, but does not bias the estimated coefficients.
Second, merger and acquisition activity increased during our sample period, which
may affect our measured control and distance derivatives for reasons unrelated to
technological progress. For example, control may have improved over time if nonleadbanks that were poorly controlled early in the sample period were merged with better-
controlled affiliates or with the lead bank later in the sample period. Similarly, the
agency costs of distance may have decreased over time because the banks that
were the hardest to control at a given distance were merged out of existence. We
address this issue by estimating Equation (1) for both the full data set and a
survivor data set that includes only nonlead banks that survived through 1998
(regardless of when they entered the data set and regardless of whether they were
still nonlead banks in MBHCs in 1998).
Third, a maintained assumption is that the geographic distribution of nonlead
affiliate banks is exogenous and is not influenced by the absolute or relative levelsof lead and nonlead bank efficiency. We recognize that this is an abstraction: at the
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
12/32TABL
E1
DefinitionsandSummaryStatisticsforVariablesinEquation(1)
Mainsample
Smallbanksample
FullDataSet(N
13411)
SurvivorDataSet(N
4363)
FullD
ataSet(N
17499)
SurvivorDataSet(N
7948)
Variabledefinition
Mean
Std.Dev.
Mean
Std.Dev.
Mea
n
Std.Dev.
Mean
S
td.Dev.
NONLEADRANK
Profitefficiency
0.607
0.296
0.599
0.288
0.61
3
0.278
0.595
0.270
rankofnonleadbanksinMBHCs.
LEAD
RANK
Profitefficiencyrank
0.439
0.301
0.416
0.303
0.52
06
0.283
0.542
0.274
ofleadbank(largestbankintheMBHC
).
DISTANCE
Distanceinmilesbetween
238.92
377.36
241.65
398.21
99.55
181.37
84.03
16
0.34
non
leadbankandleadbank.
lnDIS
TANCE
Naturallogof(DISTANC
E1).
4.576
1.505
4.601
1.433
3.82
2
1.305
3.757
1.168
tT
imevariableinyearsstartingin1985.
6.940
3.859
9.360
3.798
6.97
1
3.866
8.780
3.811
BKASS
Grosstotalassetsofnonlead
866,264
3,109,132
1,078,035
4,609,004
46,534
24,603
43,656
24,219
affiliatebank(thousandsof1998dollars).
lnBKASS
NaturallogofBKASS.
12.703
1.104
12.750
1.190
10.57
8
0.627
10.503
0.643
HCASS
Sumofgrosstotalassetsfor
22,587,297
36,069,278
21,903,028
45,581,663
3,688,328
10,792,546
1,630,531
8,905,602
all
banksinMBHC(thousandsof
199
8dollars).
lnHCASS
NaturallogofHCASS.
15.975
1.516
15.638
1.628
13.28
4
1.824
12.592
1.476
MSA
1ifnonleadbankisinaMetropo
litan
0.732
0.443
0.692
0.462
0.38
0
0.485
0.311
0.463
StatisticalArea(MSA).
HERF
AverageHerfindahlindex,weigh
ted
0.184
0.116
0.189
0.108
0.24
4
0.171
0.253
0.167
by
shareofnonleaddepositsineach
MS
Aornon-MSAcountyinwhichit
ope
rates.
(Con
tinued)
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
13/32TABL
E1
Cont
inued
Mainsample
Smallbanksample
FullDataSet(N
13411)
SurvivorDataSet(N
4363)
FullD
ataSet(N
17499)
SurvivorDataSet(N
7948)
Variabledefinition
Mean
Std.Dev.
Mean
Std.Dev.
Mea
n
Std.Dev.
Mean
S
td.Dev.
MNPL
Averagemarketnonperforming
0.022
0.012
0.019
0.011
0.02
1
0.014
0.018
0.012
loanratio,weightedbyshareofnonlead
deposits
ineachMSAornon-MSAcountyinwh
ich
ito
perates.
BKMERGE
1ifaffiliateinmergerinprior
0.196
0.397
0.212
0.409
0.02
3
0.149
0.022
0.147
threeyears.
HCMERGE
1ifchangedhighholdingcompany
0.296
0.456
0.238
0.426
0.34
4
0.475
0.318
0.466
affiliationinthepriorthreeyears.
NUM
AFFS
Numberofaffiliatesunderthesame
20.620
19.425
17.148
16.451
10.16
2
14.652
5.406
8.801
highholder.
UNIT
B
1ifunitbankingstate.
0.101
0.302
0.029
0.169
0.15
1
0.358
0.069
0.254
LIMITB
1iflimitedbranchingstate.
0.437
0.496
0.389
0.488
0.54
5
0.498
0.536
0.499
INTERSTATE
1ifinterstatebankholding
0.812
0.391
0.922
0.268
0.74
2
0.438
0.863
0.344
com
panyexpansionisallowed.
ACCESS
PercentU.S.bankingassetsin
states
0.342
0.308
0.453
0.325
0.30
0
0.312
0.388
0.333
withaccesstononleadbanksstate.
Notes:
ThedataareannualobservationsfornonleadbanksinMBHCsfortheyears1985-98.Theleadb
ankisdefinedasthelargestbankingaffiliateintheMBHC,andanonleadbankisanycommercialbanking
affiliate
inanMBHCotherthantheleadbank.TheMainSampleincludesannualobservationsofnonlea
dbankswithgrosstotalassets(GTA)inexcessof$100million,whereastheSmallBankSample
includes
observa
tionsinwhichGTAislessthanorequalto$100
million(real1998dollars).TheFullDataSetsin
cludeallobservations,whiletheSurvivorDataSetsincludeonlynonleadbanksthatsurvivedthrough1998.
