Monetary policy transmission to bank lending and

Monetary policy transmission to bank lending and cross-country heterogeneity in the Euro area Pietro Grandi1 1

LEMMA, University Paris 2, Pantheon-Assas June 6, 2018 Abstract

Is the transmission of monetary policy to bank lending heterogeneous across the Euro Area? This paper employs annual bank-level data to test whether the bank lending channel of monetary policy was heterogeneous in the Euro area over the period 2007-2016. To do so, we follow a simple procedure that allows direct testing of the presence of heterogeneity of monetary transmission across countries by defining a single, bank-specific criterion to select a sub-sample of banks most likely to be sensitive to the bank lending channel. We find robust evidence that the transmission of monetary policy to bank lending was heterogeneous across countries that were differently exposed to the sovereign debt crisis – on average the same 1% policy rate cut leads to a 1.9% increase in lending by banks located in non-stressed countries and to a 0.9% increase for banks located in stressed countries. Unconventional monetary policy – as measured by the ECB shadow rate – also appears to be unevenly transmitted across countries. These findings cannot be ascribed to differences in macroeconomic and financial conditions across countries, to balance sheets differences across banks, or to differences in credit demand at the local level. Keywords: Bank lending channel, monetary policy transmission, cross-country heterogeneity, financial structures. JEL classification: E52,E58,E42,F33

1

Introduction An important measure of the effectiveness of monetary policy is the extent to which it is transmitted via the banking system to the real economy in the form of loans to households and non-financial corporations (NFCs). That such transmission is homogeneous across regions remains a challenging requirement in the European Monetary Union where the single monetary authority – the European Central Bank – conducts policy for a group of 19 member countries whose banking systems are not yet completely integrated. As a consequence, the same monetary policy may have different effects on credit outcomes in different countries: in other words, monetary transmission to bank lending may at times be heterogeneous. While diversity in banking systems is a long-standing feature of the union,1 the impact of asymmetric financial shocks may widen differences across countries and further prevent the even transmission of monetary policy. This issue was particularly relevant between 2010 and 2013, when the region’s sovereign debt crisis led to significant fragmentation on financial and credit markets. The ensuing asymmetry between countries with different sovereign exposures motivated concerns among policymakers that “[the ECB] faced severe impairments to the transmission of monetary policy across the euro area, with marked heterogeneity from country to country”(Draghi, 2014). Using a panel of more than 2500 banks from all 19 Euro Area countries covering 2007-2016, we study whether the response of bank lending to monetary policy differed between stressed and non-stressed countries in the Euro Area. The empirical strategy consists of two steps: first, the policy response of bank lending is estimated as the reaction of financially constrained banks to monetary shocks – the bank lending channel (Kashyap and Stein, 1995); second, cross-country heterogeneity is tested by looking at differential bank-lending channel responses by banks located in stressed vis-`a-vis banks located in non-stressed countries. Results indicate that monetary transmission to lending is weaker for banks located in stressed countries as compared to banks located in non-stressed countries: on average the same 1% policy rate cut leads to a 1.9% increase in lending by non-stressed-country banks as opposed to a 0.9% increase for stressed-country banks, everything else held constant. Unconventional monetary policy — as captured by the ECB shadow rate – also appears to be unevenly transmitted to bank lending across country groups. Our results are robust to controlling for varying macroeconomic and financial conditions across countries, balance sheets differences across banks, and differences in credit demand at the local level. Our paper provides a methodological contribution by devising a simple procedure to test for the presence of cross-country heterogeneity in the bank lending channel of monetary policy. Unlike standard strategies that test bank lending response to monetary policy conditional on separate bank-specific balance sheet characteristics such as bank size, liquidity and capitalisation (Altunbas et al., 2009; Gambacorta, 2005; Gambacorta and Marques-Ibanez, 2011; Gambacorta and Shin, 2016), we restrict our test on a subsample of banks that are simultaneously small, illiquid and undercapitalised. We consider this approach particularly suited to the task of gauging heterogeneous responses to monetary policy within a monetary union as it sidesteps national idiosyncrasies inherent in each Euro area banking systems. The paper has implications for policy. Insofar heterogeneity in monetary transmission across financial systems can complicate the task of monetary policymakers (Cœur´e, 2017; Guiso et al., 2000), explicit consideration to the degree of heterogeneity should inform the optimal monetary policy strategy in a monetary union. For instance, as proposed by Gambacorta (2003), if the members of a monetary union have asymmetric bank lending channels, the optimal monetary policy would be influenced not only by the magnitude of the variance of the shock, but also by its point of origin. In this respect, recent initiatives aimed at deepening financial and banking integration such as the Banking Union and the Capital Markets would greatly help reducing the extent of heterogeneity in monetary transmission. Second, even lacking greater Euro area integration in finance and banking, tackling major sources of bank distress at the country-level (e.g. non-performing loans and sovereign exposures) would lessen or remove barriers to the even transmission of monetary policy to bank lending across the monetary union. The remainder of the paper is organized as follows. The first section presents some stylised facts on bank lending in the euro area, while the second section reviews the related literature. Data and identification strategy are discussed in the third section, while the empirical analysis is conducted in the fourth section. The fifth section concludes. 1

See, for instance, Cecchetti (1999) and Ehrmann et al. (2002) .

2

1

Bank lending and heterogeneity in the Euro area

This paper is motivated by the observed heterogeneity in bank lending observed since 2010 in the Euro Area. During the Euro Area sovereign debt crisis (2010-2012), financial and real asymmetries became evident between countries with different exposures to sovereign risk.2 On the financial side, banks located in sovereign-stressed countries were highly exposed to the suffering of their own sovereign. With domestic government securities amounting to the lion share of their fixed income portfolio, stressedcountries banks saw their balance sheets rapidly deteriorating as the value of their bond holdings – widely used as collateral – started plummeting. Soon banks in the periphery and sovereigns became ‘joined at the hip’ (Mody and Sandri, 2012). The ensuing financial fragmentation is regarded to have determined, for banks, different degrees of access to external finance in different countries (Durr´e et al., 2014). On the real side, economic conditions varied remarkably across countries, with the recession being for some countries more severe and enduring than for others. Such heterogeneity was reflected on retail credit markets where a moderate and short-lived decline in bank loans in non-stressed countries contrasted with a severe and persistent credit crunch in stressed countries (Gilchrist and Mojon, 2014). The ECB responded by launching a mix of conventional and unconventional measures to provide cheap liquidity to banks and sustain credit supply to the real economy. Moreover, by restoring the viability of money and sovereign markets, the ECB sought to preserve the integrity of monetary transmissions in the Eurozone (Durr´e et al., 2014). [Figures 1,2,3 and 4 about here] In spite of the expansionary stance of the single monetary policy (see figure 1), bank lending remained substantially heterogeneous across countries. Figure 2 shows that while following the outbreak of the Great Financial Crisis in 2008 bank lending plunged for all Euro Area countries alike, from 2010 a clear divergence emerged: while non-stressed countries banks experienced only a moderate and short-lived decline in lending, lending in stressed countries was considerably restricted, resulting in a severe and persistent credit crunch.3 Similarly, real interest rates on new lending to NFCs was similarly divergent across stressed and non-stressed countries, with corporate borrowing costs remaining substantially higher in the latter (see figure 3). Such divergence remained very persistent, arising in mid-2010 and carrying through the present day: on average, between January 2010 and November 2017, bank lending growth in stressed countries was 7.46% slower and average interest rates 0.57% higher than in non-stressed countries. The extent of cross-country heterogeneity in bank lending was economically relevant. This can be appreciated by looking at figure 4, where the dispersion of the growth rates of adjusted loans to NFCs across euro area countries is compared with the dispersion of real GDP growth.4 For the whole period, cross-country heterogeneity in loan growth was constantly higher, in absolute terms, than heterogeneity in GDP growth. Furthermore, loan growth dispersion was especially high in the run up to the Great Financial Crisis, during the sovereign debt crisis (2010-2012) and during the following recovery (2014-2016). A central question is whether the observed developments were mainly determined by supply or demand forces. A possible explanation may have heterogeneous credit flows across countries being driven by loan demand, as within sovereign stressed countries domestic firms may reduce investment and credit demand. In this case, subdued levels of credit flows in stressed countries may simply reflect lower credit demand by domestic households and firms. Yet, comparing figures 2 and 3 provides some prima facie evidence against the credit demand story, as prices (inflation-adjusted interest rates) and quantities (loan growth) move in opposite direction which suggests a downward supply shift. Furthermore, the figures highlight a price (interest rates) differential between stressed and non-stressed countries matching a quantity differential (credit flows) between stressed and non-stressed countries: 2

Throughout the paper we work with two country groups: “stressed” countries (Cyprus, Greece, Ireland, Italy, Latvia, Lithuania, Portugal, Slovenia and Spain) and “non-stressed” countries (Austria, Belgium, Estonia, Finland, France, Germany, Luxembourg, Malta, the Netherlands, and Slovakia). As in Altavilla et al. (2017) we define as “stressed” (i.e. subject to high sovereign stress) countries whose 10-year sovereign yield exceeded 6% for at least one quarter in our sample period (2007-2016). Such categorisation is analogous to “peripheral” and “core” countries as found in Lane (2012) and De Grauwe (2013). 3 It can be noted that prior to 2008 bank lending was actually growing faster for banks in stressed countries, in good measure reflecting the overheating in the banking sector related to the build-up of housing bubbles in countries such as Spain and Portugal. 4 Adjusted loans refer to loans adjusted for sales and securitisation, as well as for the impact of “notional cash pooling” positions that result from cash management services provided by certain banks to corporate groups. Adjusted loans thus give a better signal of underlying credit developments for euro area borrowers and improve the comparability of country-level data (ECB, 2017).

