Competition, ownership and bank performance in transition

Transcription

1 Competition, ownership and bank performance in transition by Steven Fries,* Damien Neven** and Pau Seabright*** June 2004 Abstract This paper examines factors that infuence the revenues and costs of banks in the postcommunist transition. The anaysis yieds severa key findings. First, in an eary subperiod soon after the start of transition, the average margins earned on oans and deposits by private banks grouped by their origin and ownership are cosey correated with the margina costs of providing these services. Newy estabished banks had reativey ow margins and margina costs compared with privatised banks. In the ater sub-period, the variation in margina costs among private banks decines as does that of margins. State banks operated in both sub-periods with reativey ow average margins compared with their estimated margina costs. In addition, the anaysis of bank margins reveas that competitive pressures among banks increased significanty over time. These findings are consistent with the market entry of new private banks putting downward pressure on margina costs and margins and with competitive pressures reducing the differences among banks over time. JEL codes: G2, L1, L8, P2 Keywords: banking, revenue functions, cost functions, transition. * Corresponding author, European Bank for Reconstruction and Deveopment, One Exchange Square, London EC2A 2 JN, United Kingdom, emai: friess@ebrd.com. ** Graduate Institute of Internationa Studies, University of Geneva, and CEPR. *** Université de Tououse I and CEPR. The vauabe research assistance of Nadia Aeshina is gratefuy acknowedged. We aso thank Erik Bergöf, Wendy Carin, Christa Hainz, Mark Schaffer Caudia Senik- Leygonie and Anita Taci for hepfu comments and suggestions. The funding of the Japan-Europe Cooperation Fund is gratefuy fuy acknowedged. The usua discaimer appies.

2 1. Introduction Since its start in 1989, the post-communist transition has transformed fundamentay East European banking systems. Under the command economy, state authorities directed credit aocation with scant regard for capacity to repay, using state banks to channe funds to state (or sociay) owned enterprises for inputs and investments authorised under panning. To direct resources in this way, banks speciaised by economic sectors (or foreign trade), rather than diversified across them. State savings banks speciaised in coecting deposits from househods, athough most savings was forced and done by the state. The payment system consisted of a cash circuit for househods and commercia transfers among enterprises handed by the centra bank. At the same time, without a profit incentive, state banks were not encouraged to compete for oans and deposits or to contro costs. Because of this structure of sociaist banking systems, and with the start of transition towards a market economy, banks had to restructure fundamentay their outputs and use of inputs. Governments and centra banks in Eastern Europe have impemented severa types of poicies aimed at transforming sociaist banking systems into market-oriented ones (see, for exampe, Anderson, Bergöf and Mizsei, 1996, EBRD, 1998, and Bergöf and Boton, 2002). Banking systems were iberaised by freeing interest rates and decentraised by transferring commercia banking activities from the centra bank to state banks. State banks were restructured and privatised and new private banks, both domestic and foreign, were aowed to enter the markets. Moreover, to support arms-ength ending reationships between banks and their borrowers and to foster confidence of depositors in banks, ega frameworks were overhaued (incuding the 2

3 strengthening of creditor rights) and systems of prudentia reguation and supervision were initiated. In broad terms, the main poicy instruments to promote the transformation of banking were therefore iberaisation, privatisation, competition and fundamenta institutiona change. To understand this process of change in post-communist countries, we examine the changing nature of competition in their banking systems. To do so, we mode and investigate empiricay the structure of margins on oans and deposits and the costs of providing these services. This approach buids on the extensive empirica studies of competition in banking in industriaised economies beginning with Shaffer (1982) 1 and the recent studies of competition in banking in deveoping countries and transition economies by Geos and Rodós (2002), Yidirim and Phiippatos (2002b) and Drakos and Konstantinou (2003). These studies find that banking systems are often characterised by imperfect competition. In a word-wide study of the factors that infuence the extent of competition in banking, Caessens and Laeven (2003) shows that banking systems with greater foreign entry and fewer activity restrictions tend to be more competitive. In addition to this empirica iterature, theoretica anayses of banking in transition economies emphasise the importance of both competition and institutiona deveopment in the transformation of banking and the deveopment of market-based financia intermediation. In a mode of coateraised ending by oigopoistic banks, Hainz (2003a) shows that, if creditor rights are weaky protected, banks have more market power other factors being equa and are abe to extract more rents from their borrowers, thereby hoding back the scae of intermediation. Hainz (2003b) aso demonstrates that in mode of spatia competition among banks the intensity of 1 The empirica iterature on competition in banking in industria countries is extensive. Recent 3

