THE PROFITABILITY OF EUROPEAN BANKS: A CROSS-SECTIONAL AND DYNAMIC PANEL ANALYSIS*

Transcription

1 The Manchester School Vol 72 No. 3 June THE PROFITABILITY OF EUROPEAN BANKS: A CROSS-SECTIONAL AND DYNAMIC PANEL ANALYSIS* by JOHN GODDARD University of Wales Swansea PHIL MOLYNEUX University of Wales, Bangor and JOHN O. S. WILSON University of St Andrews The profitability of European banks during the 1990s is investigated using cross-sectional, pooled cross-sectional time-series and dynamic panel models. Models for the determinants of profitability incorporate size, diversification, risk and ownership type, as well as dynamic effects. Despite intensifying competition there is significant persistence of abnormal profit from year to year. The evidence for any consistent or systematic size profitability relationship is relatively weak. The relationship between the importance of off-balance-sheet business in a bank s portfolio and profitability is positive for the UK, but either neutral or negative elsewhere. The relationship between the capital assets ratio and profitability is positive. 1 INTRODUCTION Competition in European banking markets has intensified significantly in recent years (De Bandt and Davis, 2000). Deregulation, technological change and the globalization of goods and financial markets have affected all aspects of the operation of banks, and accordingly have impacted on profitability. Banks are now able to participate in what were previously regarded as inaccessible domestic and foreign markets. This has caused the lines of demarcation between different types of bank (and other financial sector institutions) to become blurred, and has led to greater uniformity in the types of financial services and products available to customers. In addition, technological developments have transformed the possibilities for economies of scale and scope. As part of the globalization phenomenon, foreign bank involvement in domestic banking markets has increased, intensifying competition and reducing margins. *Manuscript received ; final version received We are grateful to two anonymous referees for helpful comments on an earlier draft of this paper. The usual disclaimer applies. Phil Molyneux acknowledges the financial support of the Foundation BBVA. Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK, and 350 Main Street, Malden, MA 02148, USA. 363

2 364 The Manchester School Banks have responded to rising competitive pressure by offering a wider range of products and services and conducting a significant proportion of their business off balance sheet (OBS). Demsetz and Strahan (1997) examine the role of diversification in US banking and find that large banks had lower capital reserves and were more active than their specialized counterparts in high risk lines of business such as derivatives. Klein and Saidenberg (1997) find that diversified banks were less profitable on average. Hughes et al. (1999) find that while growth through product and geographic diversification reduced bank risk, efficiency tended to improve as a result of geographic diversification. Goddard et al. (2001) highlight diversification and other trends in European banking. Many banks have increased in size significantly, reduced average costs and diversified in order to maintain competitiveness on a Europe-wide scale. The economics of banking literature acknowledges various determinants of bank profitability. These include the size of the bank; the extent to which the bank is diversified; the attitude of the bank s owners and managers towards risk; the bank s ownership characteristics; and the level of external competition the bank encounters (McKillop and Ferguson, 1993; Rhoades, 1997; Goddard et al., 2001). We use cross-sectional and dynamic panel estimation to investigate selected determinants of profitability in six major European banking sectors: Denmark, France, Germany, Italy, Spain and the UK, for the period The results suggest that despite intensifying competition there is still significant persistence of abnormal profit from year to year. Although there are some significant size profit relationships in some of the estimations, overall the evidence for any consistent or systematic size profitability relationship is unconvincing. The relationship between the relative size of a bank s OBS portfolio and its profitability is positive for the UK, but negative for some other countries, where banks seem to have experienced mixed results from diversification into OBS activity. The relationship between the capital assets ratio (CAR) and profitability is positive. Finally, although in Germany savings and cooperative banks underperformed relative to commercial banks, there is little evidence of a systematic relationship between ownership type and profitability elsewhere. The rest of the paper is structured as follows. Section 2 provides a selective review of previous theoretical and empirical literature on the determinants of bank profitability. Section 3 describes the empirical model. Section 4 describes the data set and estimation techniques and presents the estimation results. Section 5 concludes. 2 THE DETERMINANTS OF BANK PROFITABILITY Most early research into the determinants of the performance of banks was based on the structure conduct performance (SCP) paradigm, and focused on the interpretation of a positive empirical relationship between concentra-

