Finix

Journal of Banking & Finance 44 (2014) 114–129

Contents lists available at ScienceDirect

Journal of Banking & Finance

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j b f

Macro-financial determinants of the great financial crisis: Implications for financial regulation q

http://dx.doi.org/10.1016/j.jbankfin.2014.03.001 0378-4266/� 2014 Elsevier B.V. All rights reserved.

q We would like to thank the Editor, an anonymous referee, Luc Laeven, Ross Levine, Marco Pagano, Andrea Sironi, Randy Stevenson, Gianfranco Torriero, Giuseppe Zadra and seminar participants at IFABS Conference and ISTEIN seminar for helpful comments. This paper’s findings, interpretations, and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank and the Italian Banking Association. ⇑ Corresponding author. Tel.: +39 02 58362725.

E-mail addresses: [email protected] (G. Caprio Jr.), [email protected] (V. D’Apice), [email protected] (G. Ferri), [email protected] (G.W. Puopolo).

Gerard Caprio Jr. a, Vincenzo D’Apice b,c, Giovanni Ferri d,e, Giovanni Walter Puopolo f,⇑ a Williams College, United States b Economic Research Department of Italian Banking Association, Italy c Istituto Einaudi (IstEin), Italy d LUMSA University of Rome, Italy e Center for Relationship Banking & Economics – CERBE, Italy f Bocconi University, CSEF and P. Baffi Center, Italy

a r t i c l e i n f o

Article history: Received 15 April 2012 Accepted 4 March 2014 Available online 29 March 2014

JEL classification: G01 G15 G18 G21

Keywords: Banking crisis Government intervention Regulation

a b s t r a c t

We provide a cross-country and cross-bank analysis of the financial determinants of the Great Financial Crisis using data on 83 countries from the period 1998 to 2006. First, our cross-country results show that the probability of suffering the crisis in 2008 was larger for countries having higher levels of credit deposit ratio whereas it was lower for countries characterized by higher levels of: (i) net interest margin, (ii) concentration in the banking sector, (iii) restrictions to bank activities, (iv) private monitoring. The bank-level analysis reinforces these results and shows that the latter factors are also key determinants across banks, thus explaining the probability of bank crisis. Our findings contribute to extend the analyt- ical toolkit available for macro and micro-prudential regulation.

� 2014 Elsevier B.V. All rights reserved.

1. Introduction ment (BCBS, 2010a), has focused more on the stability of the finan-

As much as it was known that the Great Depression of the 1930s was the acid test for any reputable macroeconomic theory, the out- break of the Great Financial crisis in 2008 has shaken not only financial institutions, but also long-held beliefs and theories on how the regulation of the financial system should be structured, with renewed emphasis on macro-prudential supervision and reforming micro-prudential regulation.

In turn, the financial regulatory reforms have sparked a vibrant debate among institutions, academics and practitioners. On the one hand, the Basel Committee, starting with its consultative docu-

cial system, arguing that the costs of the new regulation will be much lower than the relative benefits (see BCBS, 2010b; MAG, 2010). On the other hand, the banking industry argues that the new measures could put economic growth at risk imposing high costs on the financial intermediaries and, in turn, on economic sys- tems (IIF, 2010). In the middle, some academics argue that the prin- ciples implicitly or explicitly subscribed by the Basel Committee may be questionable to secure more resilient financial systems (see among others Ferri, 2001; Barth et al., 2004, 2006; Caprio, 2010).

In this debate, we study whether a wide set of banking indica- tors, such as business model, funding strategy, market structure, efficiency, stability, profitability, regulation, the quality of gover- nance and a measure of financial globalization could explain the ex-post incidence of the crisis both across countries (that is, at the macro-level) and across banks (that is, at the micro-level), and be added to the analytical toolkit available for prudential supervision. Specifically, in the cross-country analysis we investi- gate the (macro) financial determinants of the probability that a country experienced the crisis in 2008, as reported by Laeven and Valencia (2010), using data on 83 countries from 1998 to 2006. In the cross-bank analysis, by contrast, we pursue a twofold

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 115

objective: first, using the information on the 10 largest banks by average total asset during 1998–2006 for all the countries consid- ered in our sample we focus on the determinants of the probability that a bank experienced some form of distress during the Great Financial crisis, to understand whether the main results obtained in the cross-country analysis hold also with bank-level data. Sec- ond, we can address the potential problem of omitted variables arising from the cross-country analysis.

A novel feature of our approach with respect to the related lit- erature1 consists in measuring the financial indicators used as explanatory variables taking into account all the information relative to the 9 years that preceded the Great Financial crisis and not just to the most recent years before its outbreak. In fact, we firmly believe that the early signals of what happened, starting from 2008, were already embedded in the financial characteristics of the countries and their banks several years before the crisis erupted. Then, the use of such ‘‘back-in-time variables’’ is justified by the fact that these contain information about (i) the health of the financial system in the past and (ii) how this evolves over time. As a consequence, they may be useful in understanding the genesis of the crisis.

Our cross-country analysis shows that, first, countries with a higher credit/deposit ratio had higher probability to be in crisis in 2008. Next, a few determinants negatively impinged on the probability of crisis. Specifically, such probability was lower for countries with a higher level of net interest margin, higher level of concentration in the banking sector, higher level of private mon- itoring, and more restrictions on bank activities. Moving to the cross-bank analysis, it is important to underline that our micro- level evidence contributes to reinforce these results by showing that they hold not only across countries, but also across banks. In other words, we find that the financial factors found at the coun- try-level are also key determinants at the bank-level, thus explain- ing the probability of bank crisis as well.

In particular, among the various determinants of the crisis, a crucial role is played by the net interest margin indicator. This factor tends to be more significant the greater the importance of deposits. In fact, banks that had a large and stable deposit base likely paid less for funds (thus reaching a higher level of net inter- est margin) than the ones who had to rely on wholesale markets, which proved to be more volatile. At the same time, net interest margin also tends to be lower in banking systems more extensively engaged in securitization, both directly as securitization fees dis- place interest earnings (and interest on the securities is accruing to off-balance sheet entities), and indirectly as securitization boosts the supply of credit from non-bank entities which leads, other things equal, to a decrease in the lending rates.

The rest of the paper is structured as follows. Section 2 provides a review of the literature. Section 3 describes the data and the dif- ferent models employed in the econometric specifications, focus- ing first on the analysis across countries and then across banks. Section 4 looks at the empirical results, while Section 5 offers some robustness checks. Finally, Section 6 concludes discussing some lessons and policy implications.

2. Literature review

Over the years, several scholars have studied financial crises, focusing on their possible causes and above all on predicting their time of occurrence. Historically, however, the economic analysis showed more success at identifying the incidence of the crises across firms, banks or countries (i.e. cross-sectional) rather than at forecasting the timing of crises (i.e. in time-series analysis). For instance, focusing on the financial crisis of 2008, Rose and Spiegel

1 See among the others Rose and Spiegel (2009), Claessens et al. (2010), Barth et al (2004) and Beck et al. (2006).

.

(2009) use a latent variable approach to investigate whether a wide number of factors could have predicted the incidence and the sever- ity of the crisis for many countries. They find few clear reliable indi- cators of the incidence of the great recession in the pre-crisis data: more precisely, only the natural logarithm of 2006 real GDP per capita and the size of the equity market run-up prior to the crisis result in a significant causality with the severity of the crisis.

In this regard, the closest paper to our cross-country analysis is Barth et al. (2004). Using their database on bank regulation and supervision in 107 countries to assess the relationship between specific regulatory and supervisory practices and banking-sector development, efficiency, and stability, they show that the likeli- hood of suffering a major crisis is greater the more countries: (1) restrict bank activities (or prevent or discourage diversification of income through non-traditional activities); (2) put limits on for- eign bank entry/ownership; (3) exacerbate moral hazard via a more generous deposit insurance scheme. On the other hand, nei- ther capital stringency nor official supervisory powers – which approximate respectively pillars one and two of Basel II – are robustly linked to banking crises when controlling for other super- visory/regulatory policies. Similarly, there is no significant associa- tion between private-sector monitoring and the likelihood of a banking crisis and only a weak positive relationship between gov- ernment ownership and the likelihood of a crisis.

Our macro-level analysis differs from theirs along several dimensions. First, we focus on a different crisis episode, that is the Great Financial crisis started in 2008. Second, we use a different set of macro-financial indicators as possible explanatory variables of the probability for a country to be in crisis in 2008. Third, we do not restrict the observations to a precise year (for example 1999 as done by the already cited authors), but rather we take the annual mean of these financial factors from 1998 to 2006, to take into account the long-term evolution of the financial sector before the crisis broke out internationally. Finally, we reinforce our cross- country results by also investigating the determinants of the crisis at the bank-level.

Before the great financial crisis broke out in 2008, Demirguc- Kunt and Detragiache (1998) investigate the relationship between banking crises and measures aimed at increasing the level of finan- cial liberalization in 53 countries during the period 1980–1995. They find that banking crises are more likely to occur in liberalized financial systems. However, they do not consider data on regula- tion and supervision. Mehrez and Kaufman (2000) employ a mul- tivariate probit model for 56 countries from 1977 to 1997 to examine how the level of corruption (i.e. transparency) affects the likelihood of financial crises. They report that, in countries where the government policy is characterized by lack of transpar- ency, banks have incentives to raise credit above the optimal level, thus increasing the probability of a banking crisis.

Using data on 69 countries from 1980 to 1997, Beck et al. (2006) study the impact of bank concentration, bank regulation, and national institutions on the probability that a country can experi- ence a systemic banking crisis. They also examine the international differences in bank capital regulations, rules restricting bank entry, regulatory restrictions on bank activities and the overall institu- tional environment. They show that crises are less likely to occur in economies characterized by: (1) more concentrated banking sys- tems; (2) fewer regulatory restrictions on banks (i.e. lower barriers to bank entry and fewer restrictions on bank activities); (3) national institutions that facilitate competition. In addition, Shehzad and De Hann (2009) analyze the impact of financial reform on systemic and non systemic banking crises in 85 coun- tries, from 1973 to 2002, finding that certain types of financial reform reduce the likelihood of crisis.

Focusing on the Great Financial crisis, Giannone et al. (2011) study cross-country differences in output loss between 2008 and

116 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

2009 using indices of country risk for more than one hundred countries. They find that the set of policies that favour liberaliza- tion in credit markets are negatively correlated with the output growth in 2008 and 2009.

Moreover, Claessens et al. (2010) investigate the causes of a broader set of financial crises finding that the recent crisis has four major features similar to earlier episodes. First, in most countries, asset prices rapidly increased before the crisis. Second, several key economies experienced episodes of credit booms ahead of the cri- sis. Third, there was a dramatic expansion in a variety of marginal loans. Fourth, the regulation and the supervision of financial insti- tutions failed to keep up with developments. They also find that the recent crisis was different from the previous ones in, at least, four new aspects. First, there was a widespread use of complex and opaque financial instruments. Second, the interconnectedness among financial markets, nationally and internationally, with the United States at the core, had increased in a short time period. Third, the degree of leverage of financial institutions accelerated sharply. Fourth, the household sector played a central role.

