Abstract

Bank profitability in the USA was extremely high in the pre-crisis period, yet this did not prevent the current crisis. It has become clear that these profits were on shaky grounds and also that bank profits were not used to buttress banks’ capital bases. This paper analyses the effects of structure on profitability from 1994 to 2005. Bank-level panel data are used to test the effects of concentration, market power, bank size and operational efficiency on profitability. Efficiency is not found to be a strong determinant of profitability, suggesting that banks’ high profits during this period were not ‘earned’ through efficient performance. Robust evidence is found that concentration increases bank profitability. This holds even when the largest banks are excluded from the sample, suggesting that the relationship between concentration and profitability acts in a generalised structural way and that the higher profits arising from concentration are at the expense of the rest of the economy. The analysis points to various policy implications relevant to the current crisis, in particular in terms of the legitimacy of expectations of the restoration of pre-crisis profit rates and the need for much stronger regulation of the banking sector, especially in terms of the structure of the sector, pricing behaviour and use of profits.

Introduction

In the decade or so prior to the financial crisis, US banks enjoyed a period of virtually unprecedented profitability. The current crisis has exposed the vulnerability of banks’ balance sheets despite these extraordinarily high profits. The increasing leverage of the financial sector is likely to have increased bank profits yet also left banks more exposed.

This study investigates the structural determinants of bank profitability during these ‘fat years’. We focus in particular on whether there were ‘rents’ arising from concentration in the banking sector that could explain anything in terms of banks’ profitability during the pre-crisis period. To the extent that concentration affected profitability, we investigate whether it is just the largest banks that benefitted from concentration or whether concentration benefitted the banking sector as a whole vis-à-vis other sectors of the economy.

An understanding of the relationship between market structure and bank profitability could have policy implications relevant to the current crisis and in the subsequent period. If the evidence suggests that it is a concentrated structure that raises bank profitability, as opposed to higher profitability being the result of efficiency or scale effects, then this might be interpreted as pointing to a greater focus on competition policy and other regulatory interventions to reduce bank concentration. This is particularly germane to current debates around the ‘rescue’ of the financial sector, both in terms of whether previous levels of profitability should be restored and also in terms of what measures should be taken to change the structure and conduct of the sector, and what conditions should be tied to public assistance that may be provided. The analysis of the extent to which bank profits in the pre-crisis period were based on banks’ efficient performance is also relevant to these considerations. Furthermore, understanding whether any ‘rents’ arising out of concentration in the banking sector accrue to the largest banks at the expense of other banks, or whether the entire banking sector gains such rents at the expense of the rest of the economy, can give insights into the extent to which bank concentration brings costs to the ‘real economy’, and this can also have policy implications.

This study uses quarterly bank-level data to conduct panel regressions. Most other studies have relied on either cross-sectional or aggregated time series data. A new index of concentration provides a good measure of the depth and intensity of concentration. The specifications used here draw on and extend the existing literature. Through the testing of regressors for endogeneity and the use of instrumentalisation and generalised method of moments (GMM) techniques, the paper provides a more rigorous treatment of potential problems of simultaneity, which is critical to drawing any firm conclusions from the econometric analysis. A potentially important contribution of the paper is the disaggregation of the banking sector into a stratum of the largest banks and the rest, which allows for an analysis of the relationship between the concentration at the top end of the banking sector and the profitability of the rest of the sector. The findings have implications for understanding the determinants of the high levels of bank profitability in the pre-crisis period and also possible implications for the rest of the economy.

The next section reviews relevant developments in the US banking sector. Section 3 briefly reviews the relevant literature. The empirical analysis is presented in Section 4 and Section 5 concludes.

Overview of developments in the US banking sector

Legislative and regulatory developments facilitating an increase in concentration

There has been an overall movement toward deregulation, especially from the 1980s onwards. This deregulation in all likelihood contributed to the current financial crisis, as well as opening the way to increased concentration. Two aspects of legislative and regulatory changes were of particular relevance in this regard. First, the weakening or elimination of previous limitations on activities permitted to be undertaken by single institutions. This facilitated conglomeration, convergence and rising concentration. Second, the removal of geographical restrictions, particularly on interstate banking, also contributed to increasing concentration.

The Depository Institutions Deregulation and Monetary Control Act (DIDMCA) of 1980 phased out interest-rate ceilings for deposits, eliminated usury ceilings, increased the scope of thrift institutions’ powers and increased the deposit insurance limit. The Garn-St Germain Depository Institutions Act of 1982 was introduced in the context of a crisis among thrifts. It allowed banks to purchase failing banks and thrifts across state lines, facilitating a rise in bank concentration. The legislation also abolished statutory restrictions on real estate lending by national banks and loosened the limits on loans to single borrowers. By 1986, Regulation Q ceilings setting maximum rates on deposit accounts had been phased out, and regulations inhibiting competition between different types of depository institutions in different markets and products were relaxed.

A key court ruling in 1993 (Independent Insurance Agents of America v. Ludwig) allowed national banks to sell insurance from small towns, aiding the trend towards conglomeration. This was further taken forward in subsequent court rulings in 1995 (Nations Bank v. Valic) and 1996 (Barnett Bank v. Nelson) that allowed banks to sell annuities and repealed state restrictions on bank insurance sales.

In 1994 the Riegle Community Development and Regulatory Improvement (CDRI) Act overhauled regulatory structures and processes with an overall deregulatory effect. The Riegle-Neal Interstate Banking and Branching Efficiency Act (IBBEA) of the same year authorised interstate banking and branching (for both US and foreign banks) to be phased in over a 3-year period. Under this legislation a bank holding company meeting certain standards was allowed to acquire a bank in any state. This effectively repealed the Douglas Amendment. Interstate acquisitions were permitted (with some restrictions, but with these restrictions subject to waiver or override by states). The legislation also opened the way for banks meeting required standards to merge across states (as of 1 June 1997), effectively repealing the McFadden Act.

The Gramm-Leach-Bliley Act (GLBA) of 1999 (also known as the Financial Services Modernisation Act) repealed the Glass-Steagall Act. This crucial change allowed different types of financial institutions to affiliate or merge with one another, thus removing key restrictions on conglomeration and facilitating increased concentration. Specific provisions included the authorisation of bank holding companies to act as financial holding companies; ending regulations barring the merger of banks, insurance companies and securities firms; lifting some restrictions governing non-bank banks; allowing a national bank to engage in new financial activities in a financial subsidiary; and allowing national banks to underwrite municipal revenue bonds.

In addition to allowing for increased concentration, the overall deregulatory thrust in legislative developments also facilitated an increase in leverage in the years preceding the present crisis. In particular, the GLBA permitted banks to borrow in order to fund both traditional and non-traditional financial investments, to a greater extent than was previously the case. Increased leverage facilitated higher bank profits but also significantly increased their vulnerability and was no doubt a contributing factor to the current crisis. Furthermore, it has become apparent that banks’ high profits during these ‘fat years’ preceding the crisis (as will be discussed further below) were not used to shore up banks’ capital base and underlying solidity, and the relatively deregulated environment of this period allowed them to get away with this. This regulatory failure must surely be considered a contributory factor to the present crisis.

Empirical trends

Figure 1 shows a precipitous and continuous drop in the number of banks from the mid-1980s on. Whereas in the previous half-century the number of commercial banks hovered within a fairly narrow band, the number fell dramatically from about 1985 onwards. The fall in the number of institutions is due both to failures (particularly in the late 1980s and early 1990s) and to consolidation through mergers and acquisitions (especially from the early 1980s to the early 2000s). The number of commercial bank mergers reached an all-time high in the late 1980s, and this led to an increase in the level of bank concentration. The rate of bank mergers was very high by historical standards between the mid 1980s and early 2000s.

Fig. 1.

Number of commercial banks and savings institutions, 1934–2007.

The following institutional definitions apply to all data presented in this section: (i) commercial banks includes the following groups of banks in the continental US operating under licenses issued by the Treasury or by state banking authorities: national banks, state-chartered commercial banks, loan and trust companies, stock savings banks, private banks under state supervision, and industrial banks. (ii) FDIC-insured savings institutions includes all institutions insured by either the Bank Insurance Fund (BIF) or the Savings Association Insurance Fund (SAIF) that operate under state or federal banking codes applicable to thrift institutions.

Source: FDIC data.

Fig. 1.

Number of commercial banks and savings institutions, 1934–2007.

