The fall of Spanish cajas: Lessons of ownership and governance for banks

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

Journal of Financial Stability xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Journal of Financial Stability journal homepage: www.elsevier.com/locate/jfstabil

The fall of Spanish cajas: Lessons of ownership and governance for banks夽 Alfredo Martín-Oliver a,∗ , Sonia Ruano b , Vicente Salas-Fumás c a b c

Universitat de les Illes Balears, Spain Banco de Espa˜ na, Spain Universidad de Zaragoza, Spain

a r t i c l e

i n f o

Article history: Received 29 October 2015 Received in revised form 16 February 2017 Accepted 16 February 2017 Available online xxx JEL classification: G21 Keywords: Ownership of banks Governance Banking crisis Spain Cajas Business models

a b s t r a c t Ownership, governance, and institutional diversity among banks are a subject of public and regulatory concern. This paper addresses this issue by using a case study of Spain, where the retail banking market was split evenly between shareholder and stakeholder banks before the crisis. We examine how institutional diversity mattered in the accumulation of risk in the pre-crisis years, in the severity of losses caused by the crisis, and in the resilience to recover from the losses. The method of analysis consists in linking the risk position of the banks in the pre-crisis period and the losses arising during the crisis to the decisions of banks to migrate from business models based on deposit financing to models based on market-debt financing. We find that cajas migrated to more vulnerable business models following the strategy of the shareholder banks, but the losses in the crisis were much higher in the former than in the latter. The paper interprets this result as evidence that what matters the most about the ownership of banks is their resilience in bad times. © 2017 Elsevier B.V. All rights reserved.

1. Introduction The research claims that the ownership and governance of banks are greatly responsible for the causes and consequences of the financial crisis (Kirkpatrick, 2009; Mulbert, 2010; Berger et al., 2012; Hopt, 2013). However, the reforms proposed to improve corporate governance do not converge toward common ground on what good governance means. In this respect, the Walker Report (2009) for the United Kingdom adopts the shareholders’ perspective on good governance and recommends that banks should be managed under the single goal of profit maximization. The

夽 This paper is the sole responsibility of its authors, and the views represented ˜ or the Eurosystem. here do not necessarily reflect those of the Banco de Espana We thank Iftekhar Hasan, Managing Editor, and two anonymous referees or their constructive comments, which helped us to improve the manuscript. We also ˜ for their comments to an earlier version of thank Jesús Saurina and Carlos Ocana the paper. Alfredo Martín-Oliver acknowledges the financial support from project MEC-ECO2013-44409, and Vicente Salas-Fumás acknowledges support from the research project ECO2009-13158 MICINN-PLAN NACIONAL I+D and from the DGAFSE through the CREVALOR research group. Any errors in the paper are the authors’ only responsibility. ∗ Corresponding author at: Universitat de les Illes Balears, Ctra. Valldemossa km. 7.5, 07122 Palma de Mallorca, Islas Baleares, Spain. E-mail address: [email protected] (A. Martín-Oliver).

Basel Committee (2010) and the European Commission (2010) adopt a different view and recommend that the banks’ boards and senior managers perform their duties while taking into account the interests of shareholders, that is, depositors and other relevant stakeholders. The Liikanen (2012) report to the European Commission praises institutional diversity as the best organizational structure for the banking industry. The diversity of views on what good governance means for banks reflects the lack of robust evidence on how the form of ownership and governance mechanisms affect the individual performance of banks and, more importantly, the stability of the whole industry.1 This paper draws from the Spanish banking crisis to examine the relevance of ownership and governance for financial stability in banks. The Spanish case is worth studying because of the severity of the crisis and because the banking industry was a good case of institutional diversity before the outburst of the crisis. The banking industry is split between for-profit and not-for-profit banks controlled by stakeholders.2 The latter are called cajas. In the crisis, the ex-post losses of government aid to compensate for losses in

1

For reviews, see Berger et al. (2005) and Butzbach and Mettenheim (2014). Shareholder banks and cajas together concentrate more than 90% of the total bank assets; the rest of the banks’ assets belong to credit cooperatives, subsidiaries, and branches of foreign banks. 2

http://dx.doi.org/10.1016/j.jfs.2017.02.004 1572-3089/© 2017 Elsevier B.V. All rights reserved.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17 2

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

the value of banks’ assets were much higher for cajas than for shareholder banks: 85% of banks’ total assets seriously damaged by the crisis belong to cajas and only 15% to corporate banks. These high losses, together with some notorious cases of banking malpractices, turned into a loss of social confidence in the cajas. Today, after some regulatory reforms that transformed cajas into shareholder banks, the traditional institutional diversity of Spanish banking has disappeared.3 Our main interest in this paper is to understand the cajas’ responsibility for the crisis, why they were more damaged than for-profit banks, and why some cajas performed differently than others. However, we conduct this analysis of cajas in the context of changes occurring in the whole banking industry since the Euro. For this reason, the first part of the paper highlights the changes in all Spanish banks from deposit-dependent to market-dependent financing of bank credit (Almazán et al., 2015). The sample period starts in 1999 and finishes in 2007, the year previous to the crisis. The analysis evaluates how the migration of banks to models more dependent on market financing made them more vulnerable to negative external shocks. We investigate the possible reasons for the migration to new business models and measure the effect of the external shock of the crisis on the performance of banks in each business model. Next, the paper focuses on the comparison between shareholder banks and cajas and why cajas suffered more. We identify two possible explanations for why the cajas did not survive the financial crisis. One states that the uniqueness of the cajas’ ownership and governance led them to make business decisions different from those made by shareholders banks when the Euro removed the constraints to market finance for all Spanish banks. Thus, the higher losses for cajas might be due to different ex-ante critical business decisions. The other hypothesis is that the business decisions previous to the crisis were similar in cajas and banks, but cajas adjusted less effectively to the external shock of the crisis. The lower capability of cajas to adjust and respond to an external shock might be a consequence of their unique ownership and governance features. The empirical evidence in this paper rejects the hypothesis that the cajas behaved differently from banks in the pre-crisis period, and supports that the higher losses for cajas can be explained because their unique ownership and governance made them less resilient to external shocks. Our results can be of interest to regulators because they provide insights about the ownership and governance of banks as mediators in the transmission of external shocks to financial stability. We also examine the effects of the interaction between the form of ownership and the business models of banks (Ayadi et al., 2011, 2012; Llewellyn, 2013; Köhler, 2015) on the build-up of risk during the period of economic expansion and on the losses caused by those risks. The conclusion is that business models alone are not sufficient to fix selective regulatory requirements. Although Spain lost institutional diversity in the banking industry, there are other countries where saving banks and credit cooperatives continue to play a relevant role in the provision of banking services for which the regulatory lessons from our research are relevant. The rest of the paper is organized as follows: Section 2 contains a review of the literature and outlines the paper in the context of the research. Section 3 presents the results of the cluster analysis used to identify the business models of Spanish banks, including a comparison of clusters in operational decisions and performance. Section 4 examines the migration of Spanish banks toward riskier business models and evaluates the ex-post losses in each business model during the crisis. Section 5 analyzes the differences between cajas and corporate banks in choosing business models as well as

3 The fall of Spanish cajas in the crisis is more striking because for many years, cajas profitably gained market share at the expense of banks.

the heterogeneity in governance among cajas as it relates to ex-ante risk decisions and ex-post losses. Finally, the conclusion summarizes the main results of the paper. 2. Literature review and research design 2.1. Literature review This section reviews the literature that studies the effect of ownership and governance on the banks’ risk-taking decisions. Banks, like any business firm, make decisions that involve a trade-off between risk and return. The legal status of limited liability corporations, which bounds the amount of losses but leaves the benefits unbounded, creates incentives for shareholders to make decisions with excessive risks in relation to the expected returns. This phenomenon of risk shifting in nonfinancial firms (Jensen and Meckling, 1976) is exacerbated in banks because their business model is highly leveraged by nature with a large fraction of insured debt (deposits) (Merton, 1977; Bhattacharya and Thakor, 1993), and because of easy risk manipulation (Myers and Rayan, 1998). Additionally, the private costs of bankruptcy to shareholders of banks from risk-taking decisions are much lower than the social ones.4 The singularity of banks justifies the study of the effect of ownership and governance on performance, as opposed to the general literature on this topic.5 In the research on ownership, one line of interest has been the comparison of the contributions to financial stability of stateowned banks versus private banks.6 On the positive side, state ownerships could lead politicians to control the decisions of banks in such a way that the social negative effects of risk shifting are internalized to induce higher financial stability. On the negative side, if bureaucrats pursue their own interests that do not necessarily coincide with social interests, state ownership can be a source of additional fragility in banks (La Porta et al., 2002). Within privately-owned banks, there are different types of ownership forms: cooperatives, saving banks, and shareholder banks. Cooperatives and saving banks are a priori candidates to have less risk-taking than shareholder banks because they have a more stable deposit and customer base, focus on capital preservation, and they care more about consumer surplus than profits. Moreover, in a perfect world, shareholders of banks directly benefit from risk-shifting with higher shares prices. Within for-profit banks, the concentration of shareholdings can also affect risk-shifting. Managers of banks with dispersed shareholdings receive a small fraction of the private benefits from risk shifting, and they are more conservative in making risky decisions, compared to banks with large shareholders that have direct influence on the banks’ decisions. The empirical evidence confirms these predictions. Hesse and Cihák (2007), in a sample of banks from OECD countries, find that cooperative banks are more stable than banks owned by corporate shareholders. Iannotta et al. (2007) find that cooperative banks are the most stable banking group, followed by shareholder banks and state-owned banks, after examining a sample of large European banks. García-Marco and Robles-Fernández (2008) find that Spanish saving banks are less exposed to financial risks than shareholder

4 The difference between private and social costs of bank failures have to do with the external effects ignored by banks’ owners and managers, including the disruptions to the payment system, disruptions in credit flows, and contagion effects. Laeven and Valencia (2013) estimate average fiscal costs of resolving banking crises in the past 40 years as 13% of countries GDP. 5 See Adams and Mehran (2003) for the US, and Mulbert (2010) in a more general context. 6 La Porta et al. (2002) find that, on average, more than 41% of the assets of the largest ten banks in each country were either owned or controlled by the state in 1995.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

3

Fig. 1. Time evolution of weighted average of variables used to identify business models.