Moving to bank-level data, Beltratti and Stulz (2012) analyze 98 large banks from 20 countries and investigate whether bank per- formance is related to bank-level governance, country-level gover- nance, bank balance sheet and profitability characteristics before the crisis, and country-level regulation. Their key results are: (1) banks that the market favored in 2006 showed especially poor returns during the crisis; (2) banks with more shareholder-friendly boards performed worse during the crisis; (3) banks in countries with stricter capital requirement regulations and with more inde- pendent supervisors performed better; (4) banks in countries with more powerful supervisors experienced worse stock returns; (5) large banks with more Tier 1 capital and more deposit financing at the end of 2006 showed significantly higher returns during the crisis.

Moreover, also focusing on micro-level data, De Jonghe (2010) analyzes the impact of revenue diversity of financial corporations on the banking system stability and investigates why some banks are better able to shelter themselves from the storm. He shows that the shift to non-traditional banking activities, which generate com- missions, trading and other non-interest income, increases individ- ual banks risks and thus reduces the stability of the financial system. Finally, DeYoung and Torna (2013) investigate whether and how banks’ shift from traditional to nontraditional income sources contributed to the failure of hundreds of US depository institutions between 2008 and 2010.

3. Data and methodology

In this section we describe the data and the econometric models we employ in the analysis (i) across countries and (ii) across banks, whereas in Section 4 we show the corresponding results. We start from the cross-country analysis.

4 For four countries, and more precisely Macau, Malta, Bahrein and Oman, we had to determine the status crisis/no crisis on our own by analyzing the financial stability review provided by their central banks. In fact, all the information corresponding to the independent variables is available whereas the information about the dependent variable is not available in Laeven and Valencia (2010).

5 We give here the synthetic description of the variables. For more details, see Appendix B.

6 This variable only includes customer deposits and does not include interbank

3.1. Cross-country analysis

In the macro-level analysis we employ aggregate (that is, coun- try-level) information on the country’s financial system from the period 1998 to 2008 for 83 countries, including OECD as well as non-OECD and developing countries.2,3

2 See Appendix A for the list of countries included in the sample. 3 We select 1998 as starting year because the measurement of our regulation

variables begins in that year. In fact, we borrow these variables from the three surveys conducted by Barth et al. (2004). Specifically, survey I was conducted to assess the state of regulation in 1998, survey II to assess the state of regulation in 2002, and finally, survey III to assess the state of regulation in 2005.

In order to identify the macro and financial structure factors contributing to the Great Financial crisis, we run cross-country regressions on the determinants of the probability that a country experienced the crisis in 2008. Specifically, for 83 countries, we estimate several probit models in which the dependent variable, that is CRISIS, is a dummy equal to one if the country is classified as either borderline crisis or systemic crisis according to the definition introduced by Laeven and Valencia (2010),4 and zero otherwise. Moreover, we investigate the impact of the degree of financial globalization on the probability of crisis by estimating an instrumental-variables probit model accounting for this variable’s potential endogeneity. Finally, we also test the role of the country’s governance quality as explanatory variable of the probability of crisis through the indices provided by Kaufman et al. (2010).

Here are the variables employed in the cross-country specifications.

As independent variables we employ a wide set of country indicators to take into account the various characteristics of the national financial systems, such as, e.g., banking efficiency, stability, profitability, market structure, quality of governance and regulation. In particular, we use the following banking indica- tors, all measured as means over the period 1998–20065:

(i) NET_INTEREST_MARGIN, measuring the country bank’s net interest revenue, as a share of its interest-bearing assets, to proxy the banking system orientation towards traditional activity;

(ii) ROA, return on assets (net income to total assets); (iii) ROE, the country return on equity (net income to total

equity); (iv) COST_INCOME, total costs as a share of total income of all the

country’s banks, proxying bank efficiency; (v) Z-SCORE, the aggregate bank z-score;

(vi) CREDIT_DEPOSIT, the country’s loan/deposit ratio.6 While a high ratio indicates high intermediation efficiency, a ratio sig- nificantly above one suggests reliance on possibly unstable non-deposit funding (see Beck et al., 2000; Beck et al., 2010; Merrouche and Nier, 2010)7;

(vii) CONCENTRATION, the share of the country’s three largest banks in all country’s banks assets. While big banks could reduce risk via enhanced asset diversification, they could raise risk if managers and shareholders anticipate ‘‘too-big- to-fail’’ policies by regulators;

(viii) BANK_ASSETS_GDP, the ratio between the country’s deposit money bank assets and its GDP.

We also take into account the degree of financial globalization with the following variable:

INT_DEBT_ISS_GROSS_GDP, the gross flow of international bond issues by the country scaled by its GDP, proxying the degree to

deposits. 7 The ratio of credit to deposit measures how much non-deposit funding are used

to increase domestic credit. These alternative sources of funding include short-term debt (e.g. commercial paper and asset-backed commercial paper) and long-term debt (e.g. bonds). Though desirable, a breakdown of funding into short-term and long-term instruments is not available either from International Monetary Fund, such as International Financial Statistics, or from international bank-level databases, such as Bankscope (see Huang and Ratnovski, 2009).

Table 1 Summary statistics.

Mean Std. Dev. Min. Max. Observations Annual Average Change

Dependent variables CRISIS (Probit Models) 0.23 0.42 0 1 83 CRISIS ORDERED (Ordered Probit Model) 0.36 0.71 0 2 83 CRISIS_COST_GDP (Tobit Model) 1.11 3.2 0 18.5 83

Independent variables Banking variables NET_INTEREST_MARGIN 4.31 2.35 0.88 12.78 83 �0.07 ROA 1.03 1.13 �1.82 4.38 83 0.13 ROE 10.73 10.09 �26.57 50.08 83 1.17 COST_INCOME 67.74 14.91 29.78 108.45 83 �1.28 Z-SCORE 11.25 6.63 4.2 39.17 81 �0.8 CREDIT_DEPOSIT 102.19 38.6 35.83 248.78 83 0.51 CONCENTRATION 67.42 18.95 24.8 100 83 �0.63 BANK_ASSETS_GDP 70.25 43.15 15.99 185.11 83 1.63 INT_DEBT_ISS_GROSS_GDP 21.44 25.62 0.04 145.98 73 2.26

Banking regulation variables RESTRICTION 7.43 1.74 3.33 12 83 PRIVATE MONITORING 8.23 1.29 5.5 11 83 CAPITAL 6.15 1.49 3 10 83 ENTRY 7.41 0.87 3.67 8 83 SUPERVISION 10.94 2.31 5 14.25 82

Governance quality variables KKZ_MEAN 0.45 0.86 �1.4 1.93 83 KKZ_RULELAW 0.45 0.93 �1.39 1.95 83 KKZ_REGQUAL 0.57 0.81 �1.3 1.92 83 KKZ_VOICE 0.42 0.86 �1.56 1.63 83 KKZ_POLSTAB 0.2 0.9 �2.1 1.58 83 KKZ_GOVEFF 0.55 0.94 �1.4 2.16 83 KKZ_CONCORR 0.48 1.04 �1.17 2.48 83

Table describes the summary statistics of the variables used in our cross-country analysis. For each country, the independent variables are computed as the annual average over the period 1998–2006. For the Independent variables ‘‘Banking variables’’ we also computed the Annual average change, defined as the average change of the variables between two consecutive years. Detailed definitions of the variables are given in Appendix B.

8 All the models are estimated with heteroschedasticity-robust standard errors.

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 117

which a country’s financial system is interlinked with international financial markets.

We also use five measures of bank regulation, taken from Barth et al. (2004), and computed as mean of their three surveys:

(i) RESTRICTION, the value of the ‘‘Overall Restrictions’’ index, measuring the extent of regulatory restrictions on bank activities in securities markets, insurance, real-estate, and owning shares in non-financial firms;

(ii) PRIVATE MONITORING, the ‘‘Private Monitoring’’ index, mea- suring the degree to which regulations empower, facilitate, and encourage the private sector to monitor banks;

(iii) CAPITAL, the ‘‘Capital Regulation’’ index, which can be con- sidered as a proxy of Basel Pillar 1;

(iv) ENTRY, the ‘‘Entry Requirements’’ index, proxying the hur- dles for entrants to get a bank license;

(v) SUPERVISION, the ‘‘Official Supervisory’’ index, measuring the degree to which the country’s bank supervisor has the authority to take specific actions. It can be seen as a measure of Basel Pillar 2.

Regarding the institutional quality of the country we use the indices of governance quality provided by Kaufman et al. (2010): ‘‘Voice and Accountability’’, ‘‘Political Stability and Absence of Vio- lence’’, ‘‘Government Effectiveness’’, ‘‘Regulatory Quality’’, ‘‘Rule of Law’’ and ‘‘Control of Corruption’’.

Table 1 reports the summary statistics (together with their clas- sification group) of all the variables we just described. Only in the case of the Independent Variables ‘‘Banking Variables’’ we also compute the annual average change, defined as the average change

of the variables between two consecutive years. In particular, the latter statistics provides some information about the (average) evolution of the variables over 1998–2006, i.e. if they have increased or diminished and the extent of this change.

We start our investigation of the determinants of the probabil- ity that a country experienced the crisis in 2008 by estimating the following probit model8:

ProbðCRISISc ¼ 1jXÞ¼ Uða þ b1NET INTEREST MARGINc þ b2CREDIT DEPOSIT c þ b3CONCENTRATIONc þ b4RESTRICTIONc þ b5PRIVATE MONITORINGcÞ; ð1Þ

where the dependent variable, CRISISc, is a dummy equal to 1 if the country c is classified as crisis and 0 otherwise, U is the standard normal cumulative distribution function, and X is the set of explan- atory variables.

The choice of the explanatory variables is motivated by several reasons. First, we believe that one of the most important banking causes of the recent crisis is the shift from the ‘‘originate to hold (OTH)’’ model to the ‘‘originate to distribute (OTD)’’ model (see for example Berndt and Gupta, 2009; Mian and Sufi, 2009; Stiglitz, 2010; FSA, 2009; D’Apice and Ferri, 2010; Trichet, 2009). Therefore, the variable NET_INTEREST_MARGIN allows testing whether higher incentives to perform traditional banking activities could be a deterrent against the crisis. In fact, a lower net interest margin implies higher incentives for traditional banks to look for other income sources (that is, ‘‘searching for yield’’) and to shift

118 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

to new business models (Beck et al., 2000, 2010; Gambacorta and Marques-Ibanez, 2011).

Moreover, one of the most striking features of the great finan- cial crisis was the impact on the money markets and global liquid- ity (Cecchetti, 2009; Allen and Carletti, 2008; Brunnermeier, 2009). Thus, we use the variable CREDIT_DEPOSIT to account for the role of maturity mismatching. In fact, a ratio significantly above one suggests that private sector lending is funded with non-deposit sources, which could result in funding instability (Beck et al., 2000, 2010).

Another important aspect of the country’s resilience to the crisis concerns the structure of the banking system, and in particular the degree of bank concentration (Carletti, 2010). In truth, economic theory provides two conflicting predictions on the relationship between concentration and stability. On the one hand, for the char- ter value hypothesis (Allen and Gale, 2000a, 2000b, 2004) concen- tration enhances stability, whereas, on the other hand, the optimal contracting hypothesis (Boyd et al., 2005) argues exactly the oppo- site. In our empirical work, the use of the variable CONCENTRA- TION is meant to capture the structure of the banking system.

