Content uploaded by Robert Morrison
Author content
All content in this area was uploaded by Robert Morrison on Aug 14, 2020
Content may be subject to copyright.
US COMMUNITY BANK PROFITABILITY: A CROSS-
SECTIONAL AND DYNAMIC PANEL ANALYSIS OF
RURAL AND METROPOLITAN BANKS
Robert D. Morrison, University of Texas Permian Basin
Diego Escobari, University of Texas Rio Grande Valley
ABSTRACT
This study compares 5,286 community banks operating in rural and metropolitan
counties from 2000 through the end of 2013 on the variables contributing to bank profitability
using pooled OLS, pooled time-series OLS, and dynamic panels methodologies. Following the
SCP and competition-fragility literature, one would expect a difference in the variables
contributing to profitability. The size of the coefficients indicates that the variables contributing
to profitability differ in magnitude when comparing community banks in metropolitan counties to
those in rural counties. Both the pooled and time-series OLS models indicate that bank size
contributes to profitability more in metropolitan areas; however, on average, rural banks have
higher return on assets, higher net interest margins, and higher non-interest income. These
findings provide some support for the competition-fragility argument that more competition in
banking, as seen in metropolitan areas, leads to lower net interest margins. Arguably, the higher
net interest margins and contribution of non-interest income to profits in the concentrated rural
bank markets supports the structure-conduct-performance paradigm that when few competitors
exist in a market, they are more likely to collude, implicitly or explicitly, to extract higher profits.
The findings of this study indicate that community banks are not a homogenous group and
highlight the importance of segregating rural and metropolitan banks when examining the US
community banking industry.
INTRODUCTION
The US Banking industry in the US has undergone dramatic changes over the past 30
years as restrictions of both the geographic area of operation and the scope of financial services
banks can offer have changed dramatically. Until 1911, states regulated banks in the US. Even
after federal regulation a two-tiered banking system of both state and federally chartered banks
existed and depression era federal regulations limited banks to whatever the state they operated
in allowed in terms of geographic areas. The result was a large number of small banks serving
communities across the nation. Beyond that, Great Depression era Glass-Steagall Act of 1933,
limited the scope financial activities in which commercial banks could participate. Although an
in-depth discussion is beyond the scope of this paper, those limitations diminished from the
1930s to the 1980s through various court decisions and legislative and regulatory changes. In the
1980s a series of legislative initiatives, leading up to the Gramm-Leach-Bliley Act of 1999,
eliminated most of the remaining limitations on the geographic scope of banks and restrictions
on what services entities in the financial services sector could offer. What followed was a
massive progression of acquisitions and mergers as commercial banks, investment banks, and
insurance companies combined into comprehensive financial services firms.
In a quest to cover the nation or particular regions of it, publicly traded banks acquired
banks across the nation with the vast majority, 87% of branches, being in metropolitan areas.
This resulted in a 59% decrease in the number of bank charters and over 80% of all bank assets
held by only 107 banks. The remaining 6,356 remaining small banks held only 14% of bank
assets. Nonetheless, these small community banks play an important role in the US economy
because they continue to provide the vast majority of funding to small businesses and small
businesses continue to employee the vast majority of people in the US. In addition, more of the
US population is migrating to metropolitan areas, and that is likely where community banks
encounter the greatest competition from the massive nationwide and regional banks. Therefore, it
is important to understand how deregulation has changed the competitive environment of
community banking and examine the two distinct environments, rural and metropolitan, where
community banks operate. Previous studies have treated community banks as a homogenous
group despite the fact that metropolitan community banks account for over 80% of US bank
failures (Morrison, Jung, Jackson, Escobari, & Sturges, 2016). Using FDIC variables that
contribute to bank profitability, this study demonstrates that there is a difference in the two
competitive environments and highlights the need to segregate when conducting research on US
community banks.
LITERATURE REVIEW
Structure-Conduct-Performance and Bank Deregulation
Due to the evolution of banking regulation in the US, the restrictions on geographic
operating area resulted in most US banks being small banks with tight ties to the communities
that they operated in. Great Depression era legislation, the Glass-Steagall Act of 1933, also
limited the scope of bank activities by prohibiting commercial banks from engaging in
investment banking (Calomiris, 2010). The Douglas Amendment in 1956 allowed states to
establish the guidelines under which banks from other states could do business; however, the
banking industry remained highly regulated and the vast majority of US banks operated in single
counties or metropolitan areas with only a few competitors. During this same timeframe,
legislative activity in the area of anti-trust made inter-industry data available for researchers to
analyze using cross-sectional approaches (e.g., Bain, 1951, 1956). These studies provided insight
into the relationship between competitor concentration in a particular industry, also referred to as
the market structure, and profitability. The use of observable industry structure indicators, such
as concentration ratios, to measure the degree of competition lead to the development of the
structure-conduct-performance paradigm (SCP) (Schmalensee, 1982, 1985, 1989). From one
point of view, in highly concentrated markets competitors can collude, implicitly or explicitly, to
extract higher profits. In contrast, profits may be the result of efficiencies that result from
economies of scale in plant, firm, and advertising efforts.
