Financial Liberalization and Banking Crises

Transcription

1 Financial Liberalization and Banking Crises Choudhry Tanveer Shehzad a and Jakob De Haan a,b1 a University of Groningen, The Netherlands b CESifo, Munich, Germany September 2008 Abstract We examine the impact of financial liberalization on systemic and non-systemic banking crises, using new financial liberalization measures for a sample of developing and developed countries for the period 1981 to In contrast to conventional wisdom, our multivariate (two stage) probit modeling results consistently suggest that financial liberalization reduces the likelihood of systemic crises. In various sensitivity tests, these results turn out to be very robust. However, there is some evidence that the likelihood of non-systemic crisis increases after financial liberalization. JEL Classifications: E44, G21, G28, F36 Keywords: Banking Crises, Financial Liberalization, Financial Fragility Corresponding author : C.T.Shehzad, Faculty of Economics and Business, University of Groningen, PO Box 800, 9700 AV Groningen, The Netherlands; c.t.shehzad@rug.nl. 1 We are highly grateful to Abdul De Guia Abiad from the International Monetary Fund for his generous permission to let us use his data on financial liberalization. We also thank Axel Dreher for providing us his data on IMF and World Bank programs.

2 1. Introduction Financial liberalization is generally believed to improve financial sector development, which, in turn, will enhance economic growth. However, some authors argue that liberalization induces risk-taking behavior and may cause banking crises (Demirgüç-Kunt and Detragiache, 1998, 2000; Mehrez and Kaufmann, 2000). However, the financial liberalization data used in these studies was quite limited and rather subjective. Using better data, we argue that financial liberalization reduces the likelihood of systemic banking crises. Our indicator of financial liberalization is based on the data set of Abiad et al. (2007) indicating the extent to which a financial system is liberalized. It is an extended and updated version of the database as used by Abiad et al. (2005), covering various dimensions of the financial system. The measures relate to the presence of (i) credit controls and reserve requirements, (ii) interest rate controls, (iii) entry barriers, (iv) state ownership in the banking sector, (v) capital account restrictions, (vi) prudential regulation and supervision of the banking sector, and (vii) securities market policy. Employing this data set, we analyze the impact of financial liberalization on systemic and non-systemic banking crises. Our data on banking crises come from Honahan and Laeven (2005). We analyze systemic and non-systemic banking crises in 33 countries during the period 1981 to Our research questions are: (1) does financial liberalization affect the likelihood of a banking crisis, and if so, are there differences among the various dimensions of financial liberalization that we distinguish, and (2) are systemic and non-systemic crises affected in the same way by financial liberalization? Our results suggest that liberalization reduce the likelihood of systemic crises. In various sensitivity tests, these results turn out to be very robust. In contrast, there is some evidence that the likelihood of non-systemic crisis increases after financial liberalization. The rest of the paper is organized as follows. Section 2 provides a discussion on the determinants of banking crises and their link with financial liberalization and introduces our measures for financial liberalization and banking crises. Section 3 describes the empirical specification and explanatory variables used and section 4 reports the results for systemic and non-systemic crises. Section 2

3 5 discusses possible endogeneity issues, while section 6 offers some further robustness checks. Finally, section 7 concludes the paper. 2. Financial Liberalization and Banking Crisis In their pioneering study, Demirgüç-Kunt and Detragiache (1998) analyze the empirical relationship between banking crises and financial liberalization using data from for 53 countries. Their findings suggest that banking crises are more likely to occur in liberalized financial systems. They also find that the impact of financial liberalization on a fragile banking sector is weaker where the institutional environment is strong. However, the indicator of financial liberalization of Demirgüç-Kunt and Detragiache (1998) can be criticized as they took the first year in which some interest rates were liberalized as the beginning date of financial liberalization. Though interest rate liberalization is important, it is quite a narrow definition of financial liberalization, covering only a minor part of financial sector reform. Mehrez and Kaufmann (2000) examine how absence of corruption (i.e., transparency ) affects the probability of a financial crisis. Using multivariate probit modeling for 56 countries during , they report a higher probability of a crisis following financial liberalization during the following five years. Moreover, they also find that crisis probability is higher in countries with poor transparency than in countries that are transparent. Mehrez and Kaufmann (2000) also show that the results of Demirgüç-Kunt and Detragiache (1998) are sensitive to sample selection and the definition of financial liberalization used. They provide their own dating of financial liberalization and construct their liberalization measure on the basis of these dates. Focusing on the link between currency and banking crises, Kaminsky and Reinhart (1999) analyze 76 currency crises and 26 banking crises for 20 countries during 1970 to mid One of their main findings is that financial liberalization often precedes banking crises. Their proxy for financial liberalization is two-year lagged domestic credit growth. However, Demirgüç-Kunt and Detragiache (2000) show that a multivariate logit model of banking crises probabilities results in lower type-i and type-ii errors than the Kaminsky and Reinhart (1999) approach. 3

