Report No. 40076-BR Brazil The Industry Structure of Banking Services June 26, 2007 Brazil Country Management Unit Poverty Reduction and Economic Management Unit Latin America and the Caribbean Region Document of the World Bank Table of Contents Executive Summary................................................................................................................ 1 2. 1. Introduction.................................................................................................................... 4 7 3. The BrazilianBanking System: StylizedFacts.............................................................. Assessingthe Degreeof Competition by Market Segment......................................... 3.1 Evolution of Loan Ratesby Market Segment andIndividualProduct ..................14 3.2 .............14 17 4. Estimation of Performanceby Market Segment.......................................................... Assessment of Price Competition inCorporate vs.Retail Loan Products 23 4.1 Overview of the Approach ..................................................................................... 23 4.2 Main Findings ........................................................................................................ 26 4.3 32 5. Drivers of Performanceby Market Segment ............................................................... Discussion of Findings........................................................................................... 34 5.1 RetailBanking ....................................................................................................... 34 5.2 Corporate Banking ................................................................................................. 36 6. Conclusions andPolicy Implications........................................................................... 39 Appendix I: BrazilianBanking Systemfrom an International Perspective- The Additional Comparisons ....................................................................................................... 42 Appendix II: Description of the Methodology to Estimate Performanceby Market Segment.............................................................................................................................. 44 Appendix III: Sensitivity Analysis inthe PerformanceEstimation...................................... 51 References .......................................................................................................................... 56 Listof Figures Figure 2.1: Total Assets of the Commercial Banking System (2005. US$ billion)............8 Figure 2.2: Bank Assets to GDP (2005) ............................................................................. 8 Figure 2.3: Composition of Assets (% of total assets. 2005) .............................................. 9 Figure 2.4: Composition of Credit (% of total credit. 2005) .............................................. 9 Figure 2.5: Cost to Income Ratio (2005) 10 Figure 2.6: Operating Expensesto Total Assets Ratio (2004).......................................... .......................................................................... 10 Figure 2.7: NPL Ratio (% of Total Loans. 2005) ............................................................. 11 Figure 2.8: Bank Equity to Assets (2005)......................................................................... Figure 2.9: Bank Provisions to NPLs (9%. 2005) .............................................................. 11 Figure 2.10: Non-Performing Loans by Market Segment (2005)..................................... 11 11 Figure 2.11: PerformanceRatios of Top Ten BrazilianBanks......................................... 12 Figure 2.12: Composition of Credit by Market Segment (9%. 2005) ................................ 13 Figure 3.1: Corporate Loans: Average All-Inclusive Rates for Banking System (199544-200644) ............................................................................................................. 14 Figure 3.2: RetailLoans: Average All-Inclusive Rates for Banking System (199544-200644) ............................................................................................................. 15 Figure 3.3: Corporate LendingRates on Average Follow the SELIC More Closely than Retail Lending Rates (199544-200644) .......................................................................... 15 Figure 3.4: The SELIC versus the StandardDeviation of LoanRates (199544- 200644) ............................................................................................................................ 17 11 .. Figure 4.1: Structure o f Assets.......................................................................................... 27 Figure4.2: Structure of Liabilities.................................................................................... 27 Figure4.4: Net Interest Income for Corporate Business Line (BRLmillion) ..................28 Figure 4.3: Net Interest Income for Retail Business Line (BRL million) ........................ 28 Figure 4.5: Margin Decomposition for Retail (BRLmillion).......................................... 29 Figure4.6: MarginDecompositionfor Corporate (BRLmillion) .................................... Figure 4.7: Net IncomeDecomposition for Retail Business Line (BRLmillion) ............29 30 Figure 4.8: Net Income Decompositionfor Corporate Business Line (BRL million) ......30 Figure 4.9: EstimatedReturn on Capital for Different Business Lines (2005) ................31 42 Figure 1.2: Composition of Deposits (% total deposits. 2005) ......................................... Figure I.1: Composition of Liabilities (2005) ................................................................... 42 Figure 1.3: Market Share of Top 3 Banks ......................................................................... 43 Figure1.4: Market Share of Top 10Banks....................................................................... Figure 1.5: Herfindahl Index o f Market Concentration..................................................... 43 Figure 1.6: Ownership of Top Ten Banks......................................................................... 43 43 Figure 11.1:Decompositionof Interest Marginbetween Lending andBorrowing Activities ........................................................................................................................... 50 Listof Tables Table 3.1: Description of MainLoan Product Categories ................................................ 16 Table 3.2: Regression Results........................................................................................... Table 3.3: Coefficient Values and Tests o f Hypotheses................................................... 19 20 Table 3.4: Variable Definitions......................................................................................... 20 Line ................................................................................................................................... Table 4.1: Banking System Balance Sheet (December 2005) -Allocation by Business Table 5.1: Profitability and Drivers for different Market Segments................................. 25 38 Table 11.1:Brazilian BankingSystem - ConsolidatedFinancial Statements (2004- 44 Table I11.la: Sensitivity Analysis inthe Performance Estimation.................................... 2005) ................................................................................................................................. 54 Table III.lb: Sensitivity Analysis inthe Performance Estimation (cont.) ........................ 55 iii Executive Summary Understanding the industry structure of banking services in Brazil is an important task both for the financial community at large and for country specialists. The Brazilian financial system i s the largest and most sophisticated in Latin America. Brazilian banks successfully navigated through several episodes of hyperinflation in the 1980s by creating extensive networks to reap the gains of float, while they have also adapted to the more stable environment o f the last decade. However, the cost of financial intermediation, in terms o f both absolute interest rates and spreads, remains among the highest in the world. In addition, there exist large differences in rates between corporate and retail lendingproducts, driving a wedge in access to credit between different groups of firms and individuals. Identifying and analyzing the various contributory factors for this phenomenon, particularly the role played by the industry structure of banking services, i s therefore an important exercise. Unlike many studies, this one gets to the market level in order to derive sensible policy implications. While other studies have attempted to assess the degree of competition, the causes o f spreads, business trends, and other factors in the Brazilian banking system at an aggregate level, they have not analyzed individual markets. Macroeconomic factors seem to dominate the level o f spreads at high interest rate levels, but as rates decrease microeconomic factors become more relevant. In order to provide relevant policy recommendations, one needs to study individual market segments since the banking system i s only an umbrella covering markets with very different characteristics, financial performance, and drivers. The study focuses in particular on the two most important banking market segments: Corporate and Retail. Each of the two segments i s defined at a high level due to data availability constraints, but they align fairly closely with actual bank practice since banks typically have separate business lines catering to the needs (loans, deposits, payments) o f these distinct types o f clients. In order to pursue this task, the study combines an industrial organization perspective with a practitioner approach. Firstly, stylized facts about the Brazilian banking system are identified and compared with a selected peer group o f developed and developing countries. Secondly, the degree of price competition i s assessed via an econometric model that estimates the sensitivity o f bank revenues by market segment to an increase in the main cost drivers. Thirdly, the performance o f different market segments i s estimated by disaggregating all the revenues, costs, and risks of the banking system into different lines o f business in order to obtain key profitability ratios such as the return on capital and assets. While high profitability is insufficient per se to infer the level of competition by market segment, it i s typically positively correlated to market power, and it represents one o f the most relevant indicators used by banking practitioners and analysts. Finally, a review of the existing financial sector literature and interviews with senior management o f eight Brazilian banks o f different size and ownership structures, serve as a consistency check on the study's findings and are used to identify the maindrivers ineach market segment. 1 From a benchmarking perspective, this study concludes that the Brazilian banking system is fairly similar to the peer group, except for the relative scarcity of credit and higher operational expenses and non-performing loans. In spite of its large size, Brazil i s similar to other countries in terms o f the degree o f concentration (Herfindahl Index of 900), as well as the efficiency ratio (Cost-to-income of 62 percent) and capital adequacy (Capital ratio of 15 percent). However, credit i s relatively low compared with other countries (34 percent o f assets) and a significant portion of bank assets i s invested in government bonds (43 percent of assets). The moderate efficiency ratio is actually hiding high operational expenses that are compensated with high revenues, seemingly revealing room for efficiency gains. Non-performing loans have been declining in recent years, but are still at the higher end o f the sample o f countries at 4.1 percent, partly because o f high differences between the Retail and Corporate segments (8.2 and 2.7 percent respectively). The largest Brazilian commercial banks exhibit significant differences in financial performance that are largely explained by their different segment focus, thus calling for a deeper understanding o f the drivers and profitability in each market segment. The findings of the econometric model corroborate the hypothesis that retail products are significantly less prone to movements in the cost of funds than corporate products. Based on the Panzar and Rosse (1987) model, the sensitivity of bank revenues by market segment to an increase in the cost drivers was estimated. This method i s based on the assumption that, under perfect competition, an increase in marginal costs would be reflected one-to-one in higher prices. While the results need to be interpreted with some caution given the relatively short time series available and other data constraints, they show that retail loan products (especially Cheque Especial and Credit0 Pessoal) are significantly less prone to movements in the cost of funds than corporate products, implying that Corporate has a higher degree of price competition than Retail. Inparticular, the results show that on average 40 percent of any decreasehcrease in the cost of funds is reflected in lowerhigher rates charged to customers for corporate loan products, whereas this pass-through declines to 20 percent inthe case o f retail loans. A key finding of the financial analysis is that the retail business line exhibits significantly higher returns than corporate, in spite of being costlier. Retail i s the largest business line in the banking system (around 40 percent o f total assets) with a different assetniability structure from Corporate. Although Retail i s more costly than Corporate both in terms o f operational expenses (it absorbs 80 percent of operational costs) and credit expenses (it accounts for almost two-thirds o f total loan loss provisions), a combination o f higher lending rates and fees (it generates 70 percent of the total interest margin)more than compensates these additional expenses. As a result, the profitability of Retail i s significantly higher, with a risk-adjusted return on capital reaching 38 percent compared to 21 percent in Corporate. This finding i s in line with that of the econometric approach and appears robust to sensitivity scenarios. While not unusual in the international context, Brazil appears exceptional regarding the extent of these differences inperformance. 2 A host of factors that influence revenues, costs, and risks explains the distinct characteristics and financial performance of different markets. These drivers can be common across business lines (reserve requirements, taxation, directed lending), but they can also differ substantially by - and even within a - business line. They include, among others, inherent client and transaction features, regulation (including the type and degree of government involvement in the financial system), and the state o f the financial infrastructure. All these drivers are relevant in determining the degree o f effective competition and financial access in particular market segments and, by implication, market structure and relative profitability. In Retail, the ability to keep clients in the network, the efficiency of managing large volumes o f standardized transactions, the ability of capturing information on borrowers, and the level of standardization o f risk management are key levers. In Corporate, the ability to manage the overall relationship with individualized customers (who are more mobile and have more choices) in order to maximize cross-selling opportunities, the effectiveness of managing low volumes of complex transactions, and the reliance on credit infrastructure (reliability o f corporate financial information, creditor rights, etc.) are some o f the most important factors. Both system-wide and market segment-specific policy implications can be drawn from the analysis. The study illustrates how differences across market segments - which tend to be averaged out in aggregate analysis - need to be taken into account when designing public policy. At the banking system level, there i s a need to strengthen the oversight framework and institutional capacity to promote competition in the financial sector. With regards to market segment-specific issues, government policies that encourage price competition among banks - especially in the retail segment - should address those factors that drive up revenues, costs, and risks, and hence adversely impact efforts to ensure wide and affordable access to banking services across Brazil. Examples o f such policies, which are elaborated in the study, include: further promoting the portability of bank accounts (e.g., by separating the corporate from the retail relationship o f companies and their employees); permitting positive credit information sharing; expanding payment system interconnection and improving the retail payments mix; developing the role of capital markets in the provision o f long-term finance; improving corporate financial reporting; strengthening legal rights and judicial procedures for contract enforcement; and carefully reassessing the nature and extent of government interventions over the medium to longer term in the Brazilian banking system (Le., directed lending schemes, reserve requirements, special regulated deposits, and the tax structure). Going forward, some of these policies will become even more relevant as the different markets continue to evolve. Business perspectives of the Retail and Corporate markets are quite different. In Retail, credit to employees of large companies and civil servants seems to be progressively reaching maturity, leading to a next wave o f credit expansion via independent workers and SMEs for which the credit infrastructure (positive information sharing, creditor rights) i s very relevant. In the corporate banking market, fee-based services to finance corporations through capital markets will likely gain share with respect to traditional lending products, stressing the importance of capital markets policy initiatives. 3 1. Introduction 1. Understandingthe causes of highinterestspreads and low access to credit in Brazil is an important task both for the financial community at large and for countryspecialists.The Brazilian financial system is by far the largest and arguably the most sophisticated in Latin America'. Brazilian banks successfully navigated through several episodes o f hyperinflation inthe 1980sby creating sophisticated networks to reap the gains of float, while they have also adapted to the more stable environment of the last decade. However, the cost of financial intermediation, in terms of both absolute interest rates and spreads, remains among the highest in the world, and credit i s still scarce for a significant proportion o f the economy2. While spreads have recently declined, they have not yet achieved a reduction to sufficiently moderate levels by international standards. Identifying and analyzing the various contributory factors for this phenomenon, including the role played by the industry structure o f banking services, i s therefore an important exercise for policy purposes. 