Dataar
efromtheU.S.commercialbankCallReports,theFDICSummaryofDeposits,andauthorscalculations.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
14/32
1496 : MONEY, CREDIT, AND BANKING
margin, well-run organizations maybe more likelythan average to acquire andmanage
affiliate banks located far from headquarters. To the extent that this effect ispresent in our data, then our estimated distance derivatives (Equation 5) will under-
state the agency costs of distance because longer distances will tend to be found in
organizations that are better able to manage distance-related costs. Hence, finding
a negative sign for the distance derivative is a strong result, as it indicates that
the agency costs of distance more than offset the managerial efficiency effect. Simi-
larly, the distribution of nonlead affiliate banks within an MBHC may be endogenous
to past acquisition patternsfor example, if inefficient MBHCs systematically
acquired more efficient affiliates in order to import managerial talent and best
practices. To the extent that this phenomenon is present in our data, then MBHCs
with inefficient lead banks would contain a mix of relatively inefficient nonlead
banks (i.e., the original affiliates) and relatively efficient nonlead banks (i.e., the
newly acquired affiliates), causing a dispersion in NONLEADRANK away from
LEADRANK that would bias our estimated control derivative (Equation 2) toward
zero. Hence, finding a positive sign for the control derivative is a strong result.
Because we cannot directly observe the implementation of new technologies at
banks and bank holding companies, we must rely on the time variable tas a proxy
for technological progress. As stated above, this is common practice in the empirical
microeconomic and macroeconomic literatures. We acknowledge that this is an
imperfect measure because it pools technological advance with other intertemporal
changes in conditions faced by banks. For this reason our right-hand-side specifica-
tion of Equation (1) includes a number of variables that control for other important
environmental changes over time, including state and interstate regulatory restric-
tions (UNITB, LIMITB, INTERSTATE, ACCESS), local competitive conditions
(HERF), and local economic conditions (MNPL). In addition, we also check the
robustness of our results by estimating Equation (1) and the derivatives (Equations
27) for sub-periods of our 1985-98 data set.
2.2 Summary Statistics
The summary statistics displayed in Table 1 allow us to compare the banks in
the main and small bank samples, as well as the banks in the full and survivor data
sets. The survivor data sets have fewer than half as many observations as the fulldata sets because of the substantial consolidation of the banking industry. These
consolidation effects may have an especially large impact on nonlead banks in
MBHCs, as these nonlead banks were especially prone to disappear over time as
geographic deregulation allowed MBHCs to combine affiliates at greater distances.
Consistent with the consolidation, the mean value of t is around 9 for the survivor
data sets versus around 7 for the full data sets, as the survivor data sets exclude
more of the early observations. Note that average bank efficiency is about the same in
the survivor data sets and the full data setsthis reflects the fact that we are
measuring efficiency using ranks rather than levels.
Most of the comparisons between the main and small bank samples yield theexpected findings. Compared to the nonlead banks in the main sample, the small
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
15/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1497
nonlead banks tend to be (1) located much closer to their lead bank, (2) in more
highly concentrated rural markets, and (3) in much smaller MBHCs with fewer totalaffiliates. The average efficiency across the two samples is very similar because the
efficiency ranks are measured only against other banks in the same size grouping
(above or below $100 million in GTA). Finally, the nonlead affiliates have higher
cost and profit efficiency ranks on average than the lead banks in both the main
sample and the small bank sample. This could reflect a number of different factors.
Back-office operations and headquarters functions such as payroll processing, mar-
keting, and legal support may not be fully accounted for in intra-organizational
accounting (e.g., incorrect or incomplete transfer pricing), resulting in financial state-
ment subsidies flowing from lead banks to nonlead affiliates. In addition, it is likely
that our efficiency model does not fully specify the wide scope of financial products
and services produced at large lead banks (e.g., private banking, proprietary mutual
funds, merger finance, global services). Finally, the high mean efficiency levels of
nonlead banks relative to lead banks could simply be an arithmetic result of the
fact that efficient banking organizations (i.e., high efficiency ranks at both the lead
and nonlead banks) tend to have relatively large numbers of affiliates.6 We include a
number of right-hand-side variables in our regression Equation (1) to try to mitigate
these potential effects.
Figure 1 shows how the average distance between nonlead affiliates in MBHCs
and their lead banks has evolved over time. For the main sample, the average distance
increased by about 60%80% from 1985 to 1998, consistent with the geographic
Fig. 1. Distance in miles between lead banks and nonlead banks within MBHCs by year, 1985-98.
6. Berger and DeYoung (2001) found that the affiliates of banking organizations with 10 or moreaffiliates tend to exhibit above-average efficiency.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
16/32
1498 : MONEY, CREDIT, AND BANKING
deregulation of U.S. banking discussed above. In contrast, for the small bank sample,
the increase in distance was only on the order of about 15%30%.
7
As discussedabove, the geographic expansion via acquisition of small bank affiliates may have
proceeded more slowly because small banks may be more difficult to control and
subject to more agency costs of distance because of their specialization in local
relationships based on soft information. For both samples, the average distance in
the survivor data sets tends to be less than in the full data sets, a finding that is
broadly consistent with our conjectures about the deleterious effects of distance on
organizational management and control.