3 The Profitability of European Banks 365 tion and profitability. According to the collusion hypothesis, a small number of banks may be able to collude either implicitly or explicitly, resulting in higher interest rates charged on loans, lower rates paid on deposits, higher fees and so on. Collusion is more difficult if the number of banks is large. In contrast, according to the efficiency hypothesis a positive concentration profitability relationship may reflect a positive relationship between size and efficiency. It is therefore uncertain whether the high profits of large banks are a consequence of concentrated market structures and collusion, or superior production and management techniques that reduce costs, creating high returns. The existence of the competing collusion and efficiency interpretations spawned a substantial body of research which sought to resolve the issue on empirical criteria. Smirlock (1985), for example, finds a positive relationship between market share and profitability; an insignificant relationship between concentration and profit; and a negative relationship between the interaction of concentration and market share with profit. These findings suggest rejection of the collusion hypothesis. 1 Rhoades (1985), on the other hand, suggests that a positive empirical relationship between market share and profit does not reflect differences in efficiency between banks, but rather advantages arising from product differentiation, enabling some banks to raise prices. Berger (1995a) finds that US bank profitability was positively related to market power (and x-efficiency) during the 1980s. The level of external competition a bank encounters is likely to be an important determinant of its profitability, in the short term at least. Empirical research based on the SCP paradigm often uses observable industry structure indicators (such as concentration ratios) to measure the degree of competition, but without any guarantee that the observed values of such measures represent the market equilibria that are the focus of the corresponding theory (Brozen, 1971). Meanwhile the contestable markets literature emphasizes the influence of potential as well as actual competition on profitability (Baumol et al., 1982). Potential competition depends largely on the size of entry barriers, and is not directly observable. Similar reasoning motivates the empirical persistence of profit literature, in which the degree of first-order serial correlation in firm- or industry-level time-series profit data indicates the speed at which competition causes above- or below-average profit in one year to dissipate subsequently (Mueller, 1986, 1990; Geroski and Jacquemin, 1988). While there is an extensive manufacturing persistence of profit literature, only a handful of studies investigate persistence of profit in banking. Levonian (1993) and Roland (1997) find that US bank profits converged to average values more slowly than is suggested by most manufacturing studies. Berger et al. (2000) find that barriers to competition and 1 Evanoff and Fortier (1988), Bourke (1989) and Amel and Froeb (1991) also tend to favour the efficiency hypothesis.

4 366 The Manchester School information constraints influence the degree of persistence. As part of a bivariate analysis of the interactions between bank profitability and growth indicators, Goddard et al. (2004) also find evidence of significant persistence in bank profits data. Interest in the efficiency of banks has spawned a substantial literature examining economies of scale (size) and scope (product mix), and technical, economic and x-efficiency. Most researchers using data up to the mid-1980s find that scale economies were evident at low asset size levels but became exhausted as size increased. 2 Using data, Scholtens (2000) finds that the profits of European banks classed as small (in terms of assets) grew faster than those of the larger banks, while Williams (2003) finds the opposite for foreign banks based in Australia. Using data, Goddard et al. (2001) find that scale economies and productive efficiency in European banking were positively related to profits, but ceteris paribus smaller banks were more profitable than their larger counterparts. Berger and Humphrey s (1997) review finds consistent evidence that large banks are more efficient on average than small ones, but it is less clear whether large banks benefit significantly from scale economies. Profitability is more likely to be enhanced by emulating industry best practice in terms of technology and management structure than by increasing size per se. As margins and fees tended to tighten in many domestic banking markets during the 1980s and 1990s, many banks responded by implementing strategies of product diversification, merger and overseas expansion in an attempt to defend their profitability (Santomero and Eckles, 2000; Hughes et al., 2002). Financial services diversification allows managers to offer a wider range of services and spread the risks of lending across a larger number of asset categories, reducing monitoring costs (Diamond, 1984). For many European banks the importance of OBS business, including loan commitments, letters of credit, derivatives and the creation of marketable securities from bundles of smaller loans, increased rapidly during the 1990s. Using 1995 and 1996 European banking data, Vander Vennet (1998) finds that financial conglomerates were more efficient on average than their specialized competitors. In principle a bank s capacity to absorb unforeseen losses determines its level of risk. Several ratios are commonly used to proxy for risk, including the CAR and the liquidity ratio. In theory an excessively high CAR could signify that a bank is operating over-cautiously and ignoring potentially profitable investment opportunities. A bank holding a relatively high pro- 2 For example, Smirlock (1985) finds that the larger US banks were less profitable than the smaller ones during the period Berger and Hannan (1989) find a negative relationship between the extent to which large banks dominate and the interest rates paid to depositors for the period For the same period, however, Jackson (1992) finds that, in local US markets dominated by large banks, depositors were paid relatively high interest rates on savings.

5 The Profitability of European Banks 367 portion of liquid assets is unlikely to earn high profits, but is also less exposed to risk; therefore shareholders should be willing to accept a lower return on equity. Rhoades (1985) finds a positive relationship between risk and profit for US banks for Using international data for , Bourke (1989) finds that both capital and liquidity ratios were positively associated with profitability. In contrast, Molyneux and Thornton s (1992) European analysis found that profitability was negatively associated with liquidity. In Granger-causality tests, Berger (1995b) finds a positive association between CAR and return on equity (ROE), and proposes several theoretical explanations. For example, expected bankruptcy costs may be relatively high for a bank maintaining CAR below its equilibrium value. A subsequent increase in CAR should lead to an increase in ROE by lowering insurance expenses on uninsured debt. Alternatively, according to a signalling hypothesis, bank managers may have both private information as to the bank s future profitability and a stake in the bank s value through personal share ownership or options. It may be less costly for managers of low risk banks to signal quality by maintaining a high CAR than for managers of high risk banks. This may create a signalling equilibrium involving a positive association between CAR and ROE. European banking is one of the few industries in which private, public and mutual firms operate together in a competitive market. However, there is little empirical guidance as to whether there are systematic differences in the performance of banks with different ownership features (Altunbas et al., 2001; Carbo et al., 2002). Seminal work by Jensen and Meckling (1976), Fama (1980) and Fama and Jensen (1983) suggests that a lack of capital market discipline weakens owners control over management, leaving management free to pursue its own interests with few incentives to be efficient. Given that public and mutual banks pursue clearly stated social and economic development objectives (and are also subject to a lack of capital market discipline) one might expect them to have different performance characteristics to profit-maximizing private sector banks. Nicols (1967) and O Hara (1981) suggest that US mutual firms are likely to be more efficient than their private sector counterparts. Mester (1989, 1993) finds that mutual firms are more efficient, while Cebenoyan et al. (1993) suggest there is no difference between the efficiency of mutual and joint stock Savings and Loans banks. Conflicts of interest between owners and managers may sometimes make the relationships between profitability and other variables difficult to disentangle. For example, while owners seek to maximize profit, managers might be willing to sacrifice profit so as to reduce risk by undertaking more secure activities, or to maximize utility through expense preference behaviour (Berger and Hannan, 1998; Hughes et al., 2002). Overall, while a divergence in performance features across ownership types and governance structures might be expected, there is little or no consensus in the empirical literature