The following institutional definitions apply to all data presented in this section: (i) commercial banks includes the following groups of banks in the continental US operating under licenses issued by the Treasury or by state banking authorities: national banks, state-chartered commercial banks, loan and trust companies, stock savings banks, private banks under state supervision, and industrial banks. (ii) FDIC-insured savings institutions includes all institutions insured by either the Bank Insurance Fund (BIF) or the Savings Association Insurance Fund (SAIF) that operate under state or federal banking codes applicable to thrift institutions.

Source: FDIC data.

While the causes of these changes fall beyond the scope of this study, we can briefly consider some likely explanations. Deregulation—in particular the removal of geographic restrictions on banking and the weakening of restrictions on permissible banking activities facilitated an increase in mergers and in concentration more generally. One demonstration of the impact of deregulation on concentration is to be found in the jump in mergers (specifically interstate mergers) with the implementation of the IBBEA. Deregulation can be considered particularly important in facilitating rising concentration, in that other factors that could potentially have spurred concentration could not have done so in practice without the weakening or removal of previous legislative and regulatory barriers to this. Furthermore, a regulatory approach that (even implicitly) regards some institutions as ‘too-big-to-fail’ could incentivise growth above a threshold where economies of scale might otherwise be realised.

In addition, developments in the structuring of managerial salaries and bonuses typically incentivised expansion and empire-building, even beyond a point of benefitting banks’ bottom line in the medium- to long-term. Globalisation may have increased economies of scale, particularly among very large banks or niche banks that either compete internationally or whose domestic markets foreign banks were especially active in. Technological developments could also have affected consolidation and concentration through the effects of technology on costs of entry, economies of scale and transaction and screening costs. Macroeconomic conditions contributing to the bank and thrift crises that saw high rates of institutional failure and mergers and acquisitions could also have led to an increase in concentration. Moreover, the virtual explosion in financial sector stock prices from the mid-1990s on seems to have provided banks with the resources to fund takeovers (although the associated increase in new banks may have dampened the net effect of this on actual levels of concentration). Interest rate volatility at times such as the mid-1980s may also have favoured large and diversified financial institutions able to take advantage of economies of scope and scale (at least up to a point).

In the past, periods in which there have been high rates of bank failure or bank difficulties have led to an increase in concentration, because of mergers and acquisitions of failed or failing banks. The current crisis has already prompted several large-scale bank mergers and acquisitions. The long-term effects of the current crisis on bank industry structure remains to be seen, but it is quite foreseeable that it will bring about a process of consolidation and a rise in concentration (unless there are specific policy interventions to avoid such an outcome). Considering the likelihood of the crisis leading to an increase in concentration, it becomes all the more important to analyse the relationship between concentration and profitability.

Figure 2 shows trends in the profitability of commercial banks and savings institutions. This data is for the entire banking sector and shows aggregate figures (i.e. not averages across banks). An upward trend in profitability over time is evident. The major exception is the downturn in profitability during the bank and thrift crisis of the second half of the 1980s and early 1990s: both measures of profitability for both types of institutions fell sharply after 1985, reaching the lowest point in 1987.1 Both measures of thrift profitability were negative between 1987 and 1992, whereas commercial bank profitability apparently approached zero in certain years. However, by the early 1990s profitability reached a new historical high, only beginning to decline (and dramatically so) in 2006 and 2007 as the current crisis began. It remains to be seen how much lower bank profitability has since fallen, and will still fall, and whether it will recover to pre-crisis levels.

Fig. 2.

Profitability of commercial banks and savings institutions, 1934–2007.

(A) Return on assets (%); (B) return on equity (%).

Source: calculated from FDIC data. ROA (return on assets) measured as net income as a percentage of total assets, and ROE (return on equity) measured as net income as a percentage of total equity capital.

Fig. 2.

Profitability of commercial banks and savings institutions, 1934–2007.

(A) Return on assets (%); (B) return on equity (%).

Source: calculated from FDIC data. ROA (return on assets) measured as net income as a percentage of total assets, and ROE (return on equity) measured as net income as a percentage of total equity capital.

An important consideration regarding profitability trends, underlined by the current crisis, is the extent to which apparent bank profits in the pre-crisis period are in fact ‘real’. The extent of banks’ vulnerability, as exposed through the crisis, certainly raises questions around banks’ ‘real profitability’ in the pre-crisis period.

One aspect of this is that it is not clear to what extent the profits that banks recorded in their balance sheets were manipulated through ‘creative accounting’, with, for example, expected future earnings rather than actual current earnings booked in the current period. The hiding of risks and liabilities off balance sheets would have also affected the veracity of profit data.

A second issue is around the solidity of the foundations of banks’ high profits in the pre-crisis period. The nature and the pricing of banks’ assets, some of which have since been exposed as toxic, means that bank profits were built on a flimsy basis (even if such profits were ‘real’) and were not sustainable in the medium- to long-term, as has been revealed with the breaking of the crisis. This is obviously related to the issue of misrepresentation of banks’ real positions, in that toxic assets were systematically overpriced. Another aspect is the extent to which high profits in the pre-crisis period were based on high leverage, which could have swelled profits but at the expense of dramatically increased vulnerability of banks, as has since become evident.

From the available data it is not possible to ascertain the extent to which banks’ reported profits are in fact genuine; more evidence on this is likely to emerge in the coming period. However, the issues discussed above—both in terms of dubious reporting of profits, and in terms of the flimsy foundations of profits, need to be borne in mind in interpreting data on bank profitability.

Literature review: concentration and profitability in banking

We briefly review the theoretical approaches in the industrial organisation literature concerning the concentration–profitability relationship. This literature falls into two broad approaches: the Market Power (MP) and Efficiency Structure (ES) paradigms. These have very different understandings of the direction of causality between concentration and profitability. In a MP paradigm the direction of causality runs from the market structure of an industry to its behaviour, which affects its performance. A concentrated structure is conducive to the use of market power in ways that may enhance banks’ profitability. An ES paradigm, by contrast, would see causality as running from individual firms’ efficiency to their market share and profitability.

Two distinct approaches can be distinguished within the MP paradigm: the Structure–Conduct Performance (SCP) hypothesis, and the Relative Market Power (RMP) hypothesis. According to the SCP approach, the level of concentration in a banking market gives rise to potential market power by banks, which may raise their profitability. Higher concentration can lower the costs of collusion (explicit or tacit), which gives rise to monopoly rents. Concentration thus affects profitability through firms’ pricing behaviour, with concentrated market structures more conducive to allowing banks to make ‘abnormal’ profits. Whereas the SCP hypothesis would predict generic benefits to banks arising from higher concentration, the RMP hypothesis sees any benefits as accruing to individual banks based on their own market share. According to the latter approach, only large banks can influence prices and increase profits.

There are also two distinct approaches within the ES paradigm: the X-efficiency and scale-efficiency hypotheses. According to the X-efficiency approach, more efficient firms are more profitable because of their lower costs. Such firms tend to gain larger market shares, which may manifest in higher levels of market concentration, but without any causal relationship from concentration to profitability. The scale efficiency approach within the ES paradigm emphasises economies of scale rather than differences in management or production technology. Larger firms can obtain lower unit costs and higher profits through economies of scale. As these firms have higher market shares, which may manifest in higher concentration, there may be an apparent—yet spurious—relationship between concentration and profitability. According to the ES approaches, a positive correlation between concentration and profitability thus need not indicate a causal economic relationship, especially not through market power.

These approaches have distinct policy implications. If high profits can be attributed to high levels of concentration, this might suggest a stronger role for competition policy and other forms of regulation. However, if high profitability is explained by banks’ efficiency, this would not provide support for stronger regulation, and might suggest that competition policy or other types of regulatory interventions intended to reduce concentration or the size of large banks could actually reduce efficiency.

The empirical evidence in the existing literature on the relationship between concentration and profitability is mixed. Smirlock's seminal paper (1985) argues that there is no causal relationship between concentration and profitability. Using data from 2,700 unit state banks operating in a particular region over the period 1973–78, Smirlock finds that once market share is controlled for, concentration does not contribute to explaining bank profit rates.

In an important contribution to this literature, Berger (1995) includes measures of concentration, market share, X-efficiency, as well as scale-efficiency in a single specification in order to test all four hypotheses. He tests this specification using 30 separate cross-sectional datasets. The results show some support for the RMP hypothesis and partial support for the X-efficiency approach. However, Berger notes that neither of these hypotheses are very important in explaining bank profitability, as the respective variables explain little of the variance of profitability.