3. MEASURING BANK EFFICIENCY RANKS
We base our analysis of lead and nonlead bank performance on the profit efficiency
concept, rather than the often-used cost efficiency concept because profits are concep-
tually superior to costs for evaluating overall firm performance. The economic goal
of profit maximization requires that the same amount of managerial attention be
paid to raising a marginal dollar of revenue as to reducing a marginal dollar of
costs. As well, if there are substantial unmeasured differences in the quality of services
across banks and over timewhich is almost surely the casea bank that pro-
duces higher quality services should receive higher revenues that compensate it for
at least some of its extra costs of producing that higher quality. Such a bank may
be measured as less cost efficient because of the extra associated costs but willappropriately be measured as more profit efficient if its customers are willing to
pay more than the extra costs in exchange for the higher quality. As discussed below,
we also perform all the tests using cost efficiency as a robustness check, with
similar findings.
We start with a profit function at time tof the form:
ln(it t) ft(wit,yit,zit,vit) lnuit ln it, (8)
where is bank profit andt is a constant for year t that makesit t positivefor all banks (so that the dependent variable expressed as a natural log is defined);
iindexes banks ( 1,N);fdenotes some functional form; wis the vector of variable
input prices faced by the bank; y is the vector of its variable output quantities;
z indicates the quantities of any fixed netputs (inputs or outputs); v is a set of
variables measuring the economic environment in the banks local market(s);
lnu is a factor that represents a banks efficiency; and ln is random error that
incorporates both measurement error and luck. We estimate the profit function
separately each year using OLS, treating(ln u ln ) as a composite error term.The profit efficiency of bank i at time t is based on a comparison of its actual
profits (adjusted for random error) to the maximum potential (best-practice) profits
7. The overall increase from 123.35 to 188.91 miles between 1985 and 1998 cited in the introductionis based on the weighted averages from the full data sets for the main and small bank data sets.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
17/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1499
it could attain given its output bundle and other exogenous variables (w, y, z, v).
Assuming temporarily that we have an estimate of the efficiency factor ln u
it, theprofit efficiency of bank i at time twould be given by:
PROFEFFitit
maxt
{exp[f(wit,yit,zit,vit)] exp[lnuit]}
{exp[f(wit,yti,zit,vit)] exp[lnumaxt]} (9)
where umaxt is the maximum uitacross all banks in the sample. PROFEFFcan be
thought of as the proportion of a banks maximum profits that it actually earns. 8
We take two additional steps to arrive at the estimates of the efficiency ranks,
NONLEADRANK and LEADRANK, that are used in our analysis of the changesover
time in the control and distance derivatives. First, we use the residuals from OLSestimation of Equation (8) as estimates of the profit efficiency factor lnuit. Second,
we create a rank ordering of the banks in each year and peer group (main sample
or small bank sample) based on those residuals. Profit efficiency ranks measure how
well a bank is predicted to perform relative to other banks in their main sample or
small bank sample peer group for producing the same output bundle under the same
exogenous conditions in the same year. These ranks are calculated on a uniform
scale over[0, 1], using the formula(orderit 1)(nt 1), whereorderitis the placein ascending order of the ith bank in the tth year in terms of its profit efficiency and
ntis the number of banks in the relevant sample in year t. Thus, bankis efficiency
rank in year t gives the proportion of the banks in its peer group in year twithlower efficiency (e.g., a bank in year twith efficiency better than 80% of its peer
group has a rank of 0.80). The worst-practice bank in the relevant sample (lowest
PROFEFF) has a rank of 0, and the best-practice bank (highest PROFEFF) has a
rank of 1.
We use efficiency ranks, rather than efficiency levels, because ranks are more
comparable over time. The main purposes of this research are to test for intertemporal
changes in parental control and the agency costs of distance on nonlead bank
performance, which we proxy by changes in NONLEADRANK/LEADRANKandNONLEADRANK/lnDISTANCE over time. Efficiency levels may change
considerably over time with the interest rate cycle, real estate cycle, macroeconomiccycle, and changes in bank regulation. Thus, comparing efficiency levels at different
times might confound changes in the distribution of efficiency with changes in
parental control or with changes in the agency costs of distance. Using efficiency
ranksthat is, ranking banks within a uniform [0,1] distribution in each timeperiodgenerally neutralizes this problem by focusing on the relative order of
the banks in the efficiency distribution at a given time.
8. This is often called alternative profit efficiency because we specify output quantitiesy, ratherthan output prices p, in the profit function. We use this concept primarily because output prices aredifficult to measure accurately for commercial banks and because output quantities are relatively fixed
in the short run and cannot respond quickly to changing prices as is assumed in the standard profitefficiency. Prior research generally found similar results for estimates of standard and alternativeprofit efficiency (e.g., Berger and Mester 1997).
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
18/32
1500 : MONEY, CREDIT, AND BANKING
Our method is a special case of the distribution-free efficiency measurement
approach that uses a single residual instead of averaging residuals over a numberof years. The single-residual method here is less accurate than the usual distribution-
free approach, which averages out more random error by using more years of data.
