Foreign banks and the doom loop

Albertazzi, Ugo; Cimadomo, Jacopo; Maffei-Faccioli, Nicolò Working Paper Foreign banks and the doom loop Working Paper, No. 2/2022 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Albertazzi, Ugo; Cimadomo, Jacopo; Maffei-Faccioli, Nicolò (2022) : Foreign banks and the doom loop, Working Paper, No. 2/2022, ISBN 978-82-8379-224-9, Norges Bank, Oslo, https://hdl.handle.net/11250/2997492 This Version is available at: https://hdl.handle.net/10419/264944 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/deed.no Foreign banks and the doom loop NORGES BANK RESEARCH 2 | 2022 UGO ALBERTAZZI, JACOPO CIMADOMO AND NICOLÒ MAFFEI-FACCIOLI WORKING PAPER NORGES BANK WORKING PAPER XX | 2014 RAPPORTNAVN 2 Working papers fra Norges Bank, fra 1992/1 til 2009/2 kan bestilles over e-post: [email protected] Fra 1999 og senere er publikasjonene tilgjengelige på www.norges-bank.no Working papers inneholder forskningsarbeider og utredninger som vanligvis ikke har fått sin endelige form. Hensikten er blant annet at forfatteren kan motta kommentarer fra kolleger og andre interesserte. Synspunkter og konklusjoner i arbeidene står for forfatternes regning. Working papers from Norges Bank, from 1992/1 to 2009/2 can be ordered by e-mail: [email protected] Working papers from 1999 onwards are available on www.norges-bank.no Norges Bank’s working papers present research projects and reports (not usually in their final form) and are intended inter alia to enable the author to benefit from the comments of colleagues and other interested parties. Views and conclusions expressed in working papers are the responsibility of the authors alone. ISSN 1502-8190 (online) ISBN 978-82-8379-224-9 (online) Foreign Banks and the Doom Loop∗ Ugo Albertazzi†, Jacopo Cimadomo‡, Nicol`o Maffei-Faccioli§ March 2022 Abstract This paper explores whether foreign banks stabilise or destabilise lending to the real economy in the presence of sovereign stress in the domestic economy and abroad. In this context, the presence of foreign intermediaries poses a fundamental, yet unexplored, trade-off. On the one hand, domestic sovereign shocks are broadly inconsequential for the lending capacity of foreign banks, given that their funding conditions are not hampered by such shocks. On the other, these intermediaries may react more harshly than domestic banks to a deterioration in local loan risk and demand conditions, or import shocks from their own sovereign. We exploit granular and confidential data on euro area banks operating in different countries to assess this trade-off. Overall, it is found that, under certain conditions, the presence of foreign lenders stabilises lending, thus mitigating the doom loop. JEL classification: E5, G21 Keywords: Sovereign stress, International banks, Lending activity. ∗We would like to thank Carlo Altavilla, Niccol`o Battistini, Agostino Consolo, Justas Dainauskas, Ettore Dorrucci, Christophe Kamps, Nadine Leiner-Killinger, Cosimo Pancaro, Beatrice Pierluigi, Antonio Riso and other participants at the presentations held at the ECB, the Fiscal Policy Workshop of the University of York and the Macro Club of Universitat Aut`onoma de Barcelona for helpful comments and suggestions. Nicol`o Maffei-Faccioli gratefully acknowledges the Fiscal Policies Division of the ECB for its hospitality. This paper should not be reported as representing the views of Norges Bank, the European Central Bank or the Eurosystem. The views expressed are those of the authors and do not necessarily reflect those of Norges Bank, the European Central Bank or the Eurosystem. †European Central Bank, Stress Test Modelling Division. Email: Ugo.Alb[email protected]. ‡European Central Bank, Fiscal Policies Division. Email: Jacop[email protected]. §Norges Bank, Research Unit. Email: [email protected]. 1 1 Introduction The “sovereign-bank nexus”, or “doom loop”, has been at the centre of the economic and policy debate in advanced economies, in particular since the 2010-2012 European sovereign debt crisis. When the sovereign debt market experiences periods of stress, the lending capacity of local banks tends to be impaired. The resulting credit crunch leads to a deterioration of the economy, which eventually ends up exacerbating the stress in the sovereign debt market even further.1Recently, the Covid-19 crisis has been accompanied by government interventions all around the globe via massive guarantee programmes for the banking sector, with possible risk of reactivation of the loop between the public sector and banks. This risk is also amplified by the fact that domestic banks have absorbed a large portion of the additional supply of government securities, reflecting the sizeable increases in the government debt-to-GDP ratio observed in advanced economies during the crisis.2 In this paper, focusing on the euro area, we explore whether an integrated banking system characterised by the presence of cross-border banks would help to absorb idiosyncratic shocks in sovereign debt markets, thus supporting the provision of credit to non-financial corporations. Indeed, the implications for the sovereign-bank nexus of a banking sector populated by foreign institutions are far from trivial. Several channels are at work and all need to be assessed. First, a “loan supply channel” is at play. Following an increase in domestic sovereign stress, this channel works through a deterioration of the funding capacity of domestic banks (but not of foreign banks). Indeed, given that the domestic sovereign is perceived as the ultimate explicit or implicit guarantor of local bank liabilities, when its creditworthiness worsens, so does the ability of domestic banks to raise funds. Moreover, a decline in the valuation of government bonds adversely affects bank capital and liquidity positions due to the banks’ exposure to the domestic sovereign and in relation to the wide utilisation of government securities as collateral in secured liquidity and funding transactions. With regard to foreign banks, sovereign stress in host countries is largely inconsequential for their lending capacity, given that their liabilities are guaranteed by other governments and their holdings of sovereign bonds are typically very 1See, among others, Lane (2012), Acharya et al. (2014), Battistini et al. (2014), Angelini et al. (2014), Fahri and Tirole (2019), Anderson et al. (2020). 