6 368 The Manchester School as to whether private banks are better performers than their mutual and cooperative sector counterparts. 3 AN EMPIRICAL MODEL OF BANK PROFITABILITY This section describes an empirical model for the determinants of profitability in European banking. The source of data is the Bankscope database, compiled by International Bank Credit Analysis Limited. The content of the model is as follows: p (1) it, = f( pi, t-1, sit,, oit,, d1, i, d2, i) where p i,t is the profit of bank i in year t, measured using ROE; s i,t is the natural logarithm of total assets in euros; o i,t is the nominal value of OBS business divided by (total assets + the nominal value of OBS business); c i,t is CAR; d 1,i = 1 for savings banks, 0 elsewhere; and d 2,i = 1 for cooperative banks, 0 elsewhere. s i,t is included as an explanatory variable to capture any relationship between bank size and profitability. A positive relationship between size and profit could be explained by several factors. Large banks may benefit from scale or scope economies. In addition, large banks may be able to exert market power through stronger brand image or implicit regulatory (too-big-to-fail) protection. Abnormal profits obtained through the exercise of market power in wholesale or capital markets may also contribute to a positive size profitability relationship. Alternatively if large banks encounter diseconomies of scale, the size profit relationship could be negative. Many European banks have widened their product offerings to become universal banks during the 1990s (Stiroh, 2000). It seems desirable to take account of the scale of this type of activity in any performance analysis. The nominal value of the OBS business component of the term in o i,t takes account of non-interest income and fee generating services from various contingent liabilities such as letters of credit, derivatives, securities underwriting, insurance and other types of non-traditional banking activities. This is the most readily available measure of the scale of contingent liability activity. Unfortunately, in contrast with some US banking data sets, Bankscope does not provide a consistent breakdown of the components of this measure. The present analysis therefore relies on the aggregate measure. The term in c i,t allows for a relationship between a bank s CAR and its profitability. In this context, CAR serves as a crude proxy for risk. For the reasons outlined in Section 2, either a positive or a negative relationship between CAR and ROE is possible. Finally, the ownership dummy variables d 1,i and d 2,i allow for the possibility that different ownership types impact on performance through variations in institutional objectives or management structures. The pooled cross-sectional time-series structure of the data set enables

7 The Profitability of European Banks 369 several variants of the relationship summarized in (1) to be estimated, as follows. Pooled cross-sectional time-series model, estimated using ordinary least squares (OLS) pit, = a1 + a2pit, -1 + a3sit, + a4oit, + a5cit, + a6d1, i + a7d2, i + uit, i = 1,..., N, t = 2,..., T (2) Cross-sectional model, estimated using OLS pi* = b1+ b3si* + b4oi* + b5ci* + b6d1, i + b7d2, i + wi i = 1,..., N (3) Dynamic panel model, estimated using the generalized method of moments (GMM) pit, = g1 + g 2pit, -1 + g 3sit, + g 4oit, + g 5cit, + hi + vit, i = 1,..., N, t = 2,..., T where p i* = S T t=1p i,t /T, and similar definitions apply for other variables. Equation (2) is the pooled model, based on an assumption that crosssectional variation (between banks) in any of the independent variables has the same implications for profit as variation over time in that variable for an individual bank. It is also likely to be informative to relax this restriction, however, by estimating the separate cross-sectional and dynamic panel specifications, (3) and (4) respectively. 3 Equation (3) is the cross-sectional model, in which the profitability equation is estimated as a relationship between the individual bank mean values of all variables. Any variation for an individual bank over time (or within bank variation) is eliminated in the computation of the individual bank means. Because there are no dynamics in (3), the lagged profit variable is omitted. Equation (4) is a dynamic panel model, in which a complete set of individual bank effects, h i, controls for all crosssectional (or between banks ) variation. The ownership dummies d 1,i and d 2,i have no time dimension and are therefore omitted from (4). Equation (4) is equivalent to a respecified version of (2), with all variables expressed as deviations from individual bank means. (4) 3 Within bank and between banks variation in the dependent variable and in each independent variable can be identified using a one-way analysis of variance decomposition. For the dependent variable p i,t,for example, this implies the following relationship between the total, within and between variation, respectively: N T ÂÂ Â ÂÂ 2 2 ( pit, - p** ) = T ( pi* -p**) + ( pit, -pi*) i= 1 t= 1 i= 1 i= 1 t= 1 where p ** = S N i=1s N t=1p i,t /NT. The components of these sums of squares represent the sums of squared values of the dependent variable in the three variants of the profits model, (2), (3) and (4) respectively. T N T 2