In the 1980s, there was a movement to enhance competition in the financial services
industry. During the legislative process Stephen Friedman (1981), the Securities and Exchange
Commission Commissioner at the time, commented that in the future only ten large banks would
cover the US. Federal Reserve researcher Alton Gilbert (1984) reviewed 45 SCP studies on the
banking industry to examine the issues of collusion and efficiencies through achieving
economies of scale. He found that the studies on the influence of market structure were highly
variable, but did not seem to support that competition concentration leads to collusion in the
banking industry and that single small banks do not appear to be more costly to operate than a
branch of a large bank. Gilbert (1984) did caution that the studies reviewed did not provide a
solid basis to generalize about large banks operating branches across the nation. As the result of
a series of legislative actions from the Depository Institutions Deregulation and Monetary
Controls Act (DIDMCA) of 1980 to the Gramm-Leach-Bliley Act of 1999 Congress deregulated
the US financial services industry. It turns out that Stephen Friedman was wrong only about the
number of banks blanketing the nation, as of 2020 it is 4 instead of 10; JP Morgan Chase, Bank
of America, Citi Group, and Wells Fargo. At the end of 2011, only 107 banks held 80% of
industry assets and federally insured bank and thrift charters fell from 17,901 in 1985 to 7,353 in
2011. However, despite the industry consolidation and increased competition, locally owned
community banks have not disappeared. Despite only holding 14% of total bank assets, they are
the most common FDIC insured institution and supply most of the credit to small businesses in
the US (FDIC CBS, 2012).
Beyond deregulation, technology has dramatically changed the competitive environment
of banking in the last 10 to 15 years. Internet banking has gone from a novel concept to a service
that bank customers expect. More recently, smartphones have enabled mobile banking and the
ability to take a photo of a check to deposit it. Combined with mobile electronic payments this is
quickly making visits to a physical bank a rare event. On the one hand, technology can bring cost
reductions that lead to greater efficiency; however, the initial capital investment and the need for
highly skilled, therefore costly, support staff can put technology implementation out of the reach
of small banks. Community banks in large metropolitan areas would arguably have a larger
customer base and assets to cover technology implementation and support cost; however, those
are the community banks most likely confronting the highest concentration of competition from
the large nationwide and regional banks. This is because the large banks have focused on
acquisitions in metropolitan areas while avoiding the small rural communities. Therefore, this
study compares the factors contributing to community bank profitability on rural versus
metropolitan areas.
Determinants of Community Bank Profitability
Studies examining bank profitability have mostly used the SCP paradigm focusing on
market concentration and bank efficiency (e.g., Berger, 1995a; Smirlock, 1985). As discussed
previously, the dispute lies in the underlying causation of market power or efficiency through
economies of scale. However, regardless of the level of market concentration, exogenous
economic conditions affect community bank profits; however, when faced with favorable
economic conditions, managerial skill will result in some banks performing better than others
(Kupiec & Lee, 2012). Although return on equity (ROE) and return on assets (ROA) are often
used to measure firm profitability, the study of community banks brings an interesting problem
because about one-third of small banks are Type-S corporations. Because Type-S corporations
act as a pass-through entities that pay no income tax at the corporate level and pass the profits on
to shareholders who pay income tax at the individual level, comparing ROA or ROE between
Type-S and Type-C banks would be erroneous. Therefore, this study uses pre-tax ROA as a
measure of profitability (FDIC variable ptxroa).
Traditionally, banks make profits by operating as financial intermediaries by paying
interest on deposits and loaning those funds out at higher rates. As a result, the gross profit from
interest comes from the difference in those rates, which is the net interest margin (FDIC variable
NIMY). In highly competitive markets banks would offer higher interest rates to attract
depositors; however, by the same reasoning, to attract good clients to lend to banks would have
to offer attractive loan rates and the net interest margin would be lower in these markets.
However, partly due to competition and partly due to deregulation, banks have turned to
generating income through non-interest activities that range from fees on services to operations
in the forward and futures markets (FDIC variable noniiay). As is the case in any business,
operating expenses reduce the gross profits and in banking terminology these are non-interest
expenses (FDIC variable nonixay); the more efficient a bank is the lower its relative non-interest
expense. Efficiency can come through reaching economy of scale and bank asset size maybe
used as a proxy (FDIC variable asset5).
Given that the interest income is the difference in the rates paid on deposits and the
interest charged for loans and that higher riskier loans pay higher interest rates, banks can
arguably increase profitability by taking on riskier loan portfolios. Because of competition for
deposits, there is a lower limit of what a bank can pay and retain sufficient deposits to lend. This
is the basis of the charter value or competition-fragility views (Hellmann, Murdock, & Stiglitz,
2000; Keeley, 1990). Because deposit insurance can act as a put option that limits bank
shareholder losses to the capital invested, banks may take on more risk and maintain lower
capital to asset ratios (CAR). While the literature is not conclusive (Canoy, van Dijk, Lemmen,
de Mooij, & Weigand, 2001; Carletti & Hartmann, 2003), Berger (1995b) found that higher CAR
correlated with higher profits. One possible explanation is that higher CAR leads to lower
insurance premiums, and that contributes to higher profits. Under either argument, CAR is an
important factor in explaining bank profitability (FDIC variable eqv).
MODELS
The data comes from the FDIC quarterly Performance and Conditions Ratios reports.
Because this study focuses only on community banks, we restrict the data to those banks that met
the definition of community banks in the 2012 FDIC Community Banking Study that reported
for the fourth quarter of 2012. The data is from individual banks and excludes bank holding
companies. To avoid the issues related to ratios with De Novo banks, we excluded institutions
that joined the FDIC after January 2, 1998. A dummy variable indicated whether the bank
operated in a rural (0) or metropolitan (1) county. The data contains 296,098 observations from
the quarterly FDIC Performance Reports from 5,286 unique community banks operating from
2000 through the end of 2013.
