Concentration and Competition in the Banking Sector: Evidence from Chile. Jean Sepúlveda-Umanzor* and Alejandra Soto P.

Transcription

1 Concentration and Competition in the Banking Sector: Evidence from Chile Jean Sepúlveda-Umanzor* and Alejandra Soto P. We thanks comments and suggestions received at the 2008 annual meeting of the Chilean Economics Association, the 2008 annual meeting of the Southern Economics Association, and the Universidad de Talca s macroeconomic seminar. The usual disclaimer applies. * Corresponding author: School of Business and Economics, Universidad del Desarrollo, Ainavillo 456, Concepcion, Chile. Phone: , Fax: , jeansepulveda@udd.cl

2 Abstract We use Panzar and Rosse methodology to estimate the degree of competition in the Chilean banking sector, for the period We estimate the H-statistic and find that in spite of rising levels of concentration and fewer banks competing in that sector, the degree of competition has intensified. Thus, in the Chilean banking sector, higher concentration levels are associated with higher degrees of competition. We also try to explain in terms of macroeconomic variables, finding that higher levels of concentration, and money market rates are associated with higher levels of competition, while output growth is related to competition in a nonlinear manner. Key Words: Panzar and Rosse, Competition, Concentration, Banking. JEL Codes: D4, G21, L11, O16.

3 Introduction During the last decade, the number of competing firms in several industries of the Chilean economy has drastically decreased. The banking sector has not been isolated from this trend, and as such it has witnessed a drop in the number of banks in the system, and an increase in concentration levels. By March of 1986, 46 banks were operating in the industry (one of them state-owned), this number reduced to 26 by the end of And while in 1986, eleven banks accounted for 75 percent of the total loans in the system, at the end of 2006, five banks brought about 76 percent of the loans in the system. Furthermore, the Herfindahl-Hirschman ( ) index has increased from 944 to 1309 during the same period. Rising concentration levels have placed banks under the scrutiny of market analysts and politicians alike. The usual argument that a more concentrated industry is associated with less competition has emerged in the country. As Levine (2000) argues ``the issue of banking sector concentration has stepped into Chile's policy spotlight." Banks play a crucial role in an economy. By funneling funds throughout the system, they are able to facilitate the intertemporal allocation of consumption and physical capital, promote economic growth and therefore enhance welfare. A banking system with low levels of competition and thus low levels of efficiency will hinder the ability of an economy to growth. This study addresses the issue of banking concentration and competition in the Chilean banking sector during the period. We investigate whether banking concentration in the Chilean economy has negatively impacted the degree of competition in that sector. We quantify market concentration by using the index and obtain a measure of competition by using the Panzar and Rosse's H-statistic.

4 The paper continues with a review of the literature on concentration and competition, section 3 offers an overview of the Chilean banking sector during the period, section 4 briefly explains the Panzar and Rosse (1987) methodology, section 5 outlines the econometric strategy, section 6 shows the data and results, and section 7 concludes. 2.- Literature Review Historically the level of concentration in a sector has been associated with the level of competition in it. Empirically, the use of indicators such as the index, the number of firms in an industry, and the different measures of market concentration respond to this logic; by measuring concentration we can infer the degree of competition. Concentrated markets have been thought to be less competitive, however, from a theoretical point of view, the link between concentration and competition is less than clear. To explain the structure-conduct-performance paradigm, the literature offers to avenues: (i) The Baumol, Panzar and Willig (1982) s theory of contestable markets argues that it is not the number of incumbents in an industry what determines competition, but the existence of potential entrants. As long as there is potential entry in a market, firms on it will behave competitively even if they are only a few of them. (ii) The efficiency hypothesis (see Demsetz, 1973) argues that concentrated industries may not be the result of collusion or lack of competition, but of more efficient firms growing at the expenses of less efficient ones. That is, concentrated industries may be more efficient and thus more competitive than fragmented industries. Since the theoretical answer to the relation between competition and concentration is ambiguous, the answer has to come case by case. To do that, Bresnahan (1982), and Panzar and Rosse (1987) derive methods, from microeconomic foundations, to study competition.