Finally, many analyses argue that the flaws in the regulatory framework played a very important role in leading to the crisis (Demirguc-Kunt and Serven, 2009; Coval et al., 2009; Buiter, 2007; De Michelis, 2009). To account for the country’s regulatory regime, we use the corresponding indices provided by Barth et al. (2004). In our base model (1), RESTRICTION proxies the regulatory restrictions on banking activities, whereas PRIVATE MONITORING proxies Basel Pillar 3.

We present the results of cross-country model (1) in Section 4.1.

3.1.1. The link with the financial globalization In the previous analysis of the macro-financial determinants of

the crisis we disregarded the role played by the degree to which a country’s financial system is interlinked with international financial markets. Of course, this could be a primary factor, especially considering that international contagion was one of the chief channels to spread the crisis worldwide. Thus, here, we extend our cross-country analysis and include this indicator together with the other determinants of the crisis. Specifically, we follow the related literature and measure the country’s degree of financial globalization using the gross flow of international bond issues as percentage of its GDP, that is the variable INT_DEBT_ISS_GROSS_GDP.

In any case, the inclusion of this independent variable may raise problems of endogeneity. On the one hand, indeed, an extensive literature underlined the possibility that the extent of financial globalization of a country depends not only on its level of develop- ment and its policies, but is also influenced by deeper fundamental factors.9 On the other hand, however, it is also true that, in our mod- els, the explanatory variables are measured as the average of the annual observations from 1998 to 2006, a circumstance which could attenuate (or even eliminate) the inverse causality effect of the prob- ability of crisis on the independent variables (measured much before 2008). To better understand this issue, we compute the cross-coun- try correlation between the variable INT_DEBT_ISS_GROSS_GDP observed in 2008 and its annual mean from 1998 to 2006. We find that the resulting correlation is very high (about 0.95), indicating a strong persistence of these fundamental factors cited by the litera- ture,10 and thus suggesting that the inclusion of this variable could indeed raise problems of endogeneity.

9 See for example Acemoglu and Johnson (2005), Collins (2005), Faria and Mauro (2005), Kose et al. (2006), Spiegel (2008), Tytell and Wei (2005), and Wei (2006).

10 Actually, the variable INT_DEBT_ISS_GROSS_GDP is quite persistent over time. In fact, all cross-country correlations between any two years of this variable within the period 1998–2006 are higher than 0.7. Results are available upon request.

In order to avoid biased estimates when measuring the proba- bility of crisis in 2008, we control for the endogeneity issue by introducing several instrumental variables denoting the various characteristics of the countries. In our setting, this translates into using an instrumental variables probit model, which is practically equivalent to estimating simultaneously the following two equations:

INT DEBT ISS GROSS GDPc ¼ a þ c1 NET INTEREST MARGINc þ c2CREDIT DEPOSIT c þ c3CONCENTRATIONc þ c4RESTRICTIONc þ c5PRIVATE MONITORINGc þ c6LEGAL ORIGIN � SOCIALIST c þ c7ETHNIC FRACTIONALIZATIONc þ c8SCALED CAPITAL DISTANCEc þ ec;

ð2Þ

ProbðCRISISc ¼ 1jXÞ ¼ Uða þ b1 INT DEBT ISS GROSS GDPc þ b2NET INTEREST MARGINc þ b3CREDIT DEPOSIT c þ b4 CONCENTRATIONc þ b5RESTRICTIONc þ b6PRIVATE MONITORINGcÞ: ð3Þ

In Eq. (2), we linearly estimate INT_DEBT_ISS_GROSS_GDP using as control variables all the regressors of the base model (1) plus the variables: (i) LEGAL ORIGINS – SOCIALIST taken from La Porta et al. (1997 and following updates), capturing the legal characteris- tics of former-socialist countries, (ii) ETHNIC FRACTIONALIZATION taken from Alesina et al. (2003), measuring the degree of ethnic het- erogeneity in the countries, and (iii) SCALED_CAPITAL_DISTANCE. The latter variable is computed as the distance between the capital of each country and the USA capital divided by the highest distance from Washington available in our sample, to guarantee the homo- geneity of this measure with the other variables of our setting. In our framework, this ratio captures the distance of the country from the origin of the recent crisis (scaled by the maximal distance from the USA).

In Eq. (3), the probit model uses the value of INT_DEBT_ISS_ GROSS_GDP from Eq. (2) together with the major cross-country determinants of the crisis found in the previous section, whereas the dependent variable does not change.

The aforementioned variables prove good instruments in our regression.11 In fact, to avoid endogeneity, it is crucial that the instruments are at the same time: (1) VALID (i.e., correlated with the financial globalization proxy), and (2) EXOGENOUS (i.e., uncorre- lated with the probability of the crisis).

We control that our instrumental variables satisfy both condi- tions, and find that they are: (1) related to financial globalization (that is, they are valid), and (2) uncorrelated with the error term (that is, they are exogenous),12 thus supporting the view that, in our framework, they are indeed good instruments.

We report the cross-country results of the link with the degree of financial globalization in Section 4.1.1.

3.1.2. The link with the quality of governance Another important macro-variable in determining the probabil-

ity that a country experienced the crisis in 2008 could be the

11 They are also among the most widely used in the related literature (see Beck et al., 2006; Ponczek and Mattos, 2009; Glaeser et al., 2004).

12 Results are not reported and are available upon request.

15 In fact, it is not suitable to assign an identical value to all banks belonging to the same country independently of the financial situation associated to the individual banks (that is, independently of whether each bank is in distress or not).

16 In other words, the dependent variable would exactly be a linear combination of such explanatory variables.

17 The ‘‘random-intercept model’’, also called ‘‘mixed-effects model’’, contains both

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 119

quality of governance. In this section we precisely address this issue by investigating the role of this factor as possible cross- country determinant of the crisis, using as a proxy the governance indices provided by Kaufman et al. (2010).

We start by testing the explanatory power of the variable KKZ_MEAN (which is the simple average of the six indicators pro- vided by Kaufman et al. (2010)13 when added to our cross-country model (Eq. (1)). However, since the correlation between this indica- tor and the variable NET_INTEREST_MARGIN is very high (i.e. �0.60), we follow a standard two-step procedure to deal with the issue of multicollinearity among the two variables. Namely, we first regress KKZ_MEAN on NET_INTEREST_MARGIN and a constant, that is:

KKZ MEANc ¼ c0 þ c1 NET INTEREST MARGINc þ nc; ð4Þ

then, we use the estimated residual nc from (4) as explanatory var- iable in the following probit model:

ProbðCRISISc ¼ 1jXÞ ¼ Uða þ b1NET INTEREST MARGINc þ b2 CREDIT DEPOSITc þ b3 CONCENTRATIONc þ b4RESTRICTIONc þ b5 PRIVATE MONITORINGc þ b6 KKZ MEAN RESIDUALcÞ; ð5Þ

where the variable KKZ_MEAN_RESIDUAL is indeed the estimated residual nc from (4). In this way, KKZ_MEAN_RESIDUAL captures the effect of governance quality that is not explained by the variable NET_INTEREST_MARGIN.

After controlling for multicollinearity, we can properly investi- gate the explanatory power of the quality of governance as possi- ble cross-country determinant of the crisis. Section 4.1.2 reports the results.

3.2. Cross-bank analysis

The analysis across banks has a twofold objective: first, by investigating the determinants of the probability that a bank expe- rienced some form of distress during the crisis, it allows us to understand whether the results obtained in the cross-country anal- ysis hold also with bank-level data, that is whether the main deter- minants identified above are also ‘‘micro-founded’’; second, it also allows us to address the potential problem of omitted variables arising from the cross-country analysis.

In fact, one of the main drawbacks of the cross-country analysis developed so far is exactly the potential problem of omitted vari- ables, since our macro-approach does not allow the inclusion of dummy variables identifying each country. Therefore, the effects attributed to the variables explicitly considered in the previous sections could instead be originated by other country characteristics.

In this section we follow Laeven and Levine (2009) and collect information on the 10 largest banks by average total asset during 1998–2006 for all the countries considered in our sample. Because some countries have data on fewer than 10 banks, our final sample consists of 755 banks across 83 countries.14 Using the same defini- tions provided in the cross-country analysis (see Section 3.1 and Appendix B) but obviously referring to each bank in the sample, we compute the variables ROA_BL, ROE_BL, COST_INCOME_BL, Z-SCORE_BL, NET_INTEREST_MARGIN_BL and CREDIT_DEPOSIT_BL.

13 See Appendix B for further details. 14 Focusing on the 10 largest banks enhances comparability among countries, as in

Laeven and Levine (2009), and at the same time avoids an unbalanced distribution of tail events. Moreover, our sample accounts for a huge portion of the total banking system assets in each country.

On the contrary, the variables CONCENTRATION, RESTRICTION, PRI- VATE MONITORING, CAPITAL, ENTRY and SUPERVISION cannot be computed for each bank, since they are defined only at the country level. Thus, in the latter case, we assign the same value to all banks belonging to the same country. Consistently with the methodology described in the cross-country analysis, all the bank-level variables are computed as the annual mean over the period 1998–2006.

In the cross-bank analysis, however, we cannot use the same dependent variable employed so far, i.e. CRISIS, because first, that variable is defined only at the country level,15 and second, regress- ing CRISIS on regressors like country dummies, RESTRICTION and MONITORING is not statistically feasible since the corresponding model would generate perfect (or deterministic) dependence between CRISIS and such regressors.16

For these reasons, here we use as dependent variable CRISIS_BL, a dummy variable being 1 if the bank failed or received a govern- ment recapitalization during the crisis and 0 otherwise. In fact, such variable not only is defined at the bank-level but, certifying that the bank experienced some form of distress during the crisis, it is very much in line with the dependent variable used in the cross-country analysis.

In order to investigate whether the results obtained in the macro-level analysis hold also with bank-level data, we estimate the following cross-bank probit model:

ProbðCRISIS BLi ¼1jXÞ¼Uðaþb1NET INTEREST MARGIN BLi þb2 CONCENTRATIONi þb3CREDIT DEPOSIT BLi þb4 RESTRICTIONi þb5PRIVATE MONITORINGiÞ; ð6Þ

presenting the corresponding results in Section 4.2. Next, we address the omitted variable problem and run a bank-

level random-intercept model17 that allows controlling for all the unmeasured factors associated with each country. In our setting, by allowing each country to have a different intercept, such model is essentially equivalent to the approaches based on the introduction of country dummies.

In practice, we estimate the following (random-intercept) pro- bit model:

ProbðYb;c ¼ 1jXb;c; ucÞ¼ Uða þ bXb;c þ zb;c ucÞ; ð7Þ

where c is the index identifying the country, b is the index identify- ing the bank, Yb,c corresponds to the dummy variable CRISIS_BL, Xb,c is the set of bank-level variables described above together with the country-level regulatory variables defined in Section 3.1, and zb,c are the covariates corresponding to the random effects and can be used to represent the random intercept in each country. Specifically, in random-intercept models, zb,c is simply the scalar 1.