4 On the basis of a panel analysis, Caprio and Martinez (2000) find that government ownership of banks increases the likelihood of banking crisis. However, Barth et al. (2004) using cross-country analysis, do not find that government ownership is significantly associated with increases in bank fragility once they control for the regulatory and supervisory environment. There are also some papers that do not explicitly include financial liberalization as a potential determinant of banking crises. A good example is the recent study by Beck et al. (2006) who examine the impact of national bank concentration, bank regulations, and national institutions on the likelihood of a country suffering systemic banking crises. They use data from 1980 to 1997 for 69 countries and report that crises are less likely in economies with more concentrated banking systems. Moreover, they find that regulatory policies and institutions that discourage competition are associated with greater banking system fragility. The studies discussed above use different indicators of banking crises. Our indicator of banking crises is based on the Honohan and Laeven (2005) dataset that updates the work by Caprio and Klingebiel (1999), distinguishing between systemic and non-systemic banking crises that have occurred since the late 1970s. 2 This database is one of the most comprehensive banking crises databases. In our analysis of the relationship between (systemic and non-systemic) banking crises and financial liberalization we use a sample of 33 countries during 1981 to This selection is primarily dictated by the availability of the financial liberalization index, to be discussed hereunder, and the availability of control variables. 3 Table A- 1 in the Appendix identifies the years in which the countries in our sample had a crisis. Our data on financial liberalization come from Abiad et al. (2007) who distinguish seven dimensions of the extent to which the financial sector has been liberalized that are graded on scale from 3 (fully liberalized) to 0 (not liberalized). Apart from distinguishing between different dimensions of financial liberalization, the database has the advantage that it allows for policy reversals. The first dimension of liberalization refers to credit controls and excessively high reserve 2 Caprio and Klingebiel (1999) define a systemic banking crisis as a crisis in which much or all bank capital been exhausted. Honohan and Laeven (2005) use the same definition. 3 Abiad et al. (2007) provide data from 1973 to 2002 for 42 countries. However, various data series in the World Banks World Development Indicators and the International Monetary Funds International Financial Statistics Database are only available from 1981 onwards and not for all countries, restricting our dataset to 33 countries. 4

5 requirements (referred to as credit controls henceforth) focusing on the presence of specific credit ceilings or floors and level for reserve requirements. The second dimension is about interest rate controls examining whether they are administered by government and whether there are floors, ceilings or bands present. The third dimension is entry barriers, which is based on licensing requirements and restrictions on geographical outreach activities. The fourth dimension covers state ownership in the banking sector, i.e., the share of the assets of the banking sector controlled by state-owned banks. The fifth dimension refers to capital account restrictions and other restrictions on international capital flows. The sixth dimension captures prudential regulations and supervision of the banking sector, including compliance with the Basel standards, and executive influence on the banking supervisory agency. The final dimension refers to securities market policy covering the auctioning of government securities, debt and equity market development, and openness to foreign investors. Table 1 shows the averages of these measures over the period of our analysis, i.e., , while Table 2 shows the correlation coefficients. Table 2 makes clear that the various dimensions of the extent to which a financial system has been liberalized clearly differ from one another. Overall financial liberalization has also been taken from Abiad et al. (2007) and consists of the sum of the scores of the various liberalization dimensions. In our empirical analysis we take the change of the various measures as our indicators of financial liberalization. 5

