Market Structure and Quality: An Application to the Banking Industry

Market Structure and Quality: An Application to the Banking Industry Astrid A. Dick ∗ This version: November 2002 (First draft: April 2002) Abstract This paper presents empirical evidence consistent with the predictions of the endogenoussunkcostmodelofSutton(1991),withanapplicationtobanks. Inparticular, banking markets remain concentrated regardless of market size. Given an asymmetric oligopoly where dominant and fringe firms coexist, the number of dominant banks remains unchanged with market size, with only the number of fringe banks varying across markets. Such structure is sustained by competitive investments in quality, with the level of quality increasing with market size and dominant banks providing higher quality than fringe banks. The analysis has implications for antitrust policy. DivisionofResearchandStatistics,FederalReserveBoard,Washington,D.C.,astrid.dick@frb.gov. This ∗ work is part of one of the chapters of the author’s 2002 PhD dissertation. She is grateful to Susan Athey andNancyRosefortheirinsightful commentsandideas, aswellastoAllenBerger for hissuggestions. The views herein, as well as any errors, are the author’s.

1 Introduction The work of Sutton (1991)1 provides a theoretical framework to explain why some markets remain concentrated as they grow in size, as well as how quality investments2 (endogenous sunkcosts) changewithmarket size. In particular, wheneverquality is arelevant part of the product and requires incurring fixed costs, the Sutton model predicts that markets remain concentrated and quality investments increase as market size grows. These relationships are interesting for understanding the process behind a given market structure, and because quality competition is just another dimension of differentiated product markets. While Sutton’s work makes robust predictions across a broad class of competition models about the relationshipbetweenmarket concentration andmarket size, aswell asquality investment and market size, the empirical literature documenting such relationships has nevertheless been scant. While concisely and clearly spelled out in the theory, appropriately establishing the predictions of the endogenous quality model in empirical settings has been difficult. Sutton (1991) provides a cross-country analysis of various industries in trying to find empirical counterparts to the theory. As he points out, however, his analysis confronts significant measurement problems. Recently, Ellickson (2001) applies the Sutton model to the empirical study of U.S. supermarkets. His work is the first to test the theory’s predictions on a large data set of markets within a single industry. This paper uses Sutton’s framework tobuild on the empirical work in the literature, with an application to the banking industry, taking a cross-section of U.S. metropolitan markets. The banking industry is a good application because of the large number of independent markets of varying sizes, and the available bank level data. This work is also of use to the industry from a policy perspective, as it can provide a tool in the design of regulation and antitrust analysis of mergers and acquisitions by aiding regulators — who currently focus on market concentration — in identifying the relevant variables of analysis. Moreover, it is an attempt to analyze and measure service quality in banking. 1Sutton (1991) builds especially on two earlier papers by Shaked and Sutton (1983, 1987). 2Notethat"quality"isusedherebroadlytoincludenotonlyproduct/serviceattributesrelatedtoquality per se, but also other activities such as marketing efforts of brand building and advertising. 1

The results suggest that the industrial structure of banking markets can be explained by the endogenous sunk cost model. In particular, there exists a lower bound to concentration, as banking markets remain concentrated across all market sizes. The basic structure of bankingmarkets is characterized by the coexistence of a few, large dominant banks — defined as those who jointly control over half of the deposits in a given metropolitan market — with a number of smaller, local banks which constitute a fringe. Given this concentrated structure ofasymmetricoligopoly, theequilibriumnumberofdominantbanksremainsunchangedwith market size, with only the number of fringe banks varying across markets. Moreover, this market structure appears to be sustained by competitive investments in quality, such as branch networks, branch staffing and geographic diversification (with some indirect evidence suggesting branding and advertising). In particular, the level of bank quality increases with market size, and dominant banks are found to provide a higher level of quality than fringe banks. Furthermore, banks do not appear to carve out areas within the relevant geographic bankingmarket, but rathercompetewitheachotherclosely. Intermsof theproduct market, however, dominant and fringe banks appear to focus on a few different sectors. Theanalysishassomedirectimplicationsforantitrustpolicy. Theintroductionofquality investment in the study of competition alters certain relationships between the number of firms, market concentration and conduct that have been believed to exist by the bank regulatory authorities.3 Both the Department of Justice and the Federal Reserve Board4 focus on market concentration to determine whether a contemplated merger might cause antitrust concerns [see Amel (1997)]. Yet a relevant question might be whether the new bankwouldbecomeadominantfirmorinsteadbepartofthefringe, aswellasconsiderations regarding market size and quality provision. For example, will the formation of the new firm imply the reduction of the number of dominant firms in the market to one? If the postmerger firm becomes dominant, will it have competition from other dominant firms? Will the new firm join the fringe instead? Will the merger increase the ATM network available to consumers? 3For example, the antitrust policy based on findings that markets with fewer firms tend to have lower deposit rates and higher loan rates [Rhoades, 1996; Amel, 1997]. 4The FederalDeposit InsuranceCorporation and theOfficeoftheComptroller of theCurrencyalsohave regulatory authority but have not been very active in antitrust enforcement in recent years. 2

This paper also sheds light on the empirical finding that larger banks charge significantly higher fees than smaller banks.5 The reasons usually speculated for this occurrence include locational differences between larger and smaller banks, the better service quality of bigger banks, and the fact that larger organizations tend to depend less on retail customers for funds. The findings here indicate that dominant firms, which tend to be large banks, do charge higher fees yet invest more in quality. While it might be presumed that the direct cause of quality is bank size, it appears that quality is the result of banks’ competitive investment in endogenous sunk costs, which gives rise to barriers to entry and allows for market structure nonfragmentation with increases in market size. This, in turn, allows those banks that invest more in quality to hold large market shares and become big banks. Therestofthepaperisorganizedasfollows. Section2outlinesthetheoreticalframework. Section 3 describes the data and provides a discussion of endogenous sunk costs in banking. Section 4 provides supporting evidence for the various predictions of the endogenous sunk cost model in terms of banking markets. Section 5 provides an analysis of competition inside banking markets, while 6 analyzes some of the implications for antitrust policy. Finally, section 7 concludes. 2 Theory of market structure and quality The work of Sutton provides a framework for analyzing market structure as market size grows, and how this relates to quality investments, whenever quality requires that firms incur fixed costs. This theory guides the empirical analysis in this paper. Central to it is the notion that some sunk costs are incurred with a view to enhancing consumers’ willingnessto-pay forthe firm’s products, andas aresult represent a firm’s choice variable (therefore are “endogenous”). The key to the theory is that while exogenous costs have a fixed magnitude irrespective of market size, endogenous sunk costs vary as market size changes (though both arefixedwithrespecttooutput). Drawingadistinctionbetweenthesesunkcosts, themodel makes robust predictions, across a broad class of competition models, about the relationship 5See Hannan (2001, 2001) for an analysis of retail fees of depository institutions for 1994-2001, based on asurveyofretail feesandservicescommissionedbytheFederalReserveBoardonanannualbasis. Hannan finds that large banks charge higher fees, on average, than small banks. 3