In our bank-level setting, the term uc � N(0, r2u) is the error related to the specific country, it has zero mean and is common to all banks from the same country. In other words, it identifies all the unmeasured factors associated with country c that affect the probability for a bank of that country to be in crisis. On the con- trary, eb,c � N(0, r2e ) concerns the error related to bank b in country

fixed effects and random effects. The fixed effects are analogous to the standard regression coefficients and are estimated directly, whereas the random effects are not directly estimated but are summarized according to their estimated variances. Specifically, random effects may take the form of random intercepts. In a cross- section model, random effects are useful for modelling intra-group correlation; that is, observations in the same panel are correlated because they share common panel- level random effects.

19 The reader should bear in mind that our results pertain to the financial crisis reaching its climax in 2008 and generalizing them to less extreme crises cases may be inappropriate.

20 We also try to interact this variable with CONCENTRATION to test whether bigger

120 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

c, it has zero mean and identifies all the unmeasured bank-specific factors which affect the probability of bank crisis.18

Moreover, compared to a standard probit model, in random- intercept models there is just one additional parameter to esti- mate, that is r2u. Specifically, if the estimation of r

2 u turns out to

be significant, it implies that each country would have a different intercept (more precisely, the intercept of country c would be a + uc), meaning that all the unmeasured factors associated with each country, thus captured by its intercept, would indeed affect the probability for a bank to be in crisis. On the contrary, if the esti- mation of r2u turns out to be insignificant, then all unmeasured factors associated with each country do not influence the probabil- ity of bank crisis.

All the results regarding the cross-bank analysis are described in Section 4.2.

4. Results

In this section we report the results of the various models described in the previous sections. We start from the cross-country results.

4.1. Cross-country determinants of the crisis

The evidence corresponding to the probit model (1) is reported in column Probit 1 of Table 2. First, the coefficient of NET_INTER- EST_MARGIN is negative and statistically significant. This indicates that countries with a higher level of net interest margin had a lower probability to be in crisis in 2008. A higher net interest income is associated with less securitization, and might also be picking up the impact of competition (e.g. less competitive systems such as Australia and Canada survived the crisis quite well, even though this evidence might be related to other features of their regulatory or institutional settings as well). Indeed, a higher level of net interest margin represents a strong incentive for banks to undertake traditional activities, such as loans, instead of riskier non-traditional activities such as securities trading (IMF, 2012; Gambacorta and Marques-Ibanez, 2011).

Second, the coefficient of CREDIT_DEPOSIT is positive and statis- tically significant, meaning that countries with a higher credit/ deposit ratio had a higher probability to be in crisis in 2008. As a matter of fact, a ratio significantly above one suggests reliance on possibly unstable non-deposit funding. While some of these alter- native sources may be as stable as customer deposits, i.e. retail bonds funding, many other can prove highly volatile, such as in the case of interbank or money market loans. Thus, considering that the crisis featured a dramatic drop in the availability of whole- sale funding, our evidence suggests that in most countries the alternative funding sources were of the volatile type.

Third, the coefficient of CONCENTRATION is negative and statisti- cally significant, indicating that countries with a higher level of con- centration in the banking sector had a lower probability to be in crisis in 2008, a finding consistent with the absence of crisis, for example, in Australia or Canada. This result seems in line with the empirical evidence provided by Beck et al. (2006) who find that cri- ses are less likely to occur in economies with more concentrated banking systems. Indeed, a more concentrated banking system implies that the bank’s charter value is higher and, as a conse- quence, the incentives for bank owners and managers to take exces- sive risk are lower. Apparently, the beneficial bank’s charter value effect on risk taking seems to have prevailed – in our sample of countries – on the possible additional detrimental effect passing

18 In fact, the random-intercept probit model (7) may also be stated in terms of the latent linear response Y⁄b,c = a + bXb,c + uc + eb,c, with Yb,c = 1 if Y

⁄ b,c > 0 and Yb,c = 0

otherwise, where Y⁄b,c denotes the latent variable.

through the impact of the level of competition on manager compen- sation schemes. Compensation policy seemed, in fact, to be a crucial factor during the crisis: countries with a lot of merger activity in banking tended to have compensation systems that favored growth of banks’ balance sheets, i.e. rewarding return without much atten- tion, if any, to the risk, and these countries seemed to have the most severe problems during the crisis (see Barth et al., 2012).

Fourth, the coefficient of RESTRICTION is negative and statisti- cally significant. This suggests that more restrictions on bank activ- ities lowered the probability for the country to be in crisis in 2008. In other words, the regulatory-induced specialization in the bank- ing sector enhances financial stability.19 However, it is important to notice that this result contrasts with earlier studies (Barth et al., 2006), and might in fact be a proxy for enforcement, that is countries that cared about imposing activity restrictions might have been enforcing other regulations, and thus it might be the enforcement rather than the restrictions per se that matters. Indeed, Barth et al. (2012) cite numerous breakdowns in enforcement of regulation as major causes of the recent crisis. Unfortunately, we do not have good direct measures of such enforcement to test this interpretation.

Fifth, the coefficient of PRIVATE MONITORING is negative and statistically significant, meaning that countries with a higher level of private monitoring had a lower probability to be in crisis in 2008. This result appears in contrast with the previous literature, which found private monitoring important in a variety of areas but practically with no influence on the stability of financial sys- tems (see Barth et al., 2006). However, we should recall that earlier research did not pertain to the evidence related to the recent crisis. Yet, this finding is in line with the goal of the third pillar of the Basel II Capital Accord.

In addition to the baseline model (1) discussed above, we esti- mate 9 cross-country probit models to investigate whether the other variables described in Section 3.1 have a role in explaining the probability of country crisis. Specifically, in each model we add to the five determinants of Eq. (1), in turn and one per time, one of the following variables: COST_INCOME, Z_SCORE, ROA, ROE, BANK_ASSET_GDP, CAPITAL, ENTRY, SUPERVISION, GDP_L and POP_L. The latter macro-variables GDP_L and POP_L are respectively the logarithm of country’s GDP and the logarithm of country’s population. Columns Probit 2 to Probit 10 of Table 2 show interesting evidence. First, the results of the baseline model are robust to the inclusion of these other regressors. Second, macro-variables like measures of bank efficiency (COST-INCOME), stability (Z-SCORE), profitability (ROA and ROE), bank size (BANK_ASSETS_GDP),20 the logarithm of GDP and the logarithm of population are never significant at the conventional levels. More- over, controlling for the five determinants characterizing the probit model (1), the coefficients of the variables CAPITAL and ENTRY turn out to be positive and significant.21 In particular, with regard to CAP- ITAL, this suggests that higher levels of initial capital restriction increased the probability for the country to be in crisis in 2008.

Indeed, one lesson we learnt from the great financial crisis is that capital adequacy was almost irrelevant in determining the occurrence of the crisis. For example, Northern Rock got in trouble just few weeks after the bank had announced plans to return excess capital to shareholders. On the other hand, the importance of capital in lowering the risk-taking approach followed by banks

and more concentrated banking systems can explain the probability of the crisis. However, also the interaction variable is not significant. The result is not reported in the tables but it is available upon request.

21 However, none of these two variables showed a significant coefficient if regressed individually against the dependent variable.

Table 2 Cross-country analysis.

Variables Probit 1 Probit 2 Probit 3 Probit 4 Probit 5

NET_INTEREST_MARGIN �0.2414** �0.2316** �0.2506** �0.2704** �0.2402** (0.1117) (0.1175) (0.1145) (0.116) (0.1107)

CREDIT_DEPOSIT 0.0174*** 0.0174*** 0.0167*** 0.0180*** 0.0173***

(0.0045) (0.0046) (0.005) (0.0047) (0.0046) CONCENTRATION �0.0235** �0.0231** �0.0239** �0.0244** �0.0232*

(0.0115) (0.0112) (0.0114) (0.0119) (0.0119) RESTRICTION �0.4298*** �0.4343*** �0.4611*** �0.4125*** �0.4355***

(0.1252) (0.1264) (0.1239) (0.1247) (0.1275) PRIVATE MONITORING �0.3858** �0.3766** �0.4079** �0.3605** �0.3886**

(0.1734) (0.1662) (0.1759) (0.1649) (0.1807) COST-INCOME �0.004

(0.0109) Z-SCORE �0.0193

(0.0238) ROA 0.2215

(0.1851) ROE �0.0037

(0.0203) Constant 5.9711*** 6.1322*** 6.7117*** 5.5357*** 6.0496***

(2.0933) (2.2314) (2.0606) (1.9792) (2.2514) Observations 83 83 81 83 83 Pseudo R-squared 0.4176 0.4184 0.4023 0.4258 0.418

Probit 6 Probit 7 Probit 8 Probit 9 Probit 10

NET_INTEREST_MARGIN �0.2465* �0.2567** �0.2476** �0.2543** �0.2434** (0.1311) (0.1199) (0.1195) (0.1203) (0.1059)

CREDIT_DEPOSIT 0.0174*** 0.0184*** 0.0164*** 0.0208*** 0.0172***

(0.0048) (0.0052) (0.0045) (0.0051) (0.0053) CONCENTRATION �0.0237** �0.0251** �0.0225* �0.0235** �0.0217*

(0.0118) (0.0122) (0.0118) (0.0119) (0.0111) RESTRICTION �0.4338*** �0.4658*** �0.4575*** �0.4983*** �0.4480***

(0.1255) (0.1211) (0.1191) (0.1304) (0.1344) PRIVATE MONITORING �0.3867** �0.3898** �0.3899** �0.3871** �0.4189**

(0.1724) (0.1838) (0.1779) (0.1709) (0.1731) BANK ASSETS_GDP �0.0003

(0.0052) CAPITAL 0.2319*

(0.129) ENTRY 0.3645*

(0.2059) SUPERVISION 0.1576

(0.1007) GDP_L 0.0213

(0.1085) POP_L 0.0504

(0.1802) Constant 6.0551*** 4.8571** 3.5369 4.3758* 5.3179*

(2.1986) (2.3979) (2.5927) (2.5379) (2.9828) Observations 83 83 83 82 82 Pseudo R-squared 0.4176 0.4536 0.4376 0.4407 0.418

Table shows the estimation of several cross-country probit models, starting from our baseline model (Eq. (1)) shown in Column Probit 1. In all these models, the dependent variable is CRISIS (which is a dummy variable equal to 1 if the country is classified as either borderline crisis or systemic crisis by Laeven and Valencia and 0 otherwise). Robust standard errors are reported in parenthesis. Summary statistics are given in Table 1 and the definitions of explanatory variables are provided in Appendix B. * Statistical significance of the parameter at 10% significance level. ** Statistical significance of the parameter at 5% significance level. *** Statistical significance of the parameter at 1% significance level.

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 121

works only in specific circumstances. In fact, Laeven and Levine (2009), combining the Bankscope data with the regulatory data- base, show that higher capital does lead to safer banks when there is a strategic owner, but when ownership is highly diversified, higher capital is associated with more risk taking. In other words, when there is no strategic (that is large) owner, everyone free rides and looks for higher returns. On the contrary, banks keep a more conservative approach when a strategic owner has control rights, which is like having a high franchise value, just as some concen- trated-ownership banks adopted the practice of plowing more bonuses into deferred equity to keep managers more risk averse.