3

from 2012 lending growth to NFCs in stressed countries falls unabated, while at the same time the average (inflation-adjusted) interest rate on loans to NFCs increases from 2% to almost 5%. Overall, the steeper contraction in lending in stressed country seems more consistent with a shift in credit supply (by contrast, a shift in credit demand would cause prices and quantities to move in the same direction). While aggregate data tend to provide only suggestive evidence of microeconomic developments, we can already surmise that much of the variability of bank lending across countries cannot be attributed to changes in loan demand. [Figure 5 about here] Similar evidence is provided by the ECB Bank Lending Survey.5 Figure 5 plots the net percentage of banks reporting increases or decreases in credit demand over the previous quarter, distinguishing between banks located in stressed versus non-stressed countries. The two series are highly correlated, suggesting credit demand seemingly moved in tandem.6 If anything, loan demand appears to have been higher for stressed countries between mid-2013 and mid-2014, when the wedge in lending between stressed and non-stressed countries approaches its largest level. Overall, survey evidence also suggests that credit demand was not likely to be a key determinant of the heterogeneity in credit outcomes across the Euro area. A set of policy questions then arises: How can geographically heterogeneous credit outcomes follow homogeneous policy impulses? Does bank lending respond differently to monetary policy in different countries? The remainder of this paper seeks to shed light on these questions.

2

Literature review

Our research straddles between two large literatures: the first assesses the transmission of monetary policy to bank lending (the bank lending channel), while the second is concerned with the consequences of heterogeneity and fragmentation in a monetary union. The bank lending channel is the traditional mechanism through which monetary policy directly influences bank lending. In its original form (Bernanke and Blinder, 1988; Bernanke and Gertler, 1995; Kashyap and Stein, 1995) the channel stems from financial frictions in the market for uninsured nondeposit funding imposing an external finance premium on banks exposed to informational asymmetries. The channel hence predicts monetary policy to have a stronger effect on these banks, namely in the form of a larger shift in their loan supply schedule.7 Empirical studies use bank-level data and focus on a subset of banks that are presumably exposed to information problems on wholesale funding markets. For instance, small banks may have difficulty accessing external funds given their simple capital structure (Kashyap and Stein, 1995), illiquid banks usually face higher costs to offset deposit losses (Gambacorta, 2005; Jim´enez et al., 2012; Kashyap and Stein, 1995, 2000; Kishan and Opiela, 2000), while poorly capitalized banks are believed to face an external finance premium as they are perceived to be riskier (Gambacorta and Shin, 2016; Jim´enez et al., 2014, 2012; Peek and Rosengren, 2005). For all these reasons, small, illiquid and poorly capitalized banks are considered as more exposed to information problems, more sensitive to policy impulses and hence the ideal conduit of the bank lending channel. A second related literature considers the monetary policy implications of heterogeneous banking structures within a monetary area. It is generally recognised that the even transmission of monetary policy within a single currency area requires synchronized business cycles and similar economic and financial structures across member states (De Santis and Surico, 2013). While there is evidence that euro area economies are sufficiently synchronized as well as economically and financially integrated (Baele et al., 2004; De Grauwe and Mongelli, 2005; De Haan et al., 2008; ECB, 2017), differences across 5 The bank lending survey (BLS) provides information on supply and demand conditions in the euro area credit markets as well as on euro area banks’ lending policies, such as the credit standards banks apply when approving loans, the terms and conditions of new loans to enterprises and households and an assessment of demand for bank loans. Since 2003, the BLS is conducted four times a year. The survey is addressed to senior loan officers of a representative sample of around 140 euro area banks and takes into account the characteristics of each country’s national banking structures (ECB, 2017). 6 The pair-wise correlation between the two series is 0.45 and is statistically significant at the 0.001 significance level. 7 The bank lending theory further assumes that financially constrained banks will tilt their asset allocation in a particular way in response to a monetary policy shock. As banks face an asset choice (holding liquid bonds or earning a spread on illiquid loans) and a liability choice (deposit or non-deposit funding), the risk of future deposits shortfalls and payments mismatches prompts banks to hold extra bonds as a liquidity buffer. By lowering the external finance premium and relaxing the liquidity constraint, monetary expansions should, ceteris paribus, encourage reallocation away from bonds and towards illiquid (but more profitable) assets such as loans (Butt et al., 2015).

4

banking systems and the lack of integration in retail credit markets persist to the present day and may introduce an element of heterogeneity in the transmission of the common monetary policy to bank lending across countries (De Santis and Surico, 2013). This hypothesis was originally put forth by Cecchetti (1999) and Kashyap and Stein (2000) who argued that differences in structures of financial intermediation, legal systems, corporate finance practices and capital market development between EMU countries may give rise to differences in the way monetary policy is transmitted to bank loans across the union. The rejoinder to this hypothesis was that even if financial structures differ substantially across countries, economic and financial convergence in the euro area would imply that any difference in the transmission mechanism will eventually disappear over time (Ciccarelli and Rebucci, 2002; Dornbusch et al., 1998). European monetary integration prompted interest in the study of differences in the bank lending channel between euro area countries and the extent of cross-countries heterogeneity therein. Among theoretical works, Gambacorta (2003) explicitly models a currency union composed of countries with heterogeneous bank lending channels and finds that the optimal single monetary policy should take differences across financial structures into account as they are bound to affect the geographical distribution of the policy final effect. More recently Cargoet and Poutineau (2017) model a two-country DSGE model to analyse the impact of homogenous unconventional monetary policies on standard macroeconomic variables in a heterogeneous monetary union where heterogeneity arises from endogenous regional credit disruptions. They find that a common credit policy is a second best solution when compared to country-specific credit policy measures. On the empirical side, while sharing very similar methodological approach, evidence on the BLC and heterogeneities in the euro area is mixed.8 Amongst country-specific studies, using panels of, respectively, Italian, Spanish and French banks for periods between 1990 and 2001, Gambacorta (2005); Hernando and Mart´ınez-Pag´es (2001) and Loupias et al. (2002) detect a bank lending channel working only for illiquid banks. Yet, cross-countries studies reveal a different picture. Using balance sheet data to investigate the bank lending channel, De Bondt (1999) finds some evidence for a bank lending channel for Italian, French and German banks, while Altunbas et al. (2002) detect a bank lending channel working through size only in Italy and, partially, in Spain. In a very comprehensive study, Ehrmann et al. (2002) use annual bank balance sheet data owned by the Eurosystem to explicitly address cross-country heterogeneity in monetary transmission by estimating and comparing the bank lending channels for France, Germany, Italy and Spain. The resulting evidence shows that the effect of monetary policy on bank credit was quite homogeneous across countries both prior and after the introduction of the euro. Ten years later, in a very similar exercise De Santis and Surico (2013) find the opposite result. Drawing from a sample over the period 1999-2011, the authors report that the impact of monetary policy is heterogeneous across countries, with the bank lending channel being stronger in Germany and Italy and weaker in Spain and France. Further evidence in favor of heterogeneity of the bank lending channel in the euro area is provided by De Santis (2015) who finds that between 1999 and 2011 monetary policy easing supported relatively more the lending of small French banks and illiquid German banks with respect to Spanish and Italian banks. Interest in cross-countries heterogeneity in monetary transmission was reignited in the context of the Euro Area sovereign debt crisis and the resulting monetary policy response by the ECB. For instance, Ciccarelli et al. (2013) estimate a VAR model over the period 2002-2011 for a panel of 12 euro area countries and identify a bank lending channel at the macro-level. In particular, the authors find that the response of bank lending to monetary policy is stronger in countries under sovereign stress (Greece, Italy, Ireland, Portugal and Spain in their analysis) as opposed to other countries (Austria, Belgium, Finland, France, Germany, Luxembourg and the Netherlands), highlighting substantial heterogeneity across countries. However, the authors maintain that ECB’s unconventional monetary policy measures appear to have partially mitigated the bank lending channel between 2010 and 2011. In a recent study, using a global VAR model that exploits panel variation among all euro area economies between 2007 and 2015 and takes into account cross-countries interdependencies, Burriel and Galesi (2018) find that the Eurozone-level response of new credit operations to unconventional monetary policy is insignificant, which the author interpret as reflecting a large degree of heterogeneity across countries. 8

From a methodological standpoint, all studies are essentially applications of the original test of the bank lending channel by Kashyap and Stein (1995) onto European data.

