Does bank ownership affect lending behavior? Evidence from the Euro area. September 13, 2013

Transcription

1 Does bank ownership affect lending behavior? Evidence from the Euro area Giovanni Ferri *, Panu Kalmi **, Eeva Kerola *** September 13, 2013 Abstract We analyze the differences in bank lending policies across banks of different ownership forms using micro level data on Euro area banks over to detect possible different patterns in bank lending supply responses to changes in monetary policy. Our results identify a prevailing difference between stakeholder and shareholder banks: following a monetary policy contraction stakeholder banks decrease their loan supply to a lesser extent than shareholder banks. Distinguishing the effect within stakeholder banks reveals that cooperative banks kept smoothing the impact of monetary contraction onto their lending even during the crisis period ( ) whereas savings banks did not. The propensity of stakeholder banks to smoothen their lending cyclicality suggests that their presence in the economy can dampen credit supply volatility. JEL classification: G21; E52; L33; P13 Keywords: European banks; Monetary policy transmission; commercial banks; savings banks; cooperative banks; lending cyclicality * LUMSA University Rome ** University of Vaasa ***Aalto University 1

2 1. Introduction The lending channel literature has long held that the impact of monetary (and financial) shocks is exacerbated because banks tend to curtail their loan supply after those shocks materialize (Bernanke and Gertler, 1995; Hubbard, 1995). In turn, the pro-cyclical attitudes of bank lending (Rajan, 1994) could exert a disproportionate strain on the economy, making it harder for bank dependent borrowers e.g. the small businesses to keep relying on external finance (Gertler and Gilchrist, 1994; Berger and Udell, 1995). The aim of this study is to investigate lending supply policies of banks from Euro countries during the last 13 years of common monetary policy, and explore whether differences in loan supply decisions arise from different forms of bank ownership. In order to study banks responses to monetary policy changes between 1999 and 2011 we utilize bank-specific financial statements (BankScope database) provided by Bureau Van Dijk. As well as looking at the whole time period, we make a distinction between the time of relative financial stability (prior to the recent crisis), and the crisis period ( ) and provide evidence that there is a clear difference between these two time periods with respect to the bank lending channel. Apart from gaining knowledge of the bank lending channel and factors influencing banks lending behaviour, this appears interesting from the industrial organization and microeconomic point of view per se; it is also extremely important to reveal underlying reasons for heterogeneity across the banking sectors within the Euro area to learn how, due to differences in the monetary transmission channel, the actual monetary stance can differ across Euro countries despite the common monetary policy instruments. Previous empirical research has already confirmed (at least on US data) that banks that are small, and moreover banks that are undercapitalized and relatively illiquid, amplify monetary policy shocks more through the lending channel (see e.g. Kashyap and Stein (1995) and Kishan and Opiela (2000)). With studies on Euro area, consensus has been harder to find: to what extent does different bank balance sheet items amplify the lending channel seems to be somewhat country-dependent (see e.g. Altunbas et al. (2002), Favero et al. (1999) and De Bondt (1999)). We argue that the reason is the heterogeneity in national banking sector compositions inside Euro area, and that the differences in bank lending would not arise entirely from differences in balance sheets but also from differences in bank business models which are closely related to bank ownership forms. For one, it is possible that banks relying on a relationship lending business approach could be less willing to curtail loans to their customers with whom they tend to liaise in long-term rapport (Berger and Udell, 2002). Our paper is one of the first attempts to distinguish differences between lending supply policies depending on banks mission/ownership form 1. Specifically, moving from general to specific, we consider two breakdowns of the banks: i) a mission-based breakdown of shareholder (profit maximizing) banks vs. stakeholder banks (catering not only for their shareholders); ii) an ownership-based categorization of the stakeholder banks differentiating cooperative banks from savings banks. We find that stakeholder banks follow less procyclical loan supply policies during the whole observation period ( ) than shareholder banks; their loan supply changes reacted less to changes in short term interest rate. This finding is similar for both cooperative and savings banks. With respect to bank specific variables we find that size and capitalization seem to be highly important (the larger or the better capitalized the bank, less its loan 1 The only previous study that we are aware of is De Santis and Surico (2013), who look at the heterogeneity between lending channels of cooperative, commercial, and savings banks (among other things). 2

3 supply decisions are affected by changes in the interest rate), while a higher share of liquid assets seems to have an opposite effect (rather unconventionally). The interest rate has a stronger overall effect on loan growth changes during a time span of relative financial stability ( ); whereas the lending channel becomes weaker in absolute terms during the recent crisis ( ). Irrespective of the time period analysed, these effects are further dampened for stakeholder banks as a group, and especially for cooperative banks. Savings bank as an explanatory variable loses its statistical significance during the recent crisis and these banks behaviour seems to become statistically no different from that of their shareholder counterparts. In all, stakeholder banks (and cooperative banks in particular) thus seem to behave less procyclically and stabilize lending cyclicality on their part by smoothing out financial conditions faced by their customers. This result survives a number of robustness checks. The rest of this paper is structured as follows: in section 2 we present the discussion on the lending channel as well as we survey previous empirical literature on bank heterogeneity affecting loan supply policies. Section 3 lays out testable hypotheses regarding different mission and ownership groups. Section 4 presents the data used in the estimations and some descriptive statistics. In section 5 we perform our empirical estimations and comment on the results obtained. Section 6 concludes. 