The evidence relative to ENTRY indicates that a higher level of entry restriction increased the probability for the country to be in crisis in 2008. This result is in accordance with the mainstream

view that entry restrictions build inefficiencies and, hence, contrib- ute to instability. It should also be noticed that our result is not in contrast with the role of concentration. In fact, as shown by Beck et al. (2006), competition reduces fragility when controlling for concentration. Finally, the regulatory variable SUPERVISION is not significant at the conventional levels.

Before closing this section, it is worthwhile assessing the quan- titative importance of the cross-country determinants of the crisis within our base model (1). In this regard, we computed the mar- ginal effect, that is the change in the probability of country’s crisis for an infinitesimal change of the independent variables, obtaining that NET_INTEREST_MARGIN and CREDIT_DEPOSIT are the two most important regressors among the banking variables, whereas RESTRICTION is the most important one among the regulatory

Table 3 Cross-country analysis: financial globalization and governance quality.

Variables Probit IV Probit All Probit 11 Probit 12 Probit 13

NT_DEBT_ISS_GROSS_GDP �0.0163 0.017 (0.023) (0.0667)

NET_INTEREST_MARGIN �0.1774** �0.4399** �0.2393** �0.2309** �0.2660** (0.0789) (0.1934) (0.1075) (0.1051) (0.122)

CREDIT_DEPOSIT 0.0175*** 0.0207* 0.0140** 0.0147** 0.0153***

(0.0051) (0.0106) (0.006) (0.0059) (0.0056) CONCENTRATION �0.0189 �0.0336** �0.0308** �0.0297** �0.0277**

(0.0119) (0.0165) (0.0125) (0.012) (0.0119) RESTRICTION �0.3730*** �0.8098*** �0.3096* �0.3448** �0.2896*

(0.123) (0.2629) (0.1636) (0.1573) (0.1691) PRIVATE MONITORING �0.3491** �0.5210** �0.5440*** �0.5494*** �0.5171***

(0.1396) (0.2115) (0.1629) (0.1674) (0.1615) COST-INCOME 0.0083

(0.0222) Z-SCORE �0.0192

(0.0404) ROA 0.4895

(0.35) ROE 0.0135

(0.0338) BANK_ASSETS_GDP �0.0115

(0.0163) CAPITAL 0.2643

(0.2017) ENTRY 0.1152

(0.2516) SUPERVISION 0.2147

(0.2681) GDP_L 0.081

(0.1488) POP_L 0.1306

(0.2943) KKZ_MEAN_RESIDUAL 1.0833**

(0.5033) KKZ_RULELAW_RESIDUAL 0.8967**

(0.4453) KKZ_REGQUAL_RESIDUAL 0.9282*

(0.5123) Constant 5.5077** 2.7754 6.9808*** 7.1977*** 6.3805**

(2.3141) (7.8998) (2.3426) (2.3833) (2.3569) Observations 73 70 83 83 83 Pseudo R-squared 0.4821 0.4728 0.4673

Column Probit IV shows the results of the instrumental variables probit model (3) taking also into account the endogeneity issue due to the inclusion of the variable INT_DEBT_ISS_GROSS_GDP among the main determinants of the crisis, i.e. Eq. (2). In columns Probit 11–13 we report the results corresponding to the role of the quality of governance as possible cross-country determinant of the crisis. Finally, in column Probit All we test whether all the macro-variables belonging to the categories ‘‘Banking Variables’’ and ‘‘Banking Regulation Variables’’ shown in Table 1 have a role in determining the probability for a country to be in crisis in 2008 when considered jointly, that is all together as explanatory variables. In all these probit models the dependent variable is a dummy equals to 1 if the country is classified as crisis and 0 otherwise. Robust standard errors are reported in parenthesis. Summary statistics are given in Table 1 and the definitions of the explanatory variables are provided in Appendix B. * Statistical significance of the parameter at 10% significance level. ** Statistical significance of the parameter at 5% significance level. *** Statistical significance of the parameter at 1% significance level.

122 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

variables. Column D Probit of Table 4 shows the results and high- lights that a marginal increase of NET_INTEREST_MARGIN deter- mines a 4.4% reduction in the probability of crisis, whereas a marginal increase of CREDIT_DEPOSIT determines a 0.3% increase in such probability. Moreover, a change of NET_INTEREST_MAR- GIN, CREDIT_DEPOSIT and RESTRICTION in the range of �0.5 to +0.5 standard deviations from the mean implies, respectively, a change of 10.5%, 12.4% and 13.9% in the probability of crisis.

4.1.1. The degree of financial globalization: empirical evidence In this section we show the empirical evidence corresponding

to the instrumental variable probit model, i.e. Eqs. (2) and (3), employed in Section 3.1.1 to investigate the degree to which a country’s financial system is interlinked with international finan- cial markets.

As highlighted in column Probit IV of Table 3, we find that the results of the baseline model (1) remain almost unchanged since the variable INT_DEBT_ISS_GROSS_GDP is insignificant. This sug-

gests that the probability for a country to be in crisis in 2008 is not influenced by the degree to which the country’s financial sys- tem is interlinked with international financial markets, after con- trolling for the relevant macro-determinants of the crisis. The only difference with the model outlined in Eq. (1) is that the vari- able CONCENTRATION becomes insignificant in this specification. For what concerns the instrumental variable approach, we find that the correlation coefficient between the Eqs. (2) and (3), i.e. rho, is equal to 0.85 while the likelihood ratio test of whether such coefficient is significantly different from zero gives a p-value of 0.09, thus confirming our suspect that the variable INT_DEBT_ISS_ GROSS_GDP is indeed endogenous.

Before closing this paragraph, and in view of the cross-country results highlighted in the previous section, we also test whether all the macro-variables discussed so far (and more precisely, those belonging to the categories ‘‘Banking Variables’’ and ‘‘Banking Reg- ulation Variables’’ shown in Table 1), when considered jointly, that is all together as explanatory variables, have a role in determining

Table 4 Marginal effects, ordered probit and tobit specifications.

Variables D Probit Ord. Probit Tobit M_Coll

NET_INTEREST_MARGIN �0.0442** �0.2526** �1.4939** (0.0196) (0.1033) (0.7019)

CREDIT_DEPOSIT .0031*** 0.0185*** 0.1076*** 0.0174***

(0.001) (0.0048) (0.0369) (0.0045) CONCENTRATION �.0043** �0.0320*** �0.1958** �0.0235*

(0.0022) (0.0121) (0.096) (0.0115) RESTRICTION �.0786*** �0.4720*** �2.7722*** �0.556***

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 123

the probability for a country to be in crisis in 2008. Therefore, we estimate a cross-country probit model which includes all the afore- mentioned variables at the same time (including as well the instru- mented measure of financial globalization) and show the results in column Probit All of Table 3. We notice that, in such a specification, the implications of our base model (Eq. (1)) still hold. In fact, the five variables characterizing our cross-country analysis are confirmed as major determinants of the crisis, whereas on the con- trary, all the other variables, as suspected, are insignificant.

(0.0269) (0.1243) (0.9715) (0.133) PRIVATE MONITORING �.0705** �0.3248** �1.565 �0.386*

(0.0287) (0.1569) (0.9675) (0.1734) NIM_RESIDUAL �0.241*

(0.1117) Constant 32.0186*** 5.8697**

(10.3615) (2.0716) Observations 83 83 83 83 Pseudo R-squared 0.4176 0.3682 0.1893 0.4176

Column D Probit shows the marginal effects of the cross-country variables char- acterizing the baseline model (1), that is the change in the probability of country crisis for an infinitesimal change of the independent variables. Column Ord. Probit shows the results of the ordered probit model in which the dependent variable is ORDERED_CRISIS, i.e. a dummy equal to 1 if the country is classified as borderline crisis, 2 if the country is classified as systemic crisis by Laeven and Valencia (2010) and 0 otherwise. Column Tobit reports the results of the cross-country Tobit model with left-censoring limit equal to 0 and dependent variable corresponding to CRISIS_COST_GDP (the variable is provided by Laeven and Valencia (2010), and measures the country’s cost of public support to the financial system in terms of its GDP). See section 5.2 for more information about this variable. Finally, Column M_Coll reports the results of the probit model (9) employed to resolve the potential multicollinearity among the explanatory variables of our baseline model. Robust standard errors are reported in parenthesis. Summary statistics are given in Table 1 and the definitions of the explanatory variables are provided in Appendix B. * Statistical significance of the parameter at 10% significance level. ** Statistical significance of the parameter at 5% significance level. *** Statistical significance of the parameter at 1% significance level.

4.1.2. The quality of governance: empirical evidence In this section we report the results characterizing the use of

the quality of governance as possible cross-country determinant of the crisis, as discussed in Section 3.1.2. We start by showing the empirical evidence corresponding to Eq. (5).

As highlighted in column Probit 11 of Table 3, we find that the coefficient of the variable KKZ_MEAN_RESIDUAL is positive and significant, implying that countries with stronger governance had a higher probability to be in crisis. This result is in line with the nature of the recent financial crisis, as it was mostly a developed countries crisis. For what concerns the other macro-variables, most importantly, the results of our base specification do not change.

We also investigate whether a more appropriate index with respect to KKZ_MEAN provides different results from the ones just discussed. In particular, given the financial nature of our study, we believe that the most appropriate indices among the ones provided by Kaufman et al. (2010) are the variables KKZ_REGQUAL and KKZ_RULELAW.22 Thus, we test their role in explaining the probabil- ity of country crisis using the same methodology employed in the case of the variable KKZ_MEAN. More precisely, since the correlation between these two indices and NET_INTEREST_MARGIN is, respec- tively, �0.56 and �0.66, we apply the same two-step procedure described in Eqs. (4) and (5). Columns Probit 12 and 13 of Table 3 show the corresponding results. We notice that, first, these two indi- ces exhibit positive and significant coefficients, in line with the results characterizing the variable KKZ_MEAN, and second, the main implications of our baseline model (1) do not change.

23 In light of this result, we also tried to identify four possible country characteristics and investigated whether they had an impact on the probability for a bank to be in crisis. More precisely: (i) We selected four possible country characteristics based on: (1) the legal origin, (2) the degree of economic freedom, (3) the degree of political freedom and (4) our cluster analysis. (ii) For each characteristic, we formed groups of countries and allocated each country to a specific group according to the value of the characteristics chosen. (iii) Finally, we included dummies representing the groups of countries (and thus the common characteristics) in our bank-level base model (Eq.

4.2. Cross-bank results

In this section we present the empirical evidence corresponding to the cross-bank analysis described in Section 3.2.

We start by showing the results of the bank-level model (6). Interestingly, in column BL1 of Table 5 we notice that all the vari- ables used in the cross-country model (1), except CONCENTRATION, are also significant when measured using individual bank data. It is worth mentioning that CONCENTRATION was already weakly sig- nificant or insignificant at all in some of the cross-country estimates seen in the previous sections. In addition, consistently with the analysis provided by the cross-country models 2–9 (see Table 2), we estimate other seven bank-level probit models in which, in turn and one per time, we add to the five explanatory variables of Eq. (6) one of the following variables: COST_INCOME_BL, ROA_BL, ROE_BL, Z_SCORE_BL, CAPITAL, ENTRY, and SUPERVISION. Columns BL2-BL8 of Table 5 highlight that our micro-results are robust to the inclu- sion of such bank-level regressors as control variables, confirming also that the macro-financial factors we have identified above are indeed key determinants at micro-level as well.