2. An important feature of the high average lending rate is that it conceals importantdifferences betweenthe spreads chargedto differenttypes of clients.The Central Bank of Brazil (Bacen) has released since 1999 a series o f papers under the research program on "Juros e Spread Bancario", which quantify the various determinants o f the spread through an accounting decomposition. According to this program, the main factors behind high interest spreads appear to be operational costs, the costs imposed by the government through explicit and implicit taxation (including unremunerated reserve requirements), and high returns. All three factors may be influenced by, as well as contribute to, the industry structure and degree o f competition in different market segments. Inparticular, there exist large differences in rates between corporate and retail lending products, driving a wedge in access to credit between corporations and individuals or small firms. The extent to which those differences could be attributed to segmentation and market power indifferent segments i s therefore an important question. 3. While other studieshaveattemptedto assess the degreeof competitioninthe Brazilian banking sector and link it to high spreads, they have not analyzed individual market segments. For example, both Nakane (2001) and Belaisch (2003) perform such an analysis at the aggregate banking level and conclude that the Brazilian banking industry neither behaves as a cartel nor i s it perfectly competitive, a finding that 'The Brazilian financial system is relatively more sophisticated than other peer countries. Aside from the data presentedin Section 2 and Appendix Iof this study that serves to corroborate this statement, Brazil fares better than most Latin American countries interms of stock market capitalization (market cap to GDP for Brazil in 2005 stood at 51% against 34% inArgentina, 29% in Colombia and 27% in Mexico), standard measures of technology penetration (e.g., the number of ATMs per 100,000people for Brazil in 2004 was 18, while in Argentina, Colombia and Mexico this figure was 15, 10, 17), and diversity of financial products, three further proxies of financial sophistication. Empiricalresults inthe World Bank's Investment Climate Assessment for Brazil (December 2005) suggest that firms of all sizes-particularly smaller ones- are credit constrained, with the cost of credit being the most commonly cited reason for not applying for a loan. 4 i s not particularly useful for policy recommendations3.The studies do not take account o f the aforementioned anecdotal evidence of strong competition among banks in the market for loans to large corporations4, in contrast to the market for loans to consumers and small firms5. Understanding the characteristics and drivers o f performance in different market segments i s critical for the formulation o f public policy to make access to credit more affordable, but there has not been - to our knowledge - a study on this issue in Brazil. 4. The objective of this study is to analyze the industry structure of banking servicesinBrazil, assessing competition and performance at a disaggregatedmarket level. In particular, the study focuses on the characteristics, financial performance and drivers - including the degree o f price competition - o f the two most important banking market segments: Corporate and Retail. Each o f the two segments i s defined at a high level due to data availability constraints, but they align fairly closely with actual bank practice since banks typically have separate business lines catering to the needs (loans, deposits, payments) o f these different markets. 5. In order to pursue this task, the study combines an industrial organization perspective with a practitioner approach. Firstly, stylized facts about the Brazilian banking system are identified and compared with a selected peer group of developed and developing countries. Secondly, the degree o f price competition i s assessed via an econometric model that estimates the sensitivity o f bank revenues by market segment to an increase in the main cost drivers (cost o f funds, labor and operational expenses)6. Thirdly, the performance of different market segments is estimated directly by disaggregating all the revenues, costs, and risks o f the banking system into different lines o f business in order to obtain key profitability ratios7. While high profitability i s insufficient per se to infer the level of competition by market segment, it i s typically positively correlated to market power, and it represents one of the most relevant indicators used by banking industry practitioners and analysts. It i s worth noting that, while methodologically different, the direct approach can be seen as complementary to the indirect or econometric approach (see chapter 4 for more details). Finally, a review o f the existing financial sector literature and interviews with senior management of eight Belaisch (2003) also distinguishes between state-owned, smalllmedium size and foreign-owned banks, and finds that the first two types behave oligopolistically while only foreign banks behave competitively. Large corporate clients can access short-term trade finance at a small premium over LIBOR, medium- term loans for working capital and business expansion at relatively narrow margins over the SELIC rate, and long-term funds from BNDES (directed lending) at even lower rates than the SELIC. Individuals face short maturities and interest rates from around 25-30 percent p.a. (per annum) on car loans, to 60-70 percent p.a. on personal loans and up to 150percent p.a. on overdrafts. This method, introducedby Panzar and Rosse (1987), is very popular inthe empirical industrial organization literature and i s basedon the assumption that, under perfect competition, an increase in marginal costs would be reflected one-to-one in higher prices and therefore revenues. The further actual market dynamics are from this ideal competitive benchmark, the lower i s the elasticity coefficient. The approach results inthe creation of stand-alone notional financial statementsand related financial ratios (e.g. pre-tax return on capital and assets) by business line as of December 2005, which can then be compared to each other and to the ones for the entire banking system. 5 Brazilian banks of different size and ownership structures, serve as a consistency check on the study's findings and are used to identify the main drivers ineach market segment'. 6. The study is in line with the World Bank's program in Brazil and complements well on-going financial sector work. In particular, one of the pillars of the World Bank's Country Assistance Strategy i s a more competitive Brazil to spur sustainable growth; this study, by tackling the issue of competition in the financial system, comes under this pillar. In addition, the study i s a natural follow-up to the recently-completed report entitled "Brazil: Interest Rates and Intermediation Spreads" by de la Torre and Claessens (World Bank, June 2006), in which the industry structure and level of competition in specific market segments are recognized as potentially important -butstillunexplored-issuesinthecurrent debateoninterestratesandspreads.Where relevant, the study also draws upon other World Bank financial sector analytic work in Brazil, as well as upon issues highlighted under the financial sector component of the two recent Programmatic Loans for Sustainable and Equitable Growth. Finally, the timing of the study i s appropriate, both because these types of microeconomic factors will become increasingly important as market rates approach single digitsg,and because the Brazilian authorities have recently announced (and are considering additional) reforms in the regulatory framework for the banking industryin order to promote competition. 7. The study is organized into five additional sections in line with the main analytical building blocks, complemented by appendices that descri,be methodological aspects in more detail. Section 2 (and Appendix I) presents some stylized facts on the Brazilian banking system - including dissagregations by market segment - and compares it to a peer group o f countries in order to identify those areas where Brazil remains an outlier. Section 3 describes and presents the results o f the industrial organization perspective (indirect approach). Section 4 (as well as Appendices I1and 111)presents the methodology and results - including sensitivity analysis - of the direct approach. Section 5 identifies and discusses the characteristics and main drivers o f revenues, costs and risks by market segment, while Section 6 concludes and presents policy implications. * Interviews were conducted both with senior bank management to get an overall perspective of the system, as well as with business line managers to obtain more detailed information. The interviews covered 2 gublic banks, 2 large domestic private banks, 1smaller domestic private bank, and 3 foreign banks. The World Bank (June 2006) report states that "the `largestbangfor the reform' buck in terms of reducing intermediation spreadswould come, at this stage,from sustainable reductions in the SELIC rate...as the SELICdeclines, these types of factors [implicit and explicit taxation, administrative and operational costs, competition, contract enforcement institutions] will increasingly become binding constraints tofurther reduction of intermediation spreads.Hence, reforms to tackle these microfactors should remain an important priority in the government's reform agenda". 6 2. The BrazilianBankingSystem: StylizedFacts 8. The analysis of stylized facts of the Brazilian banking system comprises an international comparison with a peer group of countries, as well as an in-depth assessment of selected indicators across different market segments. The largest and more developed financial systems in Latin America besides Brazil - i.e. Mexico, Argentina, Chile and Colombia - are selected for comparison purposes. In addition, the United States (largest financial system globally) and Spain (given that Brazil's financial system structure i s closer to the European universal banking model) are also included in the group of countries as international benchmarks. 9. Brazil has by far the largest banking system in Latin America (Figures 2.1 and 2.2)". System assets as of end-2005 accounted for almost US$611 billion (US$684 billion when the development bank BNDES and other consolidated level IIbanking institutions are included), which i s almost three times the size o f the second largest regional system (Mexico), and around one-half o f the total Latin American banking system assets. Relative to GDP, bank assets are also highat 74 percent of GDP, and only lower than Spain" (205 percent) and Chile (95 percent). Financial assets are complemented by a wide network o f more than 20,000 branches and 50,000 ATMs that provide a sound infrastructure and technological base to expand market access across dispersed geographical regions. There i s also an important network of correspondent bankingagencies, including lottery houses and retail stores12. 10. In spite of its large size, the Brazilian banking system presents many characteristics similar to those of other Latin American countries. The degree of c~ncentration'~ i s in line with most markets, and i s actually at the lower end o f the sample. The cost-to-income ratio - a key efficiency measure used by the industry - i s very similar to that of other countries in the sample, while the capital adequacy ratio i s also in line with its peers (see Appendix Ifor complementary benchmarking data against peer countries). loThe financial data used for the Brazilian banking system, taken from Bacen, i s at the so-called consolidated level I, which includes both independent commercial banking institutions, and banking conglomerates containing at least one commercial or universal bank. The choice of consolidated level I, which excludes BNDES, was made in order to make the ratios comparable to those of other countries. The figures for Spain exclude international operations of domestic banks; including those as well would raise the bank assets to GDP ratio for Spain above 300 percent. 12See Kumar,Nair, Parsons and Urdapilleta (2005) for more details. l3The Herfindahl index of marketconcentration shows that Brazil, with an index of 906, i s actually less concentrated than Colombia (1,642, once common ownership patterns for some banks are taken into account), Mexico (1,530) and Spain (1,394). 7 Figure2.1: Total Assets of the CommercialBanking Figure 2.2: Bank Assets to GDP (2005)14 System(2005, US$ billion) i 205% Brazil Mexico Chile Argentina Colombia USA 62% Source: Study calculations based on Banco Central do Brasil, Source: Study calculations based on Federal Comisidn Bancaria y de Valores Mhico, Banco Central de Chile, Reserve, Banco de Espaiia, ComisidnBancaria y de Banco Central de la Repliblica Argentina, SuperintendenciaBancaria Valores de MPxico, Superintendencia Bancaria de de Colombia, Banco Central de Venezuela, Superintendencia de Colombia, Banco Central de Chile, Banco Central Bancos de Panam'. do Brasil, Banco Central de la Repliblica Argentina, World Bank Financial Sector DevelopmentIndicators (FSDI). 11. However, there also exist a few salient factors that are particular to the Brazilian system and that may contribute to spreads being higher than the norm. First, bank lending is low compared with other countries. Credit, at 34 percent of total assets, i s relatively low vis-&vis other regional and global peers, while securities absorb almost 43 percent of total assets; a significant portion of banking activities i s dedicated to financing the government through, for example, investments in government securities (Figure 2.3). It i s important to note, however, that a decomposition o f credit by market segment suggests relatively higher participation o f retail credit (46 percent) intotal credit compared to other countries (Figure 2.4). Moreover, although system deposits represent a lower percentage than in other countries - Brazil stood at 49 percent of the system's liabilities compared to 71 percent in Argentina, 76 percent in Colombia and 83 percent in Spain - a large part of this difference stems from the existence of other funding sources related to directed lending schemes that are not considered intotal deposits. 14Banking assets corresponds to the relevant definition o f commercial banks ineach country of the sample to make ratios comparable among them. Datai s for non-consolidated, tier-one assets of the banking system. United States (Domestically Chartered Commercial Banks); Spain (Domestic assets of Commercial Banks, Saving Banks and Credit Cooperatives), Mexico (Banca Mdltiple), Colombia (Sistema Financiero, excluding credit unions andpublic sector special institutions), Chile (Sistema Bancario), Brazil (Consolidated IFinancial Institutions, which excludes second-tier banks), and Argentina (Sistema Financiero). 8 Figure2.3: Compositionof Assets (5% of total Figure2.4: Composition of Credit (% of total assets, 2005) credit, 2005) 1 7 23% Argentina 3 8 0 0 48% Brazil 43% 23% Brazil Chile 13% Chile 67% Colombia 35% 14% Colombia { 29b Meico 4VQ 22% Mexco i i Spain Spain 56% A 0Yo 50% 100% 0% 20% 40% 60% 80% 100% Total Credit Bonds & Notes X. OtherAssets Retail Credit Wholesale Credit # Public Credit Thecomposition of creditfigures in Mexico includes FideicomisosUDIbut excludes IPAB and FOBAPROA. Thefigures for public sector creditfor Colombiacomefrom the SuperintendenciaFinanciera. Source: Study calculationsbased on Federal Reserve,Banco de Espaiza, Comisio'nBancana y de ValoresdeMkxico, SuperintendenciaBancana de Colombia, Banco Central de Chile,Banco Central do Brasil, and Banco Centralde la RepdblicaArgentina. 12. Brazil's moderate cost-to-income ratio is actually hiding high operational expenses that are compensated with high revenues, seemingly revealing room for efficiency gains15.In 2005, the cost-to-income ratio in Brazil stood at 59 percent, which i s similar with the levels observed in Latin America and in industrialized countries (Figure 2.5). Nonetheless, this conceals structural inefficiencies that are being masked by high operating income, which itself is partly related to the very high nominal interest rates o f the system. As a comparison, the ratio of operating expenses to total assets for Brazil in 2004 (6.5 percent) was the highest in the sample, while overhead costs to assets were almost 50 percent higher than in Argentina and four times higher than in Spain (Figure 2.6). This could arise from a number o f factors, including the legacy o f high operational costs from the 1980s, expenses related to government-imposed financial taxes, regulation and directed lending requirements, credit monitoring costs due to a lack of positive information sharing, or the lack of sufficient competition to promote efficiency. This evidence therefore suggests that there remains potentially important room for efficiency gains in the banking system, especially by tackling business environment factors that feed directly into banks' operational costs. Inthis context the term efSiciency is usedas per financial analysts' definition of high costs to operate. The origin of such inefficiencies may be internal to the institution or relatedto the businessenvironment. 9 Figure 2.5: Cost to Income Ratio (2005) Figure 2.6: OperatingExpenses to Total Assets Ratio (2004) 1 Argentina i Brazd Brail 6.5% Chile Chile Colombia 2% Colombia 5.9% MeAco Mexico I Spain 56% Spain 1.6% i USA Source: Study calculations based on Federal Reserve, Banco de Es~AAA,Comisidn BAnCArki y de Valores de Mixico, Superintendencia BAncAnA de Colombia, Banco Central de Chile, Banco Central do Brasil, Banco Central de Ea Repdblica Argentina, World Bank Financial Sector Development Indicators. 13. Non-performing loans (NPLs) have been declining in recent years, but are still at the higher end of the sample of countries. The Brazilian banking system's NPL ratio as o f end o f 2005 stood at 4.1 percent of total non-earmarked loans, well above most of its regional peers (Figure 2.7) 16. A closer look at NPLs across market segments (retail and corporate) and products shows large disparities in asset quality. For instance, the NPL ratio for retail non-earmarked loans was around 8 percent in 2005, while the respective ratio for corporate non-earmarked loans stood at below 3 percent (Figure 2.10). Despite this, the quality o f banks' credit portfolio has consistently improved over the past 10 years, and the system exhibits strong capitalization (Figure 2.8) and adequate loan loss provisioning (Figure 2.9). In fact, in 2005 the Brazilian system's capital adequacy ratio was 18.2 percent, well above the minimum regulatory capital ratios imposed by Base1 (8 percent) or Bacen (11percent), and higher than the levels reported by peer countries. l6The NPL ratio is calculated as the volume of non-performing loans to total loans for the entire system and it i s a widely used measure of asset quality. The figures for Brazil exclude earmarked lending that, if included, would further raise such NPL ratio to 6.6 percent. NPLs are higher in Brazil than in peer countries partly due to the higher proportion of retail loans (see Figure 2.4). However, this is not the only factor- it i s also the case that NPLsrelated to the corporate segment are higher than inother countries, resulting in overall higher NPLs for the system. 