# Credit Supply During a Sovereign Debt Crisis

Credit Supply During a Sovereign Debt Crisis Abstract We study the patterns of credit supply in Italy following the burst of the European sovereign debt crisis in 2011. Comparing lending to the same firm, we find that domestic banks reduced credit supply, increased interest rates on credit granted, and lowered the probability of accepting loan applications more than foreign banks, which were less affected by the sovereign crisis. The credit contraction is the consequence of a largely country-specific effect, not explained by heterogeneity in bank characteristics, but associated to a generalized increase in the cost of funding of Italian banks. Looking across firms, we find that credit restrictions by domestic banks were not fully compensated by foreign banks’ lending, implying that Italian firms experienced an aggregate credit shortage. 1. Introduction The sovereign debt crisis that hit European countries in the summer of 2011 had a large effect on global economic performance and on the stability of financial markets. Economic activity in the countries affected by the crisis contracted strongly. So did bank credit. Besides lower demand, the drop in credit may have also been driven by supply factors. A credit crunch occurring during a sovereign debt crisis can be especially dangerous because it can trigger or amplify a contraction in GDP, thus fueling a negative loop between the sovereign crisis and economic activity. This threat is aggravated by the fact that, in this context, governments need to tighten fiscal policy to counteract the tensions, and the effectiveness of monetary policy might be blunted because rising sovereign yields may impair the transmission mechanism. Dell’Ariccia et al. (2013) show that, following the burst of the sovereign debt crisis, changes in the policy rate in the Euro Area had little effects on bank funding costs as the latter were increasingly driven by domestic sovereign yields. Assessing the degree to which credit supply contracts during a sovereign crisis is therefore particularly relevant. Credible identification however necessitates purging observed credit flows from demand components, and observing a subset of banks whose lending policy was plausibly not exposed to the shock. We undertake this task exploiting a rich data set of individual bank–firm relationships from the comprehensive Italian Credit Register (CR), which records lending from all banks, both domestic and foreign, operating in Italy. We first document that after the sovereign debt crisis started, Italian banks contracted credit supply about 3 percentage points more than foreign banks operating in Italy, but headquartered in countries not (or less) exposed to the sovereign crisis. Importantly, this is obtained comparing credit flows from different banks to the same firm, thus accounting for credit demand. Similar patterns emerge looking at other lending margins: domestic banks increased interest rates and reduced the probability of accepting new applications more than their foreign competitors. Second, we show that this domestic bank effect on credit granted persists even after controlling for bank individual characteristics and provide evidence that this effect reflects the increased cost of funding experienced by all domestic banks. Indeed we find that country-level measures of banks’ cost of funding have a negative effect on credit supply and their impact persists also when balance sheet features are included in the model. Moreover, being a domestic bank has a strong power in explaining the increase in the cost of funding at the bank level, again even after accounting for individual bank characteristics. We interpret the effect of this country-specific component as the impact of the sovereign debt crisis, potentially uncovering a new lending channel linking sovereign shocks to contractions to credit supply, through banks’ cost of funding. Recent evidence confirms that sovereign crises largely translate into an increase in the cost of funding for domestic intermediaries (Battistini, Pagano, and Simonelli 2014; Panetta, Angelini, and Grande 2014). This may be due to the increased riskiness of banks’ assets, which are exposed to the shock both in terms of holdings of domestic sovereign bonds and in terms of loans to domestic private sector. Moreover, the cost of funding may also increase due to the lower ability of the domestic government to support ailing intermediaries. By the same argument, the balance sheets of banks headquartered in countries that are less exposed to the sovereign crisis are less likely to respond to the shock. Furthermore, for domestic banks, access to funds might become more costly also due to the mechanical downgrades of financial institutions following the downgrade of their sovereign of residence (so-called sovereign ceiling), as highlighted in Adelino and Ferreira (2016). All these factors would imply that foreign banks may represent a suitable comparison group for estimation purposes. We find consistent evidence in the data. As shown by Figure 1, the correlation between the riskiness of the sovereign and the riskiness of the banking sector increased sharply during the sovereign debt crisis relative to the Lehman crisis (Panel A); by the same token, the association between bank retail funding costs and the premium requested on sovereign debt strongly intensified (Panel B). This strong nexus between country level sovereign risk and the cost of raising funds supports the idea of using foreign banks as a subset of intermediaries more insulated from the crisis. Figure 1. View largeDownload slide Sovereign CDS, banks CDS, and banks’ cost of funding. Figure 1(a) shows scatter plots of country average CDS spreads for 5-year senior debt of major banks (vertical axis) and CDS spreads 10-year sovereign bonds of the home country (horizontal axis) based on daily data for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data are from Thomson Reuters, banks are the major banks of each country with available CDS data for the whole sample. Unit: basis points. Figure 1(b) shows scatter plots of country spread between the yields of 10-year government bonds relative to the German (vertical axis) and the country averages of the spread between rates on new deposits by firms and households with agreed maturities of up to 1 year and the 6 month average of the EONIA rate (horizontal axis), for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data on sovereign spread are from Bloomberg, deposit rates are from ECB MFI interest rates statistics. Unit: basis point for the sovereign spread, percentage points for the deposit rate. Figure 1. View largeDownload slide Sovereign CDS, banks CDS, and banks’ cost of funding. Figure 1(a) shows scatter plots of country average CDS spreads for 5-year senior debt of major banks (vertical axis) and CDS spreads 10-year sovereign bonds of the home country (horizontal axis) based on daily data for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data are from Thomson Reuters, banks are the major banks of each country with available CDS data for the whole sample. Unit: basis points. Figure 1(b) shows scatter plots of country spread between the yields of 10-year government bonds relative to the German (vertical axis) and the country averages of the spread between rates on new deposits by firms and households with agreed maturities of up to 1 year and the 6 month average of the EONIA rate (horizontal axis), for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data on sovereign spread are from Bloomberg, deposit rates are from ECB MFI interest rates statistics. Unit: basis point for the sovereign spread, percentage points for the deposit rate. Sovereign debt crises rarely occur as stand-alone events. They are often accompanied by recessions, when demand for credit drops, and the quality of borrowers deteriorates, posing the task of disentangling demand from supply effects. We tackle these identification challenges using a rich data set of individual bank–firm relationships from the comprehensive Italian CR. This allows us to employ the identification strategy pioneered by Khwaja and Mian (2008) and to restrict our analysis to firms borrowing from at least one foreign and one domestic bank and include firm-time fixed effects to absorb all time-varying observed and unobserved firm heterogeneity. In principle, sovereign debt crises might be triggered by government-financed measures aimed at sustaining distressed banks. The subsequent weakening of public finances may in turn feed back into worsening of banks’ balance sheets (Acharya, Drechsler, and Schnabl 2014; Reinhart and Rogoff 2009). We argue that this was not the case in Italy: the total cumulative government support to the banking sector relative to GDP was negligible between 2008 and 2013 (ECB 2015). Moreover, the inclusion of firm-time and bank fixed effects attenuates the potential biases induced by a feedback from past lending policies of banks, undermining their health and thus triggering the shock to the sovereign. We also extend our main findings into three directions. First, we test if the behavior of foreign banks differs depending on whether they are incorporated as subsidiaries or as branches. We find that the differential behavior between Italian and foreign banks is due to subsidiaries, whose business model is more similar to that of Italian banks, thus reducing concerns that our findings are driven by foreign banks specializing in particular markets, such as those of Mergers and Acquisitions (M&As) or syndicated loans. Second, we look at the heterogeneity of the drop of credit across firms. We find some limited evidence that the difference in credit by domestic and foreign banks was larger for firms with less liquid buffers, and for firms with a worse credit rating. Finally, we estimate the aggregate effect of the sovereign crisis on credit supply, finding that Italian firms were unable to compensate the negative domestic bank effect tapping on credit from foreign banks. The paper is structured as follows: the next section summarizes the related literature, Section 3 describes the empirical strategy, Section 4 presents the data set and the main descriptive statistics, Section 5 contains the results of our baseline specification, their interpretation, and a set of robustness checks, Section 6 examines the effect of bank heterogeneity, Section 7 contains extensions, Section 8 presents the results on the aggregate effect, Section 9 shows results on the cost of credit, and on the probability that loans applications are accepted, Section 10 concludes. 2. Related Literature Our paper contributes to the literature studying the effects of sovereign debt crises on bank activity and real outcomes. Earlier works look at this issue using cross-country panels of different crises episodes over long periods of time, with data at the country or industry level (Arteta and Hale 2008; Borensztein and Panizza 2009). More recent papers have studied specifically the European sovereign crisis, looking at different aspects of the nexus crisis-credit-output: the effects of holdings of Greece, Ireland, Italy, Portugal and Spain (GIIPS) sovereign debt on syndicated loans by non-GIIPS banks (Popov and Van Horen 2015), the correlation between holdings of domestic sovereign bonds and credit supply (Gennaioli, Martin, and Rossi 2014), the transmission of downgrades of sovereign bonds to bank bonds and to credit (Adelino and Ferreira 2016), and the direct effects of sovereign ratings on firm outcomes, net of credit risk and fundamentals (Almeida et al. 2017). Other work focused on asset–liability management of banks during the sovereign crisis, documenting phenomena of “carry trade”, with banks borrowing short on wholesale markets and going long on GIIPS countries sovereign bonds (Acharya and Steffen 2015) and showing how U.S. branches of European banks experienced a run on their deposits, mainly from the U.S. money market funds (Correa, Sapriza, and Zlate 2012). Our paper contributes to this literature in several ways. We document that Italian banks suffered a generalized funding shock associated to the sovereign crisis and that this translated into a severe tightening of credit supply, compared with credit by foreign banks that did not suffer an equal increase in their domestic sovereign risk. The availability of a large sample of loans to nonfinancial firms from the Italian CR allows us to identify the net supply effect, in line with the bank lending channel literature pioneered by Khwaja and Mian (2008) based on loan-level data. Furthermore, because our sample contains both large and small firms, we provide a much broader picture than that obtained using syndicated loan data or information from public corporations. Moreover, we provide evidence on different margins of lending and finally estimate the real effects of the sovereign lending channel by showing that the sovereign shock had an aggregate impact on firms’ access to bank credit. Our findings complement the work by Bocola (2016) who develops a business cycle model estimated on Italian data, showing that a sovereign shock translates into higher funding costs for banks, and then into lower lending, with strong recessionary effects. Because our identification relies on comparing behavior of Italian and foreign banks, our paper also contributes to the large literature on global banks and on the international transmission of shocks. Many channels have been investigated: cross-border lending (Peek and Rosengren 1997), lending by affiliates (Peek and Rosengren 2000), internal capital markets (Cetorelli and Goldberg 2012a), and international interbank markets (Schnabl 2012). Our work concentrates on lending by foreign branches and subsidiaries operating in Italy. More recently, there has been a large interest on the specific role of global banks in the financial crisis and the Great Recession. Various papers have examined different destination markets: Cetorelli and Goldberg (2011) look at emerging economies, Cetorelli and Goldberg (2012b) at the United States, Popov and Udell (2012) at Central and Eastern Europe countries, Albertazzi and Bottero (2014) at Italy, and De Haas and Van Horen (2012) explore the international market of syndicated loans. Overall, this recent literature documents how global intermediaries contribute to “export” tensions, thus highlighting a mechanism of international transmission of shocks from the parent bank to host countries. Notably, we shed light on a “bright” side of the presence of foreign intermediaries in a country hit by a shock, showing how, being more shielded by the shock than domestic banks, foreign banks contribute to dampen the sovereign shock. Interestingly, our results can also be interpreted in line of the finding of Abbassi et al. (2016) who show that trading banks (mainly large global banks) increase their investments in risky securities to profit from fire-sales during the crisis and cut back on other activities, including credit supply to domestic firms. In the same fashion, Italian firms represent a risky investment because the country is hit by a large shock, and foreign banks could have stepped in to extract rents from credit-constrained Italian firms. This behavior could exert some negative spillovers on domestic activities of foreign banks, insofar the latter are funding constrained. In such case, the bright side of the presence of global banks for Italian firms would be accompanied by negative side effects in other markets. 3. Empirical Strategy Determining whether a credit crunch occurs during a sovereign shock poses important challenges. A first one is the identification of a set of banks more and less exposed to the shock. We argue that foreign banks operating in Italy are a good candidate to represent a set of intermediaries relatively less affected by sovereign tensions. There are various reasons why domestic banks of a country hit by a sovereign shock should be affected more severely than foreign ones operating in the same country (Battistini, Pagano, and Simonelli 2014; Panetta, Angelini, and Grande 2014). The asset side of banks located in a country hit by a shock drops in value and becomes riskier. Sovereign bonds held in the trading book are marked to market and generate an immediate loss to banks. Even if banks do not hold government debt in their trading book, they may nevertheless expect to incur future losses. Moreover, the credit risk of loans to domestic customers also rises (Bocola 2016). The increased riskiness of the asset side exerts two effects. The first one is to reduce banks’ willingness to assume additional risk. The second one, and possibly much larger than the first, operates via a higher cost of funding or even impaired access to funding. Funding cost rises also because collateralized transactions backed by government debt become more expensive, due to higher haircuts required on the collateral. Furthermore, the potential weakening of the implicit or explicit government guarantee on banks due to strains in public finances likely determines an increase in the cost of raising funds in the country. These factors affect all banks exposed to a country hit by a sovereign shock, yet typically the impact is much higher for those headquartered in that country. Domestic banks, in fact, typically hold a much larger share of their home-country sovereign securities and lend mostly to domestic borrowers. The home bias of investors, not only of banks, is a long-standing puzzle in the macro-financial literature (Coeurdacier and Rey 2013). In the case of banks, a high exposure to domestic lenders is a physiological structural characteristic due to the proximity needed in the lending process. The open question, which lies beyond the scope of this paper, is why banks do not diversify the exposure to domestic firms and households by holding securities issued by foreign entities. We take this behavior as given, because it does not compromise our identification strategy. Furthermore, another mechanism that generates an increase in the cost of funding for domestic banks only is the mechanical transmission of downgrades from sovereign to bank bonds (Adelino and Ferreira 2016). As a consequence of all these channels, domestic banks suffer from an increase in the risk of their home country government debt way more than foreign banks operating in the same country. This mostly translates into a higher cost of funding for domestic banks. We find evidence consistent with these phenomena in the data. Data on Credit Default Swaps (CDS) spreads confirm a high correlation between banks’ wholesale funding costs and sovereign CDS in the second half of 2011, following the outburst of the sovereign crisis (Figure 1, Panel (a)). In the same period also the cost of deposits, a proxy of retail cost of funding, was strongly correlated with the sovereign bonds yields (Figure 1, Panel (b)). Interestingly, compared to the Lehman crisis, the degree of correlation between measures of banks’ and sovereigns’ funding costs was much more pronounced. This suggests that a common country component may be a key driver of the higher cost of funding, possibly in addition to individual bank characteristics. Notably, foreign banks operating in Italy are headquartered in countries whose sovereign yields increased mildly (or even decreased) during the crisis (France, Germany, USA, Austria). This allows us to distinguish banks into two groups as being more and less exposed to the increase in sovereign risk: domestic (Italian) and foreign banks. There is also prima facie evidence that foreign and Italian banks behaved differently, in terms of lending policies, after the Italian sovereign risk for Italy rose in June 2011. Figure 2 shows the abrupt increase in the spread between the 10-year-Italian-government bond, and the 10-year-German-government bond. Descriptive statistics show that the variation in credit in the 6 months between December 2010 and mid-2011 and the variation observed between June and December 2011 was basically unchanged for foreign banks, although it varied remarkably for Italian banks. On average, lending by domestic banks decreased by more than three percentage points to -7% in those 6 months, as shown in Table 1; foreign banks kept lending almost steadily at -5% in the two periods. Also, graphical evidence reassures that, before the crisis, Italian banks did not display a differential trend in credit supply relative to foreign banks. If domestic banks prior to the crisis had already been in distress, their deleveraging might have started earlier than June 2011. Figure 3 shows the 6-month change in the log credit committed by Italian and foreign banks. Although prior to the crisis the two series moved similarly, since June 2011 credit from domestic banks decreased at a much faster rate than credit from foreign banks. Figure 2. View largeDownload slide Spread between 10-year Italian BTP and German Bund (percentage points). The figure shows the time series of the spread between the 10-year Italian BTP and the 10-year German Bund (percentage points). Data are from Thomson Reuters. The figure shows that the spread remained roughly constant in the first half of 2011, and started increasing abruptly, reaching historically high levels, from July 2011. Figure 2. View largeDownload slide Spread between 10-year Italian BTP and German Bund (percentage points). The figure shows the time series of the spread between the 10-year Italian BTP and the 10-year German Bund (percentage points). Data are from Thomson Reuters. The figure shows that the spread remained roughly constant in the first half of 2011, and started increasing abruptly, reaching historically high levels, from July 2011. Figure 3. View largeDownload slide Change of credit committed by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks. The unit is logpoints. Data are from the Italian CR. Figure 3. View largeDownload slide Change of credit committed by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks. The unit is logpoints. Data are from the Italian CR. Table 1. Descriptive statistics of main dependent variable. Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Notes: Delta log credit is the change in the log of credit committed to firms by banks (multiplied by 100). The change for the pre-crisis period is computed as the change between end of June 2011 and end of December 2010; the change for the pre-crisis period is computed as the change between end of December 2011 and end of June 2011. Data are from the Italian CR. View Large Table 1. Descriptive statistics of main dependent variable. Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Notes: Delta log credit is the change in the log of credit committed to firms by banks (multiplied by 100). The change for the pre-crisis period is computed as the change between end of June 2011 and end of December 2010; the change for the pre-crisis period is computed as the change between end of December 2011 and end of June 2011. Data are from the Italian CR. View Large A potential concern for this identification approach is that a sovereign shock may be driven by poor banks’ conditions due to excessive lending to weak borrowers prior to the crisis. Sovereign debt crises can indeed be fueled by banking crises, because governments disburse vast amounts of money to rescue troubled intermediaries (Reinhart and Rogoff 2009; Acharya et al. 2014). However, Italian banks weathered the post-Lehman crisis relatively unscathed (IMF 2010 Article IV consultation on Italy) and as a consequence the cumulative government support to the banking sector, between 2008 and 2013 has been a mere 0.2% of GDP (ECB 2015). Although the abrupt increase in sovereign spreads occurring at the end of June 2011 reflects many factors, including potential fears of the inability of rescuing the banking sector if needed, it can be reasonably argued that it did not reflect past or contemporaneous weaknesses of the banking sector driven by their past lending to nonfinancial firms, which is what is needed to support our identification approach. The second key identification challenge is the need to disentangle credit supply from demand. Typically sovereign tensions are accompanied by deteriorating economic conditions, inducing firms to scale down their investment plans and decrease demand for credit. Moreover, domestic banks may lend to a different set of firms (e.g. firms with weaker balance sheets, riskier firms, etc.) than foreign banks. Hence, it is critical to properly control for firm level demand for credit, for firms’ riskiness, and, more generally, for firm unobserved heterogeneity. For example, during economic downturns credit quality of existing loans deteriorates and this feeds back onto credit demand. The richness of our data set from the Credit Register allows us to control for all of these factors. We identify the impact of sovereign risk on credit supply by comparing the pre-crisis and the crisis patterns of credit supplied to the same firm by at least two banks, affected by the sovereign debt crisis to different degrees.1 The inclusion of firm-period fixed effects in all regressions, similarly to what Khwaja and Mian (2008) and more recently Jiménez et al. (2012) do, enables us to control for all firm-level unobserved heterogeneity that affects the dynamics of credit committed and of its cost in each period. Moreover, it controls for potential biases arising from the possibility that the crisis was triggered by fears about the soundness of corporate borrowers, weakening potential concerns about causality running from lenders or borrowers to sovereign yields. Following this empirical strategy, we estimate a model in which the observational unit is a credit relationship between a firm and a bank, and we compare two periods: the first half of 2011 (pre-crisis) and the second half of 2011 (crisis). Using a pre-crisis period allows us to control for existing differences in the supply of credit by Italian and foreign banks. Formally, $$\Delta \mathit{credit}_{i,j,t} =\beta _{1} \mathit{FOREIGN}_{j}+\beta _{2} \mathit{FOREIGN}_{j}\ast \mathit{CRISIS}_{t}+\alpha _{i,t}+\varepsilon _{i,j,t}$$ (1) where $$\Delta \mathit{credit}_{i,j,t}$$ is the difference in the log credit committed by bank j to firm i in period t. The dummy FOREIGN equals 1 if bank j is foreign, zero if the bank is domestic. The term FOREIGN*CRISIS is an interaction between the dummy FOREIGN and the dummy variable CRISIS which equals 1 in the second half of 2011. αi, t is a full set of firm-period fixed effects (they absorb the dummy CRISIS, which therefore does not appear in the equation above). We also run all regressions plugging bank fixed effects, which control for all bank time-invariant unobserved heterogeneity, including systematic differences in banks’ business models, geographical reach, etc.2 All regressions also include variables capturing relationship-level characteristics. The first is the share of total credit to firm i supplied by bank j (SHARE). On the one hand, SHARE measures the exposure of bank j towards firm i, and this is negatively correlated with loan growth; on the other hand, SHARE may be a proxy of the strength of the bank–firm relationship, therefore suggesting a positive relationship with credit growth. The second variable is the share of drawn over credit committed by bank j to firm i (DRAWN/COMMITTED). This control measures how intensively available credit lines are used. If a line of credit is already used close to the available limit, firms may be more likely to apply for an extension. The third variable is the share of overdraft over total committed credit by bank j to firm i (OVERDRAFT). Overdraft loans (revolving credit lines) may be more volatile than other forms of loans. Including these regressors helps controlling for differences in bank–firm relationships across domestic and foreign banks. We also include the interactions between these relationship-level controls and the dummy CRISIS. The main parameter of interest is β2, which captures the differential behavior of foreign banks relative to domestic banks during the crisis. It is identified on firms that borrow from at least one Italian and one foreign bank in at least one period.3 The key identifying assumption is that firms do not have a bank-specific demand for credit (Bonaccorsi di Patti and Sette 2016, Jimenez et al. 2010, Khwaja and Mian 2008). To reduce concerns that this assumption may be violated in our setting, we include a set of relationship-level controls (SHARE, DRAWN/COMMITTED, OVERDRAFT) that capture systemic differences in the structure of bank–firm relationships. Moreover, our findings on lending from branches and subsidiaries (Section 7.1) and those on the cost of credit (Section 9.1) contribute to attenuate possible concerns about the violation of this assumption. Our identification allows us to check more accurately whether Italian and foreign banks behaved similarly before the crisis; because the series shown in Figure 3 do not account for the different composition of firms borrowing from the two types of banks, we also compute the dynamics of credit committed in deviation from firm-period averages. We expect credit from domestic and foreign banks, net of firm effects, to move similarly until June 2011, and to start diverging afterwards. This is precisely what happens, as shown in Figure 4 (see Khwaja and Mian 2008 for a similar representation of the data). This is the graphical counterpart of the baseline model (equation 1 above). Figure 4. View largeDownload slide Change in credit committed, net of firms-period effects, by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks, and net of their means. For every firm, the average firm-level credit committed is subtracted by credit committed by Italian and foreign banks. The stocks of demeaned credit committed are then collapsed across all firms. Then changes relative to the June 2011-level (start of the crisis) are computed. The unit is percentage points. Data are from the Italian CR. Figure 4. View largeDownload slide Change in credit committed, net of firms-period effects, by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks, and net of their means. For every firm, the average firm-level credit committed is subtracted by credit committed by Italian and foreign banks. The stocks of demeaned credit committed are then collapsed across all firms. Then changes relative to the June 2011-level (start of the crisis) are computed. The unit is percentage points. Data are from the Italian CR. 4. Data and Descriptive Statistics Our empirical analysis is based on data at the bank–firm relationship level on outstanding loan amounts from the comprehensive Italian CR.4 The data set includes both granted and drawn credit. We focus on credit granted (credit commitments), because this better captures a decision of banks to supply credit. Drawn credit is influenced by the decision of the borrower to use available lines, which is largely affected by demand. December 2010–June 2011 represents the pre-crisis period, June 2011–December 2011 represents the crisis period. We do not extend the sample beyond December 2011, because on December 22 the ECB enacted the first 3-year Long-Term Refinancing Operation, which eased tensions in funding markets, and thus confounded the effect of the sovereign shock. We do not extend the sample before 2011 to avoid that our results might be influenced by events occurring in previous periods. In particular, 2010 was characterized by a slow recovery from the impact of the Lehman shock, and 2009 was part of the “full” crisis period following the default of Lehman. We include all nonfinancial firms with outstanding credit in the CR, including very small firms, both incorporated firms and sole proprietorship.5 Because our identification strategy relies on comparing the behavior of a foreign and an Italian bank lending to the same firm, we select firms that borrow from at least one Italian and one foreign bank in each period.6 This is not a restrictive condition. Table A.1 in the Online Appendix shows characteristics of firms in our sample and of firms borrowing only from Italian banks. The table shows normalized differences (Imbens and Wooldridge 2009) of the balance sheet items indicated in each row. Although firms in our sample are larger and have a slightly lower ratio of operating profits to interest expenses than firms borrowing only from Italian banks, differences are not statistically significant for any of the firm characteristics.7 We aggregate credit at the banking group level, because lending and funding policies are typically decided at the headquarter level. If a firm borrows from two banks belonging to the same banking group, we consider this as a single relationship. Overall, our sample includes 664,198 bank–firm relationships over the two periods, involving 164,470 firm-period couples, with 92,620 distinct firms sampled at least in one period.8 Median credit committed was around 800,000 euros in June 2011, consistent with the presence of a large number of small firms in the sample. Firms were borrowing on average from four banks. Half of the sample firms operate in the services sector, 30% in manufacturing and energy, 12% in construction, 8% in agriculture. Key balance sheet characteristics of firms in the sample are shown in Table 2. Table 2. Descriptive statistics of firms. Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Notes: The table shows descriptive statistics of the firms of our sample that are included in the Firm Register (CERVED) for which we can observe balance sheet information. Leverage is defined as total assets−equity−debt versus shareholders divided by total assets. Liquidity is the sum of cash and cash equivalents. Z-score is Altman Z-score as computed by CERVED, and it represents an estimate of firms’ default probability. It is computed on a scale from 1 to 9 and higher values indicate a higher probability of default. View Large Table 2. Descriptive statistics of firms. Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Notes: The table shows descriptive statistics of the firms of our sample that are included in the Firm Register (CERVED) for which we can observe balance sheet information. Leverage is defined as total assets−equity−debt versus shareholders divided by total assets. Liquidity is the sum of cash and cash equivalents. Z-score is Altman Z-score as computed by CERVED, and it represents an estimate of firms’ default probability. It is computed on a scale from 1 to 9 and higher values indicate a higher probability of default. View Large Table 3. Home country of the banks included in the sample and changes in spreads. Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Notes: The table shows the number of bank–firm relationships included in the sample, by banks’ home country. Spreads are the difference between the yield of the 10-year government bond of the country the bank is headquartered in and the 10-year German Bund. The change in spreads is computed as the difference between the average spread in March 2011 and the average spread in January 2011 for the pre-crisis period and as the difference between the average spread in September 2011 and the average spread in July 2011 for the crisis period. Averages are computed on daily data on spreads from Thomson Datastream. View Large Table 3. Home country of the banks included in the sample and changes in spreads. Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Notes: The table shows the number of bank–firm relationships included in the sample, by banks’ home country. Spreads are the difference between the yield of the 10-year government bond of the country the bank is headquartered in and the 10-year German Bund. The change in spreads is computed as the difference between the average spread in March 2011 and the average spread in January 2011 for the pre-crisis period and as the difference between the average spread in September 2011 and the average spread in July 2011 for the crisis period. Averages are computed on daily data on spreads from Thomson Datastream. View Large Table 3 shows the distribution of bank–firm relationships by home country of the lender. More than a quarter of the relationships involve foreign banks. The majority are French, followed by German, then US and Austrian. Table 3 also shows the change in the spread of the 10-year sovereign security relative to the 10-year German Bund of the same maturity, between January and March 2011 for the pre-crisis period, and between July 2011 and September 2011 for the crisis period. The spread increased sharply, by almost 200 basis points, for Italy (see also Figure 2), and to a lesser extent for Slovenia (110 basis points), Japan and Spain (98 and 83 basis points, respectively). Prior to the crisis, spreads remained roughly unchanged. To better assess to what extent foreign banks differ from domestic ones, it is useful to describe other characteristics of the two groups. The importance of foreign banks in Italy increased steadily in the past 20 years. The share of total assets that they held increased from below 4% in 1992 to 17.5% at the end of 2011, when two subsidiaries of foreign banks ranked among the top ten banking groups operating in Italy. A large fraction of the liabilities of branches and subsidiaries of foreign banks is represented by interbank transfers from their headquarters that raise funds either in their home country or in the international wholesale markets (70% for branches and 40% for subsidiaries); local retail funding, which may also have been affected by the sovereign debt crisis, is much less important for foreign banks relative to Italian ones. Subsidiaries of foreign banks have a business model similar to that of domestic banks. Typically subsidiaries of foreign banks were formerly domestically owned banks acquired by a foreign bank holding company, thus they have an extensive network of outlets and had been active in Italy for a long time, so they had established tight relationships with domestic borrowers over time. Among the subsidiaries of foreign banks included in our sample are: BNL-BNP Paribas, Cariparma-Credit Agricole, Deutsche Bank Italia, Santander, that are reasonably comparable with large and medium Italian commercial banks. The entry of foreign banks in Italy stopped at the beginning of the financial crisis, so changes in ownership did not occur during our sample period. Overall, there is limited evidence that the composition of loans of domestic banks changed significantly after acquisitions by foreign banks. In 2011, subsidiaries of foreign banks were supervised by Italian authorities, which were also responsible of their resolution in case of distress. However, they could also indirectly rely on the public support of their headquarters’ country if their default threatened the viability of the whole banking group. Panel A of Table 4 shows descriptive statistics on the activity of foreign and Italian banks included in our sample. Overall we include 567 banks, 49 of which foreign. Foreign banks hold a lower share of loans to households but a similar share of loans to nonfinancial firms and a similar average number of branches as domestic banks.9 Table 4. Descriptive statistics of domestic and foreign banks. Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Notes: Panel A shows descriptive statistics of the loan portfolio of domestic and foreign banks operating in Italy. Data are from June 2011 Supervisory reports submitted by intermediaries to the Bank of Italy. The data for foreign banks refer only to the banks’ operations in Italy. Total domestic loans include loans to Italian nonfinancial firms, households and nonprofit organizations. Panel B shows descriptive statistics of the main balance sheet variables for domestic and foreign banks operating in Italy. Data for Italian banks are from the consolidated Supervisory reports submitted to the Bank of Italy. Data for foreign banks are from Bankscope, and are at the consolidated level including operations outside Italy. Data on holdings of GIIPS (Greece, Ireland, Italy, Portugal and Spain) sovereign bonds for foreign banks have been hand-collected from banks’ published balance sheet. Data refer to 567 banks and are from the June 2011 balance sheet. T1 RATIO is the ratio of Tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, GIIPS HOLDINGS is the ratio of holdings of sovereign bonds from GIIPS to total assets, ROA (return on assets) is the ratio of profits (or losses) to total assets, SIZE is the natural logarithm of total assets. View Large Table 4. Descriptive statistics of domestic and foreign banks. Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Notes: Panel A shows descriptive statistics of the loan portfolio of domestic and foreign banks operating in Italy. Data are from June 2011 Supervisory reports submitted by intermediaries to the Bank of Italy. The data for foreign banks refer only to the banks’ operations in Italy. Total domestic loans include loans to Italian nonfinancial firms, households and nonprofit organizations. Panel B shows descriptive statistics of the main balance sheet variables for domestic and foreign banks operating in Italy. Data for Italian banks are from the consolidated Supervisory reports submitted to the Bank of Italy. Data for foreign banks are from Bankscope, and are at the consolidated level including operations outside Italy. Data on holdings of GIIPS (Greece, Ireland, Italy, Portugal and Spain) sovereign bonds for foreign banks have been hand-collected from banks’ published balance sheet. Data refer to 567 banks and are from the June 2011 balance sheet. T1 RATIO is the ratio of Tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, GIIPS HOLDINGS is the ratio of holdings of sovereign bonds from GIIPS to total assets, ROA (return on assets) is the ratio of profits (or losses) to total assets, SIZE is the natural logarithm of total assets. View Large Panel B of Table 4 shows descriptive statistics of the main balance sheet characteristics of the banks included in our sample. These data are on a consolidated basis and are from June 2011, yet variation over our sample period is very limited. Relative to domestic banks, foreign banks are on average larger, less capitalized, more relying on interbank funding, and less exposed to GIIPS sovereign securities. They do not differ significantly in terms of profitability, as measured by ROA. These differences are much less pronounced when foreign banks are compared to the subset of larger Italian banks. The top 50 domestic banks differ significantly from foreign banks only along the exposure to risky sovereign debt (higher for domestic banks) and in size (smaller for domestic banks). 5. Results for the Baseline Model Results from the estimation of equation (1) are reported in Table 5. The first column shows the effect of the dummy FOREIGN on the growth of credit committed without fixed effects of any sort; even in this basic specification, the coefficient of the dummy FOREIGN without interactions is not statistically significant and large and significant when interacted with the dummy CRISIS, showing that the behavior of domestic and foreign banks differs during the crisis. Results are quantitatively and qualitatively unchanged once we take into account observed and unobserved heterogeneity at the bank, firm, and time level. Similar results are indeed obtained when we plug firm fixed effects, which absorb all time-invariant observed and unobserved firm heterogeneity (column 2) and when we allow this heterogeneity to be time-varying (column 3). The difference in the estimates is not large, suggesting that firm demand for credit does not play a very strong role. This finding is in line with other work using CR data (Bentolila, Jensen, and Jiménez 2017, Cingano, Manaresi, and Sette 2016, Jiménez et al. 2010). The coefficients of the dummy FOREIGN and of the interaction FOREIGN*CRISIS change very little when relationship-level controls (and their interaction with the dummy crisis) are also included in the regression (column (4)), to account for specific relationship features that might vary across lending banks and over time.10 Table 5. Baseline. (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. In the seventh column, the sample includes firms in the Firm Register that borrow from banks for which we observe all balance sheet data; in the 8th column, the sample includes firms in the Firm Register for which we observe balance sheet data. CRISIS is a dummy variable equal to one if data are from the June 2011–December 2011 period. SHARE OF TOTAL CREDIT is the share of total credit committed by the bank to the firm, DRAWN OVER COMMITTED is the ratio of drawn to committed credit in the relationship, OVERDRAFT is the share of overdraft loans to total loans granted by the bank to the firm. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Table 5. Baseline. (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. In the seventh column, the sample includes firms in the Firm Register that borrow from banks for which we observe all balance sheet data; in the 8th column, the sample includes firms in the Firm Register for which we observe balance sheet data. CRISIS is a dummy variable equal to one if data are from the June 2011–December 2011 period. SHARE OF TOTAL CREDIT is the share of total credit committed by the bank to the firm, DRAWN OVER COMMITTED is the ratio of drawn to committed credit in the relationship, OVERDRAFT is the share of overdraft loans to total loans granted by the bank to the firm. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large The specification of column (4), which we consider our benchmark, indicates that during the crisis the behavior of the two types of banks differs: credit committed by foreign banks grew by about 3 percentage points more than credit committed by domestic banks. This is an economically significant effect, because the average rate of growth of credit during the crisis is −6.6%. Moreover, the measure of credit supply is the growth rate (log change) in the stock of credit commitments in each bank–firm relationship, and a 3% increase is a sizable amount. Results are also robust to the inclusion of bank fixed effects (column 5), which absorb the dummy FOREIGN. Indeed we do not observe difference in the coefficients across columns (4) and (5). This suggests that bank time-invariant unobserved heterogeneity is not strongly correlated with the impact of the shock, which appears to mainly depend on the nationality of the bank holding company. Column (6) shows estimates from a regression including a full set of firm*bank fixed effects, together with the firm*time fixed effects. This is a very demanding specification that allows controlling for all unobserved time invariant characteristics of individual firm–bank relationships, for example the possibility that certain banks specialize in certain types of loans, or that certain relationships have a longer duration than others, etc. Remarkably, results still hold, although the size of the effect is somewhat smaller than in the baseline: credit committed by foreign banks grew by about 1.7 percentage points more than credit by domestic banks. The difference with the baseline estimates is not statistically significant, though, and the estimates of the other relationship level controls strongly increase in size, likely due to a high correlation with the bank–firm fixed effects. For robustness, we also run our benchmark specification of column (4) on the two subsets of observations for which we avail of either bank or firm balance sheet characteristics and that will be used for estimation in later sections. Results on these two subsets, respectively shown in columns (7) and (8), are basically unchanged. As an additional robustness check, we run a placebo experiment using the periods before June 2011 to test whether the different behavior of domestic and foreign banks started precisely after the burst of the sovereign debt crisis. We consider the period December 2009–December 2010, setting the fictitious event at June 2010. Next, we extend the period to June 2011, and we set the event at June 2010 or at December 2010. In all cases (Supplemental Table S2), neither the dummy FOREIGN, nor the interaction between the dummy FOREIGN and the dummy CRISIS are significant. Coefficients are also small in size. Finally, to further test the robustness of the main results, we also run regressions taking averages of credit committed in the period before (the six month from January 2011 to June 2011) and after (the six months from July 2011 to December 2011) the crisis and computing log changes of the average credit committed and as shown in Table A.3 in the Online Appendix. We also run the baseline regression comparing foreign banks and the subsample of the top 50 domestic banks, those for which differences in balance sheet characteristics are much more limited, except for the holdings of GIIPS sovereign bonds and for size, and results (not reported) hold through. 6. Bank Heterogeneity Table 5 suggests that bank’s nationality drives lending policies during the sovereign debt crisis. In this section, we investigate to what extent the dummy FOREIGN is capturing bank characteristics that affect lending growth differentially across domestic and foreign banks when the crisis hit and we provide further evidence on the drivers of this foreign bank effect. We explore the first issue by augmenting our benchmark regression (column 4 of Table 5) with time-varying bank balance sheet characteristics that are likely to influence lending.11 This set of variables includes the ratio of GIIPS sovereign debt holdings to total assets (GIIPS HOLDINGS) to measure the direct exposure of banks to sovereign debt, the Tier 1 ratio (Tier 1 capital to risk-weighted assets, TIER 1), which is a measure of capitalization, the ratio of interbank funding (deposits and repos) to assets (INTERBANK), which captures banks’ reliance on wholesale funding, the most volatile funding component that dried-up sharply in the second half of 2011 (Cappelletti 2013 and Correa, Sapriza, and Zlate 2012), return on assets (ROA) as a measure of profitability, a set of dummies for the quartiles of bank’s assets to control for bank size, and a dummy identifying mutual banks (all bank-level controls, including the size and mutual bank dummies, are interacted with the dummy crisis). Results are shown in Table 6. The interaction between the dummy FOREIGN and the dummy CRISIS is always positive and significant, indicating that the main result is robust to the inclusion of time-varying bank-level controls. Furthermore, bank-level characteristics play a limited role in explaining the different behavior of domestic and foreign banks. In particular, banks less reliant on interbank funding and banks that are more profitable tend to lend more, but the impact of these characteristics does not vary across the two periods that we consider. Interestingly, GIIPS HOLDINGS is never significant, indicating that the holdings of sovereign bonds of “peripheral” countries has no effect on credit supply once we control for the dummy FOREIGN. Importantly, the effect of the interaction FOREIGN*CRISIS is still positive and significant when all bank controls are included (column 5) and even when bank fixed effects are plugged in (column 6).12 Furthermore, the size of the coefficient of the interaction FOREIGN*CRISIS is very stable across specifications and numerically very close to that of the baseline, indicating that the inclusion of individual bank characteristics has little correlation with the effect of being a foreign or domestic bank. Table 6. Regressions with bank balance sheet variables. (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on bank characteristics. FOREIGN is a dummy equal to 1 if the bank is foreign, CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. GIIPS HOLDINGS is the ratio of bank’s portfolio holdings of government bonds of Greece, Ireland, Italy, Portugal and Spain to total assets. T1 RATIO is the ratio of tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, ROA is return on assets. