Does cross-ownership affect competition? Evidence from the Italian banking industry

Transcription

1 Int. Fin. Markets, Inst. and Money 17 (2007) Does cross-ownership affect competition? Evidence from the Italian banking industry Francesco Trivieri Dipartimento di Economia e Statistica, Università della Calabria, Ponte P. Bucci, Cubo 1-C, Arcavacata di Rende (CS), Italy Received 2 June 2005; accepted 1 September 2005 Available online 3 October 2005 Abstract The purpose of this paper is to empirically investigate the effects of cross-ownership among Italian banks on competition in the national banking sector. This aim is pursued by measuring and comparing the degree of competition between banks involved in the web of cross-ownership and banks that are not involved. The bank s degree of competition is measured by applying the methodology developed by [Panzar, J.C., Rosse, J.N., Testing for monopoly equilibrium. Journal of Industrial Economics 35, ] The econometric results provide empirical evidence that, in the period , Italian banks involved in cross-ownership were less competitive than the other national credit firms, thus supporting the view that cross-ownership may represent an obstacle to industrial competition Elsevier B.V. All rights reserved. JEL classification: G21; G34; L20 Keywords: Banking; Cross-ownership; Competition 1. Introduction In the course of the 1990s, the Italian banking system underwent profound changes at normative and institutional levels, which led among other things to a significant relaxation of the entry barriers, to the liberalisation of bank branching, to the redefinition of ownership structure and to a large number of mergers and acquisitions. The effects of these transformations and, in particular, of those linked to the process of consolidation have been studied by many authors (see, among others: Resti, 1997; Angelini and Tel.: ; fax: address: francesco.trivieri@unical.it /$ see front matter 2005 Elsevier B.V. All rights reserved. doi: /j.intfin

2 80 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) Cetorelli, 2000; Messori, 2001; Sapienza, 2002; Focarelli et al., 2002; Focarelli and Panetta, 2003). In spite of an increasing interest in the literature, works dedicated to studying the causes and effects of the reorganisation and consolidation processes in the Italian banking system have largely neglected to analyse an issue which is closely tied to such operations: the formation of a complex web of cross-ownership linking the large banking groups in the country, meaning by cross-ownership the shares held, directly or indirectly, by the same subjects in more groups. Generally speaking, the suspicion that cross-ownership among firms may have anticompetitive effects in an industrial sector finds support in the argumentations and in the theoretical results of several contributions in the industrial economics literature. Bruzzone (1999) observes that cross-ownership, in the same way as interlocking directorates, can favour exchanges of practical information which can become collusive strategies among firms. Also Gilo (2000) and Gilo and Spiegel (2003) show that cross-ownership may reduce competition, since they facilitate (tacit) collusion among firms. Maxwell et al. (1999) and O Brien and Salop (2000) point out that, when major investors own shares in multiple firms in the same industry, cross-ownership changes competitive interest and moves the market equilibrium closer to the monopoly solution. Regardless of the implications on competition, the issue of cross-ownership among firms is of considerable interest for other reasons. First, cross-ownership can obstruct the contestability and, therefore, the efficient allocation of firms property rights (for a theory on the efficient allocation of firms property rights see Hart and Moore, 1990). Secondly, with reference to the Italian banking system, cross-ownership assumes major relevance if one considers that, in many cases, it can be traced down to the main banking foundations, strange beasts that for several reasons are unable to guarantee an effective corporate governance of banks (see Messori, 1998, 1999, 2000). This paper deals with the issue of the effects of cross-ownership on competition. In particular, the aim of the work is to carry out an empirical investigation to assess whether, during the 1990s, cross-ownership among Italian banks had influence on competition in the national banking sector. At the present time, this type of analysis is completely missing in the literature concerning the banking industry in Italy. The purpose of this study is pursued by measuring and comparing the degree of competition between banks involved in cross-ownership and credit firms that are not involved. The analysis refers to the years , that is to the period in which cross-ownership became an important phenomenon in the Italian banking sector. The bank s degree of competition is measured by applying the methodology developed by Rosse and Panzar (1977) and Panzar and Rosse (1982, 1987) in the context of the so-called New Economic Industrial Organizations (NEIO). The paper is structured as follows. The next section presents a brief analysis dealing with the issue of cross-ownership in the Italian banking sector. The third section illustrates the strategy of Panzar and Rosse to measure the level of competition in an industry. The fourth section describes the data, the sample and the econometric methodology used in the empirical analysis. The fifth section presents and discusses the main results obtained and the sixth provides some final considerations. 2. Cross-ownership in the Italian banking industry The formation of a thick web of cross-ownership, which currently connects the main Italian banks, is closely linked to the ongoing process of restructuring of the banking system in Italy. Over the last 10 years, significant events that have changed the face of the Italian banking industry were primed by the complex normative reform process that started at the beginning of