Nier (2000) uses risk-adjusted measures of profitability in comparing the profitability of European and UK banks, using a panel data set of 63 banks from 13 countries. Nier finds some evidence in support of the effects of concentration on profitability, as well as some support for the product differentiation hypothesis, and for measures of efficiency and factor costs. Jeon and Miller (2002) study the relationship between banking concentration and average bank profitability on a state-by-state basis, and find strong support for a positive relationship between concentration and profitability. In addition, through temporal causality tests they find that bank concentration leads bank profitability. They conclude that their evidence supports the market-power hypothesis. In an overview of the literature, Berger (2004) notes that many studies found that US banks in more concentrated local markets tend to have pricing structures consistent with the exercise of market power under the SCP hypothesis, but that when banks’ market shares were included in the regression equation, there were no longer strong relationships between concentration and profitability. Studies since the early 1990s have attempted to control for X-efficiency and scale efficiency, and have found some—generally weak—support for the effects of both market power and efficiency on profitability.

Most econometric analyses of the relationship between concentration and profitability use either cross-sectional or aggregated time series methods. There is surprisingly little analysis using panel data, particularly given the advantages of these methods. Furthermore, there appears to be limited econometric analysis employing formal tests of endogeneity and appropriate instrumentalisation of key variables. This could be considered particularly important given the nature of the relationships being studied and possible issues of simultaneity. In terms of results, there are no generalisable empirical results that seem to emerge from the literature on the relationship between concentration and profitability among US banks (and certainly not for the recent period that this paper studies).

Empirical analysis

Econometric methods

Before proceeding to the details of the empirical analysis we give a brief overview of the econometric methods used. This is especially relevant in two respects. First, the use of various panel estimation techniques (especially GMM) is an improvement on previous studies, which tend to rely on time-series or cross-sectional analysis. Second, we pay particular attention to issues of potential endogeneity, as this is important not only econometrically but in terms of the substantive issues of the causal relationships between concentration and profitability.

Equations were estimated using four econometric techniques. First, ordinary least squares (OLS) regressions (both static and dynamic) with two-way fixed effects. Three GMM methods were also used, as discussed further below. The use of GMM methods is important in terms of dealing with possible endogeneity of regressors. The endogeneity issue is especially pertinent in this study because of the nature of the economic relationships being investigated. As discussed in the literature review, the direction of causality between concentration and profitability is a critical and contentious question in the literature. The issue of possible simultaneity in regression specifications thus presents a challenge for deriving strong conclusions from regressions that appear to find a causal relationship from concentration to profitability, yet do not take account of possible issues of endogeneity and simultaneity.

In order to deal with this issue, regressors were instrumentalised and tested for endogeneity.2 None of the regressors were found to be endogenous. This was an important finding for two reasons. First, it suggests that (at least on this count) the results from the OLS regressions are not necessarily inconsistent. The OLS results are thus reported in this paper along with the GMM results. Second, in terms of economic implications, it may suggest that if market share and concentration are not endogenous to profitability, then a finding that concentration does appear to raise profitability may be robust and economically significant as opposed to just a spurious correlation or reverse causality.

The finding of no endogeneity was actually to be expected for the concentration variable in particular, given that it is defined for each of the commercial banking and savings institution sectors for any given quarter. The way in which this variable was constructed makes it virtually exogenous to the profitability of any single institution in any given quarter.3 The question of possible endogeneity was thus more relevant to the other regressors.

Despite the finding of no endogeneity of regressors, they were still treated as either endogenous or predetermined and instrumentalised accordingly (with lags of their own levels and differences as discussed further below). Even if not endogenous they could still be predetermined. Furthermore, the Durbin-Wu-Hausman test for endogeneity may not be entirely conclusive, particularly as it is sensitive to choice of instrumental variables. Possible simultaneity is a particularly important issue for this study, so maximum caution was exercised. Instrumentalisation is also integral to the GMM techniques, which are preferred for additional reasons (such as the appropriate treatment of a dynamic structure).

Three different GMM methods were used. The first was a standard two-step efficient GMM estimator (both static and dynamic). Second, the Arellano-Bond estimator was used in implementing ‘difference GMM’. The Arellano-Bond dynamic panel-data estimator is derived using lagged levels of the dependent variable as well as the predetermined and (originally) endogenous variables and differences of the strictly exogenous variables. This method was implemented with both the one- and two-step estimator. The results were generally very similar, but the results reported here for the Arellano-Bond estimations are from the one-step procedures, which are superior to the two-step in terms of inference. The third method used is the Arellano-Bover/Blundell-Bond method of ‘system GMM’, which has superior efficiency. Both the one- and two-step estimators were implemented in system GMM. Further, a finite-sample Windmeijer correction was made to the two-step covariance matrix, which improves accuracy.

Various methods of outlier identification4 were used and the baseline regressions rerun excluding the outliers identified through each method. The results were found to be consistently robust (with minor variations) with all outlier identification methods used. Having confirmed this, the results reported here are from the full dataset.

Data

The data is from Standard and Poor's Bank Compustat dataset, which provides detailed balance sheet and income statement for individual banks. This allows for more sophisticated analysis than would be possible with aggregated time-series data. Results reported here are all from the quarterly dataset.

The period of analysis, 1994–2005, is determined in the first instance by the availability of quarterly data. Nevertheless, this period turns out to be particularly appropriate for the analysis, given that it comes after the atypical plunge in profitability from which banks had recovered by the early 1990s (see Figure 2) and before the downturn in profitability associated with the current financial crisis; distinct dynamics were likely to have been at play during both of these episodes, which it might not be appropriate to generalise.

The following categories of financial institutions are included in the sample: commercial banks; national commercial banks; savings institutions not federally chartered, savings institutions that are federally chartered; state commercial banks. Life assurers were excluded from the sample, as were banks that were classified as American Depository Shares (ADSs) and American Depository Receipts (ADRs), in order to focus the analysis as much as possible on the US domestic banking market.

Banks without a minimum of six consecutive quarters of full data were excluded from the sample. This was done both in order to avoid banks with excessively patchy data, and, in particular, to minimise any unintended changes to the sample when lag structures are introduced. The small number of banks that did not have positive mean profitability over the sample period were also excluded. The sample size, of banks with valid observations for all the variables included in the specification, is 644.

Description of variables

Table 1 summarises each of the variables used in the regressions. Further detail on the construction of variables is provided in Appendix 1 for the key explanatory variables, and summary statistics for all variables are shown in Table A1.

Table 1.

Variables used in the regressions

Variable Meaning Comments 
Dependent variables   
πi,tm Profitability of bank i in quarter t using measure m Profitability is measured using two different measures: πi,tROA is return on assets (ROA) and πi,tROE is return on equity (ROE).* 
Explanatory variables   
concI,tx Index of market concentration in sector I in quarter t using measure x concI,t1 represents an index of the weighted share of the largest 2% of institutions in sector I in each quarter in the total assets of that sector in that quarter, where sector I is either commercial banks or savings institutions. concI,t2 is similarly constructed but represents the largest 5% of each type of institutions in each quarter. 
CRa Standard concentration measure CR10 for the commercial banks and CR3 for savings institutions (given the differential number of banks in these sectors). This is included for comparison purposes. 
MSi,t Market share of bank i in quarter t The share of the aggregate net income of banks for quarter t accounted for by bank i. 
lnTAi,t Size of bank i in quarter t Natural log of the (real) total assets of bank i in quarter t 
(lnTAi,t)2 Size of bank i in quarter t, squared Natural log of the total assets of bank i in quarter t, squared 
OPEFFi,t Measure of the operational inefficiency of bank i in quarter t Ratio of its total other expenses to net income. 
Control variables   
CAi,t Capital asset ratio of bank i in quarter t  
CDTAi,t Cash and dues from banks as a percentage of total assets, for bank i in quarter t  
TIATAi,t Total invested assets as a percentage of total assets, for bank i in quarter t  
PEi,t Price earnings ratio of bank i in quarter t  
Variable Meaning Comments 
Dependent variables   
πi,tm Profitability of bank i in quarter t using measure m Profitability is measured using two different measures: πi,tROA is return on assets (ROA) and πi,tROE is return on equity (ROE).* 
Explanatory variables   
concI,tx Index of market concentration in sector I in quarter t using measure x concI,t1 represents an index of the weighted share of the largest 2% of institutions in sector I in each quarter in the total assets of that sector in that quarter, where sector I is either commercial banks or savings institutions. concI,t2 is similarly constructed but represents the largest 5% of each type of institutions in each quarter. 
CRa Standard concentration measure CR10 for the commercial banks and CR3 for savings institutions (given the differential number of banks in these sectors). This is included for comparison purposes. 
MSi,t Market share of bank i in quarter t The share of the aggregate net income of banks for quarter t accounted for by bank i. 
lnTAi,t Size of bank i in quarter t Natural log of the (real) total assets of bank i in quarter t 
(lnTAi,t)2 Size of bank i in quarter t, squared Natural log of the total assets of bank i in quarter t, squared 
OPEFFi,t Measure of the operational inefficiency of bank i in quarter t Ratio of its total other expenses to net income. 
Control variables   
CAi,t Capital asset ratio of bank i in quarter t  
CDTAi,t Cash and dues from banks as a percentage of total assets, for bank i in quarter t  
TIATAi,t Total invested assets as a percentage of total assets, for bank i in quarter t  
PEi,t Price earnings ratio of bank i in quarter t  
*

Some studies in the literature use one or the other of the measures ROE and ROA, but both were used in this study.