2. Effect of the lending channel and previous empirical evidence There has been an increased interest during the past few decades on financial sector s (especially banks ) role in the monetary policy transmission process. In his seminal paper Bernanke (1983) analysed the relative importance of monetary versus financial factors during the Great Depression and his study gave support to the credit view, which argued that financial markets were imperfect, so that the Modigliani-Miller assumptions did not hold and finance did actually matter (Freixas and Rochet, 2008). Empirical research has then induced a debate between the so called money view and a set of alternative theories referred to as the broad lending channel. The broad lending channel emphasizes the role of the supply of bank funds to firms and takes into account asymmetries of information and market imperfections. Implicit assumptions of the lending channel are that prices are rigid, the central bank can influence directly the volume of credit by adjusting reserves, and loans and securities are imperfect substitutes both for borrowers and for banks. Moreover, it is usually the (ex-ante) riskier and smaller firms that cannot obtain market finance as easily as credit from the banking sector, and are thus mostly affected by the lending channel mechanism. External finance is more expensive than internal finance, unless the external finance is fully collateralized. The higher cost of external finance reflects the agency cost of lending, which is the inevitable deadweight loss arising from asymmetric information (Bernanke et al., 1994). Mostly smaller firms and firms with lower net worth are hurt by an economic downturn. Research on non-financial firms that face capital market imperfections has pointed out that shocks to internal liquidity should have larger impact on the investment behaviour of smaller companies who are more likely to have harder time accessing external sources of finance (Kashyap and Stein, 1995). For financial institutions, the situation is no different. Smaller banks find it harder to gather non-deposit funding in times of distress. Thus bad times in financial markets and the real economy are more likely to affect small, undercapitalized banks. In contrast, Romer and Romer (1990) argued that banks confronted with a decrease in their deposits would be able to simply substitute this decrease with other types of liabilities, such as certificates of deposits (CDs), which are not subject to reserves. However, banks liabilities are in fact not perfect substitutes, so a decrease in deposits cannot be matched with CDs or issuing new equity. They differ e.g. by riskiness and maturity (see Stein,

4 and Van den Heuvel, 2002). As Kashyap and Stein (1995) state, investors purchasing CDs must concern themselves with the quality of the issuing bank. With any degree of asymmetric information, standard sorts of adverse selection will arise and tend to make the marginal cost of external financing an increasing function of the amount raised. Thus a decrease in deposits leads banks to reduce their supply of credit for households and firms. Hence a reduction in the supply of bank credit, relative to other forms of credit, is likely to increase the external finance premium of the private sector and to reduce real activity (Bernanke and Gertler, 1995). The broad lending channel usually neglects banks equity, and treats bank capital as an irrelevant balance sheet item (Van den Heuvel, 2002). Especially after the strong debate over the Basel II accord that is claimed to emphasize the pro-cyclical effects of monetary policy with its capital requirements banks capital should be an aspect of great interest in the monetary policy transmission. The bank capital channel is based on three hypotheses: i) an imperfect market for bank equity; ii) a maturity-mismatch between banks assets and liabilities (usually long-term loans vs. short-term deposits); and iii) a direct influence of regulatory capital requirements on the supply of credit. The bank capital channel works in the following way: as market interest rates increase, an even lower fraction of loans can be renegotiated with respect to deposits (because of the maturity-mismatch), and thus banks face a cost due to the maturity transformation that reduces profits and then capital. If equity is sufficiently low (and banks cannot easily issue new shares) banks reduce their supply of lending, because of the bank capital ratios required by regulators. While a large number of studies have already found that lending channel is in place and is working to amplify monetary policy shocks through the banking sector, a small number of studies have also tried to figure out whether there are differences between different kinds of banks and between banking sectors across countries. So far, empirical research has been trying to identify differences in the lending channel and the impact of monetary policy depending on banks size, capitalization, and/or liquidity. Kashyap and Stein (1995) find that monetary policy shocks affect differently large and small banks. Small banks presumably face higher agency costs of raising uninsured funds, and thus their balance sheets are more affected (Bernanke et al., 1994). Kashyap and Stein (1995) also find that the impact of monetary policy on credit supply is more pronounced for banks with less liquid balance sheets (banks with lower ratios of cash and securities to total assets). Kishan and Opiela (2000) differentiate banks by their size and capital leverage ratio and conclude that capital is important in assessing the impact of policy on loan growth and in determining the distributional effects of monetary policy. Low-capitalized