3 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) the 1990s. The issuing of European directives in the banking field, the enacting of the Amato-Carli law (legge 218/90) and the introduction in 1993 of the TUB (Testo Unico in Materia Bancaria) are only some of the most important reforms that profoundly modified the national banking regulations in force at the time. Within the context of the new legislative framework, two important processes played a crucial role in transforming the ownership structure of the Italian banks: the selling-off of state held banking shares and the process of consolidation in the national credit sector. The former began in 1993 and was largely completed in the four year period (at the end of this period the state share in the Italian banking sector was roughly 0.10%); the latter involved mainly small and medium sized banks in the early 1990s, and the biggest Italian banks since In most of the cases involving the largest banks, the process of off-loading state held shares in the industry was carried out mainly through private negotiations aimed at identifying the group of controlling shareholders. This procedure, jointly with mergers and acquisitions involving the country s main banks, led to a situation where a restricted number of shareholders ended up with a relevant portion of the shares in nearly all the largest national banking groups, thus giving rise to the maze of cross-ownership one can see today in the Italian banking sector (for a more detailed description of the privatisation and consolidation processes, and the role these played in the formation of cross-ownership in the Italian banking industry, see Inzerillo and Messori, 2000; Messori, 2001). The major cross-ownership which, on 31 December 2000, linked the largest national banking groups are shown in the following table. In line with Maxwell et al. (1999) and O Brien and Salop (2000), Table 1 was obtained by considering as cross-ownership the shares, directly or indirectly, held by same subjects in more than one banking group. It is important to note that, for various reasons, Table 1 provides a simplified representation of cross-ownership in the Italian banking system at the end of the last decade. In the first place, Table 1 reports only the shares held by the major shareholders, that is by the so-called relevant shareholders (according to the Italian legislation, relevant shareholders are those holding directly or indirectly more than 2% of the capital of a firm and, therefore, are obliged to communicate to the CONSOB, the commission regulating the Italian Stock Exchange 1 ). In other words, Table 1 does not consider any cross-ownership involving owners holding under 2% of the total shares (the only exception to this is the participation of Assicurazioni Generali in Bipop-Carire). Secondly, in some cases, the shareholders indicated in Table 1 are not the ultimate owners, i.e. subjects at the top of the ownership chain (Faccio and Lang, 2002), but rather direct shareholders, that is to say the subjects that figure in the first instance in the ownership structure of banking groups. This implies that Table 1 does not take into account cross-ownership that can be traced to subjects at the top of the ownership structure of direct shareholders, the most important of these being those referable to shareholders in Mediobanca (the most important Italian financial credit bank). 2 The search for ultimate owners would have introduced numerous complications into the analysis without bringing significant advantages for the empirical investigation carried out in this paper. 1 Apart from Banca Cardine and Carinord, the banking groups indicated in Table 1 were all quoted on the Italian Stock Exchange on For example, Radici Pesenti figures as a direct shareholder both in the ownership structure of Banca Lombarda e Piemontese and in that of Mediobanca, the latter being, in turn, a direct shareholder in several banking groups. This cross-ownership is not reported in Table 1.

4 Table 1 Cross-ownership in the Italian banking system ( ) Banking group Owner Fondazione Cariplo Banca Intesa Compagnia di San Paolo San Paolo IMI-BN Fondazione MPS Ente CariFirenze Mediobanca Montedison b Generali c Gruppo (via Spafid) a Agnelli BANCA INTESA UNICREDIT SAN PAOLO IMI-BANCO DINAPOLI BANCA DI ROMA-MCC BANCA MPS BANCA NAZIONALE DEL LAVORO BANCA CARDINE # BIPOP-CARIRE BANCA LOMBARDA E P CdR FIRENZE CARINORD # CREDEM Source: CONSOB and Il Sole 24 Ore. Notes: all the figures are in percentage terms. Symbol (#) denotes that the banking group was not quoted on the Italian Stock Exchange on Symbol ( ) denotes a majority ownership share. a Mediobanca owned the 100% of shares. b Mediobanca owned the majority of shares (14.53% directly and 4.35% via Spafid). c Mediobanca owned the majority of shares (10.09% directly and 5.15% via Spafid). Reale Mutua ABN Amro 82 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007)

