Cost and profit efficiency of financial conglomerates and universal banks in Europe

Transcription

1 1 Cost and profit efficiency of financial conglomerates and universal banks in Europe Rudi Vander Vennet Department of Financial Economics 2000/81 January 2000 Abstract In contrast to the U.S. where universal banking has been legally prohibited, the Second Banking Directive allows European banks both to form financial conglomerates and hold equity stakes in non-financial companies. This paper analyzes the cost and profit efficiency of European financial conglomerates and universal banks. We find that conglomerates are more revenue efficient than their specialized competitors and that the degree of both cost and profit efficiency is higher in universal banks than in non-universal banks. These results indicate that the current trend towards further de-specialization may lead to a more efficient banking system. An investigation of the equity betas under varying business cycle conditions supports the hypothesis of superior monitoring capabilities on the part of universal banks. Finally, profit regressions suggest that operational efficiency has become the major determinant of bank profitability and that oligopolistic rents have become less prevalent in European banking. JEL G21 Keywords: universal banking, financial conglomerates, cost efficiency, profit efficiency, cost of capital Comments by Allen Berger, Claudio Borio, Jean Dermine, Reint Gropp, Phillipp Hartmann and participants at the 1998 SUERF Conference in Frankfurt and the 1999 CEPR/JFI conference in INSEAD are gratefully acknowledged. Correspondence: Rudi Vander Vennet, University of Ghent, Hoveniersberg 24, 9000 Ghent, Belgium, rudi.vandervennet@rug.ac.be D/2000/7012/02

2 2 Cost and profit efficiency of financial conglomerates and universal banks in Europe Rudi Vander Vennet University of Ghent 1. Introduction Financial conglomerates are financial institutions that offer the entire range of financial services. Next to performing the traditional banking operations, they may sell insurance, underwrite securities, and carry out security transactions on behalf of their clients. Universal banks are allowed to hold equity stakes not only in financial but also in non-financial firms. They may vote the shares they own and, if proxy provisions exist, also those they hold in trust for other agents. Nowadays, many countries permit financial conglomeration and universal banking, including all EU member states and, e.g., Switzerland. As a result of deregulation and intensified competition, the debate about conglomeration and universality is also high on the agenda in the US. And the adoption of a market-based or a bank-based financial system remains an important question in transition economies. The outcome is important for the structure of world banking since the strategic options for banks in terms of functional diversification depend to a large extent on the regulatory environment in which they operate. In the EU, conglomeration and universal banking are allowed by the Second Banking Directive (1989) which has been implemented by all member states. In this directive, the EU has adopted a broad definition of credit institutions, corresponding to the German model of universal banking. As a consequence, banks, investment firms and insurance companies may hold unlimited reciprocal equity participations, implying that there are no limits on the formation of financial conglomerates. The holding of shares in non-financial firms is, however, subject to certain limits. Individual stakes in industrial and commercial firms should not exceed 15% of the bank's capital, while the sum of these participations must remain below 60% of the capital. Within this institutional setting, the monetary union and the introduction of the euro is prompting banks with varying degrees of functional and geographic specialization to restructure. The question is whether a generalized shift to universal banking would benefit the EU economies in terms of efficiency and risk of the financial system. For individual financial institutions the strategic tradeoff is whether becoming universal is optimal to remain competitively viable. In this respect the actual behavior of EU banks displays marked differences. A number of banks are refocusing towards greater specialization. Others are opting for a strategy of diversification, often through a merger with or the acquisition of insurance companies and/or investment firms. A number of recent (mega)mergers in the Benelux and Switzerland are examples of the ongoing bancassurance conglomeration trend (Fortis-Generale Bank, Crédit Suisse-Winterthur). Throughout the 1990s large continental European banks have acquired a series of, often London-based, securities brokers or investment banks. In the US, the debate on the repeal of the Glass-Steagall (1933) provisions separating commercial and investment banking has been intensified in recent years. Already in 1991, a U.S. Treasury Report advocated the abolishment of the barriers between banking and commerce and supported the establishment of Financial Service Holding Companies with insurance and investment banking