10 Figure 2.7: NPLRatio (% of Figure 2.8: BankEquity to Figure 2.9: Bank Provisions to Total Loans, 2005) Assets (2005) N p L s (%, 2005) 1 Argentina 6.6 Argentina P% Argentina I6 Brazil I 4 1 Brazil 9% Braal 153 Chile Chile 66 Colombia 3.2 Colombia Colombia Mexico M eico Mexico Spain P% Spain 3x3 USA USA Source: Financial Soundness Source: Study calculations based on Source: Financial Soundness Indicators (IMF) Bank Equity to Assets Ratio Indicators (IMF). calculated directly from the Federal Reserve,Banco de Espaiia, Comisidn Bancaria y de Valores de M&xico, Superintendencia Financiera de Colombia, Banco Central de Chile, Banco Central do Brasil, Banco Centralde la Republica Argentina Figure 2.10: Non-Performing Loans by MarketSegment (2005) Corporate 2.7?A Retail 1 (R$ 8.23% P7,498,628) NPLs are only for non-earmarked products andfor categories D and over (90 days past-due). The NPL ratio for the entire banking system (both earmarkedand non-earmarked) in 2005 amounted to 66percent. Source: Bacen. 14. The largest Brazilian commercial banks exhibit significant differences in performance that are largely explained by their different orientation and segment focus. The top ten commercial institutions in Brazil are categorized as follows: two institutions are public (Banco do Brad and CEF), five are domestic private (Bradesco, 11 Ita5 Unibanco, Safra, and Votorantim), and three are foreign (Santander,ABN AMRO, and HSBC). I t is interesting to note that all three foreign institutions grew by acquiring established domestic banks (Banespa, Bunco Real, and Bamerindus) rather than organically via a greenfield ~peration'~; salient feature of the banking system tends to this emphasize the importance o f being an incumbent. Furthermore, banks that focus on the retail segment have costs and NPLs that are higher than the rest, but they tend to perform better (Figures 2.11 and 2.12)18. While evidence at this level i s largely circumstantial, such differences in profitability, especially in the presence o f high overall interest rates, call for a deeper understanding o f the drivers and profitability ineach market segment. Figure2.11:PerformanceRatios of Top Ten Brazilian Banks Credit to Assets (2005) Cost to Income (2005) NPLs (2005) 1 BancodoBrasil (Pu) 36% Caixakonomica Federal (Pu) 77 8 22.0% BRADESCO(Pr) 44% ITAU(Pr) SANTANDERBANESPA(F) UNIBANCO(Pr) 40% 48% I SAFRA( Pr) HSBC(F) VOTORANTIM(Pr) ROE(2005) ROA (2005) I BancodoBraal (Pu) 26.8% CaixakonomicaFederal (Pu) 26.0% BRADESCO(Pr) 31.8% i ITAU(Pr) 5.3% SANTANDERBANESPA(F) 21.6% 4.3% UNIBANCO(Pr) 24.2% ABNAMRO(F) SAFRA(Pr) HSBC(F) VOTORANTIM(Pr) i (Pu}Public Bank; (Pr}Private Bank; (F}Foreign Bank Source: Banco Central do Brasil, Bankscope. 17 This i s also true for the largest localbanks which have grown by merging smaller institutions, albeit over a longer period of time. '*For instance, Ztuu', a leadingretail bank, exhibits a NPL rate that i s higher than average at 10.8 percent, but also has above-averageprofitability with an ROEof 35 percent and ROA of 2.9 percent. Conversely, Sufya, which has a mostly corporate focus, has a NPL rate of 5.9 percent but lower profitability, with an ROE of 17percent andROA of 1.6percent. 12 Figure 2.12: Composition of Credit by Market Segment (%, 2005) BB 60.79.0 2.9% CEF 1I BRADESCO .. 54.6% 3 1.Pb KAU 02% SANTANDERBANESPA 63.490 $ 1.00b UNIBANCO 1 03% ABN AMRO 0.5% SAFRA 88.61 0.390 VOTORANTlM f 13% Retail Wholesale E Governmnt Source: Bank annual reports and interviews. 13 3. Assessing the Degreeof Competitionby Market Segment 15. This chapter presents a measure of competition by estimating the sensitivity of bank revenues by market segment to an increase in the cost drivers. The method, introduced by Panzar and Rosse (1987), i s based on the assumption that, under perfect competition, an increase in marginal costs would be reflected one-to-one in higher prices and therefore revenues (total elasticity equal to one). The further actual market dynamics are from this ideal competitive benchmark, the lower i s the elasticity coefficient (zero in the extreme case o f monopoly). 3.1 Evolution of Loan Rates by Market Segment and IndividualProduct 16. The very high average lending rate in Brazil conceals large differences among the terms of credit offered to corporate versus retail customers. Therefore, it i s particularly instructive to take a closer look at the wide spectrum o f lending rates and their evolution over time". A first observation that can be drawn from looking at the time the main corporate loan types tend to follow movements in the SELICgo very closely, series plotted in Figures 3.1, 3.2 and 3.3 below i s that interest rates char ed on some of whereas the main retail loan products are priced well above their corporate counterparts and tend to appear largely unaffectedb y changes inthe relevant cost o f funds (see Table 3.1 for a description of each loan product). Figure 3.1: Corporate Loans: Average All-Inclusive Ratesfor BankingSystem (I99544-2006Q4) 1135% p t Mney re 115% 95% ProndssoriasRe 75% -+-Conta garantida 55% 35% 15% 5 Em Em 2m~m8 8 u K 2 5 v 2 zo 5o 2o $ m - b o N - - = t m N - b o m - - = t m 7 m m m G 3 m o o o o o o o o o o o o o o o o o - - - - - w m N N N m m N N m Source: Central Bank of Brazil (Bacen) 19 The analysis in this chapter includes only financial products (pre-fixed) within the context of non- earmarked loans. These products comprise around 40 percent of total loans and advances as of December 2006. Directed (ear-marked) loans are not considered due to lack of sufficient data and due to the fact that their rates are administratively determined and are therefore insensitive to changes inmarket rates. 2o The SELIC (Sistema Especial de Liquida@o e Custodia) i s Bacen's overnight baserate for rep0 operations. The overnight interbank rate (CDI) tends to track the SELIC rate very closely. 14 Figure 3.2: Retail Loans: Average All-Inclusive Ratesfor Banking System(1995Q4-2006Q4) I Figure3.3: CorporateLendingRateson Average Follow the SELIC MoreClosely than Retail Lending Rates (199544-200644) 20% 00% 80% 60% 40% 20% 0% 1&Taxa Selic --@-- Corporate+R e a Source: Central Bank of Brazil (Bacen) 15 Table 3.1:Description of MainLoan Product Categories Category Description Volume (BRLMillion)*' Hot Money Short-term loan with no collateral. Can be asked for at any time by companies 286 respecting its credit limit. N o need to define the use for the money. Duplicatas Discount of Receivables. The bank lends money equivalent to the amount of an 12,012 invoice that the company has the right to receive inthe future. The company sells a future cash flow to receive cash inadvance. Nota Discount of Promissory Notes. Similar to the previous product but the nota 160 promissdria promissoriadoes not require any contract. The bank buys aright to receive a specific payment and provides the advanced cash to the company. Capital de Working Capital. Credit line to meet the cash flow needs of a company, 23,620 Giro financing its operational cycle. Requires a collateral. Conta Overdraft Line of Credit. Can be used with no collateral at any time until the 17,171 Garantida credit limit is reached. Nowadays some banks have a version with a collateral. Aq. Bens PJ Credit line for companies for the purchase of goods. There i s always a collateral 10,581 that depending on the good can be a nota promissoria. Vendor Credit line to allow companies to finance their clients, allowing the seller to 8,477 receive cash immediately and sell with a longer payment due date for their clients. Collateral i s a must inthis product. Limitedto 180 days. The purchaser receive a boleto bancario for payment directly to the bank. Cheque Overdraft for individuals. 11,738 Especial Credit0 Very broad range of products for consumer financing. The usual product did not 78,895 Pessoal require any collateral but nowadays banks offer other options at lower rates considering collaterals like vehicles. Limited to 24 months. Aq Veiculos Consumer loans for financing purchase of vehicles. 63,472 PF Aq Outros Credit line for the purchase of goods other than vehicles. A collateral i s 10,763 BensPFZ2 required, usually the good itself but sometimes a notapromissoria. 17. As the level of the SELIC has progressively decreased over time, corporate loan rates appear to have adjusted downwards much faster than rates for retail loans (see Figure 3.3). This may help explain why the dispersion among loan rates (largely driven b y differential terms of credit offered to corporate versus retail customers) has increased over the last decade in Brazil. In general, a significant negative correlation can be observed over time (correlation coefficient = -0.60) between the level o f the SELIC and the standard deviation of loan rates (see Figure 3.4). "AsofDecember31,2006. 22Aq BensPFi s the sumof Aq Veiculos PF and Aq Outros BensPF, with volume of BRL74,235 million 16 Figure 3.4: The SELIC versus the StandardDeviation of LoanRates (1995Q4-2006Q4) 0.29 0.27 v) Q 0.255 5 0.23 c 0.21 0.199Q B 0.17 +Taxa Selic Cross-sectionaldisDersion of rates/ ~ Source: Central Bank of Brazil (Bucen) 3.2 Assessment of Price Competition in Corporate vs. RetailLoan Products 18. Recent studies have analyzed the competitive structure of banking services in Brazil, albeit mostly at the aggregate level. Nakane, Alencar and Kanczuk (2005) estimate two alternative oligopoly models (Bertrand and collusion) and try to determine which of them better fits the actual price-cost margins observed in the industry. The analysis i s disaggregated by 3 product markets (demand deposits, time deposits and loans) and 3,252 Brazilian municipalities. While some data i s available at the municipal level (e.g., ATM and branch density, GDP, and population), other variables and particularly product prices are only observed at the bank level for the whole country. The authors find that the Bertrand model overestimates the degree of market power for time deposits and loans. Bertrandcompetition seems to be a good description o f the way banks set service fees. Belaisch (2003) applies the Panzar and Rosse methodology to Brazil using bank-level data from Bankscope. The author finds positive evidence of the presence of a noncompetitive market structure in the Brazilian banking system, particularly in state-owned banks and small/medium-sized banks that are found to behave oligopolistically. Nakane (2001) applies a dynamic version o f the Bresnahan (1982) and Lau (1982) approach to assess the degree of competition in the Brazilian banking industry as a whole. Error-correction models of both the demand for bank loans and the bank interest rate spread are estimated on seven years o f aggregate monthly data. The author finds evidence that the banking industry in Brazil can be described as highly competitive, although not perfectly competitive; the hypothesis that Brazilian banks behave collusively i s strongly rejected. In summary, the main conclusions of previous studies are not particularly clear to guide policy making, since the typical finding i s that the Brazilian banking industryneither behaves as a cartel nor i s perfectly competitive. 17 19. The present study extends and complements the previous ones by analyzing the degree of competition in corporate and retail loan products following a version of the Panzar and Rosse (1987) methodology. Data limitations imply that the regression framework, as described in this chapter, i s somewhat different from the original setup - for example, the dependent variable in the original setup i s loan revenues and not loan rates. However, while we are not using a measure o f unit revenue as a dependent variable, we do control for changing loan amounts per product as a regressor. It is also worth noting that related literature - for example, Claessens and Laeven (2004) -usesthe ratioofgrossinterest revenueto totalassets as adependent variable toproxy for the output price o f loans. Given the unavailability o f individual product rates by banks, this study does not use bank-level variables as explanatory variables in the regression. 20. A panel regressionis estimatedusingquarterly data for the period 2003-2006 from Bacen to evaluate the sensitivity of individual product rates to movements in the main bank-wide cost factorsz3.The novelty of this regression, compared to the literature, i s that the cross-sectional dimension captures the different loan product rates for an average bank as opposed to an average rate (across all products) for different banks, as it has usually been done so far by, for example, Belaisch (2003) and Claessens and Laeven (2003). ln(1oan ratesit) = ai labor,) + yRetail + + ln(cost of fundst) + p2 ln(cost of equipment,) + p3 ln(cost of jq Retail x ln(cost of fundst) + Retail x ln(cost of equipmentt) + 3 Retail x ln(cost of labor,) >90 dayspast dueit) + + 6 ln(1oans outstanding as % of totali,) $1 ln(GDP growth,) + h ln(inflationt) + ~i~ + & In(% of loans i indexes individual product rates and t indexes time 21. Due to the unavailability of sufficient information on bank revenues and costs disaggregatedat the product level, publicly available product rates are used as a dependent variable and the three main cost componentsincluded as regressorsare measured for the whole banking system over time. While this could be improved if we obtained more disaggregated data on revenues and costs by market segments, we believe this approach may offer a useful comparison of competitive conditions across market segments to assess, in particular, which bank product rates are more/less affected by bank-wide cost pressures over time. To this purpose, we also include in the regression a dummy for retail products interacted with each of the three main cost components, in order to test whether competitive conditions (and therefore the elasticity o f product rates to input prices) are significantly different between the corporate and retail loan markets. We also include additional explanatory variables, such as performing and non-performing loans outstanding for each product category, to control for any size or risk characteristics 23 The estimation period for the loan rates by categories i s from 2003 to 2006 because of noisy NPLs data inthe early years (2000-2002) due to the restructuringof public banks (particularly Caixa) and the acquisition of Sgo Paulo State's bank Banespaby Santander. The unavailability of bank-level data on NPLs, or alternative measuresof risk, i s a constraint for the econometric analysis although the overall pattern seems to confirm the conclusions of the study that the retail sector i s significantly less competitive than the corporate segment. 18 o f individual market segments that may affect pricing. Fixed effects by product category are added to capture any other contractual features (such as differences in collateral, maturity, etc.), which can independently affect loan rates. Finally, we also add GDP growth and inflation since all rates considered are nominal and we would like to account for the influence of cyclical macroeconomic fluctuations. 22. Regression results are reported in the second column of Table 3.2. Meanwhile, Tables 3.3 and 3.4 below contain more details on the definitions of the variables. Table 3.2: RegressionResults Dependent variable: Loan rates Return on assets by product category (Panel regression with fixed (Time-series regression with effects) robust standard errors) Explanatory Variables: Cost of funds 0.39*** 0.00"" Cost of equipment -0.05 -0.01 Cost of labor 0.35** 0.01 Dummy for Retail products interacted with: cost of funds -0.18*** cost of equipment 0.03 cost of labor -0.14 Loans outstanding by product -0.12*** category as a % of total loans % loans over 90 days past due by 0.03** product category Total loans as a % of assets -0.02 Total deposits as a % of GDP 0.00 Total number of banks -0.08 GDP growth -0.40""" -0.01 Inflation 3.36*** 0.06 R-squared 0.76 0.54 Above results are tests of market power and long-term equilibrium. ***, ** and * indicatestatistical significance at the 1%, 5% and 10% confidence level respectively. All variables are in natural logs. Inthe case of GDP growth, inflation and ROA, we use the approximation In(l+variable), as these variables occasionally show negative values. Data i s quarterly for the period 2000-2006 and is obtained from the public online databaseof the Central Bank of Brazil. The estimation period for the loanrates by categories i s from 2003 to 2006, and from 2000 to 2006 for the long-term condition. 19 Table 3.3: Coefficient Values and Tests of Hypotheses Valueof Coefficients:Retail Segment Cost of funds 0.21 *** Cost of fixed capital -0.02 Cost of labor 0.21 ** H-Statistics(Retail) 0.40 *** Valueof Coefficients:CorporateSegment Cost of funds 0.39 *** Cost of fixed capital -0.05 Cost of labor 0.34 *** H-Statistics(Corporate) 0.69 *** Test of Hypotheses p-value H-statistic (Corporate) = H-statistic (Retail) 0.00 Corporate 0.00 H-Statistic = 0 0.00 H-Statistic = 1 Retail H-Statistic = 0 0.00 H-Statistic = 1 0.00 ttt,**, Significantat the 1,5,and 10percent levelrespectively. f Table 3.4: Variable Definitions I Loan rates Annualized quarterly loan rates for each loan category. Rates by product category include financial taxation, commissions, and fees. Rates are publicly available for each loan category only at the banking-system level, averaging across banks using as weights the flow of new loans originated by eachbank on each day. Return on assets Quarterly ratio of pre-tax income over total assets for the aggregate banking system excluding BNDESand credit cooperatives (Consolidated I) Cost of funds Unit cost of borrowed funds (total cost of deposits, acceptancesand repos divided by their stocks outstanding) ICost of labor I Unitcost of labor (total payroll expensesas a ratio of total assets) Cost of equipment Unitcost of equipmenvfixed capital (total administrative and other operational expensesas a ratio of fixed assets) 20 23. Results tend to confirm our early hypotheses. Three main outcomes from the econometric analysis are worth highlighting: Monopolistic competition dominates the corporate market segment. In fact, the sum of (PI + p 2 + p3), Le. the H-statistic (degree o f price competition), equals 0.69, which implies that the average degree o f price competition for the corporate market segment lies between 0 and 1. The hypotheses that the H-statistic for the corporate segment i s zero or one are rejected at the 1 percent level of ~ignificance~~.The costs o f funds and labor appear to be the most important drivers of movements in product rates, whereas changes in the cost of fixed capital/equipment do not seem to have a significant impact. Retail loan products are significantly less sensitive to movements in the cost of funds compared to corporate products25.Regression results in Table 3.2 and 3.3 suggest that, on average for the corporate market segment, 39 percent o f any increase/decrease in the cost o f funds i s reflected in higher/lower rates charged to bank customers, whereas this pass-through coefficient declines to only 21 percent (=0.39-0.18) in the case o f retail. This seems to confirm our hypothesis of weaker price competition in retail banking markets. Moreover, the degree of price competition for retail (0.40) i s lower than the H-statistic for the corporate market (0.69). Both statistics suggest that these markets are further away from the perfect competition paradigm (H-statistic=l). Among the additional control variables included in the regression, the significant negative coefficient on the level of loans outstanding by product category (as a percentage of total loans) could be read as an indication of prevailing demand forces (growing/decreasing demand for relatively moreAess affordable products).26 The coefficient on the stock of non- perfonning loans by market segment has the right sign and i s statistically significant, which show that asset risk has an effect on interest rates levels. 