between market concentration andmarket size, aswell as the equilibriuminvestment in sunk costs and market size whenever costs have endogenous components. In particular, exogenous sunk costs, on the one hand, are defined as those setup costs or fixed outlays associated with acquiring a single plant of minimum efficient scale that do not vary with market size. Endogenous sunk costs, or quality investments whose magnitude is chosen by the firm, on the other hand, are defined as costs that can change a firm’s demand, such as R & D, advertising and direct service upgrades. The central focus of the theory then lies in unraveling the way in which these two elements of sunk costs, exogenous and endogenous, interact with one another to determine the equilibrium market structure in an industry. For the case of exogenous sunk costs, the central prediction of the theory is that an increase in market size relative to setup costs may lead to indefinitely low levels of market concentration.6 In the case of endogenous sunk costs, however, this property breaks down. Here the model predicts that markets remain concentrated regardless of market size, as competition among firms leads to escalating investment in quality. In particular, the conclusions of the model are: (1) Market structure does not fragment as market size increases, and therefore there exists a lower bound to the equilibrium level of concentration in the industry, no matter how large the market becomes; (2) firms engage in acompetitive escalation of investment in quality as market size increases, creating barriers to entry; (3)theequilibriumnumberoffirmsinthemarketremainsapproximatelythesameregardless of market size. These conclusions are robust to a very broad class of oligopoly models with various degrees of product differentiation, toughness of price competition and strategic symmetry/asymmetry. For a detailed study of these models, see Sutton (1991).7 6Inanindustrywherethereareonlyfixedsetupcosts(exogenouscosts)andtheproductishomogeneous, the equilibrium number of firms should increase with market size, while market concentration should asymptote to zero. In a differentiated product setting, however, the existence of only exogenous components to sunk costs leads to multiple equilibria, ranging from concentrated to fragmented market structures. 7Notethatendogenoussunkcostsareinterpretedbroadly. Thefindingthatmarketsremainconcentrated as they grow in size is in itself an indication that there is strategic interaction in the industry. This is important as it provides an explanation for market concentration that is not the usual efficiency/economies 4

3 The banking industry: data and background 3.1 Data sources The data are based on a cross-section for 19998 and are taken from several sources. The variablesusedheretoanalyzemarketstructureinclude: 1)bankcharacteristics, derivedfrom balance sheet and income statement information from the Report of Condition and Income (Call Reports) from the Federal Reserve Board; 2) branch deposits, taken from the Federal Deposit InsuranceCorporation(FDIC)Summaryof Deposits; and3)demographicvariables, taken from the U.S. Census and the Bureau of Economic Analysis. The sample includes all metropolitan markets and all FDIC insured-commercial banks in the U.S. The Appendix shows summary statistics for the variables used in the analysis, as well as a description of the variables. Given the format of the data, there are several possible levels of aggregation that could be used as the unit of analysis. My approach is to define the relevant geographic banking market at the level of the metropolitan statistical area (MSA), a geographic unit defined by the U.S. Census Bureau that consists of a large population nucleus, together with adjacent communities, that comprise one or more counties. This market definition is supported by surveys of consumers and businesses as well as the bulk of the empirical banking literature.9 3.2 Basic characteristics of banking markets In the U.S. there are about 330 MSA banking markets, which represent 83 percent of total U.S. dollar deposits. The average number of banks in an MSA is 20, with as few as two banks in Lewiston-Auburn, ME, and with as many as 255 in Chicago, IL. Table 1 shows the distribution of MSA markets in terms of the number of banks in the market. On average, an MSA has a total of 140 branches. Adjusting by population, there is an average of 28,000 of scale one. What this strategic interaction is (e.g. quality per se, advertising, first-mover advantage) depends on the application. Here lies what is one of the great contributions of Sutton’s work. 8The data are for the second quarter, which is chosen here because some the variables of interest are reported only then. 9For a detailed discussion on relevant geographic market definition, see Dick (2002) and the references therein. 5

persons per bank in a given MSA, and 4,600 per branch. As measured by population, the bulk of markets has asize between 100,000and 500,000 people.10 Table 2 shows a tabulation of MSAs by various population size categories. The average Herfindahl-Hirschman index (HHI)11 across MSA markets is around 1900, with market concentration going from as low as 584 in Chicago, IL, which has 255 banks, to almost 7800 in Pittsfield, MA with only three banks.12 The last column of Table 2 shows the average HHI for each market category by population, while Table 3 depicts some percentiles for the distribution of the HHI across MSA markets (with a standard deviation of 800). Definitions: dominant and fringe firms Banking markets usually hold dozens of firms, yet many of these firms hold a very small portionofthemarket. Asaresult, it isrelevant tomakeadistinctionbetweenthelatterand those firms that head the market in terms of market share. For this purpose, I define two types of banking firms that will be used in the analysis: dominant and fringe. Dominant firms are defined as the set of firms that jointly hold over half of the market in terms of deposits. All other firms are fringe firms. For robustness purposes, some other definitions of dominant firms will be used as well later in the analysis.13 Market equilibrium Sutton’s theory of market structure applies to markets in equilibrium. In the case of the banking application here, the underlying assumption is that the industry reached an equilibrium in 1999, the year of the analysis. While changes in the industry continued 10The average MSA size is about 1940 square miles. 11The Herfindahl-Hirschman index is a concentration measure constructed as the sum of the squares of themarketshareofdepositsatthelocalmarketlevel. Here,followingthepracticeoftheAntitrustDivision, I multiply it by a factor of 10,000. 12The Antitrust Division defines the threshold of a highly concentrated market at 1800. In the case of bank mergers, the Antitrust Division has used a screen of 1800/200 over the past several years. That is, in most cases they will not conduct a full investigation unless in at least one market: (i) the post-merger HHI is at least 1800; (ii) the merger produces a change in the HHI of at least 200. 13In particular, two other definitions of dominant firm will be utilized to test whether the results here are sensitivetothedefinitionofdominantfirmgiveninthetext: (i)followingtheDepartmentofJusticeandthe Federal Reserve Board’s definition, a dominant bank is that whose market share is at least twice as large as theshareofthesecond-largestcompetitorinthemarket(fromthe“CaseworkManual”formergerproposals of the Federal Reserve Board); and (ii) a dominant firm is that with the largest market share in a market (or alternatively, those with the largest two/three market shares). 6

to occur after 1999, the assumption seems reasonable given the tremendous shake-out the sector experienced throughout the last three decades, and in particular in the last ten years, with the introduction of nationwide branching throughout 1994-1997.14 Figure 1 shows the number of bank mergers per year since 1993.15 There is an average of 360 mergers per annum, and the number of mergers per year decreases steadily since 1994. Moreover, in 1999, there is a decrease of over 60 percent in the number of mergers from the previous year, and of 70 percent since 1993.16 3.3 Endogenous sunk costs in banking Banksdiffergreatlyintermsoftheservicequalitytheyprovidetotheircustomers. Withina given market, a set of very diverse banks tend to coexist, with some being small, local banks withafewbranches, andotherslargeandcoveringextensivegeographicareas, withextensive ATM and branch networks. Banks also differ in terms of the expertise and customer care offered at the branch, the size of branch personnel (which is related to waiting times and the availability of human interaction), financial advise, as well as advertising/brand investments and overall service quality. Endogenous sunk costs, indeed, are expected to be a significant component of total banking costs. Branch and ATM network At least some of the branch and ATM installation costs, which affect the bank’s demand by attracting new customers, are clearly sunk. Once built, it is hard to recoup the incurred 14Regulatory restrictions affecting the ability of banks to diversify geographically have decreased dramatically. Deregulation of unit banking and limited branch banking occurred gradually throughout 1970-1994 in most states. Intrastate branching deregulation began in some states even before the 1970’s, while interstate banking started as early as 1978. The process of deregulation of geographic expansion culminated in 1994with the passage of theRiegle-Neal Interstate Banking and BranchingEfficiencyAct, whichpermitted nationwide branching as of June 1997. 15Theinformationonthefigureisbasedontheauthor’scalculationusingBankingHoldingCompanydata from the Federal Reserve Board. 16Thereareafewcaveatstonoteaboutthisassumption. First,mergerstakeawhiletosettle,andmergers do occur in 1999. Second, 1999 is a boom year in the business cycle. However, the market structure in 1993 (in terms of a dominant firm vs. fringe framework) is found to be similar to that of 1999, even though 1993 is not a boom year. In addition, I find that the firms that have negative (accounting) profits in 1999 and that would likely exit the market, are part of the fringe. As a result, the basic market structure between dominant and fringe firms, documented later in this paper, should not be affected by these developments in the industry in any significant manner. 7