3 3 subsidiaries as well as commercial banking subsidiaries. But it was also suggested that banks and their holding companies should not be permitted to hold equity claims on non-financial firms. A number of academic studies have concluded that universal banking would be beneficial for the US. Benston (1994) and Saunders and Walter (1994) argue that a move to universal banking would enhance the static and dynamic efficiency of the financial services sector, without increasing the risks to financial system stability. Until recently, however, the legislator has opposed such an evolution towards universal banking 1. The merger between Citicorp and Travelers may speed up the decision process. In Japan regulatory change is inevitable in the face of the severe structural weaknesses of the financial system that surfaced during the Asian economic turmoil, but its direction remains unpredictable. The arguments in the debate are well known (Saunders, 1994). Proponents of universal banking argue that it provides discipline to corporate management, helps in the restructuring of corporations more efficiently than stock markets, allows economies of scale and scope across financial services, and promotes financial stability and economic development. Opponents question the cost advantages and invoke conflicts of interest and concentration of power as the main drawbacks. However, many authors have noted the absence of empirical studies measuring the gains from the activities of universal banks (Benston, 1994; Saunders, 1994). This paper provides new evidence by investigating the performance and efficiency of European financial conglomerates and universal banks. For this purpose the EU constitutes an ideal setting because deregulation has considerably expanded the authorized functional scope of financial services firms. A number of previous studies have examined universal banking on a country-by-country basis (Saunders and Walter, 1994; Allen and Rai, 1996). However, even in countries where universal banking is allowed, financial conglomerates, universal banks and specialized banks coexist. Therefore, our approach is to analyze EU banks based on their degree of diversification and universality as revealed by their institutional structure and by their operational and financial characteristics instead of only referring to the country in which they are headquartered. In what follows the subsamples of European financial conglomerates, universal banks and specialized institutions are analyzed in terms of their relative cost and profit efficiency. We estimate cost and profit functions and examine the relationship between profitability and a series of relevant variables for different subgroups of European banks. The main findings are that financial conglomerates are more revenue efficient than their more specialized competitors and that the degree of both cost and profit efficiency is higher in universal banks than in non-universal banks. These results indicate that the current trend towards further de-specialization may lead to a more efficient banking system. An analysis of stock market data suggests that the higher observed revenue efficiency of universal banks may be linked to their superior ability to deal with moral hazard through monitoring. Finally, the profit regressions show that operational efficiency has become the major determinant of bank profitability and that oligopolistic rents have become less prevalent in European banking. The paper proceeds as follows. Section 2 reviews the economics of non-specialized banking and discusses the existing empirical evidence. Section 3 offers an operational definition of financial conglomerates and universal banks and describes the data. Section 4 outlines the methodology to test various hypotheses related to cost and profit efficiency. Section 5 presents the empirical results. Section 6 concludes. 1 The Federal Reserve has used its powers under Glass-Steagall to increase the ceiling on the proportion of total revenue a commercial bank can earn from brokering from 10% to 25%. Although US banks are more restricted than European banks, they do undertake a number of nontraditional activities (Rogers 1998).

4 2. The economics of non-specialized banking 4 Economically, the formation of financial conglomerates would be beneficial if there are positive cost and/or revenue effects from combining various financial service activities. Consolidated revenues would be improved if the income-generating capacity of the combined institutions, e.g., through diversification and cross-selling, is enhanced. Similarly, the operating costs of financial conglomerates would be lower relative to specialized banks if integration allows the realization of operational synergy, e.g. through economies of scale and scope. Funding costs may also be lowered due to reputation effects or market power. Universal banks may exhibit superior performance if informational or other advantages associated with equity holdings produce positive spill-overs to the traditional and non-traditional banking activities they undertake. Economies of scale exist if, assuming a constant product mix, a bank faces declining average costs as its size expands. Pure scale economies cannot be invoked as a reason for the formation of financial conglomerates, but they may explain the growth of these financial firms. In fact, since conglomerates and universal banks tend to be relatively large institutions, the scale argument has intuitive appeal as an explanation for growth. Moreover, technological advances may be an obvious catalyst for increased size. Economies of scope capture the effect of a change in a bank's product mix on aggregate costs. If an expanded set of products and services is produced in a more efficient way by financial conglomerates, cost may be lowered. The sharing of inputs such as labor, technology and information across multiple outputs constitutes the major source of such potential cost savings. The presence of economies of scale and scope in banking remains a controversial subject. The early U.S. studies found that economies of scale were exhausted at relatively small output levels (see Clark (1988) for a review). More recently, studies including larger banks have found evidence of scale economies up to the 2-6 billion USD asset range (Noulas, Ray and Miller 1990; Hunter, Timme and Yang 1990). Apparently, regulatory and technological changes have shifted the optimal scale in banking. Shaffer (1988) finds evidence of scale economies up to the $60 billion size range for the 100 largest U.S. banks. Berger and Mester (1997) find substantial unexploited cost scale economies for fairly large sizes of bank in the 1990s, suggesting a change from the 1980s. However, studies of US banks cannot provide evidence on the cost characteristics in nonspecialized financial institutions because regulatory constraints have prohibited conglomeration and universal banking. In a study based on non-u.s. data, Saunders and Walter (1994) find economies of scale up to 25 billion USD in loans for the world's 200 largest banks. Vander Vennet (1994a) finds similar results for a sample of 1500 EU banks. Lang and Welzel (1995, 1996) find scale economies among German universal banks up to a size of 5 billion DEM and significant scale economies for a sample of relatively small Bavarian cooperative banks. Based on similar methods, the bulk of U.S. studies have concluded that economies of scope in banking, if at all present, are exhausted at very low levels of output (see Berger, Hanweck and Humphrey 1987; Mester 1992; Berger, Hunter and Timme 1993). However, Clark (1988) noted that some studies do find evidence of cost complementarities between certain product pairs. Recent studies tend to support this finding (Hughes and Mester 1994). One exception is a study by Kolari and Zardkoohi (1987) in which economies of scope between 10 and 50% for all size ranges and output pairings are reported. In Europe, Muldur (1991) concludes that cost complementarities exist between certain product pairs but they tend to compensate each other resulting in near zero net benefits. For British building societies diseconomies of scope are the prevalent outcome in Drake