Our approach should also yield ranks that are very similar to those yielded by the
stochastic frontier approach, an approach that is commonly used to estimate effi-
ciency in a single cross-section.9
We estimate the profit function using the Fourier-flexible functional form, which
has been shown to fit the data for U.S. banks better than the more commonly
specified translog form (e.g., Berger and DeYoung 1997). We specify three variable
input prices (local market prices of purchased funds, core deposits, and labor); four
variable outputsy(consumer loans, business loans, real estate loans, and securities);
three fixed netputsz(off-balance-sheet activity, physical capital, and financial equity
capital); and an environmental variable STNPL(ratio of total nonperforming loans-
to-total loans in the banks state). By using local market input prices, rather than the
prices paid by each bank, the efficiency ranks will reflect how well individual banks
price their inputs.
We calculate efficiency ranks using virtually all U.S. commercial banks for every
year, 1985-98, although our regressions only include the efficiency ranks of banks
that are lead banks or nonlead affiliates in MBHCs. The profit functions are
always estimated separately for the main sample and the small bank sample because of
the differences in products and markets discussed above.
4. EMPIRICAL RESULTS
Table 2 displays the estimated parameters of regressions using 1985-98 U.S.
banking data. Equation (1) is estimated four times: for the full and survivor data
sets from the main sample (GTA $100 million in real 1998 dollars) and for the
full and survivor data sets from the small bank sample (GTA $100 million inreal 1998 dollars). The key exogenous variablesLEADRANK, lnDISTANCE,
and tare all interacted, so no one coefficient on these variables can be easily
interpreted by itself. We focus on the control and distance derivatives derived from
the formulas shown in Equations (2)(7). To avoid confounding our estimates ofparental control and agency costs of distance with the effects of changes in the
exogenous variables over time, we evaluate the control and distance derivatives at
the overall 1985-98 sample means for the exogenous variables (with the exception
in some cases of the time variable t).
4.1 Control and Distance Derivatives
The top panel in Table 3 shows the estimated control derivatives at the mean of
tfrom Equation (2),NONLEADRANKLEADRANKtmean.When evaluated at
9. This is because the stochastic frontier approach also forms the expectation of the inefficiencyterm ln uit based on the observed residual, and this ranking is in the same order as the residuals forany distributions imposed on ln u and ln .
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
19/32TABL
E2
OLS
EstimatesfromtheNONLEADRA
NKRegressionsfortheMainSa
mpleandSmallBankSample
MainSample
SmallBankSample
FullDataSet(N
13411)
Survivo
rDataSet(N
4363)
FullDataSet
(N
17499)
SurvivorDataSet(N
7948)
Coefficient
t-stat.
Coeffic
ient
t-stat.
Coefficient
t-stat.
Coefficient
t-stat.
Interc
ept
1.262***
3.87
1.893
***
3.76
1.585***
3.73
1.757***
2.89
LEAD
RANK
0.048
0.60
0.185
0.82
0.257***
4.00
0.117
0.85
lnDIS
TANCE
0.036***
4.76
0.052
**
2.36
0.036***
4.54
0.058***
3.03
LEAD
RANK*lnDISTANCE
0.029*
1.72
0.019
0.40
0.029*
1.86
0.031
0.92
t
0.102***
7.97
0.113
***
4.19
0.076***
6.25
0.075***
3.37
t*LEA
DRANK
0.127***
4.93
0.102
*
1.85
0.025
1.15
0.062
1.63
t*lnDISTANCE
0.012***
4.68
0.013
**
2.22
0.012***
3.94
0.015***
2.74
t*LEA
DRANK*lnDISTANCE
0.020***
3.70
0.013
1.11
0.003
0.49
0.017*
1.76
1/2t2
0.012***
6.72
0.014
***
4.23
0.006***
3.44
0.006**
2.02
1/2t2
*LEADRANK
0.017***
4.78
0.016
**
2.48
0.001
0.20
0.003
0.73
1/2t2
*lnDISTANCE
0.002***
4.14
0.001
**
2.19
0.001**
2.13
0.001
1.58
1/2t2
*LEADRANK*lnDISTANCE
0.003***
3.78
0.002
*
1.75
0.000
0.03
0.001
1.00
lnBKASS
0.092**
2.17
0.108
1.58
0.286***
3.44
0.269**
2.28
1/2ln
BKASS2
0.005
1.63
0.008
*
1.68
0.024***
2.99
0.023**
2.02
lnHCASS
0.015
0.47
0.059
1.17
0.106***
5.46
0.172***
5.75
1/2ln
HCASS2
0.001
0.66
0.002
0.74
0.010***
7.05
0.016***
7.40
MSA
0.011
1.58
0.010
0.92
0.011**
2.13
0.021***
3.04
HERF
0.176***
6.73
0.269
***
5.63
0.065***
4.42
0.037*
1.83
MNPL
1.881***
8.46
0.532
1.26
0.825***
5.04
0.190
0.70
BKMERGE
0.028***
4.58
0.031
***
3.08
0.007
0.54
0.002
0.09
HCMERGE
0.031***
5.92
0.066
***
7.00
0.008**
1.98
0.025***
4.26
NUM
AFFS
0.007***
14.46
0.011
***
11.65
0.010***
16.43
0.022***
20.05
1/2N
UMAFFS2
0.000***
11.17
0.000
***
8.78
0.000***
10.53
0.001***
15.27
UNIT
B
0.156***
11.50
0.175
***
6.12
0.084***
8.87
0.069***
4.64
LIMITB
0.006