banks are perceived as riskier by the market and have thus greater difficulty issuing bonds, therefore being unable to shield their credit relationships. All these studies focusing on the United States stress the fact that the lending channel is more important for small banks, and especially for those that are undercapitalized or relatively illiquid. However, results are far less unanimous when looking at the banking sector in Europe. Financial sectors differ quite a bit between Europe and the US, and especially the fact that firms rely much more heavily on bank credit in Europe than they do in the US would certainly lead us to expect some differences in the estimation results. The entire financial system is much more bank-based in Europe than in the United States, where financial market financing of the corporate sector is more developed. For example, according to EBF(2012), the total assets of the banking sector in Europe account for 350% of aggregate GDP, whereas the same figure for the United States is 77%. Also, the share of total bank loans to GDP is 139% in Europe compared to 59% in the United States. Among others, Altunbas et al. (2002) use the BankScope database and classify banks according to asset size and capital strength. They find that undercapitalized banks (of any size) tend to respond more to changes in policy 4

5 through the lending channel, and that this is more prevalent in the smaller EMU countries. Favero et al. (1999) use individual bank balance sheet data (from BankScope database) in selected European countries (Germany, France, Italy, and Spain). Studying only a monetary tightening period in year 1992 they find that there are differences in banks responses across countries. Small banks in Germany, Italy and Spain (although to a lesser extent) maintain or increase their loan supply by raising new deposits, while banks in France use their excess capital to maintain existing lending levels. De Bondt (1999) finds strong support for an existing lending channel especially for Germany, Netherlands, and Belgium for and that the loan supply effects are stronger for small and illiquid banks. King (2000) confirms the importance of bank size and liquidity, but he finds them to be most effective in France and Italy. Ehrmann et al. (2001) study bank lending in euro area countries and find that monetary policy shocks do alter banks credit supply, the effects being most pronounced with illiquid banks, while the size of the bank or its capitalization do not seem to matter. Gambacorta and Mistrulli (2004) study the existence of cross-sectional differences in the response of monetary policy and business cycles owing to a different degree of bank capitalization on Italian banks ( ). They find that well-capitalized banks can better smooth their lending from monetary policy shocks as they have easier access to non-deposit fund-raising, and thus they can view other types of liabilities more of a substitute for deposits. They also conclude that non-cooperative banks behave more pro-cyclically when supplying credit, due to their stronger dependency on non-deposit forms of external funds and their lower proportion of long-term lending relationships. Gambacorta (2005) studies Italian banks, and finds evidence that heterogeneity in the monetary policy transmission exists. Lending is smoother for well-capitalized banks that are seen as less risky by the market and are better able to raise uninsured deposits. Liquid banks can further protect themselves from monetary policy tightening by simply drawing down cash and securities. He further concludes that size does not matter for lending supply policies. Fungacova et al. (2013) look at the interaction of competition and lending channel in 12 euro countries between , and find that before 2007, lending channel was enhanced in competitive markets. Apart from a few papers, to our knowledge, there are no empirical studies focusing on the possible heterogeneous effects on the strength of the lending channel across banks of different ownership groups. Ashcraft (2006) looked at affiliation across banks in the US, and found that banks affiliated with a multibank holding company react less sensitively to monetary policy contractions because they have access to larger internal capital markets. De Bondt (1999) and Schmitz (2004) included foreign ownership as one explanatory variable, former dealing with US data and latter with 10 EU accession countries in Schmitz (2004) found that foreign owned banks reacted more to euro-area interest rate changes than their domestic owned counterparts. De Bondt (1999) found stronger evidence for a lending channel when foreign owned banks were omitted from the sample; concluding that international banks have better opportunities to borrow elsewhere than even large domestic banks. Bertay et al. (2012) looked at state owned banks in 111 countries during and found that lending by state banks is less procyclical than the lending of private banks; furthermore lending by state banks located in high-income countries is even countercyclical. Given the central role of stakeholder banks in many European banking markets (e.g. Ayadi et al., 2010), it would be important to know whether the monetary policy transmission differs across ownership structure. To our knowledge, there is only one paper looking at this, and this is the recent work by De Santis and Surico (2013). They look at banking sectors in Spain, Germany, Italy, and France during and conclude that lending channel is strongly affected by heterogeneity with respect to market concentration, bank balance sheet characteristics, and bank typology (commercial, cooperative, and savings banks). They run separate regressions 5