The results of the random-intercept model described in Eq. (7) are instead shown in Table 6. We notice that the estimated stan- dard deviation ru (shown in the second part of the table labeled ‘‘Random-effects’’) is significant, being equal to 1.93 with a stan- dard error of 0.49, and represents the estimated standard deviation

22 See the definition of the two variables in Appendix B.

in the intercept. This result, together with the evidence provided by the likelihood-ratio test, highlights the presence of random effects at the country-level and suggests that all the unmeasured factors associated with each country affect the probability of bank crisis. Moreover, the first part of the table shows the estimated coefficients of the cross-bank variables we identified. Except CON- CENTRATION, they are all significant, meaning that our results are robust when considering unmeasured effects at the country level. In other words, on the one hand we showed that there exist unmeasured factors associated to each country that affect the probability for a bank to be in crisis. Nevertheless, when we take into account such factors, thus allowing each country to have a dif- ferent intercept, our results do not change qualitatively.23

5. Robustness checks

In this section we investigate whether our macro-results are sensitive to different definitions of the country crisis. Therefore, we perform two model specifications as robustness checks: an ordered probit model (in which the dependent variable is a dummy valued zero in case of no country crisis, one in case of borderline crisis, and two in case of systematic crisis), and a tobit model (in

(6)). Unfortunately, none of the proposed country-characteristics turned out to be significant. The corresponding analysis and results are not reported in the paper and are available upon request.

Table 5 Bank-level probit.

Variables Probit BL1 Probit BL2 Probit BL3 Probit BL4 Probit BL5 Probit BL6 Probit BL7 Probit BL8

NET_INTEREST_MARGIN_BL �0.1644*** �0.1582*** �0.1667*** �0.1691*** �0.1718*** �0.1627*** �0.1707*** �0.1658*** (0.0369) (0.0366) (0.0427) (0.0379) (0.0384) (0.0379) (0.0369) (0.037)

CREDIT_DEPOSIT_BL 0.0047*** 0.0039*** 0.0047*** 0.0047*** 0.0044*** 0.0046*** 0.0047*** 0.0045***

(0.0012) (0.0013) (0.0012) (0.0012) (0.0012) (0.0012) (0.0012) (0.0012) CONCENTRATION 0.0056 0.0054 0.0055 0.0051 0.0087* 0.0057 0.0059 0.0045

(0.0046) (0.0046) (0.0047) (0.0046) (0.0048) (0.0045) (0.0046) (0.0042) RESTRICTION �0.2064*** �0.2014*** �0.2067*** �0.2116*** �0.2290*** �0.2046*** �0.2106*** �0.2000***

(0.0431) (0.0442) (0.0429) (0.0433) (0.0482) (0.0432) (0.045) (0.042) PRIVATE MONITORING �0.1871*** �0.2037*** �0.1871*** �0.1901*** �0.1661*** �0.1801*** �0.1827*** �0.2114***

(0.0619) (0.0628) (0.0619) (0.0614) (0.0629) (0.0651) (0.0621) (0.0689) COST/INCOME_BL �0.0021

(0.0035) ROA_BL 0.0118

(0.0724) ROE_BL 0.0069

(0.007) ZSCORE_BL �0.0018

(0.0013) CAPITAL 0.0234

(0.0471) ENTRY 0.1726*

(0.093) SUPERVISION �0.0438

(0.0328) Constant 0.9948* 1.3006** 1.0006* 1.0142* 0.9886 0.7705 �0.3115 1.7138**

(0.5951) (0.6258) (0.5898) (0.6035) (0.6555) (0.7426) (0.9375) (0.799) Observations 755 744 755 754 684 755 755 755 Pseudo R-squared 0.2108 0.1911 0.2108 0.2148 0.2287 0.2113 0.2178 0.2147

Table shows, using information on individual data of the banks (micro-level analysis), the estimation of several probit models, starting from the baseline bank-level model (Eq. (6)) shown in Column Probit BL1. In all these models, the dependent variable is CRISIS_BL which is a dummy variable equal to 1 if the bank failed or received a recapitalization by the government during the crisis and 0 otherwise. Robust standard errors are reported in parenthesis. The corresponding methodology and variables are described in Section 3.2. * Statistical significance of the parameter at 10% significance level. ** Statistical significance of the parameter at 5% significance level. *** Statistical significance of the parameter at 1% significance level.

24 As reported in Laeven and Valencia (2010), there are three types of support: (i) liquidity support, (ii) gross restructuring costs, and (iii) asset purchases and guarantees. We disregard liquidity support for three reasons. First, strictly speaking, it is provided by the Central Bank and not by the Government. Second, as such, it is impossible to break it down by country across the Euro zone. Third, more importantly, liquidity support may be a really transient measure whose cost could be inherently limited. Thus, we use the conservative measure of Government support given by the sum of restructuring costs and asset purchases (we also disregard the entity of the guarantees as it is difficult to quantify what their real ex-post cost would actually be).

124 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

which the proxy of country crisis is the ratio of financial support offered by the government to the country’s GDP).

5.1. Borderline versus systemic country crisis

As with most situations in life, there is black and white but there is also grey. In fact, some countries crossed the Great Financial crisis quite well, while, among those bowing down, some suffered more. In this section we account for the different degree of intensity of the country crisis and distinguish between countries experiencing a borderline crisis and countries truly subject to a systemic crisis, as reported by Laeven and Valencia (2010). To accomplish this task, we estimate the same cross-country probit model (1) using a differ- ent dependent variable, i.e. ORDERED_CRISIS, which is a dummy equal to 1 if the country is classified as borderline crisis, 2 if the country is classified as systemic crisis and 0 otherwise.

Column Ord. Probit of Table 4 shows that the main results char- acterizing our cross-country analysis of the crisis are not sensitive to the distinction between borderline versus systemic crisis, thus reinforcing the role of our financial determinants. In other words, the important characteristics of the financial systems outlined in Sections 3.1 and 4.1 (maturity mismatch, business model, etc.) keep playing a crucial role also on countries not so severely affected by the crisis.

5.2. The cost of the crisis

As second robustness check, we consider an alternative proxy of the country crisis based on the extent to which the Government had to employ public finances to avoid the meltdown of the national banking system. This indicator reflects a certain type of costs associated with the crisis and captures the willingness of

the governments to help the private banking system overcoming the financial turmoil. We measure the country’s cost of the crisis as the sum of the various forms of support provided by the Govern- ment to the national banking system expressed in terms of the country GDP.24 By definition, this variable has a lower bound of zero in case of countries either not in crisis, or, alternatively, giving no financial support.

From an econometric standpoint, the lower bound on the range of the dependent variable suggests using a Tobit rather than an OLS specification. We follow this approach and estimate a Tobit model in which the dependent variable is CRISIS_COST_GDP, i.e. the coun- try’s cost of public support to the financial system as a ratio to its GDP, whereas the explanatory variables are the five determinants of the crisis shown in Eq. (1).

Column Tobit of Table 4 highlights that the results obtained with the baseline model (1) remain almost unchanged. The only difference is that PRIVATE MONITORING is no longer significant at the conventional levels (however, it has a p-value of 11%).

5.3. Multicollinearity

Our independent variables characterizing the cross-country model (1) draw on a quite heterogeneous set of financial

Table 6 Cross-bank analysis: random-intercept probit model.

Mixed- effects 1

Mixed- effects 2

Mixed- effects 3

Mixed- effects 4

Mixed- effects 5

Mixed- effects 6

Mixed- effects 7

Mixed- effects 8

Fixed-effects variables NET_INTEREST_MARGIN_BL �0.3813** �0.3631** �0.3793** �0.3983** �0.3994** �0.3781** �0.3969** �0.3840**

(0.1636) (0.163) (0.1664) (0.1643) (0.1662) (0.165) (0.1641) (0.1634) CREDIT_DEPOSIT_BL 0.0084*** 0.0076** 0.0084*** 0.0085*** 0.0078** 0.0084*** 0.0085*** 0.0084***

(0.003) (0.0033) (0.0031) (0.003) (0.0031) (0.0031) (0.003) (0.003) CONCENTRATION 0.0036 0.003 0.0037 0.0022 0.009 0.0043 0.0052 0.0013

(0.0203) (0.02) (0.0204) (0.0203) (0.0196) (0.0205) (0.0202) (0.0206) RESTRICTION �0.7790*** �0.7565*** �0.7791*** �0.7874*** �0.7758*** �0.7709*** �0.7851*** �0.7548***

(0.2907) (0.2865) (0.2909) (0.2895) (0.2797) (0.2933) (0.293) (0.2881) PRIVATE MONITORING �0.5702* �0.5873* �0.5710* �0.5710* �0.5279* �0.5288* �0.5595* �0.6012*

(0.3096) (0.3048) (0.3102) (0.3072) (0.2979) (0.3162) (0.3066) (0.3139) COST/INCOME_BL �0.0002

(0.0115) ROA_BL �0.0123

(0.1914) ROE_BL 0.0231

(0.0217) ZSCORE_BL �0.0022

(0.0029) CAPITAL 0.195

(0.2403) ENTRY 0.7781

(0.564) SUPERVISION �0.0836

(0.1616) Constant 5.6642 5.7775 5.6683 5.582 5.4415 3.9669 �0.259 6.8461

(4.0077) (3.9826) (4.0123) (3.9857) (3.8831) (4.4391) (5.5221) (4.6632) Random-effects

std. dev. ru 1.948869 *** 1.899072*** 1.950832*** 1.937124*** 1.841586*** 1.969257*** 1.901419*** 1.920985***

(0.4933981) (0.4834884) (0.4948426) (0.4905849) (0.4739024) (0.4976826) (0.4789264) (0.4910517) LR test vs. base regression

(Prob>=chibar2) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

Observations 755 744 755 754 684 755 755 755

Table shows the estimation of several random-intercept probit models based on the following: Prob(Yb,c = 1|Xb,c,uc) = H (a + bXb,c + zb,cuc), where c is the index identifying the country, b identifies the bank, Yb,c corresponds to the dummy variable CRISIS_BL, Xb,c is the set of bank-level variables described in Section 3.2 together with the country-level regulatory variables defined in Section 3.1, zb,c are the covariates corresponding to the random effects. Finally, the country error term is uc � N(0, r2u). Robust standard errors are reported in parenthesis. The corresponding methodology and variables are described in Section 3.2. * Statistical significance of the parameter at 10% significance level. ** Statistical significance of the parameter at 5% significance level. *** Statistical significance of the parameter at 1% significance level.

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 125

indicators. For example, on one side we have country-level banking variables like NET_INTEREST_MARGIN, COST_INCOME, ROA and ROE, observed annually and related to the aggregate balance sheet of the banks, so belonging to the category of financial performance indicators. On the contrary, the regulatory variables RESTRICTION, PRIVATE MONITORING, CAPITAL, ENTRY and SUPERVISION are the indices reported in the three surveys by Barth et al. (2004), and somehow reflect the countries’ financial background (structure). Given the different nature of our independent variables, one may suspect that, at the country level, the latter variables (‘‘i.e. struc- ture variables’’) could not only have a direct impact on the proba- bility for a country to be in crisis in 2008, but might also affect some of the ‘‘performance variables’’, giving rise to a problem of multicollinearity.