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Balance sheet data for domestic banks are from the Supervisory reports submitted to the Bank of Italy, for foreign banks are from Bankscope. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Table 6. Regressions with bank balance sheet variables. (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on bank characteristics. FOREIGN is a dummy equal to 1 if the bank is foreign, CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. GIIPS HOLDINGS is the ratio of bank’s portfolio holdings of government bonds of Greece, Ireland, Italy, Portugal and Spain to total assets. T1 RATIO is the ratio of tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, ROA is return on assets. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Balance sheet data for domestic banks are from the Supervisory reports submitted to the Bank of Italy, for foreign banks are from Bankscope. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Overall this evidence confirms that foreign banks, less affected by the sovereign shock, increased their credit supply more than domestic banks during the sovereign crisis and that being FOREIGN is not just disguising balance sheet differences between domestic and foreign banks. Next, we explore to what extent the foreign bank effect is due to a differentiated pattern of bank funding cost at the country level, as suggested by the aggregate evidence shown in Figure 1. As a first test, we substitute the dummy FOREIGN with country-level averages of banks' cost of funding.13 Table 7 shows that an increase in the sovereign spread has a negative effect on the supply of credit during the crisis. As yields on corporate bonds, including bank bonds, raise with the yield on sovereign bonds, this suggests an impact on lending through the higher cost of issuing bonds for banks. Similarly, an increase in the cost of deposits and, in particular, in the cost of deposits to households (again averaged at the country-level) has a negative impact on bank lending. These results also hold when focusing only on the crisis periods and when controlling for bank balance sheet characteristics (Tables A.5 and A.6 in the Online Appendix). Table 7. Effect of banks’ cost of funding. (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on different measures of banks’ cost of funding. DELTA SPREAD is the change in the spread between the 10-year sovereign bond of the country the bank is headquartered in and the 10-year German Bund; DELTA DEPOSITS is the change in the average cost of deposits and DELTA DEPOSITS HOUSEHOLDS is the change in the average cost of deposits from households, both measured in the country the bank is headquartered in. These two measures are from the ECB MFI interest rates statistics. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). All relationship level controls are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Table 7. Effect of banks’ cost of funding. (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on different measures of banks’ cost of funding. DELTA SPREAD is the change in the spread between the 10-year sovereign bond of the country the bank is headquartered in and the 10-year German Bund; DELTA DEPOSITS is the change in the average cost of deposits and DELTA DEPOSITS HOUSEHOLDS is the change in the average cost of deposits from households, both measured in the country the bank is headquartered in. These two measures are from the ECB MFI interest rates statistics. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). All relationship level controls are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large To further corroborate the hypothesis that the foreign bank effect captures a larger increase in the cost of funding for domestic banks, we run a bank-level regression of a measure of banks’ average cost of funding on the dummy FOREIGN and bank-level balance sheet variables.14 Results in Table 8 show that the dummy FOREIGN has a negative and significant coefficient in all specifications. This suggests that the change in the average cost of funding of foreign banks has been significantly lower than that of Italian banks after the crisis, and that this common country effect is over and above due to differences in individual bank balance sheet characteristics. Table 8. Change in banks’ cost of funding. (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 Notes: The table shows results of regressions of the change in the average cost of funding at the bank level on the dummy FOREIGN and on bank balance sheet characteristics. Column (1) shows results for the period June 2011–December 2011 (crisis period), columns (2) and (3) for both the June 2011–December 2011 and the December 2010–June 2011. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. In columns (2) and (3), the mutual bank dummy and the bank size dummies are also interacted with the dummy CRISIS. Data are available for Italian banks and for subsidiaries of foreign banks. Standard errors double clustered at the bank level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large Table 8. Change in banks’ cost of funding. (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 Notes: The table shows results of regressions of the change in the average cost of funding at the bank level on the dummy FOREIGN and on bank balance sheet characteristics. Column (1) shows results for the period June 2011–December 2011 (crisis period), columns (2) and (3) for both the June 2011–December 2011 and the December 2010–June 2011. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. In columns (2) and (3), the mutual bank dummy and the bank size dummies are also interacted with the dummy CRISIS. Data are available for Italian banks and for subsidiaries of foreign banks. Standard errors double clustered at the bank level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large 7. Extensions We extend our baseline results in two directions. First, we explore differences in the behavior of foreign banks depending on whether they are incorporated in Italy as subsidiaries or branches. Second, we explore heterogeneous effects across firms of different size and financial strength. 7.1. Branches and Subsidiaries To shed more light on our results, we exploit the variability within foreign banks, classifying them separately into branches and subsidiaries. The latter are very similar to domestic banks in terms of extension of their network of outlets and business model, the former concentrate their activity in certain areas of the country and are typically specialized in specific market segments, such as syndicated loans, leasing, etc. This test is important both as a robustness check of our baseline, and as a contribution to the literature on global banks. Table 9 shows results of our analysis. Columns (1) and (3) display estimates from regressions run on the subsample of firms borrowing from at least one domestic bank and at least one subsidiary of foreign banks. Results are similar to those of the baseline regressions. Columns (2) and (4) display estimates from regressions run on the subsample of firms borrowing from at least one domestic bank and at least one branch of foreign banks. In this case, we find no difference in the credit quantity supplied by domestic banks and by branches of foreign banks. Table 9. Distinguishing branches and subsidiaries. (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN and on bank characteristics. Columns (1) and (3) include firms borrowing from Italian banks and from subsidiaries of foreign banks. Columns (2) and (4) include firms borrowing from Italian banks and from branches of foreign banks. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. Balance sheet data are from the Supervisory reports submitted to the Bank of Italy. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large Table 9. Distinguishing branches and subsidiaries. (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN and on bank characteristics. Columns (1) and (3) include firms borrowing from Italian banks and from subsidiaries of foreign banks. Columns (2) and (4) include firms borrowing from Italian banks and from branches of foreign banks. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. Balance sheet data are from the Supervisory reports submitted to the Bank of Italy. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large These results indicate that the effect found in the baseline regression is mainly driven by a different behavior of Italian banks relative to subsidiaries of foreign banks. By contrast, we find a smaller difference between domestic banks and branches of foreign banks, despite the fact that the latter enjoy better access to funding than domestic banks. We interpret these results as evidence that the organizational structure of foreign banks is relevant for lending decisions. Foreign banks’ organizational form has been found to be an important driver of their lending policy in particular during crises: Cetorelli and Goldberg (2012b) find that parent banks, when hit by a funding shock, reallocate liquidity towards affiliate locations that are important for the parent bank revenue streams that are then relatively protected from liquidity reallocation within the organization. The presence of a subsidiary is a proxy for the importance of the Italian market in the portfolio of the global bank. These findings are important because they attenuate concerns that our main results could be driven by those foreign banks specialized in particular markets, such as those of M&As or syndicated loans (branches). The difference in credit supply is found precisely when we compare Italian banks to the set of foreign banks that are more similar to Italian banks, the subsidiaries. Moreover, these results suggest that for foreign banks to play a mitigating role in the transmission of shocks to the domestic banking sector, they have to be well established in the country, with a large network of outlets, possibly reflecting a higher ability to collect soft information about borrowers (Beck, Ioannidou, and Schäfer 2017). 7.2. Firm Heterogeneity We further extend our results to test whether Italian banks reduced credit mainly to certain categories of borrowers. This is an important extension for two sets of reasons. First, it allows to understand to what extent the drop in credit by domestic versus foreign banks has been heterogeneous depending on firm size. Previous work (Iyer et al. 2014, Kahle and Stulz 2010) shows that the biggest brunt of credit shocks is borne by small firms, which are also less able to substitute bank credit with other sources of finance than larger firms. If this is the case, policy measures aimed at increasing the ability of smaller firms to access external funds would be especially important. Second, it allows us to ascertain to what extent the higher credit by foreign banks has gone to financially sound rather than to weaker firms. Again, this is crucial to uncover whether foreign banks were able to take more risk, thanks to their lower exposure to the shock, or whether they “cream-skimmed” borrowers, exploiting their higher ability to grant credit. To perform these tests, we merge our data with balance sheet information from the Firm Register (Cerved). We run the baseline regressions splitting the sample of borrowers into large/small firms, high/low leverage firms, high/low roa, high/low ebitda to interest expenses, bad/good Z-score.15 Results are shown in Table 10. We find little difference in lending policies across firm size (columns 1 and 2) and across firms ability to repay interest expenses with operating profits (columns 7 and 8). The difference in lending of foreign banks compared to Italian banks is instead numerically larger in the case of firms with lower liquidity and of those with bad Z-score, although none of the differences is statistically significant. This provides some limited evidence suggesting that Italian banks cut credit more to more fragile firms relative to foreign banks. Table 10. Heterogeneous effects across firms. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN, splitting the sample according to the firm characteristics listed in each column. Large or High indicates above the median, Small or Low indicates below the median. “Bad Z-Score” identifies firms with a Z-Score in the bottom three classes. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. All relationship level controls are also interacted with the dummy CRISIS. Firm balance sheet data are from the Firm Register (CERVED). Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large Table 10. Heterogeneous effects across firms. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN, splitting the sample according to the firm characteristics listed in each column. Large or High indicates above the median, Small or Low indicates below the median. “Bad Z-Score” identifies firms with a Z-Score in the bottom three classes. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. All relationship level controls are also interacted with the dummy CRISIS. Firm balance sheet data are from the Firm Register (CERVED). Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large 8. The Aggregate Effect The results discussed so far are based on coefficients estimated comparing the behavior of a domestic and a foreign bank lending to the same borrower (within estimator). However, firms might compensate the reduction in credit from domestic banks with more loans from foreign banks. Estimates from a simple firm-level regression is likely to be biased, though, because changes in the log of total credit at the firm level also reflect firm-level demand for credit, changes in firm financial strength, and the like. A method to estimate unbiasedly the firm-level (aggregate) impact of the supply shock induced by the crisis on the growth of credit commitments has recently been proposed by Jiménez et al. (2010). However, their methodology does not allow to easily obtain standard errors and thus to conduct inference. In this paper, we use an equivalent estimation procedure, which directly yields standard errors for the unbiased effect of the proxy for the exposure of firms to the credit supply shock (Cingano, Manaresi, and Sette 2016). We first estimate firm-fixed effects from the baseline model at the bank–firm level. Then we plug these estimated firm effects in a firm-level equation in which the dependent variable is the growth of total credit committed to firms by banks (including new relationships) and the measure of the exposure to the credit supply shock is the initial share of credit committed by foreign banks.16 Standard errors are estimated by block-bootstrapping at the level of the main bank (the bank with the largest share of credit), to take into account the fact that firm fixed effects are estimated regressors. Formally, from the baseline model (equation 1), we obtain an estimate of the firm-period fixed effect $$\hat{\alpha }_{i,t}$$. As a second step, we estimate \begin{equation*} \Delta \mathit{credit}_{i,t}=\beta _{1}\overline{ \mathit{FOREIGN}_{i}}+\beta _{2}\overline{ \mathit{FOREIGN}_{i}}\ast \mathit{CRISIS}_{t}+\hat{\alpha }_{i,t}+\varepsilon _{i,t}, \end{equation*} where $$\overline{ \mathit{FOREIGN}_{i}}$$ is the share of credit committed by foreign banks, that is the average at the firm level of the dummy FOREIGN weighted by the share of credit to the firm held by each bank. A more detailed description of this approach can be found in the Appendix. Results are shown in Table 11. Column (1) shows results of a regression excluding the estimated firm effects. The interaction term between the dummy FOREIGN and the dummy CRISIS is positive and significant. This indicates that firms are not able to fully substitute credit from domestic banks by increasing credit from foreign banks. However, as argued above, this result may be biased. Column (2) shows estimates including the firm effect. Table 11. Aggregate effect. (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 Notes: The table shows results of the regressions where the dependent variable is the growth of credit committed at the firm level. All independent variables are firm-level averages, weighted by the share of credit committed by each bank to the firm at the beginning of each period (December 2010 pre-crisis period, June 2011 post-crisis period). In particular, FOREIGN is the share of credit committed by foreign banks to the firm. FIRM EFFECT is a vector including the firm fixed effects estimated from equation (1). Data are from the Italian CR. Block-bootstrapped (clustered at bank level) standard errors in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large Table 11. Aggregate effect. (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 Notes: The table shows results of the regressions where the dependent variable is the growth of credit committed at the firm level. All independent variables are firm-level averages, weighted by the share of credit committed by each bank to the firm at the beginning of each period (December 2010 pre-crisis period, June 2011 post-crisis period). In particular, FOREIGN is the share of credit committed by foreign banks to the firm. FIRM EFFECT is a vector including the firm fixed effects estimated from equation (1). Data are from the Italian CR. Block-bootstrapped (clustered at bank level) standard errors in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large The interaction between the dummy FOREIGN and the dummy CRISIS is still positive and significant, although the size of the coefficient is smaller: if the share of credit that a firm obtained before the crisis from foreign banks increases by one standard deviation (23 percentage points), credit growth during the crisis is about 0.8 percentage points higher. This is a large effect because the median credit growth at the firm-level in the crisis period is −2.8% (the mean is −3.8%). Finally, the estimated firm effect is highly significant and positive, consistent with the hypothesis that it is capturing firm–level demand for credit. Overall, these results show that firms have not been able to fully substitute credit from domestic banks with credit from foreign banks: during the sovereign crisis, the “aggregate” effect on the supply of credit at the firm-level was strong. In principle, firms may have been able to substitute bank credit with market-based external finance. This has been the case in other countries during the 2007–2008 financial crisis. Adrian, Colla, and Shin (2013) show that U.S. borrowers substituted from bank-based finance to market-based finance; Abbassi et al. (2016) find that German firms that experience credit reduction due to securities trading by banks during the crisis partly compensated it by issuing bonds. The substitution of bank credit with other sources of external finance is especially difficult during the sovereign crisis for Italian firms. First, most of them, including those in our sample, are small and medium-sized enterprises (see Table 2) for which bank credit represents the only source of external finance, possibly complemented by trade credit. Second, in the second half of 2011 in Italy the cost of other sources of finance increased significantly also for nonfinancial borrowers. In particular, bond yields raised at unprecedented highs together with sovereign yields, both because of the sovereign ceiling phenomenon (Adelino and Ferreira 2016) and because of the higher riskiness of Italian nonfinancial borrowers during the crisis. The difficulty of substituting bank credit with other sources of finance for Italian firms in our sample period is confirmed by the low level of bond issuance during 2011, basically negligible for SMEs (Bank of Italy, 2014). 9. Other Margins As a last step in our analysis, we test the impact of the sovereign shock on other key margins of lending: interest rates and the probability a loan application is accepted. 9.1. Effect on Interest Rates We first test the impact of the sovereign shock on interest rates. We use information on the interest rates from the CR, which collects these data from a representative sample of banks (110 intermediaries, some of which belonging to the same bank consolidated entity, representing over 80% of the total market for loans in Italy). The data set contains information disaggregated by loan type (revolving credit lines, term loans, loans backed by receivables).17 We estimate a version of equation (1) in which the dependent variable is alternatively the change in the Annual Percentage Rate (APR) on revolving credit lines and that on term loans (see Khwaja and Mian 2008 or Chodorow-Reich 2014 for a similar approach). We compute the change in the APR for the pre-crisis (June 2011–December 2010) and crisis (December 2011–June 2011) periods. We use the APR net of fees and commissions, which are typically computed on credit granted while the interest rates we observe are estimated on the basis of the actual usage of the credit line. Our results also hold if we use interest rates gross of fees and commissions. Importantly, all regressions still include firm*period fixed effects. Results are shown in Table 12. Columns (1) and (2) show estimates of the regressions for the change in the APR on revolving credit lines. Column (1) does not include bank fixed effects. Before the crisis, there was no difference between domestic and foreign banks. After the crisis, foreign banks increased rates on revolving credit lines by about 21 basis points less than domestic banks lending to the same firm. Column (2) shows the same regression including bank fixed effects. Results are essentially unchanged. Columns (3) and (4) (the latter includes bank fixed effects) show estimates of the same regressions run for the change in the APR (net of fees and commissions) on term loans. Results are confirmed: foreign banks increase interest rates by about 15 basis points less than domestic banks.18 Table 12. Effect on the cost of credit. (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 Notes: The table shows regressions of the change in the Annual Percentage Rate (APR) on revolving credit lines (column 1 and 2) and on term loans (column 3 and 4) granted by banks to nonfinancial firms in Italy on the variable FOREIGN, a dummy equal to 1 if the bank is a foreign bank. Changes are computed between December 2011 and June 2011 for the crisis period and between June 2011 and December 2010 for the pre-crisis period. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). The regressions include interactions of the relationship level controls and the dummy CRISIS. The sample includes bank–firm relationships from the subsection of the Italian CR reporting information on interest rates over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large Table 12. Effect on the cost of credit. (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 Notes: The table shows regressions of the change in the Annual Percentage Rate (APR) on revolving credit lines (column 1 and 2) and on term loans (column 3 and 4) granted by banks to nonfinancial firms in Italy on the variable FOREIGN, a dummy equal to 1 if the bank is a foreign bank. Changes are computed between December 2011 and June 2011 for the crisis period and between June 2011 and December 2010 for the pre-crisis period. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). The regressions include interactions of the relationship level controls and the dummy CRISIS. The sample includes bank–firm relationships from the subsection of the Italian CR reporting information on interest rates over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large 9.2. Probability of Accepting a Loan Application In this section, we examine whether Italian and foreign banks displayed any difference in the likelihood to grant loans to new clients. In line with Jiménez et al. (2012), we use data on loan applications recorded in the CR to analyze the probability of acceptance of new credit. We collect data on all the requests recorded as loan applications between October 2010 and March 2011 and between July 2011 and December 2011, pre-crisis and crisis period, respectively. If we observe that subsequently banks grant credit to the applying firms, we classify the application as accepted. 19 Our dependent variable is a dummy equal to 1 if the application of firm j to bank i is accepted, 0 otherwise, and we estimate a linear probability model.20 We estimate models both including and excluding firm*time fixed effects. Including firm*time fixed effects is important to control for applicant unobservables, but forces us to use the sample of firms posting at least two loan applications in a period. Such firms may be different, likely worse, than the average firm applying for a loan.21 A stand-out descriptive feature of the frequency of accepted applications is that it sharply dropped during the crisis, to 9% between June 2011 and March 2012 from the 37% observed in the three previous quarters. Results shown in columns (1) to (4) of Table 13 confirm the evidence of the aggregate data: foreign banks were less likely to accept a loan application prior to the crisis, and the difference disappeared during the sovereign shock, suggesting that domestic banks tightened their acceptance rate significantly after the sovereign debt crisis. Results are similar across regressions including and excluding firm*period fixed effects, and including and excluding bank fixed effects. Table 13. Probability of accepting a loan application. (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 Notes: The table shows regressions of a dummy variable equal to 1 if the loan application has been accepted by the bank on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. The sample includes information on loan applications and on granted credit from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1. View Large Table 13. Probability of accepting a loan application. (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 Notes: The table shows regressions of a dummy variable equal to 1 if the loan application has been accepted by the bank on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. The sample includes information on loan applications and on granted credit from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1. View Large Overall, these results indicate that domestic banks, more affected by the shock, diminished their propensity to accept loan applications and raised interest rates more than foreign banks. 