Journal of Banking & Finance 43 (2014) 124–136

Contents lists available at ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

The effect on competition of banking sector consolidation following the financial crisis of 2008 Carlos Pérez Montes ⇑,1 Financial Stability Department, Banco de España, Calle Alcalá 48, 28014 Madrid, Spain

a r t i c l e

i n f o

Article history: Received 4 November 2013 Accepted 9 March 2014 Available online 19 March 2014 JEL classification: G21 G01 L13

a b s t r a c t Consolidation of the Spanish banking sector after the financial crisis of 2008 raises concerns about potential negative effects on competition. I use structural econometric methods to examine these anti-competitive concerns in the Spanish mortgage market. I estimate a mixed-logit model of mortgage demand and recover bank-level cost information with a strategic model of price competition. Counterfactual experiments reveal that the observed increase in concentration is associated only with small variations in mortgage rates and market shares, staying far from collusive levels. This moderate change in industry conduct implies a small direct effect of consolidation on bank exposures to mortgage risk. ! 2014 Elsevier B.V. All rights reserved.

Keywords: Mortgage Banking Crisis Sector consolidation GMM Mixed-logit Counterfactual analysis

1. Introduction The merger and liquidation of banks are important policy options in the toolkit of regulatory authorities that confront a financial crisis. The policy decisions modifying the structure of the banking sector do not only have immediate consequences on the solvency of banks and the fiscal cost of state-support programs, but they also have a long term effect on banking competition. The changes in the banking sector structure after a severe crisis will have an impact on the financial costs of future borrowers and also the incentives of surviving banks to provide credit. This article applies a structural econometric framework to analyze the effect on mortgage market competition of the increased concentration in the Spanish banking sector that resulted from the financial crisis of 2008.

⇑ Tel.: +34 91 338 8780.

E-mail address: [email protected] 1 I wish to thank Gabriel Jimenez, Daniel Pérez, Jesús Saurina, Carlos Trucharte, the editor and two anonymous referees for their helpful comments. This article is my sole responsibility and, in particular, it does not necessarily represent the views of the Bank of Spain or the Eurosystem. http://dx.doi.org/10.1016/j.jbankfin.2014.03.004 0378-4266/! 2014 Elsevier B.V. All rights reserved.

Financial crisis and the structure of the banking sector are connected in both current policy debate and historical experience.2 During a financial crisis, regulatory incentives to contain the fiscal cost of public support and private motives to capture market share can induce the acquisition of banks with low capital by institutions with stronger financial positions. If the financial crisis produces bank failures, the liquidation and sale of these entities will lead to further increases in concentration. For high enough levels of economic losses, injections of public capital will be used as a complement to industry consolidation to restore adequate capital levels. The international financial crisis of 2008 and the end of the national real state boom produced a gradual response from Spanish banking regulators. The initial efforts on 2009 were focused on the consolidation of the sector through the creation of a public fund for restructuring Spanish banks. This consolidation process targeted specially savings banks that had expanded during the pre2 Calomiris (2000) provides an historical overview of the relation between limits to banking sector consolidation and financial stability in the U.S. Drees and Pazarbasioglu (1995) describe bank consolidation in the Nordic Crisis of the 1990s. The Proposal on a Directive for a European Bank Resolution framework in European Union Commission (2012) includes the possible sale and liquidation of banks as part of crisis resolution (see Section 4.4.10. of the proposal) of the European Union Commission.

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

vious decade and faced governance problems. As bank losses mounted in the period of 2010–2012, direct capital injections of government funds were completed and the combination of bank groups proceeded. This process lead to the consolidation of the main Spanish commercial and savings banks into fourteen groups by the end of 2012. This radical change in market structure opens up the question of whether the more concentrated bank sector behaves collusively and allocates credit very differently than the structure existing before 2008. I use counterfactual analysis to evaluate the possible anti-competitive effect on the mortgage market of the bank consolidation process in Spain after 2008. A strategic model of interest rate setting is employed to compute the counterfactual equilibrium mortgage rates and volumes that would have been set in the period 2004–2010 by the concentrated industry that emerged as result of the financial crisis. These counterfactual computations can then be compared with actual interest rates and volumes. The strategic model is evaluated with demand and cost estimates from bank-level data on the volumes and interest rates of new mortgages granted. A mixed-logit model for mortgage demand is estimated, allowing to control for the effect on loan demand of household income. I also control for the effects on demand estimates of endogenous mortgage rates with the use of GMM techniques. The expected costs for banks of new mortgages are recovered through the combination of demand estimates with the assumption of profit maximizing behavior of banks when setting mortgage rates. Given the demand estimates, the observation of a bank granting a certain volume of mortgage loans at a given interest rate reveals the cost level that makes these lending decisions profit-maximizing. This structural approach is in line with the methods popularized in empirical industrial research by Berry (1994) and Berry et al. (1995). Estimation results reveal that the demand for mortgage loans is relatively elastic, with an elasticity with respect to own loan rate of approximately !3:76. The estimated cost for banks of mortgage lending is in the range of 1.62–4.54%, with a significant correlation with interbank rates. This information is then used in the calculation of equilibrium rates for the counterfactual consolidated structure, finding a minimum variation in loan rates and volumes with respect to actual data in the period 2004–2010. The average increase in rates is only of 0.07% and the volume of new mortgages decreases on average less than 20 million euros per month. A market structure with higher degree of concentration than actually observed in 2012, and with weak savings banks concentrated in a large group, yields higher average changes for mortgage rates (+0.21%) and volumes (!51 million euros per month). This result is still found to be far from collusive rates and volumes, lessening the concerns about the anti-competitive effects of the crisis resolution on the Spanish mortgage market. However, it must also be noted that the switch to concentrated market structures does not imply a dramatic redistribution of new mortgages across entities, so the weighted expected default rates are not affected significantly. This result suggests that the change in market structure alone cannot be expected to improve credit quality without the revision of credit risk management and supervisory rules. The design of a resolution framework for banks by a financial regulator will ideally take into account its impact on market competition, in line with the analysis in the current article. However, a regulator might assign a relative low priority to evaluating competition when pressed to achieve urgent policy goals in the middle of the crisis, e.g., the protection of the payment system or avoidance of negative externalities from bank failure. If this is the case, it is still possible to complete the analysis of competition after the immediate risk of bank failure has been averted. If the market structure after resolution of the crisis is found to generate

125

inefficient credit growth or distortions in mortgage rates, a range of corrective actions can be taken. These actions could range from closer antitrust scrutiny to use of prudential regulation to require higher capital charges as function of bank conduct in credit markets. The rest of the article is organized as follows. Section 2 surveys the banking literature related to the present article. Section 3 describes the available data set and provides basic industry background. Section 4 presents the formal model of the mortgage market. Section 5 presents estimation methods and results for mortgage demand and implicit mortgage costs. Section 6 contains counterfactual analysis to measure the effect of sector consolidation on competition and default rates. Section 7 concludes. 2. Research on bank mergers and competition The analysis of U.S. bank mergers has received considerable attention since the passing of the Riegle–Neal (R–N henceforth) Interstate Banking and Branching Efficiency Act of year 1994. This act progressively developed a national banking market by lifting restrictions to the interstate operation of branches and banking sector consolidation. Berger et al. (1999) and Amel. et al. (2004) provide comprehensive reviews of the research on bank mergers after the R–N Act, with a focus on the U.S. Empirical research in the decade after the R–N Act studied the relation of bank mergers with cost and profit efficiency, economies of scale and scope, and market power. Prominent works include Akhavein et al. (1997), Berger and Humphrey (1999), Berger and Mester (1997), Berger and Hannan (1998) and Prager and Hannan (1998). The consolidation of the financial sector has remained an internationally relevant issue in more recent periods, and particularly in the context of the financial crisis of 2008. DeYoung et al. (2009) review the post-2000 literature and find that American bank mergers can increase efficiency, but that evidence on shareholders’s gains from mergers in the U.S. is mixed. For Europe, they find stronger evidence of both efficiency and shareholders’s gains. Claessens (2009) reviews recent research measuring competition in the financial sector across countries, e.g., Demirguc-Kunt et al. (2004) and Bikker and Spierdijk (2008). He also advocates lower institutional dispersion, harmonization across markets and separation from prudential regulation in the design of competition policy for the financial sector. The financial crisis of 2008 has already generated multiple studies of its causes and consequences.3 However, there has been to my knowledge limited progress in the assessment of the effect of the crisis on credit market competition. Calomiris and Pornrojnangkool (2005) point out that regulatory concerns about anti-competitive effects of concentration in lending markets before the crisis were also limited, and provide evidence that these effects can be significant for some borrowers.4 The empirical studies of European bank mergers are generally consistent with the analyses of U.S. data samples in their design and objectives. However, the institutional framework and market structure differ from the U.S. and the heterogeneity across countries conditions the analysis. Campa and Hernando (2006) find that mergers in the European financial sector in the period 1998–2002 are associated to increases in operational performance and return on equity. Hagendorff and Keasey (2009) find that European bank mergers are more closely associated with cost-cutting than Amer3 Spiegel (2011) summarizes several theoretical and empirical contributions. Santos (2011) and Murillo et al. (2011) examine the effects of the crisis on lending to firms in the U.S. Wheelock (2011) documents continued consolidation in the U.S. after the crisis of 2008. 4 Calomiris and Pornrojnangkool (2005) study the effect of the merger of Fleet and BankBoston in New England with granular firm borrower data, finding an heterogenous effect of the increased concentration on the interest spreads earned on firms of different sizes.