4 competition depends not ony on the number of banks (that is, the average distance between them), but aso on the extent and effectiveness of creditor rights. One impication of this mode is that the intensity of competition among a given number of banks in the market increases as the quaity of institutions improves. A key feature of both modes is that a better institutiona framework (that is, the more effective protection of creditor rights) strengthens the competitive pressures among banks. In this paper, we deveop a mode of monopoistic competition in banking and use it to investigate the equiibrium structure of margins earned on oans and deposits. This mode enabes us to identify the associations between the origin and ownership of banks, the extent of competition and their margins and operating costs. The origin and ownership variabe is used as an indirect measure of the capacity of a bank to differentiate its oans and deposits and to contro its costs. The mode aso aows us to test for the extent of product differentiation and the intensity of competition associated with the number of banks in a market. We aso investigate the structure of operating costs in transition banking. To impement empiricay our mode of monopoistic competition in banking and to investigate the structure of bank costs, we use a pane dataset consisting of 272 banks in 15 East European countries (Bugaria, Croatia, the Czech Repubic, Estonia, FYR Macedonia, Hungary, Kazakhstan, Latvia, Lithuania, Poand, Romania, Russia, the Sovak Repubic, Sovenia and Ukraine) and covering the years 1995 to Fries and Taci (2004) deveoped the data set, which is unique in its coverage of timevarying bank ownership in transition economies. Their reated study examines the reative cost efficiency of banks within the transition economies. The empirica estimations of our mode of monopoistic competition in contributions incude Bikkar and Groeneved (2000), De Bandt and Davis (2000) and Bikkar and Haaf (2002). The references in these papers provide a more extensive guide to this iterature. 4

5 transition banking and of the cost function, both of which aow for structura changes between to sub periods that divide the entire sampe approximatey in haf, yied a number of interesting resuts. First, in the earier sub-period, the average margins on oans and deposits earned by banks grouped by their origin and ownership are significanty correated with the estimated margina costs of providing these services. Moreover, newy estabished banks have reativey ow margina costs and margins compared with those of privatised banks. In the ater sub-period, the variation in margina costs among banks decines, as does that of margins. State banks operated in both sub-periods with reativey ow average margins compared with their estimated margina costs. Moreover, the anaysis of bank margins indicates that competitive pressures among banks increased significanty over time. Thee findings are consistent with the entry of new private banks putting downward pressure on margins and margina costs and with competitive pressures reducing the differences among private banks that remained in the market. The increase in competitive pressures over time coincided with improvements in the quaity of institutions. State-owned banks that remained in the market, however, may not have faced the same constraints of profitabiity as private banks. 2. Literature on competition, profitabiity and costs of banks in transition The existing empirica iterature on transition banking that is cosey reated to this study consists of three strands. The first examines the extent of competition in these banking markets. The second strand studies the reative profit and cost efficiency of banks within transition economies. The third considers factors that infuence net interest margins in transition banking. In the first strand, Geos and Rodós (2002), Yidirim and Phiippatos (2002b) 5

6 and Drakos and Konstantinou (2003) use the methodoogy of Panzar and Rosse (1987) to assess the competitive conditions in transition banking. The Panzar-Rosse H-statistic measures the percentage change in a bank s equiibrium revenues associated with a one per cent change in a the bank s input prices. This statistic can be used to infer the competitive structure of the industry in which the bank operates. An H-statistic vaue of one is associated with perfect competition and a vaue of zero or ess with monopoy or perfect cousion. Vaues in the range between zero and one characterise cases of monopoistic competition. The Geos and Rodós study covers three transition economies (the Czech Repubic, Hungary and Poand, as we as five Latin American countries) and the period 1994 to For the transition economies, the estimated vaues of the H- statistics ie between the vaues of one (perfect competition) and zero (monopoy); athough, in the case of Hungary the H-statistic is sufficienty cose to one that the hypothesis of perfect competition cannot be rejected. Moreover, this study finds that the competitive conditions remain broady stabe between two sub-periods, 1994 to 1996 and 1997 to The studies of Yidirim and Phiippatos and of Drakos and Konstantinou cover most of the transition economies covered in this paper. Their respective sampe periods are aso broady simiar. Both find that most banking markets in Eastern Europe are characterised by monopoistic competition. However, for FYR Macedonia and the Sovak Repubic, Yidirim and Phiippatos cannot reject the hypothesis that banks act as if they were monopoies or perfecty cousive oigopoies. Drakos and Konstantinou cannot reject this hypothesis in the case of Estonia and Latvia. Unike Geos and Rodós, Yidirim and Phiippatos aso find that the extent of competition in transition banking has increased over time. 6

7 Studies of the reative profit and cost efficiency of banks within transition economies incude Grigorian and Manoe (2002), Yidirim and Phiippatos (2002a), Bonin, Hasan and Wachte (2004) and Fries and Taci (2004). Again, they cover most of countries covered by this study and their sampe periods are broady simiar. The efficiency measures and estimation methodoogies used in these studies vary. The studies of Fries and Taci and of Grigorian and Manoe examine cost efficiency using the stochastic frontier approach and data enveopment anaysis, respectivey. Yidirim and Phiippatos and Bonin, Hasan and Wachte use the stochastic frontier approach to anayse both cost and profit efficiency. Yidirim and Phiippatos aso empoy the distribution free approach. Each of these studies finds significant variation in across countries in bank efficiency and Fries and Taci, Grigorian and Manoe, and Yidirim and Phiippatos seek to expain this variation by examining country eve variabes as correates of bank efficiency or costs. Bonin, Hasan and Wachte, which focuses primariy on the effects of foreign ownership, simpy aow for fixed effects for both country and time. Grigorian and Manoe observe that bank cost efficiency is significanty and positivey associated with GDP per capita and weaky and positivey associated with a measure of progress in institutiona reform. Fries and Taci find no association between cost efficiency and GDP per capita, but a significant non-inear association between progress in institutiona reform and costs (decining eary in the process of banking reform and then increasing). Regarding the association with measures of market competition, the findings are mixed. Fries and Taci and Yidirim and Phiippatos find that greater competition in a banking market (measured by the share of foreign banks in tota assets and the Panzar-Rosse H-statistic, respectivey) is associated with greater cost efficiency. In addition, Yidirim and Phiippatos find that greater competition is 7