5

3

Data and identification

Through the bank lending channel expansionary monetary policy can boost lending by affecting banks’ access to external finance. However, the bank lending channel may be heterogeneous across countries. In this paper we ask the following question: Was monetary transmission to bank lending different between banks located in stressed versus non-stressed countries? To answer, we use bank balance sheet data and a novel methodology based on the bank lending channel theory.

3.1

Data

We use annual data on euro area banks containing balance sheet and income statement information. Our main data source is Fitchconnect, a comprehensive commercial database provided by Fitch Solutions, a provider of credit market data and part of the Fitch Group. The original Fitchconnect sample covers 3110 commercial, savings and cooperative banks from all 19 euro area member countries for the period 2006-2016. We merge this database with Orbis Bank focus (previously Bankscope) to recover information on bank specialisation (e.g. commercial, savings or cooperative banks), public listing, role in the banking group and identity of the parent bank.9 We are able to match 2629 banks (85% of the original sample), of which two third are matched through shared unique identifiers (the ESCB Monetary and Financial Institutions identifier, the Legal Entity Identifier and the Ticker code), while the remaining banks are matched through a “fuzzy merge”.10 As in Chodorow-Reich (2014) we fuzzy merge these banks using country and consolidation identifiers along with a bigram string comparator score of the bank name as reported in each database. We then follow Gambacorta (2005) and Ehrmann et al. (2002) and remove all nonsense observations such as negative entries for total assets, loans as well as leverage and liquidity ratios (doing so removes 3650 observations). Second, to keep the cross-section dimension as large as possible, we restrict the analysis to unconsolidated accounts. Third, we winsorise loan growth, total assets as well as leverage and liquidity ratios at 99% to minimise the incidence of outliers. The final sample is a balanced panel counting 2544 banks adding up to a total of 25440 bank-year observations over the 10 year period of interest. [Table 1 about here] The representativeness of our sample is documented in table 1. On average, bank loan data covers 16% of corresponding national lending aggregates or 26% of the Euroarea loan market once we weigh by each country’s nominal GDP.11 We have higher (weighted average) coverage of non-stressed countries, with bank loans in the sample representing 16% and 31% of stressed and non-stressed countries’ totals,respectively. We further collect macroeconomic data from a number of public sources: real GDP and inflation rates as well as an index on housing prices are taken from Eurostat, while data on money market rates (Euribor and Eonia) and Eurosystem assets are collected from the European Central Bank SDW.12

3.2

Identification

In this section we discuss our strategy to identify the bank lending channel of monetary policy for banks in stressed and non-stressed countries. Specifically, we will discuss identification issues related to isolating the BLC and lending supply effect and propose our solutions.

3.2.1

Identifying the bank lending channel across the Euro area

The bank lending channel requires some banks to be more sensitive to monetary policy at the margin. Therefore, the channel is traditionally tested by looking at the differential lending response to monetary policy by banks conditional on either their size, liquidity, capitalization or credit risk (Altunbas 9 This is the variable Global Ultimate Owner defined as the global credit institutions having an ownership stake in the bank larger than 25%. 10 A bigram string comparator calculates the fraction of consecutive character matches between two strings. We perform the fuzzy merge using the Stata ado file reclink2 written by Micheal Blasnik. To ensure accuracy, we also perform a clerical review of all matches. 11 While our sample is not as representative as that of Altavilla et al. (2017), our coverage is slightly higher if compared with the 10% of the syndicated loan data used by Popov and van Horen (2013) and Acharya and Steffen (2015). 12 We describe these variables in details in appendix C

6

et al., 2009; Gambacorta, 2005; Gambacorta and Marques-Ibanez, 2011). However, if one is interested in testing and comparing the bank lending channel across countries, issues arise about the suitability of using bank financial characteristics separately to measure a bank’s sensitivity to monetary policy. To begin with, some banks may simultaneously be small but liquid, or big but overcapitalized, or overcapitalized but low-risk. If such occurrences are systematic in some banking systems, the bank lending channel-informational content of any individual measure per se would be confounded for that country, thus complicating cross-country comparisons. Evidence suggests this is indeed the case for some euro area countries. In Italy, for instance, small banks are often liquid and highly capitalized: banks of small size rooted in the territory are good at drawing resources from local deposit markets and are thus particularly liquid. Additionally, small Italian cooperative banks can be prevented from distributing dividends and thus tend to capitalize retained earnings (Gambacorta, 2005). By the same token, German banks tend to have low capital, but this is usually compensated by the low riskiness of their asset structure (Ehrmann et al., 2003). There are also specific problems related to the informational content of bank-specific characteristics when they are used as standalone proxies. For instance, bank networks and relationship lending may distort the informational content of bank size. Large banks in upper tiers of the network usually serve as liquidity providers to small banks in lower tiers in case of liquidity deficits, potentially neutralizing the latter’s liquidity constraint (Ehrmann et al., 2003). Furthermore, banks that engage in relationshiplending – typically small banks in the euro area (Ongena and Smith, 2000) – may provide implicit guarantees to their customers that they will grant access to credit regardless of liquidity and monetary policy conditions (Rajan and Zingales, 1998). As a result, bank size may not be a good proxy to evaluate the distributional effects of monetary policy across banks.13 Bank capital can also be an ambiguous indicator: under Basel II and Basel III risk-based capital requirements, after a monetary expansion banks near the capital regulatory floor may be less able to increase lending - thus lowering capitalassets ratios - unless provided with additional capital. Additional liquidity may instead be allocated to assets free from capital requirements, such as government bonds. Accordingly, no differential effect of policy on lending would be captured via bank capital (Van den Heuvel, 2002). Overall, to test for the presence of a bank lending channel across countries, a more convenient procedure is to define a necessary criterion that is more stringent than those implied by standard methods. This paper departs from previous literature by proposing an alternative identification strategy for the bank lending channel whereby a single criterion selects a sub-sample of banks that are most likely to be sensitive to the bank lending channel of monetary policy. The logic is as follows: if through the bank lending channel monetary policy has any differential effect at all on some banks, then it should definitely be observed for banks that are simultaneously small, illiquid and under-capitalized. If no effect is detected for these banks, then it is very unlikely there is an active bank lending channel of monetary policy at all for the period considered. [Table 2 about here] Defining the criterion involves two steps. In the first step (summarised in table 2), three thresholds are defined to classify banks as small, illiquid or under-capitalized. For size, following Kashyap and Stein (2000) banks are considered “small” if, in any given year, they lie in the bottom 95% of the distribution of total assets for all countries.14 For capitalisation, banks are marked as under-capitalized when, in any given year, the distance between their regulatory capital ratio from the regulatory floor is in the bottom quartile of the distribution of distances for all countries, as in Borio and Gambacorta (2017) and Brei and Gambacorta (2016).15 Finally, banks are considered illiquid if, in any given year, 13 Additionally, in the euro area government guarantees and deposit insurance cover banks irrespective of their size: accordingly, deposit accounts at small banks should not be considered any riskier than deposits accounts at big banks (Ehrmann et al., 2003) 14 For 2016 this corresponds to a size cut-off around €14 billion. For comparison, the ECB bank size group classification used for supervisory and prudential statistics considers a bank small if, in any given year, its consolidated assets are less than 0.005% of total consolidated assets of EU banks; medium-sized if its assets are between 0.005% and 0.5% of total consolidated assets of EU banks; and large if its assets are greater than 0.5% of total consolidated assets of EU banks. For 2016, these cut-offs corresponds to €2.1 billion and €215 billion, respectively. Hence, our “small” category is slightly larger than that used by the ECB. 15 The regulatory floor correspond to the minimum requirements for risk-weighted capital ratios for Basel I: Tier 1/RWA > 4% (Borio and Gambacorta, 2017). In our sample, the banks at the 25th percentile of the distribution of regulatory distance have a tier 1 ratio of 11% and a leverage ratio (equity capital/total assets) of 7.7%. To maximise the sample size, we replace the tier 1 ratio-based criterion with a leverage ratio-based cut-off whenever observations for the former variable are missing. Results are actually more conservative if we do not implement this procedure.