6 for each country and for each typology of banks and find inter alia that the interest rate channel in Spain is rather non-existing, that commercial banks react to interest rate changes only remotely irrespective of the country, and that loan supply decisions are most affected by monetary policy actions especially among relatively illiquid and less capitalized cooperative and savings banks in Germany as well as smaller savings banks in Italy. Our empirical strategy is different: we estimate the whole panel at once and allow for heterogeneous responses to monetary policy shocks between banks of different ownership types, controlling for differences in banks balance sheets and demand conditions across countries 2. This enables us to draw more directly inferences about the relative differences between stakeholder and shareholder banks loan supply policies than the approach followed by De Santis and Surico (2013). 3. Implications of differences in mission and ownership We argue that in order to study differences between banks lending policies, one should pay attention to bank ownership form and structure. Next we make some assumptions on how these differences might be affecting bank lending both during time of relative financial stability and during time of crisis Differences during financial stability (traditional monetary policy) We first concentrate on times of financial stability, when interbank markets are functioning normally and there are no economic or financial crises. Then, a monetary policy contraction (an increase of the short term money market interest rates) will decrease liabilities available for banks. This drop in liquidity will force banks to make adaptations on their asset side in order for their balance sheets to be in equilibrium. Stakeholder banks are more involved in relationship lending (see e.g. Amess, 2000) and thus hold longer term objective functions than shareholder banks, and could be more prone to smooth out financial constraints for their borrowers in order to maximize the long term values of their borrower-lender relationships (Boot, 2000; Petersen and Rajan, 1994; Gambacorta and Mistrulli, 2004). Stakeholder banks could thus be more willing to sacrifice other assets so to keep their lending volume rather intact. On the contrary, shareholder banks, while focusing on maximizing profits could then more easily cut back lending if that would result in lower short term costs (following the theory of bank capital channel). Following results on affiliated banks in US (Ashcraft, 2006) we could hypothesize that since there exists different kind of network formations especially inside the cooperative banking group (Desrochers and Fischer, 2005), they could be less hit by the liquidity shock since they could access the internal capital markets of their banking group and thus weaken the effect of their individual balance sheet constraints. Also on one hand, savings banks are on average less liquid and lack more capital (their equity to total assets ratio and the share of liquid assets are on average lower than for cooperative banks) and could be thus forced to cut back their assets more during a monetary policy contraction. On the other hand however, savings banks are in government ownership especially in Germany and Austria (municipal and/or regional) and could be thus more prone to smooth their lending supply (Bertay et al., 2012). 2 A further difference between De Santis and Surico (2013) and our paper stems from re-classification of several banks. It seems that their database was built taking the ownership classifications of the banks as provided by BankScope. On the contrary, we found that ownership was misclassified for some of the banks in BankScope and recoded those banks accordingly; see footnote 4 below. Beside the other outlined differences, this factor could also help explain possible divergences between our results and the ones in De Santis and Surico (2013). 6

7 Hypothesis 1: During financial stability, a monetary policy contraction is more effective on the loan supply of shareholder banks (rather than stakeholder banks), but the relative difference between savings and cooperative banks remains rather vague Differences during financial instability (unconventional monetary policy) Straight after 2008, in the beginning of the recent crisis interbank markets literally collapsed, and liquidity was extremely scarce. Monetary policies were eased by central banks all over the world, interest rate ultimately hitting its zero lower bound especially in Europe and in the United States, leaving conventional monetary policy tools powerless. Therefore, unconventional measures were taken. These included credit easing, quantitative easing, and signaling. Throughout the crisis, the European Central Bank has been making Long Term Refinancing Operations (LTROs), where it has supplied European banks with cheap loans up to 3-years of maturity, taking government securities, mortgage securities and other secure commercial papers as collateral. Although shareholder banks could have been initially more at risk because of their less retail oriented business models, it could be ultimately so that the liquidity shortage hit stakeholder banks harder because of their problems of issuing new equity promptly (practically nonexistent); especially if they are low capitalized. However, irrespective of the state of the economy, the longer term objective of stakeholder banks could lead to less tightening of credit supply, especially to small and medium sized enterprises that form the bulk of stakeholder banks non-financial firm borrowers. By having on average relatively more secure assets on their balance sheets, stakeholder banks could have better access to extra liquidity provided by the ECB during the recent crisis. Cooperative banks could be thought to be in a more favorable position than savings banks, because of their ownership characteristics. Cooperative banks are owned by their members (who are also usually their depositors). Although rights of members to profits are (typically) much more limited than they are at shareholder banks, cooperative banks may distribute part of their surplus to their members directly or implicitly by charging lower service fees or giving members more favorable interest rates. Among savings banks, profit distribution is absent altogether. Common to all savings banks is that they are formally non-profit institutions and owned usually by some private entity or government. These differences in ownership structure give rise to the proposition that cooperative banks would be able to better commit their customers especially during times of financial turmoil and thus keep their insured deposits (making the bulk of their funding), being less affected by the illiquidity of short-term credit markets. Also publicly owned savings banks are heavily dependent on the health of their domestic economy and in the worst case scenario could be burdened by highly leveraged owners (governments) in crisis countries. Hypothesis 2: During times of financial distress, stakeholder banks are less inclined to decrease their loan supply (than are shareholder banks), and cooperative banks exhibit this feature with particular intensity (more so than savings banks). 7