8 negativey associated with profit efficiency. This finding is consistent with competition reducing both costs and margins. In contrast, Grigorian and Manoe observe that higher banking market concentration is associated with great cost efficiency. The association between ownership and efficiency is broady consistent across the four studies. A find that banks with majority foreign ownership are more cost efficient than those with majority domestic ownership. However, Yidirim and Phiippatos and Bonin, Hasan and Wachte observe that majority foreign ownership of banks is not associated with greater profit efficiency. In addition, Fries and Taci find that private banks newy estabished after the start of transition (new banks) are more cost efficient than state bank and privatised banks with majority domestic ownership, but ess efficient that privatised banks with majority foreign ownership. However, Grigorian and Manoe observe no difference in the cost efficiency of new banks and that of od banks (privatised and state owned). In the third strand of existing empirica research on transition banking, Drakos (2003) examines the net interest margins of banks in eeven post-communist countries over period 1993 to This study finds that net interest margins decrease significanty over time and that bank ownership has a significant effect. In particuar, Drakos observes that state-owned banks set significanty ower net interest margins that do other banks. 3. A mode of monopoistic competition in banking In this section, we derive a mode of banking market equiibrium under monopoistic competition that can be used to examine the structure of revenues earned by banks and their costs. A bank is regarded as muti-product firm that manages both 8

9 its assets and iabiities, incuding ending and deposit taking. A bank can aso invest in non-oan assets, such as securities. At the same time, a bank faces the baance sheet constraint that tota assets must be equa to tota iabiities pus equity capita. To satisfy the baance sheet constraint, it has the opportunity of borrowing or ending in the inter-bank market. The profit of a bank therefore incudes the returns obtained from its ending activities and non-oan assets, the interest paid on its deposits, and the interest earned from or paid out on its net position in the inter-bank market, as we as the operating cost of undertaking its activities. Let D, L and N denote deposits, oans and non-oan financia assets of bank i in country j at time t, excuding its gross inter-bank positions. R is the inter-bank rate in country j at time t, whie d r, r and n r are respectivey the interest paid on deposits and the interest earned on oans and nonoan assets. The expression C ( D L, N, W ) is a set of factor prices. Lasty, The profit of a bank,, refers to operating costs, where E ijc is the equity capita of a bank. Π can accordingy be written as W Π = r L + r n N r d D ( D, L, N W ) R ( L + N D E ) C,, (1) where the expression ( L N D E ) + represents the net debtor position of the bank in the inter-bank market. With a view to empirica estimation, the profit function of a bank can be rewritten as 9

10 Π = n d ( r R ) L + ( r R ) N ( R r ) D + R E ( D, L, N W ) C,, (2) In order to examine how a bank earns its profits, we consider separatey the revenues and costs of the bank. Given equation (2), the revenues of a bank, REV, are simpy n d ( r R ) L + ( r R ) N ( R r ) D Rtt E REV = +. (3) Since revenues, oans, non-oan assets, deposits and equity are observabe, this equation can be estimated directy. If this were done for our sampe of banks, the coefficients on the oan, non-oan financia asset and deposit variabes woud be estimates of the average margins that the banks have earned on these activities. The coefficient for bank equity woud be an estimate of the inter-bank rate or the opportunity cost of bank equity. We now assume that each bank sets the rates for its oans and deposits to determine the respective margins, but that the returns on non-oan financia assets and the inter-bank rate are exogenous to each bank, This assumption refects the empirica evidence that banking markets in most countries are characterised by monopoistic competition, incuding the post-communist countries in eastern Europe (see Geos and Rodós, 2002, Yidirim and Phiippatos, 2002, and Drakos and Konstantinou, 2003). Accordingy, we aow for the oan and deposit margins charged by each bank and the amount of its oans and deposits to be determined jointy by the interaction between a bank s suppy curve and the demand that it faces. In other words, we focus on ending 10