The methodology in this paper follows that used by Goddard, Molyneux, and Wilson
(2004) to evaluate the determinants of profitability of banks across European countries. The
content of the model is as follows:
∏i,t = f (∏ i,t-1 , s i,t, oi,t , ci,t d 1,i ) (1)
Where ∏i,t is the profit of the bank i in year t, as measured by pre-tax return on assets; s i,t is
the natural logarithm of total assets average over the preceding five years; oi,t is the off balance
sheet or non-interest income; ci,t is CAR; and d 1,i = 1 for metro and 0 for rural. The inclusion of
s i,t captures any relationship between bank size and profitability. Following the SCP literature, a
positive sign may indicate that large banks may benefit from economies of scale or scope or they
may benefit from brand image. In the alternative, a negative sign may indicate that size results in
diseconomies of scale.
Since deregulation began, banks have increased income via non-interest income
generated through fees for services and various contingent liabilities such as letters of credit, and
other non-traditional banking activities including operations in the forward and futures markets.
In competitive markets, non-interest income may play an important role in profitability. CAR is a
crude proxy for risk; however, the competition-frailty view argues that less CAR contributes to
profitability while the lower deposit insurance premium view argues that higher CAR results in
greater profitability. Nonetheless, the goal of this study is not to resolve these differences but to
better understand the factors that contribute to bank profitability in community banks operating
in rural and metropolitan areas.
The pooled cross-sectional time-series structure of the data set enables the estimation of
several variants of the relationship summarized in (1).
Pooled cross-sectional time-series model, estimated using OLS
∏i,t = α1+ α2 ∏ i,t-1 + α s i,t + α oi,t + α ci,t + α d 1,i + ui,t
i= 1,……N, t = 2 ……T (2)
Cross-sectional model, estimated using OLS
∏i,t = β1+ β s i,t + β oi,t + β ci,t + β d 1,i + wi,t
i= 1,……N (3)
Dynamic panel model GMM
∏i,t = ϒ1+ ϒ 2 ∏ i,t-1 + ϒ s i,t + ϒ oi,t + ϒ ci,t + η i + vi,t
i= 1,……N, t = 2 ……T (4)
The pooled model, equation (2), assumes that cross-sectional variation in any
independent variable has the same implication for profit variation over time in that variable for
an independent bank. During the period from 2000 to 2013, there were major shocks that
included a terrorist attack and a banking crisis that resulted in two recessions. Given that banking
profits correlate with economic expansion and recession (Kupiec & Lee, 2012), the use
individual bank differences from yearly means of all banks in the sample removes the exogenous
effects of the economic cycle; in other words, economy-normalized values. Estimating the
equations using both the data as reported and differenced from yearly means for all community
banks provides some ability to understand how economic expansion and contraction effects
profitability in rural and metropolitan banks differently.
RESULTS
Table 1 reports the summary data on the untransformed dependent and independent
variables used in the empirical model. Table 1 reports the summary data for all community banks
(observations = 296,098) and for community banks operating in the rural (observations =
160,142) and metropolitan (observations = 135,696) areas. This data indicates that on average,
rural banks have higher return on assets, higher net interest margins, and higher non-interest
income. These findings provide some support for the competition-fragility argument that more
competition in banking, as seen in metropolitan areas, leads to lower net interest margins.
Arguably, the higher net interest margins and contribution of non-interest income to profits in the
concentrated rural bank markets supports the structure-conduct-performance paradigm that when
few competitors exist in a market, they are more likely to collude, implicitly or explicitly, to
extract higher profits.
roaptx asset5 noniiay eqv nimy nonixay observations
mean 1.358655 229266.6 0.809128 10.97072 3.987371 3.065103 296,098
sd 3.483859 428960.3 5.615186 3.809603 0.955995 3.807264
min -212.39 1055.25 -23.02 -1.69 -3.24 -0.23
max 419.01 1.30E+07 1066.4 95.9 72.64 1099.33
mean 1.434388 146835.1 0.691578 11.07118 4.026 2.944052 160,402
sd 1.883635 205449.8 0.86468 3.568647 0.918214 1.136296
min -141.32 1055.25 -6.63 -0.62 0 0
max 53.86 4511235 87.28 81.55 72.64 72.64
mean 1.269134 326706.2 0.94808 10.85198 3.941708 3.208193 135,696
sd 4.719703 578010.3 8.239063 4.072919 0.996888 5.483222
min -212.39 2816 -23.02 -1.69 -3.24 -0.23
max 419.01 1.30E+07 1066.4 95.9 29.02 1099.33
Rural Community Banks
Metro Community Banks
All Community Banks
Table 1
Descriptive Statistics
Pooled OLS Regressions
Tables 2 through 7 report the results of pooled OLS regressions for both the economy-
normalized data, which is the difference in the individual bank value and the mean for the year of
all banks on for that variable.
Source SS df MS Nuber of obs = 296098
F(5,296092) = .
Model 2586444.01 5 517288.8 Prob > F = 0.0000
Residual 1007365.93 3.402205 R-squared = 0.7197
Adj R-Squared = 0.7197
Total 3593809.94 12.1373 Root MSE = 1.8445
roaptx Coef. Std. Err. t P>(t)
lnasset5 -0.0525325 0.0031807 -16.52 0.0000 -0.5876650 -0.4629850
noniiay 0.998067 0.0012559 794.68 0.0000 0.9956054 1.0005290
eqv 0.0013112 0.0009108 1.44 0.1500 -0.0004738 0.0030963
nimy 0.8185951 0.0036912 221.77 0.0000 0.8113605 0.8258297
nonixay -1.002629 0.0018636 -538 0.0000 -1.0062810 -0.9989760
_cons 0.9590993 0.0447775 21.42 0.0000 0.8713366 1.0468620
[95% Conf. Interval]
POOLED OLS ALL BANKS USING NON-ECONOMY-NORMALIZED
Table 2
Source SS df MS Nuber of obs = 296098
F(5,296092) = .