costs. As Radecki et al. (1996) point out, the typical bank branch costs roughly $1 million to build. While a portion of this expense is for equipment, which may be removed and installed elsewhere, most of it covers construction costs. There is also plenty of anecdotal evidence suggesting that branches represent sunk costs. For instance, it represented one of the main arguments for internet banking (The European Internet Report, Morgan Stanley Dean Witter, June 1999). While the cost of opening a single branch might not be exactly fixed with respect to output, a bank’s overall branch density cost is likely to be largely independent of output levels. In other words, branch and ATM networks should be at least somewhat independent of the number of customers using them in the sense that while a consumer might do most of her banking with a single bank branch, she should still value the convenience of her bank’s branch density in the area as well as its ATM network.17 Even if there is a certain number of customers that a single branch can service, it is unlikely to be binding in practice.18 This is suggested by the popularity in recent years of the in-store or supermarket branch —a full service branch located within alarge retail outlet— as a way to expand customer bases relative to a conventional bank branch [Radecki et al., 1996]. Banks find them attractive not only for cost reduction purposes, but also because they provide access tolargeflows of potential andexistingcustomers (even thoughthey have smaller staffs than branches): the typical supermarket averages 20,000to 30,000 customers a week, whilethe typical bankbranchaveragesjust 2,000to4,000weeklycustomers [Williams, 1997]. Advertising Advertisingislikelytobeanothercomponentofsunkcosts. Unfortunately, dataonbank advertising expenditures are scant. While there seems to be a lot of heterogeneity across banks in terms of their investment in advertising, the available statistics suggest that the average dollar figure is small relative to other operating costs. According to surveys carried 17Ifthereisanyrelationshiptooutputlevels,ATMsarelikelytobelessincrementaltocoststhanbranches, though the number of ATMs is likely to be highly correlated to branches. 18Output is usually measured in terms of dollar volumes, so the link between branch/ATM costs and the number of customers is even less direct, even if there is a given number of customers that can be served per branch that is also binding. 8

out by the American Bankers Association, roughly one percent of bank operating costs on average was devoted to advertising in 1996, while total bank marketing expenditures were close to 4 billion dollars in 2001. While these numbers are rather small, anecdotal evidence suggests that in the nineties bank “marketing has moved from a back room operation ... to a front line strategic function.”19 For instance, according to National Leading Advertising, BankAmerica Corp. was the 125th leading U.S. advertiser in 1996, with total expenditures of $145 million. Advertising outlays might also be highly correlated with the number of bank branches in light of the anecdotal evidence on the greater role of the branch in the bank’s advertising decisions.20 As described by Radecki et al. (1996), a typical branch has expenses of around$700,000peryear, and while thelargest component ofthiscostisstaffcompensation, advertising is usually part of it. Furthermore, branches are to banks a form of advertising itself. There is plenty of anecdotal evidence about how banks hope to woo customers using their branches, usually withstylishmerchandisingandcustomerservice.21 Banksbecomemorevisibletoconsumers through their branches, and in fact, many banks put clocks outside their branches for this reason. Branding Branding, which requires fixed cost outlays, is also significant in banking, as a wealth of anecdotal evidence suggests, with banks investing a growing fraction of their resources by engaging in branding campaigns and brand building, as well as the development of inhouse brand marketing departments and branding strategies.22 Further evidence on the 19“The Banks, They Are A’ Changin’,” D. Asher, Newspaper Association of America, 2003. 20“With micromarketing, the promotional decisions are shifted from the corporate staff to the individual branches,wheremoreisknownaboutcustomersandprospects,suchaswheretheyliveandwhattheybuy... Thereareless[sic]expensivetelevisioncommercialsandhighlyeffectiveoutdoordisplays”(from“ItPaysto Think Small in Marketing,” K. Pelz, American Banker, March 4, 2002). 21Forexample,“... ahandfuloflargeinstitutionsareplanningaggresivecampaignstobuildmarketshare” (“SomeGiants PlanningAd Assaults; They Hope to Gain Market Share as Others Retrench,” E. Braitman, American Banker, November 15, 1990). As part of this strategy, many banks have even tried installing coffee shops and “investment bars” within their branches (“Bank branches take a page from retail’s book,” San Francisco Business Times, Sept. 2001). 22For example, a search on bank branding on the American Banker magazine database throws out thousands of related articles for recent times, suggesting the prevalence of branding as a part of bank business. 9

importance of branding is provided by the way banks that merge choose their new brand name, according to bank periodicals. Usually, they choose the name that customers are more familiar with and/or is the strongest brand.23 Data The data available do not allow for a complete and direct measure of sunk costs, but some observable bank characteristics should provide an approximation. I use here several bank attributes as quality correlates,24 including: (i) a bank’s branch density in the MSA market, defined as the number of branches per square mile in the MSA; (ii) the number of employees per branch; (iii) the age of the bank, which might proxy for bank experience/branding; (iv) the geographic diversification, measured as the number of states in which the bank operates; (v) salary per employee. From the consumer’s perspective, more of each one of these attributes is likely to be a good thing. Branch density25 and geographic diversification are expected to capture the quality of the overall bank network, as they are related to the number of branches in a bank’s local markets and should be highly correlated with the ATM network as well. Moreover, while there are no data to measure advertising expenditures, the number of branches might be highly correlated with advertising (either actual dollar outlays and/or under the interpretation of branches as advertising). 23A good example is that of the large NationsBank and BankAmerica merger in 1998: they chose the BankAmerica name because of “its longer history” and “its patriotic feel which has more intrinsic appeal thantheNationsBankname”(“BrandNametoBeUnveiledinAdsTonight,”C.Guillam,AmericanBanker, Sept. 30, 1998). 24Dick (2002) also finds that branch density, the number of states of operation, age and employees per branch are bank attributes which are significant in affecting a consumer’s deposit institution choice. Assuming that consumers choose a bank for deposit services in order to maximize a linear utility function, Dick (2002) uses a logit-based model of choice to derive bank demand as a function of bank attributes and prices. Shedefines thebankingmarketat the level of the MSA/non-MSAruralcounty; usesaggregatebank data, as opposed to actual consumer choices; and imposes some strong assumptions on the distribution of idyosyncratic consumer preferences. 25Note that while the number of states in which abank operates might be restricted by the region’sregulatory regime at the time, by 1999 virtually all U.S. states allowed for nationwide branching. Furthermore, the relevant comparison hereiscross-sectional and within amarket,whereallfirmsareexposed tothe same regulatory regime. 10

The number of employees per branch26 should capture some of the quality provided at the branch, since the larger the branch staff, the lower waiting times should be. While it is not possibleto measure brandingdirectly, bank age27 is expected to berelated to bank experience and its service quality, and/or the importance of branding, since a bank that has been around longer is more likely to have greater prestige and prominence than a younger bank. Expertise can be offered by any bank, but older banks might be particularly good if there is a learning curve. Furthermore, bank age might play a role in light of some theoretical work that suggests that bank entrants face a “lemons” problem derived from their inability to distinguish new borrowers from old borrowers who have been rejected by their previous bank [Dell’Ariccia et al. (1999), Marquez (2002)]. Older banks might know their customers better and therefore be able to custom fit their products better, therefore providing higher quality to the consumer. Salary paid to the bank’s employees should be correlated to quality, as more highly qualified employees, who might provide better service and expertise, should be more expensive. This could also be correlated with the degree of sophistication of the products offered by the bank. A word about a bank’s market presence and its quality provision is appropriate here. The fact that the definitions of dominant and fringe firm are based on deposit market share, and that some of the quality components are based on network size such as branch density and number of states of operation, might raise questions about a potential circularity of reasoning when carrying out tests that relate market dominance with quality levels. In particular, is quality choice driven by a bank’s market share? As already mentioned, there is abundant anecdotal evidence suggesting that banks mainly open branches in the hope of attracting new customers, as opposed to responding to the needs of their existing customer base. Branches, in fact, are thought to represent a form of advertising. In particular, a bank might set a high-quality target by offering a number of branches to its consumers in a local market, regardless of its current market share. This bank might then over time see its 26While a given number of customers might require a minimum number of employees at the branch, anything above that level of employment should be part of quality of service. 27While consumers might not be perfectly aware of a bank’s age, they should be able to discern between relatively young and well-established, older banks. 11