Table A1.

Summary statistics

Variable  Mean SD Min Max Observations 
Dependent variables       
ROE Overall 4.53 3.26 –48.04 60.80 N  = 20,781 
 Between  1.86 0.03 14.45 n  = 644 
 Within  2.70 –45.28 57.85 T  = 32.27 
TGIITA Overall 0.01 0.01 –0.01 0.59 N  = 20,775 
 Between  0.00 0.00 0.03 n  = 644 
 Within  0.01 –0.02 0.56 T  = 32.26 
TGIIE Overall 0.11 0.08 –2.09 1.27 N  = 20,760 
 Between  0.05 0.01 0.36 n  = 644 
 Within  0.06 –2.20 1.12 T  = 32.24 
Explanatory variables       
conc1 Overall 2.56 1.16 1.00 5.63 N  = 20,796 
 Between  0.98 1.93 5.37 n  = 644 
 Within  0.75 –0.72 4.88 T  = 32.29 
conc2 Overall 2.44 1.10 1.00 5.37 N  = 20,796 
 Between  0.94 1.84 5.13 n  = 644 
 Within  0.70 –0.60 4.60 T  = 32.29 
MS Overall 0.07 0.44 –5.54 13.75 N  = 20,794 
 Between  0.30 0.00 4.46 n  = 644 
 Within  0.27 –7.04 9.70 T  = 32.29 
lnTA Overall 6.94 1.61 2.66 13.90 N  = 20,796 
 Between  1.47 4.00 13.59 n  = 644 
 Within  0.39 4.14 9.42 T  = 32.29 
lnTA2 Overall 50.77 25.82 7.08 193.33 N  = 20,796 
 Between  23.13 16.02 184.60 n  = 644 
 Within  5.92 11.42 85.33 T  = 32.29 
OPEFF Overall 372.01 2340.26 –68158.82 142438.50 N  = 20,773 
 Between  519.12 –4401.15 4021.28 n  = 644 
 Within  2292.57 –66714.57 139277.90 T  = 32.26 
Control variables       
CA Overall 9.61 3.85 –1.33 97.78 N  = 20,781 
 Between  3.18 3.34 31.26 n  = 644 
 Within  2.38 –18.27 95.66 T  = 32.27 
CDTA Overall 4.19 3.09 –0.13 36.26 N  = 20,777 
 Between  2.51 0.48 30.27 n  = 644 
 Within  1.85 –6.45 28.04 T  = 32.26 
TIATA Overall 89.63 7.49 0.00 98.75 N  = 20,796 
 Between  6.06 19.75 97.96 N  = 644 
 Within  4.40 –1.51 137.16 T  = 32.29 
PE Overall 60.67 105.60 –2951.00 3031.00 N  = 20,148 
 Between  53.02 –625.00 874.48 n  = 644 
 Within  100.34 –2874.05 2757.87 T  = 31.29 
Variable  Mean SD Min Max Observations 
Dependent variables       
ROE Overall 4.53 3.26 –48.04 60.80 N  = 20,781 
 Between  1.86 0.03 14.45 n  = 644 
 Within  2.70 –45.28 57.85 T  = 32.27 
TGIITA Overall 0.01 0.01 –0.01 0.59 N  = 20,775 
 Between  0.00 0.00 0.03 n  = 644 
 Within  0.01 –0.02 0.56 T  = 32.26 
TGIIE Overall 0.11 0.08 –2.09 1.27 N  = 20,760 
 Between  0.05 0.01 0.36 n  = 644 
 Within  0.06 –2.20 1.12 T  = 32.24 
Explanatory variables       
conc1 Overall 2.56 1.16 1.00 5.63 N  = 20,796 
 Between  0.98 1.93 5.37 n  = 644 
 Within  0.75 –0.72 4.88 T  = 32.29 
conc2 Overall 2.44 1.10 1.00 5.37 N  = 20,796 
 Between  0.94 1.84 5.13 n  = 644 
 Within  0.70 –0.60 4.60 T  = 32.29 
MS Overall 0.07 0.44 –5.54 13.75 N  = 20,794 
 Between  0.30 0.00 4.46 n  = 644 
 Within  0.27 –7.04 9.70 T  = 32.29 
lnTA Overall 6.94 1.61 2.66 13.90 N  = 20,796 
 Between  1.47 4.00 13.59 n  = 644 
 Within  0.39 4.14 9.42 T  = 32.29 
lnTA2 Overall 50.77 25.82 7.08 193.33 N  = 20,796 
 Between  23.13 16.02 184.60 n  = 644 
 Within  5.92 11.42 85.33 T  = 32.29 
OPEFF Overall 372.01 2340.26 –68158.82 142438.50 N  = 20,773 
 Between  519.12 –4401.15 4021.28 n  = 644 
 Within  2292.57 –66714.57 139277.90 T  = 32.26 
Control variables       
CA Overall 9.61 3.85 –1.33 97.78 N  = 20,781 
 Between  3.18 3.34 31.26 n  = 644 
 Within  2.38 –18.27 95.66 T  = 32.27 
CDTA Overall 4.19 3.09 –0.13 36.26 N  = 20,777 
 Between  2.51 0.48 30.27 n  = 644 
 Within  1.85 –6.45 28.04 T  = 32.26 
TIATA Overall 89.63 7.49 0.00 98.75 N  = 20,796 
 Between  6.06 19.75 97.96 N  = 644 
 Within  4.40 –1.51 137.16 T  = 32.29 
PE Overall 60.67 105.60 –2951.00 3031.00 N  = 20,148 
 Between  53.02 –625.00 874.48 n  = 644 
 Within  100.34 –2874.05 2757.87 T  = 31.29 

Note: see Table 1 for explanations of variables. SD, standard deviation.

Specification and interpretation

The baseline specification is as follows: 

graphic
This regresses profitability on five explanatory variables of interest, which are the concentration index (‘conc1’ or ‘conc2’ in separate regressions), market share, log of total assets, the quadratic of the log of total assets, and the measure of operational inefficiency; and four additional ‘control’ variables, which are the capital-asset ratio, ratio of cash and due from banks to total assets, the ratio of total invested assets to total assets, and the price–earnings ratio, as well as two-way fixed effects by bank and by time.5

Two alternative measures of concentration are used as the explanatory variable of primary interest. A detailed explanation of the construction of these indices is in Appendix 1. ‘conc1’ measures concentration among the largest 2% of commercial banks and savings institutions, respectively, that are in the sample, while ‘conc2’ measures the degree of concentration among the largest 5%. The concentration indices were constructed in a new and arguably superior way to the measures commonly used in such studies. One advantage of the new proposed measures is that they give a better sense of the depth and intensity of concentration than the CRx measures typically used in the literature. The concentration indices incorporate the asset shares of each of the banks within the top 2% or 5%, respectively, weighted so that the largest banks within the largest 2% or 5% count more, rather than simply the cumulative asset share of the top x banks as in CRx measures.

The primary purpose of this specification is to test the relationship between concentration and profitability. The specification is designed to simultaneously test the four competing hypotheses in the literature—the SCP, RMP, scale-efficiency and X-efficiency hypotheses—by nesting all four hypothesis in the specification simultaneously. Following Berger (1995), this seeks to avoid ambiguity as to whether an apparent relationship between concentration and profitability is simply indicative of another underlying relationship. This specification thus allows stronger conclusions to be drawn about any causal relationship between concentration and profitability.