Model 2583750.98 5 516750.19 Prob > F = 0.0000
Residual 878563.16 2.967196 R-squared = 0.7462
Adj R-Squared = 0.7462
Total 3462314.13 11.6932 Root MSE = 1.7226
droaptx Coef. Std. Err. t P>(t)
dlnasset5 -0.0428519 0.0030309 -14.14 0.0000 -0.0487924 -0.0369113
dnoniiay 0.9984751 0.0011749 849.87 0.0000 0.9961724 1.0007780
deqv 0.0042829 0.000853 5.02 0.0000 0.0026110 0.0059549
dnimy 0.8220789 0.0035092 234.26 0.0000 0.8152010 0.8289569
dnonixay -1.003457 0.0017441 -575.34 0.0000 -1.0068760 -1.0000390
_cons -5.41E-06 0.0031656 0.00 0.9990 -0.0062099 0.0061991
[95% Conf. Interval]
POOLED OLS ALL BANKS USING ECONOMY-NORMALIZED
Table 3
Source SS df MS Nuber of obs = 160402
F(5,160396) = 7987.28
Model 113453.59 5 22690.72 Prob > F = 0.0000
Residual 455662.02 6 2.8409 R-squared = 0.1994
Adj R-Squared = 0.1993
Total 569115.61 3.5481 Root MSE = 1.6855
roaptx Coef. Std. Err. t P>(t)
lnasset5 -0.0450983 0.0045527 -9.91 0.0000 -0.0540215 -0.0361751
noniiay 1.077926 0.0073078 147.50 0.0000 1.0636030 1.0922490
eqv -0.0027492 0.0012271 -2.24 0.0250 -0.0051543 -0.0003441
nimy 0.9005445 0.0056208 160.22 0.0000 0.8895278 0.9115613
nonixay -1.132732 0.0064456 -175.74 0.0000 -1.1453650 -1.1200990
_cons 0.9422624 0.0623413 15.11 0.0000 0.8200747 1.0644500
[95% Conf. Interval]
Table 4
POOLED OLS RURAL BANKS USING NON-ECONOMY-NORMALIZED
Source SS df MS Nuber of obs = 160402
F(5,160396) = 9984.79
Model 113154.89 5 22630.78 Prob > F = 0.0000
Residual 363544.70 6 2.2665 R-squared = 0.2374
Adj R-Squared = 0.2373
Total 476699.59 2.9719 Root MSE = 1.5055
droaptx Coef. Std. Err. t P>(t)
dlnasset5 -0.0447315 0.0041818 -10.7 0.0000 -0.0529277 -0.0365354
dnoniiay 1.085784 0.0065819 164.97 0.0000 1.0728840 1.0986850
deqv -0.0001967 0.0010996 -0.18 0.8580 -0.0023518 0.0019584
dnimy 0.9196494 0.0051538 178.44 0.0000 0.9095480 0.9297508
dnonixay -1.149914 0.0058443 -196.76 0.0000 -1.1613690 -1.1384590
_cons 0.0220373 0.003977 5.54 0.0000 0.0142424 0.0298321
[95% Conf. Interval]
POOLED OLS RURAL BANKS USING ECONOMY-NORMALIZED
Table 5
Source SS df MS Nuber of obs = 135696
F(5,135690) = .
Model 2472735.17 5 494547.035 Prob > F = 0.0000
Residual 549951.71 4.053 R-squared = 0.8181
Adj R-Squared = 0.8181
Total 3022686.89 22.2756 Root MSE = 2.0132
roaptx Coef. Std. Err. t P>(t)
lnasset5 -0.0604381 0.0049594 -12.19 0.0000 -0.0701585 -0.0507178
noniiay 0.9916712 0.0014136 701.54 0.0000 0.9889007 0.9944418
eqv 0.0017462 0.0013703 1.27 0.2030 -0.0009395 0.0044320
nimy 0.7977294 0.005632 141.64 0.0000 0.7866908 0.8084681
nonixay -0.9910865 0.0021221 -467.03 0.0000 -0.9952458 -0.9869273
_cons 1.071039 0.0709305 15.1 0.0000 0.9320160 1.2100610
[95% Conf. Interval]
POOLED OLS METRO BANKS USING NON-ECONOMY-NORMALIZED
Table 6
Source SS df MS Nuber of obs = 135696
F(5,135690) = .
Model 2470699.04 5 494139.81 Prob > F = 0.0000
Residual 512765.94 3.7790 R-squared = 0.8281
Adj R-Squared = 0.8281
Total 2982464.98 Root MSE = 1.944
droaptx Coef. Std. Err. t P>(t)
dlnasset5 -0.0412878 0.0048799 -8.4600 0.0000 -0.5085240 -0.0317232
dnoniiay 0.9913288 0.0013664 725.5100 0.0000 0.9886507 0.9940069
deqv 0.0044963 0.001328 3.3900 0.0010 0.0070992 0.0070992
dnimy 0.7958965 0.0055364 143.7600 0.0000 0.8067478 0.8067478
dnonixay -0.9904214 0.0020519 -482.7000 0.0000 -0.9862998 -0.9863998
_cons -0.0414837 0.0055142 -7.5200 0.0000 -0.0306759 -0.0306759
[95% Conf. Interval]
POOLED OLS METRO BANKS USING ECONOMY-NORMALIZED
Table 7
In the pooled OLS analysis from both the economy-normalize and non-economy-
normalized data, the coefficient for net interest margin (nimy) is significantly higher for rural
banks and the coefficient for bank size (asset5) is negative for both rural and metropolitan banks.