banks. Beck et al. (2009), with data from German banks, also find that shareholder banks take more risks than savings and cooperative banks. Finally, the empirical evidence also confirms that banks tend to be riskier if they are controlled by large shareholders instead of by managers (Saunders et al., 1990; Anderson and Fraser, 2000 for early evidence; and Laeven and Levine, 2009; Beltratti and Stulz, 2012, for evidence on the recent crisis). As for the characteristics of boards and the risk-taking of banks, there is no clear evidence that the size of boards and the ratio of outside directors affect banks’ performance and risk-taking behavior (Adams and Mehran, 2003; Minton et al., 2012). However, the power and influence of risk committees on boards is associated

with lower risk in banks (Ellul and Yeramilli, 2013). Some papers find a positive relation between risk-taking behavior and the general and specific human capital on the board of directors. Minton et al. (2012) find a positive association between the experience of independent directors and the volatility of shareholder returns. ˜ and Garicano (2010), García-Meca and On the other hand, Cunat Sánchez-Ballesta (2014) and García-Cestona and Sagarra (2014) do not find conclusive evidence on the influence of human capital of the chairman of the board on the performance of Spanish cajas. Hau and Thum (2009) find that a lack of financial experience of board members in German banks positively relates to bank losses. Overall,

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

4

Table 1 Average values of the variables used in the analysis by cluster: 1992–2007. Period 1992–94

Equity/Assets Loans/Assets Loans/Deposits Net Interbank/Assets Bank Loans/Assets Securitiz/Assets Observations Share in total assets

Cluster 1

Cluster 2

Cluster 3

Cluster 4

0.067 0.299+++ 0.447+++ 0.238+++ 0.310+++ 0.000 56 0.167

0.061+ 0.493++ 0.690+++ 0.116+++ 0.173++ 0.000 168 0.730

0.074+ 0.508++ 1.341+++ −0.050++ 0.131++ 0.000 36 0.085

0.060 0.389+++ 2.377+++ −0.075++ 0.179+ 0.000 16 0.018

Cluster 1

Cluster 2

Cluster 3

Cluster 4

0.065+ 0.353+++ 0.517+++ 0.195+++ 0.273+++ 0.000 162 0.221

0.059+ 0.552++ 0.862+++ 0.051+++ 0.144+ 0.000 297 0.760

0.059 0.488++ 2.435+++ −0.079++ 0.154+ 0.000 35 0.014

0.064 0.516+ 4.306+++ −0.106++ 0.130+ 0.000 28 0.005

Period 1995–98

Equity/Assets Loans/Assets Loans/Deposits Net Interbank/Assets Bank Loans/Assets Securitiz/Assets Observations Share in total assets

Period 1999–2002

Equity/Assets Loans/Assets Loans/Deposits Net Interbank/Assets Bank Loans/Assets Securitiz/Assets Observations Share in total assets

Cluster 1

Cluster 2

Cluster 3

Cluster 4

0.063 0.436+++ 0.665+++ 0.094+++ 0.252+++ 0.005+ 178 0.286

0.060 0.692++ 1.016+++ −0.012+++ 0.097+++ 0.019+++ 195 0.674

0.066 0.696++ 1.760+++ −0.176+++ 0.141++ 0.010+ 43 0.035

0.062 0.609+++ 3.231+++ −0.249+++ 0.176++ 0.000 24 0.005

Cluster 2

Cluster 3

Cluster 4

Period 2003–2007 Cluster 1 Equity/Assets Loans/Assets Loans/Deposits Net Interbank/Assets Bank Loans/Assets Securitiz/Assets Observations Share in total assets

+

0.052 0.282+++ 0.635+++ 0.095++ 0.408+++ 0.001+++ 57 0.050

+++

0.060 0.698+++ 1.064+++ 0.028++ 0.101++ 0.015+++ 168 0.230

++

0.051 0.768++ 1.518+++ −0.061+++ 0.075+++ 0.035++ 149 0.410

0.046++ 0.754++ 2.299+++ −0.129+++ 0.095++ 0.077++ 85 0.310

Differences in mean values of the variables across clusters have been tested. The number of cross super-indexes (+ , ++ , +++ ) indicates the number of bilateral comparisons for which the average of a cluster is statistically different at the 10% significance level. Cluster 1 refers to the retail-deposits business model, cluster 2 to the retail-balanced model, cluster 3 to the retail-diversified model, and cluster 4 to the retail-market model.

the evidence on the human capital of directors and the risk-taking of banks is mixed. Several papers study the link between compensation and the risk-taking of banks in the pre-crisis period and the level of losses during the post-crisis period. Fahlenbrach et al. (2012) find that banks with a higher loss in share prices had CEOs with incentives in better alignment with the interests of shareholders. Cheng et al. (2010) find a strong correlation between different measures of risk-taking and managerial compensation in banks that size cannot explain. The authors interpret the results as evidence that shareholders intentionally supported risk-taking, which is consistent with their incentives to shift risks. 2.2. Our research Spanish cajas are savings banks that fall into the category of private not-for-profit commercial banks (Hansmann, 2009): cajas sell banking products under competitive conditions, do not receive external financial aid, and their directors hold the residual decision rights but cannot appropriate the residual profits. Cajas can do the

same banking activities as shareholder banks, and they can operate under the same regulatory conditions. Given that cajas do not have shareholders, they cannot issue new equity, and regulatory capital required for compliance must come from retained earnings. Earnings not retained in the business must be dedicated to social activities, in the broad sense, although regulations set a limit on the earnings spent on social activities to a maximum of 50% of accounting net profits. The residual (ownership) decision rights of cajas are held by the general assembly of elected members, by the members of the board of directors, and by the members of the board of control. The members of each governance body are representatives elected or nominated from different stakeholder groups: founding entity (nominated), depositors (elected), employees (elected), public authorities, city halls, and autonomous governments (nominated). The inside directors and managers of the cajas are not members of the governance bodies, and when they attend the meetings, they do so with a voice but without a vote. Once elected or nominated, all the members of the governance bodies are in place

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

5

Table 2 Transition matrix of banks across business models, BM: 2003–2007 compared to 1999–2002. Business Model in 2003–2007

Business Model in 1999–2002

BM1

BM2

BM3

BM4

Total

BM1 BM2 BM3 BM4

28.6 4.20 0 0

42.9 10.4 7.70 14.3

17.9 56.3 15.4 0

10.7 29.2 76.9 85.7

100 100 100 100

Total

10.4

19.8

35.4

34.4

100

Percentage of assets that remain in the business models banks were in 1999–2002 (in the main diagonal) and the percentage that migrate to other models in 2003–2007.

to make decisions and actions “in the general interest of the caja”, not to the benefit of the group they represent. The literature review suggests different hypothesis on the risk taking behavior of cajas. On the one hand, not-for-profit stakeholders’ controlled cajas would have lower incentives to risk shifting and to assume risks in general, than shareholder banks. On the other hand, if cajas are state-owned banks as some authors suggest7 (Illueca et al., 2013), the public authorities could have influenced their lending decisions toward excessive risks relative to expected returns. These two competing hypothesis are not mutually exclusive. Some cajas might be more politically influenced in their lending decisions than others and concentrated higher risks. The methodology consists in examining the banks’ choice of their business model. A banking business model consists in a pattern of assets and liabilities adopted by one or several banks that differs from the pattern adopted by other banks, each with different combinations of expected return and risk (Ayadi et al., 2011, 2012; Llewellyn, 2013; Roengpitya et al., 2014). The use of business models is justified in that firms make strategic decisions that involve a set of variables at the same time, rather than deciding on one by one (Buch et al., 2013; Blundell-Wignall et al., 2014). The combination of ownership and governance with business model choices in explaining the performance of banks is new with respect to the literature reviewed above. Other papers examine the effects of business models on performance (risk and return) considering only variables of the composition of banks assets (Shleifer and Vishny, 2010; Diamond and Rajan, 2011) or of the composition of banks’ liabilities,8 but ignore the ownership and governance dimension.

We use a cluster analysis to identify the business models and the banks grouped in each (Everitt et al., 2011). The database is elaborated from non-consolidated bank accounts (i.e., income statements, balance sheets and other complementary information), except data on regulatory capital, which is at the consolidated level. Banks in the sample include cajas and stockholder banks, including subsidiaries of foreign banks, during the period from 1992 to 2007. We split the data into four shorter sub-periods and do a cluster analysis on each of them: 1992–1994 (recession), 1995–1998 (recovery), 1999–2002 (moderate growth), and 2003–2007 (exponential growth). This division of time allows for the comparison of the pre-Euro (1992–1998) and the post Euro periods (1999–2007) and to analyze the evolution over time of banks’ migration across business models. We consider that a business model is the result of banks’ decisions on the sources and the uses of the funds, which are reflected in the composition of the assets and liabilities of their respective balance sheets. Next we identify the variables that can capture the strategy of the banks and that will serve as instruments in the identification of the clusters. The choice of the variables and instruments and the resulting clusters takes place through an iterative process described in Appendix A. The process takes place within some imposed constraints, such as the maximization of the ratio of between-within dispersion across clusters (Calinski-Harabasz Index), a constant number of clusters over time, the economic sense of every cluster in the context of existing business models in banking, and the reduction of the number of instruments minimizing the information loss. The outcome of the iterative process is the following variables as instruments for the cluster analysis:

3. Business models in the Spanish banking industry

1. Equity as a percentage of total assets. Capital and reserves from retained earnings divided by total assets. It is the component of regulatory capital with a higher loss-absorbing capacity, and it is complementary to the accounting leverage ratio. Ceteris paribus, banks with higher risk exposure and banks with higher risk aversion should choose a higher equity ratio, while complying with the regulatory capital requirements. 2. Loans as a percentage of total assets. Banks grant loans and invest in securities. Banks in the retail banking business have a higher proportion of loans in their total assets, while investment banks have a higher proportion of their assets in trading securities. 3. Loan-to-deposit ratio. Collecting deposits to grant loans to business and families characterize retail banking. A value lower than one for this variable indicates that the amount of deposits exceeds the amount of granted loans. So, the excess of funds is used in trading activities or to lend to other banks. In contrast, a

3.1. Identification of business models through a cluster analysis A business model is a particular way of performing banking sufficiently differentiated from other alternatives, and also sufficiently generalized and extended in the industry to deserve attention. Banks that adopt the same business model share differentiated strategies, operations, and performance. The volume of bank assets in each business model can differ over time as banks migrate from one model to the other. Retail banking, investment banking, and wholesale banking are common business models among banks (Ayadi et al., 2012, 2011; Llewellyn, 2013; Ayadi and de Groen, 2014; Roengpitya et al., 2014; Köhler, 2015).9