market share grow as a result of offering a higher quality product.28 4 Empirical results: market structure, quality and market size Given the theoretical framework of Sutton, and the interpretation given here to the cost structure of banks, it is expected that: 1. banking markets remain concentrated as market size grows; 2. the number of dominant banks remains more or less constant across market sizes; 3. this market structure is sustained by increases in banks’ fixed costs outlays for quality investment as market size grows. 4.1 Market structure across market sizes In this section I provide supporting evidence for the first prediction based on Sutton’s endogenous sunk cost model, that there is a minimum level of concentration which is never violated no matter how large the market becomes. Figure 2 shows the relationship between concentration and market size. The former is measured by the HHI, while the latter is measured in terms of the log of market population, where the log is taken to facilitate appreciation of the figure. The figure depicts the HHI observed in markets with as few as 57,000 people and as many as 9 million people. Apparently, there is a lower bound to concentration throughout all market sizes.29 Indeed, as depicted in the last column of Table 2, the average HHI shows little variation across various market size categories. 28The alternative of “quality follows size” might still be feasible under certain scenarios of bank entry. Forinstance,somebanksmayhavebecomelargethroughfirst-mover advantage(suchaseconomiesof scale, switching costs). 29Similar findings are obtained when using a C4 (sum of largest four market shares) and a C1 (maximum market share) measure of market concentration. Indeed, the C4 measure never goes below 40 percent, showing little variation across markets with a few thousand to millions of people. 12

FINDING 1: There exists a lower bound to concentration in banking markets, as market structure does not fragment with market size. 4.2 Number of firms across market sizes In this section I provide evidence for the remarkable fact that across all market sizes, the number of dominant banks remains roughly the same. Moreover, a similar dominant firmfringe structure arises in all markets. Table 4 presents a tabulation of markets according to population and number of dominant firms.30 This table provides evidence of a striking fact: regardless of market size, the bulk of markets (87 percent of the MSAs) have either two or three dominant firms. Moreover, the correlation between the population and the number of dominant firms in a market is almost zero. This is particularly interesting when contrasted with a model without quality competition but just exogenous fixed costs, where the number of firms should grow with market size given that the number of consumers served per firm should be the same for all markets. Deposit Lorenzcurves31 provide another way toappreciatethefact that few firms control most of the market, regardless of the number of firms serving it. Figure 3 shows a Lorenz curve for deposits, where firms are ranked on the x-axis according to their share of market U.S. dollar deposits, while the y-axis shows the cumulative share of deposits. Given the large number of MSA markets, for ease of analysis the figure depicts only six markets, one for each market size category32 (as defined in Table 2). The only apparent difference among the markets is in the length of the tail of the curve, which grows in the number of firms serving the market. Below the 50 percent cumulative share line, markets differ little. The above description indicates that as markets grow, the number of dominant banks remainsvirtuallyunchanged. Naturally,asmarketsgrowinpopulationsize,theyalsotendto 30Ellickson (2001) finds a similar structure for supermarkets. 31In a market with symmetric firms, the Lorenz curve would actually be a straight line, since all firms would have the same market share. Thus, the closer the curves get to the y-axis, the more asymmetric, and therefore, the more concentrated the market becomes. 32ThemarketschosenineachcategoryarethosethataremostrepresentativeoftheLorenzcurvestructure within their population size category, both in terms of the number of firms and the market population. However, even if markets were chosen randomly, the figure would be similar. The markets shown in the figure are, in decreasing order by population size: Philadelphia, PA; Fortlauderdale, FL; Vallejo-Fairfield- Napa, CA; Hunstville, AL; Punta Gorda, FL, and Pocatello, ID. 13

expandinthenumberofbanks, yetthisgrowthisonlyreflectedinthelengthofthetailofthe fringe, and does not affect the dominant-firm fringe structure observed in smaller markets. Indeed, the number of firms in a market is highly correlated with population size (0.77), yet the number of dominant firms is almost independent of population and the total number of firms in the market. If economies of scale were the explanation for why large markets have such a small number of dominant firms, one should then observe smaller markets tending toward monopoly (in the sense of having only one dominant firm). Yet, smaller markets appear to have the same number of dominant firms as larger markets; in fact, there is no single MSA market with a single dominant firm. FINDING2: Given a concentrated structure of asymmetric oligopoly where dominant and fringe firms coexist, the equilibrium number of dominant banks remains virtually unchanged with market size, with only the number of fringe banks varying across markets. Thus, the basic dominant firm-fringe structure does not vary across market sizes. 4.3 Sunk costs across market sizes In this section I provide supporting evidence for another expectation about banking markets based on the endogenous sunk cost model: the larger the size of the market, the greater the sunk costs incurred by banks in equilibrium. In the current setup, this prediction can be brokenupintotwoimplications: (i)asmarketsizeincreases, thelevelofabank’sendogenous sunk costs increases, and (ii) dominant firms incur a higher level of sunk costs than fringe firms. Table5reportsMSAlevelregressionsofqualitycorrelatesonthelogofpopulation.33 The coefficient on population is highly significant for branch density, number of states of bank presence, and salary per employee, suggesting that these quality correlates increase with market size, as the model predicts under endogenous sunk costs. In terms of city-specific effects, the results imply roughly that for a doubling of population size there is a 3.5 times increase inthebranch density of the average bank inthemarket, as well as a $2,000 increase in the average salary per employee, and an increase in the geographic coverage. 33These regressions include MSA income per capita (natural logs) to control for MSA characteristics. 14

Usingafewotherdefinitionsofdominantfirmtotestwhethertheresultsherearesensitive to the particular definition of dominant firm, I find that the above-mentioned relationship between market size and quality is robust to various definitions (results not shown). In particular, following the Department of Justice and the Federal Reserve Board’s definition, a dominant bank is defined as that whose market share is at least twice as large as the share of the second-largest competitor in the market (only 57 banks fall into this category, however), and as alternative definitions, a dominant firm is defined as that with the largest market share in a market (or alternatively, those with the largest two/three market shares). Table 6 shows means for the various components of the measure of quality, for both dominant and fringe firms. Dominant banks appear to provide more branches, which, in turn, have more employees, and they also tend to be more geographically diversified, have been around longer, and pay higher salaries to their employees.34 To test for the significance of these attribute differences, Table 7 shows the results from estimating quality correlates of bank (cid:1) as a function of an indicator variable for whether the bank is a dominant firm (in which case the variable takes on the value of one), including MSA fixed effects. All the specifications depict a positive and highly precise coefficient estimate for the dominant firm indicator, suggesting that dominant firms provide a significantly higher level of quality.35 In particular, dominant firms tend to have higher branch density, more employees per branch, are older and more geographically diversified.36 Moreover, after controlling for MSA fixed effects, dominant firms appear to pay salaries that are on average almost $ 5,000 higher than those paid by fringe firms. This is a particularly interesting result if there remains any concern about the potential circularity between the main quality measure here and the definition of dominant firm, as salary per employee should be unrelated to the definition of dominant firm based on market share. Among other quality-related characteristics, dominant firms also appear to serve rural 34While Sutton’s model provides some clear predictions about market structure, it tells little about what determines who becomes a dominant firm. The fact that dominant firms tend to be older might suggest the existence of a first-mover advantage into local markets, sustained not only through customer switching costs but also through informational barriers as in Dell’Ariccia et al. (1999). 35Results are shown for MSA fixed effects regressions only, given that most bank attributes are measured atthebanklevel,andasaresult,shownomarketvariation,whichisrequiredtoestimatebankfixedeffects. 36Geographic diversification is measured as the number of states in which the bank operates. However, results are similar for the measure based on the number of MSAs in which the bank has branches. 15