The higher net interest rate margin in rural areas where banking competition is more
concentrated supports the SCP paradigm. The negative coefficient for bank size (asset5) provides
some support for the position that large banks may encounter diseconomies of scale. Kupiec and
Lee (2012) found a curvilinear relationship between size and profitability in community banks
where banks as small as $300 million in assets achieved a significant proportion of the gain in
profits while banks over $1 billion in assets were less profitable. As expected, the coefficient for
non-interest expense is negative in all tables. The fact that CAR (eqv) varies in the level of
significance across the different analyses is interesting and calls for further investigation. It is
noteworthy that there were changes in capital requirements after the 2008 financial crisis and this
warrants comparison before and after the changes to gain a better understanding of these results.
Pooled Time Series OLS Regressions
Tables 8 through 13 report the results of pooled time-series OLS regressions for both the
economy-normalized data, which is the difference in the individual bank value and the mean for
the year of all banks on that variable, and the data without any adjustment. Because this is
quarterly data, we lag the dependent variable, pre-tax ROA, by 4 observations to capture the
profit from one year before.
264808
Group variable: crossid Number of groups = 5466
R-sq: within = 0.1913 17
between = 0.9837 avg = 48.4
overall = 0.6838 max = 52
10222.39
0.0000
roaptx Coef. Std. Err. t P>(t)
roaptx
L4. -0.0151202 0.001825 -8.28 0.0000 -0.1869720 -0.0115432
lnasset5 -0.2956692 0.0128646 -22.98 0.0000 -0.3208835 -0.2704548
noniiay 1.016689 0.0051139 198.81 0.0000 1.0066660 1.0267130
eqv 0.0239564 0.0021096 11.36 0.0000 0.0198216 0.0280912
nimy 0.8786408 0.0068121 128.98 0.0000 0.8652892 0.8919924
nonixay -1.079463 0.0053617 -201.33 0.0000 -1.0899720 -1.0689540
_cons 3.559793 0.161227 22.08 0.0000 30243793 3.875794
sigma_u 0.5451846
Sigma_e 1.8953757
rho 0.07641415 (fraction of variance due to u_i)
F test that all u_i = 0 :
F(5465, 259336) = 2.57
Prob > F = 0.0000
POOLED TS OLS ALL BANKS USING NON-ECONOMY-NORMALIZED
Table 8
[95% Conf. Interval]
Fixed-effects (within) regression
Number of obs =
Obs per group: min =
F(6,259336) =
Prob > F =
264808
Group variable: crossid Number of groups = 5466
R-sq: within = 0.2077 17
between = 0.9858 avg = 48.4
overall = 0.7145 max = 52
11331.89
0.0000
droaptx Coef. Std. Err. t P>(t)
droaptx
L4. -0.0014562 0.0018147 -0.80 0.4220 -0.0080131 0.0021006
dlnasset5 -0.2532024 0.0167308 -15.13 0.0000 -0.2859942 -0.2204105
dnoniiay 1.00999 0.0047955 210.61 0.0000 1.0005910 1.0193900
deqv 0.0263796 0.0019936 13.23 0.0000 0.0224722 0.0302870
dnimy 0.9175164 0.0065621 139.82 0.0000 0.9046549 0.9303780
dnonixay -1.076489 0.0050965 -211.22 0.0000 -1.0864780 -1.0665000
_cons -0.0057685 0.0034263 -1.68 0.0920 -0.0124839 0.0009469
sigma_u 0.50932716
Sigma_e 1.7628287
rho 0.07704665 (fraction of variance due to u_i)
F(5465, 259336 ) = 2.74
POOLED TS OLS ALL BANKS USING ECONOMY-NORMALIZED
Table 9
F test that all u_i = 0 :
Prob > F = 0.0000
Fixed-effects (within) regression
Number of obs =
Obs per group: min =
F(6,259336) =
Prob > F =
[95% Conf. Interval]
142943
Group variable: crossid Number of groups = 3106
R-sq: within = 0.1137 1
between = 0.6879 avg = 46
overall = 0.177 max = 52
2989.01
0.0000
roaptx Coef. Std. Err. t P>(t)
roaptx
L4. -0.286746 0.0025348 -11.31 0.0000 -0.0336429 -0.2370630
lnasset5 -0.2600908 0.0179843 -14.46 0.0000 -0.2953396 -0.2248420
noniiay 1.177087 0.0108638 108.35 0.0000 1.1557940 1.1983800
eqv 0.010226 0.0029607 3.45 0.0001 0.0044231 0.0160290
nimy 0.9447741 0.009168 103.05 0.0000 0.9268051 0.9627431
nonixay -1.21727 0.0105515 -115.36 0.0000 -1.2379500 -1.1965890
_cons 3.302526 0.2191399 15.07 0.0000 2.873016 3.732036
sigma_u 0.45151986
Sigma_e 1.7541192
rho 0.0621403 (fraction of variance due to u_i)
F test that all u_i = 0 :
Prob > F = 0.0000
F(3105, 139831 ) = 2.13
Fixed-effects (within) regression
POOLED TS OLS RURAL BANKS USING NON-ECONOMY-NORMALIZED
Number of obs =
Obs per group: min =
F(6,139831) =
Prob > F =
[95% Conf. Interval]