5 5 (1992). Lang and Welzel (1995, 1996) report the absence of scope economies in German universal banks, but they do find such economies in small cooperative banks. Saunders and Walter (1994) find important diseconomies of scope between loans and fee-earning business for the world's largest banks, many of which are universal. Scale and scope economies, however, only refer to the static effect of size and activity mix on costs. Faced with a rapidly changing competitive and regulatory environment, the improvement of operational or X-efficiency may be even more important to ensure the competitive viability of banks, especially since a number of studies have documented that technical inefficiencies (excess use of inputs) and allocative inefficiencies (suboptimal input mix) may be large and even dominate scale and product mix economies (Berger and Humphrey, 1991). Increasing competitive pressure and technological advances will force banks to shift to an institutional form that allows maximum X-efficiency. The question is whether financial conglomeration or universal banking offer a sustainable advantage over specialist suppliers. One argument relates to corporate governance and the working of the takeover market. If specialized financial firms (banking, insurance, securities business) are sheltered from acquisition, e.g., due to legal barriers to takeover, inefficient managers are protected and agency costs are high. If cross-activity mergers are allowed, managers of financial firms incur stronger monitoring by the takeover market. Saunders (1994) argues that allowing banks to be acquired by other financial companies or even commercial firms would impose monitoring and create incentives for efficiency and value-maximizing behavior. It would also reduce expense-preference behavior, which has been found to be present in banking (Arnould 1985; Akella and Greenbaum 1988). Often, the formation of a financial conglomerate constitutes an occasion for focused rationalization programs, a phenomenon that has also been observed in EU bank mergers by Vander Vennet (1996). Over the past several years, substantial effort has gone into the measurement of X-efficiency of financial institutions. Berger and Humphrey (1997) survey 130 studies on efficiency, using data from 21 countries, from multiple time periods, from various types of institutions including banks, savings institutions, and insurance companies, and using various efficiency concepts and measurement methods. It appears that the average inefficiency in banking ranges between 20-25% of total costs. Berger, Hancock and Humphrey (1993) find that larger banks are more efficient. However, since there are large disparities between banks of similar size, they indicate that the way individual banks are run is much more important than their form of organization or size, as such. These data, however, mostly cover U.S. banks and make no explicit distinction between universal and non-universal banks. One of the few European studies compares the efficiency of banks in Norway, Sweden and Finland. Berg et al. (1993) find that Swedish banks are generally more efficient than Norwegian and Finnish banks. All of these countries have universal banking. Allen and Rai (1996) document wide variations in country-specific efficiency for 194 banks in 15 countries. They also find that large banks in separated banking countries (i.e., countries that prohibit integration of commercial and investment banking) are significantly less efficient than other bank groups for the period In a review article, Benston (1994) concludes that the data on economies of scale and scope and X-efficiency indicate some advantage for universal banks over specialized banks. However, considering that specialized banks are able to survive in direct competition with universal banks, the author concedes that the efficiency advantages of neither form of banking appear to be overwhelming. Clearly, a more direct assessment of efficiency in universal versus non-universal banks is needed.