7

their liquidity ratio is smaller than 20% as in Kashyap and Stein (2000). In our sample this corresponds to banks laying in the three bottom quartiles of the distribution by liquidity ratio. In the second step, an indicator variable (BLC banks) is defined to take value 1 for all banks that are, in any given year, simultaneously small, under-capitalized and illiquid, and zero otherwise. That is: ( 1 if a bank is small & illiquid & undercapitalised BLC − banksit = 0 otherwise The criterion finally selects a subsample of 468 ”BLC-banks” corresponding to 18% of the original sample. All tests of the bank lending channel will hinge upon comparing the lending response of this subsample vis-`a-vis other banks. This approach has two main advantages and one drawback. Amongst the advantages, i) it allows doing away with all individual suitability issues linked to the choice of any particular measure of bank sensitivity to monetary policy used separately; ii) it is invariant to the idiosyncrasies inherent in each banking system. In doing so, this approach resurrects previous work on the bank lending channel whose focus was on banks that were simultaneously small and illiquid (Kashyap and Stein, 2000) or small and undercapitalised (Kishan and Opiela, 2000). On the other hand, the main drawback of this approach is selection bias. Being based on thresholds unrelated to any specific national banking system, by construction the criterion may prove too stringent for some banking system and too lax for others. That is, the criterion may select few banks in big, highly capitalized and liquid banking systems, and many in small, over-capitalized and illiquid systems. In this case the selected subsample would be endogenously skewed towards some countries, misrepresenting the original sample and impairing the validity of our tests. [Table 3 about here] We look into this issue in table 3, where we list the number of banks sorted by country, alongside their share of all banks in the sample. We intend to compare our BLC subsample of 468 banks with the original full sample of 2544 banks. Specifically, we are interested in determining if our subsample misrepresents the original sample thus biasing results in a particular direction. Column 5 of table 3 proposes a clear-cut measure of misrepresentation, by reporting the differences between the shares of total banking population of the two samples: negative numbers correspond to under-representation, and positive numbers to over-representation. One observation stands out: the criterion gives larger relative importance to non-stressed countries vis-`a-vis stressed countries: the latter end up being relatively under-represented in the BLC subsample. This rules out the concern that our subsample is skewed towards stressed-countries banks and alleviates the concern that our heterogeneity test may be guided by sample selection.

3.2.2

Identifying credit supply: within-city estimation

While we may be able to identify the bank lending channel, we may fail to disentangle loan supply from loan demand. In the bank lending channel literature it is common to assume that credit demand faced by banks is homogenous within countries and exogenous to bank-specific characteristics (Altunbas et al., 2009; Ciccarelli et al., 2013; Ehrmann et al., 2002; Fung´acˇ ov´a et al., 2014; Gambacorta, 2005; Gambacorta and Marques-Ibanez, 2011) . However, this assumption is problematic: for example, if customers of small banks tend to increase their demand for funds more promptly in the face of a monetary loosening, it is impossible to identify the bank lending channel. Furthermore, as discussed above, differences in credit demand may be responsible for lending divergence between stressed and non-stressed countries. Indeed, as emphasised by Altavilla et al. (2017), at times of sovereign stress firms and households located in stressed countries are likely to curtail investment and consumption. To the extent this translates into relatively lower demand for credit in stressed countries as opposed to non-stressed countries, this may explain the observed heterogeneous reduction in lending. While, as argued in section 1, loan demand does not appear to have an overriding influence on differences in bank lending across countries, information in our datasets and the unconsolidated nature of our data allows us to take a further step to control for credit demand at the bank-level. Following the contribution of Khwaja and Mian (2008), recent research significantly improved on the task of isolating credit supply through using loan-level data on a country-by-country basis (Iyer et al., 2014; Jim´enez et al., 2014, 2012). While we do not have access to euro area-wide loan-level data, we employ a strategy similar in spirit to Khwaja and Mian (2008) and in practice to Drechsler et al.

8

(2017): a within-city estimator. The idea is as follows: as noted by Fung´acˇ ov´a et al. (2014), many banks’ relevant deposits and lending market are often local. Moreover, geographical distance from borrowers has been recognised to be an important determinant of the availability and pricing of bank loans. This may come from inter-bank competition (Lederer and Hurter, 1986) or from monitoring costs increasing with lender-borrower distance (Sussman and Zeira, 1995). For instance, Degryse and Ongena (2005) find that a bank’s loan rates decrease with the distance between the corporate borrower and the lender. Building from this notion, we propose to use banks’ city branch location to control for local credit market conditions: using city×year fixed effects we will compare the lending response of different banks located in the same city in any given year, assuming they face similar lending opportunities and loan demand, thus leaving credit supply identified. Two important caveats apply. First, geographical locations as reported in our database may actually be unrelated with banks’ main credit market. Second, even if there is such a connection, lending opportunities for larger banks and bank holding companies are likely to be only weakly related to the local market in their jurisdiction. We address these concerns in two ways: first, in appendix D we present empirical evidence supporting our assumption that credit market conditions for banks in our sample are indeed related to their particular location; second, in section 4.4 we will explicitly take into account banks’ degree of independence and whether they belong to a group.

4

Empirical analysis

In this section we lay out our empirical analysis. We start with descriptive statistics and then introduce an econometric model whereby we formally test for heterogeneity of the bank lending channel of monetary policy.

4.1

Descriptive statistics

Table 4 presents descriptive statistics of some key balance sheet features for banks in our sample. Some clear patterns emerge. First, by construction BLC banks are smaller, less liquid and less capitalised. However, there are no great differences between the BLC-banks in stressed and non-stressed countries: while the average BLC-bank in stressed-countries is on average smaller (€743 million vs €1056 million), between country-group differences in capitalisation and liquidity are rather less conspicuous (8% vs 6% and 3% vs 2% for leverage and liquidity ratios, respectively). Second, the average BLC-bank has a larger share of loans on balance sheet (66% vs 58%) and expands lending at a faster rate (6% vs 4%). Again this remains the case for both banks in stressed and non-stressed countries. Third, banks in stressed countries tend to have a much larger exposure to government securities and NPL than banks in nonstressed countries (18% vs 3% and 11% vs 4%, respectively). BLC banks are not special in this respect – these features seem to be shared by all banks located in stressed countries. Taken at face value, this suggests that any differential bank lending response to monetary policy between the two country groups cannot be directly attributed to wide differences in observable bank characteristics, but rather to country location. [Table 4 about here] Figure 6, 7 and 8 plots some of these statistics over time for BLC-banks and other banks across country groups. For instance, sorting between stressed and non-stressed countries and plotting loan growth rates over time, figure 6 reveals a break in 2011, the start of the sovereign debt crisis with the Greek public debt crisis. Again, the divergence concerns all banks along country-group lines, with little difference between BLC-banks and other banks. We further document the extent of the divergence between country groups on banks’ balance sheet by looking at their sovereign debt and non-performing loans (NPL) exposures. Figures 7 and 8 reveal that all banks in stressed countries were significantly more exposed to domestic sovereign risk and NPL.16 Again, we stress that segmentation appears to be really between country-groups rather than between banks, as BLC and other banks are similarly exposed to sovereign and credit risk: in other words, trends are more country-dependent than balancesheet dependent. 16 We are confident that both sovereign risk and NPL are domestic phenomena in the sense that domestic banks loan portfolios are mostly composed of loans to domestic households and NFCs (transnational lending accounts only for a small percentage of overall lending (Bignon et al., 2013) and that most of banks sovereign securities are of their own sovereign, as reported by (Altavilla et al., 2017).

9

[Figures 5,6, and 7 about here] Overall, descriptive statistics suggest that whether a bank operates in a stressed or a non-stressed country is more informative than whether a bank is financially constrained or not. This observation tends to validate our identification – that is, differentiating banks by balance sheet characteristics does not alter the overall bank lending divergence observed between stressed and non-stressed countries. In other words, we will test for differential lending behaviour between banks that are very similar from a financial standpoint but that happen to operate in different countries (stressed or non-stressed countries).

4.2

Econometric model

We introduce here the econometric model that tests for the presence of heterogeneity in the bank lending channel of monetary policy across the Euroarea. Building from traditional estimations of the bank lending channel,17 model (1) is designed to detect heterogeneous bank lending channels between core and peripheral countries: ∆ log Loansict = αi + γct + β1 ∆MPt + β2 ∆MPt × BLC-banksit−1 + β3 ∆MPt × BLC-banksit−1 × Stressedi + β4 Xit−1 + β5 Yit + ict

(1)

Note that i indicates banks, c cities and t years. The dependent variable (∆ log Loans ) is the annual growth rate of bank net loans and is regressed on a measure of monetary policy (∆MP ) – in first differences to avoid spurious correlation – alongside two interaction terms and two vectors containing lagged bank- and contemporaneous country-level control variables, the latter vectors denoted as X and Y, respectively. We now discuss in detail the main variables and how they enter the model. Our main explanatory variable is an indicator of ECB monetary policy (∆MP ): across specifications we will alternate multiple measures. Our main variable is the annual change in money market rates: the 3-months EURIBOR rate and the EONIA interbank rate. Highly correlated with the ECB’s main refinancing rate, both EURIBOR and EONIA represent benchmark interest rates for bank’s funding and are widely used to study monetary transmission (see, for instance, Altunbas et al. (2009) and Jim´enez et al. (2014)). However, merely relying on interest rate indicators may not be sufficient to capture the full scope of monetary policy between 2007 and 2016. As discussed, from 2009 onwards the ECB embarked in a major program of unconventional monetary policy (UMP) measures aimed at rebooting monetary transmission and combating financial segmentation across the Euro area. In the face of such major developments, the policy rate approached the zero lower bound (ZLB) while the ECB continued to enact a series of unconventional policies designed to supplement the interest rate tool. As a result, as apparent in figure 1, money market indicators showed little variability especially after April 2012, and may under-represent the full scope of monetary policy activity during and after the sovereign debt crisis. Unfortunately, however, we do not have direct bank-level observations on banks’ participation into ECB unconventional programs: hence we use two aggregate measures. First, we take the ratio between the Eurosystem’s total assets and each country’s nominal GDP as a proxy for non-standard policy measures. As argued by Gambacorta and Marques-Ibanez (2011) and Peydro et al. (2017) this variable seeks to gauge the effect of large scale unconventional monetary policy measures adopted by the ECB through the corresponding expansion of the Eurosystem’s balance sheet.18 Third, we use the shadow rate constructed by Wu and Xia (2016). Building a nonlinear term structure model to analyse an economy operating near the zero lower bound (ZLB) for interest rates, Wu and Xia (2016) show that the shadow federal funds rate calculated by their model is closely correlated with key macro variables of interest since the Great Recession as with federal funds rate prior to 2009. The shadow rate can be taken as a summary indicator for conventional interest rate policy up to 2009 and for ECB’s unconventional monetary policy measures at the ZLB, including Long Term Refinancing Operations (LTROs), Targeted LTROs (TLTROs), the Asset Purchase Program (APP), the Negative Interest Rate Policy (NIRP) and forward guidance. As in Peydro et al. (2017), we employ the shadow rates for Europe.19 The bank lending channel and heterogeneity of monetary policy on lending are captured by the interaction terms. The first interaction term has the monetary policy indicator interacted with the 17