5 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) Finally, apart from a few cases (Credem, CdR di Firenze, Bipop-Carire), the above table involves only the large scale Italian banks. 3 By observing Table 1 it is evident how, at the end of the last decade, lying at the heart of the principal cross-ownership in the Italian banking industry is a small number of the most important banking foundations, Mediobanca and a few others. The banking foundations, born out of the reform at the beginning of the 1990s, became owners of the public banks when the latter were transformed from credit bodies into public corporations. Despite the different pieces of legislation over the last decade -aimed, amongst other things, at inducing the banking foundations to get rid of their shares in banks -they still possess significant holdings, if not the majority of the shares, in numerous banking firms. The suspicion that cross-ownership may have negative effects on competition among Italian banks finds support in the results of some recent theoretical contributions in the industrial economics literature. First, as Bruzzone (1999) observes, cross-ownership - in the same way as interlocking directorates (that is the sharing, among competitors, of some subjects in key administrative or management positions even in absence of ownership sharing) 4 can favour exchanges of practical information which can become collusive strategies among firms. Also Gilo (2000) and Gilo and Spiegel (2003) show that cross-ownership may reduce competition, since they facilitate (tacit) collusion among firms. Maxwell et al. (1999) and O Brien and Salop (2000) point out that, when major investors own shares in multiple firms in the same industry, cross-ownership changes competitive interest and moves the market equilibrium closer to the monopoly solution. To investigate if cross-ownership among Italian banks has had influence on competition in the national banking sector is the purpose of this paper. The aim is pursued on the empirical level by measuring and comparing the degree of competition between banks involved in cross-ownership and those that are not involved. The analysis refers to the years , that is to the period in which cross-ownership became an important phenomenon in the Italian banking sector. The banks degree of competition is measured by applying the methodology developed by Rosse and Panzar (1977) and Panzar and Rosse (1982, 1987), which is explained in the next section. 3. Empirical strategy: the Panzar and Rosse statistic The numerous contributions of the economic literature that have aimed to analyse the issue of competition in an industrial sector, both on the theoretical and empirical level, can be distinguished at least in three theoretical approaches. The first, known as Structure-Conduct-Performance paradigm (SCP), claims that the degree of competition in a market can be explained in terms of the conduct of firms which is, in turn, determined by the structural characteristics of the market itself (number and size of firms, condition of price and demand, etc.). According to this paradigm, structural changes that lead to a concentration in an industrial sector may facilitate collusive behaviours between firms and, therefore, lead to a reduction in the degree of competition. In contrast with the SCP paradigm, other contributions aim to show that greater concentration in an industry does not imply a reduction in competition but rather an increase in the level of efficiency in the sector as a whole. This is because it is the most efficient firms that increase 3 The largest five of them (the top five in Table 1) held, at the end of 2000, over 54% of the total assets of the entire banking sector. 4 On this issue, see among others Fich and White (2001) and, with reference to the Italian banking sector, Bianco and Pagnoni (1997) and Barbi (2000).

6 84 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) their market shares at the expense of their less efficient competitors. According to this paradigm, known as the efficient-structure hypothesis (ESH), greater concentration tends to emerge as a consequence of a more vigorous competition between firms in the market. 5 The third approach to the issue of industrial competition has been proposed in the context of the studies of the New Economic Industrial Organisation (NEIO). This line of research, unlike the other two aforementioned approaches, bases the analysis of competition in markets on nonstructural models, that is on theoretical schemes which do not rely exclusively on information concerning the structure of markets. Studies developed within the ambit of the NEIO approach use analytical techniques that are linked to two basically empirical methodologies: one based on the conjectural variations proposed by Lau (1982) and Bresnahan (1982), 6 and the other focused on the use of the H statistic of Rosse and Panzar (1977) and Panzar and Rosse (1982, 1987). If the first methodology allows us to obtain an indicator of the market power of firms from the estimation of three simultaneous equations (demand, supply and prices), the second offers a way to discriminate between different market structures based on the specification of the reduced-form revenues function at the individual firm level. Of the two methodologies, the H test of Panzar and Rosse presents the advantage that for its implementation -data availability becomes much less of a constraint, since revenues are more likely to be observable than output prices and quantities or actual cost data (Shaffer, 2004 presents a detailed analysis concerning the advantages and disadvantages of the two NEIO tests for market power). Derived from static (oligopoly) models, which determine equilibrium output and equilibrium number of firms by maximizing profits both at firm and industrial level, the statistic of Panzar and Rosse allows us to identify the nature of the market structure in which a firm operates by estimating the sum of the elasticities of the firm s revenues with respect to the firm s input prices. More formally, denoting with R the firm s revenues, with w a vector of K input prices (k =1,..., K), with Z and S two vectors of exogenous variables shifting firm s revenue and cost functions, respectively and with ε the error term, the reduced-form revenues function for the firm i is given by: R i = f (w i,z i,s i,ε) (1) Panzar and Rosse (1987) argue that the market power of a firm can be measured by the extent to which changes in factor prices (dw ki ) are reflected into equilibrium revenues (dr i ) earned by the firm itself. By indicating H as the sum of elasticity of total revenues with respect to input prices: K ( R ) H = i w ki w k=1 ki R (2) i the authors prove that: H is equal to 1 when firms operate under perfect competition (Shaffer, 1982 proves that the same result can be obtained for a natural monopolist operating in a perfectly contestable market); H is negative or equal to zero in the case of monopoly or perfectly collusive oligopoly (Vesala, 1995 proves that the same result holds for monopolistic competition without the threat of entry and Shaffer, 1983 shows that H is negative also for any conjectural variations 5 For a recent review of the debate between the structure-conduct-performance paradigm and the efficient-structure hypothesis, see Gilbert and Zaretsky (2003). 6 For applications of this methodology to the banking industry see, among others, Shaffer (1989, 1993, 2001), Berg and Kim (1994), Coccorese (2002) and Gruben and McComb (2003).