126

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

ican bank mergers, which are more focused on generation of additional revenues. Wilks (2009) outlines the strength of the European model of competition policy and considers how the financial crisis challenges this model. European country-specific studies often exploit unique datasets collected by national regulators. For example, Focarelli and Panetta (2003) use an Italian dataset of bank deposits at the local market level and find that bank mergers have a temporary adverse effect on deposit rates, which is more than compensated in the long term by efficiency gains. Other studies of European mergers include Focarelli et al. (2002) for Italian banks, Cuesta and Orea (2002) for Spanish banks, Vander Vennet (1996), Vander Vennet (2002) and Becalli and Frantz (2009) for European banks. The literature has been traditionally focused on bank deposit markets, but there has also been progress in the study of the effect of mergers on credit markets. Allen et al. (2013) find that a merger of Canadian banks has an average effect of 6 bp on the margins earned on mortgage contracts, but this effect is unequally distributed across borrowers, with almost no impact on the borrowers of the top of the margin distribution. Scharfstein and Sunderam (2013) measure across U.S. counties the effect of concentration and mergers on the sensitivity of mortgage rates to mortgage backed security (MBS) yields, which proxy for the costs of financing mortgages. The authors find that lower competition and bank mergers reduce the sensitivity of mortgage rates and refinancing activity to MBS yields. Kahn et al. (2005) find that increased local market concentration as result of bank mergers has a negligible impact on the rates of auto loans, which are set in a national market, and a negative impact on the rates of unsecured personal loans, which can be strategically important to win regulatory approval of a merger. For the Belgian SME sector, Degryse et al. (2011) find in a microlevel dataset that bank mergers have an heterogenous effect on borrowers with single and multiple bank relationships. Single relationship borrowers are less able to maintain or switch their bank relations after the merger, and they face lower credit availability and business performance. Xie (2007) and Erel (2011) examine micro-datasets of commercial and investment loans to U.S. firms and they find that mergers reduce on average loan spreads, indicating that efficiency gains are at least partially passed to borrowers. However, these reductions disappear in the long run, and significant market overlap between target and acquirer can lead to increases in the spreads. Sapienza (2002) finds comparable results to the studies of Xie (2007) and Erel (2011) for loans to Italian firms and mergers in the period 1989–1995. Carow et al. (2006) analyze a firm level sample of business loans and they observe that bank mergers involving large acquirers are associated with lower performance of the firms with loans in the acquired institution. The reviewed articles in this recent finance literature use datasets at the loan or local market level, which allow to define control and treatment groups of borrowers that are affected differently by bank mergers. For example, Allen et al. (2013) assign a borrower to the treatment (control) group if both merging banks were (not) present in her local market prior to the merger. Scharfstein and Sunderam (2013), Xie (2007) and Erel (2011) use measures of overlap in the bank deposits business to identify borrowers or local markets that are potentially affected by a bank merger.5 The availability of bank level data for mortgage rates and market shares in Spain, and the systemic nature of the crisis and consolidation process, which make difficult to define control and treatment groups, lead me to use a different approach that relies on a structural model 5 Studies of price differences as a function of merger activity have also been conducted in bank deposit markets, e.g., Prager and Hannan (1998),Park and Pennacchi (2009), Craig and Dinger (2009), and in other industries, e.g., Jimenez and Perdiguero (2012) study the effects of a merger on gasoline prices in localized geographical markets.

of price competition and simulation of bank merger effects. This literature with micro-level data is still relevant for the current work, as it indicates that the aggregate simulated effects that are the focus of this article will be probably distributed unequally across markets and borrowers. Structural frameworks with discrete choice models have also been recently applied to the analysis of bank deposit markets.6 Zhou (2008) uses a structural model of price and branch density choice together with a mixed-logit model to measure the impact on bank deposit markets of mergers of large U.S. banks in the period 1994–2004. He gauges the precision of the model predictions with historical data. Ishii (2008) also uses a mixed-logit supply model of bank deposits in the analysis of ATM expansion strategies of banks. Dick (2008) and Ho and Ishii (2011) examine the effect of the RN on U.S. depositors’ welfare with nested logit and mixed-logit specifications. Aguirregabiria et al. (2012) estimate a factor model for bank deposits to evaluate the geographic risk diversification possibilities of U.S. banks before and after the Riegel–Neal Act. Structural studies of the banking sector also include Gowrisankaran and Krainer (2011) for ATM networks, Ackerberg and Gowrisankaran (2006) for electronic payment systems, and Cohen and Mazzeo (2007) for branching decisions. I follow this recent empirical literature in the use of structural models to better quantify the consequences in the mortgage market of the consolidation of Spanish banks. 3. Data and industry background 3.1. Retail mortgage loans in Spain Mortgage loans for home acquisition by Spanish households amounted to 620 billion Euros in 2010, which represented 76% of the loans to households by credit institutions and 34% of total lending to the private sector in Spain. This retail market is dominated by bank deposit institutions, which provided an average of 98.7% of the mortgage loans for home acquisition in the period 2004– 2010, leaving a marginal role to non-bank financiers. The prominent role of bank deposit institutions in the mortgage market is facilitated by an extensive branch network (approximately 44,000 branches at the end of 2009), with a branch density above the average of the European Union.7 The vast majority of the mortgage loans in Spain have variable interest rates associated to the Euribor interbank rate. From the interest rate reports collected by the Bank of Spain, 91% of the loans for home acquisition granted in the period 2003–2010 had an initial fixation period of the mortgage rate of one year or less. If an initial fixation period of 5 years or less is considered, 98% of the loans for home acquisition granted in this same period are covered. Mortgage contracts in Spain are also characterized by the strong protection of lender rights, with personal liability of borrowers for the full value of the loan instead of the limited recourse to collateralized assets in other jurisdictions.8 6 Farrell and Shapiro (1990) already indicate the importance of measuring the welfare effects of mergers rather than using proxy measures of competition, e.g., concentration ratios. The use of structural methods allows to advance in this line of research, supplementing indirect welfare measures such as the weighted market shares in Farrell and Shapiro (1990). 7 See the Statistical Bulletin Sections 4.13 and 4.14 in Bank of Spain (2013) for information on the total stock of mortgage loans, total credit and the distribution of credit by type of credit institution. Vives (2012) and pp. 98–100 in Chapter 2 of Maudos and Fernández de Guevara (2008) document the high branch density of the Spanish banking system relative to other European countries. From Vives (2012), the average population per bank in Spain in 2007 was below 1000, whereas the European average was 2720. 8 Interest rate reports are described in detail in subSection 3.2. After the initial fixation period expires, annual revisions of the mortgage rate are the norm. Oliver Wyman (2012, pp. 31–32) and IMF (2012, pp. 15–16) recognize strong lender protection as a feature of the Spanish mortgage market.

127

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136 Table 1 Summary statistics.

Share of New Mortgages (%) Mortgage Rate (%) Branches Employees Employees per Branch Labor Cost (million Euros) Branch Network Cost (million Euros) Deposit Rate (%) Euribor 12 Months (%) GDP Growth (%) Median Household Income (thousand Euros) 1 Year Treasury Bill (%) 10 Year Bond (%)

Mean

Std. dev.

Min

Max

1.65 4.05 664 3992 5.76 153 92 2.81 3.01 1.75 12.05 2.62 3.99

2.45 1.08 887 5645 1.42 284 172 1.11 1.29 2.75 1.10 1.15 0.38

0.01 1.65 41 260 3.79 3 2 0.48 1.22 -4.50 10.48 0.74 3.09

18.80 6.98 5567 30588 11.63 2230 1558 9.93 5.39 4.30 13.30 4.55 4.80

Note. Sources (Bank of Spain, National Institute of Statistics, Eurostat). Share of New Mortgages is the ratio of new mortgages granted by each sample bank and the total system flow of new mortgages. Mortgage Rate is the initial interest rate charge on new mortgages to households in a given month. Branches, Employees and Employees per Branch are respectively the total number of branch offices, total employees and ratio of employees and branch offices of each sample bank. Labor Cost and Branch Network Cost are the quarterly costs associated to employees and branch operations. Deposit Rate is the interest rate compromised on new bank deposits from households in a given month. Euribor 12 Months is the Euribor interbank interest rate for a 12 month maturity. GDP Growth is the annual growth in real GDP computed at the end of every quarter. Median Household Income is the 5th decile of the income distribution of Spanish households. 1 Year Treasury Bill and 10 Year Bond are the interest rates paid on Treasury Bills and Spanish Government Bonds with maturities of one and ten years.

3.2. Data set The regulation of the Bank of Spain, BoS henceforth, in accordance with European Union Directives, requires banks to provide monthly reports of the volume and interest rate of new mortgages granted to households with the purpose of home acquisition.9 I use this information to form a data set of the volume, market share and interest rates of new mortgages for the fifty-five main commercial and savings banks in the Spanish banking sector for every month in the period 2004–2010. The market share of a bank in a given month is calculated as the ratio of its individual volume to the total volume of new mortgages for home acquistion in a month for all reporting institutions. The interest rate of a bank in a given month is calculated as the weighted average by volume of the actual interest rates in all the new mortgages for home acquisition granted by this entity in a given month.10 The use of the actual volumes and rates applied by banks every month avoids the need to use proxies for mortgage rates such as the ratio of interest revenues and the stock of loans to households. The main data set is complemented with macroeconomic variables, data from income statements and data on deposit rates. The household income data is drawn from the European Statistics on Income and Living Conditions, SILC henceforth, collected by the Eurostat. The data in the SILC contain equivalised household income, which corresponds with the gross household income adjusted by the number of members of the household. Real GDP growth at the end of every quarter is collected from the National Institute of Statistics of Spain. Interbank rates and the interest rates of Spanish public debt are obtained from the Statistical Bulletin of the BoS. Banks have to file confidential income statement data to the BoS every quarter. I collect information on labor costs, branch

9 Interest rate reporting requirements are set up in the BoS orders 4/2002 and 1/ 2010, in accordance with European Central Bank Regulations 290/2009. Banks with assets in excess of 1.5 billion euros and euro denominated deposits in excess of 500 million euros are compelled to report interest rate forms. The BoS can also request this information to banks that are not above the threshold. See Kingdom of Spain (2010a). 10 From Part 2 of Norm 3 in Circular 1/2010, the reported interest rate will equate the present value of the borrower’s money inflows and outflows that result from the mortgage contract, excluding associated fees. This is a pure measure of the interest cost of the mortgage.

costs, number of branches and number of employees from these reports. Monthly interest rate reports to the BoS are also used to form a series of deposit rates paid by banks on the new deposits contracted by households in a given month. Table 1 provides summary statistics for the main variables. The study covers the years 2004–2006 with an ongoing real state boom, and a period of severe financial crisis in 2007–2010. Fig. 1 plots the evolution of the loan volume and interest rate in the mortgage market during this period. The volume of new mortgages granted per month increases over 60% from the first quarter of 2004 to the end of year 2006, and it then starts a marked decline that leads to a 50% reduction from 2006 to 2010. This reversal in the series of mortgage volumes coincides with the first symptoms of the financial crisis in 2007, and the hardening of the financing conditions for banks. The rise of the Euribor 12 months interbank rate in the period 2007–2008 up to a maximum of 5.35% in 2008 is indicative of the financing difficulties for banks in the period. The subsequent decrease of interbank rates to 1.25% at the end of June of 2010 has not lead to an increase in the volume of new mortgages. This suggests that either banks cannot access the interbank market at these rates, or that profitable lending opportunities are scarce due to stagnant economic activity. The majority of mortgages in Spain are granted under a variable interest rate regime with the Euribor 12 months as a reference. This leads to a fast adjustment of the effective interest rate cost of mortgages for households to financial market conditions. Given this contractual setting, the initial mortgage rates are also referenced to the interbank rate at the date the mortgage is granted. The evolution of the average mortgage rate is thus parallel to the Euribor 12 months, as seen in Fig. 1. I also use data from the Credit Register of the BoS, CR henceforth, to calculate the mortgage default rate at the individual bank level for the end of every quarter in the period January 2004–June 2010.11 Fig. 2 plots the median and the top and bottom deciles for the mortgage default rates at the end of every quarter in the sample period. As the economic activity in Spain deteriorated and GDP 11 The default status in the CR is determined through the application of the Basel II definition for default events, i.e., obligations that are past due more than 90 days, or those that are considered to be highly unlikely to be repaid. See Basel Committee on Banking Supervision (2006), paragraphs 452–453, page 100. For more information on the research use of the CR, see Jimenez and Saurina (2006), Jimenez et al. (2006), and Jimenez and Mencía (2009).