See, for instance, Gambacorta (2005). While the main focus studies of the bank lending channel typically focus on conventional monetary policy such as the interest policy (an exception is Butt et al. (2015)) from 2010 onwards the ECB launched several unconventional monetary measures that were explicitly designed to act in complementarity with conventional policy (Albertazzi et al., 2016). 19 Shadow rates data for Europe are downloaded from https://sites.google.com/site/jingcynthiawu/home/wu-xia-shadow-rates. 18

10

bank lending channel selection criterion (BLCbanks) as defined in section 3.2.1 and is lagged one year to attenuate endogeneity concerns. The second interaction term compounds the first term with a dummy (Stressed) which takes value 1 if the bank is located in a stressed country and 0 otherwise. X is a vector of bank-level control variables controlling for balance-sheet characteristics: we include total assets, liquidity ratio (cash, interbank liquidity and securities over total assets), leverage ratio (equity capital over total assets), deposits over total liabilities and loans over total assets. All variables are entered with one lag both to attenuate endogeneity concerns and because lending decisions conditional upon bank-specific features usually take place with a lag (Gambacorta, 2005). Y is a vector of country-level control variables including real GDP growth, inflation, sovereign spread20 and an index of housing prices. Finally, to isolate credit supply we include include bank and city-year fixed effects (αi and γct ) in order to absorb unobservable time-varying city-specific credit demand. For the purpose of our research, three testable hypotheses fall out of model (1) and depend on the sign and statistical significance of the beta coefficients. ∂∆ log Loansit = β1 + β2 BLCbanksit−1 + β3 BLCbanksit−1 × Stressedi ∂∆M Pt H.I The interest rate channel: β1 < 0 For economic theory, an interest rate cut is associated with higher bank lending by all banks. H.II The bank lending channel: β2 < 0 If the bank lending channel is active, an interest rate cut should have a larger expansionary effect on bank lending by banks selected by the criterion (BLC banks), ceteris paribus. H.III Heterogeneous bank lending channel: β3 significantly different from zero. If the bank lending channel is heterogeneous across stressed and non-stressed countries, the bank lending channel should be stronger in a country-group and weaker in the other.

4.3

Main results

This section presents our main results. We estimate model (1) with OLS and alternatively cluster standard errors at the bank- and country-level. Table 5 reports our main results. Looking at the estimates in the first column, monetary policy has the expected negative effect on lending growth for all Eurozone banks: a 1% interest rate cut leads to 1.4% increase in bank lending. Other estimates are also in line with standard results (e.g. Ehrmann et al. (2002); Gambacorta (2005)). First, bank lending appears to be cyclical, as higher real GDP growth and inflation rates are associated with higher bank lending rates and the housing price index is positively related to current lending, in line with the notion that swings in property prices are positively associated with new lending. Additionally, we find that small banks tend to lend out more, while bank capitalisation and liquidity are not significantly related to the flow of bank loans. Importantly, there is no evidence of a bank lending channel working for the Euro area as a whole, since the coefficient on the first interaction term (β2 ) is not statistically different from zero. [Table 5 about here] However, once we introduce the second interaction term in the second column (MP × BLC banks ×Periphery), the bank lending channel coefficient (β2 ) turns negative and statistically significant, while the coefficient on the second interaction term (β3 ) is positive and statistically significant. This finding suggests that the transmission of monetary policy to bank lending is weaker for banks located in stressed-countries as compared with banks located in non-stressed countries. Indeed, as the specification now discriminates between country-groups, the coefficient on the first interaction term represents the bank lending channel for banks located in non-stressed countries. Conversely, the coefficient on the second interaction directly implies that the bank lending channel was weaker for banks located in stressed-countries. The cumulative effects are reported at the bottom of table 5: on average, the same 1% policy rate cut leads to a 1.9% increase in lending by banks located in non-stressed countries and to a 0.9% increase for banks located in the periphery. In column 3 we control for credit demand by introducing city fixed effects and cluster at country level. Point estimates are slightly larger and qualitatively unchanged: the bank lending channel effect of monetary policy appears to be larger for banks in non-stressed countries as compared to their 20

A widely used measure of sovereign risk,we define spread as the difference between the German and the country’s equivalent 10-year government bond yield, annual averages (Albertazzi et al., 2014).

11

counterparts in non-stressed countries. Moreover, summing up coefficients reveals that a 1% drop in interest rates leads to a 2% increase in lending by banks located in non-stressed countries, while the effect for banks in non-stressed countries now is no longer significantly different from zero. Furthermore, our within-city specification implies that such heterogeneity cannot be ascribed to banks in stressed countries facing systematically lower local credit demand. If anything, controlling for local credit demand increases the point estimate for heterogeneity, suggesting that simple bank fixed effects estimates are likely to be conservative and underestimate the extent of heterogeneity between stressed and non-stressed countries. Finally, in the fourth column we saturate the model specification by including city-year effects.21 Since coefficients on monetary policy and country-variables cannot be estimated due to perfect collinearity with city-year fixed effects, we rely on city-year fixed effects to soak up all time varying shocks at the city level, including credit demand and lower lending opportunities. Our results are left qualitatively unchanged, although the point estimates are somewhat larger. Overall, these results contrast with the evidence on macro data found by Ciccarelli et al. (2013) while are consistent with the bank-level findings of De Santis (2015); De Santis and Surico (2013), although their evidence is based on country-by-country regressions for only four Eurozone countries (Germany, France, Italy and Spain). Next we consider alternative measures of monetary policy. We keep the same specification as that in the third column with bank and city fixed effects and keep clustering at the country-level. In the fifth column of table 5 we replace the euribor rate with the eonia rate to make sure our results are robust to the particular inter-bank rate adopted. In the sixth column we introduce the scaled size of the Eurosystem’s balance sheet as proxy for unconventional monetary policy and we interact it in the same way as done for the interest rate policy indicator. The effect of unconventional monetary policy on bank lending is not statistically significant, nor is any differential effect across countries recorded. In the seventh column we reinsert our conventional monetary policy indicator to capture the interplay between conventional and unconventional policy: we find the same results pointing to cross country heterogeneity, whereas again no significant coefficient is detected on unconventional monetary policy. While these findings seemingly point to the ineffectiveness of unconventional monetary policy on bank lending, we take this conclusion with a pinch of salt. Indeed, unlike the interest rate policy, the implementation of unconventional monetary policies policies such as LTROs, TLTROs and APP was capillary and had substantial cross-sectional variation across banks (see for instance Andrade et al. (2015)), in that different banks may have participated differently in different programs. Therefore, the absence of significance may be due to the low informational content of our country-level proxy. Finally, in column 7 we present results on the shadow rate. Because by construction the shadow rate is almost perfectly collinear with the Euribor rate,22 we simply include the shadow rate (in first differences) as the sole monetary policy variable. Results confirm our heterogeneity story: the coefficients are smaller but qualitatively similar to those of table 5: a 1% drop in the shadow rate increases lending growth by 1% for non-stressed countries banks while the effect is not statistically different from zero for banks located in stressed-countries.