7 Considering cajas as state-owned bank is not technically correct, given that their legal nature is of private foundations performing commercial activities. 8 Brunnermeier (2009), Diamond and Rajan (2009), Gorton (2009), Beltratti and Stulz (2012) analyze the effect of liability composition on performance before the crisis. Ivashina and Scharfstein (2010), Allen et al. (2014) analyze the effect after the crisis. 9 Retail banks are characterized by serving customers with traditional products such as deposits, saving and loans, and payment services, using a dense network of relatively small branches extended through local, regional, national and, in some

cases, international geographic markets. Investment-oriented banks focus on trading activities and rely on different sources of funding, specially issuing debt. Finally, wholesale banks concentrate their activities in market segments of institutional clients, such as governments, corporations or other financial institutions. They get funds from the debt and the wholesale markets. Some banks specialize in one business model and others do business with all of them, universal banks.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

6

Fig. 2. Average values of variables used to construct the clusters across business models (BM). All banks: Period 2003–2007.

higher value of this ratio than one means that the bank is granting more loans than collecting deposits and, consequently, relying on market debt to finance the gap. 4. Bank loans minus bank liabilities as a percentage of total assets (Net interbank). The numerator of this ratio accounts for the net position of the bank in the interbank market. A positive (negative) value indicates that the bank is a net lender (borrower) in the interbank market. Information for this variable is complemented with information on loans and liabilities separately. All together they indicate the activity of the bank in the wholesale market. Fig. 1 shows the asset-weighted mean values of these variables for every year for the three ownership forms of banks in the database. The figure illustrates a decreasing time trend in the equity-to-total asset ratio, an increasing time trend in the proportion of loans in the assets of banks as well as in the ratio of loans to deposits (especially after year 2000), and a decreasing trend in the net interbank position. The cajas present some unique features compared with banks. First, the equity ratio of cajas increases up to year 2000 and then it decreases up to 2007. Second, the cajas had a larger balance of deposits compared to loans up to the mid-1990s. Since then, loans have increased more than deposits, and by 2007, the ratio of loans to deposits was close to that of corporate banks. 3.2. Results of the cluster analysis Table 1 shows the average value for each cluster of instruments. It also includes two additional variables (i.e., proportion of loans to other banks, and the proportion of securitized assets) that were initially included in the list of potential instruments. The number of crosses (+) indicates the number of clusters whose mean value is statistically different (p < 10%) to the one where the “+” is placed.10 Fig. 2 offers a visual representation of the profiles of the four clusters

10 For instance, the average of loans/assets in Cluster 1 for 1992–1994, displays three (+) which means that the value of 0.299 is statistically different from the average value of the variable in the other three clusters. By the same token, the average value of the variable in Cluster 2 displays two (+), meaning that 0.493 is statistically

Fig. 3. Time evolution of the distribution of assets across business models: All banks.

and illustrates the mean values of the variables by clusters for the sample period from 2003 to 2007. Cluster 1 comprises banks with a relatively low volume of loans in total assets, which also present a ratio of loans to deposits clearly lower than one. Banks in this cluster lend the excess of liquidity to other banks, so their net position in the interbank market is positive. Further, banks in this cluster lend high amounts in the interbank market and do not securitize loans. The equity ratio is in line with the ratio for the rest of the groups, with some exceptions in the final period. We call this business model retail-deposits. Cluster 2 comprises banks with a relatively high volume of loans and also a relatively high volume of deposits (ratio of loans to deposits close to one). Banks have similar volumes of borrowing and lending in the interbank market, that is, the net interbank position is close to zero. Banks in the cluster follow the practice of originateto-hold, and their securitization activity is almost nonexistent. The equity ratio does not differ substantially from the ratio of the rest of groups. We call this business model r etail-balanced. Banks in Cluster 3 differ from banks in Cluster 2 in that the volume of deposits is lower than the volume of loans, and part of the deficit is covered with funds obtained in the interbank markets. These banks lend similar relative amounts in the interbank market as those in Cluster 2, but they borrow larger amounts. Since 1999, banks obtained part of their funds by issuing market debt and securitization. We call this business model retail-diversified, since banks finance their assets with funds from multiple sources. Banks in Cluster 4 have a similar lending activity relative to total assets compared to Cluster 3, but their volume of deposits relative to loans is lower. For this reason, banks have a more negative position in the interbank market and, on average, must issue relatively more market debt than those in Cluster 3. The amount of issued securities is relatively small before 2003. After this year, the banks

different from two other clusters (Cluster 1 and 4) and not different from the value of one cluster (Cluster 3).

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx Table 3 Average values of size and growth across business models. Shareholder banks and cajas, 2003–2007. BM1 Size Assets per bank (mD ) Annual growth rates of: Assets Loans Number of branches

BM2

BM3

Table 4 Average values of indicators on market and product specialization across business models. Shareholder banks and cajas, 2003–2007. BM1

BM2

BM3

BM4

Loan markets (%)1 Firms Construction and Real State Consumer Governments Mortgage loans

42.7++ 6.8+++ 26.3+++ 0.10+++ 18.2+++

48.7++ 24++ 12.2++ 2.50+++ 38.4+++

52.7++ 28.3++ 11.1++ 1.50+++ 36.1++

53.8++ 26.6++ 17.3+++ 0.7+++ 29.0+++

Customers base Size of branches (m.Euros) Braches in towns <1000 h Growth of branches out of origin Deposits per account Loans per account

376.1+++ 0.00+++ 0.02+++ 12.02+ 97.64++

26.2++ 26.0++ 0.14+ 9.30+++ 51.9+++

31.3++ 21.0++ 0.15++ 11.8++ 71.3+++

89.6+++ 8.83+++ 0.115++ 13.66++ 97.1++

Revenues and price (%) Fees/Ordinary margin Interest rate of loans Interest rate of deposits

59.4+++ 8.20+++ 1.90++

23.8++ 6.40++ 1.80++

26.9+++ 6.20++ 2.20+++

34.0+++ 5.80++ 2.90+++

BM4

12.1++

13.7++

27.9++

41.2++

11+++ 15.6++ 0.6+++

14.9+++ 18.8+ 3.4+

19.4++ 21.8+ 0.4+

20.9++ 22.1+ 3.1+

We test the differences in mean values of variables across business models. The number of cross super-indexes (+ ,++ ,+++ ) indicates the number of bilateral comparisons for which the average of a business model is statistically different at the 10% significance level.

in Cluster 4 have the lowest equity ratio and the highest leverage. We call this business model retail-market. Table 1 shows that in the years 1992–2002, at least 90% of the assets of the banks in the sample are concentrated in business models (from now on, BM) 1 and 2 (retail-deposits and retail-balanced). In period 2003 to 2007, the industry experienced important changes. The number and characteristics of business models did not change (averages values of variables in each cluster do not change too much), but many banks changed their business model migrating from BMs 1 and 2 to BMs 3 and 4. On average, for the five-year period 2003–2007, BM 1 concentrates only 5% of the total assets while the remaining 95% of the assets are distributed almost evenly among the other three business models, with 72% of the assets of Spanish banks assigned to BMs 3 and 4. Five years earlier these two business models had only 4% of the total industry assets. Table 2 and Fig. 3 show a transition matrix across clusters from period 1999–2002 to period 2003–2007. The table confirms that persistency is very low, except for banks in BM 4. The migration over time from BMs 1 and 2 to BMs 3 and 4 is evident from the data. In 2007 the assets of banks are mostly distributed between the business model of market debt (around 60% of assets) and that of diversified finance (around 30%). The remaining 10% split equally between the models of deposits and balanced. 3.3. Comparison of size, operating decisions, and performance across models This section presents the results of comparing the average values across business models for variables related to size and growth (Table 3), product market scope (Table 4), and performance (Table 5). The information is limited to the most recent period of 2003–2007. A more detailed definition of the variables together with statistics for the whole sample period are presented in the Appendices B1 and B2, respectively. 3.3.1. Size and growth The values of the size and growth variables in Table 3 reflect the high balance sheet expansion experienced by Spanish banks in the period 2003–2007. This period coincided with the credit boom. The average sizes increase from BM 1 to BM 4 and result in part from the higher growth rates of banks in BMs 3 and 4: An average cumulative annual growth rate of 20.9% (11%) for banks in BM 4 (BM 1) means that after five years, the initial size of the bank is multiplied by a factor of 2.5 (1.6). The average annual growth rate of loans is about 20%, which is similar across most business models. However, BM 1 is statistically lower (15.6%). The average annual growth of the number of branches is between 3% and 4% in all business models except for BM 1 where the growth rate is close zero. The high expansion of the banks’ assets in the period 2003–2007 was possible because banks borrowed funds from market sources, so

7

1 Percentage of total loans. We test the differences in mean values of variables across business models. The number of cross super-indexes (+ ,++ ,+++ ) indicates the number of bilateral comparisons for which the average of a business model is statistically different at the 10% significance level.

Table 5 Average values of performance variables across business models. Shareholder banks and cajas: 2003–2007. BM1

BM2

BM3

BM4

Solvency (regulatory) and liquidity Solvency ratio (%) Equity/Regulatory capital (%) Liquid assets/Assets (%)

14.2+ 67.4 7.60+

11.8++ 65.4++ 8.60++

11.0++ 55.3+ 5.20+

9.7+++ 57.5+ 4.60+

Operating efficiency Operating costs/Operating margin (%) Total Factor Productivity

62.7++ 5.88

59.4++ 0.29+

53.3++ 4.01

49.4++ 9.82+

Rates of return Accounting ROA (%) ROA (before) (%) Accounting ROE (%)

0.80 1.06++ 14.7

0.93 1.28+ 12.5++

0.98 1.37+ 15.2++

1.03 1.49+ 17.7++

Risk Z-score (accounting) Z-score (before) RWA/Assets (%) NP Loans/Loans (%)

38.3++ 30.0+ 23.7+++ 1.79

51.6+ 41.0+++ 76.3++ 0.77+

55.6++ 40.0++ 79.1++ 0.67

71.7++ 27.0++ 62.2+++ 0.58+

We test the differences in mean values of variables across business models. The number of cross super-indexes (+ ,++ ,+++ ) indicates the number of bilateral comparisons for which the average of a business model is statistically different at the 10% significance level.

the pattern of growth rates and sizes in Table 3 are the counterpart of the migrations from BMs 1 and 2 to BMs 3 and 4. 3.3.2. Markets and product specialization The first block of variables in Table 4 refers to the composition of the portfolio of loans; the second block refers to the customer base; and the third one to the specialization in terms of sources of revenues and in pricing behavior. We observe that banks in BMs 2 and 3 tend to follow a similar specialization strategy in markets and products, while banks in BMs 1 and 4 follow a more differentiated one. While banks in BMs 1 and 4 specialize more in consumer loans and less in government and mortgage loans, those in BMs 2 and 3 serve a large base of consumers in urban and rural areas and earn relatively high intermediation margins (difference in the interest rates of loans and deposits). They sell traditional bank products, including mortgages, and have a substantial presence in the construction and real estate markets and in lending to the government. Banks in BM 1

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17 8

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

Fig. 4. Average values of the variables used to construct the clusters across business models.