markets much more frequently than fringe firms (80 percent of dominant firms operate in at least one rural market vs. 39 percent of fringe), which might be considered by some customers as a useful service, as well as operate in many more MSAs across the country (89 percent of dominant firms operate in more than one MSA vs. 55 percent of the fringe). The above findings suggest that the observed market structure cannot be merely explained by economies of scale operating on the technological side. Dominant and fringe banks appear to be different not only in terms of their scale of operation, but also in terms of quality of service, with dominant banks choosing to provide a higher level of quality than fringe firms. The unfragmented market structure that holds throughout various market sizes is apparently sustained by investments in larger networks and better service. FINDING 3: The market structure documented earlier is sustained by competitive investments in quality. In particular, the level of bank quality increases with market size and, moreover, dominant banks appear to provide a higher level of quality than fringe banks. 5 Competition analysis: Carving out of “neighborhoods” and product markets The previous sections established that banking markets remain concentrated regardless of market size, and that roughly the same number of dominant banks serve each market, as predicted by the endogenous sunk cost model. This structure, however, is consistent with various models of “localized” competition. One might ask, for instance, whether firms are able tocarve out geographicareas (“neighborhoods”) orproduct marketswithintherelevant geographic market. Using much of the insight provided by Ellickson (2001) in his study of market segmentation for supermarkets, in this section I examine the following: whetherdominant firmscontrol geographicareasorinsteadcompeteheadonwitheach • other within a given MSA; whether dominant and fringe firms serve different geographic areas within the MSA; • 16

whether dominant firms carve out a different product market from fringe firms; • whether there are differences between dominant and fringe firms in terms of prices, • costs and performance. Do dominant firms control geographic areas or compete head-to-head within a given MSA? While the bulk of the evidence suggests that the relevant geographic market is at the MSA level, one might ask whether dominant firms either segment the market or compete head to head with each other within a given MSA (in the least, this is useful as a sensitivity analysis of the results on market structure to the particular relevant market definition). For instance, suppose that in a given market, dominant bank (cid:2) has ten branches. Then another dominant bank (cid:3) in that market, with ten branches as well, could have each one of them located nearby to bank (cid:2)’s branches, or alternatively, located in very different areas or “neighborhoods” of the MSA. In order to explore this, each MSA is broken down into cities (or towns) and counties. There are 8803 cities and 883 counties for the 331 MSAs present in the sample. Cities are rather small sections within the MSA, with an average of 27 cities per MSA.37 Counties are much larger areas, comprising several cities and towns. An average MSA has between two to three counties. It is worth noting that in the analysis that follows, any reference to dominant or fringe firm refers to the definition provided earlier, done at the level of the MSA. Table 8 shows cities and counties grouped by the number of firms serving them, and provides the average number of dominant firms in each category. The first column shows the number of cities/counties that fall in each category based on the number of firms that serve the area (for instance, there are 3842 cities and 12 counties that are served by a single bank, where the bank is either dominant or fringe). The second column shows the number of dominant banks, on average, in a given area (for example, in cities with two to five banks, there is one dominant bank on average, or 1.2, as indicated on the table). The third column 37In the Boston MSA, for instance, some cities and towns include: Boston, East Boston, Braintree, Brookline, Cambridge, Belmont, Chelsea, and Newton. 17

also provides the number of dominant firms but conditional on there being at least one dominant firm in the area. The results from this table suggest that dominant banks do not carve out geographic marketnicheswithintheMSA.First,countiesservedbyonlyonefirmarefew,and,moreover, theyaremostlycontrolledbyfringefirms. Inparticular, only12outofatotalof883counties actuallyhaveasinglefirm,andoutofthese12counties,onlytwoarecontrolledbyadominant firm. Cities with a single firm represent 44 percent of all cities, and only one third of these cities have a dominant firm as the monopolist. Note, however, that over 96 percent of these monopoly cities have only one single branch in them. This suggests that the area of these cities is indeed very small –an area with a single branch can hardly be a carved-out market “niche.” Second, outside of these monopoly areas, the number of dominant firms is above one in most cities and counties, as evidenced in the second and third columns of the table. The average number of dominant firms is 1.5 in cities and 2.1 in counties. Conditional on there being two or more firms in the area, only 16 percent of cities and less than 5 percent of counties have a single dominant firm. Conditional on there being at least one dominant firm in the area, there is an average of 2.3 dominant firms in counties, and 1.8 in cities. That is, if there is one dominant firm in a given area, it is likely there is another dominant firm. This fact is relevant if one believes that competition from another dominant firm is important in curtailing the market power of an incumbent dominant firm. These findings suggest that at various levels of disaggregation within the MSA, dominant banks do not appear to hold distinct geographic areas, and instead seem to compete head on with each other. Do dominant and fringe firms serve different geographic areas within a given MSA? Analternativepossibilitytomarketsegmentationisthatdominantandfringefirmsmight serve distinct geographic areas within the MSA. This possibility is easily ruled out by the data. 18

First, most areas have dominant firms overlapping with fringe firms. Monopoly areas, as mentioned earlier, are rare. Areas with multiple firms but with only one firm type represent a small portion (14 percent of cities, and 8 percent of counties), and are mostly served by fringe firms. Moreover, these areas tend to be geographically small, with two to three banks serving them, and one or two branches per bank. Second, dominant and fringe firms tend to locate their branches near each other. Figure 4 shows the location of each branch throughout the Boston MSA market, which is fairly representative of other MSA markets in this respect. The circles in the figure represent branches belonging to Boston’s dominant banks, while the triangles depict branches of the fringe. The amount of overlapping that these two types of banks have all over the MSA is striking: right next to most circles of the figure there is a triangle. This suggests that dominant firms tend to compete with fringe firms very closely, by locating their branches near each other. The evidence indicates that even at the level of analysis of such a small unit as the city, dominant firms do not appear to be segmenting the market from those of fringe firms, but rather tend to serve the same geographic areas. Indeed, the basic dominant-fringe firm structure documented at the level of the MSAappears tobe relevant even within the smaller geographic area of the county. Do dominant firms carve out a different product market from fringe firms? This section explores whether dominant firms serve different customers from those of fringe firms. Table 9 shows several balance sheet items for both types of institutions that provide insight into their asset portfolio and product mix. Loans, commitment lines and time deposits may all be thought of as bank products. In terms of this output set, one significant difference between dominant and fringe firms is in the proportion of assets allocated to commitment or credit lines (an off-balance sheet item): while dominant firms allocate over 60 percent of their assets to commitment lines, fringe firms dedicate about half of this. Given the nature of a commitment, this might be suggestive of a difference in service quality between the two firm types (emphasizing earlier 19