8 4. Data and descriptive statistics For this paper we use microlevel data based on financial statements derived from BankScope, provided by Bureau van Dijk. These data (at unconsolidated level) include annual observations from 12 Euro area countries 3 over the period of covering 4,352 individual banks. As stated by Brissimis and Delis (2010), two recent papers (Ashcraft, 2006 and Gambacorta, 2005) provide discussion and evidence that annual observations are robust to be used in lending equations; thus validating their (and our) use of the BankScope database. The initial ownership classifications are drawn from BankScope where we have done certain corrections and amendments based on our earlier work (Ferri et al. (2012)) 4. Table 1 shows how banks of different ownership have been distributed by countries and the value of their total assets in our data (annual average taken over observation period ). Bulk of our stakeholder bank observations come from Germany (2246 German stakeholder banks out of 3491 in total). Italian stakeholder bank observations are also abundant but to a far lesser extent (703 Italian stakeholder banks). Shareholder bank observations are more dispersed between different countries (Germany and France having the most observations). Spanish savings bank sector is large measured by total assets; 55 Spanish savings banks combined annual average amount of total assets is more than fourfold to that of 65 Italian savings banks and more than tenfold to that of 77 Austrian savings banks equivalents, respectively. Table 2 gives summary statistics (broken banks into stakeholder vs. stakeholder, then to cooperative and savings banks) of the most relevant bank-specific variables. Loan growth during the observation period ( ) has been fastest on average for shareholder banks, around 9.75% on yearly basis, but also the standard deviation is much larger than with stakeholder banks; implicating higher volatility. Savings banks loan growth was on average almost 3 percentage points lower than for their cooperative counterparts. Shareholder banks are bigger (in terms of log of total assets) than stakeholder banks on average, and savings banks are larger than cooperative banks. Looking at the balance sheet composition and the share of loans on banks total assets in particular, we see distinctive differences between stakeholder and shareholder banks: loans make up to 60% of stakeholder banks total assets, while the number for shareholder banks is only 45% on average. Looking at the capitalization (share of equity to total assets), shareholder banks share is 14% on average, while the ratio is as low as 6% amongst savings banks. Finally looking at banks liquidity measure (share of liquid assets 5 to total assets), shareholder banks are on average far more liquid than stakeholder banks (60% and 33% for shareholder and stakeholder banks, respectively). Next we concentrate more on the loan supply of banks of different ownership groups and on how it has evolved during the last 13 years. Figure 1 presents the observations of loan growth of stakeholder vis-à-vis shareholder banks and for cooperative and savings banks separately during As we can see, although we have 3 Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, and Spain. 4 For example we have corrected for the following flaws in BankScope initial classifications: 1) some Austrian co-operative banks under the Raiffeisen group were classified as savings banks in BankScope. 2) especially in Italy, many banks classified as savings banks in BankScope are essentially retail commercial banks. 3) the Caisse d Epargne Group in France is still classified as savings bank in BankScope albeit it converted into co-operative ownership in ) some banks that are relevant to our analysis could also be found in other specialization classifications (than solely from commercial banks, cooperative banks, and savings banks) such as bank holding companies, governmental credit institutions, and mortgage banks. For more details, see Ferri et al. (2012). 5 Our measure of Liquid assets is taken directly from BankScope and it includes: Trading securities at fair value through income (plus) loans and advances to banks (plus) reverse repos and cash collateral (plus) cash and due from banks (minus) mandatory reserves included above. In most previous studies, liquid assets are stated to include cash, securities (often only government bonds), and interbank lending; additional items included in our measure do not distort results, but produce a rather similar liquidity ratio compared to e.g in Gambacorta (2005), and around 0.4 in Ehrmann et al. (2003). 8

9 more observations in total for stakeholder banks than for shareholder banks; the shareholder bank loan growth observations are much more dispersed between -5% and +5%; while that of stakeholder banks rest more uniformly between -1% and +2%. For cooperative and savings banks differences are rather vague; although it seems that with cooperative banks loan growth has been higher on average throughout the observation period. In addition, we can get an idea on the degree of stability of lending policies across ownership/organizational bank classes by calculating the coefficient of variation (standard deviation/mean) of the absolute change of loan supply. The figures reported in Table 3 tell us that stakeholder banks are much more stable than shareholder ones, with a coefficient of variation of as against As to cooperative banks vs. savings banks, there is a slight difference with the former (2.0733) being a little bit more stable than the latter (2.3468). We now turn to our empirical estimations, where we study distinctions between loan supply policies of different ownership groups by looking at monetary policy effects on banks loans supplied using data from bank balance sheets. 