|
Article Excerpt
WHEN BANKS COMPETE, does it make banking crises more frequent? A rationale for regulating banks is that too much competition among them could threaten the solvency of individual institutions and ultimately threaten stability of the banking system. This paper examines empirically, using data on the likelihood and timing of banking crises, whether increased competition indeed leads banks to take on greater risks than would be prudent from a systemic perspective.
Several theoretical studies argue that increased competition prompts banks to embark on riskier business strategies (e.g., Smith 1984, Keeley 1990, Repullo 2004) and that less concentrated banking systems are more prone to experience crises (e.g., Allen and Gale 2004). The main argument in this literature is the so-called "franchise value" hypothesis: greater market power allows banks to protect their franchise value by generating larger capital buffers, which makes them act more prudently and pursue low-risk strategies (Boot and Greenbaum 1993, Hellman, Murdoch, and Stiglitz 2000, Matutes and Vives 2000). Other theoretical studies reject this view and emphasize that banks in uncompetitive markets are more prone to originate risky loans that set the stage for subsequent problems in the system (Caminal and Matutes 2002). Similarly, Mishkin (1999) argues that banks in concentrated systems are more likely to be subject to regulators' "too big to fail" policies that encourage risk-taking behavior by bank managers. (1)
A number of studies have tested these competing hypotheses empirically. Using concentration as a proxy for (the lack of) competition, De Nicolo et al. (2004) illustrate that more concentrated systems are more likely to experience crises. (2) By contrast, Beck, Demirguc-Kunt, and Levine (2006, 2007) present strong evidence that concentrated banking systems are more stable. Examining empirically the relationship between competition and concentration, Claessens and Laeven (2004) find that concentration is in fact a poor proxy for competition, and that concentration and competition describe different characteristics of banking systems. Similarly, Bikker (2004) highlights that relying on concentration as a measure of competition gives rise to misleading inferences and measurement problems since concentration measures such as concentration ratios tend to exaggerate the level of concentration in small countries and are increasingly unreliable when the number of banks is small. These difficulties reflect that concentration measures computed on the country level are at best a very noisy proxy for competition, and indicate the need to employ direct measures of competition derived from bank data. (3)
Therefore, the recent literature on competition differentiates between competition and concentration (Berger et al. 2004). However, none of these studies specifically tests for the direct relationship between the risk of a crisis and competitive conduct of banks, (4) even though the aforementioned empirical studies by Beck, Demirguc-Kunt, and Levine (2006, 2007) go some way in this direction by incorporating factors that are likely to impact upon competition. (5)
The purpose of our paper is to provide improved empirical evidence on these conflicting views and findings. First, we employ the Panzar and Rosse (1987) H-statistic to gauge competition. This measure is considered superior to previously used proxies for competition since it is derived from bank-level data and accounts for bank-specific differences in production functions (Claessens and Laeven 2004). Consequently, it overcomes criticism put forward against concentration ratios as it does not require assumptions about the market. (6) Second, we use duration analysis with time-varying covariates to examine the timing of crises. Drawing upon duration analysis yields insights into the probability of observing a crisis, given that no crisis occurred in the country until that year. Further, using time-varying covariates in duration analysis accounts for multiple observations per country and can be considered to be more appropriate for the panel data structure of our data set than logit models where the estimation is run for a pooled sample (e.g., Demirguc-Kunt and Detragiache 1998, 2005). While duration analysis has been used at the micro level to estimate "time until failure" of banks (Lane, Looney, and Wansley 1986), we are not aware of any macro-level studies that draw upon this methodology. Third, our analysis reinvestigates the nexus between concentration and financial stability and explores whether concentration and competition measure different characteristics of banking systems by simultaneously incorporating variables that capture both bank conduct and bank-market structure.
Using data for 45 countries for the period 1980-2005, we find that competition reduces the likelihood of a crisis and increases time to crisis. Our results also indicate that concentration decreases the crisis probability and increases time to crisis. These results are consistent with the findings by Beck, Demirguc-Kunt, and Levine (2006, 2007) regarding the effect of concentration and reject the "franchise value" hypothesis (and the findings by de Nicolo et al. 2004 and by Boyd and de Nicolo 2005). In addition, our study substantiates the assertion by Claessens and Laeven (2004) that concentration and competition describe different characteristics of banking systems as we find independent effects arising from the two on both the likelihood and timing of systemic crises. These results are robust to a broad set of robustness checks with several alternative samples, alternative sampling periods, alternative modeling approaches and, finally, controlling for design features of the regulatory and institutional environment. Our supportive evidence for the view that more competitive systems are less susceptible to systemic crises suggests that policies that promote competition among banks may have the potential to also strengthen banking system stability.
The remainder of the paper is organized as follows. We present an exposition of our econometric approach, including calculation of the H-statistic, in Section 1. Section 2 provides an overview on the data set and summary statistics. We report the results and a variety of robustness tests in Section 3. Section 4 offers concluding remarks.