11 and deposit taking as the activities in which banks can potentiay exercise market power. Banks are assumed to be price takers in the market for non-oan assets, such as government securities, and in the inter-bank market. The estimated coefficients for oans and deposits in the revenue equation can therefore be seen as equiibrium margins and variation in the structure of these equiibrium margins across banks and countries and over time can be further expored. In particuar, the effects of exogenous variabes on equiibrium margins can be identified by specifying an equiibrium margin function. In what foows, we specify a mode of monopoistic competition among banks and derive equiibrium margins as a function of the underying parameters of the mode. It is important to emphasis that this estimation aows for neither the direct identification of market power nor the estimation of suppy functions. This woud require the estimation of a structura mode where demand and suppy functions are jointy estimated, using observed margins and quantities (see Bresnahan, 1989, for an exposition of this approach and Neven and Röer, 1999, for appications to banking). This cannot be done in this paper because we do not have data on oan and deposit margins. Nevertheess, the structure of the equiibrium margins may give some indirect insight into the nature of competition in transition banking. 2 Consider now the oan market (the anaysis can be appied in the same way to the deposit market). Assume that a bank takes the inter-bank rate and the rate on nonoan assets as given and that its oan poicy is independent of its strategy with respect to deposits. This wi be the case if the cost function is separabe in oans and deposits. 3 Assume that its margina operating costs with respect to oans is constant (a 2 This approach is therefore semi-structura and simiar in this respect to Panzar and Rosse (1987). 3 This simpifying assumption is reaxed in the empirica impementation of the mode. If the costs of deposit taking and ending are not separabe, as avaiabe evidence indicates (see, for exampe, Fries and Taci, 2004), the margina cost of a oan woud be a function of both outstanding oans and deposits. 11

12 inear margina cost coud aso be accommodated). The profit of the bank in the oan market is then given by ( M c ) L Π =, (4) where M r R is the bank s margin in the oan market and c itc is the margina cost of making a oan. Assume further that each bank faces an inverse demand function of the type M = a L λ L (5) ijc where L denotes the tota voume of oans sod by a other banks in the same country and time period and where 0 λ < 1. < This demand specification is adapted from Shubik and Levitan (1980) and aows for product differentiation. A bank may be abe to differentiate its oans in such a way that the demand curve is shifted out and the intercept a increases. The specification aso aows for reduced substitution between a bank s oans and those of its competitors in the market (that is, λ fas). Such reduced substitution can be associated with product differentiation (for exampe, through advertising) or market segmentation (for exampe, because of a imited number of competitors). The characteristics of each bank s oans that determine such differentiation cannot be observed directy. Faced with this demand specification, each bank wi maximise profit by soving the foowing first order condition 12

13 a 2 L λ L c = 0. (6) Summing up the first-order conditions for a banks in country c at time t yieds ( c ) 2 L ( n 1) L = 0 a λ, (7) where ( c ) ( a c ) a, i country j at time t and L is tota oans provided by the banking system in n is the number of banks in the country at that time. The voume of tota oans in equiibrium is therefore given by L * = a 2 + λ c ( n 1). (8) Combining equation (8) with the first order condition for each bank yieds as the equiibrium amount of oans made by bank i in country j at time t L = ( a c ) λ ( a c ) 2 λ ( 2 λ )[ 2 + λ ( n 1) ]. (9) From equations (5) and (9) a bank s equiibrium oan margin can therefore be expressed as 13

14 M = a ( 1 λ )( a c ) ( 2 λ ) λ ( a c ) ( 2 λ )[ 2 + λ ( n 1) ]. (10) parameters, The comparative static properties of the mode with respect to the fundamenta a, c and n,are straightforward to cacuate. For the equiibrium oan M M margins it is possibe to show that > 0, > 0 a c M and < 0. In other n words, a bank s equiibrium oan margin increases with its abiity to increase the demand for its oans and with margina oan costs and decreases with the number of competitors in the market. It can aso be shown that a bank s profits in a market equiibrium are increasing in its abiity to shift out its demand curve and to contro its Π costs; that is, > 0 a Π and < 0. A profit maximising bank woud therefore c seek to reduce both its cost and margins in order to boost its market share. It is aso possibe to express the equiibrium oan margin as a function of the equiibrium oan market share, * L L, which is observabe but depends on the underying mode parameters. In this expression, it is straightforward to show that the oan margin is an increasing function of the market share and the strength of the association between oan market share and margin increases as the vaue of decreases. The correation between oan margin and market shares can therefore be used as an indirect test for extent to which the ending services provided by banks in a market are substitutes. In the empirica impementation of the revenue mode, we use observabe variabes that are ikey to be correated with the underying structura parameters of λ 14

15 the mode, since the parameters, a, c and λ, cannot be observed directy. In particuar, we use the origin and ownership of banks as a variabe that may be correated with the abiity and incentive for banks to increase the demand for their oans and to manage their costs, that is a and c. In addition, we use the market share of a bank in the oan market to revea whether imited substitutabiity of ending services among banks is significant in determining margins. The number of banks in a country j at time t (reative to the country s popuation), N, is used as a measure of market competition. We contro for other country eve factors using fixed effects for countries and time. For costs, we use a standard trans-og specification. In particuar, the cost function takes the form nc t n t s 0 + β s s n Qs, + χ m m nwm, s t s, t n Qs, n = α β Q t, n m t n 1 2 m n m, n nwm, nwn, + δ s m s, m nqs, χ nw (12) m where Q s, are output quantities (that is, the amounts of oans, non-oan financia assets and deposits) and W, are input prices (wages and the cost of physica capita). u In estimating equation (12), we impose constraints on symmetry, β = and s, t β t, s χ =, and, homogeneity in input prices, χ = m, n χ n, m n m m 1, and adding-up, n m n, n = χ n, m = δ m, s = m χ. m 0 n m In the empirica impementation of this cost function, we aow for estimated parameters to vary with bank origin and ownership as a partia consistency check of the estimated revenue function, in which the equiibrium oan and deposit margins 15