7 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) oligopoly or even for short-run competitive equilibrium); H is positive and less than 1 in the case of monopolistic competition with freedom of entry. Although a proof of these various outcomes is technical (see Panzar and Rosse, 1987; Vesala, 1995 for details on the formal derivation of the H statistic), the intuition lying behind them is straightforward. In long-run competitive equilibrium, firms operate in the minimum of the average costs, which then is also equal to the price. Under the hypothesis that function costs are homogeneous of degree one in factor prices (or that the production functions are homothetic), an increase in input prices raises equiproportionally both marginal and average costs, while the optimal output of any individual firm will not change. Exit of some firms increases the demand faced by each of the remaining firms, leading to an increase in prices and revenues by the same amount as the rise in costs (H = 1). By contrast, H will be negative (or zero) for a monopoly or a perfectly collusive oligopoly: a rise in input price increases marginal costs, reduces equilibrium output and subsequently the firm s revenues. Finally, if the market structure is characterized by monopolistic competition, then the H statistic will lie between 0 and 1: under monopolistic competition revenues will increase less than proportionally to changes in input prices, as the demand facing individual firms is inelastic. Generalising a result of Panzar and Rosse (1987), Vesala (1995) proves that the H statistic is an increasing function of the absolute price elasticity of demand, that is an inverse measure of market power exercised on the part of firms. This implies that values of H between 0 and 1 can be interpreted as a continuous measure of level of competition, in the sense that higher values of the statistic indicate stronger competition than lower values (Bikker and Haaf, 2002). From an econometric point of view, the monopoly model is not consistent with the data when the H < 0 hypothesis is rejected, while the monopolistic competition model is the only one consistent if the hypotheses H < 0 and H = 1 are both rejected. Finally, the rejection of the hypothesis H 1 rules out all the three models (Panzar and Rosse, 1987). A crucial feature of the H statistic is that it must be undertaken on observations that are in long-run equilibrium. Indeed, as Panzar and Rosse (1987) explain, this hypothesis is important for the cases of perfect competition and monopolistic competition, while it does not constitute a fundamental prerequisite in the case of monopoly since H 0 is a long run condition for monopoly. To test if observations are in long-run equilibrium, one can assume that competitive markets equalise the return rates across firms, so that in equilibrium these rates should not be correlated with input prices. Empirically, this test can be carried out by using an indicator of firm return as dependent variable in the estimation of H. In this context, H = 0 implies that the data are in equilibrium (Shaffer, 1982). As claimed by several authors, various assumptions need to be made when the Panzar and Rosse methodology is applied to the banking sector (see, among others: De Bandt and Davis, 2000; Hempell, 2002; Gelos and Roldos, 2004). A first one is that banks are treated as single product firms. This requires that, consistent with the so-called intermediation approach, banks are viewed as producing intermediation services (in this case, loans and investments are used as output measures) using labour, physical capital and financial capital (deposits and funds from financial markets) as inputs. 7 The level and nature of competition in the loan market and that in the deposit market are taken to be entirely independent. The use of bank specific input prices indicates that banks are not necessarily price-takers in factor markets, or may face local factor 7 According to another approach to bank output measurement, known as production approach, banks use capital and labour to produce different categories of loan and deposit account (see Colwell and Davis, 1992).