In summary, the SCP hypothesis would predict a positive and significant coefficient on ‘conc’, but would not be associated with explicit predictions concerning the other explanatory variables. The RMP hypothesis would predict a positive and significant coefficient on the ‘MS’ (market share) variable, and does not require any particular results in terms of the other variables. The scale efficiency hypothesis would predict a positive and significant coefficient on ‘lnTA’ (natural log of the total assets of the bank), would not predict a negative (and significant) coefficient on lnTA2 (at least not at a low turning point), and would predict that positive and significant coefficients would not be found on ‘conc’ or on MS. The X-efficiency hypothesis would predict a negative and significant coefficient on ‘OPEFF’ (note that the way that this variable is constructed actually measures operational inefficiency of a bank) and would predict that that the coefficients on conc and MS would not be positive and significant. The latter two hypotheses predict that including the relevant variables in the specification (size in the case of the scale efficiency hypothesis and operational efficiency in the case of the X-efficiency hypothesis) would collapse any apparent relationship between concentration and either profitability or bank market shares.

This specification is used first to test the relationship between each of concentration, market share, size and operational efficiency for the banking sector as a whole. Thereafter, it is re-estimated excluding the largest banks (which are included in the concentration indices), in order to test the effects of concentration at the top on the rest of the banking sector.

Results from the baseline specification

As discussed in Section 4.1, various estimation techniques were used. The results shown here (Tables 2345678) are those using two-step efficient static GMM, one-step Arrellano-Bond dynamic GMM, and two-step Arrellano-Bover/Blundell-Bond dynamic GMM. Furthermore, only the results for return on equity (ROE) as the dependent variable are shown; the results with return on assets (ROA) are consistent with these. (Full results are available from the author on request.)

Table 2.

Baseline specification with two-step efficient GMM

Depvar ROE ROE 
conc1 0.3002 (0.000)  
conc2  0.3242 (0.000) 
MS 0.1312 (0.27) 0.1315 (0.27) 
lnTA 1.0071 (0.034) 1.0105 (0.033) 
lnTA2 –0.078 (0.010) –0.0781 (0.010) 
OPEFF –0.0012 (0.031) –0.0012 (0.032) 
CA –0.2518 (0.00) –0.2517 (0.00) 
CDTA –0.0306 (0.37) –0.0297 (0.39) 
TIATA –0.0206 (0.32) –0.0205 (0.32) 
PE 0.0141 (0.002) 0.0141 (0.002) 
Observations 17,118 17,118 
No. of banks 627 627 
Depvar ROE ROE 
conc1 0.3002 (0.000)  
conc2  0.3242 (0.000) 
MS 0.1312 (0.27) 0.1315 (0.27) 
lnTA 1.0071 (0.034) 1.0105 (0.033) 
lnTA2 –0.078 (0.010) –0.0781 (0.010) 
OPEFF –0.0012 (0.031) –0.0012 (0.032) 
CA –0.2518 (0.00) –0.2517 (0.00) 
CDTA –0.0306 (0.37) –0.0297 (0.39) 
TIATA –0.0206 (0.32) –0.0205 (0.32) 
PE 0.0141 (0.002) 0.0141 (0.002) 
Observations 17,118 17,118 
No. of banks 627 627 

Notes: Robust P values in parentheses; see Table 1 for explanations of variables; where the number of banks varies between specifications, this is due to lag structures—the sample is, however, standard across specifications.

Table 3.

Baseline specification with one-step Arellano-Bond dynamic GMM

Depvar ROE ROE 
LD.depvar 0.0642 (0.008) 0.0644 (0.0080) 
D.conc1 0.6375 (0.000)  
D.conc2  0.7002 (0.000) 
D.MS 1.0114 (0.000) 1.0117 (0.000) 
D.lnTA 3.5096 (0.000) 3.489 (0.000) 
D.lnTA2 –0.3357 (0.000) –0.3357 (0.000) 
D.OPEFF 0 (0.74) 0 (0.76) 
D.CA –0.0334 (0.24) –0.0328 (0.25) 
D.CDTA 0.0005 (0.99) 0.0008 (0.98) 
D.TIATA 0.0108 (0.76) 0.0112 (0.75) 
D.PE –0.0003 (0.62) –0.0002 (0.64) 
Observations 18,756 18,756 
No of banks 640 640 
Depvar ROE ROE 
LD.depvar 0.0642 (0.008) 0.0644 (0.0080) 
D.conc1 0.6375 (0.000)  
D.conc2  0.7002 (0.000) 
D.MS 1.0114 (0.000) 1.0117 (0.000) 
D.lnTA 3.5096 (0.000) 3.489 (0.000) 
D.lnTA2 –0.3357 (0.000) –0.3357 (0.000) 
D.OPEFF 0 (0.74) 0 (0.76) 
D.CA –0.0334 (0.24) –0.0328 (0.25) 
D.CDTA 0.0005 (0.99) 0.0008 (0.98) 
D.TIATA 0.0108 (0.76) 0.0112 (0.75) 
D.PE –0.0003 (0.62) –0.0002 (0.64) 
Observations 18,756 18,756 
No of banks 640 640 

Notes: see footnote of table 2.

Table 4.

Baseline specification with two-step Arellano-Bover/Blundell-Bond dynamic GMM

Depvar ROE ROE 
L.depvar 0.1375 (0.000) 0.1377 (0.000) 
conc1 0.5076 (0.005)  
conc2  0.5144 (0.007) 
MS 1.0281 (0.000) 1.0303 (0.000) 
lnTA 4.4489 (0.000) 4.4949 (0.000) 
lnTA2 –0.27 (0.000) –0.2731 (0.000) 
OPEFF 0.0002 (0.036) 0.0002 (0.037) 
CA –0.0377 (0.20) –0.0363 (0.21) 
CDTA –0.0275 (0.49) –0.0247 (0.53) 
TIATA –0.0171 (0.60) –0.0163 (0.61) 
PE –0.0015 (0.050) –0.0014 (0.054) 
Constant –13.2241 (0.000) –13.4113 (0.000) 
Observations 19,510 19,510 
No. of banks 644 644 
Depvar ROE ROE 
L.depvar 0.1375 (0.000) 0.1377 (0.000) 
conc1 0.5076 (0.005)  
conc2  0.5144 (0.007) 
MS 1.0281 (0.000) 1.0303 (0.000) 
lnTA 4.4489 (0.000) 4.4949 (0.000) 
lnTA2 –0.27 (0.000) –0.2731 (0.000) 
OPEFF 0.0002 (0.036) 0.0002 (0.037) 
CA –0.0377 (0.20) –0.0363 (0.21) 
CDTA –0.0275 (0.49) –0.0247 (0.53) 
TIATA –0.0171 (0.60) –0.0163 (0.61) 
PE –0.0015 (0.050) –0.0014 (0.054) 
Constant –13.2241 (0.000) –13.4113 (0.000) 
Observations 19,510 19,510 
No. of banks 644 644 

Notes: see footnote of table 2.

Table 5.

Economic significance of baseline results

Depvar Conc index Estimated coefficient on concentration Effect of 5% increase in concentration on profitability Increase in profitability as % mean profitability Effect of 10% increase in concentration on profitability Increase in profitability as % mean profitability 
ROE conc1 0.5076 0.02538 0.560 0.05076 1.120 
ROE conc2 0.5144 0.02572 0.568 0.05144 1.135 
Depvar Conc index Estimated coefficient on concentration Effect of 5% increase in concentration on profitability Increase in profitability as % mean profitability Effect of 10% increase in concentration on profitability Increase in profitability as % mean profitability 
ROE conc1 0.5076 0.02538 0.560 0.05076 1.120 
ROE conc2 0.5144 0.02572 0.568 0.05144 1.135 
Table 6.

Baseline specification excluding banks included in the concentration indices, with two-step efficient GMM

Depvar ROE ROE 
conc1 0.2636 (0.001)  
conc2  0.2274 (0.004) 
MS 0.095 (0.85) 5.0591 (0.003) 
lnTA 0.6721 (0.16) 1.1517 (0.037) 
lnTA2 –0.0543 (0.071) –0.0888 (0.015) 
OPEFF –0.0016 (0.0021) –0.0013 (0.001) 
CA –0.2689 (0.000) –0.2675 (0.000) 
CDTA –0.0211 (0.57) 0.0131 (0.73) 
TIATA –0.0176 (0.43) 0.0067 (0.77) 
PE 0.0153 (0.000) 0.0101 (0.002) 
Observations 16,595 15,730 
No. of banks 617 597 
Depvar ROE ROE 
conc1 0.2636 (0.001)  
conc2  0.2274 (0.004) 
MS 0.095 (0.85) 5.0591 (0.003) 
lnTA 0.6721 (0.16) 1.1517 (0.037) 
lnTA2 –0.0543 (0.071) –0.0888 (0.015) 
OPEFF –0.0016 (0.0021) –0.0013 (0.001) 
CA –0.2689 (0.000) –0.2675 (0.000) 
CDTA –0.0211 (0.57) 0.0131 (0.73) 
TIATA –0.0176 (0.43) 0.0067 (0.77) 
PE 0.0153 (0.000) 0.0101 (0.002) 
Observations 16,595 15,730 
No. of banks 617 597 

Notes: see footnote of table 2.