6 6 Next to potential cost advantages, a more efficient combination of financial products and services may also entail revenue gains. The ability for conglomerate financial services providers to market and distribute the full range of banking, securities and insurance services may increase their earnings potential. On the demand side, customers may value a bundled supply of financial services more highly than dispersed offers by separate firms for reasons of transaction and information costs. Canals (1993) finds that the increased revenues obtained from new business units has contributed significantly to improving bank performance in recent times. Gallo et al. (1996) find that mutual fund activities also increased the profitability of banks. Moreover, the combination of banking, insurance and securities activities may lead to a more stable profit stream, since the revenues stemming form different products in a conglomerate organization are usually imperfectly correlated. Saunders and Walter (1994) find that expanding banks' activities reduces risk, with the main risk-reduction gains arising from insurance rather than securities activities. Boyd et al. (1993) find that simulated mergers of BHCs with life insurance or property/casualty insurance firms may reduce risk, but that mergers of BHCs with securities firms would likely increase risk. Benston (1989) reports that returns for combined commercial and investment banking would be significantly higher, without a compensating increase in overall risk. A full universal bank may also optimize the efficiency of the information exchange with corporate customers. Financial intermediaries deal with incomplete and asymmetric information by becoming delegated monitors (Diamond, 1984). In a bank-based corporate finance system, the information flowing from firms to banks is often produced through multiple contacts within the framework of a long-term lender-borrower relationship. Hence, relationship banks should have an informational advantage in the monitoring of moral hazard. Empirically, it has been observed that companies may benefit because bank monitoring may overcome problems of financial constraints and asymmetric information (see James, 1987; James and Wier, 1990, Lummer and McConnell, 1990). In a universal bank system, the bank-firm relationship can be enhanced by adding finance-related services (issuing and placement of securities, advisory services, risk management facilities, guarantees and contingent credit lines). Once the banker also becomes a shareholder, sometimes including a presence on the board of directors, the full insider status should improve information flows even further. Monitoring then becomes a variable cost to the bank (Steinherr and Huveneers 1994). Using data for banks in eight countries (including German, Dutch and Swiss universal banks), Dewenter and Hess (1998) find evidence that the equity market risk of transactional banks relative to relationship banks rises during economic contractions. These results support the notion that relationship banks monitor moral hazard more effectively than transactional banks. 3. Data and definition of financial conglomerates and universal banks Following the previous discussion the purpose is to classify European banks into more or less homogeneous categories in order to allow meaningful comparisons between specialized financial institutions and their more diversified competitors. However, we do not classify banks as universal based on their country of origin because even in countries permitting universality, many banks opt to remain (or become) specialized. Our preferred approach is to make the actual classification of each bank depend on its type of corporate organization and on a number of financial indicators derived from the annual statements. This procedure leads to a double regrouping of the sample of European banks : (1) financial conglomerates versus specialized banks and (2) universal banks versus specialized banks.

7 7 Obviously, the legal environment in which banks operate determines the organizational form they can adopt. The liberalization of the range of activities that banks can engage in, either directly or through ownership stakes, has been most fully achieved in the EU. Both the remaining legal distinctions between various types of credit institutions (e.g., commercial banks, savings banks, credit unions) as well as the legal demarcations between commercial banking, securities business and insurance have been abolished by the Second Banking Directive enacted in As a result, banks can perform securities and insurance activities directly or through subsidiaries. The Directive also regulates the holding of equity stakes in non-financial enterprises. Yet, despite regulatory harmonization, the corporate structure of financial services companies still differs across countries, mainly reflecting historical differences (Borio and Filosa, 1994) 2. There are roughly three organizational ways of combining commercial banking, investment banking and insurance: (1) in house, via a department of the bank, (2) via a separately capitalized subsidiary of the bank, or (3) via a separately capitalized affiliate of the bank holding company (see Saunders and Walter, 1994). A fully-fledged financial conglomerate would combine banking, insurance and securities activities as in-house departments. Contrary to conventional wisdom, German universal banks do not conform to this type. They conduct only their merchant banking and securities operations in house while insurance, mortgage banking and investment funds are usually supplied through affiliate companies. Moreover, many German banks do not in practice carry out the full range of universal banking business (Edwards and Fischer, 1994). The same if true for other countries in which universal banking is legally allowed. As a result, the classification of banks should not be based on their country of origin or their institutional type. Our preferred approach is to delineate banks by their revealed degree of functional diversification and universality, based on observed organizational and financial characteristics. We distinguish three major areas of financial services: traditional banking, insurance and securities-related activities. We define specialized banks as those mainly engaged in traditional intermediation activities (transformation of deposits in loans). Financial conglomerates are defined as financial services firms which conduct at least two of the three major activities. While some operations may be exercised through a subsidiary, the criterion is that the parent institution must consolidate non-traditional bank activities in its annual statement. Conglomeration is thus associated with the potential conduct of a range of financial services comprising deposittaking and lending, trading of financial instruments and their derivatives, underwriting of new debt and equity issues, brokerage, investment management, and insurance. Universal banks are defined as diversified banking institutions that also hold equity stakes in non-financial companies. This definition follows Steinherr and Huveneers (1994) who assert that the key feature of universal banking is the range of activities and, in particular, the holding of equity shares large enough to monitor corporations as equity owner 3. Operationally, a financial institution is labeled a financial conglomerate (FC) when two conditions are met. First, the bank is engaged in non-traditional banking through an in-house department or 2 3 In most European countries, the ability to perform investment banking functions had been broadened since the mid-1980s. In Germany and several Nordic countries, few if any restrictions have existed. The deregulation of banking and insurance combinations has usually proceeded more slowly. The rules governing the production or the in-house provision of insurance services have usually been less flexible than those related to the distribution of insurance products or ownership linkages between banks and insurance companies. Allen and Gale (1995) distinguish between transactional and relationship banks. Relationship banks, such as the German, Dutch and Swiss main banks, provide both debt and equity financing to firms, have long-lasting ties with them, serve on corporate boards and remain committed in periods of financial distress.