0.81
0.028
*
1.90
0.002
0.23
0.014
1.44
INTERSTATE
0.087***
8.53
0.094
***
4.47
0.046***
6.01
0.019
1.58
ACCESS
0.041***
2.86
0.002
0.10
0.001
0.07
0.020
1.33
Adjus
tedR-Squared
0.243
0.284
0.173
0.234
Note:Datasetsareunbalancedpanels,1985-98.Superscripts***,**,and*indicatesignificantdifferenc
efromzeroat1%,5%,and10%
levels.Coefficie
ntsforstatedummyvariablesnotreported.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
20/32TABL
E3
MeasuredValuesoftheControland
DistanceDerivatives
Mean
25thpercentile
50
thpercentile
75thpercentile
A.Co
ntrolderivativesevaluatedatvalues
fromthesampledistributionoflnDISTANCE
Main
sample,fulldataset
0.235***(19.30)
0.271***(19.42)
0.23
1***(18.98)
0.195***(14.17)
Main
sample,survivordataset
0.282***(13.59)
0.313***(13.52)
0.28
1***(13.52)
0.249***(10.57)
Smallbanksample,fulldataset
0.237***(22.20)
0.272***(22.49)
0.23
6***(22.16)
0.197***(15.75)
Smallbanksample,survivordataset
0.264***(17.46)
0.312***(18.31)
0.26
5***(17.55)
0.210***(11.81)
B.Distancederivativesevaluatedatvalues
fromthesampledistributionofLEA
DRANK
Main
sample,fulldataset
0.006**(2.16)
0.007**(2.03)
0.00
5*(1.91)
0.018***(5.09)
Main
sample,survivordataset
0.013***(3.01)
0.002(0.33)
0.01
3***(2.92)
0.024***(4.32)
Smallbanksample,fulldataset
0.002(0.63)
0.010***(3.12)
0.00
3(1.18)
0.013***(4.20)
Smallbanksample,survivordataset
0.006*(1.67)
0.010**(2.08)
0.00
8**(2.19)
0.022***(4.94)
Notes:
PanelAdisplaysmeasuredcontrolderivativesat
themeanoft(NONLEADRANKLEADRANK
tmean
)
fromEquation(2),evaluatedandtestedatthemeanandthe25th,50th,and75thpercentiles
fromthe
sample
distributionoflnDISTANCE.Forthemainsample,fulldataset,thesevaluesare4.58,3.75,4.67,
and5.49,respectively.Thepointsofevaluationaresimilarfortheothersamples.PanelBdisplays
measured
distancederivativesatthemeanoft(NONLEADRANKlnDISTANCE
tmean
)fromEquation(5),evaluatedandtestedatthemeanandthe25th,50th,and75thpercentilesfromthesampledistributionofLEA
DRANK.
Forthe
mainsample,fulldataset,thesevaluesare0.44,0.15,0.42,and0.71,respectively.Thepointsof
evaluationaresimilarfortheothersamples.The
valuesinparenthesesaret-statistics.Superscripts***,**,
and*indicatesignificantdifferencefromzeroat1%,5%
,and10%
levels.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
21/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1503
the mean of lnDISTANCE (first column), all four of the control derivatives are
positive and between 0 (no control) and 1 (very good control), as expected. Allfour are also all statistically significantly different from 0 at the 1% level. The
positive, statistically significant control derivatives indicate that after accounting
for other factorssuch as distance, local market economic conditions, bank and
MBHC size, and state regulationslead bank efficiency and nonlead bank perfor-
mance are significantly positively related. This suggests that senior managers at the
lead banks are able to transfer their policies, practices, and procedures to management
at their nonlead affiliate banks to at least some degree.
The magnitudes of these control derivatives are similar for the main and small bank
samples, suggesting that, on average, affiliate size is not an important determinant of
the ability of headquarters managers to control the efficiency of affiliate banks. Thecontrol derivatives are materially larger for the survivor banks relative to the full
data set, consistent with our expectations that MBHCs in which control was weak
should be more likely to exit the industry via failure or takeover. In all cases,
the derivatives appear to be economically significant. For example, the control
derivative of 0.235 in the first cell of panel A indicates that a 10 percentage point
increase in lead bank efficiency rank would yield a predicted increase of 2.35
percentage points in the efficiency ranks of each of its nonlead bank affiliates.
Evaluating the control derivative at the 25th, 50th, and 75th percentiles of the
sample distribution of lnDISTANCE (last three columns) shows how the magnitude of
lead bank control changes with distance. The results suggest that control diminisheswith distance, as expected. For example, as shown in the top row of panel A for
the main sample, full data set, the control derivative is 4.0 percentage points
lower for nonlead affiliates at the median distance or 50th percentile (0.231) than
for affiliates closer to headquarters at the at the 25 th percentile of distance (0.271).
Similarly, the control derivative falls by about another 3.6 percentage points for
affiliates that are even more remote at the 75th distance percentile (0.195).