findings in this paper), as opposed to a distinct product market niche. The central feature of a commitment is that a borrower has the option to take the loan down on demand over some specified period of time.38 Commitment lines of credit are of great value to a bank’s client as it allows her to obtain loans as her funding needs arise, which is a feature especially useful for customers that confront numerous contingencies in their activities. Another marked difference between dominant and fringe firms is in the proportion of small loans (defined to be less than $100,000 according to the FFIEC form reported by banks to the regulatory agencies). While 13% of business loans and 24% of agricultural loans are small in the case of fringe firms, the proportion of these kinds of loans that are small is negligible in the case of dominant firms. This suggests that dominant and fringe firms might have some distinct niches in the loan market. Based on the table, dominant and fringe banks show some other differences as well, but thesearenot as striking, andarehardly large enoughastosuggest distinct nichesintermsof theproductmarket(eventhoughtheyarestatisticallysignificant, givenvaluesofT-statistics shown on the table). In particular, dominant firms allocate a larger portion to commercial and industrial loans, which constitute an important source of funding for local businesses, and have lower liquidity as measured by the federal funds and securities holdings. Based on the analysis above, however, dominant banks, who assign a large portion of their resources to credit lines, might appeal more to consumers that need financing on demand, which will tend to be business consumers. Fringe firms might focus more on serving smaller businesses and households, as evidenced by the smaller loan size. Other differences between dominant and fringe firms To complement the analysis, I examine differences between dominant and fringe firms in terms of prices, costs and performance. Table 10 shows the various interest rates paid and receivedbybothdominantandfringebanks.39 Excludingcommercialandindustrialloans, in 38Commitments are defined as the sum of unused commitment lines and letters of credit over total loans. Loan commitments are one of the products that make commercial banks different from other competing institutions/lenders such as insurance and finance companies. 39Prices are imputed using the income/expense flows from the income statement, adjusting by the corresponding balance sheet stocks, as indicated in the Appendix. 20

which dominant banks might specialize (as mentionedabove), dominant bankscharge higher interest rates on real estate and loans to individuals (mostly credit card loans), higher fees on checking accounts, and pay lower interest rates on deposits.40 This could be related to quality differences between the two types of firms, documented earlier in the paper. Dominant banks also appear to perform much better than fringe banks in terms of accounting profits. As depicted in Table 11, while fringe banks enjoy a return on equity of 24 percent, with a large standard deviation of 41 percent, dominant banks’ profits are highly concentrated around 33 percent. In fact, while the number of dominant firms that are losing money is negligible, many of the fringe firms (over 8 percent) are making negative profits, which explainsthe higherturnoverinthe firms ofthefringe. Dominantfirmsalsoshowlower average costs, as evidenced by operating expenses as a percentage of assets, which could be suggestive of the dominant firms’ greater operating efficiency. On the other hand, dominant firms might be choosing a higher level of risk, as their credit portfolio has a slightly higher level of charge-off losses in terms of assets. FINDING 4: Banks do not carve out areas within the relevant geographic banking market, but rather compete with each other closely. However, in terms of the product market, dominant and fringe banks appear to focus on a few different sectors. 6 Implications for antitrust policy Theanalysisof thispaperhas some directimplications forantitrust policy. Theintroduction of quality investment in the study of competition alters certain relationships between the number of firms, market concentration and conduct that have been believed to exist by the bank regulatory authorities.41 Both the Department of Justice and the Federal Reserve Board42 focus on market concentration to determine whether a contemplated merger might cause antitrust concerns [see Amel (1997)]. In particular, their criteria include whether a 40Note that the equality of the rate on leases cannot be rejected at any reasonable significance level, as evidence by the value of the t-statistic shown on the table. 41For example, the antitrust policy based on findings that markets with fewer firms tend to have lower deposit rates and higher loan rates [Rhoades, 1996; Amel, 1997]. 42The FederalDeposit InsuranceCorporation and theOfficefotheComptroller of theCurrencyalsohave regulatory authority but have not been very active in antitrust enforcement in recent years. 21

proposed merger would result in a market Herfindahl index greater than 1800, or increase it by more than 200 points (“1800/200 rule”), and whether the market share of the postmerger firm would be 35 percent or more of market deposits. In the context of the present paper, a relevant question might be whether the new bank would become a dominant firm or instead be part of the fringe, as well as considerations regarding market size and quality provision. For example, will the formation of the new firm imply the reduction of the number of dominant firms to one? If the post-merger firm becomes dominant, will it have competition from other dominant firms? Will the new firm join the fringe instead? Will the merger increase the ATM network available to consumers? Moreover, whenever a proposed merger violates the above-mentioned screen, regulators consider what are supposed to be mitigating factors for the potential anticompetitive effects of the merger. These include the case of an unusually large number of competitors, under the presumption that the number of firms in a market has a positive effect on competition, as well as the case of a recent trend toward deconcentration in the market where the merger is to take place. Yet in light of the analysis of this paper, it should matter whether the new bank becomes a dominant firm, as the competition effects from other dominant firms should be quite different from those of fringe firms. In addition, a trend towards deconcentration in a market could simply be the result of fringe firm entry, which should not affect significantly the competitive environment of a market. 7 Concluding remarks This paper presents empirical evidence consistent with the predictions of the endogenous sunk cost model of Sutton (1991), with an application to banking markets. In particular, banking markets remain concentrated regardless of market size. Given a prevalent structure of asymmetric oligopoly where dominant banks –defined as those who jointly control over half of the deposits in the market– and fringe firms coexist, the number of dominant banks remains unchanged with market size, with only the number of fringe banks varying across markets. This market structure is sustained by competitive investments in quality, such as branchnetwork, branchstaffingandgeographic diversification, withthelevel of bank quality 22

increasing with market size and, moreover, with dominant banks providing a higher level of quality than fringe banks. Furthermore, banks do not appear to carve out areas within the relevant geographic banking market, but rather compete with each other closely. In terms of the product market, however, dominant and fringe banks appear to focus on a few different sectors. This paper contributes to the empirical literature on the relationships between market concentration, market size, and quality. While the theory of market structure and quality is well developed and offers robust predictions, the body of empirical work documenting them is small. Offering evidence supporting the endogenous sunk cost model, this paper provides a model that can explain the market structure of banking markets. Ellickson (2001) obtains similar findings for supermarkets, suggesting that retail competition may be well characterized by this approach. In terms of the empirical banking literature, this work also represents an attempt toanalyze and measure quality in banking services. Furthermore, the paper sheds light on the empirical finding that larger banks charge significantly higher fees than smaller banks. The findings here indicate that dominant firms, which tend to be large banks, do charge higher fees yet invest more in quality. While it might be presumed that the direct cause of quality is bank size, it appears that quality is the result of banks’ competitive investment in endogenous sunk costs, which gives rise to barriers to entry and allows for market structure nonfragmentation with increases in market size. This, in turn, allows those banks that invest more in quality to hold large market shares and become big banks. Theanalysisofthispaperisalsousefulinthecontextofthebankingliterature, whichhas relied heavily on the structure-conduct-performance paradigm, and which has also affected the way antitrust analysis is carried out. The introduction of quality in models of banking competition, which this work suggests is important, changes the relationship between the number of firms, concentration and competition. The analysis might aid regulators in identifying the relevant variables of analysis as well as asking the appropriate questions. 23

References [1] Amel, D.F. (1997). “Antitrust Policy in Banking: Current and Future Prospects,” in Federal Reserve Bank of Chicago 33rd Annual Conference on Bank Structure and Competition Proceedings. [2] Dell’Ariccia, G., E. Friedman and R. Marquez (1999). “Adverse selection as a barrier to entry in the banking industry.” Rand Journal of Economics, 30: 515-534. [3] Dick, Astrid A. (2002).“Demand Estimation and Consumer Welfare in the Banking Industry.” Finance and Economics Discussion Series, Federal Reserve Board, 2002-58. [4] Ellickson, Paul B. (2001). “Supermarkets as a Natural Oligopoly.” University of Rochester, Simon School of Business Administration, mimeo. [5] Hannan, T.H. (2002). “Retail Fees of Depository Institutions, 1997-2001.” Federal Reserve Bulletin: 405-413. [6] Hannan, T.H. (2001).“Retail FeesofDepositoryInstitutions, 1994-99.”Federal Reserve Bulletin: 1-11. [7] Marquez, R. (2002). “Competition, Adverse Selection, and Information Dispersion in the Banking Industry.” The Review of Financial Studies, 15: 901-926. [8] Radecki,L.J.,J.Wenninger, andD.K.Orlow(1996).“BankBranchesinSupermarkets.” Current Issues in Economics and Finance, Federal Reserve Bank of New York. Vol. 2: 13. [9] Rhoades, S.A. (1996). “Competition and Bank Mergers: Directions for Analysis from Available Evidence.” The Antitrust Bulletin, 41: 339-364. [10] Shaked, A. and J. Sutton (1983). “Natural Oligopolies.” Econometrica, 51:1469-1484. [11] Shaked, A. and J. Sutton (1987). “Product Differentiation and Industrial Structure.” Journal of Industrial Economics, 36:131-146. 24