8 8 consolidates at least one subsidiary active in investment banking and/or insurance. For this type of banks, the existence of a group structure, fully integrated or not, is the main feature. This condition was checked for each bank based on an in-depth analysis of its organizational structure in the annual statement 4. Second, the ratio of non-interest income in total revenues should exceed 20%. We argue that the proportion of non-interest income in total revenues is a useful indicator of relative diversification because the fee-income earned on non-traditional banking activities such as insurance and securities trading is registered as non-interest income in the annual statement. The diversification threshold of 20% is added to ensure that the non-traditional banking activities are considered by the management as strategically important. In the sample of European banks, 176 institutions fulfill both conditions and are classified as conglomerates 5 (see table 1). Universal banks (UB) are those institutions in which equity stakes in non-financial companies account for more than 1% of total assets 6. Moreover, we require that universal banks have a ratio of noninterest income in total revenues higher than 5%. This minimum diversification threshold should ensure that activities for which information advantages are expected to be present (underwriting, risk management, and other fee business) are effectively undertaken by the bank. Consistent with regulatory practice, we make no distinction according to institutional type. Hence, all subsamples may consist of a mixture of commercial banks, cooperative banks, savings institutions, and government-owned banks. As can be seen in table 1, the full sample consists of 2375 EU banks from 17 countries for which all the variables were available from their published annual statements for the years 1995 and Together, these banks cover more than 85% of aggregated bank assets in their respective countries. Average values for are used in order to alleviate the effect of idiosyncratic events. All nominal amounts were converted into ECU 8. The choice of the time period is motivated by the fact that conglomeration and universal banking are allowed at the latest since 1993 onwards in all EU countries. In many member states, however, conglomeration and/or universal banking have a longer tradition. As a consequence, if a bank is identified as a conglomerate or a universal bank in , this should reflect a deliberate strategic choice for which sufficient financial and managerial resources have been mobilized. The period is also relatively neutral with respect to the macroeconomic environment since banks were facing comparable business cycle and monetary policy conditions across the various countries 9. The bank crises that occurred in a number of countries at the beginning of the 1990s were also largely resolved 10. This is important because loan 4 I thank the bank analyst team of Fortis Bank for assistance Note that we only consider bank-initiated conglomeration, not insurance firms or investment banks expanding into banking. Borio and Filosa (1994) mention the existence of some 200 banking/insurance groups operating within the EU. Universal banks have no incentive to hold all the shares of a company because of the potential loss in the case of bankruptcy. Moreover, equity investments are counted as risk assets and are subject to capital adequacy rules. Equity stakes in the 5-20% range seem to be optimal and, indeed, are observed in Japan and Germany (Steinherr and Huveneers, 1994). Hence, using a threshold of total participations of more than 1% of assets is a reasonable cutoff. The countries are the EU member states (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxemburg, Netherlands, Portugal, Spain, Sweden, UK) plus Norway and Switzerland. Data were obtained from the Fitch/IBCA database. We eliminated all observations from the original sample in which one of the variables is more than 2.5 standard errors away from its mean value in that year. This produces a sample of 2375 financial institutions. The exchange rates of the European countries vis-à-vis the ECU is traditionally much more stable than the $ exchange rate. Over the period the average $/ECU rate was The business cycles are not identical. However, most countries were adhering to the EMU convergence criteria imposed by the Maastricht Treaty. As a result positive yield curves and positive economic growth rates were generally observed in most countries in the period.

9 9 quality and loan losses may affect observed efficiency and profit levels, but the data do not permit us to control for the bank's asset quality. Table 1 presents the bank samples and their country composition. The universal bank sample contains 1066 institutions of which 436 are German. While this finding is not surprising, it may introduce a German bias in the results for universal banks. Therefore we conduct the empirical analyses both for the full sample of universal banks and for the non-german subsample. 4. Methodology This paper investigates whether or not we can find structural differences between universal banks, financial conglomerates, and their more specialized competitors in the EU. The theoretical considerations elaborated in both previous sections suggest a number of testable hypotheses related to cost and revenue effects of conglomeration and the competitive viability of specialized versus universal banks. Three approaches are used : (1) The cost characteristics (scale and scope economies, X-efficiency) are analyzed for the various bank types based on a stochastic cost frontier. (2) The revenue and profit dynamics are investigated by means of a stochastic profit function. And (3) the relationship between profitability and a series of market and bank characteristics is investigated for specialized versus non-specialized banks. We also analyze stock market data for universal banks to investigate the link between profitability and their ability to deal with moral hazard. Combined, the results from these analyses should provide evidence regarding the alleged superiority of conglomerates and universal banks Cost efficiency First, it is asserted that there may be differences between specialized and non-specialized banks with respect to the degree of operational efficiency. To test this conjecture we estimate a cost function for the different types of banks. Cost efficiency provides a measure of how close a bank's actual cost is to what a best-practice institution's cost would be for producing an identical output bundle under comparable conditions. The measure is usually derived from a cost function in which costs depend on the prices of inputs (p), the quantities of outputs (y), risk or other factors that may affect performance (z), and an error term ε. The function can be written as C = f(p,y,z) + ε in which ε is treated as a composite error term ε = u+v, where v represents standard statistical noise and u captures inefficiency. In the parametric methods, a bank is labeled inefficient if its costs are higher than a best-practice bank after removing random error. The methods differ in the way u is disentangled from the composite error term ε. Here, we use the stochastic cost frontier as proposed by Aigner et al. (1977). In general, the non-parametric methods are less suitable because they assume away noise in the data and luck. But for our purpose, the most important drawback is that these methods generally ignore prices and, thus, can only account for technical inefficiency related to using excessive inputs or producing suboptimal output levels. As Berger and Mester (1997) observe, these methods cannot compare firms that tend to specialize in different inputs or outputs because it is impossible to compare input and output configurations without the benefit of relative prices. Moreover, Berger and Mester (1997) use the distribution-free approach as well as the stochastic frontier approach for both the translog and the Fourier specification of the cost and profit function. They conclude that the empirical findings in terms of either average industry efficiency or ranking of individual bank are similar across methods. 10 Except in a number of individual cases such as Crédit Lyonnais.