10. Concluding Remarks This paper shows that lending by Italian banks declined relatively more than lending by foreign banks after the summer of 2011, when the European sovereign debt crisis hit Italy. In particular, we find that lending by domestic banks declined by 3 percentage points more than lending by foreign banks. These findings do not just reflect differences in bank balance sheet characteristics across foreign and domestic banks, such as the strength of their capital position, profitability, holdings of sovereign bonds. In fact, our results indicate that the drop in lending has a strong country-specific component, depending on the location of banks’ headquarter. This country-specific component is related to the large increase in funding costs that affected Italian banks in association with the sovereign shock, rather homogeneously and independently of individual bank characteristics. We interpret the effect of this country-specific component as the impact of the sovereign debt crisis, potentially uncovering a new lending channel linking sovereign shocks to negative shocks to credit supply, through banks’ cost of funding. We find that such differential behavior between Italian and foreign banks is due to subsidiaries only, whose business model is more similar to that of Italian banks, providing more evidence on the higher relative importance of being headquartered in countries unscathed by sovereign shocks versus differences in business models. Furthermore, our results indicate that firms have not been able to fully substitute the decrease of loans by Italian banks with loans by foreign banks, pointing to an aggregate effect of the sovereign shock on credit supply. Finally, our findings show that the transmission of the sovereign shock also occurred through higher interest rates and through a significant drop in the acceptance rate of loan applications. Appendix Derivation of the Aggregate Effect The relationship level equation is the following: \begin{equation*} \Delta \mathit{credit}_{i,j,t}=\beta _{1} \mathit{domestic}_{j}+\beta _{2} \mathit{domestic}_{j}\ast \mathit{CRISIS}_{t}+\alpha _{i,t}+\varepsilon _{i,j,t}, \end{equation*} where $$\Delta \mathit{credit}_{i,j,t}$$ is the growth rate of credit to firm i by bank j at time t. Then, we take the average of both sides of this equation weighted by the share of credit held by each bank as follows: \begin{align*} \sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}&\ast \frac{ \mathit{credit}_{j,t}}{ \sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}=\beta \sum _{j=1}^{n_{i}} \mathit{domestic}_{j}\ast \frac{ \mathit{credit}_{j,t}}{\sum _{j=1}^{n_{i}} \Delta \mathit{credit}_{i,j,t}} \\ &+\beta \sum _{j=1}^{n_{i}} \mathit{domestic}_{j}\ast \mathit{CRISIS}_{t}\ast \frac{ \mathit{credit}_{j,t}}{ \sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}\\ &+\sum _{j=1}^{n_{i}}\frac{ \mathit{credit}_{j,t} }{\sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}\alpha _{i,t}\!+\! \sum _{j=1}^{n_{i}} \frac{ \mathit{credit}_{j,t}}{\sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}\varepsilon _{i,j,t}, \end{align*} where \begin{equation*} \sum _{j=1}^{n_{i}}\frac{ \mathit{credit}_{j,t}}{\sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}=1. \end{equation*} Simple algebra shows that the left-hand side is the growth rate of total credit obtained by firm i at time t. Then this yields \begin{equation*} \Delta \mathit{credit}_{i,t}=\beta _{1}\overline{ \mathit{domestic}_{i}}+\beta _{2}\overline{ \mathit{domestic}_{i}}\ast \mathit{CRISIS}_{t}+\hat{\alpha }_{i,t}+\nu _{i,t}, \end{equation*} which is the equation for the growth of credit at the firm level we are interested to estimate. To obtain the $$\hat{\alpha }_{i,t}$$, we estimate them from the relationship-level equation. These estimates are unbiased and consistent as the number of banks increases (provided that the number of firms does not go to infinity). As the $$\hat{\alpha }_{i,t}$$ are estimated in the relationship level equation, standard errors need to be estimated by bootstrapping to obtain correct estimates of the variance–covariance matrix. This equation is exactly valid for the growth rate of credit. We approximate it by the log change in credit. To estimate the full aggregate effect, we also take into account that part of the growth of credit is due to the starting of new credit relationships. Our approach is valid as long as the firm-specific effect is the same for old as for new relationships, possibly up to a noise term uncorrelated with both the other regressors and the firm effect. This is reasonably true for firm-specific characteristics such as firm riskiness. It must also be true for firm demand for credit, which must not be bank specific. This, however, is an identifying assumption that must hold throughout our analysis, also when we study credit supply at the bank–firm relationship level. Acknowledgments We would like to thank the editor and four anonymous referees for their helpful comments. We are especially grateful to Atif Mian for his insightful discussion of the paper at the 2013 NBER Summer Institute, and to Nicola Cetorelli, Linda Goldberg, and Steven Ongena for their detailed comments. We thank Giorgio Albareto, Martin Brown, Elena Carletti, Federico Cingano, Alessandro Conciarelli, Ricardo Correa, Matteo Crosignani, Olivier De Jonghe, Hans Degryse, Domenico Depalo, Giorgio Gobbi, Luigi Guiso, Giuseppe Ilardi, Silvia Magri, Francesco Manaresi, David Marques-Ibanez, Tommaso Oliviero, Daniel Paravisini, Alberto Pozzolo, Joao Santos, Koen Schoors, Neeltje Van Horen, participants at the 2012 CREDIT conference, the workshop “Macroeconomic policies, global liquidity, and sovereign risk”, the 6th CEPR Swiss Winter Conference in Financial Intermediation, the 3rd Mofir Workshop, the 20th Finance Forum, the FIRS 2013 Conference, the 2013 NBER Summer Institute, and seminar participants at the LSE, the Bank of Italy and the New York Fed. The views expressed in this paper do not necessarily reflect those of the Bank of Italy. Corresponding author: Enrico Sette Notes The editor in charge of this paper was Claudio Michelacci. Footnotes 1 This is not particularly restrictive because multiple lending is especially common in Italy (Detragiache, Garella and Guiso 2000, Gobbi and Sette 2014), see also Section 4 for a more detailed discussion of sample selection. 2 When we include bank fixed effects, these absorb the dummy FOREIGN, because no bank changes status (from domestic to foreign or vice versa) in our sample period. 3 Suppose firm 1 borrows from Italian bank A, and foreign bank B in June 2011. Our identification compares credit growth between June and December 2011 by bank A and B to the same firm 1. The pre-crisis period (December 2010–June 2011) allows to better control for possible different dynamics in credit supply by Italian and foreign banks, but having repeated observations for the same firm–bank pair is not strictly necessary for identification. 4 The CR lists all outstanding loan amounts above 30,000 euros that each borrower (both firms and households) has from banks operating in Italy, including branches and subsidiaries of foreign banks. Intermediaries are required by law to report this information. Data on outstanding loan amounts are available at monthly frequency and are of very high quality because intermediaries use the CR as a screening and monitoring device for borrowers. 5 We exclude firms with outstanding bad loans at the beginning of each period, because these are officially classified as losses. 6 We control for mergers and acquisitions among banks. If a firm had a relationship with a bank, and the bank disappears because it is acquired or merged, we track whether there is a new relationship with the newly formed bank, or with the acquirer, in which case we consider the relationship as still existing. 7 Imbens and Wooldridge (2009) show that a normalized difference below 0.25 indicates that the characteristics are balanced (i.e. not statistically different) across samples. 8 We select firms that borrow from at least one foreign and from at least one Italian bank in the pre-crisis period and firms that borrow from at least one foreign and from at least one Italian bank in the crisis period. Essentially we select a repeated cross-section of firms. 9 These data are unconsolidated and for foreign banks they only include the banks’ activity in Italy. 10 As to the relationship-level controls, the coefficient of SHARE is negative and significant, suggesting that banks might reduce lending more intensely towards firms they were initially more exposed to. This effect is weaker during the crisis. The coefficient of DRAWN/GRANTED is positive, but statistically significant only when bank-fixed effects are included. This is consistent with the possibility that a loan commitment is more likely to be increased if a firm is already using available commitments close to the limit. The effect is stronger during the crisis, only when bank*firm fixed effects are included (column 4). The coefficient of the share of OVERDRAFT is positive and significant (we do not have any prior about its effect on credit growth). The effect is stronger during the crisis. 11 All bank balance sheet data are from December 2010 for the pre-crisis period and from June 2011 for the crisis period. We use consolidated and unconsolidated (in case of stand-alone banks) data for Italian banks from the Supervisory reports submitted to the Bank of Italy. Consolidated data for foreign banks are from Bankscope. 12 To address the potential multicollinearity problems generated by the correlation of certain bank-level characteristics with the dummy FOREIGN, we run separate regressions inserting the bank-level variables one by one. The correlation matrix of regressors is shown in Table S4. 13 These data are from the ECB MFI interest rates statistics. 14 Data are from the Supervisory Reports. These regressions are weighted by total assets. To make these tests more comparable to those shown in Tables 5 and 6, we also re-run them weighted by the number of credit relationships, and all results hold. 15 These are dummies defined according to the median of the distribution of each variable, high indicating values above the median, low, below the median. 16 This approach is similar in spirit to that proposed by Abowd, Kramarz, and Margolis (1999) to estimate worker effects in their study of wage premia. 17 The data on interest rates include the flow of interest rates paid in from the firm to the bank and the ”products”, an accounting variable which is the loan amount outstanding times the number of days in which that amount has been outstanding. Dividing the flow of interest by the products we obtain a measure of the interest rate paid on the outstanding loans by the firm. For further details on these data, see Sette and Gobbi (2015). 18 These findings attenuate concerns about the baseline results being driven by a bank-specific demand for credit. In this case, we should find interest rates to be higher for foreign banks, because firm demand for credit to these banks increase. We find instead a combination of higher quantity and lower prices in credit relationship with foreign banks. 19 Every time a bank requests information on a borrower, the query is recorded in the CR, together with the motivation of the request, typically a loan application by a new client. For each application, we check if the bank granted any credit commitment to the loan applicant in the three months following the application. Hence, a loan application submitted to a bank, say, in December 2010, is classified as accepted if we observe that the bank grants credit to the borrower at any point in time between December 2010, the date of the request, and March 2011. 20 We cannot directly observe whether an application has been rejected. Hence, zeros include both rejected applications and applications accepted with a lag longer than three months. 21 A firm posting more than one application may do so because it expects lower chances that any given application is accepted. Posting an application may be costly because it reveals that the applicant already applied for a loan, as banks observe this information in the CR. References Abbassi P. , Iyer R. , Peydrò J. L. , Tous F. ( 2016 ). “ Securities Trading by Banks and Credit Supply: Micro-Evidence .” Journal of Financial Economics , 121 , 569 – 594 . Google Scholar CrossRef Search ADS Abowd J. , Kramarz F. , Margolis D. ( 1999 ). “ High Wage Workers and High Wage Firms .” Econometrica , 67 , 251 – 334 . Google Scholar CrossRef Search ADS Acharya V. , Drechsler I. , Schnabl P. ( 2014 ). “ A Pyrrhic Victory? Bank Bailouts and Sovereign Credit Risk .” Journal of Finance , 69 , 2689 – 2739 . Google Scholar CrossRef Search ADS Acharya V. , Steffen S. ( 2015 ). “ The Greatest Carry Trade Ever? Understanding Eurozone Bank Risks .” Journal of Financial Economics , 115 , 215 – 236 . Google Scholar CrossRef Search ADS Adelino M. , Ferreira M. ( 2016 ). “ Bank Ratings and Lending Supply: Evidence From Sovereign Downgrades .” Review of Financial Studies , 29 , 1709 – 1746 . Google Scholar CrossRef Search ADS Adrian T. , Colla P. , Shin H. S. ( 2013 ). “ Which Financial Frictions? Parsing the Evidence from the Financial Crisis of 2007 to 2009 .” NBER Macroeconomics Annual , 27 , 159 – 214 . Google Scholar CrossRef Search ADS Albertazzi U. , Bottero M. ( 2014 ). “ Foreign Bank Lending: Evidence From the Global Financial Crisis .” Journal of International Economics , 92 , S22 – S35 . Google Scholar CrossRef Search ADS Almeida H. , Cunha I. , Ferreira M. A. , Restrepo F. ( 2017 ). “ The Real Effects of Credit Ratings: The Sovereign Ceiling Channel .” Journal of Finance , 72 ( 1 ), 249 – 290 . Google Scholar CrossRef Search ADS Arteta C. , Hale G. ( 2008 ). “ Sovereign Debt Crises and Credit to the Private Sector .” Journal of International Economics , 74 , 53 – 69 . Google Scholar CrossRef Search ADS Bank of Italy ( 2014 ), Financial Stability Report, April. Battistini N. , Pagano M. , Simonelli S. , ( 2014 ). “ Systemic Risk, Sovereign Yields and Bank Exposures in the Euro crisis .” Economic Policy , 29 , 203 – 251 . Google Scholar CrossRef Search ADS Beck T. , Ioannidou V. , Schäfer L. ( 2017 ). “ Foreigners vs. Natives: Bank Lending Technologies and Loan Pricing .” Management Science , forthcoming . Bentolila M. , Jensen M. , Jiménez G. ( 2017 ). “ When Credit Dries Up: Job Losses in the Great Recession .” Journal of the European Economic Association , forthcoming . Bocola L. ( 2016 ). “ The Pass-Through of Sovereign Risk .” Journal of Political Economy , 124 , 879 – 926 . Google Scholar CrossRef Search ADS Patti E. Bonaccorsi di , Sette E. ( 2016 ). “ Did the Securitization Market Freeze Affect Bank Lending During the Financial Crisis? Evidence From a Credit Register .” Journal of Financial Intermediation , 25 , 54 – 76 . Google Scholar CrossRef Search ADS Borensztein E. , Panizza U. ( 2009 ). “ The Costs of Sovereign Default .” IMF Economic Review , 56 , 683 – 741 . Cetorelli N. , Goldberg L. S. ( 2011 ). “ Global Banks and International Shock Transmission: Evidence From the Crisis .” IMF Economic Review , 59 , 41 – 76 . Google Scholar CrossRef Search ADS Cetorelli N. , Goldberg L. S. ( 2012a ). “ Liquidity Management of U.S. Global Banks: Internal Capital Markets in the Great Recession .” Journal of International Economics , 88 ( 2 ), 299 – 311 . Google Scholar CrossRef Search ADS Cetorelli N. , Goldberg L. S. ( 2012b ). “ Follow the Money: Quantifying Domestic Effects of Foreign Bank Shocks in the Great Recession .” American Economic Review , 102 ( 3 ), 213 – 18 . Google Scholar CrossRef Search ADS Chodorow-Reich G. ( 2014 ). “ The Employment Effects of Credit Market Disruptions: Firm-level Evidence From the 2008–9 Financial Crisis .” Quarterly Journal of Economics , 129 , 1 – 59 . Google Scholar CrossRef Search ADS Cingano F. , Manaresi F. , Sette E. ( 2016 ). “ Does Credit Crunch Investment Down? New Evidence on the Real Effects of the Bank-Lending Channel .” Review of Financial Studies , 29 , 2737 – 2773 . Google Scholar CrossRef Search ADS Coeurdacier N. , Rey H. ( 2013 ). “ Home Bias in Open Economy Financial Macroeconomics .” Journal of Economic Literature , 51 , 63 – 115 . Google Scholar CrossRef Search ADS Correa R , Sapriza H. , Zlate A. ( 2012 ). ”Liquidity Shocks, Dollar Funding Costs, and the Bank Lending Channel During the European Sovereign Crisis .” Federal Reserve Discussion Paper 2012 – 1059 . De Haas R. , Van Horen N. ( 2012 ). “ International Shock Transmission After the Lehman Brothers Collapse Evidence From Syndicated Lending .” American Economic Review Papers & Proceedings , 102 ( 3 ), 231 – 237 . Google Scholar CrossRef Search ADS Dell’Ariccia G. , Goyal R. , Brooks P. Koeva , Pradhan M. , Tressel T. , Pazarbasioglu C. ( 2013 ). “ A Banking Union for the Euro Area .” IMF Staff Discussion Notes 13/01 . Detragiache E. , Garella P. , Guiso L. ( 2000 ). “ Multiple Versus Single Banking Relationships: Theory and Evidence .” Journal of Finance , 55 , 1133 – 1161 . Google Scholar CrossRef Search ADS ECB ( 2015 ). ECB Economic Bulletin, Issue 6 . Gennaioli N. , Martin A. , Rossi S. ( 2014 ). “ Sovereign Default, Domestic Banks, and Financial Institutions .” Journal of Finance , 69 , 819 – 866 . Google Scholar CrossRef Search ADS Gobbi G. , Sette E. ( 2014 ). “ Do Firms Benefit From Concentrating Their Borrowing? Evidence From the Great Recession .” Review of Finance , 18 , 527 – 560 . Google Scholar CrossRef Search ADS Imbens G. W. , Wooldridge J. M. ( 2009 ). “ Recent Developments in the Econometrics of Program Evaluation .” Journal of Economic Literature , 47 , 5 – 86 . Google Scholar CrossRef Search ADS International Monetary Fund ( 2010 ). “ Italy: 2010 Article IV Consultation .” IMF Country Report No. 10/157 . Iyer R. , Peydrò J. L. , da-Rocha-Lopes S. , Schoar A. ( 2014 ). “ Interbank Liquidity Crunch and the Firm Credit Crunch: Evidence From the 2007–2009 Crisis .” Review of Financial studies , 27 , 347 – 372 . Google Scholar CrossRef Search ADS Jiménez G. , Mian A. , Peydrò A. J. , Saurina J. ( 2010 ). “ Local Versus Aggregate Lending Channel: the Effects of Securitization on Corporate Credit Supply .” NBER Working Paper No. 16595 . Google Scholar CrossRef Search ADS Jiménez G. , Ongena S. , Peydrò J. , Saurina J. ( 2012 ). “ Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel With Loan Applications .” American Economic Review , 102 ( 5 ), 2301 – 2326 . Google Scholar CrossRef Search ADS Kahle Kathleen M. , Stulz René M. ( 2010 ). “ Financial Policies and the Financial Crisis: How Important was the Systemic Credit Contraction for Industrial Corporations? ” NBER Working Paper No. 16310 . Khwaja A. I. , Mian A. ( 2008 ). “ Tracing the Impact of Bank Liquidity Shocks: Evidence From an Emerging Market .” American Economic Review , 98 ( 4 ), 1413 – 1442 . Google Scholar CrossRef Search ADS Panetta F. , Angelini P. , Grande G. ( 2014 ). “ The Negative Feedback Loop between Banks and Sovereigns .” Bank of Italy , Occasional Paper No. 213 . Peek J. , Rosengren E. S. ( 1997 ). “ The International Transmission of Financial Shocks: The Case of Japan .” American Economic Review , 87 ( 4 ), 495 – 505 . Peek J. , Rosengren E. S. ( 2000 ). “ Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States .” American Economic Review , 90 ( 1 ), 30 – 45 . Google Scholar CrossRef Search ADS Popov A. , Udell G. F. ( 2012 ). “ Cross-Border Banking, Credit Access, and the Financial Crisis .” Journal of International Economics , 87 , 147 – 161 . Google Scholar CrossRef Search ADS Popov A , Van Horen N. ( 2015 ). “ Exporting Sovereign Stress: Evidence From Syndicated Bank Lending During the Euro Area Sovereign Debt Crisis .” Review of Finance , 19 , 1825 – 1866 . Google Scholar CrossRef Search ADS Reinhart C. M. , Rogoff K. S. ( 2009 ). “ This Time is Different: Eight Centuries of Financial Folly.” Princeton University Press . Schnabl P. ( 2012 ). “ The International Transmission of Bank Liquidity Shocks: Evidence from an Emerging Market .” Journal of Finance , 67 , 897 – 932 . Google Scholar CrossRef Search ADS Sette E. , Gobbi G. ( 2015 ). “ Relationship Lending During a Financial Crisis .” Journal of the European Economic Association , 13 , 453 – 481 . Google Scholar CrossRef Search ADS © The Authors 2017. Published by Oxford University Press on behalf of European Economic Association. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of the European Economic Association Oxford University Press

# Credit Supply During a Sovereign Debt Crisis

, Volume Advance Article (3) – Aug 17, 2017
34 pages

Loading next page...

/lp/ou_press/credit-supply-during-a-sovereign-debt-crisis-Mvntlswcy8
Publisher
Oxford University Press
Copyright
© The Authors 2017. Published by Oxford University Press on behalf of European Economic Association.
ISSN
1542-4766
eISSN
1542-4774
D.O.I.
10.1093/jeea/jvx020
Publisher site
See Article on Publisher Site

### Abstract

Abstract We study the patterns of credit supply in Italy following the burst of the European sovereign debt crisis in 2011. Comparing lending to the same firm, we find that domestic banks reduced credit supply, increased interest rates on credit granted, and lowered the probability of accepting loan applications more than foreign banks, which were less affected by the sovereign crisis. The credit contraction is the consequence of a largely country-specific effect, not explained by heterogeneity in bank characteristics, but associated to a generalized increase in the cost of funding of Italian banks. Looking across firms, we find that credit restrictions by domestic banks were not fully compensated by foreign banks’ lending, implying that Italian firms experienced an aggregate credit shortage. 1. Introduction The sovereign debt crisis that hit European countries in the summer of 2011 had a large effect on global economic performance and on the stability of financial markets. Economic activity in the countries affected by the crisis contracted strongly. So did bank credit. Besides lower demand, the drop in credit may have also been driven by supply factors. A credit crunch occurring during a sovereign debt crisis can be especially dangerous because it can trigger or amplify a contraction in GDP, thus fueling a negative loop between the sovereign crisis and economic activity. This threat is aggravated by the fact that, in this context, governments need to tighten fiscal policy to counteract the tensions, and the effectiveness of monetary policy might be blunted because rising sovereign yields may impair the transmission mechanism. Dell’Ariccia et al. (2013) show that, following the burst of the sovereign debt crisis, changes in the policy rate in the Euro Area had little effects on bank funding costs as the latter were increasingly driven by domestic sovereign yields. Assessing the degree to which credit supply contracts during a sovereign crisis is therefore particularly relevant. Credible identification however necessitates purging observed credit flows from demand components, and observing a subset of banks whose lending policy was plausibly not exposed to the shock. We undertake this task exploiting a rich data set of individual bank–firm relationships from the comprehensive Italian Credit Register (CR), which records lending from all banks, both domestic and foreign, operating in Italy. We first document that after the sovereign debt crisis started, Italian banks contracted credit supply about 3 percentage points more than foreign banks operating in Italy, but headquartered in countries not (or less) exposed to the sovereign crisis. Importantly, this is obtained comparing credit flows from different banks to the same firm, thus accounting for credit demand. Similar patterns emerge looking at other lending margins: domestic banks increased interest rates and reduced the probability of accepting new applications more than their foreign competitors. Second, we show that this domestic bank effect on credit granted persists even after controlling for bank individual characteristics and provide evidence that this effect reflects the increased cost of funding experienced by all domestic banks. Indeed we find that country-level measures of banks’ cost of funding have a negative effect on credit supply and their impact persists also when balance sheet features are included in the model. Moreover, being a domestic bank has a strong power in explaining the increase in the cost of funding at the bank level, again even after accounting for individual bank characteristics. We interpret the effect of this country-specific component as the impact of the sovereign debt crisis, potentially uncovering a new lending channel linking sovereign shocks to contractions to credit supply, through banks’ cost of funding. Recent evidence confirms that sovereign crises largely translate into an increase in the cost of funding for domestic intermediaries (Battistini, Pagano, and Simonelli 2014; Panetta, Angelini, and Grande 2014). This may be due to the increased riskiness of banks’ assets, which are exposed to the shock both in terms of holdings of domestic sovereign bonds and in terms of loans to domestic private sector. Moreover, the cost of funding may also increase due to the lower ability of the domestic government to support ailing intermediaries. By the same argument, the balance sheets of banks headquartered in countries that are less exposed to the sovereign crisis are less likely to respond to the shock. Furthermore, for domestic banks, access to funds might become more costly also due to the mechanical downgrades of financial institutions following the downgrade of their sovereign