128

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

Fig. 1. Mortgage volumes and interest rates. Note: This figure plots the Euribor 12 moths and the average interest rate on new mortgage loans (the scale is presented in percentage terms on the right axis), together with the volume of new mortgages (the scale is presented in Bn Euros on the left axis) for every quarter from Q1 2004 to Q2 2010.

Fig. 2. Mortgage default rates. Note: This figure plots the cross sectional median, 1st decile and 9th decile of the default ratio for mortgage loans at the end of every quarter from Q1 2004 to Q2 2010.

growth became negative, both the median default rate of mortgages and the dispersion of credit quality across banks increase. 3.3. Consolidation of the Spanish banking sector In this subsection, I provide a brief account of the merger processes in the Spanish banking sector that resulted from the financial crisis. This overview makes clear that the operations of savings banks were not integrated by June 2010 and that the consolidation process was subject to important revisions in the period 2010– 2012. These facts motivate me to use bank level data for estimation in the period 2004–2010 and to study the counterfactual effects of a concentrated industry structure if it had been in place before the financial crisis, rather than emerging slowly over the years 2010– 2012.12 As the deterioration of the Spanish economy became apparent in 2009, the Spanish government passed the Royal Decree Law (RDL henceforth) 9/2009, Kingdom of Spain (2009), to create the Fund for Orderly Bank Restructuring, FOBR henceforth, that would guide the consolidation of the Spanish banking sector to boost its 12 A complete account of the restructuring process of the Spanish banking sector can be found on the BoS webpage at: http://www.bde.es/bde/en/secciones/prensa/ infointeres/reestructuracion/. It is of special interest to review the informative notes on the state of the savings banks on June 2010 and July 2011, Bank of Spain (2010), Bank of Spain (2011), and the background report of the external evaluator, Oliver Wyman (2012).

solvency and efficiency. The FOBR would provide capital and supervise the mergers of banks with viable restructuring plans. It would also intervene and facilitate the liquidation of institutions with negative economic value. This consolidation process targeted savings banks, as they were regarded to suffer from significant excess capacity from expansion in the boom years and their legal structure limited their access to capital markets. In particular, savings banks were not allowed to issue equity shares, relying exclusively on retained earnings to increase core capital, and they had complex governance structures with the involvement of regional public governments. The consolidation process initiated by the RDL 9/2009 was complex, and it needed time to develop, both due to the required administrative procedure and the necessary bargaining between the managers of the former savings banks. Out of the original 45 savings banks, 38 institutions were involved in twelve restructuring processes and one of them was intervened by the BoS by June 2010. The number of savings banks fell to 18 as a consequence of the process. On July of 2010, the RDL 11/2010, Kingdom of Spain (2010b), also allowed savings banks to own a commercial bank in order to segregate their banking business into this commercial bank and facilitate access to equity markets. This segregation of the banking business required additional time and these processes were only partially completed on July of 2011. In the meanwhile, the protracted economic and financial crisis in Spain had lead the government to pass the RDL 2/2011, Kingdom of Spain (2011), that raised the capital requirements for all banking institutions. Nine

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

129

Table 2 Bank consolidation in Spain (2010–2012).

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Post-consolidation groups

Pre-consolidation individual banks

Santander BBVA Caixabank BFA-Bankia Banco Sabadell Banco Popular Ibercaja-Caja 3 Unicaja-CEISS Kutxabank Catalunyabanc NCG Banco Banco Mare Nostrum Bankinter Liberbank

Santander, Banesto BBVA, Caja Manlleu, Caja Sabadell, Caja Terrasa La Caixa, Caja Gerona, Caja Navarra, Caja Canarias, Caja de Burgos, Caja Sol, Caja Guadalajara, Banco de Valencia Caja Madrid, Bancaja, Caja de Ávila, Caja Segovia, Caja Laietana, Caja Insular de Canarias, Caja Rioja Banco Sabadell, Banco Guipuzcoano, CAM Banco Popular, Banco Pastor Ibercaja, Caja Tres, Caja de Badajoz, Caja Círculo Unicaja, Caja España, Caja Duero, Caja de Jaen BBK, Kutxa, Caja Vital, Caja Sur Caja Cataluña, Caja Manresa, Caja Tarragona Nova Caixa, Caixa Galicia Caja La General, Caja Murcia, SA Nostra, Caja Penedés Bankinter Caja Asturias, Caja Cantabria, Caja Extremadura, Caja CLM

Note: This table lists on its left column the major banking groups as of year-end 2012 resulting from the consolidation process in the period 2010–2012. The original individual banks that operated separately before 2010 are listed on its right column.

out of the 18 consolidated groups of savings banks, and four commercial banks, needed additional capital as result of the RDL 2/ 2011. The adverse evolution of the Spanish economy and the financial position of many banks lead to new mergers and the break-up of some of the pre-existing agreements during the period 2011– 2012. At the beginning of 2012, there remained uncertainty regarding the solvency of the institutions that resulted from the consolidation of the savings banks. In order to resolve this issue, the Spanish government passed in 2012 further legislation regarding the solvency of banks, the provisions for real estate assets and the reorganization of credit institutions in the RDLs 12/2012, 18/ 2012 and 24/2012 (Kingdom of Spain, 2012a,b,c). However, national reforms were not deemed sufficient to ensure the solvency of the banking system and the Spanish government requested external financial support in June 2012. In year 2012, the Spanish banking system went also through the evaluation of the IMF as part of a Financial Stability Assessment Process (FSAP) and it was scrutinized by private external evaluators.13 As result of the more negative evaluation of the solvency of the financial sector and the multiple legal changes in year 2012, banks further consolidated into 14 main groups at the end of 2012. Table 2 lists the banking groups as of year-end 2012 that constitute the output of this complicated process. 4. A model of the mortgage market I consider a discrete time economy t 2 f1; 2; . . . ; 1g with a mass M t of home-buyers in need of a mortgage loan at each period t. There is a set of J differentiated banks that can provide mortgage loans and compete in prices (mortgage rates). There is a common discount factor d that applies to all agents. I detail next the specification of demand and bank competition. 4.1. Demand for mortgage loans A potential borrower i derives an utility uijt from the grant of a mortgage loan by bank j at period t. Formally,

uijt ¼ xjt b þ kj þ kt þ njt ! rðmÞjt ait þ eijt

ð1Þ

where xjt is a vector of bank characteristics, rðmÞjt is the initial mortgage rate offered to potential borrowers, kj and kt are bank-invari13 Roland Berger and Oliver-Wyman completed top-down stress tests by June 2012. An external audit of bank accounts and a bottom-up stress test by Oliver-Wyman were completed by September 2012.

ant and time-invariant fixed effects, njt is the unobserved mean utility of a contract with bank j at time t, and eijt is a purely idiosyncratic utility shock with the standard Type-I (Gumbel) extreme value distribution. As it is standard in the literature, borrowers have access to an outside option (j ¼ 0) with zero mean value. The parameters b capture the mean effect of bank characteristics on the borrower’s utility from a mortgage contract. The price coefficient ait is assumed to be a C 1 function Gð&Þ of the income of the borrower Iit , i.e., ait ¼ GðIit Þ. If we assume ait ¼ a 2 R1 , the model reduces to the standard multinomial logit specification. If we assume ait ¼ a=Iit , the model adopts a simple mixed-logit specification. The distributional assumption on eijt allows to derive an analytic expression for the probability of contracting a mortgage with bank j conditional on the individual characteristic ait . Formally,

Prijt ðait ; &Þ ¼

1þ

expðxjt b þ kj þ kt þ njt ! rðmÞjt ait Þ PJ q¼1 expðxqt b þ kq þ kt þ nqt ! rðmÞqt ait Þ

ð2Þ

The expected market share sjt for a bank j at time t will equal the expectation of Prijt in (2) across the borrower population. That is,

sjt ¼

Z

Iit 2AI

Prijt ðait ; &Þ & fI ðIit Þ@Iit

ð3Þ

where the term ait is a function of income, i.e., ait ¼ GðIit Þ, and the distributional support and density function of Iit are denoted as AI and fI ðIit Þ. The total demand for the mortgages of a bank j at time t is simply the product of market size M t and the expected market share sjt . 4.2. Competition for mortgages I assume that banks set mortgage rates non-cooperatively at every time period t, which constitutes a differentiated temporal market. Bank managers internalize the effect of mortgage rate rðmÞjt on its own market share, as defined in (2) and (3), but they do not account for business stealing effects. In addition, banks will take into account the future evolution of interbank rates and the probability of default of mortgages. A bank j that sets an initial mortgage rate rðmÞjt at a given date t anticipates that future changes in interbank rates will generate revisions of the interest rate revenues from the mortgage portfolio. These interest rate revisions are important in the Spanish mortgage market, which relies on a variable rate contracting framework. The variation over time in financing and operating costs and the probability of borrower default introduce additional uncertainty.