16 vary with this observabe characteristic of banks. This specification for the cost function enabes us to compare the average margina costs of oans and deposit across different types of banks, one of the factors that may contribute to the variation in equiibrium margins. Again, we contro for country eve factors using fixed effects for countries and time. 4. Empirica impementation In the empirica impementation of revenue equation, we express revenues in the genera form of equation (3) and assume that banks engage in monopoistic competition for oans and deposits. The returns earned on non-oan assets and the inter-bank interest rates are assumed to be exogenous to individua banks. The equation for bank revenues therefore takes the form REV d ( ) Likt + M ( ) Dikt + ρ N + σe + ε = M.., (13) where ρ and σ are the exogenous average return that banks can earn on non-oan financia assets and the exogenous average inter-bank rate, respectivey. We write individua equiibrium oan and deposit margins as a function of the observabe characteristics of banks that may be correated with the underying parameters of the mode and measures of the intensity of competition in the banking market. Specificay, the equiibrium oan margin of bank i in country j at time t is q M = φ own, + ϕ S + γ N + η F + ε φ (14) p p p s f m 16

17 where own p is a vector of time-varying origin and ownership dummy variabes (privatised with majority foreign ownership, privatised with majority domestic ownership, newy estabished bank with majority domestic ownership, newy estabished bank with majority foreign ownership, and majority state owned), the oan market share of bank i in country j at time t, and S is N is the tota number of banks, F is the share in tota banking system assets of majority foreign-owned banks. The error term is assumed to have the usua properties. The same specification can be used for the equiibrium margins in the deposit market, simpy substituting the superscript d for. However, because the costs of deposit taking and ending are not separabe and because the amounts of oans and deposits by bank are highy correated, we estimate the mode using a singe scae variabe equa to the combined vaue of oans and deposit rather then estimate separatey the structure of margins for both oans and deposits. In so doing, we make the assumption that deposit taking and ending are considered independenty of other banking activities and that their combined margina cost is a inear function of the vaue of deposits and oans. This wi be the case if the cost of the combined deposit taking and ending activity is separabe from that of other banking activities. The estimated revenue function therefore takes the form, REV = M + d ( )( Likt + Dikt ) + ρ N + σe + ε.. (15) In the estimation, the origin and ownership, market share (weighted average of the oan and deposit market shares), number of banks (scaed by popuation) and asset share of foreign banks are interacted with the singe scae variabe. The estimated 17

18 coefficients on these variabes show how the average margin on oans pus deposits varies across banks with different observabe characteristics and with measures of the intensity of competition in the banking market. The coefficients on the non-oan assets and the equity variabes are estimates of the average rates of return and the riskfree interest rate respectivey. Given that we introduce disturbances on the margin equation, the overa error term takes the form + d ( L + D ) ε ε +. (16) To account for the induced heteroscedasticity of the errors, we estimate equation (15) using generaised east squares, and because of the presence of an error term in the coefficients on oans pus deposits, we instrument the scae variabe using its own agged vaue, the bank origin and ownership variabes, the proxy measures of banking market competition and the country and time fixed-effects. We aso instrument the average oan and deposit market share variabe since it is endogenous. A Durbin-Wu- Hausman test indicates that it is appropriate to instrument these variabes in the estimations reported beow. To estimate the revenue equation, we use the Batagi (1981) error-components two-stage east squares estimator (EC2SLS) (see Batagi, 1995, Chapter 7). Country and time fixed effects are used to contro for other country eve factors. We aso aow for bank-specific random effects in the estimations reported beow. A Hausman test ceary rejects a bank fixed-effects specification in favour of the EC2SLS estimation with random effects. In the empirica impementation of the trans-og cost function, we aow the 18

19 estimated parameters of the cost function to vary with the origin and ownership of banks, but omit most of the higher order terms in the fina trans-og specification. In preiminary estimations most of these terms were not statisticay significant. Consistent with the estimation of the revenue equation, we use as the output variabes the vaues of deposits and oans, as we as the vaue of non-oan financia assets. The two input prices are the average US doar wages in the financia sectors of the countries incuded in the sampe and the rea interest. The rea interest is the ex-post rea interest rate that prevaied in the country at the beginning of each year. As with the revenue equation, we use country and time fixed effects to contro for other country eve factors. The error term is assumed to have the usua properties. We estimate the revenue and cost equations whie aowing for bank specific effects and using both fixed and random effects specifications. As with the revenue equation, we report ony the random-effects specification because the Hausman test rejects the fixed-effects mode in favour of random effects. 5. Data sources and variabe descriptions The primary source of data on the banks baance sheets, income statements and ownership is the BankScope database produced by the Bureau van Dijk, which incudes data on 10,227 banks word-wide. The database is updated monthy and the atest issue of the BankScope database used in this study was May The BankScope data are suppemented with the data and information from annua reports of the banks and from EBRD staff research on bank ownership. The centra banks and the nationa statistica agencies of the countries provided aggregate data on their banking systems, incuding the tota number of banks, tota oans and deposits in the banking systems and share of foreign bank assets in tota bank assets. 19