In this section, we investigate the issue of potentially-correlated regressors at the macro-level and show that, once multicollinearity is taken into account, our cross-country results still hold.

In order to detect the presence of multicollinearity, we start by computing the cross-country correlation between the independent variables used in the probit model (1). We find that all the correla- tion coefficients are smaller than |0.3| except that between the variables RESTRICTION and NET_INTEREST_MARGIN which is 0.38.25 This evidence suggests that, in general, the linear relationship between any pair of these macro-variables is quite weak, and most

25 Results are not reported in the paper but are available upon request.

importantly, that the ‘‘structure variables’’ do not have a big influ- ence on the ‘‘performance variables’’. However, the presence of a cor- relation coefficient close to 0.4 might induce a minimum suspect of multicollinearity between the variables NET_INTEREST_MARGIN and RESTRICTION.

To check whether this suspect is justified, we estimate a slightly different cross-country model based on a two-step procedure. We first regress NET_INTEREST_MARGIN on RESTRICTION and a con- stant, that is:

NET INTEREST MARGINc ¼ c0 þ c1 RESTRICTIONc þ nc; ð8Þ

then, we use the estimated residual n from (8) as explanatory vari- able in the following cross-country probit model:

ProbðCRISISc ¼ 1jXÞ¼Uðaþb1NIM RESIDUALc þb2 CREDIT DEPOSIT c þb3CONCENTRATIONc þb4 RESTRICTIONc þb5 PRIVATE MONITORINGcÞ; ð9Þ

where the variable NIM_RESIDUAL is indeed the estimated residual n from regression (8).26

The rationale of this two-step procedure is quite intuitive. By regressing NET_INTEREST_MARGIN on RESTRICTION, in fact, we are isolating the effects of regulatory restrictions on the interest

26 Both Eqs. (8) and (9) are estimated with heteroskedasticity robust standard errors.

126 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

margin, and let the residual capture all the other factors affecting the dependent variable.

Column M_Coll of Table 4 shows that substituting the variable NIM_RESIDUAL in lieu of NET_INTEREST_MARGIN in Eq. (9) does not alter the validity of our cross-country base specification. In fact, the estimated coefficient of NIM_RESIDUAL is negative and significant, while all the other regressors remain qualitatively unchanged.

6. Conclusions

Although there is a growing consensus on the principle that regaining stability – today as much as it did in the 1930s – requires better regulation of the marketplace (see D’Apice and Ferri, 2010), the agreement as to what ‘better’ might entail is far from being established. More of the same seems hardly a satisfactory response. In previous decades, banking systems underwent a deep transformation progressively moving away from a type of banking centered on personal relationships to hinge on more standardized and impersonal approaches. The change, theorized by some con- sultants and academics, and designed to reach unprecedented high returns, prescribed gearing the banks with financial markets and modifying the bank business model. This had a notable impact on the transformation of some banking systems, in particular those that shifted toward a new business model (that is originate-to-dis- tribute, OTD), while others remained fastened to the traditional business model (that is originate-to-hold, OTH). Our evidence sug- gests that a more traditional banking system had a lower probabil- ity to be in crisis in 2008. Thus, a return to an old style banking, like the one prevailing during the Quiet Period in the US after the Great Depression (Gorton, 2009), is being considered in some quarters.

Furthermore, the results stated in this paper, both at the coun- try-level and at the bank-level, provide additional fuel to that dis- cussion and can help policymakers calibrate new regulations, by achieving a reasonable trade-off between financial stability and economic growth, and contribute to extend the analytical toolkit available for macro-prudential supervision and reforming micro- prudential regulation. In fact, the traditional banking business model, that is banks with higher net interest margin, has proved resilient through the crisis. At the same time, the higher capital levels prescribed by Basel III will penalize commercial banks. The need for traditional banks to increase their own funds will have two consequences. First, since capital is costly and there is a race for deposits, banks will have to increase the price of their loans, making credit more expensive with negative consequences on growth and no additional positive effects on stability. Second, there will be a tendency to reduce lending in order to shrink the denom- inator of capital ratios (de Larosière, 2011). Also, higher capital requirements, in response to the vulnerabilities highlighted in the paper, require some care. In recent years, both countries featur- ing traditional approaches to banking, such as in Ireland, Spain, and

CRISIS dummy equal to 1 if the country is clas Valencia (2010) and 0 otherwise.

NET_INTEREST_MARGIN annual mean from 1998 to 2006 of the its interest-bearing assets (source: Beck

ROA and ROE ROA = annual mean of return on assets mean of return on equity (net income

COST_INCOME annual mean value of total costs as a s (Source: Beck et al. (2010))

Z-SCORE annual mean of aggregate bank z-score ratio of return on assets plus capital-as

CREDIT_DEPOSIT annual mean from 1998 to 2006 of priva saving deposits in deposit money bank

Eastern Europe, and those which moved to a more securitized approach, as the US, plunged into serious crises. The fact that cap- ital was mostly not significant should give regulators pause: like doctors who used leeches (and thought that they were helping patients), or econometricians who go into a dark room looking for a black cat that is not there and scream ‘I’ve got it’, they may be settling for a popular cure in lieu of the one that actually works. And as Laeven and Levine (2009) find, the impact of capital requirements might vary with the ownership structure of banks.

Finally, low interest rate contexts, for instance after the bust of the dotcom and subprime bubbles, could lead banks to search for yields in more complex non-traditional activities that increase their exposure to new risks. Regulators should also consider that, compared to traditional credit risks, evaluating these new risks is more difficult, especially those related to the complex financial contracts so deeply entrenched in the OTD model, which played a major role in the recent crisis. This suggests ending or reversing prolonged periods of low interest rates, the typical background of the broader regulatory framework artificially boosting securitiza- tion, and that measures to divulge and verify more information about risk taking in banking (so that market monitoring might finally work) are desiderable.

To conclude, the Authorities should carefully ponder the finan- cial determinants of the Great Crisis, especially now that a major increase in minimum bank capital is being enforced within the framework of Basel 3 and micro-and-macro-prudential regulation is to be implemented in most countries.

Appendix A. Sample information

Countries that experienced a Systemic Crisis are: Austria, Bel- gium, Denmark, Germany, Iceland, Ireland, Latvia, Luxembourg, Netherlands, United Kingdom, USA.

Countries that experienced a Borderline Crisis: Greece, Hun- gary, Kazakhstan, Portugal, Slovenia, Spain, Sweden, Switzerland.

Other countries in the sample are: Argentina, Australia, Bahrain, Bangladesh, Belize, Bolivia, Botswana, Brazil, Bulgaria, Burundi, Canada, Chile, Colombia, Costa Rica, Croatia, Czech Republic, Egypt, El Salvador, Estonia, Finland, Guatemala, Guyana, Honduras, Hong Kong, India, Indonesia, Israel, Italy, Japan, Jordan, Kenya, Korea, Kuwait, Lithuania, Macau, Macedonia, Malaysia, Mali, Malta, Mau- ritius, Moldova, Morocco, New Zealand, Norway, Oman, Pakistan, Panama, Papua New Guinea, Peru, Philippines, Poland, Saudi Arabia, Senegal, Singapore, Slovakia, South Africa, Sri Lanka, Swazi- land, Thailand, Togo, Trinidad and Tobago, Tunisia, Turkey, Uruguay.

Appendix B

Definition of the variables used in the cross-country analysis.

sified as either borderline crisis or systemic crisis by Laeven and

accounting value of the bank’s net interest revenue, as a share of et al., 2010) (net income to total assets) from 1998 to 2006. ROE = annual

to total equity) from 1998 to 2006 (Source: Beck et al. (2010)) hare of total income of all country’s banks from 1998 to 2006

from 1998 to 2006 (Source: Beck et al. (2010)). The z-score is the set-ratio to the 5-years standard deviation of return on assets te credit by deposit money banks as a share of demand, time and

s (Source: Beck et al. (2010))

CONCENTRATION annual mean from 1998 to 2006 of the share of the country’s three largest banks in all country’s bank assets (Source: Beck et al. (2010))

BANK_ASSETS_GDP annual mean of the ratio between the country’s deposit money bank assets and its GDP from 1998 to 2006 (Source: Beck et al. (2010))

INT_DEBT_ISS_GROSS_GDP annual mean from 1998 to 2006 of the gross flow of international bond issues by the country scaled by its GDP (Source: Beck et al. (2010))

RESTRICTION mean value of the ‘‘Overall Restrictions’’ index reported in the three surveys by Barth et al. (2004). This index measures the degree to which banks face regulatory restrictions on their activities in: (a) securities markets, (b) insurance, (c) real-estate, and (d) owning shares in non-financial firms. The index can take values from 0 to 4 for each of these four sub-categories, where 4 indicates the most restrictive regulations on this sub-category of bank activity. Thus, the index of overall restrictions can potentially range from 0 to 16

PRIVATE MONITORING Mean value of the ‘‘Private Monitoring’’ index reported in the three surveys by Barth et al. (2004). The index measures the degree to which regulations empower, facilitate, and encourage the private sector to monitor banks. It reflects the information on whether: (1) bank directors and officials are legally liable for the accuracy of information disclosed to the public, (2) banks must publish consolidated accounts, (3) banks must be audited by certified international auditors, (4) 100% of the largest 10 banks are rated by international rating agencies, (5) off-balance sheet items are disclosed to the public, (6) banks must disclose their risk management procedures to the public, (7) accrued interest/principal, though unpaid, enter the income statement while the loan is still non-performing, (8) subordinated debt is allowable as part of capital, and (9) there is no explicit deposit insurance system and no insurance was paid the last time a bank failed. The private monitoring index has a minimum value of 0 and a maximum value of 9, where larger numbers are associated with a greater regulatory empowerment of private monitoring of banks

CA PITAL Mean value of the ‘‘Capital Regulation’’ index reported in the three surveys by Barth et al. (2004). This index includes information on (1) the extent of regulatory requirements regarding the amount of capital banks must hold and (2) the stringency of regulations on the extent to which the source of funds that count as regulatory capital can include assets other than cash or government securities, borrowed funds, and on whether the regulatory/supervisory authorities verify the sources of capital. Large values indicate more stringent capital regulations

ENTRY Mean value of the ‘‘Entry Requirements’’ index reported in the three surveys by Barth et al. (2004). The index essentially counts the number of requirements for obtaining a banking license: (1) draft by-laws; (2) intended organizational chart; (3) financial projections for first 3 years; (4) financial information on main potential shareholders; (5) background/experience of future directors; (6) background/experience of future managers; (7) sources of funds to be used to capitalize the new bank; and (8) market differentiation intended for the new bank

SUPERVISION Mean value of the ‘‘Official Supervisory’’ index reported in the three surveys by Barth et al. (2004). This index measures the degree to which the country’s commercial bank supervisory agency has the authority to take specific actions. It is determined by the information provided on the following features of official supervision: (1) does the supervisory agency have the right to meet with external auditors about banks? (2) are auditors required to communicate directly to the supervisory agency about elicit activities, fraud, or insider abuse? (3) can supervisors take legal action against external auditors for negligence? (4) can the supervisory authority force a bank to change its internal organizational structure? (5) are off-balance sheet items disclosed to supervisors? (6) can the supervisory agency order the bank’s directors or management to constitute provisions to cover actual or potential losses? (7) can the supervisory agency suspend the directors’ decision to distribute: a) dividends? b) bonuses? c) management fees? (8) can the supervisory agency supersede the rights of bank shareholders-and declare a bank insolvent? (9) can the supervisory agency suspend some or all ownership rights? (10) can the supervisory agency: a) supersede shareholder rights? b) remove and replace management? c) remove and replace directors? The official supervisory index has a minimum value of 0 and a maximum value of 14, where larger numbers indicate a greater power

KKZ_VOICE Mean value of ‘‘Voice and Accountability’’ index from 1998 to 2006. This index reflects perceptions of the extent to which a country’s citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media. This index ranges from �2.5 (weak) to 2.5 (strong) governance performance.

