Are Competitive Banking Systems More Stable? The Architecture of Financial System Stability: From Market Micro Structure to Monetary Policy 24 – 26 th

Are Competitive Banking Systems More Stable?

The Architecture of Financial System Stability:From Market Micro Structure to Monetary Policy

24 – 26th May, 2006, Capri, Italy

Klaus SchaeckSchool of ManagementUniversity of Southampton

Martin Cihak International Monetary FundWashington, DC

Simon WolfeSchool of ManagementUniversity of Southampton

DISCLAIMER:

This paper’s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily reflect the views of the International Monetary Fund, its Executive Directors, or the countries they represent.

Outline

(1) Literature Review

(2) Rationale/Contributions of this Research

(3) Methodological Approach

(4) Econometric Analysis

(5) Sensitivity Tests

(6) Conclusion and Future Research


Klaus Schaeck, Martin Cihak and Simon Wolfe, May 2006




Theoretical literature - diametrically opposite views

Competition = less stabilitySmith (1984), Besanko and Thakor (1993), Cordella and Yeyati (1998), Matutes and Vives (2000), Hellman et al. (2000)

Competition = more stabilityCaminal and Matutes (2002), Nagaraja and Sealey (1995), Perotti and Suarez (2002)

Complex relationship of competition and stabilityAllen and Gale (2004), Boyd et al. (2004)

Instabilities can arise in any market structureMatues and Vives (1996)

Empirical literature is characterised by studies that focus on one or two countries:

Keeley (1990) - USStaikouras and Wood (2000) – Greece and SpainBordo et al. (1995) – Canada and USHoggarth et al. (1998) – Germany and UKCapie (1995) - UK


Lack of cross-country datasets on competitive behavior made both policymakers and researchers rely heavily on concentration as a proxy for competition.

Theoretical studies:Concentration increases fragilityBoyd and de Nicolo (2005), Mishkin (1999)Concentration decreases fragilityAllen and Gale (2000, 2004), Boot and Greenbaum (1993)

Empirical studies:Concentration increases fragilityDe Nicolo and Kwast (2001), De Nicolo et al. (2004), Boyd and Graham (1991, 1996) – USConcentration decreases fragilityParoush (1995), Benston et al. (1995), Craig and Santos (1997) – USBeck et al. (2005a, 2005b) – Cross country (69 jurisdictions)



Impact of the regulatory environment is equally disputed:

Liberalization and deregulation decrease stabilityFischer and Chenard (1997)

Greater contestability and less activity restrictions increase stabilityBarth et al. (2004) Beck et al. (2005a, 2005b)

(2) Rationale



Relying on concentration as a proxy for competition gives rise to several problems

(1) The inverse relationship between competition and concentration is not empirically substantiated (Claessens and Laeven, 2004; also Beck et al., 2005a, 2005b)

(2) It propels misleading inferences and gives rise to measurement problems as the level of concentration is overstated in small countries and when the number of banks is small (Bikker, 2004).

(3) It does not measure competitive conduct of financial institutions on the marginal level. It is not derived from profit-maximizing conditions (Shaffer, 2004).

Development in the recent literature to distinguish between concentration and competition (Berger et al., 2004)

However, no study specifically tests for the relationship between banks’ competitive conduct and its implications for systemic risk in a cross-country setting.

Contributions of this study:

(1)First empirical analysis of the link between competitive conduct of banks, measured by the Panzar and Rosse

(1987) H-Statistic, and systemic risk using cross-country data (38 countries, 1980–2003).

(2)Re-examination of the relationship between concentration, competition, and crises.

(3) Methodological advancement on modelling systemic risk using a parametric duration model with time-varying covariates.

(4) Examination of the impact of the regulatory environment on the timing of systemic banking crises.



(2) Contributions




The Panzar and Rosse (1987) H-Statistic

Measures market power by the extent to which a change in factor input prices translates into equilibrium revenues by bank i.

Ri* equilibrium value of revenue of bank iwi vector of m input pricesH ≤ 0 monopoly equilibrium0 < H <1 monopolistic competition H = 1 perfect competition

Data on H-Statistic obtained from Claessens and Laeven (2004).

Measuring competition

The Panzar and Rosse (1987) H-Statistic (Shaffer, 2004):

(1) H-Statistic is analytically superior to previously used measures ofcompetition, because it is derived from profit-maximizing

equilibrium conditions.

(2) It is robust with respect to the market since it only draws upon characteristics of reduced-form revenue equations at the firm level.




Measuring banking sector stabilityWe use a dummy variable indicating a systemic banking crisis

(1) Demirgüç-Kunt and Detragiache (2005) dating scheme is used as the basic one (it is based on presence of emergency measures, large nationalizations, high NPLs, and high fiscal costs of rescues). It shows 28 systemic crises in 1980-2003 for our sample.

(2) Honohan and Laeven (2005) used as an alternative dating scheme




We model crises using duration and logit analysis.