20 In our sampe, we incude a banks in the BankScope database for which at east three years of data are avaiabe between 1994 and In addition, where banks report according to both oca accounting standards and internationa accounting standards for at east five years, we seect data in internationa accounting standard rather than nationa accounting standards for banks. This accounts for 57 per cent of the banks in the sampe. The sampe incudes 272 banks from 15 transition countries (see Tabe 1), 19 banks in Bugaria, 34 in Croatia, 23 in the Czech Repubic, four in Estonia, eight in FYR Macedonia, 23 in Hungary, nine in Kazakhstan, 18 in Latvia, 10 in Lithuania, 35 in Poand, 4 in Romania, 41 in Russia, 14 in the Sovak Repubic, 17 in Sovenia and 13 in Ukraine. A bank accounting data are in nomina terms in US doars converted at current exchange rates. The composition of banks in our sampe aso varies over the sampe period of 1995 to There are 81 banks for which data are avaiabe for the entire sampe, whie there are 191 banks that enter the sampe after 1994 and/or exit from the sampe before The additions to the sampe are not necessariy new market entrants, but rather successfu banks that are added to the BankScope scope database over time. Exits from the sampe are due primariy to either bank faiures or mergers with other banks. This method of seecting banks from the BankScope database introduces seection bias in the data, as does the seection by BankScope of banks to incude in the data set, which are primariy the arger and financiay sounder banks in the region. The estimation resuts are therefore representative not of the entire popuation of banks in transition economies, but rather of the reativey successfu top tier of banks in the region. The data on tota revenues and operating costs of banks come from the income 20

21 statement and baance sheets of the sampe banks, as reported in the BankScope database. Tota bank revenues incude net interest income pus non interest income. Costs incude genera operating expenses, but not interest expenses. Bank dividend payments are aso excuded from the measure of tota cost. We use four items from the baance sheets of banks incuded in the BankScope database. Customer oans and customer deposits incude a oans made to and deposits received from non-bank entities. Non-oan financia assets are the other earning assets of a bank ess inter-bank deposits paced with other banks. This amount incudes the securities hed by a bank. Bank equity incudes tota paid-in capita pus retained earnings. Bank origin and ownership are divided into five separate types. Private banks with no state-owned antecedents are referred to as newy estabished banks and they are distinguished by whether their majority owners are domestic or foreign entities. Private banks that were formery state-owned or part of a state-owned bank are referred to as privatised banks. They too are distinguished by whether their majority owners are domestic or foreign entities. The fifth bank ownership type is state owned. Data on the origin and ownership is from EBRD staff research and varies over time for each bank. Regarding measures of competition and market structure, the centra banks of the countries covered by the study provided data on the number of banks, share of foreign banks in tota bank assets and the tota amounts of oans and deposits in the banking systems. These data on tota amounts of oans and deposits, together with the data were used in cacuating the oan and deposit market shares of individua banks. For input prices, we use the oca currency wages in the financia sectors as reported by the nationa statistica agencies of the countries incuded in the sampe 21

22 and convert these amounts into US doars using period average exchange rates. We aso use the rea interest rate as measure of the opportunity cost of using physica capita in the production of banking services. The rea interest rate is measured ex post as the difference between the nomina interest in the money market at the beginning of the year and the consumer price infation rate in that year. In the cost equation, we aso incude the equity to asset ratio as proxy measure for other unobservabe characteristics of banks that reate to their abiity to contro costs. A higher equity ratio, for exampe, may be positivey correated with the capabiities of bank managers. Tabe 2 summarises the dataset that we use in the anaysis. It reports sampe means for the dependent and expanatory variabes for each of the five bank origin and ownership categories used in the estimations. The tabe aso provides average ratios to tota assets for revenues, operating costs and profits before taxes. These data indicate that privatised banks and state banks tend to be arger than newy estabished banks with greater market shares. They aso show that foreign banks tend to have ower revenues and operating costs reative to tota assets than do domestic banks and that state banks are on average the east profitabe banks. 6. Resuts Revenue equation Tabe 3 reports the resuts of the estimations of the revenue equation for the sampe period as whoe, but which aows the estimated parameters to vary between the two sub-periods, 1995 to 1998 and 1999 to Our preiminary investigations of bank margins indicated that they vary significanty over time. This partitioning of 4 Whie the data set incudes observations from 1994, this year is dropped from the sampe period for 22