6 YEAR Liberalization TABLE 1. LIBERALIZATION MEASURES (MEANS), Credit Restrictions Interest Rate Control Entry Barriers Supervision State Ownership Capital Flows Securities Policy TABLE 2. CORRELATION COEFFICIENTS Interest Capital Credit Rate Flows Restrictions Control Entry Barriers State Ownership Securities Policy Liberalization Supervision Liberalization 1.00 Entry Barriers Capital Flows Credit Restrictions Interest Rate Control State Ownership Supervision Securities Policy

7 3. Model Specification To analyze the impact of financial liberalization on systemic and non-systemic banking crises, we employ a panel data probit model as given in equation (1). On the LHS of the model are dummy variables representing systemic and non-systemic banking crises, respectively. The dependent variable P(i,t) takes a value of 1, if there is a crisis and zero if there is no crisis. We assume that the probability that a crisis will occur at a time t for country i is a function of a vector of n variables denoted by X(i,t), that includes our financial liberalization variables and various control variables. β is a vector of unknown coefficients and F(β X(i,t)) is the cumulative probability distribution function. The log-likelihood function of the model can be written as: Ln L = Σ t=1 T Σ i=1 n [P(i,t) ln {F(β X(i,t))} + (1- P(i,t)) ln {1-F(β X(i,t))}] (1) Using this model, the likelihood function is calculated by adaptive Gauss-Hermite quadrature. The coefficients reported indicate the expected change in the log odds 4 when there is a one unit change in the predictor variable with all of the other variables in the model held constant. In our empirical analysis, we take the change of liberalization as we are interested in how liberalization reforms affect the probability of crises. In our analysis, we examine the effect of any kind of liberalization reform in the previous five years on the probability of the occurrence of a crisis. However, in our sensitivity analysis, we also check for the effect of liberalization reforms over the preceding three years. Moreover, as we mentioned before, Abiad et al. (2007) contains seven dimensions of liberalization. Out of these seven dimensions, one is about the supervision and prudential regulation. The authors themselves acknowledge that this measure is different from the other dimensions. A higher score in this case means better (more) regulation. So in our empirical analysis, we do not treat this as a dimension of liberalization and also exclude it in calculating the overall liberalization score. However, we include supervision as a control variable because it may affect the probability of a crisis. We include various other control variables following previous studies like Beck et al. (2006) and Demirgüç-Kunt and Detragiache (2002). These variables 4 The relationship between odds ratio and coefficient can be captured by the formula as odds ratio = exp(coefficient). 7

8 include real GDP growth, the change in the terms of trade, the rate of inflation 5 (change in CPI Index), the real interest rate, the depreciation of the exchange rate, and the ratio of M2 to foreign exchange reserves. Moreover, the two-year lagged domestic credit growth rate is taken up (Kaminsky and Reinhart, 1999). Finally, we include real GDP per capita (in US$ corresponding to 1981) to control for the level of economic development. Table 3 summarizes the control variables and Table 4 shows their sources and expected signs. Variable Mean Table 3. Summary Statistics Standard Deviation Maximum Minimum Observations Liberalization (total) Supervision Credit Controls Interest Rate Controls Barriers to Entry Privatization Capital Controls Security Market Policy Terms of Trade Real GDP growth Depreciation Inflation (Adjusted) Real Interest Rate M2/Reserve Credit to Private Sector Table A-2 in the Appendix shows the correlation matrix of the control variables and our indicators of banking crises. The table shows that the controls have low correlation so that multicollinearity does not seem to be an issue. Table 4. Control variables included Variable: Expected Sign: Source: 1 Real GDP Growth - WDI 2 Real GDP/capita Level - WDI 3 Terms of Trade +/- WDI 4 Real Interest Rate + WDI 5 Depreciation +/- WDI 6 Inflation + WDI 7 Lagged Credit Growth + WDI 8 M2 to Reserve Ratio + WDI 5 The inflation rate (p) is transformed by the formula (p/100)/(1+(p/100)) to reduce the influence of extreme observations. 8