2Between 2019 and 2021, the government debt-to-GDP ratio increased by about 15 and 20 percentage points of GDP in the euro area and the US, respectively. This rise was in large part absorbed by the domestic banking sector, especially in some countries (e.g., Italy). While interest rates on government securities have been low in this period, this situation may create risks of a reactivation of the doom loop in case of future interest rate hikes. 2 small.3Therefore, according to these mechanisms, the presence of foreign banks should be beneficial in terms of the mitigation of the sovereign-bank nexus. At the same time, foreign banks may export negative shocks from their own sovereign sector to the host country, via their consolidated balance sheets (see, e.g., Fillat et al.,2018). This could imply less credit to the host country (deleveraging) but also, in some cases, a lending increase, which would reflect a “flight-to-quality” mechanism. Thus, while it seems to be uncontroversial that a country populated only by domestic banks would be highly exposed to sovereign stress originating in the same country, the effects of also having foreign banks in that system is a priori unclear according to this channel. A second channel of transmission works through borrower-specific factors, instead of lender-specific factors. This channel has a “standard” demand component and a risk component, related to the perceived creditworthiness of borrowers. The first component suggests a drop in the demand for new loans on the part of the borrowers, given the deteriorated economic situation following a shock in their own sovereign market (Bocola,2016). Indeed, when the sovereign debt market worsens, so does the economy at large. This is due, among other factors, to the fact that confidence fades, investment and consumption plans are delayed, and that the government might be forced into contractionary policies in order to restore trust in the sustainability of public finances. The second (risk-related) component implies that foreign intermediaries may react more aggressively to a deterioration in the local economic situation (Albertazzi and Bottero,2014), including when it is due to an outbreak of domestic sovereign tensions. Indeed, funds from international investors tend to be prone to sudden stops, which is well known since the seminal contribution of Calvo (1998). All else being equal, the stronger this effect, the more detrimental the presence of foreign lenders for the doom loop. It is crucial to point out that this channel, which we refer to as “loan demand and risk channel” or simply “loan demand channel”, originates from a deterioration of the borrower’s risk profile. This clearly distinguishes it from the supply channel which operates via a worsening in banks’ intermediation capacity. While a growing body of the literature focuses on the supply channel of transmission, and a few papers have investigated the demand channel, no attempt has been made to simultaneously and consistently assess them and to study the role of foreign banks in this context, which is what we do in this paper. This is interesting for three main reasons. First, it allows us to derive some insights on the benefits (and costs) of having a banking system populated by foreign intermediaries. This has potentially 3Such holdings are on average less than 0.5% of their total assets for euro area banks (see, e.g., Altavilla et al.,2017). 3 very relevant policy implications especially for the EMU, which is still characterised by an incomplete banking union, but also for several emerging market economies, which typically feature a large presence of foreign intermediaries. Second, we can identify the relative importance of both channels for lending activity. Third, and more broadly, this exercise is an indirect assessment of the fundamental trade-off in finance between the benefits of relationship-lending (as postulated in the seminal contributions by Sharpe, 1990;Rajan,1992;von Thadden,2004), and of an armlength but more diversified set of lenders (as in Detragiache et al.,2000, and Ongena and Smith,2000). We perform such an assessment looking at a specific shock, such as that materialising in the sovereign debt markets. In the literature, the sovereign-bank nexus is typically analysed based on the correlation between sovereign yields and lending in a given country. However, this does not allow to isolate the role of foreign banks and to disentangle the supply and demand channels of transmission. These channels depend on borrower-specific and lenderspecific conditions, which affect domestic and cross-border banks differently. In this paper, we propose a novel identification approach, exploiting information on lending from cross-border banks in home and host countries. Our identification strategy crucially hinges on distinguishing between sovereign stress in the home and host country. This is captured by idiosyncratic movements in the 10-year government bond yield of the home country of the cross-border banking group and the 10-year government bond yield of the host country sovereign in which such intermediary operates through foreign subsidiaries. The key identification assumption is that the home sovereign yield of cross-border banks affects their foreign lending only through the loan supply channel, while the host sovereign yield affects the local lending of foreign cross-border banks only through the loan demand channel. Our empirical approach relies on the use of a unique and granular database including monthly bank-level information on lending activity in each euro area country, for a representative sample of intermediaries including all main cross-country banking groups operating in the euro