Table 7.

Baseline specification excluding banks included in the concentration indices, with one-step Arellano-Bond dynamic GMM

Depvar ROE ROE 
LD.ROA 0.0503 (0.046) 0.0202 (0.45) 
D.conc1 0.7095 (0.000)  
D.conc2  0.8173 (0.000) 
D.MS 5.2581 (0.000) 20.6324 (0.000) 
D.lnTA 1.6675 (0.12) 4.7055 (0.001) 
D.lnTA2 –0.219 (0.008) –0.4669 (0.000) 
D.OPEFF 0 (0.78) 0 (0.73) 
D.CA –0.0577 (0.056) –0.0494 (0.17) 
D.CDTA –0.0067 (0.87) 0.0111 (0.82) 
D.TIATA 0.0139 (0.71) 0.0098 (0.82) 
D.PE –0.0002 (0.68) –0.0001 (0.78) 
Observations 18,198 17,275 
No. of banks 631 610 
Depvar ROE ROE 
LD.ROA 0.0503 (0.046) 0.0202 (0.45) 
D.conc1 0.7095 (0.000)  
D.conc2  0.8173 (0.000) 
D.MS 5.2581 (0.000) 20.6324 (0.000) 
D.lnTA 1.6675 (0.12) 4.7055 (0.001) 
D.lnTA2 –0.219 (0.008) –0.4669 (0.000) 
D.OPEFF 0 (0.78) 0 (0.73) 
D.CA –0.0577 (0.056) –0.0494 (0.17) 
D.CDTA –0.0067 (0.87) 0.0111 (0.82) 
D.TIATA 0.0139 (0.71) 0.0098 (0.82) 
D.PE –0.0002 (0.68) –0.0001 (0.78) 
Observations 18,198 17,275 
No. of banks 631 610 

Notes: see footnote of table 2.

Table 8.

Baseline specification excluding banks included in the concentration indices, with two-step Arellano-Bover/Blundell-Bond dynamic GMM

Depvar ROE ROE 
L.ROA 0.1192 (0.000) 0.1125 (0.000) 
conc1 0.4513 (0.013)  
conc2  0.4766 (0.010) 
MS 4.5576 (0.001) 25.0322 (0.000) 
lnTA 3.216 (0.060) 6.9331 (0.002) 
lnTA2 –0.1855 (0.13) –0.4906 (0.004) 
OPEFF 0.0002 (0.061) 0.0002 (0.063) 
CA –0.0418 (0.25) –0.035 (0.51) 
CDTA –0.0401 (0.25) –0.0124 (0.73) 
TIATA –0.0127 (0.68) –0.0194 (0.50) 
PE –0.0013 (0.081) –0.0013 (0.10) 
Constant –9.1089 (0.11) –19.7064 (0.005) 
Observations 18,934 17,984 
No. of banks 635 614 
Depvar ROE ROE 
L.ROA 0.1192 (0.000) 0.1125 (0.000) 
conc1 0.4513 (0.013)  
conc2  0.4766 (0.010) 
MS 4.5576 (0.001) 25.0322 (0.000) 
lnTA 3.216 (0.060) 6.9331 (0.002) 
lnTA2 –0.1855 (0.13) –0.4906 (0.004) 
OPEFF 0.0002 (0.061) 0.0002 (0.063) 
CA –0.0418 (0.25) –0.035 (0.51) 
CDTA –0.0401 (0.25) –0.0124 (0.73) 
TIATA –0.0127 (0.68) –0.0194 (0.50) 
PE –0.0013 (0.081) –0.0013 (0.10) 
Constant –9.1089 (0.11) –19.7064 (0.005) 
Observations 18,934 17,984 
No. of banks 635 614 

Notes: see footnote of table 2.

Tables 2, 3 and 4 show results for the baseline specification. Both measures of concentration have positive and highly statistically significant estimated coefficients in all specifications, including those not reported here. That is, this finding holds with both measures of profitability (ROA and ROE) as the dependent variable, with both measures of concentration, and in static and dynamic specifications. The coefficients on conc2 (the 5% measure of concentration) tend to be slightly higher than on conc1 (the narrower 2% measure), as might be expected. This suggests that the market structure of the banking sector—specifically, the degree of concentration at the top end of the banking sector—has effects in its own right on profitability, not merely through channels such as economies of scale, nor as a common outcome of an underlying determinant such as market share. Part of banks’ very high profits in the pre-crisis period can thus be regarded as rents deriving from the increasingly concentrated structure of the industry.

The estimated coefficients on MS are also positive and mostly statistically significant, possibly pointing to some market power effects on profitability. This makes sense, as it indicates that banks benefit from their own market shares in addition to the overall benefits deriving from concentration.

The estimated coefficients on lnTA are consistently positive and statistically significant, while those on lnTA2 are consistently negative and statistically significant. This is an interesting result, which suggests that banks do enjoy economies of scale but only up to a certain point, with the range including about 80% of banks in the sample. The fact that the largest banks are above a size which would appear to be ‘optimal’ in terms of economies of scale suggests that considerations other than optimal performance—perhaps such as empire-building and incentive structures—have contributed to the consolidation of giant banks, and this underlines the need for stronger regulation of banks as part of a post-crisis regulatory regime.

The estimated coefficients on operational (in)efficiency (expected to be negative) are marginally negative (in some cases appearing as zero in the results due to rounding of decimal points) and mostly statistically significant (although only at a 5% level). This suggests that the degree of operational efficiency explains little of the variation in profitability across banks and across time. The weak results for the effects of bank efficiency on profitability might be interpreted as suggesting that banks did not really ‘earn’ their particularly high profits in the pre-crisis period, and that factors other than ‘efficiency’ explain these profits. This could have implications for policy approaches to banks in the current crisis, for instance in terms of the level of profits to which banks should be returned or allowed to return.

One of the control variables included is the capital–asset ratio. This is an indicator of the degree of leverage (with leverage being higher the lower is this ratio). The results show a negative relationship between the capital–asset ratio and bank profitability, as would be expected. Given that leverage is also an indication of banks’ vulnerability, this suggests that banks’ high profits during the pre-crisis period could have come at the expense of increased vulnerability. An implication of this could be that banks’ profits in this period are not in any sense ‘normal’ (even in an era of financialised capitalism), and that there should not be an expectation of these rates being restored. Regulatory measures to avoid such high leverage in future may come at the expense of bank profitability (at least in the short- to medium-term), but this could be justifiable in terms of the stability of the financial sector.

The baseline specification was also rerun with the CRx measures typically used in the literature. A CR10 measure was used for commercial banks and a CR3 measure for savings institutions (given the smaller number of the latter in both the sample and population). Both the estimated coefficients and levels of significance are very close to those for the baseline results using conc1 and conc2.6 This verifies that the findings obtained in this study are not simply due to the new measures of concentration used.

The R2 statistics (from the OLS specifications) are generally rather low, as are found in the related literature. The inclusion of various other control variables in alternative specifications (results not reported here) did raise the R2 somewhat, but this had the disadvantage of drastically reducing the sample due to poorer coverage of these variables, and probably biasing the sample as well. Nonetheless, the main findings concerning the relationship between concentration and profitability were robust to re-estimation using a fuller specification and a much reduced sample.

Interpreting the economic significance of the results is relevant to ascertaining the strength of the relationship between concentration and profitability and the degree of importance of any implications for the real economy. The economic significance of the results can be examined in terms of the hypothetical increases in profitability that could be associated with increases in concentration. Given the way in which the concentration indices are constructed, were each bank included in the concentration indices (that is, the largest 2% of banks in each quarter for conc1 and the largest 5% for conc2) to increase their asset share by x% of their current share, the concentration index would rise by x%. (For example, a 5% rise in asset shares would mean, for the first quarter of 1994, the largest bank's asset share increasing from 6.27% to 6.59%, or that of the 20th largest bank rising from 0.32% to 0.34%, and the concentration indices rising by 5%.) The effects of a small change in the concentration indices on profitability can thus be projected for illustrative purposes. These results are summarised in Table 5, for two-step Arellano-Bover/Blundell-Bond dynamic GMM. Based on the estimated coefficients on conc1 and conc2, the table shows the increase in profitability that would be associated with a 5% or 10% increase in the asset share of each institution included in the concentration indices. The increases in profitability are also given as a percentage of mean ROE, to give a sense of their relative magnitude. For example, a 10% increase in the asset share of the largest 2% of institutions (as a percentage of their actual asset share in each quarter) would mean a 10% rise in conc1, which would be associated with increases in ROA and ROE equal to 1.18% and 1.12%, respectively, of mean profitability. In economic terms, these effects are not particularly high. They do appear credible: given the range of determinants of an individual bank's profitability, a sector-wide variable such as the level of concentration would probably not be expected to have a very high absolute effect. The increases in profitability associated with increases in concentration are broadly similar for ROA. The effects are slightly higher for conc2 (the broader concentration index) than for conc1, which makes sense given the higher number of banks included in the former index.