8 86 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) markets (De Bandt and Davis, 2000). Moreover, one needs to assume that higher factor prices are not correlated with higher revenues generated by higher quality services, since this may bias the computed H statistic. However, if the hypothesis of contestable markets is rejected, the bias cannot be too large (Molyneux et al., 1996). Further assumptions include profit maximization and normally shaped revenue and cost functions (Gelos and Roldos, 2004). Despite the equilibrium test, the above assumptions and other limitations (see Hempell, 2002; Shaffer, 2004), the Panzar and Rosse methodology presents several advantages which have given rise to numerous applications. 8 Firstly, as noted above, the estimation of reduced-form revenues equation is often possible even when structural equations cannot be estimated. Moreover, by not requiring an a priori definition of the geographic market, the H test avoids the potential bias caused by misspecification of market boundaries; the behaviour of the individual banks gives an indication of their market power (Hempell, 2002). Furthermore, as the H statistic is a direct measure of the degree of competition taking competitive effects from other institutions in its stride, the Panzar and Rosse approach does not require observations of non-bank financial institutions (Bikker and Haaf, 2002). Finally, as highlighted by Claessens and Laeven (2004), being based on bank-level data, the Panzar and Rosse methodology allows one to study differences among banks (e.g., large versus small, foreign versus domestic and so on). In this paper, the Panzar and Rosse methodology has been applied to a taxonomy of Italian banking firms that draws a distinction between banks connected by cross-ownership and credit firms untouched by this phenomenon. The data, the sample and the econometric specification used for the estimation of the H indicator are illustrated in the next section. 4. Data, sample and econometric methodology 4.1. Data and sample As mentioned above, the information concerning the ownership structure of the Italian banks, which enables us to distinguish between credit firms involved in cross-ownership and those not involved, have been gathered mainly from the CONSOB and, for the banking groups not quoted on the Italian Stock Exchange on the , from the Il Sole 24 Ore (the most important Italian financial newspaper). On the other hand, the data used in the econometric implementation of the H statistic have been taken from the database BILBANK 2000, edited by the Italian Banking Association (ABI). These data are annual and refer to the period , that is the years in which cross-ownership became an important phenomenon in the Italian banking industry (see Section 2). Excluding the ex Istituti di Credito Speciale that is a category of banks that, until the first half of the 1990s, operated only in the long-run and the branches of foreign banks, BILBANK 2000 provides balance-sheet data on 518 Italian banks for a total of 2312 observations. After deleting 82 observations, for which the values of some variables used in the regressions appeared as possible outliers, the sample for the econometric analysis is made up of an unbalanced panel of 503 credit firms, for a total of 2230 observations. Of these, 2042 are related to banks unaffected by cross-ownership (1295 of which concern a category of small cooperative banks called Banche di Credito Cooperativo, henceforth BCC) and 188 are related to credit firms affected by cross- 8 For an extensive literature review of the studies that starting with Shaffer (1981a,b, 1982) applied the Panzar and Rosse statistic to the banking industry, see Koutsomanoli-Fillipaki and Staikouras (2004).

9 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) ownership. In terms of total assets, the sample represents for each year considered more than 60% of the entire Italian credit system. De Bandt and Davis (2000) argue that it is preferable to use a balanced data set in the estimation of Panzar and Rosse s statistic, as the interpretation of H rests on the assumption that the banks have reached their steady state. This implies, for example, excluding from the sample banks born in the reference period for the analysis, since they may have an atypical behaviour (see, for example, De Young and Hasan, 1998). On the other hand, however, the use of an unbalanced data set allows us to carry out estimations with a greater number of observations. This certainly could be an advantage so long as empirical evidence is found to confirm that the banks in the sample are in long term equilibrium. As said, such a test is simple to implement. In this paper, the empirical investigation is carried out firstly by including in the sample also the new banks (those born in the 1990s), and then as a robustness check by excluding all observations of credit firms less than 10 years old. In the sample, banks affiliated to credit firms connected by cross-ownership have been treated as if they were themselves involved in cross-ownership. The hypothesis formulated here is that, if cross-ownership influences the competitive behaviour of the head of the group, it will then influence in some way the affiliated banks. 9 In relation to the year in which affiliation took place, banks belonging to groups have been treated as independent and thus, when it is the case, as not involved in cross-ownership (the basic assumption is that possible changes in the management and/or other changes connected with the acquisition might not be captured by the balance-sheet data relative to the year of the operation). As in Hempell (2002), the data set is not adjusted for bank mergers, i.e. two merging banks are treated as separate up to the year they are merged and, from then on, only the take-over bank is accounted for. Finally, it should be noted that given the limits inherent to the construction of Table 1 (see Section 2) it is not possible to exclude the possibility that the category of banks not involved in the phenomenon of cross-ownership might actually contain some banks so connected Econometric methodology In general terms, the econometric model of the Panzar and Rosse statistic may be represented by the following equation: logr it = K J α k log Wit k + β j log S j N it + γ n log Xit n + ε it k=1 j=1 n=1 i = 1,...I; t = i,...t; (3) where R is a measure of gross revenue; W is a vector of factor prices (the H statistic is given by the sum of the estimated coefficients of the variables in this vector); S is a vector of scale variables; X is a vector of exogenous and bank-specific variables that may shift the cost and revenue schedule (risk, business mix etc.); ε indicates the error term; I is the total number of banks; T is the number of periods observed. In this paper, a first estimate of Eq. (3) has been carried out by employing a specification close to that used by De Bandt and Davis (2000) (Model 1). According to them, the proxy for for 9 Zazzaro (2001) claims that it does not appear unlikely that the model of governance of an affiliated bank that is to say a bank that is not independent from an economic and strategic point of view copies that of the head of the group.