23 the sampe divides the sampe into two roughy equa haves in terms of the numbers of observations. The mode specification aows us to see how the nature of competition in transition banking and the effects associated with the origin and ownership of banks change over time. We report two versions of the revenue equation. In one specification, we aow the estimated coefficient on bank equity to be freey estimated. In the other, the coefficient on equity is restricted to be equa to the risk-free nomina interest rate (as a proxy measure of the inter-bank interest rate). The reason for imposing this restriction is that the freey estimated coefficient on bank equity tends to be significanty greater than its theoreticay predicted vaue of the inter-bank interest rate. This may refect the fact that unobservabe characteristics of banks that enabe them to earn higher revenues, such as the abiity to increase demand for bank services, are aso associated with higher bank equity. The estimations yied a number of interesting resuts regarding the effects of bank origin and ownership. The estimated coefficient on the vaue of oans and deposits is the average margin earned by state banks in the earier sub period. The estimated coefficients on the other expanatory variabes indicate how average margins change from this base vaue. In the earier sub-period and the estimation that does not restrict the coefficient on bank equity to be equa to the risk free rate, the average margin earned on oans pus deposits by state banks was 1.7 per cent above a set of constants that contro for country and time fixed effects. Other factors being equa, the average margin of newy estabished foreign banks is ower than that of state banks 1.8 per cent. The average margins of privatised banks, both domestic and foreign, are about one per cent above that of state-owned banks in the earier subperiod. In this sub-period the average margins of banks above a set of constants that the estimations because agged expanatory variabe vaues are incuded in the regressions as instrumenta variabes. 23

24 aow for country and time effects are therefore 1.7 per cent for newy estabished domestic banks, -0.2 per cent for newy estabished foreign banks, 2.8 per cent for privatised banks with majority domestic ownership, 2.7 per cent for privatised banks with majority foreign ownership and 1.7 per cent for state-owned banks. In the estimations in which the coefficient on bank equity is restricted, the same pattern in average margins is observed, athough the eve of the average margins is higher by about 1.5 percentage points. The average margins associated with bank origin and ownership in addition vary significanty over time. In the ater sub-period (that is, 1999 to 2001) and the estimations that does not restrict the coefficient on bank equity, the margins earned by private banks increases significanty. Moreover, the margins of newy estabished banks increase by more than those of privatised banks. As a resut, in the ater subperiod, there is significant convergence among the average margins earned by private banks above a set of constants that contro for country and time effects. These margins are 5.7 per cent for newy estabished domestic banks, 5.0 per cent for newy estabished foreign banks, 4.9 per cent for privatised banks with majority domestic ownership and 5.3 per cent for privatised banks with majority foreign ownership. The average margin for state-owned banks remains 1.7 per cent. In the restricted estimations, the same variation in average margins by bank ownership is observed, athough their eve is about 0.5 percentage points higher. Regarding the effects of competitive pressure as measured by the number of banks per popuation in a country and the share of foreign banks in tota bank assets, there is significant evidence of competitive pressures associated with these measures ony in the ater sub-period in the unrestricted estimations. In this sub-period, as the number of banks per miion of popuation increases by one, the margin on oans pus 24

25 deposits decines by 0.2 per cent. Simiary, if the share of foreign banks in tota bank assets increases by 10 per cent, the average margin earned on oans pus deposits decreases by about 0.9 per cent. In the earier sub-period, the estimated coefficient on the number of banks per miion in popuation is statisticay insignificant, whie on the share of foreign banks in tota bank assets is positivey signed and weaky significant. This estimation indicates that competitive pressure associated with the tota number of banks and the market share of foreign banks is stronger in the ater sub-period than in the earier one. A simiar patter is observed in the restricted estimation, athough in this specification there is a significant competitive effect associated with the number of banks per miion of popuation in both the earier and ater sub-periods. This finding of increasing competition over time in transition economies is consistent with the theoretica modes of Hainz (2003a, 2003b) to the extent that the quaity of institutions improved between the two sub-periods and the empirica findings of Yidirium and Phiippatos (2002b). The correation between the weighted average market share of oans and deposits and the average margin earned on them is positive and significant in the sampe period as a whoe in both the unrestricted and restricted estimations. In the unrestricted estimation, an increase in market share of 1 per cent is associated with a 0.1 percentage point increase in the margin earned on oans and deposits. This is consistent with the oans and deposits of banks being imperfect substitutes, a basic assumption of our mode of monopoistic competition in banking. In the unrestricted estimation, the average margin on non-oan financia assets in the earier sub-period is -2.7 per cent, whie that in the ater sub-period is 7.5 per cent. With the possibe exception of the earier sub-sampe, which incudes a period of macroeconomic instabiity in most of the countries incuded in the sampe, these are 25

26 pausibe estimate vaues. A simiar patter is observed in the estimation that restricts the coefficient on bank equity. In the unrestricted estimation, coefficient on bank equity for the earier subperiod is 33.1 per cent, whie that for the ater sub-period is 29.3 per cent. These estimated vaues of the inter-bank rate for the sampe as whoe and for the ater subperiod are somewhat high compared to the theoreticay predicted vaue. This may refect the fact that bank equity is positivey correated with other unobservabe factors that contribute positivey to the revenues of banks. 5 The estimated revenue equations aso incude the dummy variabe that indicates a bank reported according to internationa accounting standards. In the earier sub-period and the unrestricted estimation, the average margin on oans pus deposits earned by banks that report according to internationa standards is 1.6 per cent ower, other factors being equa. There is weak statistica evidence that the gap between oca and internationa accounting standards narrows in the ater sub-period. Cost equation Tabe 4 reports the estimation resuts for the trans-og cost equation for the sampe period as whoe, but which aows estimated parameters of the cost function to vary between the two sub-periods, 1995 to 1998 and 1999 to As with the revenue equation, this specification aows us to see how the costs of providing banking services changes over time. The fina form of the estimated trans-og cost function incudes the vaue of oans and deposits and the mutipe of the vaue oans and deposits. The vaue of oans 5 In the restricted estimation of the revenue function, a Hausman test marginay rejects the EC2SLS with bank random effects in favour of a bank fixed-effects specification (not reported). This suggests that there may be omitted variabes in the restricted estimation. Our interpretation of the estimation resuts therefore paces greater emphasis on the unrestricted rather than restricted estimations. 26