Fig. 5. Time evolution of the distribution of bank assets across business models.

specialize in consumer credit, operate mainly in urban areas with a small network of large branches, and their fees from services represent an important source of revenues.11 The banks in BM 4 earn a higher fraction of revenues in the form of service fees and follow an aggressive pricing policy, probably to sustain their high grow rates and because their customers are more sophisticated buyers (firms, high-volume accounts, and urban residents).

11 Banks in BM 1 in the sub-period are not representative of the banks in this model along the whole period of time. The cajas in BM 1 (that traditionally collected more deposits than the loans) migrate to other business models after 2003. Thus, the banks that continue in BM 1 after this year are a miscellaneous collection of banks, including e-banking banks.

3.3.3. Performance Table 5 displays the variables that capture the performance of banks: solvency, liquidity, operating efficiency, profitability, and risk. The level and quality of the regulatory capital, as well as liquidity ratios worsen as we move toward clusters more dependent on wholesale financing. Banks with high market debt financing show relatively high efficiency and profitability ratios and mixed results in terms of risk than banks in other clusters. The solvency ratio (ratio of regulatory capital over risk weighted assets) and the proportion of equity in the total regulatory capital present higher average values in BMs 1 and 2 and lower in BM 4 (consistent with lower equity ratio in Cluster 4, see Table 1). Overall, banks in BMs 3 and 4 show higher leverage and lower regulatory solvency ratios and hold lower liquid assets than banks in BMs 1

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx Table 6 Comparative performance among dominant business models: 1999–2002 versus 2003–2007.

Loans Const. & RE Firms1 (%) Assets rate of growth (%) Equity/Assets (%) Accounting ROA (%) Z-score (accounting) ROA (before) (%) Z-score (before)

Period 1999–2002

Period 2003–2007

BM1

BM2

BM2

BM3

BM4

26.8

33.3

49.2

53.9

49.4

11.5

11.4

15

19.4

20.1

6.21 0.71 36.0

6.12 0.93 36.0

5.97 0.62 39.3

5.08 0.6 35.2

4.51 0.62 35.3

0.92 33.0

1.48 53.0

1.28 58.0

1.37 40.0

1.49 27.0

1 Percentage of total loans. We test the differences in mean values of variables across business models. The number of cross super-indexes (+ ,++ ,+++ ) indicates the number of bilateral comparisons for which the average of a business model is statistically different at the 10% significance level.

Table 7 Severity of ex-post losses in the crisis by business models. Shareholder banks and cajas (number of banks and %). BM1

BM2

BM3

BM4

Total

Merged or acquired (A) High capital need (B) Minor capital need (C) No capital needs (D)

0 0 0 2

4 0 7 2

8 8 4 7

5 5 0 5

17 13 11 16

Total (A + B)/Total (%)

2 0%

13 30.8%

27 59.3%

15 66.7%

57 52.6%

The hypothesis of independency of the distribution of observations in rows and columns is rejected. The p-value of the 2 Pearson is 1%.

and 2. These results are in line with the findings in Table 1 where we observed a higher loans-to-deposits ratio and higher dependence on market debt for banks in BMs 3 and 4. The results also support Acharya and Richardson (2009) who find that banks use securitization to reduce the requirements of regulatory capital. There are practically no statistical differences in total factor productivity across clusters (only between BMs 2 and 4). But the ratio of operating costs to operating margins indicates that the operating efficiency increases from BM 1 to BM 4. For banks in BMs 1 and 2, the lower operating efficiency might reflect their smaller size (BM 1) and their specialization in low-volume customers and their larger network of smaller branches (BM 2). On the other hand, the higher operating efficiency of banks in BM 4 might reflect the issuance of more market debt to finance their asset investment. The reason is that getting the debt from the market requires much less resources than collecting deposits through branches. The averages of accounting ROA are not statistically different across business models. However, the ROE increases from BM 2 to BM 4 and there are statistical differences across clusters. The similar averages in ROA and higher averages in ROE are explained by the fact that the average leverage increases from BM 2 to BM 4. As well as the accounting ROA (net profits over total assets), Table 5 also includes the variable ROA before that is equal to operating cash flows over total assets. The presumption is that cash flows are more difficult to manipulate by banks than profits. The average ROA before increases from BM 2 to BM 4, and the differences are statistically significant. The comparison of this result with that of equal average accounting ROA indicates that loan loss provisions and depreciation over total assets also increase when moving from BM 2 to BM 4, which is consistent with the pattern of differences in growth rates. The Z-score is an inverse measure of the insolvency risk of banks. Banks in BMs 2 and 3 show similar average Z-scores for the two measures of ROA. We observe that banks in BM 1 present higher

9

insolvency risk than banks in BMs 2 and 3. Banks in BM 4 have a lower average Z-score when using the operating cash flow than banks in other clusters. Therefore, in terms of the relative volatility of operating cash flows, the banks in BM 4 are riskier than banks in BMs 2 and 3. As for the other two measures of risk, banks in BMs 2 and 3 show similar ratios of risk-weighted assets over book value of assets. This ratio is particularly low for banks in BM 1 (see note 12). The lowest ratio of nonperforming loans is for banks in BM 4, which is explained by their assets’ higher growth rates. Banks in BM 2 score relatively high in nonperforming loans, although the score is not too different from banks in Cluster 3. This is consistent with the previous evidence on the similar specialization in products and markets of BMs 2 and 3. 4. Migration to riskier business models: determinants and consequences The massive migration of banks toward business models more dependent on market financing occurs during the post-Euro period and particularly in 2003–2006. In this period, the ECB and other central banks had a lax monetary policy, inflation differential between Spain and the rest of EU countries was at maximum levels, and securitization (CDS, ABS, etc.) grew fast in financial markets around the world. In this scenario, the Spanish banks borrowed at relatively low rates once differences in inflation were taken into account. These rates were similar to the rest of the EU banks in nominal terms and even lower in real terms. As a result, banks’ balance sheets increased at higher rates than deposits. Spanish banks used the market-supplied debt to finance loans to real estate activities that generated and fueled the real estate bubble.12 The response of banks to the lax monetary conditions was probably rational in the sense that banks perceived the opportunity to gain market share and increase profits. Finance was abundant and cheap; and given that banks’ ratings were at maximum levels, the markets did not seem to correctly price risk. The regulation authority enforced the dynamic provisioning in order to limit credit growth and to generate a buffer of provisions to absorb the losses when risks materialized (Jiménez et al., 2017). However, the dynamic provisioning did not impede two-digit credit growth at the industry level for a prolonged period of time.13 4.1. Why banks migrate Table 6 compares the average values of selected variables in BMs 1 and 2 during 1999–2002 with those of BMs 2, 3, and 4 during 2003–2007. The questions posed are: Did banks that migrated to BMs 3 and 4 improve their performance compared to that of banks that continued in the balanced-business model of BM 2? Did banks improve their performance in the more lax monetary conditions of 2003–2007 compared with the performance in the years before? Banks in BM 2 grew at higher rates during the years 2003–2007 than in the previous period (15% and 11% of annual growth rate in total assets, respectively). Most of this growth was in loans to construction and real estate and, thus, banks in the cluster substantially increased the concentration of loans in these areas. The rates of return, both ROA accounting and ROA before decreased on

12 Existing real estate assets in Spain experienced a substantial price increase with the Euro since long living assets are those benefiting relatively more from the drop in the discount factor resulting from the lower risk premium brought by the transition from the Peseta to the Euro and by the negative real interest rate of the Spanish economy in the year of lax monetary policy. 13 Nonetheless, recent studies (Agénor and Da Silva, 2016) show that dynamic provisioning can be highly effective in terms of mitigating procyclicality and financial risks.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

10

Table 8 Distribution of bank assets across business models by ownership form (%). Period 1999–2002

Corporate Banks Cajas

BM1

BM2

BM3

BM4

25.9 31.0

70.2 67.6

3.3 1.4

0.5 0.0

BM1

BM2

BM3

BM4

9.6 0.0

13.1 34.6

38.8 42.9

38.4 22.5

Period 2003–2007

Corporate Banks Cajas

average from 0.93% and 1.48% in 1999–2002 to 0.62% and 1.28% in 2003–2007, respectively, whereas the ratio of equity over total assets slightly decreased from 6.12% to 5.97%. However, the Zscores slightly increased in the period 2003–2007, which means more stable rates of return in 2003–2007 than in 1999–2002. Overall, the evidence from Table 6 indicates that banks in BM 2 grew faster in the period of lax monetary conditions, but the trade-off between risk and return remained practically unchanged. Banks that migrate to BMs 3 and 4 grew at higher rates than banks that continued in BM 2 (20% compared with 15% in 2003–2007), mostly again with loans to construction and real estate that reached concentration levels of 50%. The averages of ROA accounting, ROA before, and of the Z-score accounting of banks in BMs 3 and 4 in 2003–2007 are practically the same as those of banks in BM 2 in the same period. However, the average value of the Z-score before decreased and the average leverage ratio (inverse of equity/assets) increased in 2003–2007 compared with the previous period, which means higher insolvency risk and higher financial risk, respectively. From the values shown in Table 6, the response to the questions raised at the beginning of this section is that the lax monetary conditions of 2003–2007 induced high growth behavior among Spanish banks that led to increased construction and real estate loans compared to the years before. However, this growth did not turn into higher profitability, neither among the banks that continued in BM 2 nor among banks that migrated to the faster growing BMs 3 and 4. Migrating banks lowered their equity capital ratios, increased their leverage, and increased their default risk. Overall, we conclude that even though banks probably behaved in an individually rational way, increasing the lending activity while migrating to market-debt business models did not reward them with a more favorable risk-return ratio. Rather the evidence indicates that for similar returns, banks increased their exposure to risk, although the price of the risk did not apparently change. 4.2. The materialization of risk during the crisis This subsection presents some evidence on whether the size of the losses experienced by Spanish banks in the crisis was proportional to the vulnerability of the business model chosen by the banks in the pre-crisis period.14 For the purpose of our analysis, we rank banks in terms of the size of the losses they experienced by taking into account the results from the stress test on capital requirements by Oliver Wyman in year 2012; and whether the banks survive after the industry restructuring or not. Banks (only parent banks, subsidiaries are

14 The severity of the external shock with the crisis was the result of the international financial crisis, of the tightening of the monetary policy by the ECB that continued during 2008, the sudden stop of the activity in the construction sector with 20% of the working population (unemployment rose above 20%), and the estimated fall of 60%, real, in real state prices from 2007 on.