24. The Panzar and Rosse (1987) methodologyalso requiresthe assumptionthat banking markets are in long-term equilibrium, which implies in particular that marginalandaveragecosts are equal.Only inthis case an H-statistic=l corresponds to the perfect competition paradigm, while a value between 0 and 1 points to a monopolistic competition environment as obtained inthe analysis above. We proceed to test the market equilibrium assumption following an approach commonly used in the economic literature, for example by Shaffer (1982), Molyneux et al. (1996) and by Claessens and 24 With respect to the retail segment, the value of the H-statistic i s 0.40. As in the case of corporate, the hypotheses that the H-statistic i s zero or one are rejected at the 1percent level of significance (see Table 3.3 for details). As mentioned inpara. 27, due to the lack o f data on revenues for financial products, interest rates by products were included as a dependent variable. In this context, care i s needed with the interpretationof results. 25 Results presented here are robust to replacing the average cost of funds variable with the SELIC rate. 26 It i simportant to note that this is only one plausible explanation of the result. Still, empirical results of a related study by the World Bank (2006) suggest that an increase in the general level of interest rates reduces firms' overall demand for credit (more reliance on internal sources of funds) and i s associated with a higher supply by banks of those loan products with safer collateral cover (e.g. discount receivables, vendor) to mitigate adverse selection problems. 21 Laeven (2004). In particular, we estimate the following regression for the period 2000- 2006: ln(ROAt) = a + /l~ln(cost of funds,) + /l3 ln(cost of labor,) + 4 ln(tota1 loans /l2 ln(cost of equipmentt) as a % of assets,) number of banks,) + $1ln(GDP growth,) ++&+ln(tota1 deposits as a % of GDP,) & ln(inflationt) + et + b; ln(tota1 where ROA i s the pre-tax return on assets and the three main cost components are defined as already described in Table 3.4. Furthermore, the evolution o f the deposits-to- GDPand loan-to-asset ratios, as well as of the number of banks inthe system, are used to control for changes inthe depth and structure o f the banking sector over time. Finally, we include GDP growth and inflation as macroeconomic control variables as well as a dummy to isolate an outlier negative observation of ROA in mid-2001, mainly due to large losses reported by CEF (NPLrestructuring) and Santander (Banespa acquisition). 25. Without the necessary information to construct a ROA measure by product category, the regression is preliminarily run on time series data for the whole banking system. Results are reported in the third column of Table 3.2 and should be taken with some caution given data constraints. While the increase in ROA over time appears related to the consolidation process in the banking system (as proxied by the declining number o f banks), the degree o f price competition (PI + p 2 + p3) is not inputprices), which is consistent with the long-term equilibrium assumption.l significantly different from zero (Le. the ROA remains largely unaffected b 7 changes in 26. In sum, the findings of the indirect approach tend to corroborate the hypothesis that the degree of price competition is significantly higher in corporate than in retail banking markets in Brazil. Preliminary evidence at the individual product level indicates that Cheque Especial and Credit0 Pessoal are among the retail products where the lack of price competition seems most evident. 27. While the overall message regarding competition in different market segments arising from the analysis seems clear, more data and additional sensitivity tests would be needed to strengthen the econometric results. Unit root tests indicate an overall stationarity of the data. However, they need to be interpreted with some caution given the relatively short time series available. In order to keep consistency with the previous regression, we have used balance sheet data for over 100 Brazilian banks in the "Consolidated I" aggregate from the Central Bank of Brazil. Unfortunately, this information i s only available with quarterly frequency between 2000 and 2006. Data from Bankscope may have longer time series coverage but a large number o f banks appear missingintheir sample. Furthermore, comparable cross-country data at the market segment level would be useful to verify to what extent weaker price competition in the retail sector and higher dispersion of rates at lower interest rate levels are results specific to Brazil or applying in general to other countries as well. This specific question, which also applies to the results of the direct approach in the next chapter, would require significant additional analysis given the amount of country-specific information needed. 27The p-value of the hypothesis test that the sum of the coefficients of the cost of funds, labor and capital i s equal to zero i s 0.29. 22 4. Estimationof Performanceby MarketSegment 4.1 OverviewoftheApproach 28. In order to obtain a deeper understanding of competitive behavior in the banking system, a financial measure of bank performance by market segment is estimated.This so-called `direct approach' consists of allocating revenues, costs and all other line items in the financial statements (balance sheet and income statement) of the banking system into different business lines2*. The approach results in the creation of stand-alone notional financial statements and related financial ratios (e.g. pre-tax return on capital and assets) by business line as o f December 2005, which can then be compared to each other and to those for the entire banking system. The objective o f this exercise i s not to pinpoint precisely the actual returns o f each business line, but rather to compare profitability across business lines - in particular, to demonstrate that retail banking i s significantly more profitable than corporate banking. 29. Although methodologically different, the direct approach can be seen as complementary to the indirect approach. Firstly, while the indirect approach can be interpreted as an exercise on market conduct (i.e. rate settinghensitivity for retail vs. corporate lendingproducts), the direct approach can be perceived as a related exercise on perjGormance (Le. profitability for retail vs. corporate banking)29. Secondly, while neither approach i s sufficient per se to determine the level o f competition by market segment, both lower lending rate sensitivity and higher business line profitability are typically positively correlated to market power. In other words, one would expect that, ceteris paribus, a more profitable business line, or one with relatively less sensitive lending rates, would be characterized by relatively lower price competition. It i s also important to note that the results o f the two approaches, as well as the process o f arriving at them, facilitate the identification o f different factors that drive performance and means o f competition in each market segment (see Chapter 5). 30. Fourlinesof businesswere identifiedfollowingstandardmarketpracticeand used in this analysis: Retail, Corporate, Government, and Treasury. Each business line has been defined such that it aligns closely with actual bank practice, i.e. each business caters to the banking needs of a specific market segment with common infrastructure that i s managed as a unit within a typical bank3': 28The broadest definition of the Brazilianbanking system (the so-called `consolidated - level 111') was adopted for purposes of analysis, which covers all banks (including BNDES), finance companies, and *'While credit cooperatives. the indirect approach focuses on the lending side, the direct approach adopts a more comprehensive view that includes other types of bank services (e.g. deposits, payments) relevant for financial performance. For comparative purposes, subsequentanalysis inthis chapter attempts to disentangle the relative contribution of loans and deposits on profitability. 30Various types of services (loans, deposits, payments, etc.) are included under the same businessline becausethey address the same client base and require the use of the same bank delivery channel and staff. Different services within a business line might generate different profitability patterns, and there might also be cross-subsidy effects; however, there i s insufficient information to undertake a more disaggregated 23 0 Retail i s the branch-based business of commercial banks, and includes deposits, loans and payments services o f individuals and small businesses31 0 Corporate i s a more centralized function that comprises all banking business (lending, cash management, etc) relatedto medium-sized and large companies 0 Government consists of all banking business related to federal, state and municipal authorities. State-owned entities are excluded from this business line since they are in practice usually mapped to Corporate. Bank investments ingovernment bonds are also excluded since they are handledby Treasury 0 Treasury includes the central trading, investment, and money management facilities, as well as all other, relatively smaller businesses supplied by some banksthat are not captured inthe other three business lines. 31. Since the primary focus is to assess differences between Retail and Corporate, the other two business lines are not analyzed in detail. As shown later in the chapter, the size of the Government business line i s very small and i s not representative o f the public sector's importance as a client (e.g., securities) and o f its influence (e.g., via bank ownership or directed lending) in the banking system. In addition, since Treasury i s effectively the residual o f all other bank businesses, its estimated financial performance cannot be taken as fully representative. 32. Various public data sources were tapped and assumptions were made in order to estimate a stand-alone notional balance sheet and income statement for each of the business lines. Although the majority of the data comes from Bacen, other sources3*were included as necessary in order to facilitate the allocation o f different line items inthe financial statements. Table 4.1 below summarizes the main allocation results, while a detailed description o f the methodology and allocation mechanisms can be found in Appendix II.Given that the allocation assumptions (particularly about costs and equity) are important, interviews with senior management of eight Brazilian banks of different size and ownership structures served as a consistency check on the overall methodology. Finally, a sensitivity analysis was undertaken on those variables with higher uncertainty and potentially higher impact on the results of the relevant market segments; the main results are summarized at the end o f this chapter, while the detailed analysis can be found inAppendix III. analysis. It i s also not required: the objective of the exercise is to compare performance across - as opposed to within - business lines that represent different market segments (retail vs. corporate). 31 Dependingon the bank, other products and services provided by the Retail businessline could include sales of investment products, insurance brokerage, payrolVemployee benefits, safe-keeping services, etc. Small businessincludes micro-enterprises and those small-sized fEms that are typically served by the same distribution channels as individuals (Le., branch networks), so they share the same cost base; this implies that the definition may vary across different banks depending on their target market and strategic approach to this market segment. See Clark et a1(forthcoming) for a similar approach. 32 These included, for example, data from the Braziliandeposit insurance agency (Fundo Garantidor de Crkditos or FGC) and annual 20-Freports filed with the US Securitiesand Exchange Commission (SEC) by Brazilian banks that are listed in the New York Stock Exchange. 24 Table 4.1:BankingSystem Balance Sheet (December 2005) -Allocation by Business Line Retail Corporate Government Treasury Cash, floating and repos I I 100% Securities and derivatives 23% 5% 1% 71% Loans andleases I 45% I 52% I 3% I Other assets 64% 18% 1% 17% Deposits 58% 35% 3% 4% Reverserepos 100% Acceptances 18% 18% 64% Borrowings 23% I I 77% Other liabilities 59% 28% 1% 13% Net WorthEquity 46% 30% 2% 22% TotalAssets /Liabilities 42% 26% 2% 30% Interest revenues - Loans and leases 69% 29% 3% - Securities andrepos 12% 5% 1% 82% - Derivatives andF/X 100% - Required deposits 73% 24% 3% Interest expenses - Deposits, accept. andrepos 35% 22% 2% 41% - Borrowings 49% 51% - Leases 80% 15% 1% 4% - F/X 100% Non-interest income 42% 8% 1% I 50% Operating expenses - Payroll expenses 15% 1% 4% - Overhead 15% 1% 4% - Other operating expenses 100% Allowance for bad credits 63% 31% 6% I Pre-tax income 60% 21% I% 18% 4.2 Main Findings 33. Retail is the largest business line in the banking system and has a very different structure from Corporate (see Figures 4.1 and 4.2). Retail represents around 40 percent of the system while Corporate accounts to only about 26 percent, while both are much bigger than the Government business line33.On the assets side, loans represent only around one third of the Retail business line as a result o f the large share of other assets that are related to its infrastructure (branches and ITC systems); by contrast, loans represent the majority of assets for the other two business lines. On the liabilities side, deposits constitute around half o f both Retail and Corporate liabilities and are proportionally higher than for the Government business line. 34. The Retail business line generates a much greater proportion of net interest income than Corporate. This can be attributed to a combination of higher loan rates and spreads (the average loan yield on performing loans for Retail i s around 45 percent) and relatively lower cost of funding (the weighted average cost of deposits, reverse repos and acceptances i s around 9 percent)34. As a result, this business line generates around 70 percent of the total interest margin o f the banking system (see Figure 4.3). Conversely, the Corporate business line exhibits much tighter terms, with relatively lower loan rates (average yield on performing loans of only 16 percent, primarily due to the presence o f significant volumes o f directed lending remunerated at below-market rates) and somewhat higher funding rates (weighted average cost of deposits, reverse repos and acceptances i s around 10 percent). As a result, Corporate i s only responsible for 20 percent o f the total bankingsystem's interest margin(see Figure 4.4). ~~~~~~ ~~ 33Analysis of the Treasury line i s omitted here since, as previously mentioned, it i s not the focus of the study and it includesa variety of other unrelatedbusinesses. 34The counterpartto lower funding costs is highreserverequirementsandother limitations (e.g., on the use of savings accounts for housingfinance) that do not allow puttingmuch of this liquidity into profitable use. 26 Figure 4.1: Structure of Assets BRL 687 EN BRL 436 BN BRL 30 BN BRL 1,675 BN 100% 90% 804'0 704'0 u1 60"a 4P L 50% CI 40% 30% 20% 10% 0% Retail Corporate Go+-emment BankingSystem Source: Study calculations based on Bacen data. Figure 4.2: Structure of Liabilities BRL 687 BN BRL 436 BN BRL 30 BN BRL 1,675 EN 100% 90% 800/0 70% 4* 60% 3 3 c 50y0 $ 40'6 30% 20% 10% 0% Retail Corporate Government BankingSystem Source: Study calculations based on Bacen data. 27 Figure 4.3: Net Interest Incomefor Retail Figure4.4: Net Interest Incomefor Corporate Business Line (BRL million) Business Line (BRL million) Interest Interest Revenues` Revenues` Interest Interest Expenses* Expenses` Interest Interest Margin Margin * Interestrevenues include interest receivedfrom performing and non-perfoming loans, remunerated reserve requirements, and inter-business line transfers. Interest expenses include interest paid on various types of liabilities (deposits, acceptances, reverse repos, borrowings used tofund directed lending) where applicable. Interest margin is the difference between interest revenues and expenses (Le. net interest income). Source: Study calculations based on Bacen data. 35. Most of the Retail business line's net interest income is actually derived from lending rather than deposit-taking, while the situation is more balanced for Corporate. Using a transfer pricing methodology that i s common in financial analysis, one can actually decompose the interest margin by business line into two parts: margin from lending versus margin from deposit-gathering activities. The lending margin consists o f the loan spread over-and-above the relevant fundingrate, while the borrowing margin i s essentially the interest benefit derived from sourcing deposits at a lower cost than the interbank market - see Appendix I1for a detailed explanation. According to this decomposition, more than 80 percent of Retail's margin comes from lending activities, while for Corporate the figure i s close to 50 percent (figures 4.5 and 4.6). It i s worth noting that, even though the benefit of deposit gathering i s quite significant (i.e., the difference between the interbank rate and the average deposit rate i s large), it i s partly negated by the fact that around 40 percent o f total deposits are `tied up' in low-yielding reserve requirements and directed lending (e.g., housing finance) obligations. 28 Figure4.5: MarginDecompositionfor Retail Figure 4.6: MarginDecompositionfor Corporate (BRLmillion) (BRLmillion) Lending Margin' Borrowing Borrowing Margin* Margin' Interest Interest Margin 84.0 Margin 254 * TheLending Murgin is the difference between loun-reluted interest revenues and fhe cost offunding them (interbank ruteforfree lending and relevunt `liedfunding' rutefor directed lending). The Borrowing Murgin is the difference between interest incomefrom deposits (both uctiiul income from reserve requirements and `notionul' income from investing freely uvailuble deposits ut the interbunk rute) and interest expenses. TheInterest Murgin is the sum of these two inurgiris (i.e. net interest income). Source: Study calculutions based on Bacen dutu. 36. The Retail business is significantly costlier to operate than Corporate. This stems both from high operational costs (as shown by higher cost-income and operational expenses/total assets ratios) and from credit expenses (Retail accounts for almost two- thirds of total loan loss provisions). The former are associated with the required supporting infrastructure (Le. the branch network and ICT), while the latter can be attributed to the relatively higher average default and loss experience o f Retail clients (itself partly due to insufficient credit history and lack o f positive information sharing, see next chapter for a discussion of the main drivers). 37. Even after controlling for risks and expenses, the Retail business line generates higher profits than Corporate because its high margins and fees more than compensate higher operating costs. This i s due to the aforementioned higher spreads, in particular the lending margin. Higher profits also stem from the much greater participation of the Retail business line in non-interest income, which comprises fees generated by loans, payments and other banking services. As a result, Retail generates around 60 percent o f the total banking system's pre-tax income. By contrast, even though the Corporate business line is responsible for less than 15 percent of operating costs and one third o f credit expenses, it only generates around 20 percent o f the total pre-tax income o f the banking system (see Figures 4.7 and 4.8). 