I. Duration analysis

The dependent variable measures the time to transition from a sound banking system to a crisis episode. Since we have multiple observations per country with up to 23 time spans in the dataset, we estimate the model with time-varying covariates. Accelerated failure time models are written in the form

where ln(tj) is the log of time to failure, xj denotes our explanatory variables, βx are the parameters to be estimated and τj is a random variable that follows a distribution.

jxjj xt )ln(



Duration analysis (cont’d)

Thus, to estimate the model, we need to determine the distribution of τj and specify it to follow the exponential distribution.

We perform a robustness test based on the Weibull distribution for the baseline hazard function and also employ a Cox proportional hazards model to examine the sensitivity of our estimates to the alternative setups.

Both additional robustness tests confirm the results obtained with the model based on the exponential distribution.





II. Logit analysis

The model computes the probability that a crisis is observed:

P(i,t) … 1 if a crisis is observed, 0 otherwise. β … vector of coefficients to be estimated X (i,t) … explanatory variables

(1) (2) (3) (4) (5) (6) (7) (8) Duration models Logit probability models

GDP growth (real) -0.0691 -0.0214 -0.0184 -0.0141 0.1315 0.0787 0.0719 0.0709 (0.0803) (0.0904) (0.0909) (0.0880) (0.0817) (0.0922) (0.0935) (0.0920) Real interest rate -0.0084 -0.0103 -0.0085 -0.0067 0.0219 0.0253 0.0235 0.0231 (0.0118) (0.0139) (0.0126) (0.0131) (0.0100)** (0.0130)* (0.0130)* (0.0130)* Inflation -0.0021 -0.0018 -0.0010 0.0002 0.0053 0.0043 0.0036 0.0033 (0.0018) (0.0017) (0.0018) (0.0022) (0.0031)* (0.0037) (0.0039) (0.0038) Terms of trade change -0.0024 -0.0029 -0.0026 -0.0025 0.0002 0.0001 -0.0000 -0.0001 (0.0017) (0.0017)* (0.0017) (0.0018) (0.0027) (0.0035) (0.0032) (0.0033) Depreciation -0.0039 -0.0025 -0.0027 -0.0030 0.0002 -0.0010 -0.0008 -0.0007 (0.0015)*** (0.0017) (0.0015)* (0.0014)** (0.0023) (0.0020) (0.0019) (0.0018) M2/international reserves 0.0138 0.0187 0.0293 0.0266 0.0057 -0.0020 -0.0110 -0.0080 (0.0296) (0.0305) (0.0324) (0.0319) (0.0239) (0.0262) (0.0326) (0.0320) Credit growth (real) -0.0014 -0.0019 -0.0020 -0.0022 0.0010 0.0016 0.0017 0.0018 (0.0004)*** (0.0005)*** (0.0005)*** (0.0006)*** (0.0008) (0.0008)** (0.0008)** (0.0008)** Moral hazard index -0.0372 0.2685 0.2602 0.0527 0.3375 0.0299 0.0601 0.1363 (0.2190) (0.2772) (0.2933) (0.3306) (0.3417) (0.3690) (0.3683) (0.3984) British legal origin -0.5227 -0.3812 -0.1951 -0.6531 1.1127 1.1103 1.0450 1.2049 (0.6629) (0.7934) (0.8777) (0.9105) (0.8397) (0.9437) (0.9338) (0.9193) French legal origin -1.1209 -1.5525 -1.3319 -1.2832 1.6562 2.2597 2.1015 2.1036 (0.6180)* (0.5790)*** (0.5518)** (0.6086)** (0.6230)*** (0.6465)*** (0.5998)*** (0.6185)*** Scandinavian legal origin -1.1441 -1.0797 -1.5069 -1.2270 1.2795 1.2279 1.5485 1.4447 (0.5604)** (0.5249)** (0.7134)** (0.7192)* (1.1575) (1.1414) (1.2624) (1.2846) H-Statistic 5.0854 4.2562 -6.3020 -5.8513 -5.2130 -1.7627 (1.7922)*** (1.9919)** (8.2633) (1.7523)*** (1.8819)*** (6.2430) Concentration 2.3486 -11.8460 -1.8366 2.9196 (2.4725) (11.2600) (2.4980) (8.6863) H-Statistic * Concentration 23.9997 -7.8430 (18.7818) (13.1666)

Main Results (4) Econometric Analysis



Robust standard errors in parentheses. Significance levels of 1, 5 and 10 percent are indicated by ***, ** and *.

Findings – core results (4) Econometric Analysis



Competitive behaviour of financial institutions matters for both the timing and the probability of observing systemic banking crises, when the level of concentration, the macroeconomic environment, deposit insurance design features and the legal origin of the country are controlled for.

Our duration analysis indicates that time to crisis increases in a more competitive environment whereas the logit model suggest that competition decreases the probability of observing a crisis.

Concentration is insignificant in both the duration and the logit probability models.

Among the control variables, terms of trade, the rate of depreciation, and credit growth enter the duration model significantly.