Table 10
142943
Number of groups = 3106
R-sq: within = 0.1349 1
between = 0.6349 avg = 46
overall = 0.1937 max = 52
3633.01
0.0000
droaptx Coef. Std. Err. t P>(t)
droaptx
L4. -0.0086261 0.0025089 -3.44 0.0010 -0.0135536 -0.0037187
dlnasset5 -0.0466864 0.023388 -17.39 0.0000 -0.4525265 -0.3608462
dnoniiay 1.194264 0.0098972 120.67 0.0000 1.1748660 1.2136620
deqv 0.0136587 0.0026671 5.12 0.0000 0.0084312 0.0188861
dnimy 0.9976664 0.008516 117.15 0.0000 0.9809752 1.0143580
dnonixay -1.256684 0.0097787 128.51 0.0000 -1.2758500 -1.2375180
_cons -0.0843095 0.0077981 -10.81 0.0000 -0.0995937 -0.0690254
sigma_u 0.51602911
Sigma_e 1.5597466
rho 0.09865752 (fraction of variance due to u_i)
F test that all u_i = 0 :
F(3105, 139831 ) = 2.39
Prob > F = 0.0000
POOLED TS OLS RURAL BANKS USING ECONOMY-NORMALIZED
Table 11
Group variable: crossid
Obs per group: min =
Prob > F =
[95% Conf. Interval]
Fixed-effects (within) regression
Number of obs =
F(6, 139831) =
121865
Group variable: crossid Number of groups = 2652
R-sq: within = 0.2468 1
between = 0.9868 avg = 46
overall = 0.7889 max = 52
6508.77
0.0000
roaptx Coef. Std. Err. t P>(t)
roaptx
L4. -0.0059829 0.0026497 -2.26 0.0240 -0.0111762 -0.0007896
lnasset5 -0.3870837 0.0204479 -18.93 0.0000 -0.4271612 -0.3470063
noniiay 0.9792774 0.0063348 154.59 0.0000 0.9668613 0.9916934
eqv 0.0349243 0.0031389 11.13 0.0000 0.0287720 0.0410765
nimy 0.8714233 0.011354 76.75 0.0000 0.8491697 0.8936768
nonixay -1.054804 0.0069426 -151.93 0.0000 -1.0684110 -1.0411960
_cons 4.574221 0.263749 17.34 0.0000 4.057276 5.091167
sigma_u 0.77265
Sigma_e 2.0447255
rho 0.12494804 (fraction of variance due to u_i)
F test that all u_i = 0 :
Prob > F = 0.0000
F(2651, 119207 ) =
POOLED TS OLS METRO BANKS USING NON-ECONOMY-NORMALIZED
Table 12
Fixed-effects (within) regression
Number of obs =
Obs per group: min =
F(6, 119207) =
Prob > F =
[95% Conf. Interval]
The time-series OLS results also show that the net interest rate margin of rural banks is
higher than that of their metropolitan counterparts. The time-series OLS data also indicates that
CAR differs from the results in the pooled cross-sectional regressions. In the time-series
regressions, CAR (eqv) is positive and significant across all regressions. An interesting result is
that lagged pre-tax ROA is negative when significant, although size of the coefficient is
relatively small. This warrants further investigation. Otherwise, the signs of the coefficients are
the same as in the cross-sectional OLS regressions with size (asset5) and non-interest expense
being (nonixay) negative.
121865
Group variable: crossid Number of groups = 2652
R-sq: within = 0.2525 1
between = 0.9906 avg = 46
overall = 0.8054 max = 52
6710.78
0.0000
droaptx Coef. Std. Err. t P>(t)
droaptx
L4. 0.0017847 0.0026506 0.07 0.5010 -0.0034105 0.0069799
dlnasset5 -0.219934 0.0270817 -8.12 0.0000 -0.2730137 -0.1668543
dnoniiay 0.972894 0.0061332 158.63 0.0000 0.9608731 0.9849149
deqv 0.0339648 0.0030848 11.01 0.0000 0.0279186 0.0400111
dnimy 0.9223176 0.0113704 81.12 0.0000 0.9000318 0.9446034
dnonixay -1.039375 0.0067974 -152.91 0.0000 -1.0526980 -1.0260530
_cons 0.025413 0.0107307 2.37 0.0180 0.004381 0.0464449
sigma_u 0.67401676
Sigma_e 1.9709322
rho 0.10472332 (fraction of variance due to u_i)
F test that all u_i = 0 :
Prob > F = 0.0000
F(2651, 119207 ) = 2.95
POOLED TS OLS METRO BANKS USING ECONOMY-NORMALIZED
Table 13
Fixed-effects (within) regression
Number of obs =
Obs per group: min =
F(6, 119207) =
Prob > F =
[95% Conf. Interval]
Dynamic Panel Estimation
The null hypothesis of the Sargan test that the over-identifying restrictions are valid were
rejected for both the non-economy-normalized and economy-normalized panel regressions;
therefore, they are not valid. The Arellano-Bond test for zero autocorrelation in first-differenced
errors revealed evidence of misspecification for the non-economy-normalized panel regressions.
However, there was no evidence of misspecification in the economy-normalized regressions.
Despite the results of the Sargan test, we follow Goddard, Molyneux, and Wilson (2004), who
encountered similar issues, and provide the results of the economy-normalized regressions with
the above caveat.