10 10 The random error term (v) is assumed to be normally distributed and the inefficiency term (u) is assumed to be one-sided. We tried both the half-normal and the exponential distribution but since the results were similar we only report those based on the half-normal distribution. The inefficiency factor (u) incorporates both allocative inefficiencies from failure to react optimally to changes in relative input prices, and technical inefficiencies from employing too much of the inputs to produce the observed output bundle. The log-likelihood function is given by where (1) i = ui + vi, σ = σu + σ v, λ = σu / ε σ (2) N is the number of banks and φ(.) is the standard normal cumulative distribution function. Jondrow et al. (1982) show that bank-specific estimates of inefficiency can be obtained as the mean of the conditional distribution 11 v A Farrell-type measure of operational efficiency can be calculated as CEFF = e -u. A CEFF score of 0.8 would mean that the bank is using 80% of its resources efficiently or alternatively wastes 20% of its costs relative to a best-practice bank. For the functional form of f(p,y,z) we tried both the standard translog and the Fourier-flexible specification (see McAllister and McManus 1992; Mitchell and Onvural 1996; Berger and Mester 1997). The Fourier functional form augments the translog by including Fourier trigonometric terms. It is a global approximation because the sin and cos terms are mutually orthogonal, so that each term aids in fitting the function closer to the true path of the data. But while formal tests indicate that the Fourier terms are jointly significant, the statistical fit, and both the average levels of measured efficiency and their dispersion are very similar for both functional forms. Hence, the results reported here are those for the following translog specification (3) 11 As Mester (1996) observes, the conditional mean is an unbiased but inconsistent estimator of u i since regardles of the number of observations, the variance of the estimator remains nonzero.

11 11 (4) where C is total costs, y i the output quantities, p j the input prices and z is financial capital. Consistent with our research topic we examine two specifications with a different set of outputs. The first specification is based on the intermediation approach and uses two traditional banking outputs (loans and securities) and three input prices (the cost of labor, physical capital and deposits). The price of labor is obtained by dividing salaries and other personnel expenses by total assets 12. The cost of fixed capital is calculated as depreciation and occupancy expenses divided by net fixed assets. The price of deposits is the ratio of interest expenses over interest-bearing liabilities. Since we study financial institutions that have expanded their scope beyond commercial bank activities, we also estimate a specification where traditional and non-traditional banking activities are treated as outputs, measured, respectively, as total interest income and total noninterest revenues. The treatment of non-interest income as a nontraditional bank output is strongly advocated by Rogers (1998) who reports that the traditional specification tends to understate measured efficiency. In both specifications we include financial capital to capture default risk and the risk preferences of bank management (see Mester, 1996; Berger and Mester, 1997). In all estimations the usual symmetry and linear homogeneity restrictions are imposed a priori Profit efficiency A second type of analysis focuses on the alleged superiority of financial conglomerates and universal banks in terms of revenue and profit efficiency. Diversification may increase the revenuegenerating capacity of conglomerates while relationship-specific advantages could increase revenues of universal banks. We investigate this issue by estimating a profit function for the various types of banks. As in the parametric cost function approach, a bank is labeled inefficient if its profits are lower than the best-practice bank after removing random error. In other words, profit efficiency measures how close a bank comes to generating the maximum obtainable profit given input prices and outputs. We follow Berger and Mester (1997) and use the concept of alternative profit efficiency which relates profit to input prices and output quantities instead of output prices. Thus, output is held constant while output prices vary and may affect profits. We opt for this specification in our modeling of European bank profitability because (1) output prices cannot be measured accurately, (2) there may be unmeasured differences in the quality of banking services, and (3) output markets are not perfectly competitive so that banks may exercise some degree of market power. Condition (2) is particularly relevant for the research topic of this paper because service quality may be one of the means by which specialized and non-specialized banks try to distinguish themselves from each other. If customers are willing to pay for high-quality services, the offering banks should be able to earn higher revenues that compensate any excess expenditures and remain competitively viable. Specified in this way, the profit efficiency measures will not penalize high-quality banks whereas the cost efficiency measure might. The occurrence of condition (3) cannot be ruled out because there is evidence indicating that a number of European banking markets may be characterized by less than perfect competition (see Molyneux and Thornton 1992; Vander Vennet 1994b). Finally, alternative profit efficiency compares the ability of banks to 12 Unfortunately, data on the number of employees is lacking for most banks.