27 is interacted with bank origin and ownership in the second sub-period, with the mutipe of oans and deposits is interacted with the bank origin and ownership variabes in both sub-periods. The expanatory variabes aso incude the vaue nonoan financia assets, the two input prices, US doar wages and the rea ex post interest rate at the country eve, and the ratio of bank equity to tota assets. The estimated parameter vaues are in most cases correcty signed and statisticay significant. An initia specification of the trans-og cost function incuded a of the higher order terms and those that aowed for interactions with bank origin and ownership and with the two sub-periods, but most were eiminated because they were statisticay insignificant. The estimation provides a number of interesting resuts regarding the association between bank origin and ownership and the costs of providing oans pus deposits. Compared with state-owned banks, most types of private banks appear to derive operating cost efficiencies from producing jointy oans and deposits at a ower cost that state-owned banks. This effect can be seen from the fact that the coefficient on the mutipe of oans and deposits is significanty negative for the three types of private banks, newy estabished banks and privatised banks with majority domestic ownership. The effect is argest for newy estabished foreign banks, but it decines somewhat in the ater sub-period. Perhaps surprisingy, privatised banks with majority foreign ownership do not appear to expoit this potentia cost compementarity as do other private banks. For the purpose of comparison with the average margins on oans and deposits estimated from the bank revenue equation, the margina costs can be cacuated from the estimated coefficients from the cost equation and the average costs and average oans and deposits for the different types of banks. For earier sub-period, the 27

28 margina cost evauated at the sampe mean of a newy estabished foreign bank in providing a unit of oans pus deposits is 1.5 per cent and that of a newy estabished domestic bank 5.8 per cent. The margina costs evauated at the sampe means for privatised banks are somewhat higher, 6.3 per cent for privatised banks with majority foreign ownership and 7.6 per cent for privatised banks with majority domestic ownership. The margina cost evauated at the sampe means of state-owned banks is 9.2 per cent. Our mode of monopoistic competition indicates that the correation between the estimated margina costs of providing oans and deposits and the average margin earned on oans pus deposits shoud be positive. 6 In the earier sub-period, the correation coefficient between these two vaues by bank origin and ownership is Excuding state-owned banks, the correation coefficient is This significant correation suggests that the origin and ownership of private banks is associated with abiity the abiity of banks to contro costs and to set more competitive margins. In the ater sub-period, the differentiation among banks in terms of their margina costs evauated at the sampe means diminishes significanty. The margina cost of newy estabished foreign banks whie significanty higher (3.0 per cent) remains beow that of other banks. This may refect that the fact that newy estabished foreign banks serve in arge part foreign firms operating in the host country but that are headquartered in their home countries and are estabished cients of the parent banks. The margina costs of other types of banks range from 5.1 per cent fore newy estabished banks to 7.2 per cent for privatised banks with majority foreign ownership. This finding is consistent with the entry of new banks and bank privatisation serving to reduce margina costs from their eve in state owned banks in 6 A direct comparison between the margins earned on oans pus deposits and their margina cost is not 28

29 the earier sub-period and with competitive pressures promoting simiar costs structures in among banks over time. In the ater sub-period the correation coefficient between the average margins and margina costs of banks grouped by their origin and ownership is essentiay zero. This absence of a correation suggests that as the costs structures among banks become more simiar other factors that infuence bank margins have a more significant effect on margins, such as the abiity to attract customers through better quaity services. The estimated cost equations can aso provide evidence of the easticity of operating costs with respect to input prices. However, the estimated coefficients on the input costs, which are time-varying country-eve variabes, are statisticay insignificant. One reason for this ack of significance may be that the country and time fixed-effects variabes are capturing the variation in input prices. In estimations that excude the country and time fixed effects but that aow for bank-specific fixed effects (not reported), the estimated coefficients on the input costs are correcty signed and statisticay significant. The cost equation aso incudes the vaue of non-oan financia assets and the ratio of equity to tota assets. Like the vaue of oans and deposits, the vaue of nonoan financia assets is positivey and significanty correated with operating costs. The estimated easticity of operating costs with respect to non-oan financia is The ratio of equity to tota assets is incuded as a proxy for non-observabe characteristics of banks that may be associated with their abiity to contro costs. The estimated coefficient on the ratio of equity to tota assets in the cost equation in the earier sub-period indicates that the operating costs of banks are negativey correated possibe because the revenue equation is estimated with fixed effects for countries and time. Therefore the constant term in the estimated revenue equation is non-zero, in contrast to the theoretica mode. 29