The bottom panel in Table 3 shows the estimated distance derivatives at the mean
oftfrom Equation (5),NONLEADRANKlnDISTANCEtmean. When evaluatedat the mean of LEADRANK (first column), these derivatives are negative and statisti-
cally significant in three of the four cases, consistent with our hypothesis thataligning the incentives of local managers with those of the organization generally
becomes more difficult with distance, resulting in lower affiliate bank efficiency. Al-
though these estimated distance derivatives are small in absolute magnitude, the
effect of a changein distance can be substantial for a banking organization that expands
over very large distances within the United States. For example, the distance
derivative of0.013 (main sample, survivor data set) indicates that a doubling of
the distance between a nonlead bank and its headquarters from about 240 miles to
about 480 miles would reduce the efficiency rank of the nonlead bank by almost a
full percentage point (0.013*ln 2). Much larger potential reductions of efficiency
rank would be predicted for cross-country acquisitions, although we caution againstextrapolating too far from the dense part of the data sample.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
22/32
1504 : MONEY, CREDIT, AND BANKING
The absolute magnitude of the distance derivative is smaller in the small bank
sample, suggesting perhaps that agency problems in monitoring banks that relymore on soft information are less sensitive to the distance from headquarters. In
addition, we find that the negative effects of distance tend to be greater in surviving
organizations and in organizations with more efficient lead banks. Notably, the
distance derivative is positive and statistically significant in three of four cases when
evaluated at the 25th percentile for LEADRANK, consistent with the possibility
raised above that for very inefficient senior management, the senior management
effect may overwhelm the general agency cost effect. That is, it may on net be
more efficient to be further away from such management to avoid the transfer of
their inefficient practices.
4.2 Intertemporal Changes in the Control and Distance Derivatives
The main focus of our research is on the changes over time in the control and
distance derivatives. We hypothesize that these derivatives should increase with t
as improvements in financial and nonfinancial technologies improved the control
of banking organizations and reduced the agency costs of distances. We calculate
each of these derivatives for each of the regressions in Table 2 and for all values
of time t. The year-to-year estimates are plotted in Figures 25, and the data
underlying these figures are displayed in panel A of Tables 4 and 5. Panel A of
these tables displays tests (3) and (6), NONLEADRANKLEADRANKt14
NONLEADRANK
LEADRANKt1 and NONLEADRANK
lnDISTANCEt14
NONLEADRANKlnDISTANCEt1, which measure the change in the control
Fig. 2. Control derivatives (Equation 2) estimated for the full data set, evaluated at the means of the data and at all
values oft, 1985-98.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
23/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1505
Fig. 3. Control derivatives (Equation 2) estimated for the survivor data set, evaluated at the means of the data and
at all values oft, 1985-98.
and distance derivatives from the beginning to the end of the 1985-98 sample
period. Panel C of these tables displays the cross-derivative tests (4) and (7),
2NONLEADRANKLEADRANKttmean and 2NONLEADRANKlnDISTANCEttmean, which measure the change in the control and distance
Fig. 4. Distance derivatives (Equation 5) estimated for the full data set, evaluated at the means of the data and at
all values oft, 1985-98.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
24/32
1506 : MONEY, CREDIT, AND BANKING
Fig. 5. Distance derivatives (Equation 5) estimated for the survivor data set, evaluated at the means of the data and
at all values oft, 1985-98.
derivatives with respect to time at the means of the data. We view tests (3) and (6)
as more important than the cross-derivative tests (4) and (7) because the (3) and(6) tests measure the effects over the entire time period, not just at the mean of the
period.10 Panel D repeats tests (3) and (6) from panel B, evaluating them at the 25th,
25th, 50th, and 75th percentiles for lnDISTANCE for the change in control derivatives
(Table 4) and the same percentiles for LEADRANK for the change in distance
derivatives (Table 5).
All four of the control derivatives mapped out in Figures 2 and 3 are generally
increasing over time and are larger at the end of the sample period than at the
beginning. The U- or inverted U-shapes of these time paths are dictated by our
quadratic regression specification. The estimated time paths decrease somewhat at
the end of the sample period for the main sample, which may also reflect the large
number of M&As made during this time period (see discussion below). We focus
primarily on the change over the entire time periodt 1 to 1 14shown in panel
B of Table 4which indicate that all four of the control derivatives increased during
the sample period, three of which were statistically significant. Moreover, these
improvements in parental control over time were economically significantthe
smallest increase was about 35%, from 0.233 to 0.314 (small bank sample, survivor
data set) while the largest increase was about 89%, from 0.160 to 0.303 (small bank
10. We recognize that the derivatives (Equations 3 and 6) are based on fitted regression values farfrom the means of the data for t(i.e., at t 1 and t 14). This reduces the precision of the estimatedderivatives, which may make it relatively difficult to reject the null hypotheses in these tests.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
25/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1507
TABLE 4
The Measured Effects of Intertemporal Changes in the Control Derivative for the Full
Data Set, the Survivor Data Set, the Main Sample, and the Small Bank Sample, for1985-98 (t 1,...,14)
Bank Sample: Data set Main Full Main Survivor Small Bank Full Small Bank Survivor
A. Control derivative evaluated at the means of the data and annual values of t
1985 0.118 0.139 0.160 0.2331986 0.148 0.175 0.174 0.2341987 0.173 0.205 0.188 0.2361988 0.195 0.230 0.201 0.2391989 0.212 0.251 0.213 0.2421990 0.226 0.266 0.225 0.2471991 0.235 0.277 0.237 0.2521992 0.240 0.283 0.248 0.2581993 0.241 0.283 0.258 0.2651994 0.238 0.279 0.268 0.2731995 0.231 0.269 0.278 0.2821996 0.220 0.255 0.287 0.2921997 0.205 0.236 0.295 0.3021998 0.186 0.211 0.303 0.314
B. Change in the control derivative from 1985 to 1998
0.068** (2.49) 0.072 (1.43) 0.143*** (6.02) 0.080** (2.29)
C. Change in the control derivative with respect to t at the means of the data
0.008*** (3.46) 0.004 (0.95) 0.011*** (6.19) 0.007*** (2.61)
D. Change in the control derivative from 1985 to 1998, evaluated at the 25 th, 50th, and 75th percentilesof the sample distribution of lnDISTANCE
25 th percentile 0.063** (2.05) 0.027 (0.44) 0.167*** (5.87) 0.153*** (3.73)50 th percentile 0.068** (2.51) 0.074 (1.47) 0.143*** (6.02) 0.082** (2.34)75 th percentile 0.073** (2.33) 0.121** (2.05) 0.116*** (4.34) 0.001 (0.00)
Notes: Panel A shows annual estimated values of Equation (2), the control derivative, evaluated at the overall means of the data for allvariables other than t. Panel B shows the estimated values for Equation (3), the change in the control derivative between t 1 and t14, along with associated t-statistics. Panel C shows the estimated for Equation (4), the change in the control derivative with respect to tat the means of the data, along with associated t-statistics. Panel D shows the change in the control derivative with respect to t at themeans of the data, evaluated at the 25 th, 50th, and 75th percentiles of the sample distribution of lnDISTANCE. Superscripts ***, **, and* indicate significant difference from zero at 1%, 5%, and 10% levels.