8 370 The Manchester School Schmalensee (1989) discusses the advantages and disadvantages of crosssectional, time-series and panel estimation of firm performance relationships in industrial organization. Cross-sectional estimation rarely, if ever, yields consistent estimates of structural parameters, partly because cross-sectional data often include deviations from long-run equilibria that tend to be correlated with the independent variables. The use of panel data enables researchers to model adjustments from disequilibrium, while the use of industry-specific data minimizes potential bias due to the presence of industry-specific unobservables. Nevertheless, cross-sectional estimation can be useful in providing stylized facts to guide the construction of dynamic models. Reviewing the new industrial organization literature that was partly motivated by dissatisfaction with SCP, Bresnahan (1989) advocates the use of time-series and single-industry data rather than cross-sectional data, for similar reasons. 4 DATA AND RESULTS Accounts data on 665 banks from six European countries (Denmark, France, Germany, Italy, Spain and the UK) were downloaded from Bankscope. 4 The sample selection criteria were as follows. Any bank which operated in a country which was a member of the European Union in 1992 and was classified as a commercial, savings or cooperative bank, and for which complete, consistent and believable data for each of the years inclusive were accessible, was selected, with the following exceptions. Banks reporting extreme values or very large unexplained changes in the values of any of the variables were excluded. Banks from Belgium, Greece, Ireland, Luxembourg, the Netherlands and Portugal were excluded because little or no data were available. Banks with operations in one country and which are centrally located in another country posed particular problems in the data selection process. European banks often have subsidiaries located in other countries (e.g. Barclays Bank España). Inclusion of these subsidiaries would introduce an element of double-counting. Subsidiaries of non-european banks located in Europe were also excluded. 5 Tables 1 and 2 report summary data (means and standard deviations) on the dependent and independent variables used in the empirical model: profitability (ROE); bank size (total assets); OBS business as a proportion of assets plus OBS business; and CAR. Summary data are tabulated for all banks, and separately for commercial, savings and cooperative banks in each 4 A smaller version of the present data set is used in the analysis of the interactions between bank growth and profitability reported by Goddard et al. (2004). 5 The operation of subsidiaries may say more about the objectives and performance of the parent institution than about the competitive environment of the country in which the subsidiary is located.

9 The Profitability of European Banks 371 TABLE 1 DESCRIPTIVE STATISTICS: 1993 DATA OBS/(OBS ROE Assets + assets) CAR 1993 N Mean SD Mean SD Mean SD Mean SD All Den: comm France: comm Gmy: comm Gmy: savings Gmy: coop Ity: comm Ity: savings Ity: coop Spain: comm Spain: savings UK: comm UK: savings Note: Assets are in million euros, current prices. Conversions are at exchange rates. TABLE 2 DESCRIPTIVE STATISTICS: 1998 DATA OBS/(OBS ROE Assets + assets) CAR 1998 N Mean SD Mean SD Mean SD Mean SD All Den: comm France: comm Gmy: comm Gmy: savings Gmy: coop Ity: comm Ity: savings Ity: coop Spain: comm Spain: savings UK: comm UK: savings Note: Assets are in million euros, current prices. Conversions are at exchange rates. of the six countries, for 1993 (Table 1) and 1998 (Table 2). In 1993 the Spanish savings banks sector was the most profitable in the sample. UK commercial and savings banks recorded average profit rates significantly higher than average, while the French and Spanish commercial banking sectors both

10 372 The Manchester School recorded negative average profits. 6 In 1998 Spanish commercial and savings banks both recorded relatively high average profit rates, while the UK commercial sector was the most profitable in the sample. Table 3 reports OLS estimations of the pooled cross-sectional time-series model, (2). Table 4 reports OLS estimations of the cross-sectional model, (3). Table 5 reports estimations of the dynamic panel model, (4). OLS is unsuitable for (4) because the inclusion of the individual bank effects in the error term creates non-zero covariance between the latter and the lagged dependent variable. Equation (4) is therefore estimated using the two-step GMM estimator described by Hansen (1982) and employed by Arellano and Bond (1991) for dynamic panel estimation with large N, small T data sets. 7 All three models are estimated using the data for all countries combined, and using the data for each of the six countries individually. Prior to estimation of (2) and (4), the data on p i,t, s i,t, o i,t and c i,t are expressed as deviations from their sample mean values in each year. This transformation, equivalent to estimating using untransformed data and year dummies, eliminates the effects of any cyclical influences on profitability common to all banks in each sample. 8 In Table 3 most of the pooled estimations test positively for heteroscedasticity. The standard errors reported for all coefficients are therefore based on White s (1980) adjustment. Three of the seven pooled estimations also test positively for first-order serial correlation, a finding that may be due to the omission of the individual effects h i from (2). In Table 4, all crosssectional estimations with the exception of Germany test negatively for 6 During the early 1990s many large French banks were forced to make substantial bad debt provisions, causing both profitability and growth to suffer, while Credit Lyonnais, one of the largest French banks, was the subject of a government bail-out (Morgan Stanley, 1995, 1996). 7 By suitable transformation of the data, the two-step GMM estimator eliminates from the estimation the individual bank effects h i. Monte Carlo studies have shown that the asymptotic standard errors obtained from the two-step estimator are downward biased in small samples, but the first-step standard errors are virtually unbiased (Bond et al., 2001). In accordance with common practice, for (4) the efficient two-step coefficient estimates together with the one-step standard errors are reported below. The GMM estimator instruments the lagged dependent variable on its own second and higher-order lags and (optionally) the lagged independent variables. The Sargan test for the validity of over-identifying restrictions described by Arellano and Bond (1991) is used to select an appropriate set of instruments. Guided by the results of this test on a variety of specifications, only the second and higher-order lags of the dependent variable are used as instruments in the estimations reported in Table 5. 8 In addition to the estimations reported in Tables 3 5, the same specifications were estimated using an alternative return on assets (ROA) profitability measure as the dependent variable. Because there is greater year-on-year variation in the numerator than in the denominators of ROE and ROA, the results using ROA were qualitatively very similar to those reported in Tables 3 and 5. As a profitability measure ROE is preferred to ROA because OBS business, which for many European banks makes a significant contribution to total profit, is excluded from the denominator of ROA.