5 Bresnahan uses a market equilibrium condition to highlight that profit maximizing firms chose, in equilibrium, prices and quantities for which the marginal cost equals the perceived marginal revenue (this is the price in perfect competition, and marginal revenue in the case of a monopolist). Bresnahan shows that movements on the demand curve around the equilibrium point reveals the competitive structure of the market. He uses this fact to show that the solution to an oligopoly situation can be estimated econometrically. Shaffer (1993) applied this approach to study the Canadian banking sector for the period, and finds that although the system has concentrated; competition has increased during the period. Panzar and Rosse (1987) use comparative statics to develop a set of testable implications for a profit maximizing monopoly. They show that the sum of the factor price elasticities of the revenue equation must be nonpositive. They expand it to demonstrate that the sum of factor price elasticities may be positive for perfect competitive, monopolistically competitive, and oligopolistic environments. Applications of the Panzar and Rosse (1987) approach are Shaffer (1982) who study a sample on New York banks and finds monopolistic competition. Nathan and Neave (1989) studies the Canadian banking system and reject monopoly power. Molyneux, Lloyd- Williams and Thornton (1994) studies the European banking system, finding monopolistic competition in most countries. Claessens and Laeven (2004) examine the competitive structure of the banking industry in 50 countries, Chile among them. They find that monopolistic competition describes the banking sector in the majority of the countries.

6 The Chilean Banking Sector, During the period, the banking sector in Chile has witnessed vast transformations. By the beginning of 1986 there were 46 banks operating in the industry, of which one of them was state-owned and eleven were foreign-owned banks. By the end of 2006 the system was composed by 26 banks, of which one is state-owned and twelve are foreign-owned banks. In 2006, the Chilean banking system generated loans for about 75 percent of GDP. This is up from near 60 percent of GDP in To do that, in a period of high GDP growth, banks expanded both its client base and its portfolio of services. Foreign banks became key players in the system, and by the end of 2006 they owned more than 35 percent of the assets managed in the Chilean banking sector. The system has experimented a gradual but sustained process of concentration during the last twenty years. This trend has also been observed in other banking systems around the world. Figure 1 shows the evolution in the number of banks in the system during the period, along with the HH index. [INSERT FIGURE 1] The index shows that the decrease in the number of banks has been accompanied by an increase in concentration. The index was 944 at the beginning of 1986, and it reached 1309 in We also calculate the concentration index. 2 In 1986 this ratio was = 57% to reach = 74% by Figure 3 displays the behavior of this index 1 An index between 1000 and 1800 represents a moderately concentrated industry according to the U.S. Justice department 2 The index measures concentration by adding up the biggest market shares in the industry.

7 overtime. [INSERT FIGURE 2]. It can be seen that this ratio has consistently increased since Both, the HH index and the C-5, measures of concentration show the same picture: the Chilean banking system has concentrated during the period under analysis. What is less clear though is whether competition has been hurt because of this trend. The Panzar and Rosse (1987) Approach Panzar and Rose (1987) propose a test of competitive structure based on the following intuition: under perfect competition and long run equilibrium, an increase in a variable input's price generates an increase in marginal costs and total revenue of the same magnitude. However, if the market is dominated by a monopoly, the increase in the price of the input will make total revenue decrease because the monopoly operates in the elastic part of the demand curve. Thus, Panzer and Rosse measure the degree of competition based on the impact of a cost's increase on total revenue. If increases in costs are associated to drops in total revenue; the industry behaves like a monopolist. On the contrary, if a change in costs is translated into a change of the same magnitude in total revenue, then the industry is competitive. Panzar and Rosse develop a test, the H test, which econometrically summarizes this intuition. This approach does not need marginal costs; it only needs changes in costs at the aggregate. Data needed to perform such a test are usually found in the financial statements of firms. What follows is a brief summary of Panzar and Rosse formal proof. Define a monopoly's profit function as, where, and are the revenue function and cost function respectively. is a vector

8 of decision variables, is a vector of exogenous variables that affect revenues, is a vector of exogenous input's prices, and is a vector of exogenous variables that affect the firm's cost function. Let and where is a scalar. By the same token,, and function. Then by definition, where represents the firm's reduce revenue and thus and since is a linearly homogenous function in, and after some algebraic manipulations, it can be shown that i.e., the sum of factor price elasticities of a monopolist's reduced form equation is nonpositive. Panzar and Rosse (1987) also show that if we assume a cobb-douglas technology, then there exist a relation between the Lerner index of monopoly power and, given by. The authors then propose that ``in a Chamberlinian equilibrium, the sum of elasticities of firm's reduced form revenues with respect to factor prices is less than or equal to unity." They also show that ``for firms observed in long-run competitive equilibrium, the sum of the elasticities of reduced form revenues with respect to factor prices equal unity." Finally, if firms in an stable, symmetric, homogenous product, conjectural variation oligopoly equilibrium, the sum of input's price elasticities of the reduced form revenue function, is negative."