Table 9 Business models and severity of ex-post losses in the crisis: cajas (number and %). BM1

BM2

BM3

BM4

Total

Merged or acquired (A) High capital need (B) Minor capital need (C) No capital needs (D)

0 0 0 0

3 0 7 1

8 7 4 6

3 5 0 0

14 12 11 7

Total (A + B)/Total (%)

0 0

11 27.3%

25 60.0%

8 100%

44 57.0%

The hypothesis of independency of the distribution of observations in rows and columns is rejected. The p-value of 2 Pearson is 0.5%.

excluded) in each business model are then classified by their level of losses: (A) banks acquired or absorbed; (B) banks with high capital requirements after the stress test, including nationalized ones; (C) banks with minor capital requirements, partial public support, and restructuring; and (D) banks with no additional capital needs. The results of the cross-tabulation appear in Table 7. Half of the banks are classified as banks with severe losses (A and B). More than 70% of the banks either disappeared because they were merged or acquired by others, or were diagnosed needing an additional amount of capital. The highest proportion of banks more severe losses within a particular business model occurs in BM 4 and is lower in BMs 1 and 2. For banks in BM 3, the likelihood of having severe losses in the crisis was 56%, which means the probability of suffering serious losses is almost equal to the probability of not needing extra capital. A Pearson correlation test rejects the null hypothesis of independence between belonging to a business model and the severity of losses in the crisis (p-value = 1.6%). The conclusion from Table 7 is that the-ex-ante high vulnerability of BM 4 is correlated with the ex-post degree of difficulties faced by the bank. In terms of systemic losses, 60% of the total assets of Spanish banks in 2007 were managed under the business model of market debt. Therefore, in the years previous to the crisis the expected proportion of total assets exposed to severe losses was 40% (60%·0.67). If we take together the exposure to losses of banks in BMs 3 and 4, then the expected losses increases up to 54% (probability of losses increases up to 60% and 90% of the total assets). Although belonging to BM 4 means a higher likelihood of severe losses in the crisis, there are five banks in this business model that passed the stress test without additional capital requirements. Similarly, the ex-post likelihood of severe losses is around 50% in BM 3, even though banks in this cluster are medium-high vulnerable to the crisis. These results show that, although differences in the business model of the bank mean different ex-ante vulnerability to severe external shocks, there might be additional factors affecting such vulnerability too. One of these factors could be the ownership form of the banks.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx Table 10 Comparison of the likelihood of experiencing higher losses during the crisis between cajas and shareholder banks, 2003–2007.

Caja Assets rate of growth Loans Const. & RE Firms NP Loans/Loans Solvency ratio Z-score (before) Pseudo R2 Observations

I

II

III

IV

0.703*** (0.233) 2.828*** (1.060) 2.918*** (0.925) 102.0*** (24.03) −17.91*** (4.740) −0.003* (0.002)

0.601 (0.449) 14.17** (5.740) 4.479* (2.361) 196.4*** (70.32) −11.02 (10.50) −0.005 (0.007)

1.572*** (0.307) 2.346* (1.380) 5.070*** (1.662) 200.8*** (53.78) −31.94*** (9.782) 0.003 (0.003)

1.681** (0.680) 18.77** (7.511) 4.812 (4.203) 317.5** (129.2) −59.52** (28.84) 0.007 (0.009)

22.5% 273

34.5% 54

42.7% 274

58.0% 54

Dependent variable Loss takes the value of one if the bank belongs to categories A or B of high losses and zero otherwise. Columns I is estimated with pooled data from 2003 to 2007, and Column II is estimated with the average values of the variables during 2003–2007. Columns III and IV replicates the estimations in Column I and II, respectively, weighting each observation by the size of the bank measured by the total assets. All of the variables are expressed in unit terms. Standard errors are in parentheses. *** Statistically significant at 1%. ** Statistically significant at 5%. * Statistically significant at 10%.

Table 11 Cross tabulation of number of cajas according to political influences and human capital of chairman and choice of the business model.

Public Origin Education Political Influence Banking Experience Management Exper

Value of Dummy

BM2

BM3

BM4

p-value

0 1 0 1 0 1 0 1 0 1

5 6 7 4 5 6 8 3 6 5

11 14 16 9 10 15 16 9 16 9

3 5 5 3 2 6 5 3 6 2

0.93 0.99 0.65 0.86 0.66

Political Influence identifies banks with a CEO with political experience, Education is a dummy that identifies banks with a CEO with a postgraduate degree, PublicOrigin identifies banks whose original funding was as a public entity; Banking Experience indicates the top manager had previous experience in the banking sector, before being appointed as CEO; Management Exper identifies those top managers that had previous experience in management in non-banking sectors. P-values correspond to the Pearson correlation test of independence among variables.

5. The relevance of the form of ownership 5.1. Shareholder banks versus cajas We now repeat the analysis presented in the previous section. Fig. 4 shows the average values across the clusters of the instruments that are separated into shareholder banks and cajas. Next, in Fig. 5 we show the time evolution from 1999 till 2007 of the assets of corporate banks and the cajas in each business model. And, Table 8 shows the distribution of bank assets across business models in 1999–2002 and 2003–2007. The observation of these figures and Table 8 makes clear that cajas imitated banks in their migration to more market debt after 2002, although with a time lag. In 2007, the distribution of the cajas’ assets across business models was: 50% in BM 4, 40% in BM 3, and 10% in BM 2, (there were no cajas remaining in BM 1). Among shareholder banks, these proportions were 75%, 15%, and 8%, respectively (2% of the assets of corporate banks were still in BM 1).

11

5.1.1. The losses during the crisis If an external shock affects the market equilibrium, all banks with similar business models are likely to experience similar effects from the shock. Table 9 presents the distribution of cajas across business models. In the group of cajas, the null hypothesis of independence between the choice of business model and the severity of the ex-post losses is rejected (Pearson correlation test with p-value of 0.5%). The likelihood of experiencing severe losses in the crisis is lower in cajas with a balanced-business model (27% in BM 2), and higher in cajas that adopt the market-debt business model (100% in BM 4). Eight out of eight cajas in BM 4 experience severe losses during the crisis contrasts with only two out of seven shareholder banks in the same situation. The five shareholder banks in BM 4 with minor losses during the crisis include the two largest Spanish banks. More than half of the cajas (57%) and a similar proportion of their total assets (55%), experience high losses in the crisis. The number of independent shareholder banks with high losses is 4 out of 11, about one third. In terms of assets, these four banks hold around 10% of the total assets of all these banks. Of the total assets with heavy losses, 85% belong to cajas and only 15% to corporate banks. Even though the exposure to ex-ante risk vulnerability was similar because of the business model choices in the period 2003–2007, the ex-post losses caused by the financial crisis were more severe for cajas than for shareholder banks. 5.1.2. Could the unique ownership and governance in cajas explain the differences in the ex-post losses? To answer the above question, we estimate a Probit model where the dependent variable Loss takes the value of one if the bank in the sample belongs to the losses categories of A or B in Table 7, and zero otherwise. The values of the explanatory variables refer to the pre-crisis period 2003–2007 when the risks are build-up. These variables include the dummy Caja that takes the value of one if the bank is a caja and zero otherwise, and a list of variables that are commonly accepted as measures of the exposure to risk: growth of total assets, proportion of loans to construction and real estate, ratio of nonperforming loans, solvency ratio, and Z-score. The analysis is based on the premise that the classification of banks in higher or lower levels of ex-post losses can be traced back to decisions made in the pre-crisis period. Because one of our sources of ex-post losses is the results from the stress tests conducted in 2012, an argument might be that too much time elapsed for this assumption to hold. However, we argue that the losses reported in the stress tests are a good approximation of the differences in loses experienced by all banks in the crisis, which is the relevant measure for our analysis. In any case, the levels of losses are discretized so the margin for errors in measurement should be relatively low. We expect that the higher rates of growth in total assets will indicate higher hidden credit risk in the portfolio of bank loans; higher concentration of loans in construction and real estate loans also add to the likelihood of serious losses. The ratio of nonperforming loans is a measure of the known credit risk of the portfolio of bank loans before the crisis. Higher values of this ratio anticipate higher likelihood that the losses from a negative external shock will also be higher. The solvency ratio is the ratio of regulatory capital of the bank over its risk weighted assets; higher solvency ratios mean that banks have more capital to absorb losses and, thus, are negatively associated with the likelihood of a severe losses. Finally, the Z-score expresses the ratio between the loss absorbing capacity of the bank (expected profits plus equity capital) and the variability (standard deviation) of profitability. The results of the Probit estimation appear in Table 10. Column I is estimated with all the information available between 2003 and 2007, whereas Column II is estimated with the average value of each variable during 2003–2007. In Column III and IV, we present

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model

ARTICLE IN PRESS

JFS-527; No. of Pages 17

A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

12

Table 12 Political and insiders’ influence variables as determinants of the severity of losses.