KKZ_POLSTAB mean value of ‘‘Political Stability and Absence of Violence’’ index from 1998 to 2006. This index reflects perceptions of the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, including politically-motivated violence and terrorism. This index ranges from �2.5 (weak) to 2.5 (strong) governance performance

KKZ_GOVEFF mean value of ‘‘Government Effectiveness’’ index from 1998 to 2006. This index reflects perceptions of the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the

(continued on next page)

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 127

government’s commitment to such policies. This index ranges from �2.5 (weak) to 2.5 (strong) governance performance.

KKZ_REGQUAL Mean value of ‘‘Regulatory Quality’’ index from 1998 to 2006. This index reflects perceptions of the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development. This index ranges from �2.5 (weak) to 2.5 (strong) governance performance

KKZ_RULELAW Mean value of ‘‘Rule of Law’’ index from 1998 to 2006. This index reflects perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence. This index ranges from �2.5 (weak) to 2.5 (strong) governance performance

KKZ_CONCORR mean value of ‘‘Control of Corruption’’ index from 1998 to 2006. This index reflects perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ‘‘capture’’ of the state by elites and private interests. This index ranges from �2.5 (weak) to 2.5 (strong) governance performance

KKZ_MEAN Mean value of the six measures provided by Kaufman et al. (2010) from 1998 to 2006

128 G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129

References

Acemoglu, D., Johnson, S., 2005. Unbundling institutions. Journal of Political Economy 113 (5), 949–995.

Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R., 2003. Fractionalization. Journal of Economic Growth 8, 155–194.

Allen, F., Gale, D., 2000a. Comparing Financial System. MIT Press, Cambridge, MA. Allen, F., Gale, D., 2000b. Financial contagion. Journal of Political Economy 108, 1–

33. Allen, F., Gale, D., 2004. Asset price bubbles and monetary policy. In: Desai,

Meghnad, Said, Yahia (Eds.), Global Governance and Financial Crisis. Routledge, London, pp. 19–42.

Allen, F., Carletti, E., 2008. The Role of Liquidity in Financial Crises. Jackson Hole Conference Proceedings, Federal Reserve Bank of Kansas City, pp. 379–412.

Barth, J., Caprio Jr., G., Levine, R., 2004. Bank regulation and supervision: what works best? Journal of Financial Intermediation 13, 205–248.

Barth, J., Caprio Jr., G., Levine, R., 2006. Rethinking Bank Regulation: Till Angels Govern. Cambridge University Press.

Barth, J., Caprio Jr., G., Levine, R., 2012. The Guardians of Finance: Getting Them to Work for Us. MIT Press.

Basel Committee on Banking Supervision (BCBS), 2010a. Basel III: A Global Regulatory Framework for More Resilient Banks and Banking Systems.

Basel Committee on Banking Supervision (BCBS), 2010b. An Assessment of the Long-Term Economic Impact of the New Regulatory Framework.

Beck, T., Demirgüç-Kunt, A., Levine, R., 2000. A new database on financial development and structure. World Bank Economic Review 14, 597–605.

Beck, T., Demirgüç-Kunt, A., Levine, R., 2010. Financial institutions and markets across countries and over time: the updated financial development and structure database. World Bank Economic Review 24 (1), 77–92.

Beck, T., Demirgüç-Kunt, A., Levine, R., 2006. Bank concentration, competition and crises: first results. Journal of Banking and Finance 30, 1581–1603.

Beltratti, A., Stulz, R.M., 2012. Why did some banks perform better during the credit crisis? A cross-country study of the impact of governance and regulation. Journal of Financial Economics 105, 1–17.

Berndt, A., Gupta, A., 2009. Moral hazard and adverse selection in the originate-to- distribute model of bank credit. Journal of Monetary Economics 56, 725–743.

Boyd, J., De Nicolo, G., Jalal, A., 2005. The Theory of Bank Risk Taking and Competition Revisited. IMF WP.

Brunnermeier, M., 2009. Deciphering the liquidity and credit crunch 2007–08. Journal of Economic Perspectives 23 (1), 77–100.

Buiter, V., 2007. Lessons from the 2007 Financial Crisis. Background Paper Submitted to the UK Treasury Select Committee, December 11, 2007.

Caprio, G. Jr., 2010. Financial Regulation in a Changing World: Lessons from the Recent Crisis. In: Paolo Savona (Ed.), Financial Collapse: How are the Biggest Nations and Organizations Managing the Crisis? Fondazione Cesifin ‘‘Alberto Predieri’’.

Carletti, E., 2010. Competition, Concentration and Stability in the Banking Sector. IstEin WP No. 8.

Cecchetti, S.G., 2009. Crisis and responses: the federal reserve in the early stages of the financial crisis. Journal of Economic Perspectives 23 (1), 51–76.

Claessens, S., Köse, M.A., Terrones, M.E., 2010. The global financial crisis: how similar? How different? How costly? Journal of Asian Economics 21, 247–264.

Collins, S.M., 2005. Comments on financial globalization, growth, and volatility in developing countries, by Eswar Prasad, Kenneth Rogoff, Shang-Jin Wei, and M. Ayhan Kose. In: Ann Harrison (Ed.), Globalization and Poverty.

Coval, J., Jurek, J., Stafford, E., 2009. The economics of structured finance. Journal of Economic Perspectives 23 (1), 3–25.

D’Apice, V., Ferri, G., 2010. Financial Instability: Toolkit for Interpreting Boom and Bust Cycle. Palgrave Macmillan Studies in Banking and Financial Institutions, London.

De Jonghe, O., 2010. Back to the basics in banking? A micro-analysis of banking system stability. Journal of Financial Intermediation 19, 387–417.

de Larosière, J., 2011. Don’t Punish the Banks That Performed Best. Financial Times, March 3.

De Michelis, A., 2009. Overcoming the Financial Crisis in the United States. OECD Economics Department W P No. 669.

Demirguc-Kunt, A., Detragiache, E., 1998. Financial Liberalization and Financial Fragility. World Bank Policy Research Paper No. 1917.

Demirguc-Kunt, A., Serven, L., 2009. Are the sacred cows dead? Implications of the financial crisis for macro and financial policies. World Bank Research Observer 25, 91–124.

DeYoung, R., Torna, G., 2013. Nontraditional banking activities and bank failures during the financial crisis. Journal of Financial Intermediation 22, 397–421.

Faria, A., Mauro, P., 2005. Institutions and the External Capital Structure of Countries. IMF W P No. 04/236.

Ferri, G., 2001. Capital Adequacy Requirements: Impact and Evolution. Opening Remarks, As Guest Editor, of the Special Issue (no. 3/2001) Economic Notes, pp. 319–326.

Financial Services Authority (FSA), 2009. The Turner Review a Regulatory Response to the Global Banking Crisis.

Gambacorta, L., Marques-Ibanez, D., 2011. The bank lending channel: lessons from the crisis. Economic Policy 26 (66), 135–182.

Giannone, D., Lenza, M., Reichlin, L., 2011. Market freedom and the global recession. IMF Economic Review 59, 111–135.

Glaeser, E.L., La Porta, R., Lopez-De-Silanes, F., Shleifer, A., 2004. Do institutions cause growth? Journal of Economic Growth 9, 271–303.

Gorton, G.B., 2009. Slapped in the face by the invisible hand: banking and the panic of 2007. Prepared for the Federal Reserve Bank of Atlanta’s 2009 Financial Markets Conference: Financial Innovation and Crisis, May 11–13, 2009.

Huang, R., Ratnovski, L., 2009. Why are Canadian Banks More Resilient? IMF WP/09/ 152.

Institute of International Finance (IIF), 2010. Interim Report on the Cumulative Impact on the Global Economy of Proposed Changes in the Banking Regulatory Framework.

International Monetary Fund (IMF), 2012. Global Financial Stability Report (October).

Kaufman, D., Kraay, A., Mastruzzi, M., 2010. The Worldwide Governance Indicators: A Summary of Methodology, Data and Analytical Issues. World Bank Policy Research Paper No. 5430.

Kose, M.A., Prasad, E., Rogoff, R., Wei, S.-J., 2006. Financial Globalization: A Reappraisal. IMF WP 06/189.

Laeven, L., Levine, R., 2009. Bank governance, regulation and risk taking. Journal of Financial Economics 93, 259–275.

Laeven, L., Valencia, F., 2010. Resolution of Banking Crises: The Good, the Bad, and the Ugly. IMF WP 10/146.

La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1997. Legal determinant of external finance. The Journal of Finance 52, 1131–1150.

Macroeconomic Assessment Group of the Financial Stability Board and Basel Committee on Banking Supervision (MAG), 2010. Assessing the Macroeconomic Impact of the Transition to Stronger Capital and Liquidity Requirements. Bank for International Settlements.

Mehrez, G., Kaufman, D., 2000. Transparency, Liberalization and Banking Crises. World Bank Policy Research p No. 2286.

Merrouche, O., Nier, E., 2010. What Caused the Global Financial Crisis? Evidence on the Drivers of Financial Imbalances 1999–2007. IMF WP/10/265.

Mian, A., Sufi, A., 2009. The consequences of mortgage credit expansion: evidence from the U.S. mortgage default crisis. Quarterly Journal of Economics 124, 1449–1496.

Ponczek, V., Mattos, E., 2009. Institutions and economic volatility: democracy versus risk of expropriation. Journal of Economic Studies 36 (2), 184–194.

G. Caprio Jr. et al. / Journal of Banking & Finance 44 (2014) 114–129 129

Rose, A., Spiegel, M., 2009. The causes and consequences of the 2008 crisis: early warning. Japan and the World Economy 24 (1), 1–16.

Shehzad, C.T., De Hann, J., 2009. Financial Reform and Banking Crises. CESifo WP No. 2870.

Spiegel, M., 2008. Financial Globalization and Monetary Policy Discipline: A Survey with New Evidence from Financial Remoteness. Federal Reserve Bank of San Francisco, WP 2008–10, July.

Stiglitz, J.E., 2010. Freefall: America, Free Markets, and the Sinking of the World Economy. WW Norton & Company, NY, USA.

Trichet, J., 2009. What Lessons Can be Learned from the Economic and Financial Crisis? Speech at the 5e Recontres de l’Entreprise Européenne, Paris, France, March 17.

Tytell, I., Wei, S.-J., 2005. Global Capital Flows and National Policy Choices. IMF, mimeo.

Wei, S.-J., 2006. Connecting two views on financial globalization: can we make further progress? Journal of the Japanese and International Economies 20, 459– 481.