area. The sample covers the period July 2007-April 2018. Our main findings suggest that, first, both the loan supply and loan demand channels are relevant drivers of lending activity. In particular, the loan supply channel explains about 20% of lending growth variation over time, while the demand channel accounts for approximately 17%. Second, we find that, on average, cross-border intermediaries deleverage foreign positions in case of stress in their home sovereign. However, when the exposure to the home sovereign is high, cross-border banks increase lending to the host countries, which could be interpreted as a result of flight-to-quality 4 or diversification strategies. Third, as regards the demand channel, our results indicate that foreign banks react more harshly to sovereign stress in the host countries. Overall, we show that - based on a stylised mean-variance model - a large share of foreign lenders in the host economy is preferable when the variance of foreign sovereign shocks is lower relative to that of domestic sovereign shocks. The remainder of the paper is structured as follows. Section 2 reviews the related literature, Section 3 presents the data and some descriptive statistics, Section 4 illustrates the empirical methodology, Section 5 shows our results and Section 6 assesses the overall benefits of a more diversified banking sector. Finally, Section 7 concludes. 2 Related literature Our paper contributes to a growing literature that studies the effects of sovereign stress on lending activity. Most studies analyse the loan supply channel and, especially, the role of sovereign exposures in the transmission mechanism. Several of these papers focus on the European sovereign debt crisis of 2010-2012. For example, Popov and Van Horen (2015) study the syndicated lending of European banks during that period and show that foreign sovereign stress affected negatively the lending of banks with sizeable holdings of government debt of stressed-countries relative to banks less exposed to that debt. Altavilla et al. (2017) illustrate that sovereign exposure considerably amplifies the transmission mechanism of sovereign stress to lending, especially in stressed countries. The banks in stressed-countries contracted both their domestic and foreign lending in response to an increase in domestic sovereign stress, and the higher their sovereign exposure, the more this is true. Bofondi et al. (2018), focusing on the Italian banking sector only, show that Italian banks tightened their credit supply and increased lending rates following the sovereign debt crisis more than foreign banks whose head institution resided in countries that were less exposed to sovereign stress. De Marco (2019) finds that losses on sovereign debt held by banks lead to a cut in lending supply to financially-constrained firms, both in stressed and non-stressed countries, and the higher the share of short-term funding of the banks, the more this is true. In a similar vein, Acharya et al. (2018) show that value impairment in banks’ exposures to sovereign debt triggered by the sovereign debt crisis and the risk-shifting behavior of weakly capitalized banks significantly reduced the probability of firms being granted new syndicated loans. Bottero et al. (2020) highlight that the shock to the banks’ sovereign portfolio caused by the 2010 Greek bailout was passed on to Italian firms through a credit contraction. This was particularly the case for banks 5 with a lower capital and less stable funding. Finally, Correa et al. (2016) find that US branches of euro area cross-border banks reduced their lending to U.S. firms due to the strong liquidity shock generated by depositors’ bank run during the European Sovereign debt crisis.4 Other papers focus on different sets of countries or sample periods. For example, Giannetti and Laeven (2012) highlight that foreign lenders, when hit by shocks that negatively affect bank wealth in their home market, have a tendency to rebalance their portfolio away from host markets to their domestic market. However, they do not focus explicitly on the transmission of sovereign shocks. Moreover, Adelino and Ferreira (2016) study the causal effect of bank credit rating downgrades on the supply of bank lending, for a panel of advanced and emerging economies. They find that banks with ratings at the sovereign bound reduce their lending significantly more than otherwise similar banks whose ratings are not at the sovereign bound following a sovereign downgrade. Finally, Gennaioli et al. (2018) provide evidence on the relationship between government bond holdings and the sovereign-bank nexus using a large bank-level dataset which comprises 20 sovereign default episodes in 17 countries between 1988 and 2012. The authors find systematic evidence of a negative relationship between sovereign holdings and lending during sovereign defaults, and show that pre-crises government bond holdings are strong predictors of a decline in bank lending. A smaller share of the literature has focused on the loan demand channel of sovereign stress transmission, but only in isolation to the loan supply channel. One of the first studies on the borrower specific determinants of sovereign stress transmission is Arteta and Hale (2008). This paper shows that sovereign crises in emerging markets were followed by a decline in foreign credit to domestic private firms in the non-financial sector. Moreover, Albertazzi and Bottero (2014) find that, in the post-Lehman collapse, foreign lenders contracted their credit supply to the same firm more than domestic banks, in response to the increase in credit risk and the deterioration of economic conditions that followed the collapse. More recently, Arellano et al. (2020) suggest that sovereign stress may be transmitted to firms not only via a credit supply crunch, but also endogenously through a contraction in the demand of labour input and intermediate good on the side of the borrowers, which therefore cut their demand for new loans. Altogether, it emerges that the lending (supply) channel of transmission has been 4˙ Zochowski, Franch and Nocciola (2021) explore the role of foreign banks but in another dimension, i.e., the cross-border propagation of prudential regulation in the euro area. They find that domestic banks reduce lending after the tightening of capital requirements in other countries. They also find that foreign affiliates increase lending following the tightening of sector-specific capital buffers in the countries where their parent banks reside. 