29 Figure4.7: Net IncomeDecompositionfor RetailBusiness Line (BRLmillion) Interest Margin Fees Gross Income 117.6 Operating Expenses Credit Allowance Net Income 28.5 Interest Margin and Fees represent net interest and non-interest income respectively. Operating expenses include payroll, overhead, tax and other operating expenses. Credit allowance refers to loan loss provisions. Source: Shidy calculations bused on Bacen data. Figure 4.8: Net Income Decomposition for Corporate Business Line (BRL million) Interest Margin Fees Gross 31.7 Income Operating v Expenses t 13.8 Credit Allowance Net Income 10,2 Interest Margin and Fees represent net interest and non-interest income respectively. Operating expenses include payroll, overhead, tax and other operating expenses. Credit allowance refers to loan lossprovisions. Sonrce: Study calculations based on Bacen data. 30 38. The Retail business line is also more profitable when judged by profitability ratios such as the pre-tax return on assets (ROA) or risk-adjusted ones such as the pre-tax return on capital (ROC). While the former indicator i s not weighted for risk, the latter attempts to capture the relative riskiness of different business lines by estimating their capital requirement for credit risk and inserting it in the denominator of the equation35. As shown in Figure 4.9 below, Retail's estimated ROC i s 38 percent, compared to only 21 percent for Corporate. While Corporate i s a nominally profitable activity, it i s worth noting that its ROC i s close to the base rate (SELIC) during 2005, which i s often used as a standard hurdle rate. The Government business line generates a lower ROC partly due to its proportionally higher NPLs and related credit expenses, while Treasury exhibits a ROC closer to the banking system average of 29 percent, but - as previously mentioned - these two business lines are not so relevant for the study. Using the ROA indicator also yields similar relative performance results for the Retail and Corporate business lines (4.1 percent and 2.3 percent respectively). Figure 4.9: EstimatedReturnon Capital for Different Business Lines (2005) Retail Corporate Government Treasury I Total a 29% SELIC = 19" Source: Study culcitlutions bused on Bucen dutu. 35 This i s also the approach undertaken by the Cruickshank Report (2000). Eventhough economic capital (i.e., capital neededto support the economic risks of the business) would be a conceptually more appropriate variable than regulatory capital, the latter is easier to estimate and i s often inpractice the binding constraint, particularly for larger, well-diversified banks. The estimated regulatory capital i s allocated to the various business lines based on their respective credit exposures measured using Basel I rules (Le., specific credit risk weights by asset type), with the residual book equity kept in Treasury. This i s also consistent with the fact that such business lines mostly incur credit risk, while Treasury typically `collects' and i s responsible for managingmarket and interest rate risk. It should be noted that the capital charge for credit risk used in the study is based on the banking system's actual capital adequacy ratio (15 percent) rather than on the regulatory minimumcapital requirement according to Basel I(8 percent) or for Brazil itself (11percent). 31 4.3 Discussionof Findings 39. The findings of the direct approach are in line with those of the indirect approach. Inparticular, although there might remain some uncertainty about the level of absolute returns or the extent o f out-performance by the Retail business line, the above estimations provide support to the hypothesis that the degree of price competition i s lower in the retail (as opposed to the corporate) banking market. The findings also substantiate the study's premise that such analysis needs to take place at a disaggregated level in order to derive meaningful results. 40. The aforementioned results are also similar to those reached by studies elsewhere. The finding that retail banking in Brazil i s higher-yielding and more profitable - even on a risk-adjusted basis - than corporate banking i s not unusual. Previous studies in other countries have confirmed this result, either directly by estimating risk-adjustedprofitability measures for different business lines or indirectly by comparing the valuations o f different financial institution^^^. Importantly, these findings have also been corroborated by various antitrust commission initiatives and other studies on specific segments of the retail banking market, such as payments services and lending to small firms37.In that sense, Brazil does not appear lo be exceptional in terms of differences in performance between business lines, although this would have to be fully substantiated by undertaking a similar type o f anal sis for other countries. What seems particular of Brazil i s the extent o f such differences3r. 41. Findings need to be treated with caution due to significant methodological and data constraints. Methodological constraints include heavy reliance on accounting statements that might significantly differ from economic reality (e.g., in the treatment o f expected credit losses), and estimation at a point in time (2005) as opposed to a full business cycle, which might distort both absolute and relative rates o f return3'. Data constraints include the unavailability o f business line-level information on volumes and rates, which necessitated makingallocation and rate assumptions. 42. In any case, sensitivity analysis supports the robustnessof the results under plausible scenarios. Analysis was undertaken for key variables that were not fully corr~borated~~ inorder to identify the robustness of different assumption^^^. Appendix III 36See, for example, Clark et a1(forthcoming), the Cruickshank Report (2000), Oliver, Wyman & Company (2003), and Morgan Stanley and Oliver Wyman & Company (2002). 37For example, the Cruickshank Report (2000) attempts to estimate financial performance for UKbanks at a disaggregated level (Le. by individual economic market and customer class) using accounting data. 38Still, Clark et a1(forthcoming) compare recent performance of eight large US banks and findthat the retail businesshas returnstwo to three times higher than non-retail. However, the same article later concedes that the evidence i s mixed about the extent of such differences. 39However, international experience fromdeveloped countries suggeststhat default rates on corporate banking are more correlated and volatile than those on retail banking. As a result, one would expect corporate banking to exhibit even better performance in `good times' than normally. This phenomenon i s also recognized inBase111, where exposures to retail and small/medium-sized firms receive a lower credit riskweight than corporate exposures and therefore have relatively lower regulatory capital requirements. 40Sensitivity analysis on variables whose values are confirmed - for example, the average interest rate of Selic for 2005 - i s not undertaken, even though these may also represent key drivers of performance. 32 provides a detailed description of the different sensitivity tests and of their outcomes. Assumptions about lending rates and the allocation o f operating expenses were found to be the most sensitive variables, in terms of influencing the relative profitability o f the Retail versus the Corporate business lines. However, the results reported above were generally insensitive to most plausible scenarios, thereby confirming their robustness. 43. The above analysis could be usefully extended in the future if sufficient data was available. Inparticular, additional valuable insights could be attained by makingthe analysis more granular. Examples include adopting a more disaggregated definition o f business lines (e.g., separating out rural, asset management, credit cards, investment banking activities etc.), or analyzing and comparing performance by groups of banks (e.g., by ownership, size, or target market segment) or by geographical regions. However, such analysis would require significantly more bank-level (and likely non-public) data than is currently available, and would need to address the additional complexities that would be introduced (e.g., allocation o f costs given the likely presence o f joint cost structures/economies of scope at a more disaggregated level, allocation o f risks, treatment of cross-subsidies etc.). 44. One avenue of future research would be to analyze the effect of government interventions on the Brazilian banking system's financial performance. Inaddition to the fact that the government represents the largest client of the banking system (government securities), it also has a very important and multidimensional influence. This includes the complex role played by state-owned banks (40 percent of total system assets), directed lending schemes (one-third o f total lending), reserve requirements (30 percent of total demand, savings and time deposits), special and regulated deposits (one- fourth o f total deposits), and the tax structure (corporate income and financial transaction taxes). Additional scrutiny on the effects o f individual types o f government intervention would shed light on the nature and extent to which each one contributes to the system's structure and performance. 41Sensitivity analysis i s undertaken at the level of individual variables; it is possible that combinations o f different variables, while insignificant on an individual basis, could also significantly affect the results. 33 5. Drivers of Performance by Market Segment 45. Several underlying causes may account for the significantly different financial performance of retail and corporate banking. Firstly, the difference in performance might be apparent rather than real because the estimated stand-alone results do not capture the complex inter-linkages and cross-saleshbsidies between business lines. For example, as was indicated in several bank interviews, corporate banking sometimes represents a `loss leader' for generating additional business booked in Retail (e.g., payroll loans) or Treasury (e.g., foreign exchange trading). This i s a valid argument in the case of Brazil, although it likely cannot explain by itself the large difference in performance across business lines. Secondly, and consistent with traditional finance theory, the discrepancy in performance might be a temporary phenomenon related to the current stage in the business cycle that will eventually sort itself out. While there might be some merit to this argument42, one would need to show that retail banking was less profitable on a risk-adjustedbasis than corporate bankingat an earlier point in the cycle, perhaps during the 2002 downturn - an unlikely result given the even higher loan spreads that existed for retail loans at that time. Finally, as described below, the various market drivers (which can include cross-sales and the business cycle mentioned above) affect business lines in different ways, and their combined effect leads to the differences in performance across bankbusiness lines. 46. Importantly, a host of factors that influence revenues, costs, and risks explains the structure and financial performance of different market segments. These factors can be common across business lines (although they might have different effects), but they can also differ substantially by - and even within a - business line. They include, among others, inherent client and transaction features, regulation (including the type and degree o f government involvement inthe financial system)43,and the state of the financial infrastructure. All these factors are relevant in determining the degree of effective ~ o m p e t i t i o nand~~ontestability~~ ~ in particular market segments and, by implication, on market structure and relative profitability (see Table 5.1 for a summary). 5.1 Retail Banking 47. Retail banking shows many aspects of a network business in which physical presence is very important. It i s characterized by significant fixed costs of operation, 42For example, Clark et a1(forthcoming) speculate that there i s a cyclical element at play in explaining the higher returnson retail activities for large U S bank holding companies, while Hirtle and Stiroh (2005) analyze a sample of 700 U S bank holding companies for 1997-2004and conclude that in the US "while retail banking may be a relatively stable activity, it is also a relatively low-return one". 43For example, McKinsey (1998 and 2006) stresses the importance of regulation as a key barrier to increased productivity growth for Brazil's capital-intensive retail banking sector. 44Competition i s effective only if the consumer has sufficient information to make rational choices and i s able to exercise them at low transaction costs. 45A contestable market has low entry and exit barriers, thereby restrainingincumbent firms from exercising monopoly power because of the credible threat of new entrants. 34 including branches, ATMs and information and communications technology (ICT) infrastructure. Yet, as the study indicates, margins and fees o f this business more than compensate the higher inherent fixed and overhead costs. Not surprisingly, efficiency i s a key driver of this business and the cost-to-income ratio i s the traditional control variable. 48. There are several drivers on the revenues side that support high interest income and fees. The small size of individual clients does not typically allow them to negotiate rates, while independent alternative non-bank providers o f such services are relatively few46.Regulation i s important since it tends to discourage client switching due to the penalizing financial transaction tax CPMF (Contribucao Provisoria sobre Movimentacao financier^)^^, the lack of sufficient positive information sharing on borrowers (cadastro p o ~ i t i v o ) ~ ~ and, the structure of payroll account relationships4'. Other potentially important switching barriers include the convenience factor and related need for a branch network by individuals, product bundling by banks (e.g. linking o f deposit accounts to cheaper payment and loan services), as well as inefficiencies in the retail payment systems infrastructure. Examples of the latter issue include a low level of interoperability between banks and clearinghouses, deficient integration among Automated Teller Machine (ATM) and Point o f Sale (POS) networks, and lack of standardization in communication protocols50. As a result, customers o f some banks cannot access other banks' networks, creating an infrastructure barrier to competition. Other factors such as branding, service quality and perceptions o f solvency also act as barriers to entry for potential competitors by leading to less price-sensitive behavior by consumers. Many o f these are common to other countries, as evidenced by international consumer surveys indicating that proximity, perceptions o f solvency, and brand image tend to be ranked above pricing as determinants o f the primarybanking relationship. 49. Although funding costs are low, operating costs inthe Retail market segment are very high. Regulation - in particular, the presence of savings and special deposit accounts - has contributed to low-cost funding. However, highreserve requirements tend to offset this advantage. On the operational side, in spite of economies of scale stemming from Brazilian banks's success at standardizing and automating important features o f the delivery system, operational costs remain very high. These include high fixed costs that are needed to set up the branch network and ICT infrastructure, as well as high variable costs that are needed to run it, which stem from high-volume, small value transactions 46Although there exist several credit providers for consumer durables (e.g. appliance stores), they tend to be associatedwith - and financed by -the banks themselves. 47Infact, CPMF is not charged ifthere is an account transfer of the same type and with the same names, but according to bank interviews it dissuadesmost retail customers from performing any financial movements. See Ebrill, Summers and Coelho (2000) for a discussion of such taxes. 48The lack of positive information limits the predictive ability of collected data and prevents consumers who honor their obligations to benefit from the `reputation collateral' that they have built. 49The current practice inBrazil is that payroll accounts tendto lock-in retail relationships. Onthe one hand, banks obtain important information on the customer while, on the other hand, customers increasingly receive payroll-linked loans and therefore tend to stay with the same bank. The potential payoffs from this "captive market" seem to be so high that, in fact, banks pay considerable amounts to corporations and statellocal governments in order to obtain those relationships. 50See Western Hemisphere Payments and Securities Clearance and Settlement Forum (2004) for more details. 35 and from an inefficient retail payments mix. In particular, the infrastructure for clearing and settling transactions among banks i s fragmented and sometimes incompatible, which reduces economics of scale. The predominant use of cash and checks as payment instruments also implies higher costs and increased operational risks. In order to cope with these costs, banks continually strive to capture and keep customers in their own network. Finally, additional costs might also exist relating to non-price competition5' and possibly to internal technical inefficiencie~~~stemming from inadequate competition in specific product markets. 50. The loan loss experience of the Retail business is relatively higher, but less volatile, than for Corporate. This segment i s prone to higher default and loss rates (albeit more granular and less correlated to the business cycle) than for corporate, as evidenced by the ratio of non-performing loans in the study. Knowledge of customer credit history, including both positive and negative information, i s relevant to any credit decision - however, Brazil lacks private credit bureaus that can fully share such information. This not only reduces access to credit to potentially creditworthy customers while increasing overall costs to the system, but also helps strengthen the customer relationship with a bank that possesses such information. 5.2 Corporate Banking 51. In contrast to retail, the corporate business line is more akin to an auction market and presents relatively more price competition. Our results strongly suggest that the corporate market has a higher degree of price competition than retail, as evidenced by tight interest rate margins and much lower profitability. Although this business i s less costly to operate, reduced margins depress returns to the point where they are comparable to those obtained simplyby investing ingovernment bonds. 52. Lower corporate loan spreads and revenues can be largely explained by regulation, as well as by greater effective competition and contestability. Anecdotal evidence from interviews suggests that Chief Financial Officers o f corporations tend to shop around for rates, and look at the banking relationship with a comprehensive view. Thus, share of wallet of the customer is a key driver for banks in this business and it encourages cr~ss-subsidization~~.There exist more substitute providers (e.g. capital markets, offshore banks) in this market segment, while entry barriers - at least on the lending side - are lower. In addition, the significant presence of directed lending (46 percent of all corporate loans as of end-2005, compared to only 20 percent for retail loans) depressesaverage loan spreads and interest income. 