[12] Sutton, John (1991). Sunk Cost and Market Structure: Price Competition, Advertising, and the Evolution of Concentration. Cambridge: MIT Press. [13] Williams, C.A. (1997). “Banks Go Shopping for Customers.” The Regional Economist, Federal Reserve Bank of St. Louis, October: 12-13. 25

Table 1: PERCENTILES FOR THE NUMBER OF FIRMS ACROSS MSA MARKETS 5% 10% 25% Median 75% 90% 95% Number of banks (Mean=20) 6 7 10 14 22 37 54 Number of branches (Mean=140) 19 27 39 67 152 343 516 NOTE.– Year: 1999. Table 2: DISTRIBUTION OF BANKING MARKETS BY POPULATION SIZE Population Number of MSA markets Percent HHI 100K or less 22 6(cid:1)65 2723 100K-200K 103 31(cid:1)12 1948 200K-500K 106 32(cid:1)02 1863 500K-1M 39 11(cid:1)78 1781 1M-2M 37 11(cid:1)18 1857 2M+ 24 7(cid:1)25 1696 Total 331 100(cid:1)00 NOTE.– Year: 1999. The last column shows the average Herfindahl-Hirschman index for each market category. 26

Table 3: HERFINDAHL INDEX PERCENTILES ACROSS MSA MARKETS 5% 10% 25% Median 75% 90% 95% 934 1124 1432 1793 2240 2817 3417 NOTE.– Year: 1999. Based on deposit shares. Table 4: BANKING MARKETS BY POPULATION AND NUMBER OF DOMINANT FIRMS Number of Dominant Firms Total Population 1 2 3 4 5 6 7 markets <100K 4 14 4 0 0 0 0 22 100K-200K 2 54 35 10 2 0 0 103 200K-500K 6 42 37 14 4 3 0 106 500K-1M 0 17 18 3 1 0 0 39 1M-2M 2 18 11 5 1 0 0 37 >2M 0 13 9 1 0 0 1 24 Total 14 158 114 33 8 3 1 331 NOTE.– Year: 1999. Dominant firms are defined as those whojointly control over half of the deposits in the market. 27

Table 5: OLS REGRESSIONS OF QUALITY ATTRIBUTES AND MARKET SIZE Dependent Variable: Branch Employees Bank’s Number Salary Density per age of per branch states employee Explanatory Variable (i) (ii) (iii) (iv) (v) Ln(population) 0(cid:1)002 15(cid:1)524 0(cid:1)339 0(cid:1)519 2(cid:1)925 (0(cid:1)001) (15(cid:1)426) − (1(cid:1)270) (0(cid:1)131) (0(cid:1)380) ∗ ∗∗ ∗∗ Ln(income p.c.) 0(cid:1)044 120(cid:1)348 11(cid:1)341 0(cid:1)732 13(cid:1)080 (0(cid:1)005) (82(cid:1)811) (6(cid:1)819) (0(cid:1)706) (2(cid:1)040) ∗∗ † ∗∗ Observations 331 331 331 331 331 R-squared 0.27 0.02 0.01 0.07 0.36 NOTE.– Year: 1999. Level of observation: MSA. significant at 10%; † *significantat5%; **significantat1%. Standarderrorsareinparentheses. The dependent variable is a market share weighted average. Salary per employee is in thousands. Branch density is number of branches per MSA square mile. 28

Table 6: QUALITY ATTRIBUTES: DOMINANT VS. FRINGE Dominant Firms Fringe Firms Variable Mean St. Dev. Mean St. Dev. T-Stat Employees per branch 42.33 313.62 25.60 189.89 2(cid:1)19 Branch density 0.0168 0.0228 0.003 0.0115 26(cid:1)43 Bank’s age 93.73 40.32 58.71 44.76 21(cid:1)79 Number of states 4.61 5.14 1.85 2.66 24(cid:1)50 Salary per employee 46,281 10,962 43,283 15,095 5(cid:1)67 Observations 869 5856 NOTE.– Year: 1999. An observation is a bank*market combination. Dominant firms are defined as those who jointly control over half of the deposits in the market. Branch density is number of branches per MSA square mile. 29

Table 7: OLS REGRESSIONS OF SERVICE QUALITY AND DOMINANT VS. FRINGE FIRMS Dependent Variable: Branch Employees Bank’s Number Salary density per age of per branch states employee Explanatory Variable (i) (ii) (iii) (iv) (v) Dominant firm indicator 0(cid:1)013 23(cid:1)629 29(cid:1)717 2(cid:1)493 4(cid:1)914 (0(cid:1)001) (11(cid:1)096) (2(cid:1)967) (0(cid:1)774) (0(cid:1)640) ∗∗ ∗ ∗∗ ∗∗ ∗∗ MSA fixed effects YES YES YES YES YES Observations 6390 6725 6724 6725 6716 R-squared 0.55 0.06 0.27 0.16 0.28 NOTE.– Year: 1999. Robust, adjusted for within-bank dependence standard errors are in parentheses. *significant at 5%; ** significant at 1%. A single observation is a bank*market combination. Dominant firms are defined as those who jointly control over half of the deposits in the market. Salary per employee is in thousands. Branch density is number of branches per MSA square mile. 30

Table 8: CITIES/COUNTIES BY NUMBER OF FIRMS N Avg. # of dominant banks per city/county All cities With at least one dom. firm CITIES/TOWNS Served by exactly 1 bank 3842 0.3 Served by 2-5 banks 3738 1.2 1.5 Served by 6-10 banks 988 2.3 2.3 Served by 11-15 banks 164 2.8 2.8 Served by more than 15 banks 71 2.7 2.7 Total cities 8803 COUNTIES Served by exactly 1 bank 12 0.2 Served by 2-5 banks 166 1.1 1.6 Served by 6-10 banks 355 2.0 2.1 Served by 11-15 banks 189 2.6 2.6 Served by more than 15 banks 161 2.9 2.9 Total counties 883 Year: 1999. 31

Table 9: PRODUCT MIX: DOMINANT FIRM VS. FRINGE Dominant firms Fringe firms Mean St. Dev. Mean St. Dev. T-Stat Assets 59(cid:2) 725(cid:3) Liquidity: Cash / Assets 0(cid:1)0572 0(cid:1)0399 0(cid:1)0532 0(cid:1)0506 2(cid:1)27 Fed. Funds + Securities / Assets 0(cid:1)2322 0(cid:1)1141 0(cid:1)2847 0(cid:1)1439 10(cid:1)30 Loans: Real estate loans / Assets 0(cid:1)3143 0(cid:1)1214 0(cid:1)3741 0(cid:1)1508 11(cid:1)16 Loans to individuals / Assets 0(cid:1)0968 0(cid:1)0708 0(cid:1)0763 0(cid:1)0741 7(cid:1)64 Commercial and industrial loans / Assets 0(cid:1)1660 0(cid:1)0804 0(cid:1)1317 0(cid:1)0918 10(cid:1)44 Leases / Assets 0(cid:1)0266 0(cid:1)037 0(cid:1)0081 0(cid:1)025 18(cid:1)86 Most business loans are small (1=yes) 0(cid:1)0046 0(cid:1)0677 0(cid:1)1342 0(cid:1)3409 11(cid:1)17 Most agricultural loans are small (1=yes) 0(cid:1)0702 0(cid:1)2556 0(cid:1)2427 0(cid:1)4287 11(cid:1)56 Commitment lines / Loans 0(cid:1)6031 0(cid:1)8162 0(cid:1)3187 1(cid:1)6702 4(cid:1)93 Time deposits over 100K / Assets 0(cid:1)0749 0(cid:1)0437 0(cid:1)1082 0(cid:1)0732 13(cid:1)05 Equity / Assets 0(cid:1)0836 0(cid:1)0191 0(cid:1)1021 0(cid:1)0699 7(cid:1)77 Observations 869 5856 Year: 1999. An observation is a bank*market combination. Dominant firms are defined as those who jointly control over half of the deposits in the market. 32