Compensation Public Origin Education Political Influence Banking Experience Management Exper Pseudo R2 Observations

I

II

III

IV

V

VI

0.280* (0.148) 0.308 (0.491) −0.218 (0.432) −0.003 (0.445) −0.383 (0.434) −0.513 (0.517)

0.243* (0.139) 0.264 (0.475) −0.171 (0.443) −0.011 (0.447) −0.285 (0.460) −0.406 (0.490)

0.280** (0.138) 0.540 (0.413) −0.172 (0.423)

0.252* (0.135) 0.458 (0.406) −0.131 (0.431)

0.283** (0.140) 0.601 (0.411)

0.258* (0.137) 0.505 (0.405)

−0.287 (0.435)

−0.214 (0.439)

15.1% 44

12.3% 164

13.2% 44

11.4% 164

12.1% 44

10.3% 164

Dependent variable Loss takes the value of one if the bank belongs to categories A or B of high losses and zero otherwise. Political Influence identifies banks with a CEO with political experience, Education is a dummy that identifies banks with a CEO with a postgraduate degree, PublicOrigin identifies banks whose original funding was as a public entity; Banking Experience indicates the top manager had previous experience in the banking sector, before being appointed as CEO; Management Exper identifies those top managers that had previous experience in management in non-banking sectors. Compensation is measured by the ratio of total compensation to the management team over the total personnel expenditures. Columns I, III, and V are estimated with the average values of the variables during 2003–2007, and Columns II, IV, VI are estimated with pooled data from 2003 to 2007. The *** statistically significant at 1%; ** statistically significant at 5%; * statistically significant at 10%. Standard errors are in parentheses.

the same estimations as in I and II but now each observation is weighted by the size of the bank as measured by total assets. The estimated coefficients of the risk variables are in general statistically significant and with the predicted sign, though the Zscore is only statistically significant in Column I. As expected, the likelihood that a bank experiences serious losses from the crisis is positively associated with high growth, higher concentration in construction and real estate loans, and higher nonperforming loan ratios in the pre-crisis period. Controlling for these variables of exante vulnerability, the likelihood that cajas experience high losses with the crisis is significantly higher than that of corporate banks.15 Our conclusion is that cajas experienced higher losses even when controlling for their risk exposure in 2007. This finding, together with the one above that cajas with the same business model experience higher losses than corporate banks, confirm that differences in the post-crisis losses of cajas cannot be attributed entirely to pre-crisis differences in the risks taken. 5.2. Differences in governance among cajas In this subsection, we examine the possible reasons behind the heterogeneous behavior and performance of cajas. In Section 2, we advanced the main distinct features of Spanish cajas within the large body of saving banks. The general presumption is that Spanish cajas operate under strong political influences. The heterogeneity observed among cajas in terms of the choice of the business model may then be related with variables indicative of the power of insiders and of political influence. The political influences in the management of cajas are captured by two dummy variables. One, Public Origin takes either a value of one if the caja was originally a public entity, and zero otherwise. The Political Experience takes a value equal to one if the chairman of the caja had direct political experience before being appointed to this position, and zero otherwise. The influence of insiders is

captured with variables on the human capital of the chairman of the board, and on the compensation of the management team, which are variables also considered in the research. We consider that lower human capital and higher compensation are indicative of higher insider control over the nomination and compensation decisions. The human capital of the chairman accounts for formal education and work experience. The Education variable takes the value of one if the chairman has a postgraduate degree and zero otherwise; the variables Banking (General) experience take the value of one if the chairman has managerial experience in banking (non-banking sectors) before occupying the current position, and zero otherwise. We collect the information needed for the measurement of these variables from web files and press reports from the period 2003–2007. The variable Compensation is measured by the ratio of total compensation to the management team over the total personnel expenditures for the period 2004–2007 (data on compensation and personnel expenditures are from company files). The null hypothesis is that political and insider influences have no effect on the choice of the business model and on the consequences of the choice in terms of the severity of the losses caused by the crisis. If the hypothesis is rejected then it should be in the direction of higher influence that indicates a higher likelihood of choosing ex-ante more vulnerable business models (belonging to BM 3 or 4), and of belonging to categories A or B. Table 11 shows the number of observations from the crosstabulation of the types of business models, the variables of political influence, and the characteristics of the chairman, along with the p-value of the test of independence. In all cases the null hypothesis of independence is not rejected at p < 10% significance level, so we do not find a statistically significant association between political influence, human capital, and the choice of a business model among cajas. We examine the differences in compensation across business models by using a regression where the dependent variable is Compensation and the explanatory variables comprise dummy variables to identify BM2, BM3, and BM4, the log of assets to control for size,

15 Sagarra et al. (2015) provide a comprehensive analysis on the prediction of failure of cajas with the crisis using financial ratios. The classification of cajas in fail or succeed is similar to that used here; they find that ex-ante profitability, solvency and quality of loans are good predictors of failure. Akin et al. (2014) provide evidence that more damaged cajas ex-post were more careless in ex-ante loan granting decisions. Neither of them compares cajas with corporate banks as we do here.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

and time dummy variables for each year in the period 2004–2007.16 The model is estimated with 164 caja-year observations. We find that the coefficients for the dummies that identify BM 3 and BM 4 are 0.378 (p < 0.4) and 0.721 (p < 0.10), respectively. Therefore, the compensation for the directors of the cajas in BM 4 is statistically and significantly higher than in the rest of the business models. This provides evidence of insiders’ influence over cajas in BM 4. Next, we examine the political and insider influences in the likelihood of the cajas belonging to the group of high or low losses from the crisis. For this purpose, we estimate a Probit model with the dependent variable Loss and the Compensation and the proxy variables for political and insider influences as explanatory variables. The results are in Table 12. The variables of political influence and human capital of the chairman are neither individually nor jointly significant statistically. A different result is obtained for the variable Compensation, whose coefficient is positive and statistically significant. In this case, the null hypothesis of no influence of the compensation variable on the ex-post losses is rejected: those cajas that paid higher salaries to their management teams are more likely to fall ex-post in the category of high losses from the crisis.17 The above evidence indicates that cajas with high insider influence migrated to high market-debt business models with the expectation of increasing the compensation of the management team. The evidence corroborates Serra-Ramoneda (2011), a former president of an important caja, who attributes a large responsibility of the fall of Spanish cajas to compensation practices adopted by cajas with the purpose of emulating shareholders banks in the pre-crisis period. The evidence is similar to that in Cheng et al. (2010) where institutional investors pushed banks managers toward riskier decisions by rewarding such behavior; Acharya and Naqvi (2012) also predict an excessive credit growth in banks where managerial compensation is correlated with size. 6. Conclusion This paper shows that shortly after the creation of the Euro, a period of low official interest rates and abundant liquidity worldwide, Spanish banks expanded their balance sheets. In this period, they lent massively to construction and real estate and financed these loans with funds from debt markets. The result was a migration of all banks to business models that had been residual in the industry until then. Both, shareholders banks and cajas migrated to business models that were more dependent on market debt, in spite of having very different ownership and governance. Therefore, the response to the softening of the constraints within the Euro was similar for the two ownership forms. In this respect, cajas repeated the pattern of following and imitating shareholder banks in their commercial practices, a pattern adopted since the liberalization of Spanish banks in the 1990s. This time, the cajas followed shareholder banks in the migration to market-dependent business models. The Spanish shareholder banks and cajas had experienced several crises in the past, and the cajas had survived them, in general with lower revealed losses and government help than banks. The difference in the current crisis is that even though shareholder banks and cajas made similar choices in business models in the pre-crisis period, cajas had more severe losses from the crisis. So

16

as

13

why did a lack of resilience determine the end of the cajas this time. This paper claims that the particular ownership and governance of cajas lowers their resilience in front of external negative shocks. The lack of resilience was particularly relevant this time for several reasons. One is the higher severity of the recent crisis that combined financial and real sector turbulences. Closely related, the high dependence of Spanish banks to debt markets and the transference of the mechanism of lending of last resort to the ECB. Finally, the restructuring of insolvent cajas through mergers with other cajas located in the same region that worked in the past did not work this time. The restructuring required the involvement of cajas from different regions, but political interests at regional levels reduced even more the resilience in responding to the negative shock. Although the research mostly considers the crisis of the Spanish banking industry as a crisis of the cajas, the truth is that not all cajas migrated to the most vulnerable business models. The paper has examined the heterogeneity observed among cajas in the precrisis period under the lens of two variables, political and insider influences in decision-making. We find no supportive evidence that potentially higher political influence and less human capital in the chairman of the board led to choices of more vulnerable ex-ante business models, and to higher ex-post losses. What we find is that the probability of a caja belonging to the category of high losses increases with the relative compensation of the management team. Therefore, we find supportive evidence that the compensating system induced growth and risk-taking behavior in the management team of those cajas that ended up with more losses. The experience of the recent crisis has made regulatory authorities at the supranational and national levels fully aware of the relevance of ownership and governance of banks for financial stability. Our results show that ownership and governance seem to matter more in bad times than in good times. The lesson we learn is that, if a particular ownership form of banks is less resilient to external shocks, then it is recommended to be more conservative in risk-taking decisions in good times. In this respect, the trade-off might be between stakeholder banks, like cajas, with advantages in good times (for example contributing to financial inclusion), and shareholder banks more resilient in bad times. Over all, the intrusion of regulation in setting rules and norms on governance of banks, including compensation policies, appears well justified from our results. The paper also justifies the serious consideration of discriminatory regulation requirements on capital and liquidity across business models and ownership forms in the context of the trade-offs just indicated. The case of Spain is a good example of the weakness of the European integration without a real European banking union. Until very recently, the national economies were responsible for the insurance of the deposits and other liabilities, independently of whether they came from residents or from non-residents. During the pre-crisis period of high credit growth, national and supranational regulatory and supervisory authorities probably did not realize that private debt in the banks’ balance sheets would turn into public debt in case of a crisis. Indeed, after the outburst of the crisis, the strength of the Spanish economy proved to be insufficient to assure the safe return of the debt from the banking system, and the impact of the crisis in the economy was magnified. The Single Supervisory Board is a step in the direction of a European Banking Union. However, the lesson learnt from the recent crisis is that European interbank markets will not operate again properly until the single Euro area safety net insures the deposits of all the banks within the Union.

The complete equation of the model that we estimate can be expressed follows: Compensationit = ˛0 + ˛1 Cluster3it + ˛0 Cluster4it + ˇ ln Assetsit +

2007 

ıj Timej + εit .

j=2004 17

We have estimated the Probit with data pooled for the four years in which we have information on the compensation variable with similar results.