10 88 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) Table 2 Definition of variables GIR Gross Interest Revenues Interest received TGR Total Gross Revenues Interest received + income from banking services + other operating income (exceptional items excluded) TA Total assets UPL Unit Price of Labour Personnel expenses to number of employees UPC (in MOD 1) Unit Price of Capital Physical capital expenditure (depreciation, write-down on intangible and tangible assets) plus other operating expenses (exceptional items excluded) to total assets UPC (in MOD 2) Unit Price of Capital Physical capital expenditure (depreciation, write-down on intangible and tangible assets) plus other operating expenses (exceptional items excluded) to fixed assets UPF Unit Price of Funds Interest paid to total funds, where total funds = Deposits + money market liabilities, the latter including subordinated debt EQ Equity to total assets LTA Loans to total assets DFT Deposits to total funds CDTD Customer deposits to total deposits OITA Non-interest income Income from banking services + other operating income (exceptional items excluded) to total assets ROA Return on assets Gross profits to total assets the dependent variable is given by the gross interest revenues (GIR), that is the interest received. Hempell (2002) questions the use of such a proxy, by indicating potential endogeneity problems it can cause given that it is expressed in levels. Vesala (1995) and De Bandt and Davis (2000), on the other hand, claim that this is the most appropriate choice, as the use of revenues scaled with total assets implies that the estimated equation no longer represents a revenue equation but rather a price equation. Furthermore, considering only the interest revenues rather than the total revenues seems to be coherent with the conceptual structure implied by the application of Panzar and Rosse s statistic to the banking sector, that is to say with the idea that the activity of financial intermediation represents the core business of credit firms (Bikker and Haaf, 2002). As far as the variables relating to the input prices are concerned, the proxies used are: the ratio of personnel expenses to the number of employees (UPL), as a proxy for unit price of labour; the ratio of physical capital expenditure and other non-interest expenses to total assets (UPC) as a proxy for unit price of capital; the ratio of interest paid on liabilities to total funds (UPF) as a proxy for unit price of funds. Total assets (TA) is used as scale variable. 10 Finally, for the product mix variables, loans as proportion of assets (LTA) and deposits as proportion of total funds (DTF) are the proxies employed. A description of all the variables used in the empirical investigation is found in Table 2, while for a synthesis of the main statistical measures concerning these variables (for the entire sample) see Table 3. Regarding the sign of the explanatory variables, the scale variable is expected to have a positive effect on revenues. A positive parameter is expected also for LTA, as more loans reflect more potential rate income. For the other regressors, some authors hold conflicting theories and some others do not have a priori expectations. 10 In their specification, De Bandt and Davis (2000) use equity and fixed assets as scale variables.

11 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) Table 3 Summary statistics Mean S.D. Min Max Cases GIR a , TGR a , GIR/TA b TGR/TA b UPL a UPC1 b UPC2 b UPF b TA a 2,730 11, , EQ b LTA b DFT b CDTD b OITA b ROA b Overall sample. GIR: gross interest revenues; TGR: total gross revenues; TA: total assets; UPL: personnel expenses to number of employees; UPC1: other non-interest expenses to total assets; UPC2: other non-interest expenses to fixed assets; UPF: interest paid on liabilities to total funds. EQ: equity to total assets; LTA: loans to total assets; DFT: deposits to total funds; CDTD: customer deposits to total deposits; OITA: non-interest income to total assets; ROA: gross profit to total assets. For more about the definition of the variables used in the estimates see Table 2. Further descriptive statistics can be provided upon request. a In millions of Euro. b Percentage. Eq. (3) has been estimated by using a fixed effects panel regression with time dummies. This approach permits to estimate the relevant parameters of the empirical model by utilising both the cross-sectional and the temporal data. Moreover, bank-fixed effects allow us to control for unobserved heterogeneity, and this is important since the regressions are otherwise to suffer from omitted variable problems. Fixed effects capture all bank specific non time-varying determinants of revenues not explicitly addressed in the regression specification. 11 Time dummies variables are added to the model in order to account for yearly macro effects. Finally, in order to carry out the equilibrium test (see Section 3), Eq. (3) has been re-estimated by using the ratio between gross profits and total assets (a proxy for the return on assets, ROA) as dependent variable. Also for this test a fixed effect model has been employed since as discussed in the next section it turns out to be the appropriate estimation approach. With respect to the equilibrium test, it should be noticed that as argued by De Bandt and Davis (1999, page 10) equilibrium does not mean that competitive conditions are not allowed to change an assumption which would be contradicted by the period that we consider, characterised by a process of structural changes. It only implies that changes in banking are taken as gradual. This assumption holds also for the analysis carried out here, given that the restructuring process that occurred in the Italian banking industry throughout the 1990s though it was characterised by an important consolidation activity and not rare entry and exit movements of banks appears to be gradual, as the data concerning this process seem to suggest (see Banca d Italia, ; De Bonis, 2003). 11 Whether the individual effects are fixed or random is tested by applying the Hausman test (see Section 5).