249338
Group variable: crossid Number of groups = 5466
Time variable: timeid
16
avg = 45.61617
max = 51
Number of instruments - 1.5e+03 2444.07
0.0000
One-step results
Robust
dr0roaptx Coef. Std. Err. z P>(z)
dr0roaptx
L1. 0.3973775 0.0441071 9.01 0.000 0.3109292 -0.4838258
L2. -0.0175836 0.0152226 -1.16 0.248 -0.0474193 0.0122521
L3. -0.01011649 0.0191132 -5.29 0.000 -0.1386261 -0.0637036
L4. 0.0369237 0.118291 3.12 0.002 0.0137390 0.0601083
dlnasset5 -2.720921 0.222671 -12.22 0.000 -3.1573480 -2.2844940
dnoniiay 0.8255317 0.051451 16.05 0.000 0.7246896 0.9263738
deqv 0.0777918 0.0186109 4.18 0.000 0.0413152 0.1142685
dnimy 0.7787569 0.465123 16.74 0.000 0.6875944 0.8699194
dnonixay -0.8508638 0.068657 -12.39 0.000 -0.9854290 -0.7162986
_cons 0.0321022 0.0397568 0.81 0.419 -0.0458197 0.1100241
Standard: _cons
Standard: D.dlnassat5 D.dnoniiay D.deqv D.dnimy D.dnoixay
(Std. Err. adjusted for clustering on crossid)
DYNAMIC PANEL ALL BANKS USING ECONOMY-NORMALIZED DATA
Table 14
Arellano-Bond dynamic panel-data estimation
Instruments for differenced equation
Instruments for level equation
GMM-type: L(2/.).dr0aptx
Number of obs =
Obs per group: min =
Wald chi 2(9) =
Prob > chi 2 =
[95% Conf. Interval]
134391
Group variable: crossid Number of groups = 3103
Time variable: timeid
1
avg = 43.31002
max = 51
Number of instruments - 1.5e+03 52572.28
0.0000
One-step results
Robust
dr0roaptx Coef. Std. Err. z P>(z)
dr0roaptx
L1. 0.4498633 0.002912 154.49 0.000 0.4441559 0.4555707
L2. -0.0310608 0.0029281 -10.61 0.000 -0.0036800 -0.0253219
L3. -0.1364222 0.0027547 -49.52 0.000 -0.1418214 -0.1310231
L4. 0.0647087 0.002365 27.36 0.000 0.0600733 0.0693440
dlnasset5 -3.173356 0.075241 -42.18 0.000 -3.3208230 -3.0258840
dnoniiay 1.065877 0.0126253 84.42 0.000 1.0411320 1.0906220
deqv 0.0288549 0.0055402 5.21 0.000 0.0179963 0.0397135
dnimy 0.78339973 0.0138901 56.4 0.000 0.7561731 0.8106214
dnonixay -1.117461 0.0127226 -87.81 0.000 -1.1424050 -1.0925170
_cons -8499601 0.029723 -40.53 0.000 -891065 -0.8088553
Table 15
DYNAMIC PANEL RURAL BANKS USING ECONOMY-NORMALIZED
Instruments for differenced equation
GMM-type: L(2/.).dr0aptx
Standard: D.dlnassat5 D.dnoniiay D.deqv D.dnimy D.dnoixay
Instruments for level equation
Standard: _cons
Arellano-Bond dynamic panel-data estimation
Number of obs =
Obs per group: min =
Wald chi 2(9) =
Prob > chi 2 =
[95% Conf. Interval]
(Std. Err. adjusted for clustering on crossid)
114947
Group variable: crossid Number of groups = 2652
Time variable: timeid
1
avg = 43.34351
max = 51
Number of instruments - 1.5e+03 1468.22
0.0000
One-step results
Robust
dr0roaptx Coef. Std. Err. z P>(z)
dr0roaptx
L1. 0.3685604 0.062088 5.94 0.000 0.0246870 0.4902506
L2. -0.0129432 0.0175742 -0.74 0.461 -0.0473881 0.0215017
L3. -0.0845967 0.0237472 -3.56 0.000 -0.0121140 -0.0380530
L4. 0.0199218 0.0136042 1.46 0.143 -0.0067419 0.0465855
dlnasset5 -1.644753 0.3033744 -5.42 0.000 -2.2387560 -1.0495500
dnoniiay 0.8151457 0.0397175 20.52 0.000 0.7373007 0.8929906
deqv 0.102721 0.0309129 3.32 0.001 0.0421327 0.1633092
dnimy 0.9003749 0.0895391 10.06 0.000 0.7248815 1.0758680
dnonixay -0.8179056 0.0532217 -15.37 0.000 -0.9222183 -0.7135929
_cons 0.5602436 0.1056296 5.3 0.000 0.3532133 0.7672738
DYNAMIC PANEL METRO BANKS USING ECONOMY-NORMALIZED
Table 16
Instruments for differenced equation
GMM-type: L(2/.).dr0aptx
Standard: D.dlnassat5 D.dnoniiay D.deqv D.dnimy D.dnoixay
Instruments for level equation
Standard: _cons
Arellano-Bond dynamic panel-data estimation
Number of obs =
Obs per group: min =
Wald chi 2(9) =
Prob > chi 2 =
[95% Conf. Interval]
(Std. Err. adjusted for clustering on crossid)
CONCLUSIONS
This study demonstrates that US community banks are not a homogenous group. Rural
and metropolitan community banks have differences on the variables contributing to profitability.
Therefore, it is important to segregate the two when conducting studies on community banking
in the US. This study compares community banks operating in rural and metropolitan counties
on the variables attributing to bank profitability using pooled OLS, pooled time-series OLS, and
dynamic panels methodologies. Following the SCP and competition-fragility literature and given
that community banks operating in metropolitan areas are facing direct competition from
massive nationwide and regional banks whereas rural community banks are not to a great extent,
one would expect a difference in the variables contributing to profitability. This study is
exploratory in nature in that the purpose is to provide informative insight into areas in need of
further research.
Overall, the three methodologies are more alike than different in that the signs of the
coefficients are substantially alike across all three methodologies. The size of the coefficients
indicates that the variables contributing to profitability differ in magnitude when comparing
community banks in metropolitan counties to those in rural counties. Both the pooled and time-
series OLS models indicate that bank size contributes to profitability more in metropolitan areas.
Perhaps, in a rural community with only a few banks size is not as important when it comes to
attracting and retaining customers. In the results from the dynamic panel analysis, metropolitan
banks have a smaller size coefficient than rural banks; however, we must view these results with
caution given the results of the Sargan test.
The results across all three methodologies provide some interesting insight into net
interest margins, non-interest income, and non-interest expenses. Traditionally, the majority of
bank profit comes from the difference in the rate paid for deposits and the rates charged for loans.