4.4

Robustness

In this section we report a series of robustness checks. First, a common concern in the literature is that current bank lending may be related to its past realisations (Altunbas et al., 2009; Ehrmann et al., 2002; Gambacorta, 2005; Gambacorta and Marques-Ibanez, 2011; Gambacorta and Shin, 2016). In that case, failing to account for lending persistency may introduce omitted variable bias into our estimation. Furthermore, there may be reverse causality between our dependent variable (lending growth), bank controls (total assets, leverage and capital ratios) and our BLC-criterion itself. It is indeed plausible that changes in current and prospective lending may in fact drive changes in bank capital, liquidity and size. In this regard, lagging bank controls by one year may not be sufficient. To tackle these issues, we re-estimate model (1) with system GMM (Arellano and Bond, 1991; Blundell and Bond, 1998). This estimator is chosen to attenuate endogeneity concerns related to reverse causality between the dependent variable and bank specific characteristics, as well as to address dynamic panel bias, i.e. the fact that lagged values of the dependent variable are mathematically correlated with bank-fixed effects. System GMM estimator mitigates endogeneity issues by instrumenting suspected endogenous variables (∆ log Loans−1 , size, liquidity, capitalization and all other bank-specific variables included in vector 21 The interbank rates are perfectly collinear with year dummies, since they are the same for all countries and only vary over time. 22 The correlation between the two rates is 0.92

12

X ) with their lags or differenced lags.23 Exogenous variables (∆MP, real GDP growth, inflation, spread and the house prices index) are instrumented by themselves.24 This approach ensures consistency and efficiency provided that the instruments used are valid and the variables do not display auto-correlation of order two (Roodman, 2006). Specifically, the paper employs the system version of the estimator, because it tends to outperform the difference GMM estimator in terms of consistency and efficiency as it uses both the difference and the levels equation (Blundell and Bond, 1998). [Table 6 about here] We report our GMM estimates in column 1 of table 6. Lagged lending growth is indeed positive and significant, confirming the presence of some inertia in bank lending flows. Our main results are, however, left qualitatively unchanged, though coefficients on monetary policy and its interactions are larger. Next we turn to consider other observable bank-specific characteristics that may explain away some of the observed heterogeneity in monetary transmission across countries. For instance, belonging to a network is likely to make a bank considerably safer than standalone banks, regardless of its observable balance sheet features. For instance, Campello (2002) and Ashcraft (2006, 2008) document that banks affiliated with multi-bank holding companies are more likely to receive capital injections when distressed through access that the parent bank has on public markets. Banks affiliated with multi-bank holding also tend to have better access to funding markets, implying they can more easily smooth monetary policy and financial shocks on their balance sheets. In this case, the presence of bank networks bears on our identification of the bank lending channel: for instance, it is possible that our procedure assigns a given bank to the “small” category when in fact it is a subsidiary of a very large bank holding company, thus casting doubts on the information content of our criterion to identify the bank lending channel. However, as originally pointed out by Kashyap and Stein (1995), any misclassification will merely have the effect of making our estimates more conservative, in that the fact of belonging to a multi-bank holding company should make it harder to identify differences across banks’ characteristics that do in fact exist. To remain on the safe side we verify that controlling for a bank’s degree of independence leaves our results unchanged by including a categorical variable (Entitytype) that contains information on whether a bank is either independent, a single location, a local branch, a controlled subsidiary, or the global ultimate owner.25 Once again, results in column 2 of table 6 remain qualitatively unvaried: monetary policy - here again proxied by the Euribor rate - has a stronger effect on BLC-banks located in non-stressed countries, but the effects is lower for BLC-banks located in stressed countries. Finally, we take bank specialisation into account, as it has been recognised as to be an important source of variation across banks (De Santis and Surico, 2013) and to have important consequences for the way a bank react to monetary policy shocks.26 If banks with a particular specialisation were less sensitive to monetary policy and these kind of banks were to be prevalent in stressed countries, this may explain away some of the cross-country heterogeneity in monetary transmission. We thus introduce specialisation fixed effects to absorb all time-invariant differences in bank lending due to bank specialisation. As reported in column 3 of table 6, results are unchanged.

23

We use lags ranging from 2 to 9 years for each suspected endogenous variable. This approach ensures consistency and efficiency provided that the instruments used are valid and the variables do not display auto-correlation of order two (Roodman, 2006). 25 Specifically, the categories are defined as : branch location, i.e. a secondary location over which headquarters have legal responsibility; controlled subsidiary, i.e. a bank that is controlled by another entity; global ultimate owner (GUO), i.e. a bank which is the ultimate owner of a corporate group; independent, that is a company which is not a GUO but which could be GUO; and single location, i.e. a banks which has no ownership links to other banks (e.g. is neither a shareholder nor a subsidiary). 26 The main specialisation categories in our database are commercial, cooperative and savings banks. As reported by De Santis and Surico (2013), Orbis Bank Focus defines commercial banks as banks mainly active in retail banking (households and SMEs), wholesale banking (large firms) and private banking. Savings banks are similarly engaged in retail banking but belong to a banking group usually characterised by a decentralised distribution network which provides local outreach. Finally, cooperative banks have a cooperative ownership structure and are also engaged in retail banking. 24

13

Conclusion This paper asked whether monetary policy may affect bank lending heterogeneously across the Euro area through the bank lending channel. Results indicate that between 2007 and 2016 the bank lending channel was stronger for banks in non-stressed countries vis-`a-vis banks located in stressed countries. Unconventional monetary policy – as captured by the ECB shadow rate – is also similarly unevenly transmitted to bank lending across countries. The finding is robust to controlling for different macroeconomic and financial conditions across countries, balance sheets differences at the bank-level and differences in credit demand at the local level. The presence heterogeneity across national financial systems poses significant challenges to the conduct of monetary policy within a monetary union. Lowering barriers to the even transmission of monetary policy to bank credit across the Euro area is therefore crucial. In this respect, initiatives aimed at deepening financial and banking integration such as the Banking Union and the Capital Markets Union are likely to greatly reduce the extent of heterogeneity in monetary transmission. Lacking greater Euro-wide banking integration, tackling major sources of bank distress at the country-level may lessen barriers to the even transmission of monetary policy to bank lending across the monetary union. In this latter respect, further research should investigate which particular bank-specific factors are driving heterogeneity in monetary transmission across the Euro area.

References Acharya, V. V. and Steffen, S. (2015). The “greatest” carry trade ever? Understanding eurozone bank risks. Journal of Financial Economics, 115(2):215–236. Albertazzi, U., Nobili, A., and Signoretti, F. M. (2016). The bank lending channel of conventional and unconventional monetary policy. Banca d’Italia - Temi di discussione, (1094). Albertazzi, U., Ropele, T., Sene, G., Signoretti, and Federico M. (2014). The impact of the sovereign debt crisis on the activity of Italian banks. Journal of Banking & Finance, 46:387–402. Altavilla, C., Pagano, M., and Simonelli, S. (2017). Bank Exposures and Sovereign Stress Transmission. Review of Finance. Altunbas, Y., Fazylov, O., and Molyneux, P. (2002). Evidence on the bank lending channel in Europe. Journal of Banking & Finance, 26:2093–2110. Altunbas, Y., Gambacorta, L., and Marques-Ibanez, D. (2009). Securitisation and the bank lending channel. European Economic Review, 53(8):996–1009. Andrade, P., Cahn, C., Fraisse, H., and MMsonnier, J.-S. (2015). Can the Provision of Long-Term Liquidity Help to Avoid a Credit Crunch? Evidence from the Eurosystem’s LTROs. Banque de France Working Papers, (540). Arellano, M. and Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. The Review of Economic Studies, 58(2):277. Ashcraft, A. B. (2006). New Evidence on the Lending Channel. ournal of Money, Credit and Banking, 38(3):751–775. Ashcraft, A. B. (2008). Are Bank Holding Companies a Source of Strength to Their Banking Subsidiaries? Journal of Money, Credit and Banking, 40(2-3):273–294. Baele, L., Ferrando, A., H¨ordahl, P., Krylova, E., and Monnet, C. (2004). Measuring European Financial Integration. Oxford Review of Economic Policy, 20(4):509–530. Bernanke, B. S. and Blinder, A. S. (1988). Credit, Money, and Aggregate Demand. The American Economic Review, 78(2):435–439. Bernanke, B. S. and Gertler, M. (1995). Inside the Black Box: The Credit Channel of Monetary Policy Transmission. Journal of Economic Perspectives, 9(4):27–48.