5. Empirical estimations and results 5.1. Empirical method Kashyap and Stein (1995) build a theoretical model and empirically test with disaggregated bank balance sheet data whether a lending channel exists in US. They argue that if the lending view is correct, one should expect the loan portfolios of banks of different sizes to respond differently to a contraction in monetary policy. Using their model as a basis, and following more recent empirical studies (see e.g. Gambacorta (2005), Gambacorta and Mistrulli (2004), Bertay et al. (2012)) we write an autoregressive model, but deviating from existing literature we test whether interest rate changes affect differently banks of different ownership form. The bank-specific variables that are found in the existing literature to be affecting lending supply the most (namely capitalization (equity to asset ratio), liquidity (liquid assets to total assets ratio), and bank size (log of total assets)) are included as controls. We test how changes in short-term interest rate affect the overall loans supplied by banks. The estimated equation is: Where loans is the dependent variable; namely the volume of loans supplied by bank i at year t. r is the short term interest rate (EONIA overnight interest rate) reflecting changes in monetary policy. As would be according to the theory of the lending channel, the coefficient α should be negative; as interest rates increase banks decrease the amount of loans supplied. Now µ captures the separate effect of interest rates for stakeholder banks; if this coefficient is positive it means that the negative effect is dampened for the stakeholder banks. This interaction term only gets values either for stakeholder banks as a group or for cooperative and savings banks separately; depending on the specification (OS DUMMY stands for ownership dummy). rgdp is the annual real GDP in country c where bank i is operating and Ygap is the output gap in percent of potential GDP 6 in the same 6 Both real GDP and output gap taken from International Monetary Fund, World Economic Outlook Database, October

10 country c. Real GDP (value of economic output adjusted for price changes) controls for economic growth accounting also for changes in the price level, while the output gap (difference between the real GDP and the potential GDP) indicates the imbalance existing in the real economy. According to Gambacorta (2005), the inclusion of demand side control variables allows us to capture cyclical movements and enables to separate the monetary policy component of interest rate changes. We also control for bank-specific variables: CAPITAL is bank i s capital position (share of equity to total assets), SIZE is bank i s size (log of total assets), and LIQUIDITY is bank i s liquidity (share of liquid assets out of total assets). We are aware that there might be a number of time-invariant bank and/or country characteristics (fixed effects) that might be correlated with the explanatory variables. The fixed effects are contained in the error term in equation (1), which consists of the unobserved bank/country-specific effects and the observation-specific errors: To cope with this problem of fixed effects, and further because of the lagged dependent variable and heteroskedasticity 7 present in the data we estimate equation (1) with Arellano-Bond type difference GMM estimator 8 (Arellano and Bond, 1991). Arellano-Bond difference GMM estimator is specifically designed for panels with large-n and short-t 9 and transforms our equation (1) into: And from equation (2) we get By first differencing the regressors, the fixed effects are removed because they do not vary with time. All these reasons ensure efficiency and consistency of our estimates provided that instruments are adequately chosen (the validity of the instruments is tested for with the Hansen test). The Hansen test of overidentifying restrictions has the null hypothesis that instruments are exogenous. A rejection of this null hypothesis implies that the instruments are not satisfying the orthogonality conditions required for their employment. A further test is the Arellano-Bond test of autocorrelation of errors, with as a null hypothesis no autocorrelation in differenced 7 Tested for with the Breusch-Pagan test (designed to detect any linear form of heteroskedasticity) and the White s test (general test for heteroskedasticity allowing for both non-linear forms of heteroskedasticity and errors to be non-normally distributed). Both test statistics had large chi-square values rejecting the null-hypotheses (constant variance for Breusch- Pagan and homoscedasticity for White s test, respectively). 8 One step difference GMM. Instruments are the dependent variable and the bank-specific variables (second and third lags for the crisis years ( ), and collapsed for the other two time spans). Especially Roodman (2009) shows how collapsing instruments can control efficiently for instrument proliferation. Restricting lag length to three for the crisis years was done in order to further cut the number of instruments. Output gap, real GDP, and the monetary policy indicator (EONIA interest rate change) are considered as exogenous instrumental variables. Standards errors are heteroskedasticity and autocorrelation robust. 9 In large-t panels a shock to the fixed effect (showing in the error term) will decline over time. Similarly, correlation of the lagged dependent variable with the error term will be insignificant (Roodman, 2006). 10