Appendix A. Cluster Method This appendix details the steps to identify the business models in the population of Spanish banks. We use cluster analysis. The

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17 14

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

process can be summarized as “guided statistical search”, which means that we set some constraints on the dimensions along which business models differ, on the instruments used to capture those dimension, and on the stability in the number of business models over time. We then proceed with the cluster analysis in an iterative way to verify the statistical soundness of the results. The steps are not too different from those proposed by reference books on cluster analysis such as Everitt et al. (2011), pages 261–262. A.1 A priori scope of business models The research confirms that the Spanish banking industry dominate the traditional business model of collecting deposits and granting loans. There are some differences among banks in the dependence on financial markets to complement deposits and in the portfolios of investment that emphasize or deemphasize securities. The first constraint is then to identify business models that group banks that depart from the traditional core model of collecting deposits and granting loans. Other studies on business models start with a broader scope of possible business models in the data set than we do in this paper, and therefore with different list of instruments. A.2 Choice and number of instruments We consider a business model the result of a subset of banks that adopt a similar strategy in their funding and lending activities and that use a strategy sufficiently different from that adopted by other banks. We separate the decision on the strategy from the decisions on the implementation of the strategy and the final performance after the strategic choice and implementation. Consistent with this view, the application of the cluster analysis requires selecting the instruments that capture only the dimensions of the strategic choice, identify the clusters with these instruments, and once identified compare the resulting clusters in operational decisions and performance (size, growth, liquidity, profitability, and risk). This distinction between the choice of the strategy, its implementation, and the resulting performance is important in understanding the difference between the list of instruments used to identify the clusters, and the list of variables that are used to compare the different businesses once they have been identified. From the related literature cited in the main text and from the observation of banks’ balance sheets, the initial list of potential instruments to elaborate the clusters is the following: equity/assets, loans/deposits, net interbank position/assets, loans assets, interbank loans/assets, interbank deposits/assets, securitized loans/assets, solvency ratio, equity/regulatory capital, and wholesale debt/liabilities. In the evaluation of the results we follow the recommendation of Everitt et al. (2011) who show that irrelevant or masking variables should be excluded as instruments in the identification of the clusters. The process of elimination and selection of instruments from the long list is as follows: First, we use a principal components analysis to see the redundancy in information among all instruments in the list. The results of the analysis show that four components cover more than 95% of the information contained in the list, and the results are similar in all subperiods. At this point we could have used

the component scores to identify the clusters but instead we use variables representative of each component because of the ease in interpretation of the business models associated with the clusters. In order to identify the instruments that are representative of the whole list we perform an iterative process of combining different instruments from each component, applying the cluster algorithm and comparing the results (similarity in the banks in each cluster, and differences in the means of the instruments across clusters, ranking of classifications using the Calinski-Harabasz Pseudo-F statistic). The final instruments make economic sense, assured the stability of clusters over time, and the Pseudo F statistic is the highest or among the highest in all cases. They are equity/assets, loans/deposits, net interbank position/assets, and loans assets. Although some instruments did not make the final list their information about the strategy of banks is not lost. To the contrary, the correlation between these instruments with one or more of the included in the list prevents this loss. For example the high loan-to-deposit ratio is correlated with the existence of alternative funding sources, like wholesale debt or securitization. Finally, considering that even though some of the instruments were not informative in the identification of the clusters, they could be so as descriptors of the respective business models associated with the clusters. Tables 1 and 2 present the mean values in each cluster of all instruments, the four used to identify the cluster and the rest. A.3 Choice of the number of clusters The main criterion followed to determine the number of clusters is the overall maximization of the Calinski-Harabasz index over time. This index means that we have looked for a number of stable clusters over time, such that the Calisnki-Harabasz index is close to its maximum in each of the four subperiods. The stability in the number of clusters over time is one of the constraints imposed in the analysis in order to facilitate the identification of the transition from one business model to another. A.4 Choice of the cluster agglomerative method The Ward’s linkage method is used to assign banks to the clusters because it is the method that gives the highest value of the Calasnki-Harabasz index. It also assures a good representation of banks in each cluster. Everitt et al. (2011) concludes that Ward’s method works well most of the time. Single, average, and complete linkage are other alternatives that we consider. We discard the single linkage because of its questionable mathematical properties and because the main bulk of banks concentrate in a single cluster. The complete and average linkages provide similar results to Ward’s linkage in terms of the distribution of observations across clusters, with slightly more banks concentrated in Clusters 2 and 3. Thus, the results are robust to the linkage method used to assign banks to the different clusters. Appendix B.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

15

Table B1 Definitions of the main variables. Size and growth Assets Growth rate loans (%) Growth rate branches (%) Loan markets Construc & RE Firms (%) Governments (%) Mortgage loans (%) Customer base Average size of branches Braches in towns <1000inh

Growth out of region branches (%) Deposits per account (%) Loans per account (%) Product and price Fees/Inter. Margin (%) Loans interest rate (%) Deposits interest rate (%) Solvency and liquidity Solvency ratio (%) Equity/Regulatory capital Liquid assets/Assets (%) Rates of return ROA (accounting) (%) ROA (before) (%) ROE (accounting) (%) Operating efficiency Operating costs/Oper margin (%) Total Factor Productivity

Risk Z-score Z-score (before) RWA/Assets (%) NPL ratio (%) Financial Structure Securitiz/Assets Equity/Asset (%) Loans/Assets (%) Loans/Deposits (%) Net Interbank/Assets (%) Bank Loans/Assets (%)

Volume of total assets at book value Year-by-year rate of growth of loans granted by the bank, where the stock of loans at book value Year-by -year growth rate of the total number of branches of the bank. Proportion of real-estate loans in the loan protfolio of the bank Proportion of loans to government institutions in the loan portfolio of the bank Proportion of mortgages in the loan portfolio of the bank Ratio of bank’s assets with respect to the number of braches For every bank and year, we aggregate the number of branches in towns smaller than 1000 inhabitants, using ˜ Information on the size (inhabitants) is taken municipality-level information from the Branches Registry of Banco de Espana. from the Spanish Statistical Institute (INE). ˜ we aggregate the number of branches Using municipality-level information from the Branches Registry of Banco de Espana, out of the origin province and compute the year-by-year rate of growth. Amount of deposits divided by the number of sight accounts Volume of loans divided by the number of loan accounts Net commissions with respect to intermediation margin, both taken from the bank’s profit and losses account. Yearly average of marginal interest rates set by banks to new loan operations granted every month. Yearly average of marginal interest rates of new deposits collected every month. Regulatory capital ratio of the bank at consolidated level Ratio of equity capital to regulatory capital. Sum of cash, reserves in central bank and bonds issued by the Government with respect to total assets Profitability of the bank measured by the return on assets. It is calculated as net profits divided by total assets of the bank. Return on assets calculated as net profits before interests, taxes and depreciation divided by total assets of the bank. It is the profitability of the bank measured by the return on equity. It is calculated as net profits divided by the equity of the bank. Ratio of operating costs divided by the operating margin Measure of Total Factor productivity of the bank in logs. It is obtained as a residual of a measure of raw productivity after controlling for differences and changes in the business models and observed characteristics of banks (Martín-Oliver et al., 2013). Z-score of the bank, in the numerator the sum of the capital ratio and the accounting ROA and in the denominator the standard deviation of ROA computed with the last 5 years of the bank information Z-score, computed with the ROA (before) Measure of the risk profile of the bank, defined as the ratio of the risk-weighted assets (i.e. denominator of the regulatory capital ratio) and the book value of the assets Ratio of non-performing loans with respect to total loans, from information of the Credit Register Proportion of ABS and MBS to total assets Ratio of equity capital to total assets of the bank Weight of the loans in the assets of the bank Volume of loans with respect to the volume of deposits Net borrowing position of the bank in interbank markets as a percentage of book value total assets. Gross borrowing of the bank to other banks as a percentage of total assets.

Measure of damages during the crisis Damage

Dummy variable that identifies banks either acquired or absorbed (A) or banks with high capital requirements after the stress test, including nationalized ones (B) Characteristics of Executives and Compensation Compensation Measured by the ratio total compensation to the management team (from reports of stock regulator, CNMV) over the total personnel expenditures for the period 2004 to 2007 Personnel Expenses, in millions of euros, as a size variable to compare with the compensation of the management team Salaries (mD ) This variable identifies banks with a CEO with political experience. Political Influence Dummy variable that identifies banks whose original funding is public, as in García-Cestona and Surroca (2008) Public Origin Education Education is defined as a dummy variable that identifies banks with a CEO with postgraduate formation. Identifies banks whose top manager had previous experience in banking before becoming president of the company Banking Experience Identifies banks whose top manager had occupied charges of responsibility in non-banking organizations before becoming Management Exper the president of the caja

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17 16

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

Table B2 Descriptive Statistics of the main variables, 1992–2007.

Size and growth Assets (mD ) Growth rate loans (%) Growth rate branches (%) Loan markets Construc & RE Firms (%) Governments (%) Mortgage loans (%) Customer base Average size of branches (mD ) Braches in towns <1000inh Growth out of region branches (%) Deposits per account (%) Loans per account (%) Product and price Fees/Inter. Margin (%) Loans interest rate (%) Deposits interest rate (%) Solvency and liquidity Solvency ratio (%) Equity/Regulatory capital Liquid assets/Assets (%) Rates of return ROA (accounting) (%) ROA (before) (%) ROE (accounting) (%) Operating efficiency Operating costs/Opermargin (%) Total Factor Productivity Risk Z-score Z-score (before) RWA/Assets z(%) NPL ratio (%) Financial Structure Securitiz/Assets Equity/Asset (%) Loans/Assets (%) Loans/Deposits (%) Net Interbank/Assets (%) Bank Loans/Assets (%) Characteristics of Executives and Compensation Compensation management (mD ) Salaries (mD ) Political Influence Public Origin Education Banking Experience Management Exper

Mean

Std.Dev.