14

Bignon, V., Breton, R., and Rojas Breu, M. (2013). Currency Union with and Without Banking Union. SSRN Electronic Journal. Blundell, R. and Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1):115–143. Borio, C. and Gambacorta, L. (2017). Monetary policy and bank lending in a low interest rate environment: Diminishing effectiveness? Journal of Macroeconomics, 21(0):29–0. Burriel, P. and Galesi, A. (2018). Uncovering the heterogeneous effects of ECB unconventional monetary policies across euro area countries. European Economic Review, 101:210–229. Butt, N., Churmz, R., Mcmahon, M., Morotzz, A., Schanz, J., Butt, N., Churm, R., and Morotz, A. (2015). QE and the Bank Lending Channel in the United Kingdom QE and the Bank Lending Channel in the United Kingdom. Bank of England - Working papers, (511). Campello, M. (2002). Internal Capital Markets in Financial Conglomerates: Evidence from Small Bank Responses to Monetary Policy. The Journal of Finance, 57(6):2773–2805. Cargoet, T. and Poutineau, J.-C. (2017). The Effects of an Homogenous Unconventional Monetary Policy on a Financially Heterogeneous Monetary Union. Working Paper. Cecchetti, S. (1999). Legal Structure, Financial Structure, and the Monetary Policy Transmission Mechanism. NBER Working Paper, (7157). Chodorow-Reich, G. (2014). The employment effects of credit market disruptions: Firm-level evidence from the 2008-9 financial crisis. Quarterly Journal of Economics. Ciccarelli, M., Maddaloni, A., and Peydr´o, J.-L. (2013). Heterogeneous transmission mechanism: monetary policy and financial fragility in the eurozone. Economic Policy, 28(75):459–512. Ciccarelli, M. and Rebucci, A. (2002). The Transmission Mechanism of European Monetary Policy: Is there Heterogeneity? Is it Changing Over Time? International Monetary Fund - Working paper, 54(2). Cœur´e, B. (2017). Convergence matters for monetary policy. In Speech given at the Competitiveness Research Network (CompNet) conference on ”Innovation, firm size, productivity and imbalances in the age of de-globalization”, Brussels. De Bondt, G. J. (1999). Banks and monetary transmission in Europe: Empirical evidence. Banca Nazionale del Lavoro Quarterly Review, 149:52–209. De Grauwe, P. (2013). Design Failures in the Eurozone: Can They Be Fixed? LEQS Paper, 57. De Grauwe, P. and Mongelli, F. P. (2005). Endogeneities of optimum currency areas: What brings countries sharing a single currency closer together? ECB Working paper, (468). De Haan, J., Inklaar, R., and Jong-A-Pin, R. (2008). Will business cycles in the euro area converge? A critical survey of empirical research. Journal of Economic Surveys. De Santis, R. A. (2015). Heterogeneity, monetary policy and banks’ balance sheets: evidence from Euro area countries. Applied Economics Letters, 22(7):549–557. De Santis, R. A. and Surico, P. (2013). Bank lending and monetary transmission in the euro area. Economic Policy, 28(75):423–457. Degryse, H. and Ongena, S. (2005). Distance, Lending Relationships, and Competition. The Journal of Finance, 60(1):231–266. Dornbusch, R., Favero, A. C., and Giavazzi, F. (1998). A Red Letter Day? CEPR Discussion Paper Series, (1804). Draghi, M. (2014). Monetary Policy Communication in Turbulent Times. In Speech at the Conference De Nederlandsche Bank, Amsterdam, 24 April.

15

Drechsler, I., Savov, A., and Schnabl, P. (2017). The Deposits Channel of Monetary Policy*. The Quarterly Journal of Economics, 132(4):1819–1876. Durr´e, A., Maddaloni, A., and Mongelli, F. P. (2014). The ECB’s Experience of Monetary Policy in a Financially Fragmented Euro Area. Comparative Economic Studies, 56(3):396–423. ECB (2017). Financial integration in Europe. Technical report, European Central Bank. Ehrmann, M., Gambacorta, L., Mart´ınez-Pag´es, J., Sevestre, P., and Worms, A. (2002). Financial Systems and the Role of Banks in Monetary Policy Transmission in the Euro Area. ECB Working Paper, 105. Ehrmann, M., Gambacorta, L., Martinez-Pag´es, J., Sevestre, P., Worms, A., Gambacorta Banca, L., Jorge, ˜ M.-P. (2003). The effects of monetary policy in the euro area. Oxford Review I., and Banco De Espana, of Economic Policy, 19(1):58–72. Fung´acˇ ov´a, Z., Solanko, L., and Weillb, L. (2014). Does competition influence the bank lending channel in the euro area? Journal of Banking & Finance, 49:356–366. Gambacorta, L. (2003). Asymmetric bank lending channels and ECB monetary policy. Economic Modelling, 20(1):25–46. Gambacorta, L. (2005). Inside the bank lending channel. European Economic Review, 49(7):1737–1759. Gambacorta, L. and Marques-Ibanez, D. (2011). The bank lending channel: lessons from the crisis. Economic Policy, 26(66):135–182. Gambacorta, R. L. and Shin, H. S. (2016). Why bank capital matters for monetary policy. Journal of Financial Intermediation, 15(0):0–1. Gilchrist, S. and Mojon, B. (2014). Credit risk in the euro area. The Economic Journal. Guiso, L., Kashyap, A. K., Panetta, F., and Terlizzese, D. (2000). Will a common european monetary policy have asymmetric effects? Temi di discussione - Banca d’Italia, (384). Hernando, I. and Mart´ınez-Pag´es, J. (2001). Is There a Bank Lending Channel of Monetary Policy in Spain? ECB - Working Paper, 99. Iyer, R., Peydr´o, J.-L., Da-Rocha-Lopes, S., and Schoar, A. (2014). Interbank Liquidity Crunch and the Firm Credit Crunch: Evidence from the 2007–2009 Crisis. Review of Financial Studies, 27(1):347–372. Jim´enez, G., Ongena, S., Peydr´o, J., and Saurina, J. (2014). Hazardous Times for Monetary Policy: What Do Twenty-Three Million Bank Loans Say About the Effects of Monetary Policy on Credit RiskTaking? Econometrica, 82(2):463–505. Jim´enez, G., Ongena, S., Peydr´o, J.-L., and Saurina, J. (2012). Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel with Loan Applications. American Economic Review, 102(5):2301–2326. Kashyap, A. K. and Stein, J. C. (1995). The impact of monetary policy on bank balance sheets*. CarnegieRochester Conference Series on Public Policy, 42:151–195. Kashyap, A. K. and Stein, J. C. (2000). What Do a Million Observations on Banks Say about the Transmission of Monetary Policy? The American Economic Review, 90(3):407–428. Khwaja, A. I. and Mian, A. (2008). Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market. American Economic Review, 98(4):1413–1442. Kishan, R. P. and Opiela, T. P. (2000). Bank Size, Bank Capital, and the Bank Lending Channel. Source Journal of Money, Credit and Banking, 32(1):121–141. Lane, R. (2012). The European Sovereign Debt Crisis. Journal of Economic Perspectives, 26(3):49–68. Lederer, P. J. and Hurter, A. P. (1986). Competition of Firms: Discriminatory Pricing and Location. Econometrica, 54(3):623.

16

Loupias, C., Savignac, F., and Sevestre, P. (2002). Is There a Bank Lending Channel in France? Evidence from Bank Panel Data. Banque de France - Working paper, 92. Mody, A. and Sandri, D. (2012). The eurozone crisis: how banks and sovereigns came to be joined at the hip. Economic Policy, 27(70):199–230. Ongena, S. and Smith, D. C. (2000). What Determines the Number of Bank Relationships? CrossCountry Evidence. Journal of Financial Intermediation, 9(1):26–56. Peek, J. and Rosengren, E. S. (2005). Unnatural selection: Perverse incentives and the misallocation of credit in Japan. American Economic Review, 95(4):1144–1166. Peydro, J.-L., Polo, A., and Sette, E. (2017). Monetary Policy at Work: Security and Credit Application Registers Evidence. CEPR Discussion Paper, (12011). Popov, A. A. and van Horen, N. (2013). The Impact of Sovereign Debt Exposure on Bank Lending: Evidence from the European Debt Crisis. De Nederlandsche Bank Working Paper, (Working paper No. 382). Rajan, R. G. and Zingales, L. (1998). Financial Dependence and Growth. The American Economic Review, 88(3):559–586. Roodman, D. (2006). How to do Xtabond2: An Introduction to Difference and System GMM in Stata. Center for Global Development, 103. Sussman, O. and Zeira, J. (1995). Banking and Development. Van den Heuvel, S. J. (2002). Does bank capital matter for monetary transmission? Economic Policy Review, 8(1):259–265. Wu, J. C. and Xia, F. D. (2016). Measuring the Macroeconomic Impact of Monetary Policy at the Zero Lower Bound. Journal of Money, Credit and Banking, 48(2-3):253–291.

17

Figures

-5

1000

0 interest rate (%)

ECB total assets (€ billions) 2000 3000 4000

5

5000

Appendix A

2004m1 2006m1 2008m1 2010m1 2012m1 2014m1 2016m1 2018m1 ECB total assets Shadow rate

Euribor rate,3m

Figure 1: ECB monetary stance

LHS: Eurosystem total assets, monthly average, EUR trillions; RHS: Euribor rate, 3-months, monthly average, percentage points; ECB’s shadow rate (Wu and Xia, 2016). Source: ECB (2017),Jing Cynthia Wu’s webpage (link

Bank lending growth to NFCs

-10

Loan growth 0 10

20

Stressed vs non-stressed countries

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 stressed

non-stressed

Figure 2: Bank lending growth to NFCs, stressed vs non-stressed countries.

Growth rate of adjusted loans to euro area Non-financial Corporations (NFCs) by country-group, year-on-year growth, monthly frequency, percentage points. Source: ECB (2017).

Real interest rates on new lending to NFCs

0

1

Interest rate (%) 2 3

4

5

stressed vs non-stressed countries

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 stressed

non-stressed

Figure 3: Real interest rates on new lending to NFCs, stressed vs non-stressed countries.

Average interest rates on lending for new business for NFCs by country-group. New business is defined as any new agreement between a household or a non-financial corporation and a bank. Inflation-adjusted, monthly frequency, percentage points. Source: ECB (2017).