51Banks not only compete in terms of prices; non-price competition can take place interms of quality of service, delivery channels, branding, product innovation, relationship management etc. 52This type of efficiency measure(so-called X-efficiency) is the effectiveness with which a given set of inputsis usedto produce outputs. 53This may also provide indications of cross-subsidies inthe payrollaccount businessmentioned above - for example, Brazilian banks may sacrifice returns on the corporate relationship in order to profit from the relationship with employees of those corporations that are served inretail through payroll accounts. 36 53. The cost of running the corporate business line is relatively low. Even though the cost o f funding i s higher than for retail due to the lack o f low-cost savings deposits, operational costs are much lower. The main contributing factor i s the nature of the delivery model (no need for extensive branch network, combined with lower-volume, higher-value transactions), although the extensive scale and type (e.g., rural credit) o f directed lending operations might add to operational costs. 54. Another important feature of the corporate market is its relatively lower (but more volatile) credit expenses. As previously shown, non-performing corporate loans are closer to international averages than those in retail. Although the information needed to assess the credit risk of firms in Brazil - particularly larger ones - i s more widely available, the impact o f individual credit exposures i s larger and more correlated to the macroeconomic environment, requiring customized risk management models with professional financial analysis. In addition, there i s significant room for improvement in the quality of Brazilian corporate financial statement^^^, which would enhance the reliability of the input data inthose models. The inability to include positive information (including by public banks) in the borrower reports produced by the credit reporting industry also hinders the predictive ability of such models. Moreover, much of corporate lending (especially to the middle market) i s collateral-driven and i s therefore more reliable on the credit infrastructure, making it vulnerable to related problems with weak creditor rights such as judicial debt collection difficulties. In fact, Brazil fares below many Latin American countries in terms of legal rights and contract enforcement proceeding^^^. Finally, the inter-linkage among different credit registries i s not yet complete, while links between credit registries and payments systems remain imperfect56. 54Examples include overlapping mandates for accounting and reporting standard-setting, as well as significant gaps between Brazil's Generally Accepted Accounting Principles and International Financial Reporting Standards. 55According to the WB Doing Business2006 statistics, Brazil's legal rights index score is 2 (versus 4.5 for the region and 6.3 for the OECD), while the cost of enforcing contracts is 16percent of debt value (versus 23 for the region and 11for the OECD) and the time required i s 616 days (versus 642 for the region and 351for the OECD). 56See Western Hemisphere Credit and Loan Reporting Initiative (2005) for more details. 37 . . . . . . . 0 0 . 0 . M i;: a 6. Conclusions and Policy Implications 55. The study confirms the need to analyze the Brazilian banking system at a disaggregated level. The banking system must not be seen as a single homogeneous business, but rather as a set o f markets in which competitive conditions can vary substantially. For example, as both the direct and indirect approaches strongly indicate, the retail and corporate market segments are characterized by different size, structure, conduct (interms of loan pricing) and performance. There are also significant differences within each o f these segments (e.g., between deposit and loan product markets), so defining the relevant market and accessing data become important issues for such analyses57. Differences across market segments in particular need to be taken into account, aggregate studies tend to average these out and thus provide a blurred picture. As such, the answer to Belaisch's (May 2003) paper entitled "Do Brazilian Banks Compete?' cannot be a simple yes or no; it i s rather a question o f degree and it depends greatly upon the specific market segment. 56. One key finding of the study is that the Retailbusiness line is less sensitive to price competition and exhibits considerably higher returns than Corporate. Even though the Retail business line i s both costlier and riskier than Corporate, it i s significantly more profitable - even on a risk-adjusted basis - due to higher loan spreads and fees. Although this result needs to be treated with some caution due to methodological and data constraints in the analysis, it appears robust to most plausible sensitivity scenarios. While this finding i s not unusualin the international context, Brazil may appear exceptional in the extent o f such performance differences. The analysis also highlights the importance of identifying the different drivers of performance and determinants of levels of (effective) price competition and contestability. Some of these factors are within the scope o f government policy, while others are inherent to the type o f clientltransaction and are therefore less amenable to change. 57. There are both system-wide and market segment-specific policy recommendations that can be drawn from the analysis. At the banking system level, there i s a clear need to strengthen the oversight framework and institutional capacity to promote competition in the financial sector, which i s currently primarily (at least until the new antitrust law for the banking sector i s approved by Congress) in the hands o f Bacen5*. Drawing a clear distinction between the criteria underlying the prudentialhupervisory and competitiodconsumer protection perspectives, as i s the practice in other countries, would be a useful practical step, which would also facilitate relevant data collection and analysis. With regards to market segment-specific issues, government policies that encourage price competition among banks - especially in the Retail market - should address those factors that drive up revenues, costs, and risks, and hence adversely impact efforts to ensure wide and affordable access to banking services. 57Moredisaggregated(but non-public)data for this study would have beenwelcome andcouldallow analysis at bankgrouplevel- for example, testing whetherMcKinsey's(1998) findingthat "public banks have contributed to the low level of price competition by creating a price ceiling [because of lower productivity] under which private banks can comfortably operate" still holds. '*SeeWorldBank (May 2006) for moredetails. A formal agreement on a reformedantitrust regimefor the bankingindustry was reachedin 2005 betweenthe antitrust authorities(Conselho Administrativo de Defesa Economica) andBacenas an interimmeasureuntilthe law i s passed. 58. Further promoting the portability of bank accounts would encourage more price competition and contestability, especially in Retail banking. As previously mentioned, current regulations discourage the transfer of accounts from one bank to another either by punishing financial transactions (CPMF tax) or by inherently locking clients with one bank as a result o f the nature o f payroll account relationships. The latter i s a particularly important consideration given that the growth of consumer lending in recent years can be attributed primarily to payroll loans59.Inthis realm, an attempt to de- link the retail and corporate relationships has already been initiated by the authorities in late 2006 and it should be considered as a first step in facilitating client switching in the banking system. 59. The issue of positive credit information sharing is crucial in facilitating access to credit, particularly for lower-income borrowers and for smaller firms. A draft "cadustro positivo" law that provides the legal foundation for the creation o f positive information databases i s awaiting approval by Congress. This initiative will allow banks to share the positive credit history on prospective borrowers, reducing information asymmetries in credit markets (particularly for lower-income borrowers) and broadening access. Passage o f this law would be particularly timely given the current move by banks to expand loans to independently employed individuals and small firms, because the market for lending to employees o f larger firms and the public sector i s progressively reaching maturity. Expanding this market i s conditional on securing access to sufficient credit information6'. In addition, strengthening the inter-linkage among credit registries as well as the links between credit registries and payments systems would increase transparency and contribute to reducing the costs and risks of bank lending, thereby promoting access and decreasing loan spreads. 60. Expanding payment system interconnection and improving the retail payments mix would also increase banking system competition and efficiency. While the technological platform for large-value transactions i s fully operational and follows best international practices, there i s ample room for improvement in the efficiency o f retail payment systems and instruments. To address these issues, Bacen i s implementing a series of reforms to modernize retail payments systems (Sistema de Pagamento Brusileiro) by encouraging integration of ATM networks, standardizing communication protocols, and promoting electronic means o f payment. Looking forward, it i s o f paramount importance that Bacen fully implements this reform agenda since it will promote efficiency and encourage greater competition among payments providers. 61. Further development of capital markets is also critical to enable them to play a more active role in the provision of long-term finance. Brazil's long history of macroeconomic instability, coupled with high interest rates and public intervention inthe market for long-term capital, has contributed to historically weak local capital markets. The recent increase in public listings o f new companies on BOVESPA (Sgo Paul0 stock exchange) has mobilized equity markets, but fixed income markets remain incipient61. In 59 Infact, payroll loans have grown dramatically from around 10percent of loans to individuals as of end- 2003 to over 25 percent as of end-2006. 6o Interviews with senior executives of banks that are active in the retail segment indicate that their focus will progressively move to independent workers and other professionals not affiliated to a corporation. 61 Trading value at BOVESPA has doubled over 2000-05, while commercial debt offerings registered with the Brazilian Securities and Exchanges Regulator (CVM) multiplied 10times, albeit from a very low base. 40 addition to continued macroeconomic discipline and development o f the domestic long- term sovereign yield curve, initiatives to address distortions inthe long-term debt market would also be beneficial. Furthermore, the potential of pension funds and insurance markets to mobilize long-term financial resources and contribute to financial development (e.g., via infrastructure and project finance) can be exploited. Assuming the current scenario of decreasing interest rates persists, capital market-based operations will likely grow more and will progressively replace traditional corporate lending in this market segment. 62. Likewise, improvements in corporate financial reporting are essentialto boost the development of capital markets and to promote sound corporate governance practices. Currently, corporate financial reporting is governed by several laws and regulations outlined inthe Corporations Law. Although this Law has been amended twice (1997 and 2001) since its inception in 1979, the legislative process can be lengthy and cumbersome, especially when contentious issues are at stake. A positive development in this realm would entail amendment of the Corporations Laws to exclude detailed accounting and financial reporting rules, which in turn should be governed by sub- legislative acts. Furthermore, the establishment of an independent standard-setting body with authority to issue standards on general corporate financial practices would facilitate Brazil's convergence with international practices6*. 63. Legal rights andjudicial proceduresfor contract enforcement should also be tackled to improve the overall efficiency of the system. Brazil fares below many Latin ' American countries in terms o f legal rights and contract enforcement proceeding^^^. Recent changes to the bankruptcy law (Law No.ll.lO1, February gth, 2005) have been beneficial, but a higher focus on enforcement i s needed, which would help reduce credit risks and costs, thereby expanding access to credit for certain borrower segments that depend on collateral-based lending (e.g., small firms). 64. Finally, the nature and extent of government interventions in the Brazilian banking system also need to be reassessed carefully going forward. In addition to the fact that the government represents the largest client in the banking system (government securities), it also has a very important, complex, and multidimensional influence via state-owned banks, directed lending schemes, reserve requirements, special and regulated deposits, and the tax structure. As indicated inthe study, the various types o f government interventions represent important drivers of performance in different market segments, necessitating a deeper analysis o f their individual impact and overall value-added. The improved macroeconomic conditions and declining interest rate environment in Brazil provide a good opportunity to tackle these medium to longer term issues. 62There are gaps between Brazil's generally accepted financial and accounting practices and International Financial Reporting Standards (IFRS).For instance finance leases are not recorded as such in the financial statements; Brazilian financial reporting standardsare less demanding in terms of disclosure; and the text of standards i s less detailed, leading to inconsistencies in their application. 63According to the World Bank's Doing Business 2006 statistics, Brazil's legal rights index score i s 2 (versus 4.5 for the region and 6.3 for the OECD), while the cost of enforcing contracts i s 16 percent of debt value (versus 23 for the region and 11for the OECD) and the time required is 616 days (versus 642 for the region and 351for the OECD). 41 Appendix I:The BrazilianBankingSystemfrom anInternational Perspective-AdditionalComparisons The Brazilian bankingsystem's deposits represent a lower proportion o f liabilities than in other countries, partly due to the role of legally mandated (non-deposit) funds that are used for directed lending purposes. In addition, retail deposits (including poupanCa) represent a significant share of total deposits. Figure I.: Composition of Liabilities 1 Figure 1.2: Composition of Deposits (% total (2005) deposits, 2005) Argentina Argentina 24% 0rail Brazil 4% Chile 3P.b Chile e% Colombia Colornbia 22% Memo Mexxo 23% Spain h A Spain S A h A USA 0% 20% 4G% 60% w/o to% 0% 20% 40% 60% 80% 100% Total Deposits Other Retail Deposits Wholesale Deposits K Public Sector Source: SuperintendenciaBancaria de Colombia- Direccidn Tkcnica SubdireccidndeAna'lisis Financier0 y Estadistica; - Superintendenciade Bancos e InstitucionesFinancierasde Chile- Revistci de Informacidn Financiera Dic. 2005- CorporateDeposits includesPension SystemDeposits;Brazil Credit GuaranteeFund (FGCJ and IMF International MonetarySurvey; Banco Centralde la RepublicaArgentina- Subgerenciade Estadisticas Monetariasy Financieras- Informacion Diana sobre Depositosy Obligaciones& Boletin Estadistico,Federal Reserve, and Superintendencia Bancaria deMkxico. Compositionof Deposits excludesfinancial systemdeposits (Argentinu, Colombia, Chile). 42 Although the Brazilian banking system includes some large banks, it i s not out of line in terms of traditional measuresof concentration. Figure 1.3: Market Share of Top Figure 1.4: Market Share of Top Figure 1.5: Herfindahl Index of 3 Banks 10Banks Market Concentration Argentina 43 Argentina Argentina i Brail 43 Brazil Braal Chile 41 Chile Chile Colombia 40 Colombia Colombia 1642 MeMco Me ~ c o E30 Spain 55 Spain 74 Spain USA 48 Source: Federal Reserve, Banco de Espafia, Comisidn Bancaria y de Valores de Mkxico, Superintendencia Bancaria de Colombia, Banco Centralde Chile, Banco Centraldo Brasil, Banco Central de la Repdhlica Argentina. Note: The Herfindahl index is calculated as the sum of the squares of market shares of each instinition. For Colombia, it is based on the share of all banks belonging to a single banking conglomerate consideredas one institution. Compared to most Latin American countries, Brazil has a high participation of private domestic and public banks. Figure 1.6: Ownership of Top Ten Banks 1 Argentina 59% Brazli 39% Chile 14% Colombia 21% Mexico Spain USA 0Yo 50% x)O% Foreign Private Domestic z Public Source: Federal Reserve, Banco de Espafia, Comisidn Bancaria y de Valores de Mkxico, SuperintendenciaBancaria de Colombia, Banco Central de Chile, Banco Central do Brasil, Banco Central de la RepdblicaArgentina 43 Appendix 11: Descriptionof the Methodology to Estimate Performance by Market Segment Financialstatements of the Brazilianbankingsystem The broadest definition of the Brazilian banking system (the so-called `consolidated - level 111') i s adopted for purposes o f analysis, which covers all banks (including BNDES), finance companies, and credit cooperatives. Inthis way, all directed lending schemes are properly accounted for as part of the banking system. The financial statements for 2004 and 2005 are collected, tabulated and re-arranged into a suitable format (Table 11.1). Table 11.1:BrazilianBankingSystem-ConsolidatedFinancial Statements (2004-2005) ASSETS (LESS BROKERAGE) 2QE INCOME STATEMENT m Cash and floating 32,617 23,410 Interest revenues 255.657 Repos 143,692 182,419 Loansand leases 149,905 Securitiesand derivatives 367,422 445,321 Securities and repos 86,204 Interbanktransactions 198,931 214,882 Derivatives 4,205 Branches due transfers 1.068 773 F/X 3,747 Net loans and leases 442,641 530,445 Requireddeposits 11,596 Gross loansand leases 474,108 569,227 Interest expenses 136,960 Loan lossprovisions 31,267 38,783 Deposits,acceptanceand reverserepos 107,060 Other credit outstanding 162,310 173,651 Borrowing 19,525 Other assets 5,615 6,127 Lease 9,806 Leased assets 20,147 33,536 F/X 569 zZ43 6t851 Net interest income 116,698 Total Assets 1,450,626 1,674,624 Non-interest income 80,132 Servicecharges 42,026 Method equity 4,468 LIABILITIES am m Otheroperatingrevenues 33,637 Deposits 571,670 682,703 Gross income 196,629 Demand 87,500 100.501 Operating expenses 115,625 Savings 159,383 169,032. Payrollexpenses 37,732 Interbank 23,623 25,605 Overhead 43,364 Time 262,824 334,379 Otheroperatingexpenses 34,529 Other 38,340 53,184 Taxexpenses 10,976 Reverse repos 194,395 243,926 Net operating income before provisions 72,229 Acceptances 37,115 39,162 Allowance for bad credits 24,545 Interbanktransactions 6,164 6,946 Net operating income after provisions 47,664 Suspense accounts branches 9,056 9,609 Non-operating income 166 Borrowings 164,445 172,947 Pre-tax income 47,850 Derivatives 15,522 25,293 Income tax 11,231 Other liabilities 284.860 327.602 Profit sharing rn M i 9 4 Net after-taxincome 33,798 1,450,171 1,674,624 I Allocationof Deposits,AcceptancesandReverseRepos Given the unavailability o f deposit information by business line, the allocation o f deposits and acceptances i s based on interviews with banks, complemented by data from Bacen, the Brazilian deposit insurance agency (Fundo Garantidor de Crkditos or FGC) and the IMF's International Financial Statistics (IFS). In particular, the Government's demand deposits are derived from Bacen reports, while the IFS provide the source o f total central Government deposits as o f end-2005. Non-Government demand and time deposits are assigned evenly between the Retail and Corporate business lines based on bank interviews, while savings deposits ("poupanp") are allocated exclusively to Retail. Acceptances are assumed to comprise o f two types: various types of `letras' (e.g. Zetras hipotecarias, Zetras de cambio etc.) insured by FGC are allocated based on their relative size based on FGC data, while all other types o f acceptances are assumed to be mainly debentures and are therefore mapped to Treasury. 