9 4. Financial liberalization and systemic banking crisis: Empirical Findings Table 5 shows our results using the panel data probit model and occurrence of systemic crises as the dependent variable. In model I, we regress systemic banking crises on control variables only, without using any liberalization measure. Our findings are in line with those of previous studies and the estimated coefficients in accordance with the expected signs as shown in Table 4. However, only three variables (GDP growth, GDP/capita, and depreciation) turn out to be significant at the 5% level of significance. In model II, we introduce our indicator of overall liberalization reforms, while in the remaining columns of Table 5 we include the various dimensions of financial reform. As suggested by Mehrez and Kaufmann (2000), we examine the impact of liberalization steps taken in the preceding five years. It turns out that overall financial liberalization has a negative impact on the likelihood of systemic crises. Next, we introduce our six liberalization measures separately into the regressions for systemic crises one by one. We observe that reforms come up significant except for barriers to entry and securities market reforms. Moreover, all liberalization measures have negative signs. All these results suggest that liberalization reduces the likelihood of systemic crises. The Wald chi-square tests and Likelihood ratio tests indicate joint significance of our models at the 1% level of significance. In the next step, we use non-systemic crises as our dependent variable. There are 14 episodes of non-systemic crises in our sample of 33 countries. Modeling non-systemic crises is a difficult task for two reasons. First, there are many factors that can cause non-systemic crises, and secondly, it is not necessary that these crises occur because of changes in macroeconomic or overall financial system variables. However, financial liberalization can be one of the causes for these crises. Table 6 presents our multivariate probit regression results for nonsystemic crises. The results indicate that the impact of liberalization is not as strong as for systemic crises. The Likelihood Ratio test is significant for all the specification at 1% level of significance, but the Wald tests are not significant even at 10% level of significance. This indicates the limited ability of a macroeconomic model to explain non-systemic crises. 9

10 According to the results, financial liberalization increases the likelihood of a non-systemic crisis, a result primarily driven by reform aimed at reducing entry barriers and credit control reforms. Interestingly, this result is opposite to our findings for systemic crises. Moreover, we also observe that real GDP growth and Real GDP per capita, which reduce the likelihood of systemic crises, appear insignificant here. 5. Endogeneity The results presented in the previous section can be subject to endogeneity because it is possible that supervisors liberalize or repress their financial systems in the wake of crises. To test for this problem of endogeneity, we use a two-step probit model with endogenous repressors. 6 For systemic crises, we use two instrument variables. The first one is from the economic freedom index dataset from the Fraser institute (Gwartney and Lawson, 2008). The economic freedom index data is available from 1970 onwards and has several dimensions of economic freedom like size of government (expenditure, taxes and enterprises), legal structure and security of property rights, access to sound money, freedom to trade internationally and regulation of credit, labor and business. We drop those dimensions of the economic freedom index that are very similar to our financial liberalization measures. The basic intuition for using this proxy is that financial sector reforms are often part of a broader economic reform program. Secondly, we use the openness of the economy (computed as the sum of exports and imports as a percentage of real GDP) as an instrument. We average both instruments over five years. 6 We implement the two-step probit model with endogenous regressors by IVPROBIT module of STATA and use robust standard errors for the clustering over countries. 10

11 Real GDP Growth (t-1) Table 5. Systemic Crisis: Panel Data Probit Model *** *** *** *** *** *** *** *** SE Real GDP Level ** ** ** ** ** ** ** ** SE Real Interest Rate ** 0.008* 0.010* ** 0.010** 0.008* SE Inflation SE M2/Reserves * * ** * * SE Terms of Trade (t-1) Credit to Private Sector (t-2) SE SE Depreciation 1.791** ** 1.581** 1.825** 1.348* 1.437* 1.596** SE Supervision SE Liberalization Reforms Credit Control Reforms Interest Rate Control Reforms *** SE ** SE *** SE 0.11 Barrier to Entry Relaxation SE Privatization *** SE Capital Control Relaxation *** SE Securities Market Reforms SE Constant * * SE Number of Observation Wald Chi- Square *** *** *** *** *** *** *** *** LR-Ratio Test 71.61*** 83.93*** 72.28*** 93.49*** 70.64*** 78.91*** 74.65*** 70.89*** stands for coefficient value and SE for Standard Errors *** indicates significance at 1% level whereas ** indicates significance at 5% and * indicates significance at 10% level 11