11 The Profitability of European Banks 373 TABLE 3 ESTIMATION RESULTS: POOLED CROSS-SECTIONAL TIME-SERIES MODEL All Denmark France Germany Italy Spain UK Banks N Observations N(T - 1) i,t-1 p *** *** *** *** *** *** *** (0.0871) (0.0645) (0.0979) (0.0887) (0.1181) (0.0472) (0.1378) s i,t (0.0020) (0.0073) (0.0093) (0.0009) (0.0031) (0.0078) (0.0039) o i,t ** ** (0.0350) (0.0486) (0.1337) (0.0287) (0.0451) (0.4520) (0.1573) c i,t ** ** ** (0.2079) (0.2910) (0.3132) (0.0815) (0.0756) (0.6239) (0.3588) d 1,i *** * (0.0087) (0.0053) (0.0110) (0.0201) (0.0272) d2,i *** (0.0098) (0.0050) (0.0108) l *** 3.78 * *** 33.1 *** *** l *** *** *** Notes: The estimated model is equation (2): p i,t = a 1 + a 2 p i,t-1 + a 3 s i,t + a 4 o i,t + a 5 c i,t + a 6 d 1,i + a 7 d 2,i + u i,t. The estimation method is OLS. Estimations are over i = 1,...,N and t = 2,...,T, N as shown, T = 7 ( inclusive). White-adjusted standard errors are in parentheses. l1 = N(T - 1)R 2 is the Lagrange multiplier test of the null of homoscedasticity, based on the auxiliary regression of squared residuals on squared fitted values, distributed c 2 (1) under the null. l 2 = N(T - 2)R 2 is the Lagrange multiplier test of the null of zero first-order serial correlation, based on the auxiliary regression of residuals on explanatory variables and lagged residuals, distributed c 2 (1) under the null.

12 374 The Manchester School TABLE 4 ESTIMATION RESULTS: CROSS-SECTIONAL MODEL All Denmark France Germany Italy Spain UK Banks N s i,* * * ** (0.0024) (0.0092) (0.0214) (0.0016) (0.0049) (0.0087) (0.0045) o i,* *** * ** (0.0443) (0.0848) (0.3323) (0.0604) (0.0912) (0.5611) (0.0985) c i,* *** (0.0888) (0.3287) (0.7398) (0.1699) (0.2360) (0.4096) (0.1106) d 1,i *** (0.0101) (0.0099) (0.0157) (0.0330) (0.0215) d 2,i * ** (0.0095) (0.0095) (0.0150) Constant *** (0.0373) (0.1242) (0.3373) (0.0303) (0.0789) (0.1441) (0.0712) l *** Notes: The estimated model is equation (3): p i* = b 1 + b 3 s i* + b 4 o i* + b 5 c i* + b 6 d 1,i + b 7 d 2,i + w i. The estimation method is OLS. Estimations are over i = 1,..., N as shown. Standard errors are in parentheses (Whiteadjusted for Germany only). l 1 = NR 2 is the Lagrange multiplier test of the null of homoscedasticity, based on the auxiliary regression of squared residuals on squared fitted values, distributed c 2 (1) under the null. heteroscedasticity. White-adjusted standard errors are therefore reported for Germany only, and unadjusted standard errors for all other countries. In Table 5, the Sargan test for the validity of the over-identifying restrictions in the instrumentation used in the GMM estimation is rejected for Spain at the 5 per cent level and for Italy and the UK at the 10 per cent level, but is accepted elsewhere. With the exception of the UK, the remaining dynamic panel estimations test negatively for second-order serial correlation and therefore satisfy this condition for the consistency of the GMM estimator. 9 In the pooled model the estimated persistence of profit coefficients a 2 are highly significant in all cases. These estimates are larger in the pooled model than their counterparts g 2 in the dynamic panel model, probably because the latter incorporates individual effects, the presence of which is (incorrectly) attributed to persistence in the former. In both sets of estimations there are quite large differences between countries in the magnitudes of the persistence coefficients. The all countries dynamic panel persistence 9 For the pooled and cross-sectional models, a standard Lagrange multiplier test of heteroscedasticity is employed, based on an auxiliary regression of squared residuals on squared fitted values. For the pooled model, a Lagrange multiplier test for first-order serial correlation is employed, based on an auxiliary regression of residuals on explanatory variables and lagged residuals. To test for serial correlation in the dynamic panel model, the procedures described by Arellano and Bond (1991) are employed. In this case first-order serial correlation in the residuals would not render the GMM estimates inconsistent, but second-order serial correlation would do so. Accordingly the test for the latter is reported in Table 5.