Table 10: PRICES: DOMINANT FIRM VS. FRINGE Dominant firms Fringe firms Mean St. Dev. Mean St. Dev. T-Stat Real estate loans 7.69% 1.08% 7.44% 2.33% 3(cid:1)04 Loans to individuals 2.28% 2.82% 1.34% 2.90% 8(cid:1)90 Commercial and industrial loans 8.88% 6.30% 15.05% 19.75% 9(cid:1)13 Leases 8.00% 15.15% 8.45% 12.27% 0(cid:1)80 Service fees 0.72% 0.34% 0.56% 0.72% 6(cid:1)47 Deposits 3.01% 0.51% 3.21% 0.70% 8(cid:1)15 Observations 869 5856 Year: 1999. An observation is a bank*market combination. Dominant firms aredefinedasthosewhojointlycontrolover halfofthedepositsinthemarket. 33

Table 11: COSTS, RISK AND PROFITS: DOMINANT FIRM VS. FRINGE Dominant firms Fringe firms Mean St. Dev. Mean St. Dev. T-Stat Operating costs / Assets 0.0668 0.0264 0.0739 0.0664 3(cid:1)11 Charge-off losses / Loans 0.0027 0.0022 0.0018 0.0041 6(cid:1)28 Profits / Equity 0.3292 0.1249 0.2359 .4098 6(cid:1)67 Observations 869 5856 Year: 1999. An observation is a bank*market combination. Dominant firms aredefinedasthosewhojointlycontroloverhalfofthedepositsinthemarket. 34

Bank mergers per annum (1993-1999) 500 450 453 438 427 400 368 358 350 344 300 250 200 150 134 100 1993 1994 1995 1996 1997 1998 1999 Figure 1: Lower bound to concentration (1999) 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 10 11 12 13 14 15 16 17 ln(MSA population) xedni lhadnifreH Figure 2: 35

Lorenz Curve for Deposits - selected MSAs by market size- 1.00 0.80 0.60 0.40 0.20 0.00 1 0 0 0 0 0 0 1 2 3 4 5 6 Number of banks (ranked by deposits) fo erahs sa stisoped evitalumuC latot Pocatello, ID Punta Gorda, FL Huntsville, AL Philadelphia, PA Fort Lauderdale, FL Vallejo-Fairfield-Napa, CA Figure 3: 36

### # $T$T##$T$T# $T ##$T#$T$T####$T# $T## # # $T$T$T $T######### $T$T$T#$T##$T$T#$T$T$T####$T$T#$T$T###$T$T$T$T$T$T#######$T$T #$T#$T## ########$T# $T######$T# # ###$T## # #$T####$T## $T # #$T$T$T $T### ###$T #$T#$T# # ## # # $T $T $T# # $T # $T$T### $T $T $T # # $T ###$T#$T##$T#$T # # # ##$T ## #$T # ##$T$T$T#$T$T##$T$T#$T#$T$T#$T#$T# # # # # # ## $T $T$T$T $T # $T $T $T $T $T## $T $T #$T$T$T$T##$T### $T $T ### # # $T# # $T #### $T $T#$T#$T##$T $T$T#$T$T # ### # # # $T $T # # $T #$T #$T #$T #$T ## $T# # # $T # $T$T# $T# # $T #$T # # # # # ## # $T $T$T # # # # # # # # ## # #$T $T $T $T ## # $T# # $T $T # # # ## # # #$T#$T# $T $T # ## # # # # # #$T # ##$T# ##$T $T $T # $T ## $T $T ## # $T$T# #$T# $T $T$T$T# # # #$T $T # # ## $T $T $T # # # # # $T $T ## ## # # $T #$T # # # $T##$T## $T $T $T # # $T # # #$T## $T # $T $T # $T $T $T$T $T $T# $T# # $T#$T # #$T # $T# # #$T # $T # $T# $T $T # $T $T # $T # # # # BOSTON MSA BRANCHES (CIRCLE=DOMINANT, TRIANGLE=FRINGE) Figure 4: 37

APPENDIX: SUMMARY STATISTICS: MSA MARKETS, 1999 Variable Mean St. Dev. Min Max Bank assets 16B 53B 1.4M 323B MSA deposits 370M 1701M 2000 87B Cash / Assets 0.05368 0.04935 0.00000 0.96094 Federal funds + securities / Assets 0.27794 0.14150 0.00000 0.98427 Real estate loans / Assets 0.36634 0.14869 0.00000 0.93675 Loans to individuals / Assets 0.07896 0.07397 0.00000 0.95678 Commercial & industrial loans / Assets 0.13609 0.09114 0.00000 0.84619 Leases / Assets 0.01050 0.02780 0.00000 0.47440 Commitment lines / Loans 0.35557 1.58856 0.00000 112.97 Most business loans are small (1 = yes) 0.11747 0.32201 0 1 Most agricultural loans are small (1 = yes) 0.22037 0.41453 0 1 Time deposits over $100,000 / Deposits 0.10392 0.07100 0.00000 0.82665 Equity / Assets 0.09968 0.06588 0.01055 0.99675 Charge-off losses / Loans 0.00194 0.00395 0.00000 0.13267 Employees per branch 28 210 0 12279 Branch density 0.00511 0.01419 0.00003 0.38544 Bank’s age 63 46 0 215 Salary per employee 43,671 14,660 839 275,429 Number of states in which bank operates 2 3 1 17 Bank operates in at least one rural area 0.4430 0.4968 0 1 Banking holding company indicator 0.8369 0.3695 0 1 Real estate loan rate 0.0747 0.0221 0.0000 0.3414 Loans to individuals rate 0.0149 0.0298 0.0000 0.6345 Commercial & industrial loan rate 0.1418 0.1859 0.0000 4.6667 Lease rate 0.0832 0.1308 -0.0160 4.0000 Service fees 0.0072 0.1155 0.0000 9.4537 Deposit rate 0.0317 0.0075 0.0000 0.1562 Operating costs / Assets 0.0365 0.0314 0.0000 1.3873 Profits / Equity 0.1240 0.1931 -3.4059 7.4284 Number of observations (bank-market) 6725 Constructed on the basis of the Federal Reserve Report on Condition and Income; U.S. Census; Bureau of Economic Analysis. 38

APPENDIX(CONT.):DESCRIPTION OF VARIABLES Variable Description Most business loans are small (1=yes) If all or substantially all of the dollar volumeofloanssecuredbynonfarmnonresidentialpropertiesandcommercialandindustrial loans have amounts of US$100,000 or less Most agricultural loans are small (1=yes) If all or substantially all of the dollar volume of loans secured by nonfarm farmland and loans to finance agricultural production and other loans to farmers have amoutns of US$100,000 or less Employees per branch Number of bank employees / Number of branches Branch density Numberofbranchesinlocalmarket/Square miles of local market Bank’s age Years since beginning of bank’s operations Interest rate on real estate loans Interest income on real estate loans / Loans Interest rate on loans to individuals Interest income on loans to individuals / Loans Interest rate on commercial & industrial loans Interest income on commercial & industrial loans / Loans Interest rate on leases Interest income on leases / Loans Service fees Service charge on deposit accounts / Deposits Deposit interest rate Interest expense on deposits (includes interest on time, savings and NOW accounts) / Deposits Operating costs Expenses including salaries, expenses on premisesandfixedassets,andotherexpenses 39