6 Ei h(i),t−1. Specifically, we estimate the following equation: Lb(i),j,t =αj,t +β1Si,t−1+β2I(i=j) + β3Si,t−1∗I(i=j)+ +β4Ei h(i),t−1+β5Si,t−1∗Ei h(i),t−1+ +β6I(i=j)∗Ei h(i),t−1+ +β7Si,t−1∗I(i=j)∗Ei h(i),t−1+γXi,j,t−1+ξb(i),j,t , (3) The first two columns of Table 4 add only I(i=j) and its interaction with Si,t−1to the specifications of Columns (2) and (3) of Table 3. We observe a more negative coefficient associated to the home sovereign yield. However, ˆ β1has a different interpretation than ˆ βof Table 3: it is now the coefficient associated to an increase in home sovereign stress conditional on bank b(i) operating in a foreign country. For the sample including only cross-border banks (Column 2), we observe a positive and significant ˆ β3. This means that, on average, foreign banks cut lending more than domestic ones, for which we observe a smaller effect, as reflected by the sum of the coefficients ˆ β1and ˆ β3. Put differently, this suggests that when cross-border banks deleverage due to a worsening of funding conditions in their home country, they do so in particular in foreign countries. While international spillovers of sovereign stress have already been emphasised in some papers (e.g. Correa et al.,2016), the stronger deleveraging operated on foreign assets is novel in the literature. Columns (3) and (4) of Table 4 add home sovereign exposure at the head bank level Ei h(i),t−1and its interaction with Si,t−1to the baseline specification. In this case, ˆ β1 represents the effects of sovereign stress on lending conditional on the head bank having no home sovereign exposure. We do not find prima facie evidence of a strengthening of the loan supply channel if sovereign exposure is higher, as reflected in a ˆ β5coefficient not statistically different from zero. This refers to the overall sample of banks. However, the picture substantially changes when we distinguish between the domestic and foreign origin of banks for the full sample (Column 5) and for the sample of cross-border banks (Column 6). In practice, we add the dummy I(i=j), its interaction with Si,t−1and a three-way interaction term between Ei h(i),t−1,Si,t−1and I(i=j). In this case, the interpretation of the coefficients and the overall effect of sovereign stress on lending is less trivial. The coefficient ˆ β1reflects the effect of the sovereign stress conditional on being a foreign subsidiary and the head bank having no home sovereign exposures. In line with the findings reported in Columns (1) to (4), the coefficient is negative and significant across the different specifications. The estimated coefficients of the interaction between the domestic bank dummy and sovereign yields, i.e., ˆ β3, and of 13 Table 4: Supply channel effects conditional on sovereign exposure and domestic vs. foreign origin of the head bank (1) (2) (3) (4) (5) (6) Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Si,t−1-0.503∗∗∗ -0.599∗∗∗ -0.477∗∗∗ -0.355∗∗ -0.879∗∗∗ -1.045∗∗∗ (-4.04) (-4.26) (-4.05) (-2.42) (-3.20) (-3.76) I(i=j) -0.144 -1.972∗∗∗ -1.695∗∗ -3.185∗∗∗ (-0.47) (-3.76) (-2.50) (-3.88) Si,t−1∗I(i=j) 0.0703 0.551∗∗∗ 0.575∗∗ 1.042∗∗∗ (0.66) (2.93) (2.32) (3.83) Ei h(i),t−1-3.794∗-1.821 -25.01∗∗ -41.21∗∗∗ (-1.88) (-0.26) (-2.26) (-3.24) Si,t−1∗Ei h(i),t−10.0475 -0.255 10.31∗∗ 16.94∗∗∗ (0.18) (-0.18) (2.36) (3.69) I(i=j)∗Ei h(i),t−122.73∗∗ 53.91∗∗∗ (2.06) (3.91) Si,t−1∗I(i=j)∗Ei h(i),t−1-10.48∗∗ -20.51∗∗∗ (-2.39) (-4.50) Only cross-border banks no yes no yes no yes Controls yes yes yes yes yes yes R20.161 0.271 0.176 0.232 0.177 0.236 Adjusted R20.0829 0.0766 0.0790 0.0510 0.0801 0.0557 F 15.87 8.299 7.000 3.413 4.606 5.609 N 18605 5388 12888 4718 12888 4718 tstatistics in parentheses ∗p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Note: The dependent variable Lb(i),j,t is the annualised 3-month growth rate of loan volumes to nonfinancial corporations of bank b(i) that operates in country jin month t.Si,t−1is the three-months moving average of the 10-year sovereign yield of country iin month t−1. I(i=j) is a dummy that equals 1 if bank b(i) is domestic in country j(i.e. i=j) and 0 otherwise. Ei h(i),t−1is the domestic (country i) sovereign exposure of the head bank h(i) of bank b(i). Columns (1) to (7) include a lag of the dependent variable, the capital ratio and the liquidity ratio in month t−1 as controls. Standard errors are clustered at the host country and time level. 14 Figure 1: Effect of the loan supply channel for cross-border banks depending on their home sovereign exposure 0 0.05 0.1 0.15 0.2 0.25 0.3 -1 0 1 2 3 4 5 6 0 0.05 0.1 0.15 0.2 0.25 0.3 -1.5 -1 -0.5 0 Note: The figure shows the percentage point effect on lending in the host country following a 100 b.p. increase in the home sovereign yield, depending on the degree of exposure to the home sovereign (Ei h(i),t−1). The coefficient ˆγf(left chart) captures lending by foreign cross-border banks, while the coefficient ˆγd(right chart) lending by domestic cross-border banks. The red vertical line corresponds to the average home sovereign exposure in the sample. interaction between sovereign exposures and sovereign yields, i.e., ˆ β5, are positive and significant. The former coefficient shows that, in response to an increase in home sovereign stress, and conditional on the head banks having no sovereign exposure, domestic lenders cut their lending in the home country less than foreign ones. This is consistent with the findings of Column (2). The latter coefficient suggests that, when home sovereign stress increases, a higher home sovereign exposure of foreign subsidiaries