13 The Profitability of European Banks 375 TABLE 5 ESTIMATION RESULTS: DYNAMIC PANEL MODEL All Denmark France Germany Italy Spain UK Banks N Observations N(T - 2) i,t-1 p *** *** *** *** *** * *** (0.0204) (0.0457) (0.0717) (0.0402) (0.0642) (0.0470) (0.0427) s i,t ** *** *** (0.0254) (0.0682) (0.1383) (0.0154) (0.0404) (0.1560) (0.0295) o i,t *** *** (0.0675) (0.0829) (0.4649) (0.0478) (0.0839) (1.1887) (0.1196) c i,t *** ** *** *** (0.1865) (0.7053) (0.9556) (0.1357) (0.2650) (1.1467) (0.2390) l * ** * l ** Notes: The estimated model is equation (4): p i,t = g 1 + g 2 p i,t-1 + g 3 s i,t + g 4 o i,t + g 5 c i,t + (h i + v i,t ). The estimation method is Arellano and Bond s (1991) two-step GMM dynamic panel estimator. Estimations are over i = 1,...,N and t = 3,...,T, N as shown, T = 7 ( inclusive). One-step Arellano and Bond standard errors are shown in parentheses (see footnote 3). l3 is the Sargan test for validity of over-identifying restrictions, distributed c 2 (14) under the null. l 4 is the test of the null of zero second-order serial correlation (Arellano and Bond, 1991, p. 282, equation (8)), distributed N(0, 1) under the null.

14 376 The Manchester School estimate of 0.26 is smaller than those reported in several earlier European manufacturing studies (Geroski and Jacquemin, 1988; Goddard and Wilson, 1999). In both the pooled and the dynamic panel models, the degree of persistence is highest for France, where historically a high level of government regulation has insulated banks from the full rigours of competition. There is no evidence of any relationship between bank size and profitability in the pooled estimations. The cross-sectional and dynamic panel estimations, however, suggest there are some significant size profit relationships within the data set. The cross-sectional all countries estimate of b 3 is positive and significant at the 10 per cent level. The corresponding estimate for Germany is negative and significant at the 10 per cent level, while the UK estimate is positive and significant at the 5 per cent level. In the relatively unconcentrated German banking sector, after controlling for ownership type the smaller banks appear to have outperformed their larger counterparts. In the more highly concentrated UK banking sector, larger banks may have benefited from scale or scope economies, or from other size-related advantages. In the other four countries, the size profit relationship appears to have been neutral. The all countries dynamic panel estimation produces an insignificant estimate of g 3. The individual country estimates of g 3 for Denmark and the UK are positive and significant at the 5 per cent and 1 per cent levels, respectively. In contrast the estimate of g 3 for Spain is negative and significant at the 1 per cent level. These rather varied crosssectional and dynamic panel results are not contradictory, since they reflect different aspects of the size profitability relationship: the cross-sectional estimation compares the average profitability of banks in different size bands, while the dynamic panel estimation reflects the direction of change in profitability for a given change in size for any individual bank from year to year. There is no reason why the cross-sectional and dynamic effects should necessarily operate in the same direction. Nevertheless, overall the evidence for any consistent or systematic relationship between size and profitability is unconvincing. The pooled estimations also suggest differences between countries in the relationship between the importance of OBS business in a bank s portfolio and its profitability. A negative and significant estimate of a 4 is obtained for Germany, and a positive and significant one for the UK. The corresponding estimates for the other countries are insignificant, and in the all countries version, pooling the German and UK data with the rest produces an estimated coefficient close to zero. Similar patterns are found in the crosssectional estimations, although the positive UK estimate of b 4 falls short of being significant. The all countries estimate of b 4 is negative and significant at the 1 per cent level. In the dynamic panel estimations, a positive and significant estimate of g 4 is obtained for the UK; for other countries these coefficients are insignificant. There is some evidence of a positive relationship between the CAR and