12 90 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007) Robustness To check if the econometric results are sensitive to the chosen model, several proofs of robustness have been carried out. Firstly, a further specification of equation (3) has been estimated (Model 2). Following Bikker and Haaf (2002), the ratio of personnel expenses to the number of employees (UPL), the ratio of physical capital expenditure and other expenses to fixed assets (UPC2) and the ratio of annual interest expenses to total funds (UPF) have been used as proxies for the price of labour, capital and funds, respectively. Once again, total assets is the scale variable. The ratio of equity to total assets (EQ) has been introduced to account for bank-specific risk. Loans as proportion of total assets (LTA) and the ratio of customers deposits on total deposits (CDTD) are the proxies for business mix. Finally, in order to take into account the influence exerted on banks marginal revenue and cost functions by non-interest income, the ratio of other income to the total assets (OITA) has been introduced as explanatory variable. Other income includes revenues from banking services and other operating income. The coefficient for OITA is expected to be negative, as the generation of other income may be at the expense of interest income (Bikker and Haaf, 2002). Following some authors (De Bandt and Davis, 2000 and Coccorese, 2004, among others) a second robustness proof has been carried out by replacing for both the models considered gross interest revenue (GIR) with total gross revenues (TGR) as dependent variable, where TGR is obtained as the sum of gross interest revenues plus non-interest income, the latter defined above. Besides, similarly to some other authors (Bikker and Groeneveld, 2000; Hempell, 2002; Gelos and Roldos, 2004), for both the specifications taken into account Eq. (3) has been re-estimated by using, as dependent variable, once GIR and once TGR, both scaled with total assets. 12 Finally, as mentioned above, all the estimations have been re-run after excluding from the sample observations of credit firms less than 10 years old. These banks are 41 involving a total of 157 observations, almost all of them (156) related to credit firms not involved in cross-ownership (138 of which concern BCC). 5. Empirical results The results obtained from the econometric analysis are shown in Tables 4 7. Indeed, they refer to the estimations carried out on the sample including also new banks, while the results found from the regressions run on the sample excluding observations of credit firms less than 10 years old are not reported, since they leave unchanged the conclusions that are discussed in this section. 13 Tables 4 6 report the estimated coefficients of the explanatory variables (the H statistic is given by the sum of the estimated parameters of UPL, UPC and UPF) and the following diagnostic tests: an F-test for the joint significance of all the regressors, including the individual and temporal dummies (model test); an F-test both for the null hypothesis H = 0 (monopoly) and for the null H = 1 (perfect competition); the Lagrange Multiplier statistic to test for the existence of individual unobserved heterogeneity, that is to test the classical regression model against the panel regression approach; finally, the Hausman statistic to test the null hypothesis that explanatory variables and individuals effects are uncorrelated, namely to evaluate if the individual effects are 12 When TGR represents the dependent variables, OITA is not included as regressor. Besides, when the dependent variable is scaled with total assets, TA does not compare on the right-hand side of Eq. (3). 13 These results are available from the author upon request.

13 Table 4 H statistic for the overall sample GIR TGR GIR/TA TGR/TA MOD1 MOD2 MOD1 MOD2 MOD1 MOD2 MOD1 MOD2 UPL (0.0010) (0.0000) (0.0020) (0.0010) (0.0010) (0.0000) (0.0020) (0.0010) UPC (0.0000) (0.0040) (0.0000) (0.0000) (0.0000) (0.0030) (0.0000) (0.0000) UPF (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H Statistic TA (0.0000) (0.0000) (0.0000) (0.0000) EQ (0.3920) (0.0260) (0.2880) (0.0150) LTA (0.2660) (0.1310) (0.0750) (0.0160) (0.2420) (0.1180) (0.0690) (0.0140) DTF (0.1900) (0.1750) (0.1600) (0.1600) CDTD (0.0170) (0.0000) (0.0000) (0.0000) OITA (0.0120) (0.0120) ADJ R Model test * (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =0 ** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =1 *** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) LM test a (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Hausman b (0.0790) (0.0000) (0.0000) (0.0025) (0.0000) (0.0000) (0.0012) (0.0007) OBS GIR: gross interest revenues; TGR: total gross revenues; TA: total assets; UPL: personnel expenses to number of employees; UPC: other non-interest expenses to total assets in model 1 and other non-interest expenses to fixed assets in model 2; UPF: interest paid on liabilities to total funds; EQ: equity to total assets; LTA: loans to total assets; DFT: deposits to total funds; CDTD: customer deposits to total deposits; OITA: non-interest income to total assets. For more about the definition of the variables used in the estimates see Table 2. All the variables are in natural logarithmics. In italics are reported the p-values of the tests. The t statistics (not reported) are based on robust standard errors (Arellano, 1987). H statistic is the sum of the estimated parameters of UPL, UPC and UPF. Time dummies included, but not explicitly reported. a Lagrange multiplier test: H 0 = classical regression model (CRM) vs. fixed effects model (FEM)/random effects model (REM). b Hausman test: H 0 = REM vs. FEM. * F-test for the joint significance of all the regressors, including individual and time dummies. ** F-test for the null H = 0 (monopoly). *** F-test for the null H = 1 (perfect competition). F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007)