reduces the contraction in lending volumes in the host country. If domestic sovereign exposures of the head bank are large (i.e., above about 6% of total assets), foreign banks actually increase their lending in the host country where they operate following negative developments in the home country’s sovereign yield. We can interpret this finding, which is new in the literature, as a “flight-to-quality” phenomenon. Finally, we observe a negative and significant coefficient for the three-way interaction term ˆ β7, especially for the sample with only cross-border banks (Column 6). This shows that domestic lenders contract lending more in the home country the higher their home sovereign exposure. We summarize these results in Figure 1, where we report the overall effect of an increase in sovereign stress through the loan supply channel for domestic and foreign cross-border lenders by varying their exposure to domestic (i.e., home) sovereign debt. This effect depends on the domestic sovereign exposure of the head bank. We construct 15 ˆγfand ˆγd, based on the estimated coefficients reported in Column (6) of Table 4: ˆγf=ˆ β1+ˆ β5Ei h(i) ˆγd=ˆ β1+ˆ β3+ (ˆ β5+ˆ β7)Ei h(i) and make domestic sovereign exposures Ei h(i)vary in the range [0,0.3], where 0 represents no exposure and 0.3 indicates a 30% share of domestic sovereign debt relative to main assets. We choose 30% as an upper bound as it corresponds to the largest sovereign exposure observed in our sample of cross-border banks. The coefficient ˆγf represents the percentage points effect on lending in the host country by foreign crossborder banks, following a 100 basis points increase in the home sovereign yield. The coefficient ˆγdrepresents the effect on the lending by domestic cross-border banks. The two charts display substantially different patterns. The left chart suggests that, for the average foreign cross-border bank in our sample (as represented by the red vertical line), there is a contraction of lending in the host country following an increase in sovereign stress in the home country. Lending in the host country still shrinks, though to a lesser degree, for home sovereign exposure between 3% and about 6%. However, if home sovereign exposure is higher than 6%, we observe a strong, positive, and significant expansion of lending in the host country. For example, for an exposure of around 15%, our findings suggest that the growth rate of lending in the host country rises by about 1.5 percentage points The picture is substantially different when we focus on domestic cross-border banks (right chart), for which the coefficient is negative, and significant for levels of domestic sovereign exposure higher than 10%. All in all, we document an important role of the loan supply channel of sovereign stress transmission for lending activity, which also depends on sovereign exposure and on the (domestic vs. foreign) lenders’ origin. 5.2 Loan demand channel Here, we report the results on the loan demand channel as a driver of lending activity. Table 5 presents the results for the baseline regression (2). The first column includes home country-time fixed effects and a lag of the dependent variable as the only control for the entire set of banks in the sample, i.e., both cross-border and non cross-border (Column 2 of Table 1). Columns (2) to (4) also include industrial production and the unemployment rate as additional controls to capture business cycle developments. The third and fourth column restrict the sample to only cross-border banks. The fourth column replaces the home country-time fixed effects with head bank-time fixed effects, 16 i.e., αh(i),t. Table 5: Demand channel effects on growth rates of loans (1) (2) (3) (4) Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Sj,t−1-0.259∗∗∗ -0.411∗∗∗ -0.371∗∗∗ -0.376∗∗∗ (-3.50) (-4.66) (-3.08) (-2.81) αi,t yes yes yes no αh(i),t no no no yes Only cross-border banks no no yes yes Controls no yes yes yes R20.146 0.147 0.176 0.263 Adjusted R20.0717 0.0720 0.0172 0.00842 F 38.31 21.03 2.602 3.333 N 23702 23702 5775 5137 tstatistics in parentheses ∗p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Note: The dependent variable Lb(i),j,t is the annualised 3-month growth rate of loan volume to nonfinancial corporations of bank b(i) that operates in country jin month t.Sj,t−1is the three months moving average of the 10-year sovereign yield of country jin month t−1. Column (1) includes a lag of the dependent variable as unique control. Columns (2), (3) and (4) include industrial production and unemployment in month t−1 as additional controls, Columns (1) to (3) have home-country fixed-effects, while Columns (4) has fixed-effects at the head-bank level. Standard errors are clustered at the host country and time level. We find a negative and statistically significant effect of an increase in sovereign stress in the host country on lending growth through this channel across the different specifications. In particular, the estimated coefficients are very similar when macroeconomic controls are included (Columns 2 to 4), regardless if we focus on cross-border banks only or use head bank-specific fixed effects. In terms of the size of this channel, we note that a 100 basis points increase in the 10-year sovereign yield of the host country leads to an average decrease in loan growth of around 0.4 percentage points. This accounts for approximately 17% of the standard deviation of loan growth, slightly less than for the loan supply channel (which was about 20%).6 Similarly to the loan supply channel, the loan demand channel of sovereign stress transmission could depend on the fact that lenders are domestic or foreign, and on sovereign exposures. However, in this context, the sovereign exposure that matters is the one to the host country where the bank operates, because the sovereign exposure to 6In Appendix B, we perform an IV regression, based on the Greek sovereign debt crisis which is used as an exogenous source of variation for yields in other fragile euro area countries, to assess the robustness of the loan demand channel of sovereign stress transmission. 