15 The Profitability of European Banks 377 profitability in the pooled estimations. The estimates of a 5 for all countries and for France and Spain are positive and significant at the 5 per cent level. In the cross-sectional estimations, the all countries estimate of b 5 is positive and significant. Positive estimates of this coefficient are also obtained for five out of six individual countries, but none of these is significant. In the dynamic panel estimations, all of the estimated values of g 5 are positive, and the coefficients for all countries, Denmark and the UK are significant at either the 5 per cent or 1 per cent levels. Assuming that CAR provides an accurate reflection of a bank s risk profile, a predominantly positive empirical relationship between CAR and profitability is surprising, although similar to the results for the USA reported by Berger (1995b). Berger s alternative theoretical explanations for a positive relationship between CAR and ROE (see Section 2) may be applicable in the present case. 10 There is no systematic relationship between ownership type and profitability according to the pooled estimation across all countries. In the all countries cross-sectional estimation, cooperative banks are less profitable than commercial and savings banks, though the effect is only significant at the 10 per cent level. According to both the pooled and cross-sectional estimations, however, savings and cooperative banks in Germany were significantly less profitable than commercial banks, after controlling for the other variables in the model. There is little evidence of systematic variation in profitability by ownership type in countries other than Germany. 5 CONCLUDING REMARKS This paper has reported cross-sectional and dynamic panel estimations of a model designed to identify selected determinants of profitability in six major European banking sectors: Denmark, France, Germany, Italy, Spain and the UK, for the period The results of the empirical analysis suggest that, despite the growth in competition in European financial markets, there is still significant persistence of profit from one year to the next. In other words, if a bank earns an abnormal profit in the current year, its expected profit for the following year should include a sizeable proportion of the current year s abnormal profit. Although competition is effective in eliminating abnormal profit eventually, the adjustments are by no means instantaneous. There is some variation between countries in effectiveness of competition in eliminating abnormal profits. The degree of persistence 10 Alternatively, CAR may simply serve inadequately as a proxy for risk. The growth of OBS business creates one type of difficulty with the interpretation of CAR. Consider the following decomposition: ROE = ROA EM, where ROE = return on equity = profit/equity, ROA = return on assets = profit/assets, and EM = equity multiplier = assets/equity = 1/CAR. Moving assets OBS tends to increase ROA and reduce EM, creating the (false) impression that the bank is earning a higher return on its assets at a lower level of risk, and rendering CAR unreliable as an indicator of the bank s true risk profile.

16 378 The Manchester School appears to be highest for France, where historically competition within the banking sector has been heavily regulated. There is some evidence of a significant size profitability relationship in some of the estimations, but overall the evidence for any systematic relationship between size and performance is unconvincing. The results are consistent with those of much of the previous empirical banking literature, suggesting that efficiency is likely to be a more important determinant of performance than size per se. The results also suggest differences between countries in the relationship between the importance of OBS business in a bank s portfolio and its profitability. The relationship appears to be positive for the UK, but either neutral or negative for other countries, where some banks that have diversified rapidly into OBS business have experienced difficulties in maintaining their profitability. Meanwhile there is evidence of a positive relationship between CAR and profitability. This finding does not reflect the expected theoretical relationship between risk and return: a high CAR should signify a bank that is operating over-cautiously and ignoring potentially profitable trading opportunities. Theoretical explanations that have been put forward to explain a similar positive empirical association between CAR and ROE in the USA consider the relationship between CAR and the costs of insurance against bankruptcy, which also impact on profitability, and a signalling hypothesis according to which managers use CAR to send signals about expected future profitability, creating a positive association between CAR and ROE. There is little evidence of any systematic relationship between ownership type and profitability. After controlling for the other variables in the model, however, German savings and cooperative banks appear to have been significantly less profitable than German commercial banks during the 1990s. A more general conclusion of the empirical analysis is that, the increasing integration of European banking markets notwithstanding, national factors still seem to play an important role among the determinants of bank performance. REFERENCES Altunbas, Y., Evans, L. and Molyneux, P. (2001). Ownership and Efficiency in Banking, Journal of Money, Credit and Banking, Vol. 33, No. 4, pp Amel, D. and Froeb, L. (1991). Do Firms Differ Much?, Journal of Industrial Economics, Vol. 39, No. 3, pp Arellano, M. and Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations, Review of Economic Studies, Vol. 58, No. 2, pp Baumol, W. J., Panzar, J. C. and Willig, R. D. (1982). Contestable Markets and the Theory of Industry Structure, New York, Harcourt Brace Jovanovich. Berger, A. N. (1995a). The Profit Structure Relationship in Banking: Tests of Market