11 residuals. Specifically, the second order test (reported here as the AR(2)) in first differences tests for autocorrelation in levels and is more relevant. Again, the failure to reject the null hypothesis is the preferred outcome Estimations results Results can be seen in Table 4 with proper diagnostics: the Hansen test does not reject the overidentification conditions and the tests for serial correlation find no second order serial correlation. We have first estimated equation (1) for the whole time span, and then for the time of relative financial stability (before 2008) and time of crisis (from 2008 onwards). The first specification includes one dummy interaction term with stakeholder bank dummy and lagged interest rate. The second specification includes two dummy interaction terms with a cooperative bank dummy and lagged interest rate as well as a savings bank dummy and lagged interest rate. First looking at the whole time span ( ) we see that a contraction (expansion) in monetary policy leads banks to reduce (increase) their loan supply. This effect is dampened for stakeholder banks as a group, as well as for cooperative and savings banks separately (their coefficients are positive and statistically significant at 1% level). This indicates that stakeholder banks loan supply is less affected by changes in interest rates than that of shareholder banks. During times of financial stability ( ) we can note that the interest rate has a larger absolute effect on loan growth changes (although it loses some of its statistical significance); i.e. the lending channel is stronger when financial markets are working more conventionally. Again it seems that for stakeholder banks as a group as well as for cooperative and savings banks separately, their loan supply is less affected by changes in interest rate. During the recent crisis ( ) this negative effect of interest rate changes and loan supply growth seems to be lower in absolute value (statistically significant at 5% level). Coefficients of the interaction terms of ownership groups and interest rates remain positive and are statistically significant (at 10% level) for the stakeholder bank group and cooperative banks. For savings banks, the coefficient is no longer statistically significant, and so they are statistically no different from their shareholder counterparts. Here also the lagged dependent variable loses its statistical significance and so we could expect some model misspecification. This result is not entirely unexpected; loss of trust towards other banks and customers has made banks reluctant to increase their loan supply although central bank has lowered interest rates almost to the zero-lower bound. This could have made the traditional methodologies of studying the lending channel inadequate and thus models with a short-term interest rate as the monetary policy instrument no longer suitable for studying the effectiveness of monetary policy. One reason could also be the small number of yearly observations because of the shorter time span especially while using difference GMM. As for the different bank-specific variables, it seems that capitalization (ratio of equity to total assets) is an important variable in explaining bank loan supply behavior especially during relative financial stability: better capitalized banks can better smooth their lending from interest rate changes. Size, on the other hand, seems to be highly important also during the crisis; reflecting the fact that the larger the bank, the less its loan supply decisions are affected by changes in interest rate. These results that bigger and better capitalized banks are less affected by interest rate changes were already found in Kashyap and Stein (1995) on US data and are further confirmed in some European studies (see e.g. Ehrmann et al., 2001). Liquidity (share of liquid assets to total assets) has a negative and statistically significant coefficient in the first four columns; banks with relatively large amount of liquid assets on their balance sheets are responding more strongly to changes in interest rate. This negative effect is however no longer statistically significant during the crisis years. This result contradicts most of the previous empirical papers on the subject (with the exception of De Santis and Surico (2013) for some of their regressions): liquidity is more often found to have a positive coefficient with banks that have relatively more liquid assets on their balance sheets are better able to shield their lending activity from changes in short term interest rates. Our findings could be explained for one by the bank capital channel. If banks have relatively 11

12 more liquid assets, the ratio of loans to total assets is already lower. Now as market interest rates increase, an even lower fraction of longer-maturity loans can be renegotiated with respect to short-maturity deposits and thus bank faces higher short term costs. These costs could then be more easily lowered by cutting back lending than selling other types of securities that are more liquid. These results indicate that even though we did find that bank-specific balance sheet variables have an impact on the strength of the lending channel (in line with previous empirical studies), banks mission and ownership forms role is just as important. In fact we argue that this could be one explanation behind the lack of consensus among empirical research done on Euro area regarding different effects of balance sheet variables (see latter part of Section 2 for an overview). In those studies, banks were treated as having identical business models, only differing for example by size or relative share of equity. However, as banking sector composition diverge between countries in Euro area (although not all inclusive, Table 1 gives a broad idea); we cannot draw inferences on bank lending channel strength by only concentrating on bank balance sheet differences Robustness Next, we provide some robustness checks in order to gain more validation for our results presented above. First, based on our results we argue that while savings banks behave like their cooperative peers before the recent crisis they become statistically no different from shareholder banks during the crisis years while cooperative banks continue to behave less procyclically. Looking at the last column in Table 4 however, one could claim that by comparing the statistical significance between the two coefficients of cooperative and savings banks (rather than vis-à-vis commercial banks) one might not find any difference. To back up our initial claim, we perform the same estimations but first by excluding savings banks from the estimation sample and only using one interaction term (cooperative bank dummy interest rate change) and second by excluding cooperative banks from the estimation sample and only using (savings bank dummy 12 interest rate change) as the interaction term. Results can be seen in Table 5, first three columns excluding savings banks