9 Methodology We follow Claessens and Laeven (2004) interpretation of the Panzar and Rosse approach. We estimate a panel model of the following form (1) where is interest revenue as a fraction of total assets of each bank at year, is the ratio of interest expenses to total deposits, is personnel expenses as a fraction of total assets, and represents other operative and administrative expenses as a fraction of total assets. As explained by Claessens and Laeven, those are proxies for output price for loans, input price of deposits, input price of labor, and input price for fixed capital respectively. represents control variables given by the ratio of equity to total assets ( ), net loans as a fraction of total assets ( ), and total assets ( ). In this setup, the value of (sum of factor price elasticities with respect to the reduced revenue function) gives all the information needed to conclude about the nature of competition. An indicates a monopoly, an represents monopolistic competition, and corresponds to perfect competition. Panzar and Rosse method is only correct under long-run equilibrium. Thus, we need to estimate whether the banking industry is in long-run equilibrium during the period of analysis. Following Claessens and Laeven (2004), we estimate (2)

10 where is the ratio of pre-tax profits to total assets. 3 In equilibrium, returns on banks assets should not be related to factor prices, and thus we test for. Data and Results Data come from the Chilean superintendence of banks and financial institutions. We have a panel of banks with monthly observations for the period We aggregate the data on a quarterly and yearly basis and use it for the estimation. Table 1 [INSERT TABLE 1] summarizes the data. We drop six banks from the data because they have less than four quarters (one year) of observations. During the period, all banks, with the exception of Banco del Estado (state-owned), are private institutions. We use a total of 680 annual observations and 2749 quarterly observations. We estimate equation (1) by both fixed-effect and random-effect models. We perform a Hausman test to decide on fixed or random effects. We obtain a = (pvalue=0.0087), which indicates that the fixed-effect method is to be preferred. Table 2 shows the results. [INSERT TABLE 2] The sum of the elasticities is = 0.64 (p-value=0.00), which indicates that competition in the Chilean banking sector can be described, during the whole period, as monopolistic competition. This value is similar to the one reported by Cleassens and Laeven (2004) for the period. Since we have an unbalanced panel, we check for sample selection bias by using the test of Nijman and Verbeek (1992). The results show that attrition can be described as random and thus it is not affecting our estimates (F=1.49, p-value=0.2224). 3 Since some s are negative, we take the natural log of

11 As stated above, the -statistic is only valid under long-run equilibrium. To test it, we estimate equation (2). The null that cannot be rejected with an F= 0.17 (p-value= ); thus we cannot reject long-run equilibrium. 4 We then obtain the average the estimated -statistic by using the cross-section of banks for each quarter. We s over periods of one, two, and three years. Figure 3 displays these averages. As can be seen, the one-year average is quite volatile and does not show a clear tendency. However, the two-year and three-year average s clearly show that competition in the Chilean banking sector, as measure by the -statistic, has increased over the period under analysis. [INSERT FIGURE 3] We also obtain the pair-wise correlation between the one-year-average quarterly - statistic, the number of banks, the index, and total assets (in logs) of the system during each quarter of the period. Table 3 shows such correlations. It can be seen that the lineal relation between the -statistic and the number of banks is negative and significantly different from zero at conventional levels. It can also be seen that there exist a positive and significant correlation between the -statistic and the level of assets in the system. Interestingly, although the correlation between the -statistic and the index, a measure of concentration, is positive, it is statistically not different from zero at conventional levels of significance. Thus, this measure indicates that for the period under analysis the level of concentration is not related to the degree of competition in the Chilean banking sector. [INSERT TABLE 3] As an aside, we investigate whether competition, as measured by the -statistic, can be explained by macroeconomic factors. We obtain a time series of s for the 81 quarters 4 The equation (2) coefficients estimates are available upon request.

12 comprised in the 1986:q1-2006:q1 period. With this series we estimate (by OLS) the following model (3) where is a vector of exogenous variables such as, proxied by the growth rate of the Index of Economic Monthly Activity ( ) on a quarterly basis, the index, the money market rate,, and total assets in the system,. is a vector of coefficients. Table 4 displays the results of the model with the best fit. Since the Breushc-Godfrey LM test indicates the presence of serial correlation at the four lag, the OLS estimation is carried out with robust standard errors (Newey-West). [INSERT TABLE 4] It can be seen that the relation between competition,, and output growth,, is nonlinear and statistically significant. Higher growth rates tend to positively affect competition, but at some point the effect becomes negative. One possible explanation is that when the economy is growing at low rates, banks would need to aggressively compete to sell their services. However, when the economy is growing at high speed, banks can easily sell more without the need to aggressively compete with each other., the money market rate, shows a positive relationship with competition. Interestingly, the one-year ago affects the level of competition today. One argument for this relationship could be that an increase in the increases the cost of funds for banks, and these tightened conditions remain in place four quarters later. Under