The effects of concentration at the top on profitability of other banks

We now investigate whether the positive overall relationship between concentration and profitability is just operating through increased profitability of the top banks or, alternatively, whether concentration has positive effects on profitability of the banking sector as a whole. One objective of this exercise is to further verify that the apparent relationship running from concentration to profitability does not derive simply from a correlation between large banks being more profitable and having a large market share, showing up in a higher concentration index, without any necessary structural relationship between concentration and overall profitability. Of broader interest and importance, this investigates empirically the extent to which the benefits of concentration for the profitability of the largest banks comes at the expense of the rest of the banking sector, and to what extent this concentration also enhances the profitability of the rest of the sector in a generalised way.

The relationship between concentration amongst the larger banks and the profitability of the rest of the banking sector is more complex than the relationship between concentration and profitability in the banking sector overall. The extent to which concentration among the top banks is favourable for the rest of the banks is interesting both analytically and in terms of policy, yet it has not been addressed in the existing literature on the relationship between concentration and profitability.

Concentration at the top is likely to affect profitability in the rest of the banking sector in multiple ways. To the extent that the largest banks benefit from concentration through intra-sectoral channels, this would come at the expense of the rest of the sector, as a form of ‘redistribution’ of profits within the banking sector. To the extent that concentration of the top banks increases their overall power vis-à-vis non-banking entities and this allows the largest banks to extract higher profits, other banks may benefit from this. More specifically, if collusive behaviour among the largest banks facilitates a pricing structure with higher interest rate spreads, and hence higher profitability, it might, as a by-product, allow smaller non-colluding banks also to have higher interest rate spreads and hence higher profitability than would otherwise be the case. Insofar as the largest banks act as price-leaders for the rest of the banking sector (or at least for the next stratum of banks), concentration among the large banks might be expected to positively affect the profitability of the rest through ‘trickle-down’ effects.

Furthermore, in a more general sense, a highly concentrated top end of the banking market may render the sector as a whole more economically and politically powerful, with benefits in terms of a more favourable legislative and regulatory environment. We discussed earlier the deregulation of the banking sector over a period of time preceding the crisis. Whether such effects would be limited to the top banks, to particular categories of large banks, or generic to the banking sector as a whole, would be conjuncturally contingent.

The overall relationship between concentration at the top and the profitability of the rest of the banking sector is thus a priori indeterminate. This is important to understanding the extent to which any positive effects of concentration on profitability accrue simply to the largest banks, or are generalised to the sector as a whole.

In order to investigate the relationship between concentration among the largest banks and profitability of the rest of banks, the baseline specification is thus re-estimated excluding those banks whose asset shares are included in the construction of the concentration indices. That is, excluding the largest 2% of each type of institution in each quarter in the case of conc1 and the largest 5% for conc2. Selected results are shown in Tables 6, 7 and 8.

An important finding is that the coefficient on concentration remains positive and highly statistically significant in all 16 estimations (i.e. for both profitability measures, both concentration indices and all four estimation techniques). This strongly suggests that the overall positive relationship between concentration and profitability is not just a reflection of the positive effects of concentration at the top on the profitability of the largest banks, but that concentration positively affects the profitability of the rest of the banking sector.

This is a particularly striking result given the indeterminate and possibly contradictory effects of higher concentration at the top on the rest of the sector. Concentration at the top may well also facilitate an intra-sectoral ‘redistribution’ of profits within the banking sector in favour of the largest banks. In fact, such dynamics are likely to be present, particularly given that the size of the estimated coefficients is weaker when the largest banks are excluded from the sample. However, the results suggest that any negative intra-sectoral effects of concentration at the top on the rest of banks are outweighed by positive ‘trickle-down’ effects of the benefits of concentration.

This result also provides strong support for the SCP hypothesis over the RMP hypothesis. If the concentration–profitability relationship were operating through the advantages accruing directly to large banks through their own individual market shares (as per the RMP hypothesis), the apparent relationship between concentration and profitability would be expected to fall away when the large banks, from whose asset shares the concentration indices are calculated, are excluded from the sample. The finding of a robust positive relationship between concentration and profitability even when these banks are excluded points strongly towards a generalised structural relationship between concentration and profitability.

It also suggests that the effects of bank concentration on bank profitability come at the expense of non-bank entities (as opposed to smaller banks). This is a significant finding, particularly in the context of the result that operational efficiency is not particularly important in explaining bank profitability. It suggests that high levels of bank concentration could be detrimental to the ‘real economy’. The ‘net rents’ arising from concentration in the banking sector would essentially come at a cost to the non-banking sectors of the economy, through lower rates on deposits and/or higher lending rates and/or higher fees and changes. This could thus be expected to have negative effects on the ‘real economy’, in terms of investment and growth. These types of effects would need to be taken into account in developing a new and stronger regulatory regime for the banking sector. Furthermore, the fact that banks apparently did not even utilise these rents to buttress their own capital base in the period preceding the crisis, instead squandering these rents in dividends and exorbitant salaries at the top, makes the finding that concentration facilitated these rents (at least to some extent) at the expense of the economy even more problematic.

Conclusion

US banks enjoyed very high profits from the early-mid 1990s until the beginnings of the current crisis. It is striking that this crisis comes on the back of a period of profitability that is unmatched historically. It is now becoming increasing clear that these profits were built on very shaky grounds and were unsustainable, in terms of inter alia the real underlying quality of banks’ assets and in terms of banks’ degree of leverage. Furthermore, the high profits that banks enjoyed during this period were apparently not used to bolster their capital base, which might have allowed the present crisis to be avoided or at least mitigated.

This paper has investigated the effects of structure on profitability of banks during this period of the ‘fat years’. The crisis is likely to lead to an increase in bank concentration, particularly through consolidation through mergers and acquisitions, at least unless there is active intervention to mitigate such a trend. This makes an analysis of the relationship between concentration and profitability especially relevant.

A positive and highly significant relationship is found between concentration and profitability, even with the inclusion of regressors associated with banks’ individual market share, size and operational efficiency. This result is robust to various econometric techniques employed, to two alternative measures of concentration, as well as to both measures of profitability specified as the dependent variable (ROA and ROE). This supports the SCP hypothesis of a causal relationship between overall concentration and profitability. The implication is that bank concentration raises profitability in a structural way—rather than simply as an outcome of banks’ individual market power associated with their own market shares, or with economies of scale or the benefits associated with higher operational efficiency simply manifesting in higher concentration. This is important to understanding the sources of bank profits in the pre-crisis decade or so.

Some support is also found for the RMP hypothesis. However, the result of a relationship between concentration at the top and the profitability of the rest of the banking sector could be interpreted as going against the RMP hypothesis. Economies of scale are found over a range that includes most banks in the sample. However, the scale-efficiency hypothesis is not valid for the sample, as the concentration–profitability relationship holds even with bank size included in the specification.

The results do not support the X-efficiency hypothesis. While coefficients on the measure of operational efficiency are generally of the expected sign, they are very small in magnitude and are not always statistically significant. This has important implications as it suggests that banks did not ‘earn’ their very high profits during the pre-crisis period through being more efficiently run, but rather through appropriating rents arising from a highly concentrated structure. This finding could be interpreted as having policy implications, in terms of the treatment of banks that find themselves in trouble in the current financial crisis, specifically in terms of the extent to which banks’ previously high profits were ‘justified’ and deserve to be restored. A similar conclusion could of course be reached on other grounds, but the results from this analysis provide one basis for such an argument.