and last three excluding cooperative banks, respectively. Excluding savings banks from the estimation sample changes very little for the separate effect of interest rate changes on cooperative banks loan supply. Interest rate change has a statistically significant and negative effect on the overall bank credit supply, and this effect is dampened for the cooperative banks throughout the estimation period, also just looking at the crisis years ( ). If we exclude cooperative banks and compare the relative difference only between savings and shareholder banks (last three columns in Table 5), things differ. Although it seems that the interest rate change effect on loan supply of savings banks is dampened if we look at the whole time span, it does no longer hold for the crisis period. We could thus conclude that the tendency of smoothing the lending supply is stronger for cooperative banks relative to savings banks during the recent period of financial instability. Second, following arguments in Jiménez et al. (2012) it could be that the large presence of German banks in our data could render interest rate changes somewhat endogenous. Being at the core of the Euro area, changes in economic and monetary conditions are likely to be more correlated in Germany than in smaller or more peripheral countries. By looking only at the non-core Euro area countries there would be more exogenous variation in monetary conditions, allowing us to better separate its effects from those of national economic conditions. Also, more than one half of our observations are from Germany, so any results could reflect German idiosyncracies. We redid our estimations first excluding Germany from the dataset and then excluding the socalled core countries that have been performing most similarly to Germany with respect to different economic

13 and financial measures (Germany, Netherlands, Finland, and Luxembourg). Results can be seen in Table 6 (excluding Germany) and Table 7 (excluding core countries). It seems that our results remained more or less intact after excluding Germany and also after excluding the rest of the core countries from our dataset when we look at the whole time period (first two columns for both tables). The negative effect of the interest rate change on banks loan supply is dampened for the stakeholder banks as a group as well as for both cooperative and savings banks separately. One difference is the statistical insignificance of both capitalization and size of banks with the subsamples. Liquidity remains the only bankspecific variable that has a statistically significant, amplifying effect on banks loan supply. Capitalization and size become statistically significant when we look solely at the time of relative financial stability. The lending channel becomes stronger in absolute terms during , cooperative and savings banks maintaining their dampening effect. Looking at time during the recent crisis (last two columns for both tables) the direct effect of interest rate changes on loan supply loses its statistical significance, albeit the interaction term between cooperative banks dummy and interest rate change is positive and statistically significant (at 10% level) when excluding Germany from the dataset. The coefficient related to savings banks interaction with interest rate is negative indicating a further amplification of the lending channel (although the coefficient is statistically insignificant). Larger banks are better able to shield their lending supply also during the crisis. With the first specification while excluding Germany in first (third) column we can note that the Hansen overidentification test is rejected at 10% (5%) level, indicating that instruments may not be valid. However, the test is again passed when we break the stakeholder banks in two (second (fourth) column). The same problem can be found for the time of relative financial stability when excluding the core countries. As a third robustness check, we wanted to see whether the problems in Spanish savings bank sector and resulting massive reforms and fusions could have affected our results. Having become universal banks, Spanish savings banks expanded their activities across Spain and abroad and contributed to the build-up of excess capacity and risk concentration in the Spanish banking system which was all revealed by the recent crisis (IMF, 2012). During the crisis, several Spanish savings banks have been turned to commercial banks, or intervened and resolved; reducing the number of institutions from 45 to 11 by May Thus we redid our estimations for the whole time period, and for and separately excluding Spain from the sample (Table 8). Again, our results seem to be rather robust to the exclusion of Spanish banks from our sample looking in particular at the results for the whole time span. Now, unlike with the sample excluding Germany or the core countries, size and capitalization maintain their statistical significance in explaining banks loan supply also after we excluded Spain from the sample. Stakeholder banks as a group and cooperative and savings banks separately seem to follow less procyclical lending policies. When we break the time span in two, statistical significance of the estimates become weaker but it still seems that stakeholder banks as a group as well as cooperative banks especially continue to supply loans less procyclically. Size is again the only bank-specific variable that remains statistically significant during the recent crisis. We also note that the Hansen test statistic rejects the null hypothesis of exogenous instruments in specification (1) for the whole time span at 5% level, but is passed when we look at cooperative and savings banks separately. As a last robustness check, albeit the use of changes in short-term interest rate as a measure of change in monetary policy ties in with previous literature analyzing lending channel at bank level (see e.g. Kishan and Opiela, 2000 and Ashcarft, 2006 among others); we wanted to check that our results prevail when using ECB s main refinancing operations (MRO) interest rate changes instead of the Eonia overnight-rate. While Eonia is a weighted average of all overnight unsecured lending transactions in the interbank market, we could run into 13