sample, full data set). Panel C of Table 4 shows similar results for the intertemporal
change in the control derivatives evaluated at the mean value of time t. Overall, thesedata are strongly consistent with our hypothesis that technological progress has
allowed banking organizations to exercise substantially more control over nonlead
affiliates over time. Notably, the measured increases in control in panels B and C are
generally greater for small bank affiliates than large bank affiliates. Although an
investigation of this difference is beyond the scope of this paper, this finding is
consistent with the possibility raised above that technological progress may have
allowed MBHCs to apply some of the hard-information techniques to small bank
affiliates for the first time, yielding substantial increases in control over these
affiliates.
The improvement in parental control over time was sensitive to the distancebetween the lead and nonlead banks, as shown by Equation (3) results in panel D
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
26/32
1508 : MONEY, CREDIT, AND BANKING
TABLE 5
The Measured Effects of Intertemporal Changes in the Distance Derivative for the Full
Data Set, the Survivor Data Set, the Main Sample, and the Small Bank Sample, for1985-98 (t 1,...,14)
Bank Sample: Data set Main Full Main Survivor Small Bank Full Small Bank Survivor
A. Distance derivative evaluated at the means of the data and annual values of t
1985 0.020 0.052 0.042 0.0351986 0.017 0.046 0.033 0.0301987 0.014 0.040 0.025 0.0251988 0.011 0.034 0.018 0.0211989 0.009 0.029 0.011 0.0171990 0.007 0.025 0.006 0.0131991 0.006 0.021 0.001 0.0101992 0.004 0.017 0.002 0.0081993 0.003 0.014 0.005 0.0061994 0.003 0.012 0.007 0.0041995 0.002 0.010 0.008 0.0031996 0.002 0.008 0.008 0.0021997 0.002 0.007 0.007 0.0011998 0.002 0.006 0.006 0.001
B. Change in the distance derivative from 1985 to 1998
0.018*** (3.18) 0.046*** (3.97) 0.047*** (8.82) 0.034*** (3.99)
C. Change in the distance derivative with respect to t at the means of the data
0.002*** (3.55) 0.003*** (3.11) 0.004*** (10.01) 0.002*** (3.14)
D. Change in the distance derivative from 1985 to 1998, evaluated at the 25 th, 50th, and 75th percentilesof the sample distribution of LEADRANK
25 th percentile 0.016** (2.27) 0.029** (2.04) 0.055*** (8.28) 0.057*** (5.90)50 th percentile 0.017*** (3.18) 0.045*** (3.94) 0.047*** (8.57) 0.031*** (3.60)75 th percentile 0.019** (2.49) 0.061*** (3.63) 0.040*** (5.62) 0.010 (0.88)
Notes: Panel A shows annual estimated values of Equation (5), the distance derivative, evaluated at the overall means of the data for allvariables other than t. Panel B shows the estimated values for Equation (3), the change in the distance derivative between t 1 and t 14, along with associated t-statistics. Panel C shows the estimated for Equation (4), the change in the distance derivative with respect tot at the means of the data, along with associated t-statistics. Panel D shows the change in the distance derivative with respect to t at themeans of the data, evaluated at the 25th, 50th, and 75th percentiles of the sample distribution of LEADRANK. Superscripts ***, **, and* indicate significant difference from zero at 1%, 5%, and 10% levels.
of Table 4. In the main sample, control improved the most for the relatively distantaffiliates, consistent with the possibility raised above that technological progress
allowed a greater increase in control for longer distances for these banks because
of the electronic transmission of hard information with little or no effect of distance. In
the small bank sample, however, we find that the control derivative improved more
dramatically over time at the relatively nearby affiliates. This is consistent with
the possibility that for small affiliates, control improvements from the export of
hard-information technologies to small affiliates may have occurred more frequently
for affiliates closer to MBHC headquarters.