1. ECONOMETRIC APPROACH
We discuss the H-statistic in Section 1.1 and present an overview of duration analysis in Section 1.2. The logit model is briefly discussed in Section 1.3.
1.1 H-Statistic
The H-statistic discriminates between competitive, monopolistically competitive, and monopolistic markets. It has been used as a direct measure of competitive conduct in the recent literature on bank competition (Molyneux, Lloyd-Williams, and Thornton 1994, Bikker and Haaf 2002, Claessens and Laeven 2004). The main reason for its popularity is its analytical strength: unlike other (indirect) measures of competition, it is derived from profit-maximizing conditions. Moreover, it is robust with respect to the market as it only requires bank-level data, so that no assumptions need to be made about the relevant market.
The H-statistic is calculated by estimating the sum of the elasticities of reduced-form revenue equations with respect to factor input prices. Underlying this approach is the idea that market power is reflected in the extent to which changes in factor prices are reflected in revenues. In other words, the H-statistic measures the ability of a bank to pass on increases in factor input prices to customers. Panzar and Rosse (1987) prove that the sum of the elasticities of revenue with respect to input prices is negative for a monopolist, equal to 1 for a competitive price-taking firm operating in long-run equilibrium, and ranges between and 1 for monopolistic competition.
For the monopoly case, Panzar and Rosse show that an increase in input prices will increase marginal cost, reduce equilibrium output, and subsequently reduce revenues. They prove that this is a general result that requires little beyond the profitmaximization hypothesis itself. As a result, the sum of the reduced-form revenue elasticities is nonpositive, so that H [less than or equal to] (H is a decreasing function of the perceived demand elasticity).
In contrast, the H-statistic is positive for monopolistic competition and for perfect competition (H is an increasing function of the perceived demand elasticity). The revenue function of individual banks depends in such circumstances on the decisions made by either actual or potential entrants. Based on comparative static properties of the Chamberlinian equilibrium model, and assuming long-run equilibrium, Panzar and Rosse show that under monopolistic competition H [less than or equal to] 1, and under perfect competition (in which bank products are perfect substitutes of one another) H = 1. (7)
The latter result has an intuitive explanation: a perfectly competitive bank that is constrained to zero economic profit at the initial price vector will adjust output prices equiproportionally and will pass along the entire cost increase to customers to remain in business in the long run. (8)
To approximate the H-statistic empirically, we follow Claessens and Laeven (2004) and estimate reduced-form revenue equations for each country of the form
In ([P.sub.it]) = [alpha] + [[beta].sub.1] 1n([W.sub.1,it]) + [[beta].sub.2] 1n([W.sub.2,it]) + [[beta].sub.2] 1n([W.sub.3,it]) + [[gamma].sub.1] 1n([Y.sub.1,it]) + [[gamma].sub.2] 1n([Y.sub.2,it]) + [[gamma].sub.3] 1n([Y.sub.3,it])) + [delta]D + [[epsilon].sub.it], (1)
where [P.sub.it] is the ratio of interest revenue to total assets (proxy for output price), [W.sub.1,it] is the ratio of interest expenses to total deposits and money market funding (to proxy for the input price of deposits), [W.sub.2,it] is the ratio of personnel expense to total assets (proxy for the price of labor), and [W.sub.3,it] is the ratio of other operating and administrative expenses to total assets (proxy for price of fixed capital), with i denoting bank i and t denoting time t. Here, [Y.sub.1,it] is a control variable for the ratio of equity to total assets, [Y.sub.2,it] controls for the ratio of net loans to total assets, and [Y.sub.3,it] is the log of total assets to capture size effects. Also, D is a vector of year dummies for 1998-2005 (the dummy for 1998 is dropped to avoid perfect collinearity). All variables enter the equation in logs. Here, H is calculated as [[beta].sub.1] + [[beta].sub.2] + [[beta].sub.3]. To mirror the approach by Claessens and Laeven (2004), equation (1) is estimated using ordinary least squares (OLS) with time dummies and GLS with fixed effects and time dummies. They rerun equation (1) with the ratio of total revenue to total assets as dependent variable, using OLS with time dummies and GLS with fixed effects and time dummies to have a more comprehensive measure of the degree of competition, since this alternative dependent variable extends to nontraditional sources of bank revenues like fee income-generating activities. (9) We also run equation (1) with the alternative dependent variable. We average the estimates of the H-statistics obtained from the four regression setups for our subsequent analyses. Since the H-statistic assumes that the market is in equilibrium, we test for this assumption by estimating the following equation
ln([ROA.sub.it]) = [alpha] + [[beta].sub.1] 1n([W.sub.1,it]) + [[beta].sub.2] 1n([W.sub.2,it]) + [[beta].sub.3] 1n([W.sub.3,it]) + [[gamma].sub.1] 1n([Y.sub.1,it]) + [[gamma].sub.2] 1n([Y.sub.2,it]) + [[gamma].sub.3] 1n([Y.sub.3,it]) + [delta]D + [[epsilon].sub.it], (2)
whereby ROA denotes the pretax return on assets (see also Molyneux et al. 1996). We use an...
|