130

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

Bank managers form an expectation of mortgage loan profits based on the available information ijt of bank j at time t. The information set ijt contains public information (macroeconomic variables, public accounts of banks, etc.), but it can also contain private information available only to the bank j. Formally, the expected profit E½pjt jijt ( is defined as:

! h i E½pjt jijt ( ¼ Mt & sjt ðrðmÞjt ; &Þ & d & rðmÞjt ! E½cðmÞjtþ1 jijt ( ! S h i X s þ d & rðmÞjt þ E½DrðmÞjtþs ! cðmÞjtþs jijt ( s¼2

ð4Þ

where cðmÞjtþs is the average cost per monetary unit at date t þ s of the mortgage originated at t (including losses from borrower default), DrðmÞjtþs is the adjustment to mortgage rates derived from interest rate changes from period t þ 2 onward and S is the average maturity of the mortgage loans originated at time t, with time index s 2 f2; . . . ; Sg. The expression in (4) can be written more compactly as:

E½pjt jijt ( ¼ Mt & sjt ðrðmÞjt ; &Þ & d) & ðrðmÞjt þ Rjt =d) Þ

ð5Þ

where d) ¼ ðd ! dSþ1 Þ=ð1 ! dÞ and Rjt is the discounted expected return correction from interest rate adjustments and liability costs according to the information set ijt .14 Formally,

Rjt ¼ !d & E½cðmÞjtþ1 jijt ( þ

S X ds & E½DrðmÞjtþs ! cðmÞjtþs jijt ( s¼2

ð6Þ

I adopt the assumption of Bertrand–Nash competition in mortgage rates across the J differentiated banks. Each bank j sets rðmÞjt to maximize the individual profits in (5). The assumptions on mortgage demand imply that (5) is continuously differentiable on rðmÞjt and the necessary first order condition for individual optimality of rðmÞjt is given by:

@E½pjt jijt (=@rðmÞjt ¼ 0 $

@sjt ðrðmÞjt ; &Þ=@rðmÞjt & ðrðmÞjt þ Rjt =d) Þ þ sjt ðrðmÞjt ; &Þ ¼ 0

ð7Þ

More generally, a bank group g formed by several individual banks will choose mortgage rates frðmÞjt gj2g to maximize P j2g E½pjt jigt (, with the expectations of fRjt gj2g now defined with respect to the information set igt of the banking group. As a result, the necessary optimality condition for each mortgage rate rðmÞjt of the banking group is:

Xh q2g

i @sqt ðrðmÞqt ; &Þ=@rðmÞjt & ðrðmÞqt þ Rqt =d) Þ þ sjt ðrðmÞjt ; &Þ ¼ 0

ð8Þ

5. Estimation and results 5.1. Estimation method I follow the approach introduced by Berry et al. (1995), BLP henceforth, to estimate the demand model for mortgage loans in Section 4. The model parameters fb; k; ag are identified through the assumption that there exists a ð1 * KÞ vector of instruments Z + fx; Wg, with W formed by excluded instruments from Eq. (1), such that E½Z T & n( ¼ 0. I use the log number of branches and employees per branch as exogenous bank characteristics x 2 Z. On the contrary, the initial mortgage rate rðmÞ is excluded from the set of instruments Z, as 14 The average maturity of mortgage loans can vary for each bank j and mortgage revisions can start after t þ 2 with no effect on the estimation exercise, as all these unobserved terms are collected in Rjt =d) . I fix these elements in (4) for ease of presentation.

it is likely to be correlated with unobserved mean utility n. This standard precaution avoids to attribute to the price variable rðmÞ variations in demand that are a consequence of quality factors of the bank (changes in reputation, promotional campaigns, etc.) that are not observed by the econometrician. I use as price instruments the quarterly growth in labor costs and branch operational costs. I also include in the list W the six-month lag of the deposit rate relative to market average, as a proxy of financing costs.15 In addition to these cost variables, I use the sum of the squared deviations of the P 2 log number of branches, i.e., q–j ðlogðBranchesj Þ ! logðBranchesq ÞÞ , as the relative differences in the bank network with respect to rivals affect the substitution possibilities of borrowers and condition competition in prices.16 In order to form a sample analog of E½Z T & n( ¼ 0, it is necessary to extract the mean value djt ¼ xjt b þ kj þ kt þ njt from the market share Eq. (3) for each period and bank. In this task, I use the contraction algorithm introduced by Berry (1994). For every candidate value of the parameters a, this procedure extracts each djt by solving the equation

es t ¼ st ða; dt Þ

ð9Þ

where the term es t stacks the market share data for period t; es t ¼ ½es 1t ; . . . ; es jt ; . . . ; es Jt (, whereas the term st ða; dt Þ stacks the model prediction of market share in Eq. (3) for period t given the vector of mean values dt ¼ ½d1t ; . . . ; djt ; . . . ; dJt (. In the sample, a given month is treated as a separate temporal market and (9) is then solved for each month providing a set of J & T values fb d1; . . . ; b dt ; . . . ; b d T g. The implementation details are relegated to Appendix A. The sample analog to E½Z k & n( ¼ 0 for the kth variable in Z is then computed as:

mk ðb; k; a; &Þ ¼

PJ PT j¼1

b a; es t Þ ! xjt b ! kj ! kt Þ

t¼1 Z k;jt ð d jt ð

J&T

ð10Þ

The objective function is formed as ! ¼ Mðb; k; a; &ÞT & N & Mðb; k; a; &Þ&, where Mðb; k; a; &Þ ¼ ½m1 . . . mk . . . mK (T is a (K * 1) vector that stacks the sample analogs of all moments and N is a (K * K) weight matrix.17 The first order condition for profit maximization in (7) can be used to infer the unobserved adjustment Rjt =d) to the return on mortgage loans from the observed rates and volumes of new mortgages. If this correction is negative, it can be interpreted as the expected discounted cost of mortgages granted by bank j at time t. Formally, the elements in (7) can be rearranged into:

Rjt =d) ¼

!sjt ! rðmÞjt @sjt =@rðmÞjt

ð11Þ

where @sjt =@rðmÞjt is estimated from the demand model. I regress the implicit returns adjustments Rjt =d) obtained through the application of (11) on bank fixed effects and the interest rates in the interbank and public debt markets. This allows to study the correlation of Rjt =d) with macroeconomic conditions and measure the sensitivity of the returns of mortgage loans to different interest rate scenarios. 5.2. Results for the Spanish mortgage market (2004–2010)

15 I consider the deposit rate relative to market average to use variation in financing costs that is specific to each bank, rather than macro-factors that would be correlated with market fixed effects. The lag is taken as precaution against possible correlation of deposit rates with contemporary factors that affect both deposit and mortgage rates. 16 Train and Winston (2007) propose adding average squared differences of exogenous variables to the set of instruments based on transformations of exogenous variables introduced in BLP (1995). P P 17 I use a 2SLS robust matrix computed as Jj¼1 Tt¼1 Z Tjt & Z jt . The observation of Z jt for bank j and period t is formed by the (1 * K) vector Z jt + ½Z 1;jt ; . . . ; Z k;jt ; . . . ; Z K;jt (.

131

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136 Table 3 Mortgage demand

b Ln (Total Branches) Ln (Employees per Branch)

(a) Coeff.

Std. error

(b) Coeff.

Std. error

(c) Coeff.

Std. error

1.0445 1.2025

0.0086*** 0.0417***

0.5343 0.7452

0.1732 0.2018

0.5526 0.6876

0.2511 0.3104

!0.4180 – Yes No

0.0258

!0.8543 – Yes Yes

0.1295

Mean

Std. dev.

Mean

Std. dev.

Mean

Std. dev.

!1.67 0.03 0.80 – 4125

0.45 0.04

!3.43 0.06 – 4.66 (3) 4125

0.93 0.08

!3.76 0.07 – 2.09 (3) 4125

0.87 0.10

*** ***

** **

a Mortgage Rate Mortgage Rate/Income Trimester Fixed Effects Bank Fixed Effects

Average Own Rate Elasticity Average Cross Rate Elasticity R2 Hansen J-Stat Number of Obs.

***

***

– !1.3133 Yes Yes

0.3571

***

Note: This table reports estimates for the model of mortgage demand in Eqs. (1)–(3). Specification (a) reports estimates for a logit demand model with exogenous mortgage rate; OLS estimation. Specification (b) reports estimates for a logit demand model with endogenous mortgage rate; GMM-IV estimation. Specification (c) reports estimates for a mixed-logit demand model with endogenous mortgage rate interacted with household income; GMM-IV estimation. Degrees of freedom for the Hansen J-Stat are provided in parentheses. ** Significant at 5% level. *** Significant at 1% level.

Table 3 presents the demand estimates for the different specifications. A simple multinomial logit specification (ait ¼ a 2 R1 ) estimated with OLS is presented in column (a). The coefficients on bank characteristics and mortgage rates are significant and the implied model elasticity for changes in the own interest rate of a bank is !1:66. This estimate can contain bias if the mortgage rate is an endogenous variable, as considered in Subsection 5.1. The specification in column (b) introduces kj for unobserved bank fixed effects and substitutes OLS with GMM estimation, with use of the set of excluded instruments W defined in Subsection 5.1.18 The coefficient on the mortgage rate more than doubles and the new estimate of average own price elasticity in column (b) increases to !3:43. The cross elasticity also was underestimated under the specification in (a), given the change from value 0:03 to 0:06. Heterocedasticity robust standard errors are used for all specifications. The specification in (c) uses a mixed-logit specification ðait ¼ a=Iit ; a 2 R1 Þ that adjusts the impact of the mortgage rate rðmÞjt on the borrower’s choice of bank by individual income Iit . The own-price elasticity estimate is now !3:76, whereas the cross price elasticity is 0:07. As it is well known, e.g., BLP (1995) or Train (2009), the mixed-logit specification allows a more flexible substitution pattern across alternatives than the standard logit. I use thus the estimates in specification (c) in the calculations in the subsequent sections. Table 4 presents the analysis of the implied return corrections Rjt that result from the application of Eq. (11) and the results of demand specification (c) in Table 3 to the data on market shares and mortgage rates. The implied return corrections are negative through the sample, indicating that adverse interest rate revisions and expected cost elements dominate possible expected gains from interest rate revisions. The Panel I of Table 4 shows that the average implied return correction decreases from !2:37 in 2004 to !4:54 in 2008, coinciding with the ascending trend of the interbank rate from 2:2% in 2004 to 5:87% in 2008. Given that interest rates in mortgage contracts are variable and referenced to the interbank rate, this observed pattern is reasonable. Bank managers would expect that the initial interest rates set up during periods of high interbank rates will be revised downward in the future, as interbank rates revert to average levels. As the interbank rate de18 The use of quarterly growth rates in costs as instruments requires to sacrifice the first quarter of 2004, and estimation is based on the period April 2004 to June 2010.