In both the pooled OLS and pooled time-series OLS models, net-interest margins contribute less
to profitability in metropolitan banks. This would conform to the competition-fragility argument
that competition in the banking sector leads to lower net interest margins. One might expect that
banks in metropolitan areas might have more opportunities to profit from non-interest fee income;
however, the results from the pooled OLS, pooled time-series OLS, and dynamic panel models
indicate that non-interest income contributes less to profitability in metropolitan banks. One
possibility might be that metropolitan banks compete with massive nationwide and regional
banks and as a result have to compete by offering free or lower cost services whereas the SCP
paradigm indicates that small banks in rural communities have a greater ability to collude on fees
such as checking, overdraft, letters of credit, and charges for other services. Non-interest expense
is negative in all results as expected. The results from the pooled OLS, pooled time-series OLS,
and dynamic panel models indicate that non-interest expense has less of an impact on profits in
metropolitan banks. Given the higher real estate and labor prices in metropolitan areas, one
might expect non-interest expense to have more of a negative impact on profits in big cities than
small towns. However, it may be possible that efficiencies achieved though economies of scale
in metropolitan banks may result in non-interest expenses being less of a factor. In the results
from the dynamic panel analysis, metropolitan banks have a larger net interest margin coefficient
than rural banks; however, we must view these results with caution given the results of the
Sargan test.
Finally, the coefficient for equity was small but positive and significant across all
methodologies, except cross-sectional OLS by type, with the coefficient being larger for
metropolitan banks. However, future research needs to examine this variable before and after the
financial crisis because there were regulatory changes that required increases in CAR after the
crisis. It would be interesting to examine the difference in CAR between rural and metropolitan
banks prior to the regulatory changes. Given the wide fluctuation in economic conditions over
the period of this study, we ran all studies using economy-normalized data where we subtracted
the individual bank numbers for each variable from the year mean for all banks. This did not lead
to any changes in the signs of coefficients; however, it is noteworthy that only the economy-
normalized dataset passed the Arellano-Bond test for zero autocorrelation in first-differenced
errors. However, both data samples failed to pass the Sargan test and as a result, one must view
the dynamic panel results with caution.
REFERENCES
Bain, J. S. (1951). Relation of profit rate to industry concentration: American manufacturing,
1936-1940. Quarterly Journal of Economics, 65(3), 293-324.
Bain, J. S. (1956). Barriers to new competition, their character and consequences in
manufacturing industries. Cambridge MA: Harvard University Press.
Berger, A. N. (1995a). The profit-structure relationship in banking—Tests of market-power and
efficient-structure hypotheses. Journal of Money, Credit & Banking, 27(2), 404-431.
Berger, A. N. (1995b). The relationship between capital and earnings in banking. Journal of
Money, Credit & Banking, 27(2), 432-456.
Calomiris, C. W. (2010). The political lessons of Depression-era banking reform. Oxford Review
of Economic Policy, 26(3), 540-560. doi: 10.1093/oxrep/grq020
Canoy, M., van Dijk, M., Lemmen, J., de Mooij, R., & Weigand, J. (2001). Competition and
stability in banking. The Hague, Netherlands: CBP Netherlands Bureau for Economic
Policy Analysis.
Carletti, E., & Hartmann, P. (2003). Competition and stability: What’s special about banking? In
P. Mizen (Ed.), Monetary History, Exchange Rates and Financial Markets: Essays in
honor of Charles Goodhart (Vol. 2). Cheltenham, UK: Edward Elgar.
FDIC CBS. (2012). FDIC Community Banking Study. Washington D.C.: Federal Deposit
Insurance Corporation.
Friedman, S. J. (1981). The illusion of intrastate banking. Washington D.C.: Securities and
Exchange Commission Retrieved from
http://www.sec.gov/news/speech/1981/050581friedman.pdf.
Gilbert, R. A. (1984). Bank market structure and competition. Journal of Money, Credit &
Banking, 16(4), 617-645.
Goddard, J., Molyneux, P., & Wilson, J. O. S. (2004). The profitability of European banks: a
cross-sectional and dynamic panel analysis. Manchester School (14636786), 72(3), 363-
381. doi: 10.1111/j.1467-9957.2004.00397.x
Hellmann, T. F., Murdock, K. C., & Stiglitz, J. E. (2000). Liberalization, moral hazard in
banking, and prudential regulation: Are capital requirements enough? American
Economic Review, 90(1), 147-165.
Keeley, M. C. (1990). Deposit insurance, risk, and market power in banking. American
Economic Review, 60(5), 1183-1200.
Kupiec, P., & Lee, Y. (2012). What factors explain differences in Return on Assets among
Community Banks? Washington D.C.: Federal Deposit Insurance Corporation.
Morrison, R. D., Jung, J., Jackson, D. O., Escobari, D., & Sturges, D. L. (2016). The competitive environment of
community banking: Impact on microenterprise access to bank financing. Proceedings of the 2016
Academy of Management Conference: Anaheim, CA. doi:10.5465/ambpp.2016.17551abstract
Schmalensee, R. (1982). Product differentiation advantages of pioneering brands. American
Economic Review, 72(3), 349.
Schmalensee, R. (1985). Do markets differ much? American Economic Review, 75(3), 341.
Schmalensee, R. (1989). Inter-industry studies of structure and performance. In R. Schmalensee
& R. D. Willig (Eds.), Handbook of Industrial Organization (pp. 959-1009). Amsterdam:
Elsevier Science Publishers.
Smirlock, M. (1985). Evidence on the (non) Relationship between concentration and profitability
in banking. Journal of Money, Credit & Banking, 17(1), 69-83.