14 Table 5 H statistic for the comparison NCRO (banks not invoved in cross ownership) versus SCRO (banks with cross-ownership) GIR MOD1 MOD2 MOD1 MOD2 NCRO SCRO NCRO SCRO NCRO SCRO NCRO SCRO UPL (0.0030) (0.0090) (0.0010) (0.0150) (0.0030) (0.0360) (0.0010) (0.0300) UPC (0.0000) (0.0130) (0.0010) (0.3320) (0.0000) (0.0060) (0.0000) (0.5470) UPF (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H Statistic TA (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) EQ (0.5700) (0.6860) (0.0500) (0.4310) LTA (0.1060) (0.0090) (0.0090) (0.0090) (0.0390) (0.1150) (0.0130) (0.1970) DTF (0.0200) (0.3800) (0.0220) (0.0990) CDTD (0.0650) (0.0400) (0.0060) (0.1280) OITA (0.0350) (0.4290) ADJ R Model test * (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) 4231 (0.0000) (0.0000) (0.0000) H =0 ** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =1 *** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) 4.89 (0.0270) (0.0000) 8.41 (0.0037) LM test a (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Hausman b (0.0553) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0052) (0.0002) OBS TGR 92 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007)

15 GIR/TA TGR/TA MOD1 MOD2 MOD1 MOD2 NCRO SCRO NCRO SCRO NCRO SCRO NCRO SCRO UPL (0.0030) (0.0190) (0.0010) (0.0450) (0.0030) (0.0280) (0.0010) (0.0250) UPC (0.0000) (0.0000) (0.0000) (0.7530) (0.0000) (0.0000) (0.0000) (0.5710) UPF (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H Statistic TA EQ (0.3960) (0.2140) (0.400) (0.0910) LTA (0.0390) (0.0470) (0.0580) (0.0530) (0.0390) (0.1480) (0.0120) (0.2690) DTF (0.0150) (0.4490) (0.230) (0.1270) CDTD (0.0430) (0.0060) (0.0060) (0.0080) OITA (0.0350) (0.8010) ADJ R Model test * (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =0 ** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =1 *** (0.0000) 8.00 (0.0047) (0.0000) 9.83 (0.0017) (0.0000) 3.05 (0.0805) (0.0000) 6.56 (0.0104) LM test a (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Hausman b (0.0000) (0.0000) (0.0000) (0.0000) (0.0009) (0.0002) (0.0026) (0.0002) OBS GIR: gross interest revenues; TGR: total gross revenues; TA: Total assets; UPL: personnel expenses to number of employees; UPC: other non-interest expenses to total assets in model 1 and other non-interest expenses to fixed assets in model 2; UPF: interest paid on liabilities to total funds; EQ: equity to total assets; LTA: loans to total assets; DFT: deposits to total funds; CDTD: customer deposits to total deposits; OITA: non-interest income to total assets. For more about the definition of the variables used in the estimates see Table 2. All the variables are in natural logarithmics. In italics are reported the p-values of the tests. The t statistics (not reported) are based on robust standard errors (Arellano, 1987). H statistic is the sum of the estimated parameters of UPL, UPC and UPF. Time dummies included, but not explicitly reported. a Lagrange multiplier test: H 0 = classical regression model (CRM) vs. fixed effects model(fem)/random effects model (REM). b Hausman test: H 0 = REM vs. FEM. * F-test for the joint significance of all the regressors, including individual and time dummies. ** F-test for the null H = 0 (monopoly). *** F-test for the null H = 1 (perfect competition). F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007)

16 Table 6 H statistic for the comparison NCRO-B (banks not involved in cross ownership, excluding BCC) vs. SCRO (banks with cross-ownership) GIR MOD1 MOD2 MOD1 MOD2 NCRO-B SCRO NCRO-B SCRO NCRO-B SCRO NCRO-B SCRO UPL (0.0440) (0.0090) (0.0270) (0.0150) (0.0160) (0.0360) (0.0100) (0.0300) UPC (0.1170) (0.0130) (0.4080) (0.3320) (0.0010) (0.0060) (0.7910) (0.5470) UPF (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H Statistic TA (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) EQ (0.2650) (0.6860) (0.9160) (0.4310) LTA (0.0050) (0.0090) (0.0100) (0.0090) (0.0020) (0.1150) (0.0020) (0.1970) DTF (0.7500) (0.3800) (0.8510) (0.0990) CDTD (0.5240) (0.0400) (0.0280) (0.1280) OITA (0.8400) (0.4290) ADJ R Model test * (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =0 ** (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) H =1 *** 1.25 (0.2636) (0.0000) 2.92 (0.0877) (0.0000) 1.29 (0.2566) 4.89 (0.0270) 4.03 (0.0448) 8.41 (0.0037) LM test a (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Hausman b (0.0009) (0.0000) (0.0001) (0.0000) (0.0001) (0.0000) (0.0023) (0.0023) OBS TGR 94 F. Trivieri / Int. Fin. Markets, Inst. and Money 17 (2007)