12 12 generate profits for the same level of outputs and thus reduces the scale bias that might be present when output levels are allowed to vary freely (Berger and Mester, 1997). As a result, the profit function uses essentially the same specification as the cost function. The dependent variable is now ln(π + π min + 1), where π min is the absolute value of the minimum value of π in the appropriate sample. In practice, the constant term π min +1 is added to every bank's profit so that the natural log is taken of a positive number. This adjustment is necessary since a number of banks in the sample exhibit negative profits in the sample period. The dependent variable is ln(1)=0 for the bank with the lowest value of π. We calculate π as all interest and non-interest earnings minus interest and operating costs. The explanatory variables remain unaltered. Consistent with our research topic we report the results for the specification based on the output-mix combining traditional and non-traditional bank activities. This produces a measure of profit efficiency (PEFF for all banks in the sample. A PEFF of 0.8 would mean that a bank is actually earning 80% of bestpractice profits or that the bank is losing 20% of possible profits due to excessive costs, deficient revenues, or both Bank profitability An alternative approach is to relate bank profitability to a series of relevant bank and market characteristics and determine whether the findings are different for financial conglomerates and universal banks as opposed to specialized banks. The relationship between bank performance and market structure has traditionally been analyzed within the structure-conduct-performance (SCP) paradigm. The assumption is that concentration facilitates collusion, which in turn leads to excess profits. The major antagonistic theory is the efficient-structure hypothesis asserting that differences in profit levels are attributable to differences in operational efficiency across banks. The empirical testing of these theories has usually been based on the following equation where CONC is a concentration measure, MS is market share and X represents a vector of control variables. It is assumed that banks with superior efficiency will increase their market share which automatically leads to a higher market concentration. If b 1 is positive and dominates b 2, this is interpreted as evidence in favor of the collusion hypothesis. If b 2 is found to be significantly positive, the efficient-structure explanation would be supported. Most US studies have found that once market share is included in the empirical analysis the concentration variable loses its explanatory power (see Smirlock 1985; Evanoff and Fortier 1988). Empirical evidence about European banking markets is scarce. Bourke (1989) finds evidence in favor of the collusion hypothesis for a sample of European, North American and Australian banks. Molyneux and Thornton (1992) also find a statistically significant positive correlation between concentration and bank returns for a sample of European banks. Lloyd-Williams, Molyneux and Thornton (1994) report that the SCP-relationship holds in the Spanish banking market. Vander Vennet (1994b) incorporates an efficiency measure in the model and finds that collusion is predominant in a number of EU countries. However, these empirical tests may fail to discriminate between the various theories because some observed outcomes are consistent with more than one hypothesis. Berger (1995a) and Goldberg and (5)