of residence (so-called sovereign ceiling), as highlighted in Adelino and Ferreira (2016). All these factors would imply that foreign banks may represent a suitable comparison group for estimation purposes. We find consistent evidence in the data. As shown by Figure 1, the correlation between the riskiness of the sovereign and the riskiness of the banking sector increased sharply during the sovereign debt crisis relative to the Lehman crisis (Panel A); by the same token, the association between bank retail funding costs and the premium requested on sovereign debt strongly intensified (Panel B). This strong nexus between country level sovereign risk and the cost of raising funds supports the idea of using foreign banks as a subset of intermediaries more insulated from the crisis. Figure 1. View largeDownload slide Sovereign CDS, banks CDS, and banks’ cost of funding. Figure 1(a) shows scatter plots of country average CDS spreads for 5-year senior debt of major banks (vertical axis) and CDS spreads 10-year sovereign bonds of the home country (horizontal axis) based on daily data for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data are from Thomson Reuters, banks are the major banks of each country with available CDS data for the whole sample. Unit: basis points. Figure 1(b) shows scatter plots of country spread between the yields of 10-year government bonds relative to the German (vertical axis) and the country averages of the spread between rates on new deposits by firms and households with agreed maturities of up to 1 year and the 6 month average of the EONIA rate (horizontal axis), for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data on sovereign spread are from Bloomberg, deposit rates are from ECB MFI interest rates statistics. Unit: basis point for the sovereign spread, percentage points for the deposit rate. Figure 1. View largeDownload slide Sovereign CDS, banks CDS, and banks’ cost of funding. Figure 1(a) shows scatter plots of country average CDS spreads for 5-year senior debt of major banks (vertical axis) and CDS spreads 10-year sovereign bonds of the home country (horizontal axis) based on daily data for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data are from Thomson Reuters, banks are the major banks of each country with available CDS data for the whole sample. Unit: basis points. Figure 1(b) shows scatter plots of country spread between the yields of 10-year government bonds relative to the German (vertical axis) and the country averages of the spread between rates on new deposits by firms and households with agreed maturities of up to 1 year and the 6 month average of the EONIA rate (horizontal axis), for the 6 months following the default of Lehman (October 2008–March 2009) and for the 6 months following the start of the sovereign debt crisis (July 2011–December 2011). Data on sovereign spread are from Bloomberg, deposit rates are from ECB MFI interest rates statistics. Unit: basis point for the sovereign spread, percentage points for the deposit rate. Sovereign debt crises rarely occur as stand-alone events. They are often accompanied by recessions, when demand for credit drops, and the quality of borrowers deteriorates, posing the task of disentangling demand from supply effects. We tackle these identification challenges using a rich data set of individual bank–firm relationships from the comprehensive Italian CR. This allows us to employ the identification strategy pioneered by Khwaja and Mian (2008) and to restrict our analysis to firms borrowing from at least one foreign and one domestic bank and include firm-time fixed effects to absorb all time-varying observed and unobserved firm heterogeneity. In principle, sovereign debt crises might be triggered by government-financed measures aimed at sustaining distressed banks. The subsequent weakening of public finances may in turn feed back into worsening of banks’ balance sheets (Acharya, Drechsler, and Schnabl 2014; Reinhart and Rogoff 2009). We argue that this was not the case in Italy: the total cumulative government support to the banking sector relative to GDP was negligible between 2008 and 2013 (ECB 2015). Moreover, the inclusion of firm-time and bank fixed effects attenuates the potential biases induced by a feedback from past lending policies of banks, undermining their health and thus triggering the shock to the sovereign. We also extend our main findings into three directions. First, we test if the behavior of foreign banks differs depending on whether they are incorporated as subsidiaries or as branches. We find that the differential behavior between Italian and foreign banks is due to subsidiaries, whose business model is more similar to that of Italian banks, thus reducing concerns that our findings are driven by foreign banks specializing in particular markets, such as those of Mergers and Acquisitions (M&As) or syndicated loans. Second, we look at the heterogeneity of the drop of credit across firms. We find some limited evidence that the difference in credit by domestic and foreign banks was larger for firms with less liquid buffers, and for firms with a worse credit rating. Finally, we estimate the aggregate effect of the sovereign crisis on credit supply, finding that Italian firms were unable to compensate the negative domestic bank effect tapping on credit from foreign banks. The paper is structured as follows: the next section summarizes the related literature, Section 3 describes the empirical strategy, Section 4 presents the data set and the main descriptive statistics, Section 5 contains the results of our baseline specification, their interpretation, and a set of robustness checks, Section 6 examines the effect of bank heterogeneity, Section 7 contains extensions, Section 8 presents the results on the aggregate effect, Section 9 shows results on the cost of credit, and on the probability that loans applications are accepted, Section 10 concludes. 2. Related Literature Our paper contributes to the literature studying the effects of sovereign debt crises on bank activity and real outcomes. Earlier works look at this issue using cross-country panels of different crises episodes over long periods of time, with data at the country or industry level (Arteta and Hale 2008; Borensztein and Panizza 2009). More recent papers have studied specifically the European sovereign crisis, looking at different aspects of the nexus crisis-credit-output: the effects of holdings of Greece, Ireland, Italy, Portugal and Spain (GIIPS) sovereign debt on syndicated loans by non-GIIPS banks (Popov and Van Horen 2015), the correlation between holdings of domestic sovereign bonds and credit supply (Gennaioli, Martin, and Rossi 2014), the transmission of downgrades of sovereign bonds to bank bonds and to credit (Adelino and Ferreira 2016), and the direct effects of sovereign ratings on firm outcomes, net of credit risk and fundamentals (Almeida et al. 2017). Other work focused on asset–liability management of banks during the sovereign crisis, documenting phenomena of “carry trade”, with banks borrowing short on wholesale markets and going long on GIIPS countries sovereign bonds (Acharya and Steffen 2015) and showing how U.S. branches of European banks experienced a run on their deposits, mainly from the U.S. money market funds (Correa, Sapriza, and Zlate 2012). Our paper contributes to this literature in several ways. We document that Italian banks suffered a generalized funding shock associated to the sovereign crisis and that this translated into a severe tightening of credit supply, compared with credit by foreign banks that did not suffer an equal increase in their domestic sovereign risk. The availability of a large sample of loans to nonfinancial firms from the Italian CR allows us to identify the net supply effect, in line with the bank lending channel literature pioneered by Khwaja and Mian (2008) based on loan-level data. Furthermore, because our sample contains both large and small firms, we provide a much broader picture than that obtained using syndicated loan data or information from public corporations. Moreover, we provide evidence on different margins of lending and finally estimate the real effects of the sovereign lending channel by showing that the sovereign shock had an aggregate impact on firms’ access to bank credit. Our findings complement the work by Bocola (2016) who develops a business cycle model estimated on Italian data, showing that a sovereign shock translates into higher funding costs for banks, and then into lower lending, with strong recessionary effects. Because our identification relies on comparing behavior of Italian and foreign banks, our paper also contributes to the large literature on global banks and on the international transmission of shocks. Many channels have been investigated: cross-border lending (Peek and Rosengren 1997), lending by affiliates (Peek and Rosengren 2000), internal capital markets (Cetorelli and Goldberg 2012a), and international interbank markets (Schnabl 2012). Our work concentrates on lending by foreign branches and subsidiaries operating in Italy. More recently, there has been a large interest on the specific role of global banks in the financial crisis and the Great Recession. Various papers have examined different destination markets: Cetorelli and Goldberg (2011) look at emerging economies, Cetorelli and Goldberg (2012b) at the United States, Popov and Udell (2012) at Central and Eastern Europe countries, Albertazzi and Bottero (2014) at Italy, and De Haas and Van Horen (2012) explore the international market of syndicated loans. Overall, this recent literature documents how global intermediaries contribute to “export” tensions, thus highlighting a mechanism of international transmission of shocks from the parent bank to host countries. Notably, we shed light on a “bright” side of the presence of foreign intermediaries in a country hit by a shock, showing how, being more shielded by the shock than domestic banks, foreign banks contribute to dampen the sovereign shock. Interestingly, our results can also be interpreted in line of the finding of Abbassi et al. (2016) who show that trading banks (mainly large global banks) increase their investments in risky securities to profit from fire-sales during the crisis and cut back on other activities, including credit supply to domestic firms. In the same fashion, Italian firms represent a risky investment because the country is hit by a large shock, and foreign banks could have stepped in to extract rents from credit-constrained Italian firms. This behavior could exert some negative spillovers on domestic activities of foreign banks, insofar the latter are funding constrained. In such case, the bright side of the presence of global banks for Italian firms would be accompanied by negative side effects in other markets. 3. Empirical Strategy Determining whether a credit crunch occurs during a sovereign shock poses important challenges. A first one is the identification of a set of banks more and less exposed to the shock. We argue that foreign banks operating in Italy are a good candidate to represent a set of intermediaries relatively less affected by sovereign tensions. There are various reasons why domestic banks of a country hit by a sovereign shock should be affected more severely than foreign ones operating in the same country (Battistini, Pagano, and Simonelli 2014; Panetta, Angelini, and Grande 2014). The asset side of banks located in a country hit by a shock drops in value and becomes riskier. Sovereign bonds held in the trading book are marked to market and generate an immediate loss to banks. Even if banks do not hold government debt in their trading book, they may nevertheless expect to incur future losses. Moreover, the credit risk of loans to domestic customers also rises (Bocola 2016). The increased riskiness of the asset side exerts two effects. The first one is to reduce banks’ willingness to assume additional risk. The second one, and possibly much larger than the first, operates via a higher cost of funding or even impaired access to funding. Funding cost rises also because collateralized transactions backed by government debt become more expensive, due to higher haircuts required on the collateral. Furthermore, the potential weakening of the implicit or explicit government guarantee on banks due to strains in public finances likely determines an increase in the cost of raising funds in the country. These factors affect all banks exposed to a country hit by a sovereign shock, yet typically the impact is much higher for those headquartered in that country. Domestic banks, in fact, typically hold a much larger share of their home-country sovereign securities and lend mostly to domestic borrowers. The home bias of investors, not only of banks, is a long-standing puzzle in the macro-financial literature (Coeurdacier and Rey 2013). In the case of banks, a high exposure to domestic lenders is a physiological structural characteristic due to the proximity needed in the lending process. The open question, which lies beyond the scope of this paper, is why banks do not diversify the exposure to domestic firms and households by holding securities issued by foreign entities. We take this behavior as given, because it does not compromise our identification strategy. Furthermore, another mechanism that generates an increase in the cost of funding for domestic banks only is the mechanical transmission of downgrades from sovereign to bank bonds (Adelino and Ferreira 2016). As a consequence of all these channels, domestic banks suffer from an increase in the risk of their home country government debt way more than foreign banks operating in the same country. This mostly translates into a higher cost of funding for domestic banks. We find evidence consistent with these phenomena in the data. Data on Credit Default Swaps (CDS) spreads confirm a high correlation between banks’ wholesale funding costs and sovereign CDS in the second half of 2011, following the outburst of the sovereign crisis (Figure 1, Panel (a)). In the same period also the cost of deposits, a proxy of retail cost of funding, was strongly correlated with the sovereign bonds yields (Figure 1, Panel (b)). Interestingly, compared to the Lehman crisis, the degree of correlation between measures of banks’ and sovereigns’ funding costs was much more pronounced. This suggests that a common country component may be a key driver of the higher cost of funding, possibly in addition to individual bank characteristics. Notably, foreign banks operating in Italy are headquartered in countries whose sovereign yields increased mildly (or even decreased) during the crisis (France, Germany, USA, Austria). This allows us to distinguish banks into two groups as being more and less exposed to the increase in sovereign risk: domestic (Italian) and foreign banks. There is also prima facie evidence that foreign and Italian banks behaved differently, in terms of lending policies, after the Italian sovereign risk for Italy rose in June 2011. Figure 2 shows the abrupt increase in the spread between the 10-year-Italian-government bond, and the 10-year-German-government bond. Descriptive statistics show that the variation in credit in the 6 months between December 2010 and mid-2011 and the variation observed between June and December 2011 was basically unchanged for foreign banks, although it varied remarkably for Italian banks. On average, lending by domestic banks decreased by more than three percentage points to -7% in those 6 months, as shown in Table 1; foreign banks kept lending almost steadily at -5% in the two periods. Also, graphical evidence reassures that, before the crisis, Italian banks did not display a differential trend in credit supply relative to foreign banks. If domestic banks prior to the crisis had already been in distress, their deleveraging might have started earlier than June 2011. Figure 3 shows the 6-month change in the log credit committed by Italian and foreign banks. Although prior to the crisis the two series moved similarly, since June 2011 credit from domestic banks decreased at a much faster rate than credit from foreign banks. Figure 2. View largeDownload slide Spread between 10-year Italian BTP and German Bund (percentage points). The figure shows the time series of the spread between the 10-year Italian BTP and the 10-year German Bund (percentage points). Data are from Thomson Reuters. The figure shows that the spread remained roughly constant in the first half of 2011, and started increasing abruptly, reaching historically high levels, from July 2011. Figure 2. View largeDownload slide Spread between 10-year Italian BTP and German Bund (percentage points). The figure shows the time series of the spread between the 10-year Italian BTP and the 10-year German Bund (percentage points). Data are from Thomson Reuters. The figure shows that the spread remained roughly constant in the first half of 2011, and started increasing abruptly, reaching historically high levels, from July 2011. Figure 3. View largeDownload slide Change of credit committed by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks. The unit is logpoints. Data are from the Italian CR. Figure 3. View largeDownload slide Change of credit committed by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks. The unit is logpoints. Data are from the Italian CR. Table 1. Descriptive statistics of main dependent variable. Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Notes: Delta log credit is the change in the log of credit committed to firms by banks (multiplied by 100). The change for the pre-crisis period is computed as the change between end of June 2011 and end of December 2010; the change for the pre-crisis period is computed as the change between end of December 2011 and end of June 2011. Data are from the Italian CR. View Large Table 1. Descriptive statistics of main dependent variable. Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Mean Median StdDev N Obs Log changes in percent All banks ΔLog Credit −5.36 0.00 42.18 664,198 ΔLog Credit—Pre crisis −4.13 0.00 41.18 332,563 ΔLog Credit—Crisis −6.60 0.00 43.12 331,635 Domestic banks ΔLog Credit −5.38 0.00 40.72 478,379 ΔLog Credit—Pre crisis −3.73 0.00 37.86 239,686 ΔLog Credit—Crisis −7.04 0.00 43.34 238,693 Foreign banks ΔLog Credit −5.31 −0.93 45.72 185,819 ΔLog Credit—Pre crisis −5.15 −0.28 48.71 92,877 ΔLog Credit—Crisis −5.47 −1.45 42.51 92,942 Notes: Delta log credit is the change in the log of credit committed to firms by banks (multiplied by 100). The change for the pre-crisis period is computed as the change between end of June 2011 and end of December 2010; the change for the pre-crisis period is computed as the change between end of December 2011 and end of June 2011. Data are from the Italian CR. View Large A potential concern for this identification approach is that a sovereign shock may be driven by poor banks’ conditions due to excessive lending to weak borrowers prior to the crisis. Sovereign debt crises can indeed be fueled by banking crises, because governments disburse vast amounts of money to rescue troubled intermediaries (Reinhart and Rogoff 2009; Acharya et al. 2014). However, Italian banks weathered the post-Lehman crisis relatively unscathed (IMF 2010 Article IV consultation on Italy) and as a consequence the cumulative government support to the banking sector, between 2008 and 2013 has been a mere 0.2% of GDP (ECB 2015). Although the abrupt increase in sovereign spreads occurring at the end of June 2011 reflects many factors, including potential fears of the inability of rescuing the banking sector if needed, it can be reasonably argued that it did not reflect past or contemporaneous weaknesses of the banking sector driven by their past lending to nonfinancial firms, which is what is needed to support our identification approach. The second key identification challenge is the need to disentangle credit supply from demand. Typically sovereign tensions are accompanied by deteriorating economic conditions, inducing firms to scale down their investment plans and decrease demand for credit. Moreover, domestic banks may lend to a different set of firms (e.g. firms with weaker balance sheets, riskier firms, etc.) than foreign banks. Hence, it is critical to properly control for firm level demand for credit, for firms’ riskiness, and, more generally, for firm unobserved heterogeneity. For example, during economic downturns credit quality of existing loans deteriorates and this feeds back onto credit demand. The richness of our data set from the Credit Register allows us to control for all of these factors. We identify the impact of sovereign risk on credit supply by comparing the pre-crisis and the crisis patterns of credit supplied to the same firm by at least two banks, affected by the sovereign debt crisis to different degrees.1 The inclusion of firm-period fixed effects in all regressions, similarly to what Khwaja and Mian (2008) and more recently Jiménez et al. (2012) do, enables us to control for all firm-level unobserved heterogeneity that affects the dynamics of credit committed and of its cost in each period. Moreover, it controls for potential biases arising from the possibility that the crisis was triggered by fears about the soundness of corporate borrowers, weakening potential concerns about causality running from lenders or borrowers to sovereign yields. Following this empirical strategy, we estimate a model in which the observational unit is a credit relationship between a firm and a bank, and we compare two periods: the first half of 2011 (pre-crisis) and the second half of 2011 (crisis). Using a pre-crisis period allows us to control for existing differences in the supply of credit by Italian and foreign banks. Formally, $$\Delta \mathit{credit}_{i,j,t} =\beta _{1} \mathit{FOREIGN}_{j}+\beta _{2} \mathit{FOREIGN}_{j}\ast \mathit{CRISIS}_{t}+\alpha _{i,t}+\varepsilon _{i,j,t}$$ (1) where $$\Delta \mathit{credit}_{i,j,t}$$ is the difference in the log credit committed by bank j to firm i in period t. The dummy FOREIGN equals 1 if bank j is foreign, zero if the bank is domestic. The term FOREIGN*CRISIS is an interaction between the dummy FOREIGN and the dummy variable CRISIS which equals 1 in the second half of 2011. αi, t is a full set of firm-period fixed effects (they absorb the dummy CRISIS, which therefore does not appear in the equation above). We also run all regressions plugging bank fixed effects, which control for all bank time-invariant unobserved heterogeneity, including systematic differences in banks’ business models, geographical reach, etc.2 All regressions also include variables capturing relationship-level characteristics. The first is the share of total credit to firm i supplied by bank j (SHARE). On the one hand, SHARE measures the exposure of bank j towards firm i, and this is negatively correlated with loan growth; on the other hand, SHARE may be a proxy of the strength of the bank–firm relationship, therefore suggesting a positive relationship with credit growth. The second variable is the share of drawn over credit committed by bank j to firm i (DRAWN/COMMITTED). This control measures how intensively available credit lines are used. If a line of credit is already used close to the available limit, firms may be more likely to apply for an extension. The third variable is the share of overdraft over total committed credit by bank j to firm i (OVERDRAFT). Overdraft loans (revolving credit lines) may be more volatile than other forms of loans. Including these regressors helps controlling for differences in bank–firm relationships across domestic and foreign banks. We also include the interactions between these relationship-level controls and the dummy CRISIS. The main parameter of interest is β2, which captures the differential behavior of foreign banks relative to domestic banks during the crisis. It is identified on firms that borrow from at least one Italian and one foreign bank in at least one period.3 The key identifying assumption is that firms do not have a bank-specific demand for credit (Bonaccorsi di Patti and Sette 2016, Jimenez et al. 2010, Khwaja and Mian 2008). To reduce concerns that this assumption may be violated in our setting, we include a set of relationship-level controls (SHARE, DRAWN/COMMITTED, OVERDRAFT) that capture systemic differences in the structure of bank–firm relationships. Moreover, our findings on lending from branches and subsidiaries (Section 7.1) and those on the cost of credit (Section 9.1) contribute to attenuate possible concerns about the violation of this assumption. Our identification allows us to check more accurately whether Italian and foreign banks behaved similarly before the crisis; because the series shown in Figure 3 do not account for the different composition of firms borrowing from the two types of banks, we also compute the dynamics of credit committed in deviation from firm-period averages. We expect credit from domestic and foreign banks, net of firm effects, to move similarly until June 2011, and to start diverging afterwards. This is precisely what happens, as shown in Figure 4 (see Khwaja and Mian 2008 for a similar representation of the data). This is the graphical counterpart of the baseline model (equation 1 above). Figure 4. View largeDownload slide Change in credit committed, net of firms-period effects, by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks, and net of their means. For every firm, the average firm-level credit committed is subtracted by credit committed by Italian and foreign banks. The stocks of demeaned credit committed are then collapsed across all firms. Then changes relative to the June 2011-level (start of the crisis) are computed. The unit is percentage points. Data are from the Italian CR. Figure 4. View largeDownload slide Change in credit committed, net of firms-period effects, by Italian and foreign banks. The figure shows the time series of the change relative to June 2011 in total credit committed to the firms of our sample, split by Italian and foreign banks, and net of their means. For every firm, the average firm-level credit committed is subtracted by credit committed by Italian and foreign banks. The stocks of demeaned credit committed are then collapsed across all firms. Then changes relative to the June 2011-level (start of the crisis) are computed. The unit is percentage points. Data are from the Italian CR. 4. Data and Descriptive Statistics Our empirical analysis is based on data at the bank–firm relationship level on outstanding loan amounts from the comprehensive Italian CR.4 The data set includes both granted and drawn credit. We focus on credit granted (credit commitments), because this better captures a decision of banks to supply credit. Drawn credit is influenced by the decision of the borrower to use available lines, which is largely affected by demand. December 2010–June 2011 represents the pre-crisis period, June 2011–December 2011 represents the crisis period. We do not extend the sample beyond December 2011, because on December 22 the ECB enacted the first 3-year Long-Term Refinancing Operation, which eased tensions in funding markets, and thus confounded the effect of the sovereign shock. We do not extend the sample before 2011 to avoid that our results might be influenced by events occurring in previous periods. In particular, 2010 was characterized by a slow recovery from the impact of the Lehman shock, and 2009 was part of the “full” crisis period following the default of Lehman. We include all nonfinancial firms with outstanding credit in the CR, including very small firms, both incorporated firms and sole proprietorship.5 Because our identification strategy relies on comparing the behavior of a foreign and an Italian bank lending to the same firm, we select firms that borrow from at least one Italian and one foreign bank in each period.6 This is not a restrictive condition. Table A.1 in the Online Appendix shows characteristics of firms in our sample and of firms borrowing only from Italian banks. The table shows normalized differences (Imbens and Wooldridge 2009) of the balance sheet items indicated in each row. Although firms in our sample are larger and have a slightly lower ratio of operating profits to interest expenses than firms borrowing only from Italian banks, differences are not statistically significant for any of the firm characteristics.7 We aggregate credit at the banking group level, because lending and funding policies are typically decided at the headquarter level. If a firm borrows from two banks belonging to the same banking group, we consider this as a single relationship. Overall, our sample includes 664,198 bank–firm relationships over the two periods, involving 164,470 firm-period couples, with 92,620 distinct firms sampled at least in one period.8 Median credit committed was around 800,000 euros in June 2011, consistent with the presence of a large number of small firms in the sample. Firms were borrowing on average from four banks. Half of the sample firms operate in the services sector, 30% in manufacturing and energy, 12% in construction, 8% in agriculture. Key balance sheet characteristics of firms in the sample are shown in Table 2. Table 2. Descriptive statistics of firms. Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Notes: The table shows descriptive statistics of the firms of our sample that are included in the Firm Register (CERVED) for which we can observe balance sheet information. Leverage is defined as total assets−equity−debt versus shareholders divided by total assets. Liquidity is the sum of cash and cash equivalents. Z-score is Altman Z-score as computed by CERVED, and it represents an estimate of firms’ default probability. It is computed on a scale from 1 to 9 and higher values indicate a higher probability of default. View Large Table 2. Descriptive statistics of firms. Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Mean Median P25 P75 Std Dev Total assets (000 euros) 21,059.47 3,442.5 1,212 10,651.5 188,749.3 Leverage 77.08 81.59 66.28 91.53 18.78 Liquidity/Assets 6.31 2.51 0.58 7.96 9.17 Ebitda/Interest Expenses 1,270.01 402.74 169.94 1,000 3,393.35 Z-Score 5.38 5 4 7 6.06 Notes: The table shows descriptive statistics of the firms of our sample that are included in the Firm Register (CERVED) for which we can observe balance sheet information. Leverage is defined as total assets−equity−debt versus shareholders divided by total assets. Liquidity is the sum of cash and cash equivalents. Z-score is Altman Z-score as computed by CERVED, and it represents an estimate of firms’ default probability. It is computed on a scale from 1 to 9 and higher values indicate a higher probability of default. View Large Table 3. Home country of the banks included in the sample and changes in spreads. Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Notes: The table shows the number of bank–firm relationships included in the sample, by banks’ home country. Spreads are the difference between the yield of the 10-year government bond of the country the bank is headquartered in and the 10-year German Bund. The change in spreads is computed as the difference between the average spread in March 2011 and the average spread in January 2011 for the pre-crisis period and as the difference between the average spread in September 2011 and the average spread in July 2011 for the crisis period. Averages are computed on daily data on spreads from Thomson Datastream. View Large Table 3. Home country of the banks included in the sample and changes in spreads. Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Country Number of relationships % ΔSpread—Pre crisis ΔSpread—crisis Basis points Basis points Austria 8,395 1.26 −0.4 33 Switzerland 207 0.03 −9.4 45 Germany 22,846 3.44 0 0 Spain 4,353 0.66 3.2 83 France 134,954 20.32 −3.7 38 UK 2,312 0.35 −44 34 Japan 463 0.07 −13 98 Netherlands 2,908 0.44 5.1 15 Slovenia 42 0.01 −7.6 110 United States 9,339 1.41 −37 8 Total foreign 185,819 27.98 Italy (domestic) 478,379 72.02 12 192 Notes: The table shows the number of bank–firm relationships included in the sample, by banks’ home country. Spreads are the difference between the yield of the 10-year government bond of the country the bank is headquartered in and the 10-year German Bund. The change in spreads is computed as the difference between the average spread in March 2011 and the average spread in January 2011 for the pre-crisis period and as the difference between the average spread in September 2011 and the average spread in July 2011 for the crisis period. Averages are computed on daily data on spreads from Thomson Datastream. View Large Table 3 shows the distribution of bank–firm relationships by home country of the lender. More than a quarter of the relationships involve foreign banks. The majority are French, followed by German, then US and Austrian. Table 3 also shows the change in the spread of the 10-year sovereign security relative to the 10-year German Bund of the same maturity, between January and March 2011 for the pre-crisis period, and between July 2011 and September 2011 for the crisis period. The spread increased sharply, by almost 200 basis points, for Italy (see also Figure 2), and to a lesser extent for Slovenia (110 basis points), Japan and Spain (98 and 83 basis points, respectively). Prior to the crisis, spreads remained roughly unchanged. To better assess to what extent foreign banks differ from domestic ones, it is useful to describe other characteristics of the two groups. The importance of foreign banks in Italy increased steadily in the past 20 years. The share of total assets that they held increased from below 4% in 1992 to 17.5% at the end of 2011, when two subsidiaries of foreign banks ranked among the top ten banking groups operating in Italy. A large fraction of the liabilities of branches and subsidiaries of foreign banks is represented by interbank transfers from their headquarters that raise funds either in their home country or in the international wholesale markets (70% for branches and 40% for subsidiaries); local retail funding, which may also have been affected by the sovereign debt crisis, is much less important for foreign banks relative to Italian ones. Subsidiaries of foreign banks have a business model similar to that of domestic banks. Typically subsidiaries of foreign banks were formerly domestically owned banks acquired by a foreign bank holding company, thus they have an extensive network of outlets and had been active in Italy for a long time, so they had established tight relationships with domestic borrowers over time. Among the subsidiaries of foreign banks included in our sample are: BNL-BNP Paribas, Cariparma-Credit Agricole, Deutsche Bank Italia, Santander, that are reasonably comparable with large and medium Italian commercial banks. The entry of foreign banks in Italy stopped at the beginning of the financial crisis, so changes in ownership did not occur during our sample period. Overall, there is limited evidence that the composition of loans of domestic banks changed significantly after acquisitions by foreign banks. In 2011, subsidiaries of foreign banks were supervised by Italian authorities, which were also responsible of their resolution in case of distress. However, they could also indirectly rely on the public support of their headquarters’ country if their default threatened the viability of the whole banking group. Panel A of Table 4 shows descriptive statistics on the activity of foreign and Italian banks included in our sample. Overall we include 567 banks, 49 of which foreign. Foreign banks hold a lower share of loans to households but a similar share of loans to nonfinancial firms and a similar average number of branches as domestic banks.9 Table 4. Descriptive statistics of domestic and foreign banks. Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Notes: Panel A shows descriptive statistics of the loan portfolio of domestic and foreign banks operating in Italy. Data are from June 2011 Supervisory reports submitted by intermediaries to the Bank of Italy. The data for foreign banks refer only to the banks’ operations in Italy. Total domestic loans include loans to Italian nonfinancial firms, households and nonprofit organizations. Panel B shows descriptive statistics of the main balance sheet variables for domestic and foreign banks operating in Italy. Data for Italian banks are from the consolidated Supervisory reports submitted to the Bank of Italy. Data for foreign banks are from Bankscope, and are at the consolidated level including operations outside Italy. Data on holdings of GIIPS (Greece, Ireland, Italy, Portugal and Spain) sovereign bonds for foreign banks have been hand-collected from banks’ published balance sheet. Data refer to 567 banks and are from the June 2011 balance sheet. T1 RATIO is the ratio of Tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, GIIPS HOLDINGS is the ratio of holdings of sovereign bonds from GIIPS to total assets, ROA (return on assets) is the ratio of profits (or losses) to total assets, SIZE is the natural logarithm of total assets. View Large Table 4. Descriptive statistics of domestic and foreign banks. Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Panel A: Activity of domestic and foreign banks Domestic Foreign Mean Std. Dev Mean Std. Dev. Loans to households/Totaldomestic loans 32.8 13.2 23.1 31.8 Loans to firms/Total domestic loans 64.3 13.5 64.8 33.0 Average number of branches in Italy 60.3 369.3 50.1 175.2 Panel B: Balance Sheet Variables of Banks T1 RATIO 14.98 5.06 11.94 3.73 INTERBANK 3.28 5.98 17.94 9.07 GIIPS HOLDINGS 14.22 9.51 1.58 2.01 ROA 0.22 0.43 0.29 0.53 SIZE 6.05 1.54 12.63 1.59 Number of banks 518 49 Notes: Panel A shows descriptive statistics of the loan portfolio of domestic and foreign banks operating in Italy. Data are from June 2011 Supervisory reports submitted by intermediaries to the Bank of Italy. The data for foreign banks refer only to the banks’ operations in Italy. Total domestic loans include loans to Italian nonfinancial firms, households and nonprofit organizations. Panel B shows descriptive statistics of the main balance sheet variables for domestic and foreign banks operating in Italy. Data for Italian banks are from the consolidated Supervisory reports submitted to the Bank of Italy. Data for foreign banks are from Bankscope, and are at the consolidated level including operations outside Italy. Data on holdings of GIIPS (Greece, Ireland, Italy, Portugal and Spain) sovereign bonds for foreign banks have been hand-collected from banks’ published balance sheet. Data refer to 567 banks and are from the June 2011 balance sheet. T1 RATIO is the ratio of Tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, GIIPS HOLDINGS is the ratio of holdings of sovereign bonds from GIIPS to total assets, ROA (return on assets) is the ratio of profits (or losses) to total assets, SIZE is the natural logarithm of total assets. View Large Panel B of Table 4 shows descriptive statistics of the main balance sheet characteristics of the banks included in our sample. These data are on a consolidated basis and are from June 2011, yet variation over our sample period is very limited. Relative to domestic banks, foreign banks are on average larger, less capitalized, more relying on interbank funding, and less exposed to GIIPS sovereign securities. They do not differ significantly in terms of profitability, as measured by ROA. These differences are much less pronounced when foreign banks are compared to the subset of larger Italian banks. The top 50 domestic banks differ significantly from foreign banks only along the exposure to risky sovereign debt (higher for domestic banks) and in size (smaller for domestic banks). 5. Results for the Baseline Model Results from the estimation of equation (1) are reported in Table 5. The first column shows the effect of the dummy FOREIGN on the growth of credit committed without fixed effects of any sort; even in this basic specification, the coefficient of the dummy FOREIGN without interactions is not statistically significant and large and significant when interacted with the dummy CRISIS, showing that the behavior of domestic and foreign banks differs during the crisis. Results are quantitatively and qualitatively unchanged once we take into account observed and unobserved heterogeneity at the bank, firm, and time level. Similar results are indeed obtained when we plug firm fixed effects, which absorb all time-invariant observed and unobserved firm heterogeneity (column 2) and when we allow this heterogeneity to be time-varying (column 3). The difference in the estimates is not large, suggesting that firm demand for credit does not play a very strong role. This finding is in line with other work using CR data (Bentolila, Jensen, and Jiménez 2017, Cingano, Manaresi, and Sette 2016, Jiménez et al. 2010). The coefficients of the dummy FOREIGN and of the interaction FOREIGN*CRISIS change very little when relationship-level controls (and their interaction with the dummy crisis) are also included in the regression (column (4)), to account for specific relationship features that might vary across lending banks and over time.10 Table 5. Baseline. (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. In the seventh column, the sample includes firms in the Firm Register that borrow from banks for which we observe all balance sheet data; in the 8th column, the sample includes firms in the Firm Register for which we observe balance sheet data. CRISIS is a dummy variable equal to one if data are from the June 2011–December 2011 period. SHARE OF TOTAL CREDIT is the share of total credit committed by the bank to the firm, DRAWN OVER COMMITTED is the ratio of drawn to committed credit in the relationship, OVERDRAFT is the share of overdraft loans to total loans granted by the bank to the firm. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Table 5. Baseline. (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.427 −1.063 −1.177 −1.086 −0.869 −0.795 (1.080) (0.918) (0.892) (0.875) (0.830) (1.007) FOREIGN BANK*CRISIS 3.000** 2.778** 2.958** 3.217** 3.076** 1.680** 3.317** 2.607** (1.317) (1.194) (1.167) (1.248) (1.246) (0.823) (1.297) (1.215) SHARE OF TOTAL CREDIT −0.131*** −0.155*** −3.796*** −0.144*** −0.161*** (0.0165) (0.0148) (0.0853) (0.0158) (0.0188) DRAWN OVER COMMITTED 0.283 1.389* 9.039*** 1.069 −0.0141 (0.842) (0.709) (0.903) (0.792) (1.029) OVERDRAFT 11.18*** 10.04*** 45.17*** 11.18*** 10.75*** (0.659) (0.696) (3.943) (0.589) (0.895) SHARE OF TOTAL CREDIT*CRISIS 0.0235** 0.0230** −0.117*** 0.0257** 0.0124 (0.00935) (0.00947) (0.0202) (0.00998) (0.0134) DRAWN OVER COMMITTED*CRISIS 1.076 1.022 2.559** 0.892 1.137 (1.317) (1.335) (1.096) (1.374) (1.543) OVERDRAFT 3.768*** 3.801*** 13.02*** 3.636*** 3.842*** (1.289) (1.281) (0.984) (1.350) (1.341) CRISIS −3.314*** −4.857*** (0.741) (0.649) CONSTANT −3.728*** (0.302) FIRM FIXED EFFECTS no yes no no no no no no FIRM*TIME FIXED EFFECTS no no yes yes yes yes yes yes BANK FIXED EFFECTS no no no no yes no no no FIRM*BANK FIXED EFFECTS no no no no no yes no no Observations 664,198 664,198 664,198 664,198 664,194 576,518 644,647 449,248 R2 0.001 0.175 0.275 0.283 0.288 0.706 0.287 0.251 The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. In the seventh column, the sample includes firms in the Firm Register that borrow from banks for which we observe all balance sheet data; in the 8th column, the sample includes firms in the Firm Register for which we observe balance sheet data. CRISIS is a dummy variable equal to one if data are from the June 2011–December 2011 period. SHARE OF TOTAL CREDIT is the share of total credit committed by the bank to the firm, DRAWN OVER COMMITTED is the ratio of drawn to committed credit in the relationship, OVERDRAFT is the share of overdraft loans to total loans granted by the bank to the firm. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large The specification of column (4), which we consider our benchmark, indicates that during the crisis the behavior of the two types of banks differs: credit committed by foreign banks grew by about 3 percentage points more than credit committed by domestic banks. This is an economically significant effect, because the average rate of growth of credit during the crisis is −6.6%. Moreover, the measure of credit supply is the growth rate (log change) in the stock of credit commitments in each bank–firm relationship, and a 3% increase is a sizable amount. Results are also robust to the inclusion of bank fixed effects (column 5), which absorb the dummy FOREIGN. Indeed we do not observe difference in the coefficients across columns (4) and (5). This suggests that bank time-invariant unobserved heterogeneity is not strongly correlated with the impact of the shock, which appears to mainly depend on the nationality of the bank holding company. Column (6) shows estimates from a regression including a full set of firm*bank fixed effects, together with the firm*time fixed effects. This is a very demanding specification that allows controlling for all unobserved time invariant characteristics of individual firm–bank relationships, for example the possibility that certain banks specialize in certain types of loans, or that certain relationships have a longer duration than others, etc. Remarkably, results still hold, although the size of the effect is somewhat smaller than in the baseline: credit committed by foreign banks grew by about 1.7 percentage points more than credit by domestic banks. The difference with the baseline estimates is not statistically significant, though, and the estimates of the other relationship level controls strongly increase in size, likely due to a high correlation with the bank–firm fixed effects. For robustness, we also run our benchmark specification of column (4) on the two subsets of observations for which we avail of either bank or firm balance sheet characteristics and that will be used for estimation in later sections. Results on these two subsets, respectively shown in columns (7) and (8), are basically unchanged. As an additional robustness check, we run a placebo experiment using the periods before June 2011 to test whether the different behavior of domestic and foreign banks started precisely after the burst of the sovereign debt crisis. We consider the period December 2009–December 2010, setting the fictitious event at June 2010. Next, we extend the period to June 2011, and we set the event at June 2010 or at December 2010. In all cases (Supplemental Table S2), neither the dummy FOREIGN, nor the interaction between the dummy FOREIGN and the dummy CRISIS are significant. Coefficients are also small in size. Finally, to further test the robustness of the main results, we also run regressions taking averages of credit committed in the period before (the six month from January 2011 to June 2011) and after (the six months from July 2011 to December 2011) the crisis and computing log changes of the average credit committed and as shown in Table A.3 in the Online Appendix. We also run the baseline regression comparing foreign banks and the subsample of the top 50 domestic banks, those for which differences in balance sheet characteristics are much more limited, except for the holdings of GIIPS sovereign bonds and for size, and results (not reported) hold through. 6. Bank Heterogeneity Table 5 suggests that bank’s nationality drives lending policies during the sovereign debt crisis. In this section, we investigate to what extent the dummy FOREIGN is capturing bank characteristics that affect lending growth differentially across domestic and foreign banks when the crisis hit and we provide further evidence on the drivers of this foreign bank effect. We explore the first issue by augmenting our benchmark regression (column 4 of Table 5) with time-varying bank balance sheet characteristics that are likely to influence lending.11 This set of variables includes the ratio of GIIPS sovereign debt holdings to total assets (GIIPS HOLDINGS) to measure the direct exposure of banks to sovereign debt, the Tier 1 ratio (Tier 1 capital to risk-weighted assets, TIER 1), which is a measure of capitalization, the ratio of interbank funding (deposits and repos) to assets (INTERBANK), which captures banks’ reliance on wholesale funding, the most volatile funding component that dried-up sharply in the second half of 2011 (Cappelletti 2013 and Correa, Sapriza, and Zlate 2012), return on assets (ROA) as a measure of profitability, a set of dummies for the quartiles of bank’s assets to control for bank size, and a dummy identifying mutual banks (all bank-level controls, including the size and mutual bank dummies, are interacted with the dummy crisis). Results are shown in Table 6. The interaction between the dummy FOREIGN and the dummy CRISIS is always positive and significant, indicating that the main result is robust to the inclusion of time-varying bank-level controls. Furthermore, bank-level characteristics play a limited role in explaining the different behavior of domestic and foreign banks. In particular, banks less reliant on interbank funding and banks that are more profitable tend to lend more, but the impact of these characteristics does not vary across the two periods that we consider. Interestingly, GIIPS HOLDINGS is never significant, indicating that the holdings of sovereign bonds of “peripheral” countries has no effect on credit supply once we control for the dummy FOREIGN. Importantly, the effect of the interaction FOREIGN*CRISIS is still positive and significant when all bank controls are included (column 5) and even when bank fixed effects are plugged in (column 6).12 Furthermore, the size of the coefficient of the interaction FOREIGN*CRISIS is very stable across specifications and numerically very close to that of the baseline, indicating that the inclusion of individual bank characteristics has little correlation with the effect of being a foreign or domestic bank. Table 6. Regressions with bank balance sheet variables. (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on bank characteristics. FOREIGN is a dummy equal to 1 if the bank is foreign, CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. GIIPS HOLDINGS is the ratio of bank’s portfolio holdings of government bonds of Greece, Ireland, Italy, Portugal and Spain to total assets. T1 RATIO is the ratio of tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, ROA is return on assets. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Balance sheet data for domestic banks are from the Supervisory reports submitted to the Bank of Italy, for foreign banks are from Bankscope. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Table 6. Regressions with bank balance sheet variables. (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 (1) (2) (3) (4) (5) (6) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −0.855 −1.101 −0.204 −0.387 0.717 (0.882) (0.979) (0.818) (0.836) (0.982) FOREIGN BANK*CRISIS 3.313** 3.716** 3.215*** 3.042** 2.670* 3.226* (1.399) (1.452) (1.210) (1.362) (1.597) (1.837) GIIPS HOLDINGS 0.00352 0.0205 −0.499 (0.0639) (0.0558) (2.456) GIIPS HOLDINGS*CRISIS −0.0214 −0.00622 0.0420 (0.179) (0.194) (35.95) T1 RATIO 0.109 −0.0603 −0.512 (0.141) (0.134) (0.357) T1 RATIO*CRISIS −0.0201 0.0777 −0.0411 (0.183) (0.160) (0.179) INTERBANK −0.132*** −0.159*** 0.0129 (0.0486) (0.0465) (0.213) INTERBANK*CRISIS 0.00412 0.0316 −0.0296 (0.0872) (0.0894) (0.0834) ROA 2.899*** 2.963*** 5.190* (1.057) (0.957) (2.831) ROA*CRISIS −2.838 −1.930 −0.843 (1.755) (1.586) (1.752) FIRM*TIME FIXED EFFECTS yes yes yes yes yes yes BANK FIXED EFFECTS no no no no no yes RELATIONSHIP CONTROLS yes yes yes yes yes yes BANK SIZE DUMMIES yes yes yes yes yes yes D MUTUAL BANK yes yes yes yes yes yes Observations 654,598 646,411 655,316 654,538 644,610 644,607 R2 0.284 0.287 0.284 0.284 0.287 0.291 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on bank characteristics. FOREIGN is a dummy equal to 1 if the bank is foreign, CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. GIIPS HOLDINGS is the ratio of bank’s portfolio holdings of government bonds of Greece, Ireland, Italy, Portugal and Spain to total assets. T1 RATIO is the ratio of tier 1 capital to risk-weighted assets, INTERBANK is the ratio of interbank funding to total assets, ROA is return on assets. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Balance sheet data for domestic banks are from the Supervisory reports submitted to the Bank of Italy, for foreign banks are from Bankscope. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Overall this evidence confirms that foreign banks, less affected by the sovereign shock, increased their credit supply more than domestic banks during the sovereign crisis and that being FOREIGN is not just disguising balance sheet differences between domestic and foreign banks. Next, we explore to what extent the foreign bank effect is due to a differentiated pattern of bank funding cost at the country level, as suggested by the aggregate evidence shown in Figure 1. As a first test, we substitute the dummy FOREIGN with country-level averages of banks' cost of funding.13 Table 7 shows that an increase in the sovereign spread has a negative effect on the supply of credit during the crisis. As yields on corporate bonds, including bank bonds, raise with the yield on sovereign bonds, this suggests an impact on lending through the higher cost of issuing bonds for banks. Similarly, an increase in the cost of deposits and, in particular, in the cost of deposits to households (again averaged at the country-level) has a negative impact on bank lending. These results also hold when focusing only on the crisis periods and when controlling for bank balance sheet characteristics (Tables A.5 and A.6 in the Online Appendix). Table 7. Effect of banks’ cost of funding. (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on different measures of banks’ cost of funding. DELTA SPREAD is the change in the spread between the 10-year sovereign bond of the country the bank is headquartered in and the 10-year German Bund; DELTA DEPOSITS is the change in the average cost of deposits and DELTA DEPOSITS HOUSEHOLDS is the change in the average cost of deposits from households, both measured in the country the bank is headquartered in. These two measures are from the ECB MFI interest rates statistics. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). All relationship level controls are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large Table 7. Effect of banks’ cost of funding. (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 (1) (2) (3)) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ Δ SPREAD 10.65** (4.845) Δ SPREAD * CRISIS −11.91** (4.720) Δ DEPOSITS 0.744 (1.434) Δ DEPOSITS *CRISIS −8.792** (3.514) Δ DEPOSITS HOUSEHOLDS 0.865 (2.215) Δ DEPOSITS HOUSEHOLDS * CRISIS −5.994* (3.113) RELATIONSHIP CONTROLS yes yes yes FIRM*TIME FE yes yes yes Observations 664,198 648,651 648,651 R2 0.283 0.287 0.287 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on different measures of banks’ cost of funding. DELTA SPREAD is the change in the spread between the 10-year sovereign bond of the country the bank is headquartered in and the 10-year German Bund; DELTA DEPOSITS is the change in the average cost of deposits and DELTA DEPOSITS HOUSEHOLDS is the change in the average cost of deposits from households, both measured in the country the bank is headquartered in. These two measures are from the ECB MFI interest rates statistics. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). All relationship level controls are also interacted with the dummy CRISIS. The sample includes bank–firm relationships from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1 View Large To further corroborate the hypothesis that the foreign bank effect captures a larger increase in the cost of funding for domestic banks, we run a bank-level regression of a measure of banks’ average cost of funding on the dummy FOREIGN and bank-level balance sheet variables.14 Results in Table 8 show that the dummy FOREIGN has a negative and significant coefficient in all specifications. This suggests that the change in the average cost of funding of foreign banks has been significantly lower than that of Italian banks after the crisis, and that this common country effect is over and above due to differences in individual bank balance sheet characteristics. Table 8. Change in banks’ cost of funding. (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 Notes: The table shows results of regressions of the change in the average cost of funding at the bank level on the dummy FOREIGN and on bank balance sheet characteristics. Column (1) shows results for the period June 2011–December 2011 (crisis period), columns (2) and (3) for both the June 2011–December 2011 and the December 2010–June 2011. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. In columns (2) and (3), the mutual bank dummy and the bank size dummies are also interacted with the dummy CRISIS. Data are available for Italian banks and for subsidiaries of foreign banks. Standard errors double clustered at the bank level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large Table 8. Change in banks’ cost of funding. (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 (1) (2) (3) Crisis only Pre- and post-crisis crisis Pre- and post-crisis with bank fe VARIABLES Δ COST OF FUNDING FOREIGN −0.227*** 0.297* (0.0770) (0.165) FOREIGN*CRISIS −0.524*** −0.513*** (0.182) (0.185) GIIPS HOLDINGS −0.00932** 0.00338 0.0366 (0.00422) (0.00721) (0.0386) GIIPS HOLDINGS*CRISIS −0.0127 −0.0207* (0.00836) (0.0106) T1RATIO −0.00808 −0.0177 0.0177 (0.00832) (0.0114) (0.0509) T1RATIO*CRISIS 0.00963 0.0217 (0.0141) (0.0178) INTERBANK 0.00979** −0.0131* 0.0373 (0.00457) (0.00744) (0.0271) INTERBANK*CRISIS 0.0228*** 0.0205* (0.00873) (0.0112) ROA −0.0513 0.0191 0.0422 (0.0652) (0.116) (0.284) ROA*CRISIS −0.0704 −0.0857 (0.133) (0.165) CRISIS 1.136*** 1.033*** (0.192) (0.215) BANK SIZE DUMMIES yes yes yes D MUTUAL BANK yes yes yes BANK FE no no yes Observations 501 1,007 996 R2 0.287 0.962 0.971 Notes: The table shows results of regressions of the change in the average cost of funding at the bank level on the dummy FOREIGN and on bank balance sheet characteristics. Column (1) shows results for the period June 2011–December 2011 (crisis period), columns (2) and (3) for both the June 2011–December 2011 and the December 2010–June 2011. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. In columns (2) and (3), the mutual bank dummy and the bank size dummies are also interacted with the dummy CRISIS. Data are available for Italian banks and for subsidiaries of foreign banks. Standard errors double clustered at the bank level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large 7. Extensions We extend our baseline results in two directions. First, we explore differences in the behavior of foreign banks depending on whether they are incorporated in Italy as subsidiaries or branches. Second, we explore heterogeneous effects across firms of different size and financial strength. 7.1. Branches and Subsidiaries To shed more light on our results, we exploit the variability within foreign banks, classifying them separately into branches and subsidiaries. The latter are very similar to domestic banks in terms of extension of their network of outlets and business model, the former concentrate their activity in certain areas of the country and are typically specialized in specific market segments, such as syndicated loans, leasing, etc. This test is important both as a robustness check of our baseline, and as a contribution to the literature on global banks. Table 9 shows results of our analysis. Columns (1) and (3) display estimates from regressions run on the subsample of firms borrowing from at least one domestic bank and at least one subsidiary of foreign banks. Results are similar to those of the baseline regressions. Columns (2) and (4) display estimates from regressions run on the subsample of firms borrowing from at least one domestic bank and at least one branch of foreign banks. In this case, we find no difference in the credit quantity supplied by domestic banks and by branches of foreign banks. Table 9. Distinguishing branches and subsidiaries. (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN and on bank characteristics. Columns (1) and (3) include firms borrowing from Italian banks and from subsidiaries of foreign banks. Columns (2) and (4) include firms borrowing from Italian banks and from branches of foreign banks. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. Balance sheet data are from the Supervisory reports submitted to the Bank of Italy. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large Table 9. Distinguishing branches and subsidiaries. (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 (1) (2) (3) (4) Subsidiaries Branches Subsidiaries Branches VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN BANK −1.841* −0.223 −0.666 1.816 (1.107) (1.264) (0.838) (1.386) FOREIGN BANK*CRISIS 5.353*** 0.528 4.383*** −0.973 (1.195) (0.950) (1.407) (1.484) GIIPS HOLDINGS −0.0263 0.0507 (0.0465) (0.0598) GIIPS HOLDINGS*CRISIS −0.0213 −0.0741 (0.196) (0.203) T1 RATIO 0.101 −0.173 (0.0893) (0.144) T1 RATIO*CRISIS 0.0412 0.343** (0.155) (0.156) INTERBANK −0.148*** −0.119** (0.0352) (0.0479) INTERBANK*CRISIS 0.101 0.00825 (0.0941) (0.0877) ROA 3.778*** 2.790*** (0.711) (1.052) ROA*CRISIS −3.148** −2.596 (1.569) (1.587) RELATIONSHIP CONTROLS yes yes yes yes BANK SIZE DUMMIES no no yes yes D MUTUAL BANK no no yes yes FIRM*TIME FE yes yes yes yes Observations 554,922 530,779 545,749 518,342 R2 0.290 0.289 0.294 0.292 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN and on bank characteristics. Columns (1) and (3) include firms borrowing from Italian banks and from subsidiaries of foreign banks. Columns (2) and (4) include firms borrowing from Italian banks and from branches of foreign banks. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. BANK SIZE DUMMIES is a set of dummies, one for each of the three largest quartiles of the banks’ distribution by size. D MUTUAL BANK is a dummy variable equal to 1 if the bank is a mutual bank. All relationship level controls, the mutual bank dummy, and the bank size dummies are also interacted with the dummy CRISIS. Balance sheet data are from the Supervisory reports submitted to the Bank of Italy. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large These results indicate that the effect found in the baseline regression is mainly driven by a different behavior of Italian banks relative to subsidiaries of foreign banks. By contrast, we find a smaller difference between domestic banks and branches of foreign banks, despite the fact that the latter enjoy better access to funding than domestic banks. We interpret these results as evidence that the organizational structure of foreign banks is relevant for lending decisions. Foreign banks’ organizational form has been found to be an important driver of their lending policy in particular during crises: Cetorelli and Goldberg (2012b) find that parent banks, when hit by a funding shock, reallocate liquidity towards affiliate locations that are important for the parent bank revenue streams that are then relatively protected from liquidity reallocation within the organization. The presence of a subsidiary is a proxy for the importance of the Italian market in the portfolio of the global bank. These findings are important because they attenuate concerns that our main results could be driven by those foreign banks specialized in particular markets, such as those of M&As or syndicated loans (branches). The difference in credit supply is found precisely when we compare Italian banks to the set of foreign banks that are more similar to Italian banks, the subsidiaries. Moreover, these results suggest that for foreign banks to play a mitigating role in the transmission of shocks to the domestic banking sector, they have to be well established in the country, with a large network of outlets, possibly reflecting a higher ability to collect soft information about borrowers (Beck, Ioannidou, and Schäfer 2017). 7.2. Firm Heterogeneity We further extend our results to test whether Italian banks reduced credit mainly to certain categories of borrowers. This is an important extension for two sets of reasons. First, it allows to understand to what extent the drop in credit by domestic versus foreign banks has been heterogeneous depending on firm size. Previous work (Iyer et al. 2014, Kahle and Stulz 2010) shows that the biggest brunt of credit shocks is borne by small firms, which are also less able to substitute bank credit with other sources of finance than larger firms. If this is the case, policy measures aimed at increasing the ability of smaller firms to access external funds would be especially important. Second, it allows us to ascertain to what extent the higher credit by foreign banks has gone to financially sound rather than to weaker firms. Again, this is crucial to uncover whether foreign banks were able to take more risk, thanks to their lower exposure to the shock, or whether they “cream-skimmed” borrowers, exploiting their higher ability to grant credit. To perform these tests, we merge our data with balance sheet information from the Firm Register (Cerved). We run the baseline regressions splitting the sample of borrowers into large/small firms, high/low leverage firms, high/low roa, high/low ebitda to interest expenses, bad/good Z-score.15 Results are shown in Table 10. We find little difference in lending policies across firm size (columns 1 and 2) and across firms ability to repay interest expenses with operating profits (columns 7 and 8). The difference in lending of foreign banks compared to Italian banks is instead numerically larger in the case of firms with lower liquidity and of those with bad Z-score, although none of the differences is statistically significant. This provides some limited evidence suggesting that Italian banks cut credit more to more fragile firms relative to foreign banks. Table 10. Heterogeneous effects across firms. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN, splitting the sample according to the firm characteristics listed in each column. Large or High indicates above the median, Small or Low indicates below the median. “Bad Z-Score” identifies firms with a Z-Score in the bottom three classes. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. All relationship level controls are also interacted with the dummy CRISIS. Firm balance sheet data are from the Firm Register (CERVED). Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large Table 10. Heterogeneous effects across firms. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Firm size Leverage Liquidity/Assets Ebitda /Int Exp Z-Score Large Small High Low High Low High Low Bad Good VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN −0.881 −0.577 −1.124 −0.536 0.0226 −1.540 −0.280 −1.372 −2.332* −0.220 (1.102) (1.045) (1.147) (0.971) (0.923) (1.201) (1.112) (1.022) (1.345) (0.952) FOREIGN*CRISIS 2.650** 2.539* 2.273* 2.883** 2.351** 2.871* 2.735** 2.518* 3.833** 2.140* (1.340) (1.413) (1.364) (1.197) (1.081) (1.468) (1.285) (1.302) (1.714) (1.115) REL-LEVEL CTRLS yes yes yes yes yes yes yes yes yes yes FIRM*TIME FE yes yes yes yes yes yes yes yes yes yes Observations 328,404 120,457 205,031 243,632 212,207 222,917 216,559 226,958 114,213 333,625 R2 0.226 0.359 0.259 0.244 0.251 0.248 0.250 0.247 0.263 0.244 Notes: The table shows results of the regressions of the change in log committed credit, as defined in Table 1, on the dummy FOREIGN, splitting the sample according to the firm characteristics listed in each column. Large or High indicates above the median, Small or Low indicates below the median. “Bad Z-Score” identifies firms with a Z-Score in the bottom three classes. CRISIS is a dummy equal to 1 if data are from the period June 2011–December 2011. Bank-level controls are defined in Table 6. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). They are also interacted with the dummy CRISIS. All relationship level controls are also interacted with the dummy CRISIS. Firm balance sheet data are from the Firm Register (CERVED). Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1 View Large 8. The Aggregate Effect The results discussed so far are based on coefficients estimated comparing the behavior of a domestic and a foreign bank lending to the same borrower (within estimator). However, firms might compensate the reduction in credit from domestic banks with more loans from foreign banks. Estimates from a simple firm-level regression is likely to be biased, though, because changes in the log of total credit at the firm level also reflect firm-level demand for credit, changes in firm financial strength, and the like. A method to estimate unbiasedly the firm-level (aggregate) impact of the supply shock induced by the crisis on the growth of credit commitments has recently been proposed by Jiménez et al. (2010). However, their methodology does not allow to easily obtain standard errors and thus to conduct inference. In this paper, we use an equivalent estimation procedure, which directly yields standard errors for the unbiased effect of the proxy for the exposure of firms to the credit supply shock (Cingano, Manaresi, and Sette 2016). We first estimate firm-fixed effects from the baseline model at the bank–firm level. Then we plug these estimated firm effects in a firm-level equation in which the dependent variable is the growth of total credit committed to firms by banks (including new relationships) and the measure of the exposure to the credit supply shock is the initial share of credit committed by foreign banks.16 Standard errors are estimated by block-bootstrapping at the level of the main bank (the bank with the largest share of credit), to take into account the fact that firm fixed effects are estimated regressors. Formally, from the baseline model (equation 1), we obtain an estimate of the firm-period fixed effect $$\hat{\alpha }_{i,t}$$. As a second step, we estimate \begin{equation*} \Delta \mathit{credit}_{i,t}=\beta _{1}\overline{ \mathit{FOREIGN}_{i}}+\beta _{2}\overline{ \mathit{FOREIGN}_{i}}\ast \mathit{CRISIS}_{t}+\hat{\alpha }_{i,t}+\varepsilon _{i,t}, \end{equation*} where $$\overline{ \mathit{FOREIGN}_{i}}$$ is the share of credit committed by foreign banks, that is the average at the firm level of the dummy FOREIGN weighted by the share of credit to the firm held by each bank. A more detailed description of this approach can be found in the Appendix. Results are shown in Table 11. Column (1) shows results of a regression excluding the estimated firm effects. The interaction term between the dummy FOREIGN and the dummy CRISIS is positive and significant. This indicates that firms are not able to fully substitute credit from domestic banks by increasing credit from foreign banks. However, as argued above, this result may be biased. Column (2) shows estimates including the firm effect. Table 11. Aggregate effect. (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 Notes: The table shows results of the regressions where the dependent variable is the growth of credit committed at the firm level. All independent variables are firm-level averages, weighted by the share of credit committed by each bank to the firm at the beginning of each period (December 2010 pre-crisis period, June 2011 post-crisis period). In particular, FOREIGN is the share of credit committed by foreign banks to the firm. FIRM EFFECT is a vector including the firm fixed effects estimated from equation (1). Data are from the Italian CR. Block-bootstrapped (clustered at bank level) standard errors in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large Table 11. Aggregate effect. (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 (1) (2) VARIABLES $$\Delta \mathit{LOG(CREDIT)}$$ FOREIGN 1.086 −0.328 (1.780) (1.084) FOREIGN*CRISIS 4.360*** 3.307*** (1.309) (1.086) CRISIS −3.323*** −0.759** (0.508) (0.335) SHARE −0.109*** −0.110*** (0.0118) (0.00693) DRAWN / GRANTED −0.0176*** 0.0149*** (0.00414) (0.00300) OVERDRAFT 0.0710*** 0.105*** (0.00617) (0.00239) FIRM EFFECT 0.689*** (0.0241) Observations 164,470 164,470 Notes: The table shows results of the regressions where the dependent variable is the growth of credit committed at the firm level. All independent variables are firm-level averages, weighted by the share of credit committed by each bank to the firm at the beginning of each period (December 2010 pre-crisis period, June 2011 post-crisis period). In particular, FOREIGN is the share of credit committed by foreign banks to the firm. FIRM EFFECT is a vector including the firm fixed effects estimated from equation (1). Data are from the Italian CR. Block-bootstrapped (clustered at bank level) standard errors in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large The interaction between the dummy FOREIGN and the dummy CRISIS is still positive and significant, although the size of the coefficient is smaller: if the share of credit that a firm obtained before the crisis from foreign banks increases by one standard deviation (23 percentage points), credit growth during the crisis is about 0.8 percentage points higher. This is a large effect because the median credit growth at the firm-level in the crisis period is −2.8% (the mean is −3.8%). Finally, the estimated firm effect is highly significant and positive, consistent with the hypothesis that it is capturing firm–level demand for credit. Overall, these results show that firms have not been able to fully substitute credit from domestic banks with credit from foreign banks: during the sovereign crisis, the “aggregate” effect on the supply of credit at the firm-level was strong. In principle, firms may have been able to substitute bank credit with market-based external finance. This has been the case in other countries during the 2007–2008 financial crisis. Adrian, Colla, and Shin (2013) show that U.S. borrowers substituted from bank-based finance to market-based finance; Abbassi et al. (2016) find that German firms that experience credit reduction due to securities trading by banks during the crisis partly compensated it by issuing bonds. The substitution of bank credit with other sources of external finance is especially difficult during the sovereign crisis for Italian firms. First, most of them, including those in our sample, are small and medium-sized enterprises (see Table 2) for which bank credit represents the only source of external finance, possibly complemented by trade credit. Second, in the second half of 2011 in Italy the cost of other sources of finance increased significantly also for nonfinancial borrowers. In particular, bond yields raised at unprecedented highs together with sovereign yields, both because of the sovereign ceiling phenomenon (Adelino and Ferreira 2016) and because of the higher riskiness of Italian nonfinancial borrowers during the crisis. The difficulty of substituting bank credit with other sources of finance for Italian firms in our sample period is confirmed by the low level of bond issuance during 2011, basically negligible for SMEs (Bank of Italy, 2014). 9. Other Margins As a last step in our analysis, we test the impact of the sovereign shock on other key margins of lending: interest rates and the probability a loan application is accepted. 9.1. Effect on Interest Rates We first test the impact of the sovereign shock on interest rates. We use information on the interest rates from the CR, which collects these data from a representative sample of banks (110 intermediaries, some of which belonging to the same bank consolidated entity, representing over 80% of the total market for loans in Italy). The data set contains information disaggregated by loan type (revolving credit lines, term loans, loans backed by receivables).17 We estimate a version of equation (1) in which the dependent variable is alternatively the change in the Annual Percentage Rate (APR) on revolving credit lines and that on term loans (see Khwaja and Mian 2008 or Chodorow-Reich 2014 for a similar approach). We compute the change in the APR for the pre-crisis (June 2011–December 2010) and crisis (December 2011–June 2011) periods. We use the APR net of fees and commissions, which are typically computed on credit granted while the interest rates we observe are estimated on the basis of the actual usage of the credit line. Our results also hold if we use interest rates gross of fees and commissions. Importantly, all regressions still include firm*period fixed effects. Results are shown in Table 12. Columns (1) and (2) show estimates of the regressions for the change in the APR on revolving credit lines. Column (1) does not include bank fixed effects. Before the crisis, there was no difference between domestic and foreign banks. After the crisis, foreign banks increased rates on revolving credit lines by about 21 basis points less than domestic banks lending to the same firm. Column (2) shows the same regression including bank fixed effects. Results are essentially unchanged. Columns (3) and (4) (the latter includes bank fixed effects) show estimates of the same regressions run for the change in the APR (net of fees and commissions) on term loans. Results are confirmed: foreign banks increase interest rates by about 15 basis points less than domestic banks.18 Table 12. Effect on the cost of credit. (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 Notes: The table shows regressions of the change in the Annual Percentage Rate (APR) on revolving credit lines (column 1 and 2) and on term loans (column 3 and 4) granted by banks to nonfinancial firms in Italy on the variable FOREIGN, a dummy equal to 1 if the bank is a foreign bank. Changes are computed between December 2011 and June 2011 for the crisis period and between June 2011 and December 2010 for the pre-crisis period. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). The regressions include interactions of the relationship level controls and the dummy CRISIS. The sample includes bank–firm relationships from the subsection of the Italian CR reporting information on interest rates over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large Table 12. Effect on the cost of credit. (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 (1) (2) (3) (4) VARIABLES Δ APR-revolving Δ APR-term FOREIGN −0.0442 −0.0165 (0.0609) (0.0158) FOREIGN*CRISIS −0.205* −0.206* −0.149** −0.149** (0.111) (0.111) (0.0618) (0.0615) RELATIONSHIP CONTROLS yes yes yes yes FIRM*TIME FIXED EFFECTS yes yes yes yes BANK FIXED EFFECT no yes no yes Observations 203,042 203,042 134,323 134,323 Notes: The table shows regressions of the change in the Annual Percentage Rate (APR) on revolving credit lines (column 1 and 2) and on term loans (column 3 and 4) granted by banks to nonfinancial firms in Italy on the variable FOREIGN, a dummy equal to 1 if the bank is a foreign bank. Changes are computed between December 2011 and June 2011 for the crisis period and between June 2011 and December 2010 for the pre-crisis period. RELATIONSHIP CONTROLS are SHARE OF TOTAL CREDIT, DRAWN OVER COMMITTED, and OVERDRAFT (defined in Table 5). The regressions include interactions of the relationship level controls and the dummy CRISIS. The sample includes bank–firm relationships from the subsection of the Italian CR reporting information on interest rates over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ****p<0.01, **p<0.05, *p<0.1. View Large 9.2. Probability of Accepting a Loan Application In this section, we examine whether Italian and foreign banks displayed any difference in the likelihood to grant loans to new clients. In line with Jiménez et al. (2012), we use data on loan applications recorded in the CR to analyze the probability of acceptance of new credit. We collect data on all the requests recorded as loan applications between October 2010 and March 2011 and between July 2011 and December 2011, pre-crisis and crisis period, respectively. If we observe that subsequently banks grant credit to the applying firms, we classify the application as accepted. 19 Our dependent variable is a dummy equal to 1 if the application of firm j to bank i is accepted, 0 otherwise, and we estimate a linear probability model.20 We estimate models both including and excluding firm*time fixed effects. Including firm*time fixed effects is important to control for applicant unobservables, but forces us to use the sample of firms posting at least two loan applications in a period. Such firms may be different, likely worse, than the average firm applying for a loan.21 A stand-out descriptive feature of the frequency of accepted applications is that it sharply dropped during the crisis, to 9% between June 2011 and March 2012 from the 37% observed in the three previous quarters. Results shown in columns (1) to (4) of Table 13 confirm the evidence of the aggregate data: foreign banks were less likely to accept a loan application prior to the crisis, and the difference disappeared during the sovereign shock, suggesting that domestic banks tightened their acceptance rate significantly after the sovereign debt crisis. Results are similar across regressions including and excluding firm*period fixed effects, and including and excluding bank fixed effects. Table 13. Probability of accepting a loan application. (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 Notes: The table shows regressions of a dummy variable equal to 1 if the loan application has been accepted by the bank on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. The sample includes information on loan applications and on granted credit from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1. View Large Table 13. Probability of accepting a loan application. (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 (1) (2) (3) (4) VARIABLES D(accept=1) D(accept=1) D(accept=1) D(accept=1) FOREIGN −0.151*** −0.111*** (0.0500) (0.0302) FOREIGN*CRISIS 0.141*** 0.0973*** 0.109*** 0.0715*** (0.0474) (0.0286) (0.0363) (0.0219) CRISIS −0.297*** −0.263*** (0.0411) (0.0288) FIRM*TIME FE no yes no yes BANK FE no no yes yes Observations 926,736 366,743 926,711 366,689 Notes: The table shows regressions of a dummy variable equal to 1 if the loan application has been accepted by the bank on the variable FOREIGN, a dummy equal to one if the bank is a foreign bank. The sample includes information on loan applications and on granted credit from the Italian CR over the period December 2010–December 2011, including all firms financed by at least one Italian and one foreign bank. Standard errors double clustered at the bank and firm-level in parentheses. ***p<0.01, **p<0.05, *p<0.1. View Large Overall, these results indicate that domestic banks, more affected by the shock, diminished their propensity to accept loan applications and raised interest rates more than foreign banks. 10. Concluding Remarks This paper shows that lending by Italian banks declined relatively more than lending by foreign banks after the summer of 2011, when the European sovereign debt crisis hit Italy. In particular, we find that lending by domestic banks declined by 3 percentage points more than lending by foreign banks. These findings do not just reflect differences in bank balance sheet characteristics across foreign and domestic banks, such as the strength of their capital position, profitability, holdings of sovereign bonds. In fact, our results indicate that the drop in lending has a strong country-specific component, depending on the location of banks’ headquarter. This country-specific component is related to the large increase in funding costs that affected Italian banks in association with the sovereign shock, rather homogeneously and independently of individual bank characteristics. We interpret the effect of this country-specific component as the impact of the sovereign debt crisis, potentially uncovering a new lending channel linking sovereign shocks to negative shocks to credit supply, through banks’ cost of funding. We find that such differential behavior between Italian and foreign banks is due to subsidiaries only, whose business model is more similar to that of Italian banks, providing more evidence on the higher relative importance of being headquartered in countries unscathed by sovereign shocks versus differences in business models. Furthermore, our results indicate that firms have not been able to fully substitute the decrease of loans by Italian banks with loans by foreign banks, pointing to an aggregate effect of the sovereign shock on credit supply. Finally, our findings show that the transmission of the sovereign shock also occurred through higher interest rates and through a significant drop in the acceptance rate of loan applications. Appendix Derivation of the Aggregate Effect The relationship level equation is the following: \begin{equation*} \Delta \mathit{credit}_{i,j,t}=\beta _{1} \mathit{domestic}_{j}+\beta _{2} \mathit{domestic}_{j}\ast \mathit{CRISIS}_{t}+\alpha _{i,t}+\varepsilon _{i,j,t}, \end{equation*} where $$\Delta \mathit{credit}_{i,j,t}$$ is the growth rate of credit to firm i by bank j at time t. Then, we take the average of both sides of this equation weighted by the share of credit held by each bank as follows: \begin{align*} \sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}&\ast \frac{ \mathit{credit}_{j,t}}{ \sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}=\beta \sum _{j=1}^{n_{i}} \mathit{domestic}_{j}\ast \frac{ \mathit{credit}_{j,t}}{\sum _{j=1}^{n_{i}} \Delta \mathit{credit}_{i,j,t}} \\ &+\beta \sum _{j=1}^{n_{i}} \mathit{domestic}_{j}\ast \mathit{CRISIS}_{t}\ast \frac{ \mathit{credit}_{j,t}}{ \sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}\\ &+\sum _{j=1}^{n_{i}}\frac{ \mathit{credit}_{j,t} }{\sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}\alpha _{i,t}\!+\! \sum _{j=1}^{n_{i}} \frac{ \mathit{credit}_{j,t}}{\sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}\varepsilon _{i,j,t}, \end{align*} where \begin{equation*} \sum _{j=1}^{n_{i}}\frac{ \mathit{credit}_{j,t}}{\sum _{j=1}^{n_{i}}\Delta \mathit{credit}_{i,j,t}}=1. \end{equation*} Simple algebra shows that the left-hand side is the growth rate of total credit obtained by firm i at time t. Then this yields \begin{equation*} \Delta \mathit{credit}_{i,t}=\beta _{1}\overline{ \mathit{domestic}_{i}}+\beta _{2}\overline{ \mathit{domestic}_{i}}\ast \mathit{CRISIS}_{t}+\hat{\alpha }_{i,t}+\nu _{i,t}, \end{equation*} which is the equation for the growth of credit at the firm level we are interested to estimate. To obtain the $$\hat{\alpha }_{i,t}$$, we estimate them from the relationship-level equation. These estimates are unbiased and consistent as the number of banks increases (provided that the number of firms does not go to infinity). As the $$\hat{\alpha }_{i,t}$$ are estimated in the relationship level equation, standard errors need to be estimated by bootstrapping to obtain correct estimates of the variance–covariance matrix. This equation is exactly valid for the growth rate of credit. We approximate it by the log change in credit. To estimate the full aggregate effect, we also take into account that part of the growth of credit is due to the starting of new credit relationships. Our approach is valid as long as the firm-specific effect is the same for old as for new relationships, possibly up to a noise term uncorrelated with both the other regressors and the firm effect. This is reasonably true for firm-specific characteristics such as firm riskiness. It must also be true for firm demand for credit, which must not be bank specific. This, however, is an identifying assumption that must hold throughout our analysis, also when we study credit supply at the bank–firm relationship level. Acknowledgments We would like to thank the editor and four anonymous referees for their helpful comments. We are especially grateful to Atif Mian for his insightful discussion of the paper at the 2013 NBER Summer Institute, and to Nicola Cetorelli, Linda Goldberg, and Steven Ongena for their detailed comments. We thank Giorgio Albareto, Martin Brown, Elena Carletti, Federico Cingano, Alessandro Conciarelli, Ricardo Correa, Matteo Crosignani, Olivier De Jonghe, Hans Degryse, Domenico Depalo, Giorgio Gobbi, Luigi Guiso, Giuseppe Ilardi, Silvia Magri, Francesco Manaresi, David Marques-Ibanez, Tommaso Oliviero, Daniel Paravisini, Alberto Pozzolo, Joao Santos, Koen Schoors, Neeltje Van Horen, participants at the 2012 CREDIT conference, the workshop “Macroeconomic policies, global liquidity, and sovereign risk”, the 6th CEPR Swiss Winter Conference in Financial Intermediation, the 3rd Mofir Workshop, the 20th Finance Forum, the FIRS 2013 Conference, the 2013 NBER Summer Institute, and seminar participants at the LSE, the Bank of Italy and the New York Fed. The views expressed in this paper do not necessarily reflect those of the Bank of Italy. Corresponding author: Enrico Sette Notes The editor in charge of this paper was Claudio Michelacci. Footnotes 1 This is not particularly restrictive because multiple lending is especially common in Italy (Detragiache, Garella and Guiso 2000, Gobbi and Sette 2014), see also Section 4 for a more detailed discussion of sample selection. 2 When we include bank fixed effects, these absorb the dummy FOREIGN, because no bank changes status (from domestic to foreign or vice versa) in our sample period. 3 Suppose firm 1 borrows from Italian bank A, and foreign bank B in June 2011. Our identification compares credit growth between June and December 2011 by bank A and B to the same firm 1. The pre-crisis period (December 2010–June 2011) allows to better control for possible different dynamics in credit supply by Italian and foreign banks, but having repeated observations for the same firm–bank pair is not strictly necessary for identification. 4 The CR lists all outstanding loan amounts above 30,000 euros that each borrower (both firms and households) has from banks operating in Italy, including branches and subsidiaries of foreign banks. Intermediaries are required by law to report this information. Data on outstanding loan amounts are available at monthly frequency and are of very high quality because intermediaries use the CR as a screening and monitoring device for borrowers. 5 We exclude firms with outstanding bad loans at the beginning of each period, because these are officially classified as losses. 6 We control for mergers and acquisitions among banks. If a firm had a relationship with a bank, and the bank disappears because it is acquired or merged, we track whether there is a new relationship with the newly formed bank, or with the acquirer, in which case we consider the relationship as still existing. 7 Imbens and Wooldridge (2009) show that a normalized difference below 0.25 indicates that the characteristics are balanced (i.e. not statistically different) across samples. 8 We select firms that borrow from at least one foreign and from at least one Italian bank in the pre-crisis period and firms that borrow from at least one foreign and from at least one Italian bank in the crisis period. Essentially we select a repeated cross-section of firms. 9 These data are unconsolidated and for foreign banks they only include the banks’ activity in Italy. 10 As to the relationship-level controls, the coefficient of SHARE is negative and significant, suggesting that banks might reduce lending more intensely towards firms they were initially more exposed to. This effect is weaker during the crisis. The coefficient of DRAWN/GRANTED is positive, but statistically significant only when bank-fixed effects are included. This is consistent with the possibility that a loan commitment is more likely to be increased if a firm is already using available commitments close to the limit. The effect is stronger during the crisis, only when bank*firm fixed effects are included (column 4). The coefficient of the share of OVERDRAFT is positive and significant (we do not have any prior about its effect on credit growth). The effect is stronger during the crisis. 11 All bank balance sheet data are from December 2010 for the pre-crisis period and from June 2011 for the crisis period. We use consolidated and unconsolidated (in case of stand-alone banks) data for Italian banks from the Supervisory reports submitted to the Bank of Italy. Consolidated data for foreign banks are from Bankscope. 12 To address the potential multicollinearity problems generated by the correlation of certain bank-level characteristics with the dummy FOREIGN, we run separate regressions inserting the bank-level variables one by one. The correlation matrix of regressors is shown in Table S4. 13 These data are from the ECB MFI interest rates statistics. 14 Data are from the Supervisory Reports. These regressions are weighted by total assets. To make these tests more comparable to those shown in Tables 5 and 6, we also re-run them weighted by the number of credit relationships, and all results hold. 15 These are dummies defined according to the median of the distribution of each variable, high indicating values above the median, low, below the median. 16 This approach is similar in spirit to that proposed by Abowd, Kramarz, and Margolis (1999) to estimate worker effects in their study of wage premia. 17 The data on interest rates include the flow of interest rates paid in from the firm to the bank and the ”products”, an accounting variable which is the loan amount outstanding times the number of days in which that amount has been outstanding. Dividing the flow of interest by the products we obtain a measure of the interest rate paid on the outstanding loans by the firm. For further details on these data, see Sette and Gobbi (2015). 18 These findings attenuate concerns about the baseline results being driven by a bank-specific demand for credit. In this case, we should find interest rates to be higher for foreign banks, because firm demand for credit to these banks increase. We find instead a combination of higher quantity and lower prices in credit relationship with foreign banks. 19 Every time a bank requests information on a borrower, the query is recorded in the CR, together with the motivation of the request, typically a loan application by a new client. For each application, we check if the bank granted any credit commitment to the loan applicant in the three months following the application. Hence, a loan application submitted to a bank, say, in December 2010, is classified as accepted if we observe that the bank grants credit to the borrower at any point in time between December 2010, the date of the request, and March 2011. 20 We cannot directly observe whether an application has been rejected. Hence, zeros include both rejected applications and applications accepted with a lag longer than three months. 21 A firm posting more than one application may do so because it expects lower chances that any given application is accepted. Posting an application may be costly because it reveals that the applicant already applied for a loan, as banks observe this information in the CR. References Abbassi P. , Iyer R. , Peydrò J. L. , Tous F. ( 2016 ). “ Securities Trading by Banks and Credit Supply: Micro-Evidence .” Journal of Financial Economics , 121 , 569 – 594 . Google Scholar CrossRef Search ADS Abowd J. , Kramarz F. , Margolis D. ( 1999 ). “ High Wage Workers and High Wage Firms .” Econometrica , 67 , 251 – 334 . Google Scholar CrossRef Search ADS Acharya V. , Drechsler I. , Schnabl P. ( 2014 ). “ A Pyrrhic Victory? Bank Bailouts and Sovereign Credit Risk .” Journal of Finance , 69 , 2689 – 2739 . Google Scholar CrossRef Search ADS Acharya V. , Steffen S. ( 2015 ). “ The Greatest Carry Trade Ever? Understanding Eurozone Bank Risks .” Journal of Financial Economics , 115 , 215 – 236 . Google Scholar CrossRef Search ADS Adelino M. , Ferreira M. ( 2016 ). “ Bank Ratings and Lending Supply: Evidence From Sovereign Downgrades .” Review of Financial Studies , 29 , 1709 – 1746 . Google Scholar CrossRef Search ADS Adrian T. , Colla P. , Shin H. S. ( 2013 ). “ Which Financial Frictions? Parsing the Evidence from the Financial Crisis of 2007 to 2009 .” NBER Macroeconomics Annual , 27 , 159 – 214 . Google Scholar CrossRef Search ADS Albertazzi U. , Bottero M. ( 2014 ). “ Foreign Bank Lending: Evidence From the Global Financial Crisis .” Journal of International Economics , 92 , S22 – S35 . Google Scholar CrossRef Search ADS Almeida H. , Cunha I. , Ferreira M. A. , Restrepo F. ( 2017 ). “ The Real Effects of Credit Ratings: The Sovereign Ceiling Channel .” Journal of Finance , 72 ( 1 ), 249 – 290 . Google Scholar CrossRef Search ADS Arteta C. , Hale G. ( 2008 ). “ Sovereign Debt Crises and Credit to the Private Sector .” Journal of International Economics , 74 , 53 – 69 . Google Scholar CrossRef Search ADS Bank of Italy ( 2014 ), Financial Stability Report, April. Battistini N. , Pagano M. , Simonelli S. , ( 2014 ). “ Systemic Risk, Sovereign Yields and Bank Exposures in the Euro crisis .” Economic Policy , 29 , 203 – 251 . Google Scholar CrossRef Search ADS Beck T. , Ioannidou V. , Schäfer L. ( 2017 ). “ Foreigners vs. Natives: Bank Lending Technologies and Loan Pricing .” Management Science , forthcoming . Bentolila M. , Jensen M. , Jiménez G. ( 2017 ). “ When Credit Dries Up: Job Losses in the Great Recession .” Journal of the European Economic Association , forthcoming . Bocola L. ( 2016 ). “ The Pass-Through of Sovereign Risk .” Journal of Political Economy , 124 , 879 – 926 . Google Scholar CrossRef Search ADS Patti E. Bonaccorsi di , Sette E. ( 2016 ). “ Did the Securitization Market Freeze Affect Bank Lending During the Financial Crisis? Evidence From a Credit Register .” Journal of Financial Intermediation , 25 , 54 – 76 . Google Scholar CrossRef Search ADS Borensztein E. , Panizza U. ( 2009 ). “ The Costs of Sovereign Default .” IMF Economic Review , 56 , 683 – 741 . Cetorelli N. , Goldberg L. S. ( 2011 ). “ Global Banks and International Shock Transmission: Evidence From the Crisis .” IMF Economic Review , 59 , 41 – 76 . Google Scholar CrossRef Search ADS Cetorelli N. , Goldberg L. S. ( 2012a ). “ Liquidity Management of U.S. Global Banks: Internal Capital Markets in the Great Recession .” Journal of International Economics , 88 ( 2 ), 299 – 311 . Google Scholar CrossRef Search ADS Cetorelli N. , Goldberg L. S. ( 2012b ). “ Follow the Money: Quantifying Domestic Effects of Foreign Bank Shocks in the Great Recession .” American Economic Review , 102 ( 3 ), 213 – 18 . Google Scholar CrossRef Search ADS Chodorow-Reich G. ( 2014 ). “ The Employment Effects of Credit Market Disruptions: Firm-level Evidence From the 2008–9 Financial Crisis .” Quarterly Journal of Economics , 129 , 1 – 59 . Google Scholar CrossRef Search ADS Cingano F. , Manaresi F. , Sette E. ( 2016 ). “ Does Credit Crunch Investment Down? New Evidence on the Real Effects of the Bank-Lending Channel .” Review of Financial Studies , 29 , 2737 – 2773 . Google Scholar CrossRef Search ADS Coeurdacier N. , Rey H. ( 2013 ). “ Home Bias in Open Economy Financial Macroeconomics .” Journal of Economic Literature , 51 , 63 – 115 . Google Scholar CrossRef Search ADS Correa R , Sapriza H. , Zlate A. ( 2012 ). ”Liquidity Shocks, Dollar Funding Costs, and the Bank Lending Channel During the European Sovereign Crisis .” Federal Reserve Discussion Paper 2012 – 1059 . De Haas R. , Van Horen N. ( 2012 ). “ International Shock Transmission After the Lehman Brothers Collapse Evidence From Syndicated Lending .” American Economic Review Papers & Proceedings , 102 ( 3 ), 231 – 237 . Google Scholar CrossRef Search ADS Dell’Ariccia G. , Goyal R. , Brooks P. Koeva , Pradhan M. , Tressel T. , Pazarbasioglu C. ( 2013 ). “ A Banking Union for the Euro Area .” IMF Staff Discussion Notes 13/01 . Detragiache E. , Garella P. , Guiso L. ( 2000 ). “ Multiple Versus Single Banking Relationships: Theory and Evidence .” Journal of Finance , 55 , 1133 – 1161 . Google Scholar CrossRef Search ADS ECB ( 2015 ). ECB Economic Bulletin, Issue 6 . Gennaioli N. , Martin A. , Rossi S. ( 2014 ). “ Sovereign Default, Domestic Banks, and Financial Institutions .” Journal of Finance , 69 , 819 – 866 . Google Scholar CrossRef Search ADS Gobbi G. , Sette E. ( 2014 ). “ Do Firms Benefit From Concentrating Their Borrowing? Evidence From the Great Recession .” Review of Finance , 18 , 527 – 560 . Google Scholar CrossRef Search ADS Imbens G. W. , Wooldridge J. M. ( 2009 ). “ Recent Developments in the Econometrics of Program Evaluation .” Journal of Economic Literature , 47 , 5 – 86 . Google Scholar CrossRef Search ADS International Monetary Fund ( 2010 ). “ Italy: 2010 Article IV Consultation .” IMF Country Report No. 10/157 . Iyer R. , Peydrò J. L. , da-Rocha-Lopes S. , Schoar A. ( 2014 ). “ Interbank Liquidity Crunch and the Firm Credit Crunch: Evidence From the 2007–2009 Crisis .” Review of Financial studies , 27 , 347 – 372 . Google Scholar CrossRef Search ADS Jiménez G. , Mian A. , Peydrò A. J. , Saurina J. ( 2010 ). “ Local Versus Aggregate Lending Channel: the Effects of Securitization on Corporate Credit Supply .” NBER Working Paper No. 16595 . Google Scholar CrossRef Search ADS Jiménez G. , Ongena S. , Peydrò J. , Saurina J. ( 2012 ). “ Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel With Loan Applications .” American Economic Review , 102 ( 5 ), 2301 – 2326 . Google Scholar CrossRef Search ADS Kahle Kathleen M. , Stulz René M. ( 2010 ). “ Financial Policies and the Financial Crisis: How Important was the Systemic Credit Contraction for Industrial Corporations? ” NBER Working Paper No. 16310 . Khwaja A. I. , Mian A. ( 2008 ). “ Tracing the Impact of Bank Liquidity Shocks: Evidence From an Emerging Market .” American Economic Review , 98 ( 4 ), 1413 – 1442 . Google Scholar CrossRef Search ADS Panetta F. , Angelini P. , Grande G. ( 2014 ). “ The Negative Feedback Loop between Banks and Sovereigns .” Bank of Italy , Occasional Paper No. 213 . Peek J. , Rosengren E. S. ( 1997 ). “ The International Transmission of Financial Shocks: The Case of Japan .” American Economic Review , 87 ( 4 ), 495 – 505 . Peek J. , Rosengren E. S. ( 2000 ). “ Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States .” American Economic Review , 90 ( 1 ), 30 – 45 . Google Scholar CrossRef Search ADS Popov A. , Udell G. F. ( 2012 ). “ Cross-Border Banking, Credit Access, and the Financial Crisis .” Journal of International Economics , 87 , 147 – 161 . Google Scholar CrossRef Search ADS Popov A , Van Horen N. ( 2015 ). “ Exporting Sovereign Stress: Evidence From Syndicated Bank Lending During the Euro Area Sovereign Debt Crisis .” Review of Finance , 19 , 1825 – 1866 . Google Scholar CrossRef Search ADS Reinhart C. M. , Rogoff K. S. ( 2009 ). “ This Time is Different: Eight Centuries of Financial Folly.” Princeton University Press . Schnabl P. ( 2012 ). “ The International Transmission of Bank Liquidity Shocks: Evidence from an Emerging Market .” Journal of Finance , 67 , 897 – 932 . Google Scholar CrossRef Search ADS Sette E. , Gobbi G. ( 2015 ). “ Relationship Lending During a Financial Crisis .” Journal of the European Economic Association , 13 , 453 – 481 . Google Scholar CrossRef Search ADS © The Authors 2017. Published by Oxford University Press on behalf of European Economic Association. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

### Journal

Journal of the European Economic AssociationOxford University Press

Published: Aug 17, 2017

## You’re reading a free preview. Subscribe to read the entire article.

### DeepDyve is your personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month ### Explore the DeepDyve Library ### Search Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly ### Organize Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place. ### Access Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals. ### Your journals are on DeepDyve Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more. All the latest content is available, no embargo periods. DeepDyve ### Freelancer DeepDyve ### Pro Price FREE$49/month
\$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off