creases abruptly in 2009–2010, the return correction becomes less negative, pointing that upward revisions from low interest rate levels in this sub-period are more likely. A higher (lower) interbank rate is also associated to more (less) elevated financing costs for banks. The Panel II of Table 4 presents a more formal analysis of the relation of return corrections Rjt and the interest rates in interbank and public debt markets. Column (a) presents the OLS projection of Rjt on the Euribor 12 Months, the interest rates on the 1 Year Treasury Bill and 10 Year Government Bond and bank fixed effects. The coefficient estimate for the Euribor 12 months is negative and significant, implying more negative return corrections when the interbank rate is higher. The coefficient for the 1 Year Treasury Bill rate is positive, whereas the coefficient for the 10 Year Government Bond rate is found negative. The variable rate regime for mortgages connects directly the Euribor 12 month and the rate revisions embedded in Rjt , and it is reasonable to find significant correlation in the data. The relation of the bank costs of the mortgage portfolio with government interest rates is less direct and a more ambiguous relation is thus estimated.19 Column (b) of Table 4 presents the projection of Rjt on market interest rates allowing for different coefficients on these variables before and after year 2007 through the inclusion of interactions of interest rates and the indicator Iyear>2007 . The indicator Iyear>2007 is also included directly to control for the average effect of the latter period on return corrections and avoid the introduction of bias in the interactive terms, as discussed for example in Braumoeller (2004), Brambor et al. (2006) and Ozer-Balli and Sorensen (2013). This specification controls then for the possibility that the severe financial crisis that has afflicted Spain since 2008 could alter the relation between return corrections Rjt and market interest rates. The estimates show that this relation is stable, with insignificant coefficients for the interactions of the Euribor 12 months, 1 Year Treasury Bill and 10 Year Government Bond with the indicator Iyear>2007 . Reported standard errors in all specifications are robust to clustering in the time dimension. 19 High government interest rates can inform of adverse financial market conditions that increase borrowing costs for all agents. However, the sensitivity of bank borrowing costs to short term and long term government rates can vary greatly as a function of portfolio composition, e.g., positions in repo markets. Additionally, public support to banks deteriorates government finances, while reducing the financing cost of banks.

132

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

Table 4 Implied mortgage return corrections. I. Average implied return correction

December-2004 December-2005 December-2006 December-2007 December-2008 December-2009 June-2010

II. Determinants of implied return corrections (a) (b)

Mean

Std. dev.

!2.37 !2.38 !3.53 !4.29 !4.54 !1.62 !1.41

0.20 0.15 0.25 0.19 0.40 0.42 0.40

Euribor 12 Months Treasury Bill 1 Year Govt. Bond 10 Years Euribor 12 Months * Iyear>2007 Treasury Bill 1 Year * Iyear>2007 Govt. Bond 10 Years * Iyear>2007 Iyear>2007 Constant R2 Number of Obs.

Coeff.

Std. error

Coeff.

Std. error

!1.2022 0.5771 !0.1234 – – – – !0.7904 0.918 4125

0.0970 *** 0.1023 *** 0.0674 * – – – – 0.243537***

!1.1338 0.4405 !0.1164 !0.0962 0.1858 0.0887 !0.5697 !0.6350 0.920 4125

0.1104 0.1207 0.0526 0.1676 0.1966 0.3386 1.2779 0.1968

*** *** **

***

Note: This table reports information on the mortgage return corrections implied by the demand estimates from specification (c) in Table 3 and the competition model in Section 4.2. The Panel I shows the cross sectional averages and standard deviations of the implied return corrections Rjt at selected dates. The Panel II shows the OLS estimates for a linear projection of implied return corrections Rjt on the Euribor 12 months, interest rate on the 1 Year Treasury Bill, 10 Year Government Bond and bank fixed effects. Standard errors are robust to clustering in the time dimension. The specification (a) considers a constant effect of interest rate variables, whereas specification (b) allows for a change in the relation between implied return corrections Rjt and interest rates after 2007 (Iyear>2007 is an indicator for dates posterior to 2007). * Significant at 10% level. ** Significant at 5% level. *** Significant at 1% level.

6. Counterfactual analysis

6.2. Policy experiments

6.1. Methodology

6.2.1. Competition analysis The application of the method in Subsection 6.1 allows to ompute for every month the Nash equilibrium mortgage rates nder the consolidated bank structure, which I denote, rCt ¼ frðmÞCjt gj¼1;...;J . The market shares evaluated at counterfactual rates r Ct are sCt ¼ fsCjt gj¼1;...;J and the corresponding volumes, qCt ¼ fqCjt gj¼1;...;J can then also be computed for every month. The volume of mortgage loans of an individual bank j in month t is calculated as qCjt ¼ sCjt & M t , where M t is the actual market size of the mortgage market in a given month. The actual data on mortgage rates, e r t ¼ fe r ðmÞjt gj¼1;...;J , market shares, es t ¼ fes jt gj¼1;...;J and volqt ¼ fe q jt gj¼1;...;J can then be used to calculate the differences umes, e between counterfactual and actual interest rates, volumes and bank profits:

I use the model of competition introduced in Subsection 4.2 to compute different counterfactual equilibria in the market for mortgage loans through the period 2004–2010 for a concentrated market structure. In particular, I merge the individual banks of the sample into consolidated bank groups and compute for every month the non-cooperative Nash equilibrium in mortgage rates that results from this structure. From the model in Subsection 4.2, each bank group chooses the mortgage rates offered by their member banks to maximize expected profits, i.e.,

max

frðmÞjt gj2g

X E½pjt ðrðmÞjt ; :Þjigt ( j

ð12Þ

where igt is the information set of the group g and rðmÞjt and pjt ð&Þ are respectively the mortgage rate and profit of an individual bank j inside the group g. The managers of group g internalize the impact on all of its members of altering the mortgage rate rðmÞjt offered by one of their individual bank brands, as recognized in the first order condition (8). A Nash equilibrium in mortgage rates at month t is a NE NE NE set of rates r NE t + frðmÞ1t ; . . . ; rðmÞjt ; . . . ; rðmÞJt g that does not leave the opportunity to a bank group of unilaterally increasing its profits by altering the rates frðmÞjt gj2g under its control. Equivalently, bank group g chooses in a Nash equilibrium rates frðmÞNE jt gj2g that maximize the expected profit in (12) taking as fixed parameters the rates offered by the remaining banks frðmÞNE jt gjRg . Appendix B details the computation of the Nash equilibrium. The computation of the Nash equilibrium is made feasible by the available estimates of loan demand sjt ðrðmÞjt ; &Þ and return corrections Rjt =d) , which allow us to calculate E½pjt ðrðmÞjt ; &Þjigt ( ¼ sjt ðrðmÞjt ; &Þ & ðrðmÞjt þ Rjt =d) Þ from (5).20 In the policy experiments below, I use the demand estimates and return corrections Rjt =d) following from the specification (c) in Table 3. 20 Note that we identify pjt ðrðmÞjt ; :Þ up to multiplication by a positive constant given that d) is unobserved. However, M t & d) is a positive constant with no effect on the choice of rðmÞjt and we can normalize expected profits by it.

Drt ¼ frðmÞCjt ! er ðmÞjt gj¼1;...;J Dqt ¼

fqCjt

Dpt ¼ fp

q jt gj¼1;...;J !e

C e jt = jt

p ! 1gj¼1;...;J

ð13Þ ð14Þ ð15Þ

e jt is defined as pCjt = p e jt ¼ where the relative profit difference pCjt = p C ) ) 21 C e e ½sjt & ðrðmÞjt þ Rjt =d Þ(=½ s jt & ð r ðmÞjt þ Rjt =d Þ(. The Panel I of Table 5 reports these counterfactual variations for different market structures for the months at the end of every semester in the sample. Firstly, I analyze the effect on mortgage market competition during the period 2004–2010 of consolidating those individual banks affected by the merger wave of 2010–2012 into the 14 groups listed in Table 2. The results, under the heading Consolidated Structure I in Table 5, indicate that the increase in concentration has an almost nihil effect on equilibrium interest rates, with an average increase of 0:07% through the sample. There is little variation in the average change in interest rates through both the sub-periods of expansion (2004–2007) and contraction (2008–2010).22 The negligible change 21 It is impossible to identify the absolute variation of profit, as the unobserved element d) is part of pjt ð&Þ in (5 ). This is no obstacle to identify the relative profit e jt . change as d) cancels out in pCjt = p 22 Both the counterfactual and actual rates vary through time as the demand and cost conditions of bank j change, but they do it roughly in paralell leading to the low time variation in Dr jt . In complementary work, I have verified the weak correlation of Dr jt with a simple measure of the specialization of bank j in mortgages, indicating that this measure is not sufficient to capture the individual bank conditions that drive the simulations.