17 the home country is already controlled for by head bank-time fixed effects. We expect this channel to be not very powerful as head banks and their subsidiaries tend to hold relatively few host country sovereign securities (see, e.g., Altavilla et al.,2017). Finally, the loan demand channel could depend on whether the host country was under stress in the sovereign debt market, in the analysed sample, or not. Specifically, we divide the sample and include a group of “stressed” countries, composed by Cyprus, Greece, Ireland, Italy, Portugal, Slovenia, and Spain. These are the countries which have been affected the most by the European sovereign debt crisis of 2011-2012 (see, e.g., Lane, 2012). To account for these different factors, we extend the baseline specification (2) as follows: Lb(i),j,t =αh(i),t +β1Sj,t−1+β2I(i=j) + β3Sj,t−1∗I(i=j)+ +β4Ej b(i),t−1+β5Si,t−1∗Ej b(i),t−1+ +β6I(i=j)∗Ej b(i),t−1+ +β7Sj,t−1∗I(i=j)∗Ej b(i),t−1+γXi,j,t−1+ξb(i),j,t, (4) where αh(i),t is the head-bank fixed effect, I(i=j) is a local lender dummy that takes value one if the home and host countries of bank b(i) are the same. We then include its interaction with the host 10-year sovereign yield Sj,t−1, the host sovereign exposure of the bank b(i) , i.e., Ej b(i),t−1, its interaction with Sj,t−1and I(i=j), and a three way interaction between Sj,t−1,I(i=j) and Ej b(i),t−1.7 The first column of Table 6 presents the estimation results in which we include only the local lender dummy, I(i=j), and its interaction with the host country sovereign yield, Sj,t−1. The second column focuses on cross-border banks that operate only in stressed countries. The interaction coefficient ˆ β3is positive and significant across specifications. This suggests that local lenders decrease credit less than foreign ones when sovereign stress in the host country increases. Quantitatively, these results suggests that following a 100 basis points increase in the host sovereign yield, lending via foreign banks contracts by 0.4 percentage points and by 0.1 percentage points via domestic banks (Column 1). Interestingly, if we focus on the sample of cross-border banks that operate in only stressed host countries, we observe a larger positive and significant ˆ β3and a more negative coefficient ˆ β1(Column 2). It seems therefore that foreign banks cut lending strongly in response to host country stress, even more so if the borrower country is stressed. This result confirms the view prevailing in international finance that foreign investors tend to be more flighty (see, e.g., Caballero and Simsek, 7Note that when i=j, equation (4) still differs from equation (3) because it includes head-bank fixed effects instead of country-specific fixed effects. 18 Table 6: Demand channel effects conditional on sovereign exposure and domestic vs. foreign origin of the head bank (1) (2) (3) (4) (5) (6) Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Sj,t−1-0.406∗∗∗ -2.349∗∗∗ -0.318∗∗ -2.312∗∗∗ -0.409∗∗∗ -2.138∗∗∗ (-3.09) (-4.15) (-2.50) (-4.06) (-2.60) (-3.68) I(i=j) -0.814∗∗ -3.162∗-1.209∗-2.410 (-2.04) (-1.81) (-1.93) (-0.97) Sj,t−1∗I(i=j) 0.306∗∗ 0.935∗∗ 0.416∗∗ 0.450 (2.29) (2.18) (2.19) (0.72) Ej b(i),t−116.47∗∗∗ -1.267 7.413 5.830 (4.75) (-0.18) (1.19) (0.26) Sj,t−1∗Ej b(i),t−1-3.275∗∗ -3.298∗∗ -0.900 -8.626 (-2.35) (-2.07) (-0.36) (-1.43) I(i=j)∗Ej b(i),t−116.71∗∗ -13.99 (2.37) (-0.59) Sj,t−1∗I(i=j)∗Ej b(i),t−1-4.166 7.759 (-1.50) (1.22) Stressed jno yes no yes no yes R20.177 0.263 0.179 0.286 0.181 0.290 Adjusted R20.0178 0.00806 0.0203 0.0558 0.0213 0.0582 F 2.467 2.265 6.958 7.568 5.317 5.003 N 5775 5137 5775 2514 5775 2514 tstatistics in parentheses ∗p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Note: The dependent variable Lb(i),j,t is the annualised 3-months growth rate of loan volumes to nonfinancial corporations of bank b(i) that operates in country jin month t.Sj,t−1is the three months moving average of the 10-year sovereign yield of country jin month t−1. I(i=j) takes value 1 if the lender is local and 0 if the lender is foreign. Ej b(i),t−1is the host (country j) sovereign exposure of bank b(i) in month t−1. All regressions include a lag of the dependent variable, industrial production and unemployment in month t−1 as additional controls. Standard errors are clustered at the host country and time level. 2020). Columns (3) and (4) of Table 6 present the results when we extend the baseline specification to account for the host country sovereign exposure of the particular bank b(i) and its interaction with the sovereign yield of the host country j. Column (3) includes all banks, while Column (4) only the stressed borrower countries. The coefficient ˆ β1represents the effects of sovereign stress on lending conditional on the bank holding no sovereign exposures to the host country. We observe a negative and significant coefficient for the different specifications. The magnitude of the estimated coefficient in the third column is similar to the one of Table 5. This is not surprising, however, as host sovereign exposure is expected to play a small role in the transmission. As before, 19 the magnitude of the effects of the sovereign stress through the loan demand channel turns out to be bigger when we focus on the stressed host countries (Column 4). The interaction coefficient ˆ β5is negative and significant: the higher the exposure in host sovereign bonds, the stronger the lending cut in response to the sovereign stress in that country. In the last two columns, we also include the local lender dummy, the interaction term with Sj,t−1and, additionally, a three-way interaction term between the host sovereign exposure of the bank, the sovereign yield and the dummy I(i=j). By introducing all the interaction terms, we observe, again, a negative and significant ˆ β1across the different specifications. In this context, ˆ β1is the effect of an increase in sovereign stress in the host country conditional on being a foreign lender with no sovereign exposure in the host country. The effect of the host sovereign yield is again stronger for stressed countries (Column 6). The local lender interaction term is positive and significant only when both stressed and non-stressed countries are included, while ˆ β5becomes insignificant for both specifications. In general, these results suggest that, when local lender dummies are introduced, the role of host sovereign exposure becomes negligible for the transmission of sovereign stress through the demand channel. 