P10th

P25th

P50th

P75th

P90th

24,400 0.176 0.035

51,300 0.080 0.030

1380 0.075 0.000

3573 0.121 0.010

8207 0.167 0.028

18,800 0.227 0.057

57,600 0.288 0.081

0.263 0.018 0.365

0.111 0.017 0.127

0.142 0.000 0.190

0.195 0.003 0.283

0.256 0.014 0.367

0.331 0.027 0.441

0.388 0.042 0.498

42.93 20.71 0.136 36.26 68.14

104.81 35.742 0.130 17.58 36.92

12.03 0 0.000 14.47 30.28

15.58 0 0.024 22.34 40.79

20.78 7 0.100 33.70 58.47

29.95 28 0.223 50.218 83.515

43.36 56 0.366 65.55 116.2

0.269 0.063 0.021

0.165 0.017 0.007

0.133 0.044 0.015

0.169 0.052 0.017

0.236 0.060 0.019

0.321 0.072 0.024

0.396 0.083 0.032

0.111 0.600 0.065

0.024 0.219 0.049

0.090 0.352 0.013

0.098 0.455 0.028

0.110 0.588 0.057

0.120 0.728 0.087

0.132 0.941 0.132

0.010 0.014 0.146

0.005 0.008 0.075

0.005 0.009 0.083

0.007 0.010 0.106

0.009 0.012 0.129

0.011 0.015 0.174

0.017 0.020 0.242

0.548 0.030

0.167 0.217

0.378 −0.240

0.486 −0.094

0.557 0.034

0.624 0.147

0.669 0.303

56.80 37.84 0.743 0.007

132.8 25.89 0.205 0.005

14.42 11.15 0.485 0.003

21.02 19.56 0.674 0.004

33.40 30.65 0.778 0.006

57.98 50.33 0.839 0.008

88.24 72.59 0.912 0.012

0.054 0.747 1.462 −0.022 0.079

0.022 0.122 0.482 0.104 0.077

0.029 0.623 0.929 −0.206 0.018

0.037 0.695 1.103 −0.054 0.033

0.049 0.766 1.338 −0.003 0.055

0.068 0.817 1.686 0.034 0.098

0.086 0.899 2.166 0.096 0.157

3296 174.5 0.576 0.397 0.375 0.335 0.357

4247 328.0 0.495 0.510 0.485 0.473 0.480

830 10.53 0 0 0 0 0

1274 31.80 0 0 0 0 0

2096 70.17 1 0 0 0 0

3553 152.1 1 1 1 1 1

5639 349.1 1 1 1 1 1

References Acharya, V., Naqvi, H., 2012. The seeds of a crisis: a theory of bank liquidity and risk taking over the business cycle. J. Financ. Econ. 106 (2), 349–366. Acharya, V., Richardson, M., 2009. Causes of the financial crisis. Crit. Rev. 21, 195–210. Adams, R., Mehran, H., 2003. Is corporate governance different for bank holding companies? Econ. Policy Rev. 9, 123–142. Agénor, P.R., Da Silva, L.A.P., 2016. Cyclically adjusted provisions and financial stability. J. Financ. Stabil. 28, 143–162. Akin, O., García-Montalvo, J., García Villar, J., Peydró, J.L., 2014. The real estate and credit bubble: evidence from Spain. SERIEs 5, 223–243. Allen, F., Hryckiewicz, A., Kowalewski, O., Tümer-Alkan, G., 2014. Transmission of financial shocks in loan and deposit markets: role of interbank borrowing and market monitoring. J. Financ. Stabil. 15, 112–126. Almazán, A., Martín-Oliver, A., Saurina, J., 2015. Securitization and Banks’ Capital Structure. Rev. Corp. Financ. Stud. 4 (2), 206–238. Anderson, R.C., Fraser, D.R., 2000. Corporate control, bank risk taking, and the health of the banking industry. J. Bank Financ. 24 (8), 1383–1398. yadi, R., Arbak, E., de Groen, W.P., 2012. Regulation of European banks and business models: Towards a new paradigm? Centre for European Policy Studies (CEPS), Brussels. Ayadi, R., Arbak, E., de Groen, W.P., 2011. Business models in European banking: A pre- and post-crisis screening. Centre for European Policy Studies (CEPS), Brussels.

Basel Committee on Banking Supervision, 2010 October. Principles for Enhancing Corporate Governance., pp. 1–34. Bhattacharya, S., Thakor, A., 1993. Contemporary banking theory. J. Financ. Intermed. 3, 2–50. Beck, T., Hesse, H., Kick, T, Von Westernhagen, N., 2009. Bank Ownership and Stability: Evidence from Germany, Working Paper. Beltratti, A., Stulz, R., 2012. The credit crisis around the globe: why did some countries perform better? J. Financ. Econ. 105 (1), 1–17. Berger, A.N., Clarke, G., Cull, R., Klapper, L., Udell, G., 2005. Corporate governance and bank performance: a joint analysis of the static, selection, and dynamic effects of domestic, foreign, and state ownership. J. Bank. Financ. 29 (8–9), 2179–2221. Berger, A., Imbierowicz, B., Rauch, Ch., 2012. The Roles of Corporate Governance in Bank Failures during the Recent Financial Crisis. Working Paper. University of South Carolina. Blundell-Wignall, A., Atkinson, P., Roulet, C., 2014. Bank business models and the Basel system: complexity and interconnectedness. OECD J. Financ. Market Trends, 1–26. Brunnermeier, M., 2009. Deciphering the liquidity and credit crunch 2007–2008. J. Econ. Perspect. 23, 77–100. Buch, C., Koch, C., Koetter, M., 2013. Do banks benefit from internationalization? Revisiting the market power-risk nexus. Rev. Financ. 17, 1–33. Butzbach, O., Mettenheim, K., 2014. Explaining the competitive advantage of alternative banks: from agency theory to institutionalist theories? In: Butzbach, O., Mettenheim, K. (Eds.), Alternative Banks in the Crisis. Pickering and Chatto, London.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004

G Model JFS-527; No. of Pages 17

ARTICLE IN PRESS A. Martín-Oliver et al. / Journal of Financial Stability xxx (2017) xxx–xxx

Cheng, I.-H., Harrison, H., Scheinkman, J., 2010. Yesterday’s heroes: Compensation and creative risk-taking. NBER Working Paper 16176. ˜ Cunat, V., Garicano, L., 2010. ¿Concedieron las cajas buenas créditos malos? Gobierno corporativo, capital humano y carteras de créditos. In: Boldrin, M., ˜ FEDEA, Díaz-Jiménez, J., Dolado, J.J. (Eds.), La Crisis de la Economía Espanola. pp. 351–398. Diamond, D., Rajan, R., 2011. Fear of fire sales, liquidity seeking, and credit freezes. Q.J. Econ. 126, 557–591. Diamond, D., Rajan, R., 2009. The credit crisis: conjectures about causes and remedies. Am. Econ. Rev. 99, 606–610. European Union, 2010. Corporate Governance in Financial Institutions: Lessons to be drawn from the Current Financial Crisis, Commission Staff Working Document 1-45. Ellul, A., Yerramilli, V., 2013. Stronger risk controls, lower risk: evidence from US bank holding companies. J. Financ. 68 (5), 1757–1803. Everitt, B.S., Landau, S., Leese, M., 2011. Cluster Analysis, 5th ed. Wiley, John and Sons Ltd, West Sussex. Fahlenbrach, R., Prilmeier, R., Stulz, R., 2012. This time is the same: using bank performance in 1998 to explain bank performance during the recent financial crisis. J. Financ. 67, 2139–2185. García-Cestona, M., Sagarra, M., 2014. Good and Bad Banks? Governance, Human Capital of Top Managers and Performance. SSRN Working Paper Series. García-Cestona, M., Surroca, J., 2008. Multiple goals and ownership structure: effects on the performance of Spanish savings banks. Eur. J. Operat. Res. 187 (2), 582–599. García-Marco, T., Robles-Fernández, M.D., 2008. Risk-taking behaviour and ownership in the banking industry: the Spanish evidence. J. Econ. Bus. 60, 332–354. García-Meca, E., Sánchez-Ballesta, J.P., 2014. Politicization, banking experience and risk in savings banks. Eur. J. Law Econ. 38, 535–553. Gorton, G., 2009. Information, liquidity, and the (ongoing) panic of 2007. Am. Econ. Rev. 99, 567–572. Hansmann, H., 2009. The Ownership of Enterprise. Harvard University Press. Hau, H., Thum, M., 2009. Subprime crisis and board (in-)competence: private vs. public banks in Germany. Econ. Policy 24 (60), 701–751. Hesse, H., Cihák, M., 2007. Cooperative banks and financial stability. International Monetary Fund. Hopt, K.J., 2013. Corporate governance of banks and other financial institutions after the financial crisis-regulation in the light of empirics & theory. J. Corp. Law Stud. 13 (2), 219–253. Iannotta, G., Nocera, G., Sironi, A., 2007. Ownership structure, risk and performance in the European banking industry. J. Bank Financ. 31 (7), 2127–2149. Illueca, M.L., Norden, L., Udell, G.F., 2013. Liberalization and risk taking: evidence from government-controlled banks. Rev. Financ., 1217–1257.

17

Ivashina, V., Scharfstein, D., 2010. Bank lending during the financial crisis of 2008. J. Financ. Econ. 97, 319–338. Jensen, M.C., Meckling, W.H., 1976. Theory of the firm: managerial behavior, agency costs and ownership structure. J. Financ. Econ. 3 (4), 305–360. Jiménez, G., Ongena, S., Peydró, J.L., Saurina, J., 2017. Macroprudential policy, counter cyclical bank capital buffers and credit supply: evidence from the Spanish dynamic provisioning experiments. J. Political Econ. (forthcoming). Kirkpatrick, G., 2009. Corporate governance lessons from the financial crisis. OECD J. Financ. Markets 1, 61–87. Köhler, M., 2015. Which banks are more risky? The impact of business models on bank stability. J. Financ. Stabil. 16, 195–212. La Porta, R., Lopez-De-Silanes, F., Shleifer, A., 2002. Government ownership of banks. J. Financ. 57, 265–301. Laeven, L., Levine, R., 2009. Bank governance, regulation, and risk-taking. J. Financ. Econ. 93, 259–275. Laeven, L., Valencia, F., 2013. Systemic banking crises database. IMF Econ. Rev. 61, 225–270. Liikanen, E., 2012. Reforming the structure of the EU banking sector. Final Report of High-level Expert Group. European Commission. Llewellyn, D., 2013. Fifty years in the evolution of banks business models. 50 Years of Money and Finance: Lessons and Challenges, SUERF European Money and Finance Forum. Martín-Oliver, A., Ruano, S., Salas-Fumás, V., 2013. Why high productivity growth of banks preceded the financial crisis. J. Finan. Interm. 22, 688–712. Merton, R.C., 1977. An analytic derivation of the cost of deposit insurance and loan guarantees: an application of modern option pricing theory. J. Bank. Financ. 1, 3–11. Minton, B., Taillard, J., Williamson, R., 2012. Board Composition, Risk Taking and Value: Evidence from Financial Firms. SSRN Working Paper Series. Mulbert, P., 2010. Corporate governance of banks after the financial crisis: theory, evidence, reforms. ECGI Working Paper Series in Law, 151. Myers, S.C., Rajan, R.G., 1998. The paradox of liquidity. Q. J. Econ 113 (3), 733–771. Roengpitya, R., Tarashev, N., Tsatsaronis, K., 2014. Bank business models. BIS Q. Rev., 55–65. Sagarra, M., Mar-Molinero, C., García-Cestona, M.A., 2015. Spanish saving banks in the credit crunch: could distress have been predicted before the crisis? A multivariate statistical analysis. Eur. J. Financ. 21 (3), 195–214. Saunders, A., Strock, E., Travlos, N.G., 1990. Ownership structure, deregulation, and bank risk taking. J. Financ. 45, 643–654. Serra-Ramoneda, A., 2011. Los Errores De Las Cajas. Ediciones Invisibles, Barcelona. Shleifer, A., Vishny, R., 2010. Unstable banking. J. Financ. Econ. 97, 306–318. Walker, D., 2009. A review of corporate governance in UK banks and other financial industry entities. Final Recommendations, London.

Please cite this article in press as: Martín-Oliver, A., et al., The fall of Spanish cajas: Lessons of ownership and governance for banks. J. Financial Stability (2017), http://dx.doi.org/10.1016/j.jfs.2017.02.004