133

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136 Table 5 Counterfactual analysis of market structure. I. Analysis of Competition

Dates June 2004 December 2004 June 2005 December 2005 June 2006 December 2006 June 2007 December 2007 June 2008 December 2008 June 2009 December 2009 June 2010 II. Analysis of Credit Quality

Dates June 2008 December 2008 June 2009 December 2009 June 2010

Consolidated Structure I

Consolidated Structure II

Cartel Structure

Drjt (%)

Dqjt (m Eur)

Dpjt (%)

Drjt (%)

Dqjt (m Eur)

Dpjt (%)

Drjt (%)

Dqjt (m Eur)

Dpjt (%)

0.068 0.066 0.078 0.075 0.086 0.065 0.083 0.076 0.080 0.079 0.093 0.084 0.081

!22.442 !21.933 !32.519 !34.660 !33.597 !14.740 !23.596 !14.683 !14.530 !7.596 !18.827 !19.088 !26.443

4.071 4.172 4.480 4.423 4.819 3.373 4.676 3.716 3.573 4.344 3.485 3.374 3.162

0.169 0.169 0.198 0.193 0.223 0.158 0.213 0.208 0.222 0.234 0.271 0.244 0.226

!53.555 !55.246 !78.502 !88.189 !94.414 !38.837 !58.976 !44.391 !45.331 !20.394 !56.137 !49.982 !61.984

8.778 9.530 10.199 10.274 11.835 7.418 10.498 10.064 9.927 12.602 10.725 10.181 8.490

4.374 5.262 5.459 5.872 6.523 7.810 6.287 6.826 6.968 7.474 4.687 4.814 4.711

!4634.031 !4808.899 !6124.032 !6495.195 !6262.142 !6487.751 !5588.428 !4377.544 !3737.567 !2322.784 !2830.118 !2509.039 !3735.960

188.075 254.318 223.156 246.549 295.522 350.890 232.350 225.441 214.918 280.972 141.669 166.677 153.736

Consolidated Structure 1

Consolidated Structure II

Cartel Structure

DPD ! Alt1ð%Þ

DPD ! Alt2ð%Þ

DPD ! Alt1ð%Þ

DPD ! Alt2ð%Þ

DPD ! Alt1ð%Þ

DPD ! Alt2ð%Þ

!0.02 !0.03 !0.03 !0.02 !0.02

!0.14 !0.19 !0.37 !0.12 !0.18

!0.10 !0.16 !0.13 !0.11 !0.10

– – – – –

0.01 0.04 0.05 0.04 0.03

– – – – –

Note: The Panel I of this table reports for the months of June and December of each year the counterfactual changes in the average mortgage rate (Dr jt Þ, total volume of new mortgage loans ðDqjt Þ and relative profit (Dpjt Þ that result from the transition of actual to hypothetical market structures. Consolidated Structure I considers a structure with individual banks merged into the groups formed in 2010–2012. Consolidated Structure II considers a structure with a consolidated group that gathers weak savings banks and merges other individual banks into the groups formed in 2010–2012. Cartel Structure considers the structure of the perfectly collusive cartel. The Panel II of this table considers the changes in the average default frequency from Eq. (17) corresponding to the change in market structure. In column DPD ! Alt1, each bank is assigned its sample default frequency as post-merger default rate PDjt in (17), whereas the default frequency of the leader of the bank group is assigned as PDjt in (17) for DPD ! Alt2.

in mortgage rates is translated into a very small a variation in loan volumes (in the range of 8–34.6 million euros in a given month) and a small relative change in profits (in the range of 3–4%). The more concentrated structure that resulted from BoS policy in response to the financial crisis of 2008 leaved competition basically unaltered. The consolidated groups internalize the profit loss from competition between their members and are thus more prone to increase mortgage rates, but this effect is very small. Competition between the banks resulting from the merger process is almost as intense as with the pre-existing regime. The concern that increased concentration in the Spanish bank sector might lead to monopolization of loan markets does not seem grounded for the case of the mortgage market. This result is conservative in that it does not consider possible cost reductions following the mergers, which could even reduce mortgage rate levels with respect to their actual levels. I consider a second counterfactual market structure in which some of the groups resulting from the historical consolidation process of 2010–2012 are further merged into a single entity. These groups include only savings banks that were perceived to have a relatively weak solvency position at the end of 2012.23 The combination of these groups into a single entity would improve their competitive position relative to rivals and it would facilitate the strengthening of their balances. Results are reported on Panel I of Table 5 under the Consolidated Structure II heading. The counterfactual variation in mortgage rates (in the range of 0.17–0.27%) is quite moderate, but it is not as small as under Consolidated Structure I. This also leads to higher reductions in loan volumes (approximately !51 million euros per month) and a more marked increase in profits with an average change close to 10:7%. This leads some support to the view that concentrating these banking groups with weak solvency positions can increase their value, but also with a higher cost for borrowers. 23 The groups considered are BFA-Bankia, CEISS, Catalunyabanc, NCG Banco, Banco Mare Nostrum and Liberbank, which ranked relatively low on the Oliver-Wyman solvency stress test of 2012.

As a benchmark for the two previous exercises, I also compute the counterfactual mortgage rates that would be charged by a perfectly colluding cartel, with results reported under the Cartel Structure heading of Panel I of Table 5. The average interest rate changes associated to the cartelization of the mortgage market are vastly superior than in the previous exercises, with a range of 4.37– 7.81%. This sizeable upward movement of mortgage rates leads to a vast reduction of the volume of mortgage loans granted by commercial and savings banks, with an average decrease of !4096 million euros per month. The relative profit changes per month are high with an average of approximately 250%, which implies that banks could more than triple their profits in the mortgage market if they colluded perfectly. This hypothetical scenario yields markedly different results than the consolidated market structure that has resulted from historical experience. 6.2.2. Analysis of credit quality I also use counterfactual computations to get an approximation to the effect of a concentrated bank structure on credit risk. Credit default does not occur immediately after granting a loan, but some time is required for the quality of a loan to be revealed. The default frequency observed in the crisis period starting in 2008 is mostly driven by loans granted in the preceding period of expansion. I build then the following indicator of credit performance for the mortgage market at a given period:

PDt ¼

PJ

P

t 0 2B qjt j¼1 PDjt & ð PJ P 0 t0 2B qjt Þ j¼1 ð

0

Þ

ð16Þ

where B is the set of months falling into the period of credit boom (months from 2004 to May 2007 in the empirical sample) and PDjt is the mortgage default frequency assigned to individual bank j in month t.24 For PDjt , I use two alternative proxies: (i) the actual data 24 Default rate data is available with a quarterly frequency so this measure is computed for months at the end of a quarter.

134

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

f jt ) and (ii) the of the individual bank default frequency (PDjt ¼ PD data of the bank default frequency for the leading bank in the group f 0 for j 2 g and bank j0 leading g to which bank j belongs (PDjt ¼ PD jt bank group g). Bank concentration can affect this measure of credit performance through both the loan volumes granted during the boom period and the change in default frequencies in the bust period. The concentrated bank group can alter not only the total volume of loans granted by its members, but also the shares of individual banks on this total volume. If each bank of group g retains separate risk management practices, it is reasonable that individual banks inside the group still face default rates observed in the data. On f 0 for the leading the other hand, it might be better to use PDjt ¼ PD jt 0 0 bank j in group g if this bank j imposes its own practices for loan screening and monitoring to the rest of group members. The use of the default frequency of the leading group as reference for all its members is applied only to Consolidated Structure I, as this group structure only applies to this experiment, e.g., it would not be very informative to assign the system average default rate to the cartel. The impact on credit performance of the change from fragmented to concentrated market structure can thus be measured with the difference:

DPDt ¼

PJ

P

C t 0 2B qjt 0 Þ j¼1 PDjt & ð PJ P C t 0 2B qjt 0 Þ j¼1 ð

!

PJ

f

P

t 0 2B q jt j¼1 PD jt & ð PJ P e 0 t 0 2B q jt Þ j¼1 ð

e 0Þ

ð17Þ

where all terms have already been defined. Panel II of Table 5 shows these comparisons for the different counterfactual market structures. The months depicted correspond to the end of the semester of the crisis years 2008, 2009 and 2010. The variation of the average default rates when individual banks are still assigned their default rates in the data f jt ) are quite small for all three scenarios. The range of (PDjt ¼ PD variation goes from !0:16% on December 2008 for Consolidated Structure II to 0:05% for the Cartel Structure on June 2009. These are small changes and indicate that the redistribution of loan volumes across the individual banks that results from sectoral consolidation is not dramatic. The increase in average default rates for Cartel Structure is explained by the redistribution of loans to banks with relatively high ex-ante return expectations which later prove to have a higher default frequency. The reduction of default rates is more significant if the merged groups that arose in 2010–2012 achieve the credit performance level of the individual banks that lead the mergers. The range of average default rate reductions is then ½!0:12%; !0:37%(, as shown under Consolidated Structure I in Panel II of Table 5. The change in the credit quality of the mortgage portfolio has thus a larger effect than the simple redistribution of mortgage loans across the banks of the consolidated groups. 6.3. Discussion of additional merger effects The simulation exercises in this section focus on the collusive effect of mergers on the pricing strategies of mortgage lenders. As noted in Peters (2006), the simulated price changes can then be taken as a measure of the reduction in competition, which is only one of the effects of the merger. The changes in perceived brand quality, costs and firm conduct can bring additional price and output effects. These limitations are also noted in Budzinski and Ruhmer (2010), who consider additional caveats such as the neglect of non-quantifiable and long run effects. For example, mergers might affect long run investment plans or innovative efficiency in the industry. The simulation exercises provide useful information about particular aspects of mergers, but the analytical conclusions must take into account their limits. The consolidated structure of the bank sector that resulted from the crisis of 2008 still leaves more than ten major banks in the

industry and these banks have differentiated brands inherited from historical institutions. The application of borrowers for mortgages through the extensive bank branch network and the use of contracts with interest rate as main characteristic, as described in Subsection 3.1, are also likely to remain as features of the consolidated sector. The model of Bertrand–Nash price competition with differentiated goods is then still a reasonable framework to analyze the post-merger bank sector, even though the available data and analytical methods do not allow to rule out completely changes in conduct. The analysis in Subsection 6.2.1 reveals that the benefits from a perfectly collusive agreement are substantial, and consolidation might facilitate that the banks bring prices closer to that solution. The analysis in Subsection 6.2.1 is not predictive, and it does not allow to determine ex-ante whether this change of conduct will apply. However, it offers a useful benchmark to detect collusion ex-post. The expected change in prices resulting from increased concentration is small under reasonable Bertrand–Nash competition, so higher ex-post increases in prices would be indicative of less intense competition than under this benchmark. A possible shift to Cournot competition is less applicable to the case at hand, as banks do not offer batches of capacity and the mortgages are not homogenous, making more difficult the empirical application of this model. The cost efficiencies that result from bank mergers are difficult to measure ex-ante and they might overcome the effects of reduced competition. As noted in Section 2, Xie (2007) and Erel (2011) find some evidence of the pass-through of the cost efficiencies of bank mergers to the rates offered to borrowers in the U.S., but they also find that this favorable price effect is temporary. The simulation methods in this section cannot be used to predict the size of cost efficiencies, but they can indicate whether reduced competition can eliminate the effect on prices of potential efficiency gains. The small price effect in the Consolidated Structure I experiment, an average rate increase of 0:07%, points that reduced price competition under the Bertrand–Nash framework cannot be expected to eliminate even moderate efficiency gains of the consolidation process of the Spanish banking sector.25 Significant price rises following the wave of mergers should be carefully examined by regulators, as they are indicative of either failure to increase cost efficiency or a competitive conduct that is more collusive than Bertrand–Nash.