12 Table 6. Non-systemic Crisis: Panel Data Probit Model Real GDP Growth (t-1) SE Real GDP Level 0.110* 0.100* 0.104* 0.102* 0.111* 0.111* 0.111* 0.109* SE Real Interest Rate SE Inflation M2/Reserves SE *** ** ** ** *** 0.056*** 0.056*** ** SE Terms of Trade (t-1) Credit to Private Sector (t-2) SE SE Depreciation SE Supervision SE Liberalization Reforms 0.062* SE Credit Control Reforms 0.307*** SE Interest Rate Control Reforms 0.1 SE Barrier to Entry Relaxation 0.279** SE Privatization SE Capital Control Relaxation SE Securities Market Reforms 0.349* Constant SE *** 1.834*** 2.060*** 1.815*** *** 1.894*** 1.843*** 1.906*** SE Number of Observation Wald Chi- Square * LR-Ratio Test 98.01*** 92.59*** 96.44*** 94.68*** *** 99.02*** 96.99*** 96.57*** stands for coefficient value and SE for Standard Errors *** indicates significance at 1% level whereas ** indicates significance at 5% and * indicates significance at 10% level 12

13 We check the validity of our instruments by the Amemiya-Lee-Newey minimum chi-square test under the null hypothesis that the used group of instruments is valid, i.e., they are uncorrelated with the error term in the structural equation. As shown in the bottom panel of Table 7, we cannot reject the null hypothesis. Next we apply the Wald test of exogeneity under the null hypothesis that the instrumented variable is exogenous. The results as shown in Table 7 suggest that only reforms aimed at reducing barriers to entry in banking system appear endogenous. Using instruments to deal with this endogeneity makes its coefficient negative and significant. The results for the other dimensions of liberalization remain similar to those reported earlier. Similarly, we also instrument liberalization measures for non-systemic crises by our economic freedom index. In this case, openness to trade turns out to be an invalid instrument as indicated by the Amemiya-Lee-Newey minimum chisquare test. As an alternative, we use the participation in an IMF or World Bank adjustment programs during the preceding five years as instrument for liberalization reforms for non-systemic crises. 7 We check the validity of instruments again by Amemiya-Lee-Newey minimum chi-square test and endogeneity by the Wald test. The results as shown in Table 8 indicate that the two instruments are valid. Interestingly, the results show that most of the reform measures used here are endogenous to non-systemic crises. It shows that often liberalization measures are taken as a consequence of non-systemic crises. Here the models are jointly significant as shown by the Wald chi-square and likelihood ratio tests. Still, our macroeconomic variables remain insignificant. Our liberalization measures are the only significant variables and all come up with positive signs, which is clearly in contrast to our results for systemic crises. We therefore conclude that the effect of liberalization on systemic and non-systemic crises is fundamentally different. 7 The data comes from Dreher, who has used them in various publications; see, for instance, Dreher (2004). 13

14 Table 7. Systemic Crisis Results with Two-Step Probit Model with Endogenous Regressors Real GDP Growth (t-1) *** *** *** *** *** *** * SE Real GDP Level SE Real Interest Rate SE Inflation SE M2/Reserves *** ** *** * ** ** ** SE Terms of Trade (t-1) SE Credit to Private Sector (t-2) * SE Depreciation 1.791** SE Supervision * * ** SE Liberalization Reforms * SE Credit Control Reforms ** SE Interest Rate Control Reforms * SE Barrier to Entry Relaxation *** SE Privatization * SE Capital Control Relaxation ** SE 0.33 Securities Market Reforms ** SE 0.53 Constant SE Wald Chi-Square 9.429*** 58.4*** 56.28*** 81.08*** *** 68.68*** *** 84.84*** Wald Test of Exogeneity P(Wald test of Exogeniety) Amemiya-Lee- Newey Test P(Amemiya-Lee- Newey Test) stands for coefficient value and SE for Standard Errors *** indicates significance at 1% level whereas ** indicates significance at 5% and * indicates significance at 10% level 14