6 Is a diversified banking system overall beneficial? The overall assessment on the role of foreign banks in stabilising the domestic banking sector in the presence of sovereign stress needs two elements. One is the set of elasticities to the sovereign shocks, which is provided in the previous analysis. One general result is that foreign lenders tend to react more strongly to sovereign yield shocks, both in terms of lending supply when the shock originates in their home country, and in terms of reaction to local demand shocks in the host country. At the same time, funding conditions for foreign banks are not affected by shocks occurring in the host sovereign debt market, and thus the loan supply channel is inactive in the presence of such shocks. A second element is a model for the stochastic processes underlying the realisation of the sovereign shocks in each economy. A stylised mean-variance model, calibrated with the parameters from the empirical analysis, is proposed here to show under which conditions a higher share of foreign banks is preferable, depending on the relative variance of domestic vs. foreign sovereign shocks. Let dand fbe the sovereign stress shocks in the domestic and foreign country respectively, which can be interpreted as changes in the sovereign yield in the two 20 countries. We assume that the shocks are normally distributed d∼N(0, σ2 d) and f∼N(0, σ2 f). The shocks are assumed to be uncorrelated, reflecting the conceptual framework of our empirical analysis, which controls for the correlation between sovereign yields via fixed effects. Let ∆Lbe the lending change in the domestic country in response to sovereign stress that originated both in the domestic economy and abroad. It will depend on the share of foreign lenders operating in an economy, which is defined as φ, and on the elasticities to the sovereign shocks estimated in the empirical analysis. In particular, changes in lending will evolve according to: ∆L=d((1 −φ)α+φβ) + f(φγ) (5) where αis the sum of supply and demand channel elasticity to a domestic shock by domestic banks (from Column (1) in Table 4 and Column (1) in Table 6, respectively), βis the demand channel elasticity to a domestic shock by foreign banks (from Column (1) in Table 6). Note that the supply channel is inactive for foreign banks, when the shock hits the host country. Finally, γis the supply channel elasticity to a foreign shock by foreign banks (from Column (1) in Table 4). This is an important element in the analysis, because foreign banks may export (to the host country) shocks which are originated in their home (foreign) country. The expected value of ∆Lis then given by:8 E(∆L) = E(d)((1 −φ)α+φβ) + E(f)(φγ) (6) Notice that E(∆L) = 0 as we assume normality of the shocks. The variance of ∆Lis given by: var(∆L) = σ2 d((1 −φ)α+φβ)2+σ2 f(φγ)2(7) Similarly to other portfolio problems in finance (see, e.g., Markowitz and Todd, 2000), we can frame the problem in terms of a “social planner” who wants to maximise expected lending and minimise the variance of lending and, to that aim, chooses the “optimal” share of foreign banks φ. Specifically, we consider the following problem: max φE(∆L)−a∗var(∆L) subject to 0 ≤φ≤1 (8) for a given value a, which weights the relative importance of lending variance.9 8Note that estimation uncertainty from the regression results in Table 4 and 6 is not taken into account here. 9The value of ais irrelevant for the exercise based on normally distributed shocks, because E(∆L) 21 We solve the problem above for the share of foreign banks in the domestic economy, φ, conditional to all combinations of σ2 dand σ2 fin the interval [0,1] and report in Figure 2 the optimal values φ∗. The yellow region in Figure 2 indicates a combination of variances for which an economy populated only by foreign banks is optimal. The figure suggests that, when the variance of the foreign sovereign shocks is relatively low, a high share of foreign banks is preferable. In the extreme case in which the variance of such shocks is zero, it will always be optimal to have an economy populated by only foreign banks because they will be completely unaffected by shocks that originated in their own country. Put differently, a large share of foreign lenders in the host economy is always preferable conditional on domestic sovereign shocks only. In general, as long as the variance of foreign shocks increases relative to that of domestic shocks, it will be progressively less convenient to have a large share of foreign banks in the local economy. Figure 2: Optimal φ∗with normally distributed sovereign shocks Note: The figure shows the optimal share of foreign banks in the local economy, φ∗, for given levels of the variance of domestic (σ2 d) and foreign sovereign shocks (σ2 f), where such shocks are assumed to be normally distributed and centered around zero. The yellow region indicates a combination of variances consistent with an environment populated by only foreign banks, while in the dark blue region it would be optimal to have only domestic banks. 7 Conclusions This paper offers new insights into the transmission of sovereign stress to lending to non-financial corporations, focusing on the role of cross-border banks. In contrast to in equation (8) is zero. In Appendix C, we show the robustness of the exercise to shocks distributed as binomial. In that case, the value of awill affect the optimisation problem. We choose in an ad-hoc way ato be equal to 0.5, the results are broadly robust to other values of that parameter. 