17 The Profitability of European Banks 379 Power and Efficient Structure Hypotheses, Journal of Money, Credit and Banking, Vol. 27, No. 2, pp Berger, A. N. (1995b). The Relationship Between Capital and Earnings in Banking, Journal of Money, Credit and Banking, Vol. 27, No. 2, pp Berger, A. N. and Hannan, T. H. (1989). The Price Concentration Relationship in Banking, Review of Economics and Statistics, Vol. 71, No. 2, pp Berger, A. N. and Hannan, T. H. (1998). The Efficiency Cost of Market Power in the Banking Industry: a Test of the Quiet Life and Related Hypotheses, Review of Economics and Statistics, Vol. 80, No. 3, pp Berger, A. N. and Humphrey, D. B. (1997). Efficiency of Financial Institutions: International Survey and Directions for Future Research, European Journal of Operational Research, Vol. 98, No. 2, pp Berger, A. N., Bonime, S. D., Covitz, D. M. and Hancock, D. (2000). Why are Bank Profits so Persistent? The Roles of Product Market Competition, Information Opacity and Regional Macroeconomic Shocks, Journal of Banking and Finance, Vol. 24, No. 7, pp Bond, S., Bowsher, C. and Windmeijer, F. (2001). Criterion-based Inference for GMM in Autoregressive Panel Data Models, Economics Letters, Vol. 73, No. 3, pp Bourke, P. (1989). Concentration and Other Determinants of Bank Profitability in Europe, North America, and Australia, Journal of Banking and Finance, Vol. 13, No. 1, pp Bresnahan, T. F. (1989). Empirical Studies of Industries with Market Power, in R. Schmalensee and R. D. Willig (eds), Handbook of Industrial Organisation, Amsterdam, Elsevier Science, Vol. 2, pp Brozen, Y. (1971). Bain s Concentration and Rates of Return Revisited, Journal of Law and Economics, Vol. 13, No. 2, pp Carbo, S., Gardener, E. P. M. and Williams, J. (2002). Efficiency in Banking: Empirical Evidence from the Savings Banks Sector, The Manchester School, Vol. 70, No. 2, pp Cebenoyan, A., Sinan, E. S., Cooperman, C. A. and Hudgkins, S. (1993). The Relative Efficiency of Stock versus Mutual S & Ls: a Stochastic Frontier Approach, Journal of Financial Services Research, Vol. 7, No. 2, pp De Bandt, O. and Davis, E. P. (2000). Competition, Contestability and Market Structure in European Banking Sectors on the Eve of EMU: Evidence from France, Germany, and Italy with a Perspective on the United States, Journal of Banking and Finance, Vol. 24, No. 6, pp Demsetz, R. S. and Strahan, P. E. (1997). Diversification, Size, and Risk at Bank Holding Companies, Journal of Money, Credit and Banking, Vol. 29, No. 3, pp Diamond, D. W. (1984). Financial Intermediation and Delegated Monitoring, Review of Economic Studies, Vol. 51, No. 3, pp Evanoff, D. D. and Fortier, D. L. (1988). Re-evaluation of the Structure Conduct Performance Paradigm in Banking, Journal of Financial Services Research, Vol. 1, No. 3, pp Fama, E. F. (1980). Agency Problems and the Theory of the Firm, Journal of Political Economy, Vol. 88, No. 2, pp Fama, E. F. and Jensen, M. C. (1983). Separation of Ownership and Control, Journal of Law and Economics, Vol. 26, No. 2, pp Geroski, P. A. and Jacquemin, A. (1988). The Persistence of Profits: a European Comparison, Economic Journal, Vol. 98, No. 391, pp

18 380 The Manchester School Goddard, J. A. and Wilson, J. O. S. (1999). Persistence of Profit: a New Empirical Interpretation, International Journal of Industrial Organization, Vol. 17, No. 5, pp Goddard, J. A., Molyneux, P. M. and Wilson, J. O. S. (2001). European Banking: Efficiency, Technology and Growth, Chichester, Wiley. Goddard, J. A., Molyneux, P. M. and Wilson, J. O. S. (2004). Dynamics of Growth and Profitability in Banking, Journal of Money, Credit and Banking, forthcoming. Hansen, L. P. (1982). Large Sample Properties of Generalized Method of Moments Estimators, Econometrica, Vol. 50, No. 4, pp Hughes, J. P., Lang, W. W., Mester, L. J. and Moon, C. G. (1999). The Dollars and Sense of Bank Consolidation, Journal of Banking and Finance, Vol. 23, No. 2 4, pp Hughes, J. P., Lang, W. W., Mester, L. J., Moon, C. G. and Pagano, M. S. (2002). Do Bankers Sacrifice Value to Build Empires? Managerial Incentives, Industry Consolidation and Financial Performance, Wharton Financial Institutions Center Working Paper Jackson, W. E. (1992). The Price Concentration Relationship in Banking: a Comment, Review of Economics and Statistics, Vol. 74, No. 2, pp Jensen, M. C. and Meckling, W. H. (1976). Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure, Journal of Financial Economics, Vol. 3, No. 4, pp Klein, P. G. and Saidenberg, M. R. (1997). Diversification, Organization and Efficiency: Evidence from Bank Holding Companies, Federal Reserve Bank of New York Working Paper. Levonian, M. E. (1993). The Persistence of Bank Profits: What the Stock Market Implies, Working Papers in Applied Economic Theory No , San Francisco, CA, Federal Reserve Bank of San Francisco. McKillop, D. G. and Ferguson, C. (1993). Building Societies: Structure, Performance and Change, London, Graham and Trotman. Mester, L. J. (1989). Testing for Expense Preference Behaviour: Mutual versus Stock Savings and Loans, Rand Journal of Economics, Vol. 20, No. 4, pp Mester, L. J. (1993). Efficiency in the Savings and Loan Industry, Journal of Banking and Finance, Vol. 17, No. 2 3, pp Molyneux, P. and Thornton, J. (1992). Determinants of European Bank Profitability: a Note, Journal of Banking and Finance, Vol. 16, No. 6, pp Morgan Stanley (1995). French Banking Long-term Under-performance May Be Over, European and UK Banking Bulletin, 6 November. Morgan Stanley (1996). French Banks: Light at the End of the Tunnel, European Financial Briefing, 21 June. Mueller, D. C. (1986). Profits in the Long Run, Cambridge, Cambridge University Press. Mueller, D. C. (1990). The Persistence of Profits in the United States, in D. C. Mueller (ed.), The Dynamics of Company Profits: an International Comparison, Cambridge, Cambridge University Press. Nicols, A. (1967). Property Rights and Behavior: Stock versus Mutual Savings and Loan Associations: Some Evidence of Differences in Behavior, American Economic Review, Vol. 57, No. 2, pp O Hara, M. (1981). Property Rights and the Financial Firm, Journal of Law and Economics, Vol. 24, No. 2, pp Rhoades, S. A. (1985). Market Share as a Source of Market Power: Implications and Some Evidence, Journal of Economics and Business, Vol. 37, No. 4, pp