7. Conclusion I find that the consolidation of the Spanish banking sector resulting from the financial crisis of 2008 has small impact on competition in the mortgage market. During the period 2004–2010, counterfactual equilibrium mortgage rates and loan volumes corresponding to the consolidated market structure that resulted from the crisis of 2008 are close to their actual realization and far from collusive levels. Concerns about the anti-competitive effects of bank consolidation are then not very relevant for the Spanish mortgage market. However, this also implies that consolidated banks do not have lower incentives to expand credit than the individual commercial and savings banks that incurred in lending excesses during the period 2004–2010. The change of the market conduct of Spanish banks in credit markets would depend on other policy measures, such as revision of risk management systems and supervisory rules. 25 In complementary work, I have simulated the rate changes Dr jt for different levels of efficiency gains (!5%, !10% and !15% reduction in expected costs) for both the Consolidated Structure I and the unconsolidated sector. The average effect of consolidation is to moderate by 0:07% the rate decrease associated to cost reduction, as anticipated from the price experiment in Subsection 6.2.1.

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

The competition analysis in the article is based on structural demand and cost estimates, which reveal that the demand for mortgage loans is relatively elastic and that the expected cost of mortgages for banks is highly correlated with interbank rates. The conclusions in the analysis are based on this particular structural model, and possible deviations from it after the consolidation process is complete must be taken into account in policy evaluation. If the consolidation process generates efficiency gains that bring costs below the structural estimates, mortgage rates are expected to decrease given the small effect of the mergers on competition. Deviation of banks to a behavior that is more collusive than the Bertrand–Nash model in the main analysis would lead to higher price rises, signaling the need for regulatory intervention. Appendix A. Contraction algorithm The solution of (9) is found as a fixed point to the following iterative algorithm:

ditþ1 t

¼

ditt

; ditt ÞÞ

! lnðst ða

þ lnðes t Þ

ð18Þ

where the superindex it indexes the number of iterations. In the sample, a given month is treated as a separate temporal market and (18) is then applied to each month to solve for the corresponding dt . I adapt the mixed-logit code of Nevo (2000) for this purpose. The integral in (3) is analytically intractable and I resort to simulation to obtain a computational analog for each sjt in st ða; ditt Þ in (18), i.e.,

sjt ða; dt Þ ’

NS X

ns¼1

Prjt

a

ns ð ns ; dt ; &Þ=NS

ð19Þ

h P where Prjt ns ðans ; dt ; &Þ ¼ expðdjt ! rðmÞjt & ans Þ= 1 þ Jq¼1 expðdqt !rðmÞqt & ans Þ(; ans ¼ GðIns Þ and ns is a simulation draw out of total simulations NS. The individual income Ins is drawn from the empirical income distribution data in the SILC. As the income data in the SILC are available annually, income draws for all the months in a year are taken from the SILC income distribution of that year. For the mixed-logit specification ans ¼ a=Ins , whereas ans ¼ a 2 R1 for the standard multinomial logit specification. As known since McFadden (1974), the mean value b d t in the standard logit specification can be recovered in a single step as b d t ¼ lnðes t Þ ! lnðes 0 Þ, where es 0 is a ðJ * 1Þ vector with each entry equal to the share of the outside good in t. Appendix B. Computation of counterfactual equilibria NE A Nash equilibrium in mortgage rates r NE t + frðmÞ1t ; . . . ; NE rðmÞNE ; . . . ; rðmÞ g for a period t is computed as a solution to the jt Jt ðJ * 1Þ system of individual first order conditions:

"

# @st ðr t ; &Þ & ðr t þ Rt =d) Þ ¼ 0 st ðr t ; &Þ þ Xt , @r t

ð20Þ

where r t + ½rðmÞ1t ; . . . ; rðmÞjt ; . . . ; rðmÞJt (; st + ½s1t ; . . . ; sJt ( and Rt =d) + ½R1t =d) ; . . . ; RJt =d) ( are ðJ * 1Þ vectors collecting mortgage rates, market share functions and return corrections of individual banks, @st ðr t ; :Þ=@r t is a ðJ * JÞ matrix with the derivatives of market shares with respect to mortgage rates, Xt is a ðJ * JÞ matrix of indicators for industry structure and , denotes element-by-element matrix multiplication. Eq. (20) is non-linear and I do not force shares st or derivatives @st =@r t to stay constant, avoiding the approximation error considered in Capps et al. (2003, p.11). 0 An element in row j and column j of @st ðrt ; &Þ=@r t contains the derivative of market share sj0 t ðr t ; Þ with respect to mortgage rate rðmÞjt . The matrix @st ðr t ; &Þ=@rt can then be written explicitly as:

135

3

2

@s1t =@rðmÞ1t .. . @sj0 t =@rðmÞ1t . .. @sJt =@rðmÞ1t 7 6 . .. . .. .. . 7 6 7 6 7 0 =@rðmÞ @s =@rðmÞ .. . @s . .. @s =@rðmÞ @st ðr t ;:Þ=@r t ¼ 6 1t Jt jt jt jt jt 7 6 7 6 . .. . .. .. . 5 4 @s1t =@rðmÞJt .. . @sj0 t =@rðmÞJt . .. @sJt =@rðmÞJt

ð21Þ

The form of own-rate ð@sj0 t =@rðmÞj0 t Þ and cross-rate 0 ð@sj0 t =@rðmÞjt ; j – j Þ derivatives follows from the mixed-logit specification:

@sj0 t =@rðmÞj0 t ¼ @sj0 t =@rðmÞjt ¼

NS X

ns¼1 NS X

ns¼1

ans & Prj0 t ns ðans ; dt ; :Þ & ð1 ! Prj0 t ns ðans ; dt ;:ÞÞ=NS ð22Þ

! ans & Prjt ns ðans ;dt ; :Þ & Prj0 t ns ðans ; dt ;:Þ=NS

ð23Þ

where Prjtns and ans are defined as in Eq. (19) in Appendix A. 0 An element in row j and column j of Xt contains an indicator 1jj0 t for whether the banks j and j0 are part of the same banking group at a period t. In Section 6, the experiment Consolidated Structure I considers a constant Xt + X1 with indicators identifying the 14 groups defined in Table 2. The experiment Consolidated Structure II considers a constant Xt + X2 , which differs from X1 in that weak savings banks are merged into a single group. The experiment Cartel Structure sets Xt ¼ 1JxJ , where 1JxJ is a matrix of ones. The accuracy of the solution for the Nash-equilibrium in mortgage rates from (20) can be verified with two alternative methods for each bank group. As indicated in Section 6, frðmÞNE jt gj2g maximizes the expected profit of the group, given the set of mortgage rates set by the rivals frðmÞNE jt gjRg . A necessary condition for optimality of frðmÞNE jt gj2g as a solution to the profit maximization problem is that the Hessian Hgt of second order conditions is negative-semidefinite. I verify that Hgt is negative-semidefinite by checking that all of its eigenvalues are non-positive.26 The Hessian Hgt is a ðJg * J g Þ matrix, where J g is the number of banks in group g, with the entry in row k 0 and column k corresponding to the cross derivative: 0

Hgt ðk; k Þ ¼

X @E½pjt ðrðmÞjt ; :Þjigt (=@rðmÞkt @rðmÞk0 t j2g

ð24Þ

Given the rival rates frðmÞNE jt gjRg in the solution to (20), it is also possible to compute the solution to problem (12) by a direct search method with no use of the derivatives of the objective function. If the solution of this approach coincides with the solution frðmÞNE jt gj2g from (20), we verify that there is no profitable deviation for group g, as required in a Nash equilibrium. References Ackerberg, D., Gowrisankaran, G., 2006. Quantifying equilibrium network externalities in the ACH banking industry. RAND Journal of Economics 37, 738–761. Aguirregabiria, V., Clark, R., Wang, H., 2012. Diversification of Geographic Risk in Retail Bank Networks: Evidence from Bank Expansion after the Riegle–Neal Act. University of Toronto. Working Paper 465. Akhavein, J.D., Berger, A.N., Humphrey, D.B., 1997. The effects of megamergers on efficiency and prices: evidence from a bank profit function. Review of Industrial Organization 12, 95–139. Allen, J., Clark, R., Houde, J.F., 2013. The Effect of Mergers in Search Market: Evidence from the Canadian Mortgage Industry. NBER Working Paper No. 19126. Amel, D., Barnes, C., Panetta, F., Salleo, C., 2004. Consolidatio and efficiency in the financial sector: a review of the international evidence. Journal of Banking and Finance 28, 2493–2519. Bank of Spain, 2010. Progress Report on the Restructuring of the Spanish Savings Banks. Bank of Spain Press Release. Bank of Spain, 2011. Note on the Savings Bank Restructuring Process. Bank of Spain Press Release. 26 See Section 1.7 of Sydsaeter et al. (2008) for results on the relation between eigenvalues and definiteness of quadratic forms.