44 Interbank deposits and reverse repos (a form of funding usingpledged securities) are also mapped to Treasury. Other deposits, which consist of so-called `special deposits' by institutions such as FAT (Fundo de Amparo ao Trabalhador) and are mostly held b y state-owned banks, are assumed to be primarily used for directed lending purposes and are therefore allocated by business line based on the type o f lending that they are assumedto support (see below for more details). As can be seen inTable 4.1, the Retail business line holds 58 percent of total deposits and 18 percent o f acceptances, while Corporate controls 35 percent and 18 percent respectively. Treasury holds 4 percent of deposits and all remaining acceptances, while Government only has a small share (3 percent) o f total deposits. Allocation of Interest Expenses on Deposits, Acceptances and Repos Two types o f data are required in order to estimate and allocate interest expenses on deposits, acceptances and repos by business line: Average 2005 balance by business line -this i s estimated by taking the simple average o f balances at the beginning and end of 2005 Average interest rate by type of deposit during 2005 - this i s estimated based on rate information in Bacen's website, as well as on yield information found inthe 20-F reports for 2005 of Brazilianbanksfiled with the SEC64. Total interest expenses by deposit product can be estimated by multiplying these two types of data; their accuracy can be assessed by comparing the resulting figure to the one published by Bacen for the banking system. The allocation o f interest expenses by product and business line i s then based on the volume o f each product that belonged to each business line as o f end-2005. As can be seen in Table 4.1, the Retail business line is charged 34 percent of all interest expenses on deposits, acceptances and repos; Corporate and Treasury are charged 22 percent and 41 percent respectively, with the remainder (2 percent) charged to Government. Allocation of Reserve Requirements on Deposits and of their Remuneration Not all demand, savings and time deposits collected by a business line are freely available to finance loans, because of the existence o f reserve requirements (RR) that need to be estimated and appropriately allocated. Each type o f deposit has its own RR, which vary with respect to the required proportion, utilization (placed with Bacen or invested in securities) and remuneration. For example, 53 percent o f all demand deposits must be kept with Bacen as RR (of which only 8 percent i s remunerated), while banks have the obligation to either lend another 25 percent for rural projects (earning a fixed rate o f 8.75 percent) or leave those funds with Bacen earning no interest (some banks ~________ The average interest rate by type of deposit is effectively a weighted average of rates of different maturities and currencies (both local and foreign currency) offered in each type of deposit. 45 actually prefer this option). This means that only 12 percent of demand deposits collected by business lines are freely available to be used for lendingpurposes65. RR are estimated by type of deposit and then allocated to business lines based on their respective deposit share. RR that are placed with Bacen appear as interbank transactions in the banking system's balance sheet, so business lines that have such RR actually book them as such on their stand-alone balance sheets. All remaining interbank transactions (Le. those that are not RR with Bacen) are assumed to be placements with other banks and are therefore allocated to Treasury. Assumptions also need to be made about the remuneration o f RR in order to estimate the income from such transactions (shown as interest revenue from required deposits in the banking system's financial statements); the average Selic rate for 2005 i s assumed to be 19 percent, while the average yield on savings accounts i s assumed to be 8 percent. This income is allocated across business lines based on their respective type and volume of RR. It is worth noting that 15 percent of time deposits must be held as RR in the form of securities investments; these investments are also allocated across business lines, while the residual volume i s assigned to Treasury. No explicit assumption about the yield o f these securities was made, so they effectively earn the average yield o f all securities on the banking system's balance sheet, which i s implicitly calculated by dividing the relevant interest income by the average volume o f these investments in2005. As can be seen inTable 4.1, the RR of the Retail business line account for 44 percent of interbank transactions and 5 percent o f total banking system securities (another 18 percent booked in Retail corresponds to housing finance-related investments - see below), and generate 73 percent of interest income on required deposits. The Corporate business line accounts for 20, 5 and 24 percent respectively. Treasury has all non-RR- related interbank transactions (34 percent) as well as securities and derivatives (71 percent), while Government has small shares of each. Allocationof Loans Bacen's monthly report on financial system credit operations i s used to allocate different types of loans (earmarked vs. non-earmarked, individuals vs. corporations, etc.) by business line. The report includes a higher credit figure than that reported by the banking system(BRL 607 billionvs. BRL 569 billion as o f end-2005) because the former i s more comprehensive and includes non-bank sources o f lending (e.g. credit unions, lending by development agencies etc.). These loans are selectively excluded from the analysis in order to arrive at the same gross credit figure as for the banking system (Le. BRL 569 billion), which needs to be allocated across business lines. Bacen's report also includes the total volume of public sector credits (BRL20.6 billion). Given that the Retail business line includes small firms, dividing up reported lending figures between individuals and firms i s insufficient for the allocation mechanism. An additional assumption that was made, based on bank interviews and 20-F reports on the 65 There are additional complications for demand deposits that are not treated in this exercise, e.g. banks have the obligation of lending an additional 2 percent of their demand deposits to microcredit. 46 distribution of loans by client size, i s that 15 percent of all corporate loans are given to small firms - with two exceptions66 - and should therefore be allocated to Retail (as opposed to Corporate). As a result of the aforementioned allocation mechanism, the Retail business line accounts for 46 percent o f the banking system's gross loans (37 percent and 9 percent are loans to individuals and small firms respectively), while Corporate and Government account for 50 and 4 percent respectively. Allocation of Non-Performing Loans and LoanLossProvisions Bacen's monthly report on financial system credit operations i s also used to measure non-performing loans (NPLs) and allocate loan loss provisions (LLP) by business line. NPLs are defined based on Bacen regulations as those loans whose credit ratings are D- H, which are loans representing higher levels of credit risk and/or are already in arrears for more than 60 days. The NPL ratios for some loan types (e.g. individuals, housing, public sector, rural etc.) can be explicitly derived from the report, while assumptions are made for all other types such that the overall estimated NPL rate for the banking system i s consistent with that reported inthe Bacen report (10.2 percent as of end-200567). Once the volume o f NPLs by business line i s calculated, the reported LLP - found in both the balance sheet and income statement - o f the banking system are allocated in the same proportion as their respective NPLs. It could be argued that retail loans, which are generally unsecured, have lower recovery rates and should therefore have proportionally higher LLP than their corresponding figure for default rates; however, to the extent that ratings and their corresponding LLP already reflect collateral coverage (i.e. they measure ExpectedLoss and notjust Probability of Default), this problem should not arise. As a result of its higher NPLrates, the Retail business line accounts for 63 percent of LLP even though it only has 46 percent of gross loans. By contrast, Corporate accounts for 31 and 50 percent o f LLP and gross loans respectively, while Government represents 6 and 4 percent of LLP and gross loans respectively. Allocation of Interest Income on Loans Two types of data are required in order to estimate and allocate interest income on loans by business line: Average 2005 volume o f performing loans by business line - this i s estimated by appropriately `scaling down' the volume of performing loans that had been previously calculated for end-2005 Those exceptions are earmarked rural lending (assumed to be divided equally between Retail and Corporate basedon interviews) and first-tier direct BNDES lending (assumedto consist entirely of Corporate loans from interviews). The reported NPLrate is higher than mentioned inchapter 2 of this study becauseit is based on stricter domestic regulations (Le. 60 rather than 90 days past due) and becauseit includes a broader definition of the banking system (Le. consolidated level I11rather than level I). 47 Average interest rate by type of loan during 2005 - this i s estimated based on rate information found in Bacen's report on financial system credit operations6*. Total interest income on loans can be estimated by multiplying these two types of data, and its accuracy can be assessed by comparing the resulting figure to the one published by Bacen for the banking system. The allocation o f interest income by business line i s then based on the volume of performing loans that belonged to each business line as of end-200569. As a result of its significantly higher lending rates, the Retail business line accounts for 69 percent o f interest income on loans and leases even though it only has 46 percent of gross loans. By contrast, Corporate accounts for 29 and 50 percent of interest income and gross loans respectively, while Government has 3 and 4 percent o f interest income and gross loans respectively. The average yield on performing loans by business line i s 45/16/22 percent for Retail/Corporate/Governmentrespectively, while the average yield on the banking system's performing loan book i s around 29 percent. As can be deduced, the results for the Corporate business line are heavily influenced by earmarked credit, which brings down the average loan yield and related interest income. Fundingof DirectedCreditanditsRemuneration In contrast to non-earmarked loans, directed credit typically has its own dedicated funding sources7o and it is provided at below-market interest rates. It is therefore important to identify the funding source of the three main directed lending schemes in order to allocate such funding- and its cost -to the relevant business line. Housing finance, as well as a small portion o f rural finance, i s assumed to be funded via voluntary passbook savings accounts (cademetas de poupanp). Given that the actual amount of housing finance i s relatively small compared to the legal requirement to use 65 percent of total poupanp for such it is further assumed that the remaining balance i s invested in low-yielding securities. Directed rural finance, primarily via Banco do Brazil, i s assumed to be funded by other ("special") deposits and demand deposits. Finally, directed long-term finance via BNDES (both first and second-tier lending) i s assumed to be funded - based on information provided in BNDES's financial statements The averageinterest rate by type of loan i s effectively a weighted average of rates of different maturities. Corporate non-earmarked loans indomestic currency are assumedto be based on a spread over the CDI rate, while all foreign currency loans are assumedto be priced off the U S Libor. Inthe case of earmarked credit with pre-defined rates (e.g. TR, TJLP etc.), assumptions were also made about the additional spread charged by banks in order to effect the transaction. 69One additional complication relates to interest income that i s actually derived from loans classified as non-performing (Le. D-Hratings). Inorder to overcome this problem, which affects the assumptions about the average interest rate of performing loans, it i s assumed that 10percent of all interest income on loans actually derives from NPLs and that its allocation across businesslines i s the same as for interest income on performing loans. 70These include constitutional funds, worker funds based on mandated contributions, legal reserve requirements, and passbook saving accounts. 71Although regulations require 65 percent ofpoupanfa funding to go to housing-related credit, banks currently enjoy several exemptions that substantially diminish the effective amount of such funding. 48 - via a combination of other ("special") deposits, borrowings and subordinated debt (classified as "other liabilities" inBacen's balance sheet for the banking system). Allocationof All Other LineItems Notwithstanding the above analysis, there remain several unallocatedline items inthe balance sheet and income statement o f the banking system. With regards to assets and liabilities, the following are the main additional assumptions that are made: Treasury i s responsible for managing all cash, floating and repos, reverse repos and interbank liabilities; it i s also responsible for the residual o f securities and derivatives, interbank assets, acceptances and borrowings that were not already allocated usingthe above methodology All other line items (e.g. other assets/liabilities, leased and fixed assets) are assumed to be allocated in the ratio of 80/15/1/4 for Retail/Corporate/Government/Treasury respectively - this i s based on bank interviews on their approximate allocations across business lines72. With regards to the income statement, the following additional assumptions are made: Treasury i s allocated all net interest revenues stemming from derivatives and foreign exchange, as well as all income from equity participations, other operating revenues/expenses and non-operatingincome Interest revenues from securities and repos, as well as interest expenses on borrowing and leases, are allocated based on their balance sheet share Service fees, as well as payroll, overhead and tax expenses are allocated in the same ratio as for all other line items on the balance sheet (Le. 80/15/1/4 for Retail/Corporate/Government/Treasuryrespectively) based on interviews. In addition, each type of asset is assigned a credit risk weight for regulatory capital purposes inorder to estimate and allocate capital73.The allocation mechanismused in the study is based on the banking system's actual capital adequacy (15 percent) rather than on the regulatory minimum capital requirement (11 percent for The estimated capital i s allocated to the Retail, Corporate and Government business lines based on their respective share of each asset type, with the residual book equity kept inTreasury75. Finally, given that the assets and liabilities o f each business line are not equal, it i s assumed that Treasury plays the `transfer clearing and pricing' role by either supplying or absorbing the difference between the assets and liabilities of each business line. As i s common practice in financial institutions, such inter-business line transfers are assumed to be remunerated at the interbank rate (CDI). 72As previously mentioned, Treasury includes the residual of all other, relatively smaller businesses that are not captured in the other 3 business lines. 73Given the unavailability of disaggregated data, assignment of risk weights takes place at a high level - for example, 100percent risk weight for all loans and othedfixed assets, 20 percent for securities etc. 74Bank book equity can be allocated across businesslines in two basic ways: (1) by allocating the minimumrequiredcapital to support eachbusinessline, with Treasury keeping `excess' book equity; (2) by allocating all book equity proportionally across business lines basedon their relative riskiness; a version of the former approach i s usedhere. 