Turning to the effects of technological progress on the agency costs of distance,
all four of the distance derivatives plotted in Figures 4 and 5 increased overtime. Panels B and C of Table 5 indicate that the intertemporal changes in the distance
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
27/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1509
derivatives were statistically significant. These changes were also economically
significant. The increase in the distance derivative between 1985 and 1998 tends tocenter around 0.04, suggesting that a doubling of distance from the lead bank reduced
the efficiency rank of a nonlead bank by about 2.8 percentage points less at the
end of the sample period than at the beginning of the sample period (0.04*ln2).
These data are consistent with our hypothesis that technological progress has reduced
the agency costs of distance, and the effects do not differ greatly for large and small
nonlead banks. As noted above, this finding may also reflect in part other causes,
such as the increased integration of state and regional economies.
The degree to which the distance derivative increased over time was somewhat
sensitive to the efficiency of the lead bank in the organization, as shown by Equation
(6) results in panel D of Table 5. For banks in the main sample, the data suggest
that efficient lead banks were able to reduce the agency costs of distance more for
large nonlead bank affiliates. This is consistent with the possibility that efficient lead
banks that likely specialize in the use of hard-information technologies also excelled at
reducing the agency costs of distance for large nonlead affiliates that also likely
specialize in these same technologies. In contrast, the data suggest that for the small
bank sample, distance-related inefficiencies declined more slowly over time for
organizations with efficient lead banks. This is consistent with the possibility that
efficient lead banks that specialize in hard-information technologies were not as
able to reduce agency costs associated with distance to small nonlead banks that
tend to use soft-information technologies.
The intertemporal increases in the control and distance derivatives shown in
Figures 25 occur mostly during the first portion of the sample period. Although
the shapes of these estimated time paths are constrained by our quadratic specification
of time, these shapes suggest that improvements in parental control and reductions
in the agency costs of distance may have been easier to achieve during the late
1980s than in the 1990s. A full investigation of this phenomenon is beyond
the scope of this study, but we suggest two reasonable hypotheses that are consistent
with the data. First, the mergers of the 1990s tended to be larger, more complex,
and involve more distant target banks (see Figure 1, especially for the main sample),
which may have posed more difficult managerial challenges than the mergers of
the 1980s. Second, as discussed above, banking was one of the first industries to takeadvantage of improvements in information processing and telecommunications,
consistent with the measured productivity gains made by banks in the late 1980s.
4.3 Robustness Tests
We performed a number of additional robustness checks that are not shown in
the tables and figures. First, we performed all of the above tests using cost efficiency
ranks in place of profit efficiency ranks, and obtained similar, but somewhat weaker
results. This suggests that technological advances as implemented at banking compa-
nies has helped improve parental control over both expenses and revenues, and
reduce the impact of distance-related agency costs on both expenses and revenues,although the improvements were generally more pronounced on the revenue side.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
28/32
1510 : MONEY, CREDIT, AND BANKING
Second, we estimated the regressions using subperiods of the data1985-1991
(the first half of the panel), 1992-98 (the second half of the panel), and 1985-96 (before full effect of Riegle-Neal Act). These regressions yielded qualitatively
similar results to those from the overall 1985-98 time period. Third, we estimated
the regressions using efficiency levels rather than efficiency ranks, and using linear
distance rather than the natural log of distance. These regressions always produced
theoretically correct signs for the control and distance derivatives, although the
behavior of these derivatives across time was somewhat less robust. Finally,
we replaced each panel regression with 14 annual cross-section regressions. 11 This
cross-sectional approach allows all of the estimated regression parameters to vary
freely from year to year and does not force the control and distance derivatives
to follow smooth parabolic paths over time, but this approach also reduces estimation
efficiency relative to our panel approach because of the fewer numbers of restrictions.
Although some of the resulting derivatives exhibited substantial noise from year to
year, the results were qualitatively similar to our panel regression results.
5. CONCLUSIONS
In this paper, we test if the data on banks in U.S. MBHCs over 1985-98 are
consistent with some hypotheses about the effects of technological progress on
the ability of the banking industry to expand geographically. The empirical results
are strongly consistent with the predictions of our hypothesis that technological
progress hasallowed banking organizations to exercise substantially more control over
nonlead affiliates over time. Specifically, our estimates suggest that the influence
of a lead banks efficiency rank on the efficiency rank of its nonlead bank affiliates
is substantial, and this ability to control its affiliates increased on the order of
50%100% over the sample period. Furthermore, our estimates suggest that senior
managers improved their ability to control both the costs and the revenues of
their nonlead affiliates over the sample period. The data are also consistent with
our hypothesis that technological progress has allowed banking organizations to
reduce the agency costs that arise when nonlead affiliate banks are located far awayfrom headquarters. These improvements appear to be more substantial on the revenue
side than on the cost side of banks income statements.
The issue of whether technological progress is facilitating the geographic expan-
sion of the banking industry has important implications. Currently, only a few
organizations in the United States have come close to expanding nationwide; only one
has approached the Riegle-Neal cap of 10% of national bank and thrift deposits in
one organization; and some banking organizations have explicitly reduced their
11. These regressions repeat the Equation (1) specification with two exceptions: we exclude the
variable tas it has no cross-sectional variation; and we exclude UNITB, LIMITB, INTERSTATE, andACCESS as they do not vary for banks within a state in a given year, making them redundant to theSTATE DUMMIES.
-
8/13/2019 Berger - 2006 - Technological Progress and the Geographic Expansion of the Banking Industry
29/32
ALLEN N. BERGER AND ROBERT DEYOUNG : 1511
geographic footprints. Similarly, very few banks have taken ad