136

C. Pérez Montes / Journal of Banking & Finance 43 (2014) 124–136

Bank of Spain, 2013. Statistical Bulletin of the Bank of Spain. Issue Number 07/ 2013. Basel Committee on Banking Supervision, 2006, Basel II: International Convergence of Capital Measurement and Capital Standards: A Revised Framework Comprehensive Version. Bank of International Settlements. Becalli, E., Frantz, P., 2009. M&A operations and performance in banking. Journal of Financial Services Research 36, 87–110. Berger, A.N., Hannan, T.H., 1998. The efficiency cost of market power in the banking industry: a test of quiet life and related hypotheses. Review of Economics and Statistics 80, 154–165. Berger, A.N., Humphrey, D.B., 1999. Efficiency of Financial Institutions: International Survey and Directions for Future Research. Board of Governors of the Federal Reserve System. Finance and Economics Discussion Paper Series, No. 11. Berger, A.N., Mester, L.J., 1997. Inside the black box: what explains the differences in the efficiencies of financial institutions? Journal of Banking and Finance 21, 895–947. Berger, A.N., Demsetz, R.S., Strahan, P.E., 1999. The consolidation of the financial services industry: causes, consequences, and implications for the future. Journal of Banking and Finance 23, 135–194. Berry, S., 1994. Estimating discrete-choice models of product differentiation. The RAND Journal of Economics 25, 242–262. Berry, S., Levinsohn, J., Pakes, A., 1995. Automobile prices in market equilibrium. Econometrica 63 (4), 841–890. Bikker, J., Spierdijk, L., 2008. How Banking Competition Changed Over Time. DNB Working Papers 167. Netherlands Central Bank. Research Department. Brambor, T., Clark, W.R., Golder, M., 2006. Understanding interaction models: improving empirical analyses. Political Analysis 14, 63–82. Braumoeller, B.F., 2004. Hypothesis testing and multiplicative interaction terms. International Organization 58, 807–820. Budzinski, O., Ruhmer, I., 2010. Merger simulation in competition policy: a survey. Journal of Competition Law & Economics 6, 277–319. Calomiris, C.W., 2000. U.S. Bank Deregulation in Historical Perspective. Cambridge University Press. Calomiris, C.W., Pornrojnangkool, T., 2005. Monopoly-Creating Bank Consolidation? The Merger of Fleet and BankBoston. NBER Working Paper No. 11351. Campa, J.M., Hernando, I., 2006. M&As performance in the European financial industry. Journal of Banking and Finance 30, 3367–3392. Capps, O., Church, J., Love, H.A., 2003. Specification issues and confidence intervals in unilateral price effects analysis. Journal of Econometrics 113, 3–31. Carow, K.A., Kane, E.J., Narayanan, R.P., 2006. How have borrowers fared in banking megamergers? Journal of Money, Credit and Banking 38, 821–836. Claessens, S., 2009. Competition in the financial sector: overview of competition policies. The World Bank Research Observer 24 (1), 83–118. Cohen, A.M., Mazzeo, M.J., 2007. Market structure and competition among retail depository institutions. Review of Economics and Statistics 89, 60–74. Craig, B.R., Dinger, V., 2009. Bank mergers and the dynamics of deposit interest rates. Journal of Financial Services Research 36, 111–133. Cuesta, R.A., Orea, L., 2002. Time-varing efficiency and stochastic distance functions: the effect of mergers on Spanish savings banks’ efficiency. Journal of Banking and Finance 26, 2231–2247. Degryse, H., Masschelein, N., Mitchell, J., 2011. Staying, dropping, or switching: the impacts of bank mergers on small firms. Review of Financial Studies 24, 1102– 1140. Demirguc-Kunt, A., Laeven, L., Levine, R., 2004. Regulations, market structure, institutions, and the cost of financial intermediation. Journal of Money, Credit and Banking 36, 593–622. DeYoung, R., Evanoff, D.D., Molyneux, P., 2009. Mergers and acquisitions of financial institutions: a review of the post-2000 literature. Journal of Financial Services Research 36, 87–110. Dick, A., 2008. Demand estimation and consumer welfare in the banking industry. Journal of Banking and Finance 32, 1661–1676. Drees, B., Pazarbasioglu, C., 1995. The Nordic Banking Crisis: Pitfalls in Financial Liberalization?. IMF Working Paper 95/61. Erel, I., 2011. The effect of bank mergers on loan prices: evidence from the United States. Review of Financial Studies 24, 1068–1101. European Union Commission, 2012. Proposal for a Directive of the European Parliament and the Council establishing a Framework for the Recovery and Resolution of Credit Institutions and Investment Firms. pp. 1–171. Farrell, J., Shapiro, C., 1990. Horizontal mergers: an equilibrium analysis. American Economic Review 80, 107–126. Focarelli, D., Panetta, F., 2003. Are mergers beneficial to consumers? Evidence from the market for bank deposits. American Economic Review 93, 1152–1172. Focarelli, D., Panetta, F., Salleo, C., 2002. Why do banks merge? Journal of Money, Credit, and Banking 34, 1047–1066. Gowrisankaran, G., Krainer, J., 2011. Entry and pricing in a differentiated products industry: evidence from the ATM market. RAND Journal of Economics 42, 1–22. Hagendorff, J., Keasey, K., 2009. Post-merger strategy and performance: evidence from the US and European banking industries. Accounting and Finance 49, 725–751. Ho, K., Ishii, J., 2011. Location and competition in retail banking. International Journal of Industrial Organization 29, 537–546. International Monetary Fund, 2012. Spain: Financial Stability Assessment. IMF Country Report No. 12/137. Ishii, J., 2008. Compatibility, Competition and Investment in Network Industries: ATM Networks in the Banking Industry. Stanford GSB Working Paper. Jimenez, G., Mencía, J., 2009. Modelling the distribution of credit losses with observable and latent factors. Journal of Empiriral Finance 16, 235–253.

Jimenez, J.L., Perdiguero, J., 2012. Mergers and Difference-in-Difference Estimator: Why Firms do not Increase Prices? Universitat de Barcelona. Research Institute of Applied Economics. Working Paper 2012/05. Jimenez, G., Saurina, J., 2006. Credit cycles, credit risk and prudential regulation. International Journal of Central Banking 2, 65–98. Jimenez, G., Salas, V., Saurina, J., 2006. Determinants of collateral. Journal of Financial Economics 81, 255–281. Kahn, C., Pennacchi, G., Sopranzetti, B., 2005. Bank consolidation and the dynamics of consumer loan interest. Journal of Business 78, 99–134. Kingdom of Spain, 2009. Royal Decree Law 9/2009 Regarding the Bank Restructuring and Strengthening of Capital of Credit Institutions. Spanish Official Gazette, No. 155. pp. 53194–53212. Kingdom of Spain, 2010a. Circular 1/2010 of the Bank of Spain regarding Statistical Information about the Interest Rates applied to Deposits and Loans of Households and Non-financial Firms. Spanish Official Gazette, No. 31. pp. 10555–10565. Kingdom of Spain, 2010b. Royal Decree Law 11/2010 Regarding the Government Bodies and other Aspects of the Juridical Regime of Savings Banks. Spanish Official Gazette, No. 169. pp. 61427–61457. Kingdom of Spain, 2011. Royal Decree Law 2/2011 Regarding the Strengthening of the Financial System. Spanish Official Gazette, No. 43. pp. 19213–19239. Kingdom of Spain, 2012a. Royal Decree Law 2/2012 Regarding the Rationalization of the Financial System. Spanish Official Gazette, No. 30. pp. 9889–9913. Kingdom of Spain, 2012b. Royal Decree Law 18/2012 Regarding the Rationalization and Sale of the Real Estate Assets of the Financial System, No. 114. pp. 35110– 35120. Kingdom of Spain, 2012c. Royal Decree Law 24/2012 Regarding the Restructuring and Liquidation of Credit Institutions. pp. 61559–61619. Maudos, J., Fernández de Guevara, J., 2008. El Sector Bancario Español en el Contexto Internacional: Evolución Reciente y Retos Futuros. BBVA Foundation Publications. McFadden, D., 1974. Conditional logit analysis of qualitative choice behavior. In: Zarembka, P. (Ed.), Frontiers in Econometrics. Academic Press, New York, pp. 105–142. Murillo, C., Giambona, E., Graham, J.R., Harvey, C.R., 2011. Liquidity management and corporate investment during a financial crisis. Review of Financial Studies 24, 1944–1979. Nevo, A., 2000. A practitioner’s guide to estimation of random-coefficients logit models of demand. Journal of Economic and Management Strategy 9 (4), 513– 548. Oliver Wyman, 2012. Asset Quality Review and Bottom-Up Stress Test Exercise. Oliver Wyman, Marsh and McLennan Companies. pp. 1–88. Ozer-Balli, H., Sorensen, B.E., 2013. Interaction effects in econometrics. Empirical Economics 45, 583–603. Park, K., Pennacchi, G., 2009. Harming depositors and helping borrowers: the disparate impact of bank consolidation. The Review of Financial Studies 22, 1– 40. Peters, C., 2006. Evaluating the performance of merger simulation: evidence from the U.S. airline industry. Journal of Law and Economics 49, 627–649. Prager, R.A., Hannan, T.H., 1998. Do significant horizontal mergers generate significant price effects? Evidence from the banking industry. Journal of Industrial Economics 46, 433–452. Santos, J.A.C., 2011. Bank corporate loan pricing following the subprime crisis. Review of Financial Studies 24, 1916–1943. Sapienza, P., 2002. The effects of banking mergers on loan contracts. Journal of Finance 57, 329–367. Scharfstein, D.S., Sunderam, A., 2013. Concentration in Mortgage Lending, Refinancing Activity and Mortgage Rates. NBER Working Paper No. 19156. Spiegel, M., 2011. The academic analysis of the 2008 financial crisis: round 1. The Review of Financial Studies 24, 1773–1781. Sydsaeter, K., Hammond, P., Seierstad, A., Strom, A., 2008. Further Mathematics for Economic Analysis. Prentice Hall. Train, K.E., 2009. Discrete Choice Methods with Simulation. Cambridge University Press. Train, K.E., Winston, C., 2007. Vehicle choice behavior and the declining market share of U.S. automakers. International Economic Review 48, 1469–1496. Vander Vennet, R., 1996. The effect of M&As on the efficiency and profitability of EC credit institutions. Journal of Banking and Finance 20, 1531–1558. Vander Vennet, R., 2002. Cost and profit efficiency of financial conglomerates and universal banks in Europe. Journal of Money, Credit, and Banking 34, 254–282. Vives, X., 2012. The Spanish financial industry at the start of the 21st century: current situation and future challenges. In: Malo de Molina, J.L., Martín-Aceña, P. (Eds.), The Spanish Financial System. Growth and Development Since 1900. Palgrave Macmillan, pp. 383–423. Wheelock, D.C., 2011. Banking Industry Consolidation and Market Structure: Impact of the Financial Crisis and Recession. Federal Reserve Bank of St. Louis Review 93 (6), 419–438. Wilks, S., 2009. The impact of the recession on competition policy: amending the economic constitution? International Journal of the Economics of Business 16 (3), 269–288. Xie, L., 2007. Bank Consolidation and Loan Pricing. Ball State University. Department of Economics. Working Paper No. 200706. Zhou, X., 2008. Estimation of the Impact of Mergers in the Banking Industry with Historical Data on Mergers. Shangai University of Finance and Economics Working Paper. .