15 Table 8. Non-systemic Crisis Results with Two-Step Probit Model with Endogenous Regressors Real GDP Growth (t-1) ** ** ** * *** SE Real GDP Level SE Real Interest Rate *** ** * ** * *** ** SE Inflation ** ** SE M2/Reserves SE Terms of Trade (t- 1) ** SE Credit to Private Sector (t-2) SE Depreciation SE Supervision SE Liberalization Reforms 0.201** SE Credit Control Reforms 0.919*** SE Interest Rate Control Reforms 0.734*** SE Barrier to Entry Relaxation 1.074*** SE Privatization 1.105*** SE 0.28 Capital Control Relaxation 0.766** SE 0.32 Securities Market Reforms 1.453*** SE Constant *** *** *** *** * ** *** SE Wald Chi-Square 67.62*** 69.51*** 167.2*** *** 82.61*** 64.99*** *** Wald Test of Exogeneity P(Wald test of Exogeniety) Amemiya-Lee-Newey Test P(Amemiya-Lee- Newey Test) stands for coefficient value and SE for Standard Errors *** indicates significance at 1% level whereas ** indicates significance at 5% and * indicates significance at 10% level 6. Robustness We examine the robustness of our results for systemic and non-systemic crises results in a number of ways. These tests indicate that our results are not sensitive to 15

16 changes in estimation methodology, sample composition, and modification of certain variables. First, we have used a conditional logit model. While this does not affect our results, the number of observations drops from 577 to 397. Arellano and Hahn (2007) and Green (2004) show that the probit estimator is not well behaved in the presence of fixed effects. However, when we check for the impact of fixed effects in our logit model it turns out that our main results do not differ in both random and fixed effects models. 8 Moreover, we have also generated bootstrapped standard errors. Despite some changes in the significance level it goes without affecting our main conclusions. Secondly, we have restricted our sample to only non-oecd countries. It reduces our number of observations from 577 to 443. The effect of liberalization is significant and negative for systemic crises (except for barriers to entry) and positive for non-systemic crises, but significant only for credit market liberalization for non-systemic crises. The results for systemic crises are provided in Table A-3 in the Appendix. Thirdly, we modify our liberalization variables and average them over three (instead of five) years. The results (available on request) are qualitatively very similar to our previous findings. As part of our sensitivity analysis, we also use two other variables that have been suggested to affect banking crises, namely transparency (as suggested by Mehrez and Kaufmann, 2000) and concentration (as suggested by Beck et al., 2006). Transparency (i.e., lack of corruption) has been taken from the International Country Risk Guide (ICRG) database; the variable takes a higher value if there is less corruption in the country concerned. As this variable is only available from 1984 onwards, our analysis is restricted by this limitation. Concentration has been taken from the World Bank s Financial Structure Database. However, this series is also not available for our full sample period, reducing the number of observations by 50%. Both of these modifications do not change our basic results. All the liberalization variables retain their sign and significance for both systemic and nonsystemic crises. Inclusion of concentration does not affect our main results again except for a minor anomaly for entry to barrier reforms which comes out with a 8 For this we build comparable regression models, by excluding GDP/capita from the random effects model and then apply conditional fixed effect logit model. 16

17 positive sign for systemic crises but as mentioned before this is the only variable which was shown as endogenous. All other liberalization measures come out with a negative sign. Inclusion of transparency does not change our conclusions although transparency comes up significantly with a positive sign for non-systemic crises; for systemic crises it turns out be insignificant. (Results are available on request). 7. Conclusions In this paper, we examine the effect of six dimensions of financial liberalization on the likelihood of systemic and non-systemic banking crises. We find that liberalization reduces the likelihood of systemic crises, which is against the commonly held view that liberalization increases the likelihood of a banking crisis. Our results also hold after controlling for a possible endogeneity problem. Moreover, we also find that systemic and non-systemic crises behave differently. If anything, liberalization has a positive impact on the likelihood of a non-systemic crisis. References Abiad, Abdul and Ashoka Mody (2005), Financial Reform: What Shakes It? What Shapes It?, American Economic Review, 95 (1), pp Abiad, Abdul, Enrica Detragiche, and Thierry Tressel (2007), A New Database of Financial Reforms, unpublished, Washington: International Monetary Fund. Arellano, M. and J. Hahn (2007), Understanding Bias in Nonlinear Panel Models: Some Recent Developments. In: R. Blundell, W. Newey, and T. Persson (eds.), Advances in Economics and Econometrics, Ninth World Congress, Cambridge University Press Barth, James R., Gerard Caprio Jr., and Ross Levine (2004), Bank Regulation and Supervision: what works best?, Journal of Financial Intermediation, 13, Beck, Thorsten, Asli Demirguc-Kunt, and Ross Levine, Ross (2006), Bank Concentration, Competition, and Crises: First results, Journal of Banking and Finance, 30, Caprio, G. and D. Klingebiel (1999), Episodes of Systemic and Borderline Financial Crises, Mimeo, The World Bank. Caprio Jr., Gerard and E.A. Martinez, (2000), Avoiding Disaster: Policies to Reduce the Risk of Banking Crises. Egyptian Centre for Economic Studies Working Paper No. 47. Demirguc-Kunt, Asli and Enria Detragiache, Enrica (1998), Financial Liberalization and Financial Fragility, World Bank Policy Research Working Paper No Demirguc-Kunt, Asli and Enrica Detragiache (2000), Monitoring Banking Sector Fragility: A Multivariate Logit Approach, World Bank Economic Review, 14(2),