75Inits function as the bank's central money management `clearinghouse', Treasury also incurs all market and interest rate risk. 49 Allocationof InterestMargin For comparative purposes, the interest margin i s decomposed into margin from lending versus margin from deposit-gathering activities using a method standard in financial practice. The lending margin consists o f the loan spread over-and-above the relevant funding rate, while the borrowing margin i s essentially the interest benefit derived from sourcing deposits at a lower cost. The funding rate for non-earmarked loans i s generally assumed to be the interbank rate CDI which in the case o f Brazil closely follows the Selic rate. For earmarked loans, we used the relevant `tied' rate (e.g., interest rate paid on poupanga deposits for housing loans). The borrowing margin represents the difference between interest income and interest expenses on deposits and other liabilities. Interest income stems from three sources: (1) remunerated reserve requirements (`tied' deposits); (2) `freely available' deposits (i.e., net of reserve and directed lending requirements), which are assumed to be remunerated at the interbank rate; and (3) any `excess' liabilities o f each business line which are transferred to Treasury and are also assumed to be remunerated at the interbank rate. See Figure II. 1. Figure11.1:Decompositionof Interest Margin betweenLending andBorrowingActivities Net Intnrmi IncnmR on Non-earmsrked Loans - (spread over Selic) Lend ng Margin Net Interesi Income on Earmarked Loan; - (spread over relevant funding irate) ~ literest Income on `Tied' Deposits - (Rese rve Requirements) literest Income on `Frie' Depcsits - (rerrunerated at 3elic) B3rro\ving Matgin Intzrest Income on Inter-Business Line -ran;fers - (rerrunerated at Zelic) Iin: erest Cxpeiijes - 50 Appendix 111: SensitivityAnalysis inthe PerformanceEstimation The results of the direct approach that are described in chapter 4 are conditional on assumptions made about the allocation o f line items in the balance sheet and income statement, as well as on specific interest rates. Thus, sensitivity analysis i s undertaken for key variables that are not fully c~rroborated~~ in order to identify the robustness and relative contribution o f the different assumptions to the results. The outcomes of the different tests are reported inTables III.1a and b below. We have tested the relative share of deposits between the Retail and Corporate business lines. As discussed above, an important part of the allocation was based on the assumption - based on bank interviews - of an even allocation for the lines of demand and time deposits, which were not previously allocated to Government or other specific rules. We have tested the effect of this assumption by shifting 10 percent o f total Retail deposits to Corporate (and vice versa). In both cases, the key result o f the returns of Retail being higher than those o f Corporate i s maintained. While in the base case the relation o f ROC between Retail and Corporate i s 38/21, in the first test the relationship moves to 34/29 and inthe other test to 42/12. In fact, one needs to assume that the Retail business has less than 20% of time and demand deposits - an unlikely assumption, which would itself cast doubt on one of the main reasons for having a branch network - in order for the two businesslines to have a similar rate o f return. Results are also fairly insensitive to changes in deposit rates. We have tested for different assumed rates for time deposits, reverse repos and acceptances such that the total interest expense which i s already known from Bacen reports remains the same. The variability in the relationship of ROCs between the Retail and Corporate businesses i s between 35/15 and 41/26 (Treasury actually exhibits the greatest volatility inROC across the two scenarios), thus confirming the robustness o f our key result. We have tested the results of changes inthe relative share of gross loans. Inthe above analysis, we have assumed that around 15 percent of loans to companies were actually loans to small firms that come under the Retail business line. We tested this assumption intwo ways: firstly, by assuming that no loans to these companies belong to Retail; and secondly, by assuming that a larger number o f loans to firms (25 percent) are actually served by the branch network and should therefore come under Retail. The results o f the tests show that the contribution of small businesses to profitability i s fairly small. In fact, the ratio o f ROCs between RetaiUCorporate moves from 38/21 in the base case to 41/18 and 37/21 inthe respective scenarios. Sensitivity analysis was also undertaken with respect to the distribution o f loan loss provisions (LLP) among different loan products, while ensuring that the overall LLP rate remains the same as the one reported by Bacen. One test increased LLP of Retail compared to Corporate, and the other did the reverse. Results are fairly insensitive to 76Sensitivity analysis on variables whose values are confirmed - for example, the averageinterest rate of Selic for 2005 - i s not undertaken, even though these may also represent key drivers of performance. 51 these tests, there i s little change in profitability by business line and therefore in the relative performance of Retail versus Corporate. Results are more sensitive to changes in certain loan rates. We have tested this assumption in two ways: (1) by increasing a key Retail lending rate (average non- earmarked lending to individuals) by 300 basis points, while adjusting downwards by 300 basis points certain Corporate loan rates to maintain overall interest revenues; and (2) by doing the reverse, Le. decreasing the specific Retail rate by 300 basis points while increasing the key Corporate loan rates by 300 basis points on average. In the latter case, the key result o f our analysis (Le. that the return o f the Retail business remains higher than the return of Corporate) i s barely maintained, implying a higher sensitivity of results than with other variables tested. Given this sensitivity, it is important to fully confirm the validity of the assumed lending rates and spreads77. Eighty percent of service fees, payroll/overhead expenses and other items are allocated to the Retail business line based on information gathered in interviews with bank executives. We have tested this assumption by decreasing the share o f Retail to 75 percent (while increasing Corporate to 20 percent), and then increasing it to 85 percent (while decreasing Corporate to 10 percent). In both cases, the Retail business remains more profitable than Corporate, with the relationship of ROC varying between 43/14 and 33/27. Only when Retail absorbs close to 90 percent o f these line items (and Corporate less than 5 percent) are similar returns obtained for the two business lines, which i s a highlydoubtful scenario. One feature of the methodology i s inter-business line transfers, i.e. the central role assigned to Treasury as a clearinghouse to ensure that assets and liabilities equal each other by business line. We have followed the standardpractice that any `excess' liabilities of each business line that are not used to fund that business line's assets are transferred to Treasury at the prevailing interbank rate o f CDI (and vice versa). Since Corporate has a sizable share of funding (deposits, as well as borrowings linked to directed lending) compared to its total loans, it i s the business that benefits the most from this remunerated transfer. We tested the hypothesis that these transfers are not remunerated, which would somewhat reduce Retail's profitability but would significantly reduce the profitability o f Corporate. Although this result i s not realistic (banks typically engage in remunerated transfer pricing across different business lines), it demonstrates (as mentioned in chapter 4) the dependence of Corporate on the liabilities side for its profitability. The reverse test i s to allocate completely to the respective business lines the investment gain that Treasury obtains from those funds transferred from the Retail and Corporate business lines, but this test does not have an important impact in profitability since the returns o f these investments are close to the C D Irate. Finally, we have tested the assumption used to allocate the banking system's capital base. In the aforementioned base case, we have allocated book equity to business lines according to the credit risk weights of the assets in each business line multiplied by the 77 Inparticular, the average2005 rate for non-earmarked loans to individuals (assumedto be 55 percent) and for non-earmarked domestic and foreign currency loans to mediudarge firms (assumed to be 23 and 7 percent respectively, representing a spread over Selic and US Libor of 4 and 3.5 percent respectively). 52 average banking system capital adequacy o f 15 percent, with all excess book equity remainingin Treasury. This figure goes inthe denominator of the ROC equation and thus helps to determine profitability by business line. For sensitivity analysis purposes, we performed two tests of this assumption: first, that the distribution o f equity i s proportional to the size o f each business line, and second, that the allocation i s based on Bacen's minimum capital adequacy ratio of 11 percent (with Treasury acting as the residual, absorbing excess capital). The ratio o f ROC for RetaiVCorporate moves from 38/21 inthe base case to 41/21 and 45/21 inthese two tests. Overall, it seems likely that the aforementioned messages stemming from the analysis (Le. higher returns in the Retail business line when compared to those of Corporate) are valid under most sensitivity scenarios, although more information i s required in order to ensure the soundness o f certain critical assumptions. 53 Table 111.la: Sensitivity Analysis inthe Performance Estimation Retail Corporate Government Treasury Base Case Market Share of Deposits 58% 35% 3% 4% Market Share of Gross Loans 46% 50% 4% 0% Market Share of Loan Loss Provisions (LLP) 63% 31% 6% 0% Market Share of InterestExpense 34% 22% 2% 41% Pre-tax Return on Assets 4.1% 2.3% 1.8% 1.7% Pre-tax Return on Capital 38% 21% 16% 24% Sensitivityto Deposits Description:10% of Retaildepositsshift to Corporate Market Share of Deposits 48% 45% 3% 4% Pre-taxReturn on Assets 4.0% 2.8% 1.8% 1.5% Pre-taxReturn on Capital 34% 28% 16% 21% Description:10% of Corporatedepositsshift to Retail Market Share of Deposits 68% 24% 3% 4% Pre-taxReturn on Assets 4.2% 1.6% 1.8% 1.8% Pre-taxReturn on Capital 42% 12% 16% 26% Sensitivityto Deposit Rates Description:Interestrate on time deposits increasedto 18% (from l6Y0)and interest rate on acceptances and on reverserepos reducedto 4% (from 8%) and 16 % (from l8Y0)respectively,such that overall interest expenses reportedby Bacen remain the same Market Share of InterestExpense 37% 25% 3% 36% Pre-taxReturn on Assets 3.7% 1.7% 0.8% 2.7% Pre-taxReturn on Capital 35% 15% 7% 38% Description:Interestrate on time depositsdecreasedto 16% and interest rate on acceptancesand other deposits increasedto 12%,such that overall interestexpensesreported by Bacen remainthe same Market Share of Interest Expense 32% 20% 2% 46% Pre-taxReturn on Assets 4.5% 2.9% 2.9% 0.6% Pre-taxReturnon Capital 41% 26% 25% 9% Sensitivityto Gross Loans Description:All loans to small firms (exceptdirected rural loans) are included in the Corporatebusiness line, as opposedto the Retail businessline under the base case Market Share of Gross Loans 40% 56% 4% 0% Pre-taxReturn on Assets 4.2% 2.2% 1.8% 1.7% Pre-taxReturn on Capital 41% 18% 16% 24% Description:The proportionof loansto firms that belongto the Retail business line is raised to 25% of the total (exceptfor directed rural loans),as opposedto 15% under the base case Market Share of Gross Loans 50% 46% 4% 0% Pre-taxReturn on Assets 4.1% 2.3% 1.7% 1.6% Pre-taxReturnon Capital 37% 21% 15% 22% Sensitivityto Loan Loss Provisions(LLP) Description:The LLP rate for non-earmarkedCorporateloans is increasedto 7% and for Retail loansto small firms decreasedto 8%, such that the overall LLPrate of 10.2% reported by Bacen remains the same MarketShare of LLP 61% 34% 6% 0% Pre-taxReturn on Assets 4.2% 2.2% 1.8% 1.7% Pre-tax Return on Capital 39% 19% 16% 24% Description:The LLP rate for non-earmarkedCorporateloans is decreasedto 5% and for Retail loansto small firms increasedto 16%, such that the overall LLPrate of 10.2% reported by Bacen remains the same Market Share of LLP 66% 29% 6% 0% Pre-tax Returnon Assets 4.0% 2.5% 1.8% 1.7% Pre-tax Returnon Capital 37% 22% 16% 24% 54 Table 111.lb: Sensitivity Analysis inthe PerformanceEstimation (cont.) Retail Corporate Government Treasury Base Case Market Share of Interest Revenue on Loans 69% 29% 3% 0% Market Share of Interest Expense 34% 22% 2% 41% Market Share of Service Charges & Op. Expenses 80% 15% 1% 4% Market Share of CapitaVEquity 46% 30% 2% 22% Pre-tax Return on Assets 4.1% 2.3% 1.8% 1.7% Pre-tax Return on Capital 38% 21Yo 16% 24% Sensitivityto Loan Rates Description:The interestrate on non-earmarkedlendingto individuals increasedto 58%. while non- earmarked domestic and foreign currency lendingto Corporates decreasedto 20% and 4% respectively, such that the overall interest incomeon loans reportedby Bacen remainsthe same Market Share of Interest Revenue on Loans 72% 26% 3% 0% Pre-tax Return on Assets 4.7% 1.3% 1.8% 1.7% Pre-tax Return on Capital 44% 11% 16% 24% Description:The interestrate on non-earmarkedlendingto individuals decreasedto 52%. while non- earmarked domesticand foreign currencylending to Corporates increasedto 26% and 10% respectively, such that the overall interest income on loans reported by Bacen remainsthe same Market Share of Interest Revenue on Loans 66% 32% 1.a% 3% 0% Pre-tax Return on Assets 3.4% 3.4% 1.7% Pre-tax Return on Capital 32% 30% 16% 24% Sensitivityto Allocation of ServiceCharges, Operating Expensesand Other Items Description:The share of the Retailbusinessline in service charges, operatingexpensesand all other line items that are allocatedsimilarlydrops by 5% to 75%,while the share of Corporate rises by 5% to 20% Market Share of Service Charges & Op. Expenses 75% 20% 1.a% 1% 4% Pre-tax Return on Assets 4.6% 1.6% 1.7% Pre-tax Return on Capital 43% 14% 16% 24% Description:The share of the Retailbusiness line in service charges, operatingexpensesand all other line items that are allocatedsimilarlyincreasesby 5% to 85%, while the share of Corporatedrops by 5% to 10% Market Share of Service Charges & Op. Expenses 85% 10% 1% 4% Pre-tax Return on Assets 3.6% 3.1% 1.8% 1.7% Pre-tax Return on Capital 33% 27% 16% 24% Sensitivityto Inter-BusinessLine Transfers Description:Treasury, in its central clearing role, does not remunerateexcess transfersto/from it in response to an imbalancebetweenassets and liabilitiesin a specific business line Pre-tax Return on Assets 3.0% 0.1% 0.7% 5.1% Pre-tax Return on Capital 28% 1Yo 6% 72% Description:Treasury, in its centralclearing role, remuneratesexcess transfers to/from it in responseto an imbalancebetweenassets and liabilities in a specific business line at its own active rate instead of CDI Pre-tax Return on Assets 4.1% 2.2% 1.8% 1.8% Pre-tax Return on Capital 38% 20% 16% 25% Sensitivityto Capital Description:The book equity of the banking system is allocated pro rata to the business linesaccordingto their size insteadof accordingto regulatoryrequirementsfor credit risk bytype of asset Market Share of Capital 39% 23% 2% 36% Pre-tax Return on Assets 3.9% 1.9% 1.5% 2.4% Pre-tax Return on Capital 41% 21% 15% 22% Description:The book equity of the bankingsystem is allocated based on Bacen'sminimumcapital adequacy ratio of 11% (with Treasury acting as the residual), as opposedto based on the average capital adequacy of the banking system (15%) Market Share of Capital 33% 22% 2% 43% Pre-tax Return on Assets 3.7% 1.8% 1.3% 2.7% Pre-tax Return on Capital 45% 21% 15% 21% 55 References Beck, Thorsten, Asli DemirguG-Kunt andVojislav Maksimovic, 2004 "Bank Competition and Access To Finance: International Evidence" Journal of Money Credit and Banking 36 (3). Belaisch, Agnes, 2003, "Do Brazilian Banks Compete?" IMF Working Paper 03/13, Washington, D.C. Berger, Allen N and Timothy H. Hannan. 1989. "The Price-Concentration Relationship inBanking." Review of Economics and Statistics 71, 291-99. Berger, Allen N. 1995 "The Profit-Structure Relationship in Banking - Tests of Market Power and Efficient-Structure Hypothesis." Journal of Money, Credit, and Banking 27,404-3 1. Berger, Allen N., Rebecca S. Demsetz, and Philip E. Strahan. 1999. "The Consolidation of the Financial Services Industry: Causes, Consequences, and Implications for the Future." Journal of Banking and Finance 23 (February), 135-94. Berger, Allen N, Leora F.Klapper, and Gregory F. Udell, 2001. "The Ability of Banks to Lend to Informationally Opaque Small Businesses," Journal of Banking and Finance 25. Bresnahan, Timothy F. 1982. "The Oligopoly Solution Concept I s Identified." Economics Letters 10, 87-92. Cetorelli, Nicola and Michele Gambera. 2001. "Banking Market Structure, Financial Dependence and Growth: International Evidence from Industry Data." Journal of Finance 56,617-648. Cetorelli, Nicola, 2001. "Does Bank Concentration Lead to Concentration in Industrial Sectors?" Federal ReserveBank of Chicago Working Paper 2001-01. Cetorelli, Nicola. 1999. "Competitive Analysis in Banking: Appraisal of the Methodologies." FederalReserve Bank of Chicago Economic Perspectives ,2-15. Chang, E.J., Solange M. Guerra, Rodrigo Penaloza and Benjamin Tabak, 2005. "Measuring Banking Concentration: The Brazilian Case." In Financial Stability Report, Banco Central do Brasil, May. Claessens, Stijn and Luc Laeven. 2004. "What Drives Bank Competition? Some International Evidence." Journal of Money Credit and Banking 36(3), 563-83. Clark T., Dick A., Hirtle B., Stiroh K. and Williams R. 2007. "The Role of Retail Banking inthe U.S. Banking Industry: Risk, Return, and Industry Structure." Federal ReserveBoard of New York Economic Policy Review, forthcoming. Cruickshank D., 2000. "Competition in UK banking: a report to the Chancellor of the Exchequer." Ebrill L., Summers V. and Coelho I. "Bank Debit Taxes in Latin America - An 2001. Analysis of Recent Trends." IMF Working Paper 01/67, May. Falkena H.et al, 2004. "Competition in South African Banking." Task Group Reportfor the National Treasury and the SouthAfrican ReserveBank. April. FitchRatings, 2006. "The BrazilianBanking System." Country Report. July. FitchRatings, 2006. "The BrazilianPrudential Regulations." Country Report. August. Gelos, R. Gaston and Jorge Roldos, 2002. "Consolidation and Market Structure in Emerging MarketBankingSectors." ZMF Working Paper No. 02/186, November. Group of Ten. 2001. "Report of Consolidation in Financial Sector," Bank for International Settlements: Basel, Switzerland. Januky. 56 Hirtle B. And Stiroh K. 2005. "The Return to Retail and the Performance o f U.S. Banks." Federal Reserve Bank of New York StafSReport No. 233, December. Kumar, Anjali, 2004. Brazil: Access to Financial Services. Report No. 27773-BR, February. Kumar, Anjali, Nair, Ajai, Parsons, Adam and Urdapilleta, Eduardo. 2005. "Expanding Bank Outreach Through Retail Partnerships, Correspondent Banking in Brazil", WorldBank Working Paper No. 85. Lau, Lawrence. 1982. "On Identifying the Degree o f Competitiveness from Industry Price and Output Data." Economics Letters 10,93-99. McKinsey & Company, "Productividade no Brasil," 1998 McKinsey & Company. 2006. "How Brazil can grow." McKinsey Quarterly 2006 No.2. Morgan Stanley and Oliver, Wyman & Company. 2002. "The Future of European Corporate & Institutional Banking: The Need to Differentiate", March. Morgan Stanley and Mercer Oliver Wyman. 2004. "US Retail Banking and Consumer Credit: An Agenda for Growth", August. Molyneux, Phillip, John Thornton, and D. Michael Lloyd-Williams. 1996. "Competition and Market Contestability in Japanese Commercial Banking." Journal of Econometrics and Business 48, 33-45. Nakane, Marcio 2001. "A Test of Competition in Brazilian Banking", Working Paper Series, Banco Central do Brasil. Nakane, Marcio, Leonard0 Alencar and Fabio Kanczuk, 2005. "Demand for bank services and market power in Brazilian banking", Banco Central do Brasil and University o f Sao Paulo, mimeo. Oliver, Wyman & Company. 2003. "Retail Banking in Europe: Markets and Management", February. Panzar, John C., and James N. Rosse. 1987. "Testing for `Monopoly' Equilibrium." Journal of Industrial Economics 35,443-56 Rocha, Fernando, 2001. "Evolution of Banking Concentration in Brazil." Banco Central do Brasil, Technical Notes n.11 Shaffer, Sherrill. 1982. "A Non-Structural Test for Competition in Financial Markets." In Bank Structure and Competition. Conference Proceedings, Federal Reserve Bank of Chicago, pp. 225-243. Western Hemisphere Payments and Securities Clearance and Settlement Forum, 2004. Payments and Securities Clearance and Settlement Systems inBrazil. September. Western Hemisphere Credit and Loan Reporting Initiative, 2005. Credit and Loan Reporting Systems inBrazil. March. World Bank, 2001. First Programmatic Financial Sector Adjustment Loan, Report No. P7448-BR, April. World Bank, 2004. Second Programmatic Financial Sector Adjustment Loan, Report No. 24067-BR, May. World Bank, 2005. Brazil: Investment Climate Assessment, December. World Bank, 2006. Brazil: DoingBusiness. World Bank, 2006. Second Programmatic Loan for Sustainable and Equitable Growth, Report No. 36059-BR, May. World Bank, 2006. "Brazil: Interest Rates and Intermediation Spreads." Economic Sector Work ledby Augusto de la Torre and Stijn Claessens. Report No. 36628-BR, June. 57