22 the supply one when focusing on loan rates. Interestingly, the demand channel effect on loan rates can be interpreted as driven by two components. On the one hand, according to the pure “demand” component of this channel, we would expect both loan volumes and loan rates to decrease when credit demand decreases due to an increase in sovereign stress. On the other, according to the “credit risk” component of this channel, we would expect loan volumes and loan rates to move in opposite directions, as this channel is driven by the supply of credit of banks (even though not through a deterioration of funding conditions). The positive and significant effect of an increase in sovereign stress of loan rates suggests that the risk component is relatively more important than the demand component. This could not be seen by simply looking at loan volumes, for which both channels imply a clear decrease. Table A2: Demand channel effects of sovereign stress on loan rates to NFC (1) (2) (3) (4) Rb(i),j,t Rb(i),j,t Rb(i),j,t Rb(i),j,t Sj,t−10.495∗∗∗ 0.535∗∗∗ 0.393∗∗∗ 0.536∗∗∗ (6.31) (5.01) (3.27) (3.84) αi,t yes yes yes no αh(i),t no no no yes Cross-border banks only no no yes yes Controls no yes yes yes R20.941 0.941 0.938 0.943 Adjusted R20.935 0.935 0.926 0.921 F 6694.6 3464.2 1528.3 1634.7 N 21743 21743 5058 4526 tstatistics in parentheses ∗p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Note: The dependent variable Rb(i),j,t is the annualised 3-month loan rate to non-financial corporations of bank b(i) that operates in country jin month t.Sj,t−1is the three months moving average of the 10-year sovereign yield of country jin month t−1. Column (1) includes a lag of the dependent variable as unique control, whereas Columns (2) to (4) include, additionally, industrial production and unemployment in month t−1 as additional controls. Standard errors are clustered at the host country and time level. B IV regressions Here, we present a robustness exercise for our loan demand channel equation (2), based on a simple IV regression. We focus on that channel only as our definition of the supply side is less affected by a possible reverse causality problem. In fact, in that framework, the shock originates in a different country than the one in which the bank operates and we believe this is largely rules out possible endogeneity problems. As instrumental variable, we use the Greek sovereign yield interacted for a dummy with 29 value 1 for the period from April 2010 to December 2011, which corresponds with the peak of the Greek sovereign crisis, and 0 otherwise. The idea is to isolate more effectively exogenous movements in the sovereign yield of other European countries, which is arguably the case when we focus on the contagion from Greece to those countries in that period. We focus on stressed countries only (Cyprus, Greece, Ireland, Italy, Portugal, Slovenia and Spain) because the first stage coefficients would not be interpretable if using the whole set of 19 euro area countries. Indeed, due to flight to quality, one may expect that stress in the Greek sovereign market would reduce - rather than increase - the yields in the core countries. We perform four exercises, which are presented in Table B1. Table B1: Demand channel effects in an IV regression (1) (2) (3) (4) Lb(i),j,t Lb(i),j,t Lb(i),j,t Lb(i),j,t Sj,t−1-0.185∗∗ -0.184∗∗∗ -0.232∗∗∗ -0.229∗∗∗ (-2.05) (-2.02) (-4.78) (-5.08) Cross-border banks only yes yes no no αh(i)no yes no yes N 2766 2766 9812 9812 tstatistics in parentheses ∗p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01 Note: The dependent variable Lb(i),j,t is the annualised 3-months growth rate of loan volumes to nonfinancial corporations of bank b(i) that operates in country jin month t.Si,t−1is the three months moving average of the 10-year sovereign yield of country iin month t−1, which is instrumented with the Greek yield for the same month. Regressions include a lag of the dependent variable as unique control. Standard errors are clustered at the host country and time level. The first column presents a simple IV regression of the growth rate of loan volumes on the host country sovereign yield where the sample is restricted to stressed host countries and cross-border banks only. In line with our baseline results, the coefficient is negative and statically significant at the 1% confidence level. In the second column, we add head-bank fixed effects in order to better control for the supply side.10 Again, the coefficient turns out to be negative and highly significant. In columns three and four we perform the same exercise of the first two columns, but for the entire banking sample in stressed host countries (including cross border and non cross-border banks). Across different specifications, the coefficients of a sovereign stress increase are negative and always highly statistically significant. 10We cannot include head bank-time fixed effects, because in this case we could not use the Greek sovereign yield as an instrument, as it is varies over time. 30 C Diversification exercise with binomially distributed shocks We report here a robustness exercise to the one of Section 6 where, instead of assuming that the sovereign shocks dand fare normally distributed and centered around zero, we assume that they follow binomial distribution, with values 0 and 1. The optimisation problem is identical to the one in (8), but now the term E(∆L) is not zero. Similarly to Figure 2, the yellow region in Figure C1 indicates a combination of variances for which an economy populated by only foreign banks is optimal. Consistently with the findings of Section 6, Figure C1 suggests that, when the variance of the foreign sovereign shocks is low (in this case, below 0.2), a high share of foreign banks is preferable. Figure C1: Optimal φ∗with binomial shocks Note: The figure the optimal share of foreign banks in the local economy, φ∗, for given levels of the variance of domestic (σ2 d) and foreign sovereign shocks (σ2 f), where such shocks are assumed to be follow a binomial distribution with values 0 and 1. 31