18 Demirguc-Kunt, Asli and Enrica Detragiachea (2002), Does Deposit Insurance Increase Banking System Stability? An Empirical Investigation, Journal of Monetary Economics, 49, Dreher, A. (2004), The Influence of IMF Programs on the Re-election of Debtor Governments, Economics and Politics, 16(1), Green, William (2004), The Behaviour of the Maximum Likelihood Estimator of Limited Dependent Variable Models in the Presence of Fixed Effects, Econometrics Journal, 7(1), Gwartney, James and Robert Lawson with Seth Norton (2008). Economic Freedom of the World, 2008 Annual Report. The Fraser Institute. Data retrieved from Honahan, P. and L. Laeven (2005), Systemic Financial Distress: Containment and Resolution, Cambridge (UK): Cambridge University Press. Kaminsky, Graciela L. (1999), Currency and Banking Crises: The Early Warnings of Distress, IMF Working Paper No. 99/178. Mehrez, Gil and Daniel Kaufman (2000), Transparency, Liberalization and Banking Crises, World Bank Policy Research Working Paper No

19 Appendix Country TABLE A1: Systemic and Non-systemic Crises Systemic Crises Non-systemic Crises Argentina , , 1995, 2001 Australia Bangladesh Bolivia , 1994 Brazil 1990, Canada Chile Colombia Costa Rica Ecuador , , Egypt Ghana Guatemala 1990s , Hong Kong , 1998 India 1993 Indonesia Israel Japan Korea, PR Malaysia Mexico , Morocco New Zealand Peru Philippines , 1998 Singapore 1982 South Africa 1989 Sri Lanka Thailand , United States Uruguay , 2002 Venezuela Zimbabwe Source: Honohan and Laeven (2005) 19

20 Table A2 Correlation of controls Credit Growth Lag GDP Growth GDP level Inflation GDP Growth 1.00 GDP level Inflation Credit - Growth Lag Systemic Crises Nonsystemic Crises 0.02 Systemic Crises Nonsystemic Crises Depreciation Real Interest Rate Terms of Trade Changes Depreciation Real Interest - Rate Terms of Trade Changes M2 to Reserve Ratio M2 to Reserve Ratio

21 Table A3 : Systemic Crisis Results with Panel Data Probit Model (For Non-OECD Countries) Real GDP Growth (t-1) *** *** *** *** *** *** *** *** SE Real GDP Level *** *** *** *** *** *** *** ** SE Real Interest Rate 0.008* 0.013** 0.008* 0.010** 0.008* 0.010** 0.011** 0.009* SE Inflation SE M2/Reserves * * * * * * * * SE Terms of Trade (t-1) SE Credit to Private Sector (t-2) SE Depreciation 1.699** * 1.552* 1.714** * SE Supervision SE Liberalization Reforms *** SE Credit Control Reforms ** SE Interest Rate Control Reforms *** SE Barrier to Entry Relaxation SE Privatization *** SE Capital Control Relaxation -0.31*** SE Securities Market Reforms ** SE Constant SE Number of Observation Wald Chi- Square *** *** *** *** *** *** *** *** LR-Ratio Test 22.04*** 27.8*** 22.88*** 32.64*** 21.8*** 25.43*** 21.44*** 23.03*** stands for coefficient value and SE for Standard Errors *** indicates significance at 1% level whereas ** indicates significance at 5% and * indicates significance at 10% level 21