18

Loan growth and heterogeneity

0

Standard deviation .1 .05

.15

(standard deviation)

2004q1

2007q3

2011q1

Real GDP growth

2014q3

2018q1

Loan growth

Figure 4: Loan growth and heterogeneity

The graph plots the standard deviation across countries of the year-on-year loan growth rate of adjusted loans by Eurozone Main Financial Institutions (MFIs) to domestic Non Financial Corporations (NFCs) against the standard deviation across countries of real GDP growth, year-onyear. Source: ECB (2017) and Eurostat (2017)

Demand for credit by NFCs Net % of banks reporting increasing demand -40 -20 0 20 40

stressed vs non-stressed countries

2005

2006

2008

2009

2011

stressed

2012

2014

2015

2017

2018

non-stressed

Figure 5: Demand for credit by NFCs, stressed vs non-stressed countries

The graph plots the difference between the share of surveyed banks that reported an increase in credit demand by NFCs over the past three months minus the share of surveyed banks that reported a reduction in credit demand by NFcs over the past three months. Source: ECB (2017).

Figure 6: Lending growth in the cross section

The graph plots annual lending growth by BLC-banks and other banks and sorts them by stressed and non-stressed countries of location. Source: Fitchconnect (2017).

19

Figure 7: Government securities on banks’ balance sheets

The graph plots the annual share of government securities held by BLC-banks and other banks over total assets and sorts them by stressed and non-stressed countries of location. Source: Fitchconnect (2017).

Figure 8: Non-performing loans on banks’ balance sheets

The graph plots the annual share of non-performing loans held by BLC-banks and other banks over total loans and sorts them by stressed and non-stressed countries of location. Source: Fitchconnect (2017).

20

Appendix B

Tables

Table 1: Sample description and representativeness For each country the table reports the number of banks and the aggregate values of loans as percentage of total aggregate data reported in the Balance Sheet Indicator statistics of the ECB, time averaged over 2007-2016. Stressed countries

N°of banks

Cyprus Greece Ireland Italy Latvia Lithuania Portugal Slovenia Spain

14 9 21 583 8 2 107 10 96

Total banks Average Weighted average All countries, average All countries, Weighted average

850

2544

Loans as % of BSI totals 15 3 10 24 7 21 15 27 8

Non-stressed countries

N°of banks

Austria Belgium Estonia Finland France Germany Luxembourg Malta Netherlands Slovakia

195 26 2 38 212 1101 79 9 15 17 1694

14 16 16 26

Loans as % of BSI totals 17 6 0,04 5 16 55 23 11 7 34 17 31

Table 2: Bank lending channel selection criterion This table summarises the three thresholds used to define the criterion to select banks most likely to be sensitive to the bank lending channel of monetary policy. Bank-specific characteristics

Thresholds

Average number of banks

Other banks

Small

Log assets ≤ 95 pct

2415

124

Undercapitalised

Capital regulatory distance ≤25 pct

436

2108

Illiquid

Liquidity ratio ≤75 pct

1859

685

BLC banks

Small, illiquid and undercapitalized

468

2076

21

Table 3: Selection bias – comparing BLC-banks with full sample BLC banks

Full sample

Countries

(1) Number of banks, BLC subsample

(2) Share of BLC subsample (%)

(3) Number of banks, full sample

(4) Share of full sample (%)

Stressed countries Cyprus Greece Ireland Italy Latvia Lithuania Portugal Slovenia Spain Non-stressed countries Austria Belgium Estonia Finland France Germany Luxembourg Malta Netherlands Slovakia Total

110 4 0 2 75 1 2 4 4 18 358 35 6 0 1 52 231 29 0 1 3 468

23,5 0,9 0,0 0,4 16,0 0,2 0,4 0,9 0,9 3,8 76,5 7,5 1,3 0,0 0,2 11,1 49,4 6,2 0,0 0,2 0,6 100

850 14 9 21 583 8 2 107 10 96 1694 195 26 2 38 212 1101 79 9 15 17 2544

33,4 0,6 0,4 0,8 22,9 0,3 0,1 4,2 0,4 3,8 66,6 7,7 1,0 0,1 1,5 8,3 43,3 3,1 0,4 0,6 0,7 100

Comparison (5) Over/under representation (=2-3) -9,9 0,3 -0,4 -0,4 -6,9 -0,1 0,3 -3,4 0,5 0,1 9,9 -0,2 0,3 -0,1 -1,3 2,8 6,1 3,1 -0,4 -0,4 0,0 100

Table 4: Descriptive statistics Descriptive statistics (mean values) Total assets Equity/assets Loans/assets NPL/loans Liquidity ratio Loans growth Securities/assets Government securities/assets

All countries All banks 1,973 0.1 0.59 0.08 0.69 0.04 0.23 0.08

BLC banks 919 0.07 0.66 0.07 0.03 0.06 0.21 0.07

Stressed countries

Other banks 2,048 0.1 0.58 0.08 0.75 0.04 0.23 0.08

22

All banks 1,748 0.11 0.63 0.11 1.01 0.03 0.2 0.18

BLC banks 743 0.08 0.71 0.1 0.03 0.037 0.16 0.12

Other banks 1,852 0.12 0.62 0.11 1.12 0.035 0.2 0.19

Non-stressed countries All banks 2,075 0.09 0.57 0.04 0.52 0.04 0.25 0.03

BLC banks 1,056 0.06 0.62 0.05 0.02 0.051 0.24 0.02

Other banks 2,133 0.09 0.57 0.04 0.55 0.04 0.25 0.03

Table 5: Main results Dep. variable: log ∆Loans ∆MP ∆MP×BLC-banks−1 ∆MP×BLC-banks−1 × Stressed Log total assets−1 Leverage ratio−1 Liquidity ratio ratio−1 Loans/total assets−1 Deposits/total liabilities−1 Real GDP growth Inflation Sovereign spread House prices index UMP

(1) (2) (3) (4) (5) MP=Euribor MP=Euribor MP=Euribor MP=Euribor MP=Eonia -1.401*** (0.228) -0.240 (0.214) -0.196*** (0.0161) 0.179 (0.118) 0.000370 (0.00144) -0.520*** (0.0344) 0.0324 (0.0407) 0.546*** (0.0850) 0.551*** (0.187) -0.352** (0.146) 0.00366*** (0.000259)

-1.427*** (0.230) -0.457* (0.250) 1.000** (0.402) -0.195*** (0.0161) 0.181 (0.118) 0.000383 (0.00144) -0.519*** (0.0344) 0.0315 (0.0407) 0.557*** (0.0861) 0.560*** (0.187) -0.354** (0.147) 0.00368*** (0.000259)

-1.577** (0.565) -0.437** (0.177) 1.415*** (0.408) -0.211*** (0.0155) 0.0595 (0.159) 0.00185 (0.00141) -0.520*** (0.0725) 0.0339 (0.0855) 0.690*** (0.155) 0.289 (0.175) -0.210 (0.234) 0.00417*** (0.000329)

-1.884*** (0.332) -0.884** (0.357)

-2.014*** (0.589) -0.599 (0.381)

11,918 0.447

8,793 0.439

-0.654*** (0.156) 2.392*** (0.555) -0.142*** (0.0441) 0.314** (0.130) 0.00172 (0.00832) -0.509*** (0.0719) 0.0529 (0.210)

-1.584* (0.754) -0.423** (0.190) 1.356** (0.489) -0.211*** (0.0172) 0.0478 (0.170) 0.00189 (0.00139) -0.525*** (0.0695) 0.0338 (0.0878) 0.663*** (0.204) 0.126 (0.196) -0.0816 (0.268) 0.00426*** (0.000324)

UMP×BLC-banks−1 UMP×BLC-banks−1 ×Stressed ∂log∆Loans ∂∆MP

for non-stressed countries

∂log∆Loans ∂∆MP

for stressed countries

Observations R-squared Bank FE City FE Year FE Cluster SE

11,918 0.446 Yes No No

2,488 0.660

8,793 0.437

(7) (8) MP=Euribor MP=Shadow rate

-1.691** (0.612) -0.461* (0.235) 1.549*** (0.401) -0.220*** -0.202*** (0.0130) (0.0146) 0.0959 0.196 (0.169) (0.142) 0.00179 0.00164 (0.00150) (0.00146) -0.497*** -0.488*** (0.0562) (0.0623) 0.00721 0.0331 (0.0928) (0.0933) 0.196*** 0.760*** (0.0566) (0.178) -0.0915 0.574** (0.362) (0.257) -1.086*** -1.009*** (0.261) (0.267) 0.00413*** 0.00423*** (0.000634) (0.000644) 0.0784 0.0847 (0.0870) (0.0848) 0.313 0.257 (0.814) (0.848) -0.0312 0.0243 (0.806) (0.842)

-2.007* (0.831) -0.651 (0.581)

Yes Yes Yes Yes No Yes Yes Yes No No Yes No Bank Country Country Country Robust standard errors in parentheses *** p