French and Scandinavian legal origin are also found to to be significant, suggesting shortened survival time of banking systems in the respective jurisdictions. The logit model furthermore indicates that increases in realinterest rates, inflation, credit growth are good precursors forbanking crises. The dummy for French legal origin enters again the equation significantly corroborating the findings from the duration analysis.



Findings – core results (cont’d) (4) Econometric Analysis

(1) (2) (3) (4) (5) (6) (7) (8) Duration models Logit probability models … … … … … … … … H-Statistic 1.2038 5.7925 4.9358 5.3638 -3.0043 -6.1835 -5.2564 -4.3798 (1.5802) (2.3317)** (2.4638)** (2.0009)*** (2.4721) (2.2136)*** (2.1689)** (2.3654)* Concentration 2.5495 4.5866 0.3286 4.8894 0.1156 -3.8898 -1.0092 -0.8001 (3.1369) (3.2740) (2.3985) (3.4130) (2.8124) (2.7648) (1.9934) (3.0274)



Robustness tests (5) Sensitivity AnalysesWe perform a set of sensitivity analyses using both the duration and the logit model employing a different coding for the dependent variable in Regression (1) and (5), excluding low income economies in Regression (2) and (6), using first differences rather than levels for the macroeconomic control variables in Regression (3) and (7), and constrain the sampling period to the time between 1985 and 2003 in Regression (4) and (8).




While the alternative crisis definition (1) and (5) does not suggest a significant relationship between competition and systemic risk, the three other robustness tests underscore the core findings and reiterate the positive relationship between competition and time to crisis and the negative relationship between competition and the probability of observing a systemic crisis!

Concentration never assumes any meaningful level of significance!

Robustness tests (cont’d) (5) Sensitivity Analyses

(1) (2) (3) (4) (5) (6) (7) (8) Duration models Logit probability models … … … … … … … … H-Statistic 4.3392 4.4300 3.8345 3.6107 -5.7529 -5.4366 -5.3887 -4.3336 (2.3380)* (2.1107)** (3.2015) (1.9182)* (2.3421)** (2.0885)*** (2.9989)* (2.0670)** Concentration 2.1585 2.7230 1.0349 2.7449 -0.4683 -2.4233 -0.2124 -2.0229 (2.3569) (2.2381) (2.9230) (2.3386) (2.3656) (2.2404) (3.2036) (2.4477) Activity restrictions -0.1077 0.3427 (0.1042) (0.1514)** Capital regulatory index -0.1924 0.4580 (0.2313) (0.3995) Foreign ownership -2.6596 0.9981 (1.6900) (1.7759) Government ownership -0.7383 0.3385 (0.5709) (0.6596)



Competitiveness, Regulation and Crises (5) Sensitivity AnalysesWe test the impact of competitiveness on the timing and likelihood of systemic crises, whilst controlling for a set of regulatory and institutional variables.


Controlling for the effect of activity restrictions, a capital regulatory index, and government and foreign ownership of banks hardly impacts upon the degree of competitive bank behaviour on systemic risk.

The significant effect of competition on the timing of banking crises is only rendered insignificant when foreign ownership enters the equation. The logit models however suggest competition increases banking system stability across all four specifications.

Concentration remains again insignificant across all regressions.

Activity restrictions are found to enter the logit model significantly with a positive coefficient. This finding is consistent with Barth et al. (2004) and Beck et al. (2005a, 2005b).



Findings - Competitiveness, Regulation and Crises (5) Sensitivity Analyses

(1) The first empirical study on the relationship between competition, measured by the Panzar and Rosse (1987) H-Statistic, and the likelihood and timing of systemic banking crises.

(2) Higher degrees of competition in banking systems are associated with increased survival time of banking systems and go hand in hand with a decrease in the probability of systemic banking crises.

(3) Results for the timing and the probability of suffering a systemic crisis are robust to using alternative samples, alternative sampling periods, and also hold when an alternative coding for the macroeconomic variables is employed.

(4) These findings also hold when concentration is accounted for and provide empirical support to the assertion that concentration and competition are different concepts.

(5) The results obtained from both methods empirically substantiate theoretical research on the ‘competition-stability’ view in the literature.

(6) No claim, that highly competitive banking systems are free of failures.

(6) Conclusion



(1) It is worthwhile to examine whether alternative measures of competitive bank conduct support our results and which levels of competition, if any, are optimal to achieve and sustain banking system stability.

(2) Studies on the firm level, using cross-country data and controlling for the regulatory environment will help explore the linkages further.

(3) The mechanism by which competition contributes to stability has to be scrutinised. For example, an analysis of the effects of competition in the short and in the long run may yield different outcomes for stability. As a complement to the 0/1 measure of financial fragility, one could use more continuous measures, such as distance to default.

(6) Future Research



Relationship between competition and fragility has to be investigated further!

Documents

Are Competitive Banking Systems More Stable? The Architecture of Financial System Stability: From Market Micro Structure to Monetary Policy 24 – 26 th