This conclusion that the relationship between concentration and profitability operates in a structural way is bolstered by the persistence of a positive relationship between concentration and profitability even when those banks whose asset shares are included in the construction of the concentration indices were excluded. This relationship has not been previously investigated in the literature; the method used here is a new way of analysing the way in which concentration affects profitability, and the findings contribute to our understanding of this relationship. The results suggest that the concentration–profitability relationship operates not simply through higher profits of those dominant banks, but that the payoffs of concentration benefit the bank sector more broadly. This finding thus provides strong support for a generalised structural and causal relationship between bank concentration and profitability, and for the SCP over the RMP hypothesis. They could also suggest that the enhanced bank profits associated with higher bank concentration come primarily at the expense of non-banking sectors rather than ‘redistribution’ within the banking sector.

A positive causal relationship between bank concentration and bank profitability—particularly insofar as the banking sector as a whole benefits from high levels of concentration—could have significant implications for the real economy. The implication is that high concentration in the banking sector is not harmless. The expectation is that bank concentration raises bank profitability through the pricing structure, on one or both sides of banks’ balance sheets. Should higher bank profits arise from higher interest rates being charged on loans, this would be expected to depress investment and growth. Insofar as bank profits are raised through lower interest rates being paid on deposits, this could reduce the rate of savings. This suggests that concentration in the banking sector could potentially exacerbate credit rationing and constrain accumulation and growth. To the extent that higher bank profits are essentially ‘rents’ deriving from high levels of concentration (as opposed to being based on better operational efficiency), these higher bank profits could be seen as a drain on the ‘real economy’.

These results point to the need for stronger regulation of concentration in the banking sector, whether through more aggressive competition policy or other instruments. Furthermore, the evidence that the issue is the level of concentration in the sector could be taken as supportive of pre-emptive regulatory interventions based on structural conditions, as opposed to interventions requiring a behavioural trigger. Typical arguments advanced as to the benefits of concentration in the industrial sector—such as enhanced competitiveness, facilitation of innovation, and economies of scale—do not necessarily apply to the banking sector, particular as it appears that economies of scale in banking only prevail up to a certain size (which is below that of the large banks in the sample).

Related to this, it is interesting that the current financial crisis has come on the back of a period of unprecedented bank profitability. Even these ‘fat years’ apparently did not set banks up in solid enough positions to be able to weather subsequent problems. If banks could enjoy a run of historically unparalleled profitability yet still succumb to crisis and collapse thereafter, this makes a case against arguments that banks should be allowed to be more profitable for the sake of the stability of the financial sector. It also highlights this period as a missed opportunity for banks, and a fundamental omission by regulatory authorities to let them get away with this, with the consequences currently being felt far beyond the banking sector (and beyond the USA). This also draws attention to the problems with focusing any rescue and support packages on restoring levels of bank profitability without addressing underlying issues of banks’ capital base as well as the structure of the sector.

I would like to thank four anonymous referees, Hashem Pesaran, Gabriel Palma and Donald Robertson for helpful suggestions and comments, and in particular James Crotty for invaluable advice on an earlier version of this paper; any errors or omissions are my own.

Appendix 1: More information on selected variables

This section provides greater detail on the construction of certain variables.

Concentration (concI,tx)

concI,t1 is an index of the weighted share of the largest 2% of institutions in sector I (by quarter) of aggregate total assets of sector I, and similarly for the largest 5% of each type of institution in each quarter. These are new measures of concentration proposed in this paper. The specific institutions included in each measure thus vary by quarter. The two sectors are commercial banks and savings institutions. This division was necessary in order to avoid collinearity problems arising from a variable being uniform across banks for every period, in models with time-fixed effects.7 The measures of aggregate total assets are derived from the FDIC statistics of the entire sectors, rather than the sample in the Bank Compustat data.

The indices were calculated as follows: 

graphic
Where ASi,t is the assets of institution i in quarter t as a share of the aggregate (FDIC-reported) assets for the relevant sector (commercial banks or savings institutions) in that quarter; and N is the number of institutions in the largest two percentiles of institutions of sector i in the case of concI,t1, or the largest five percentiles in the case of concI,t2.

The weighting in the concentration indices is thus that the asset share of the largest institution in each quarter is counted N times, the second largest institution counted (N–1) times, and so on, such that the asset share of the Nth largest institution is counted once.

concI,t2 is constructed in the same way but is broader, as it includes the asset shares of the largest 5% of institutions.

This is a new way of measuring concentration. It improves on the measures typically used in the literature (such as CR1, CR3, CR5, C7 or CR10), which do not give a sense of the depth or intensity of concentration among the n banks. Such measures are also unduly sensitive to changes such as mergers and acquisitions at the threshold of n banks.

Market share (MSi,t)

Note that this denotes the share of institution i in the aggregate net income as reported by the FDIC for quarter t, rather than the total net income of institutions in the Bank Compustat dataset.

Size (lnTAi,t and (lnTAi,t)2)

Total assets were deflated (using a quarterly GDP deflator) and then the natural log taken, and the measure squared in the case of (lnTAi,t)2.

Operational inefficiency (OPEFFi,t)

This is the ratio of an institution's total other expenses [including salaries and wages of officers and employees, pension and employment benefits, (net) occupancy expense of bank premises, total costs of furniture and equipment and other current operating expenses] to net income in quarter t.

Appendix 2: Trends in bank concentration

The chart shown in Figure A1 plots the concentration indices used in the econometric analysis.

Fig A1.

Concentration in the banking sector, 1994–2005.

Fig A1.

Concentration in the banking sector, 1994–2005.

References

Berger
AN
The profit-structure relationship in banking—tests of market-power and efficient-structure hypotheses
Journal of Money, Credit and Banking
 , 
1995
, vol. 
27
 
2
(pg. 
404
-
31
)
Berger
AN
Bank concentration and competition: an evolution in the making
Journal of Money, Credit and Banking
 , 
2004
, vol. 
36
 
3
(pg. 
433
-
51
)
Crotty
J
‘If Financial Market Competition is so Intense, Why are Financial Firm Profits so High? Reflections on the Current “Golden Age” of Finance’
2007
 
Political Economy Research Institute Working Papers No. 134
Federal Deposit Insurance Corporation
History of the Eighties—Lessons for the Future
 , 
1997
Washington DC
FDIC
Jeon
Y
Miller
SM
‘Bank Concentration and Performance’
2002
 
University of Connecticut Department of Economics Working Paper No. 2002-25
Nier
E
‘The profitability of banks: a cross-country study with a particular focus on UK banks’
2000
 
mimeo
Smirlock
M
Evidence on the (non) relationship between concentration and profitability
Journal of Money, Credit and Banking
 , 
1985
, vol. 
17
 
1
(pg. 
69
-
83
)
1
Savings institutions’ return on assets bottomed out in 1989.
2
Variables were instrumentalised in the first instance with lags of their own levels. Proposed instruments were assessed for relevance by considering the (individual and joint) significance of the excluded instruments in the first-stage regressions. The bound F-statistics, t-statistics and Shea ‘partial R2’ were examined for each of the potentially endogenous regressors in this manner. The validity of subsets of the proposed instruments was tested using the J-statistic from the Hansen test. The ‘difference-in-Sargan’ statistic test was also employed in order to test a subset of orthogonality conditions. Once appropriate instruments were found for potentially endogenous regressors, the endogeneity of these regressors were tested for. This was implemented through the Durbin-Wu-Hausman test. Suspect regressors were tested for possible endogeneity both individually and as a subset of regressors.
3
In fact, to the extent that there could be a causal relationship from profitability to concentration, this might be a negative relationship, insofar as high levels of aggregate profitability could encourage entry, leading to rising competition and lower levels of concentration. If this type of simultaneity were present, then a positive relationship between concentration and profitability could actually be underestimated.
4
Outlier identification methods used were the Cook, leverage, covratio, dfbeta, dfits, Rstandard and Rstudent.
5
Note that this analysis is not intended to analyse other underlying determinants of aggregate bank profitability in terms of macroeconomic issues, although these would no doubt be important. Macroeconomic factors are dealt with through the inclusion of time fixed effects (and these also preclude the additional inclusion of specific macroeconomic variables). The focus of this study is on analysing the determinants of profitability across banks and across time in terms of the structure of the banking sector and bank-level variables. For a broader analysis of determinants of financial sector profitability in recent years see, for instance, Crotty (2007).
6
Results not shown here for the sake of brevity, but available from the author on request.
7
An alternative that was considered was to construct separate concentration indices for different geographical regions. However, in addition to the problem that banks do compete interregionally (particularly since the full implementation of the IBBEA), this approach would have meant the loss of close to half the banks in the sample, no doubt with a selection bias in this as well, as many banks did not have a valid geographical code in the dataset.