13 these conditions, the more efficient banks will be able to avoid passing this increase to consumers. In this sense, this motivates banks to compete more. To control for size during the different periods, we include the natural logarithm of total assets. The negative and significant coefficient indicates that a bigger bank sector is associated with lower levels of competition. We also include the index in the equation. The results show a positive and statistically significant relationship, which again seems to indicate that for the period under analysis concentration has not resulted in lower levels of competition, but the opposite. Conclusions We estimate the Panzar and Rosse (1987) -statistic to measure competition in the Chilean banking system and find that for the period , it can be described by monopolistic competition. We then estimate the same measure in a quarterly basis. Averaging of those values over periods of one, two, and three years show that the degree of competition in the Chilean banking system has increased. This, in spite of the lower number of banks competing in the sector and the higher levels of concentration. We then explain the Hs, and find that GDP growth affects competition in a nonlinear way, at lower levels it is associated with higher levels of competition, but at some point GDP growth begins to hurt it. Furthermore, the money market rate is related to competition in a positive way, that is increases in the money market rate are associated with higher s. Concentration, measured by the index, shows a positive association with competition, which reinforces the notion that in the Chilean banking sector concentration has been associated with higher levels of competition.

14 References Baumol, William., John C. Panzar, and Robert D. Willig Contestable Markets and the Theory of Industry Structure, Harcourt Brace Jovanovich. Bresnahan Timothy The Oligopoly Solution Concept is Identified, Economic Letters, 10, Claessens, Stijn., and Luc Laeven What Drives Bank Competition? Some International Evidence, Journal of Money, Credit and Banking, 36: Demsetz, Harold "Industry Structure, Market Rivalry, and Public Policy," Journal of Law & Economics, University of Chicago Press 16(1), pages 1-9. Molyneux, Phil, Michael Lloyd-Williams, and John Thorton Competitive Conditions in European Banking, Journal of Banking and Finance 18, Levine, Ross Bank Concentration: Chile and International Comparisons, Central Bank of Chile Working Paper 62. Nathan, Alli, and Edwin H. Neave Competition and Contestability in Canada s Financial System: Empirical Results, Canadian Journal of Economics 22, Nijman, Theo, and Marno Verbeek Nonresponse in Panel Data: The Impact on Estimates of a Lyfe Cycle Consumption Function, Journal of Applied Econometrics 7(3), Panzar, John.C., and James Rosse Testing for Monopoly Equilibrium, Journal of Industrial Economics 35, Shaffer, Sherrill A Nonstructural Test for Competition in Financial Markets," In Bank Structure and Competition, Conference Proceedings, Federal Reserve Bank of Chicago. Shaffer, Sherrill "A Test of Competition in Canadian Banking," Journal of Money, Credit, and Banking 25 (1),

15 Figure Legends Figure 1: Number of Banks and the Herfindahl-Hirschman Index, Figure 2: C-5 (Loans) Concentration Index, Figure 3: H-Statistic for Different Aggregation Periods,

16 Figure 1 Herfindahl-Hirschman Index Number of Banks and the Herfindahl-Hirschman Index year Number of Banks Herfindahl-Hirschman Index Number of Banks

17 Figure 2 C-5 (Loans) Concentration Index C year C-5 measures the market share of the first five commercial banks

18 Figure 3: H-Statistic for Different Aggregation Periods Year H two-year-period H four-year-period H three-year-period

19 Table 1: Data summary Variable Obs Mean Std. Dev. Min Max Interest revenue to assets Interest expenses to assets Personnel expenses to assets Operative and Adm. expenses to assets Equity to assets Loans to assets Total assets ROA

20 Table 2: Panel estimation of H-Statistics (annual data) Fixed Effects Random Effects F1 F (0.000)*** (0.846) (0.000)*** (0.027)** F (0.000)*** C (0.365) C (0.000)*** (0.001)*** (0.006)*** (0.002)*** C Observations/groups R-2 H-Statistic P-value F-test (H=0) P-value F-test (H=1) (0.815) 411/ (0.000) (0.000) (0.722) 411/ (0.000) (0.000) Chi-sq p-value Chi-Sq. NV attrition F test P-value F-test (0.0087) 1.49 (0.2224) Values in parenthesis are p-values. **, ***, significant at the 5% and 1% level respectively. Chi-sq. corresponds to the the Hausman test for fixed versus random effects. NV attrition F test corresponds to the Nijman and Verbeek test for bias selection.

21 Table 3: Correlation Matrix H-Statistic and HHI for different aggregation periods. H 1 H # Banks HHI Assets # Banks (0.0073)*** HHI (0.1723) (0.0000)*** Assets (0.0077)*** (0.0000)*** (0.0000)*** Numbers in brackets are p-values for the null hypothesis that the cross-correlation is zero. ***: Significant at the 1% level.

22 Table 4: Explaining H in terms of macroeconomic variables (quarterly data) Coefficient p-value CONSTANT * IMACEC ** IMACEC *** MMR *** LN (TA) * HH ** Adj.R F Obs. 73 *, **, ***, significant at the 10%, 5%, and 1% level respectively.