13 13 Rai (1996) consider four different explanations. The traditional SCP asserts that the positive relationship between profits and market structure reflects non-competitive pricing behavior in more concentrated markets. A second theory is the relative-market-power hypothesis (RMP) which states that only firms with large market shares are able to exercise market power and earn abnormal profits. Alternatively, two efficiency explanations may account for the positive relationship between profits and either concentration or market share. The X-efficiency version asserts that firms with superior management or production technologies have lower costs and subsequently reap higher profits. Since these firms are also assumed to gain larger market shares, the market may become more concentrated as a result of competition. The scale-efficiency version allows that some firms simply produce at a more efficient scale than others, leading to lower unit costs and higher profits. Again, these firms are assumed to increase their market share, which would lead to higher market concentration. Under both efficient-structure hypotheses the positive profit-structure relationship is spurious. Finding a positive and dominating coefficient estimate for market share and an insignificant coefficient for concentration would support the relative-market-power explanation. However, one could also argue that this finding is consistent with the efficient-structure hypothesis because market share may be positively related to efficiency. One way to avoid this confusion is to include measures of concentration, market share, X-efficiency and scale efficiency simultaneously in the profit equation. The basic equation then becomes i ( CONCm MSi, EFFi, SCALEi, X m i ) ε i π f, + (6) =, where CONC m is a measure of market concentration, MS i is the bank's market share, EFF i measures operational efficiency, SCALE i is a proxy for any scale advantages, and X m,i represents a vector of market and bank-specific control variables. The efficient-structure hypothesis predicts positive signs for EFF and SCALE and zero coefficients for CONC and MS. A necessary condition for the efficient-structure hypothesis to hold is that efficiency affects both MS and CONC. These conditions will also be tested. Positive values for CONC and/or MS would lend support to the market power explanations. Based on this setting Goldberg and Rai (1996) fail to find a positive relationship between concentration and profitability and find weak support for the efficientstructure hypothesis for a sample of large banks located in 11 European countries for the period The variables are defined as follows. We use three measures of performance as the dependent variable: ROA is net income / total assets, ROE is net income / total equity, and IM is net interest margin / total assets. ROA and ROE have been used in most previous studies, they are calculated on a before-tax basis because of differences in taxation across countries. In addition, the interest margin is used as a proxy for the pricing behavior of banks for deposits and loans. If banks are able to exert market power, IM will be higher due to lower deposit rates, higher loan rates, or both. Our preferred measure for concentration is the Herfindahl index (HERF) because it provides information on the dispersion of market shares in national banking systems 13. The market share (MS) of each bank is calculated as its share in total domestic assets. Two proxies are used to capture X-efficiency. CEFF is the bank's estimated level of technical efficiency based on the stochastic cost frontier described above. However, the standard error of the CEFF estimate is not accounted for in the subsequent regression. Therefore we prefer the cost/income ratio (COSTINC) 13 Another popular indicator is the three-bank concentration ratio. However, focusing exclusively on the three largest banks is arbitrary for most European banking markets since the group of large banks, typically operating through nationwide branch networks, is usually larger. Treating countries as relevant markets may still be relatively appropriate in view of the absence of branching restrictions and the limited importance of foreign entry. This, however, may change after the introduction of the euro.

14 14 calculated as operating costs divided by gross revenues for each bank. Analysts and regulators view this indicator as the key measure of bank efficiency. Scale efficiency (SCALE) is also derived from the cost function. In order to eliminate potential outliers, we assign each bank the SCALE value obtained for the size class to which it belongs, based on the estimated group-specific cost frontier (see equation 8). Since we want to test differences in profitability across subgroups of specialized and nonspecialized banks, the crucial variables are a set of dummy variables used to identify the appropriate bank type. For financial conglomerates the FC dummy equals 1 when a bank is identified as a financial conglomerate, UB=1 for universal banks. The dummy variable is zero for specialized banks in each category. Finally, we include a number of control variables that have been used in the literature. Vander Vennet (1994) finds that the proportion of demand and savings deposits in total deposits (DEMSAV) helps explain differences in ROA 14. Berger (1995b) has shown that there is a positive relationship between ROA and the capital-assets ratio (CAR) for US banks in the 1980s Capital market evidence One possible reason why universal banks may be more profit efficient than specialized banks refers to the hypothesis that universal banks have superior monitoring capabilities. We explore this conjecture by investigating the risk exposure of universal versus specialized banks under varying business cycle conditions. In general, banks face problems of adverse selection and moral hazard caused by asymmetric information between the bank and its customers. Banks can reduce adverse selection by screening potential borrowers and may limit moral hazard by monitoring them. Universal banks provide both debt and equity financing to their corporate clients, develop longterm relationships with them and, sometimes, serve on the board of directors (Allen and Gale 1995). Transactional banks, on the other hand, only provide and monitor loans and have limited interference in the management of the corporations they lend to. The closer ties with their clients and repeated lending may give universal banks access to private information which may improve the effectiveness of their monitoring efforts. The biggest advantage of universal banks may be in the ex post monitoring of firms facing financial distress because they can better judge the true prospects of the firm and may build up renegotiation reputation (Chemmanur and Fulghieri, 1994). If universal banks are better able to deal with financial distress their cash flows will be less affected by adverse economic conditions. Specialized banks, on the other hand, are expected to be more vulnerable to economic fluctuations. Hence, the incidence of moral hazard can be conditioned on the evolution of the business cycle. The difference in the ability to deal with moral hazard should be reflected in the cost of capital of universal versus specialized banks over the business cycle. Following Flannery and James (1984) and Dewenter and Hess (1998) the empirical analysis uses an expanded market model. The cost of capital of a bank is assumed to depend on the exposure to market-wide risk plus premiums for the bank s exposures to default risk and yield curve risk (time indices are suppressed): R in = α + β * D * R + β * D * R, + β * DEF + β * TERM + ε (7) i iu u mn id d m n if n ir n i 14 This suggests that banks with access to a stable and relatively cheap pool of core deposits have a systematic profit advantage. The variable is only used in specifications with ROA and ROE as the dependent variable; IM is used to explicitly capture deposit pricing behavior.