What Determines the Composition of International Bank Flows?

K.7 What Determines the Composition of International Bank Flows? Niepmann, Friederike, and Cornelia Kerl Please cite paper as: Niepmann, Friederike, and Cornelia Kerl (2016). What Determines the Composition of International Bank Flows? International Finance Discussion Papers 1170. http://dx.doi.org/10.17016/IFDP.2016.1170 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1170 June 2016

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1170 June 2016 What Determines the Composition of International Bank Flows? Cornelia Kerl Friederike Niepmann NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

What Determines the Composition of International Bank Flows?* Cornelia Kerl** and Friederike Niepmann*** Abstract This paper studies how frictions to foreign bank operations affect the sectoral composition of banks’ foreign positions, their funding sources and international bank flows. It presents a parsimonious model of banking across borders, which is matched to bank-level data and used to quantify cross-border frictions. The counterfactual analysis shows how higherbarrierstoforeignbankentryalterthecompositionofinternationalbankflowsand may reverse the direction of net interbank flows. It also highlights that interbank lending and lending to non-banking firms respond differently to changes in foreign and domestic conditions. Ultimately, the analysis suggests that policies that change cross-border banking frictions and, thereby, the composition of banks’ foreign activities affect how shocks are transmitted across borders. Keywords: global banks, interbank market, international bank flows, cross-border banking JEL-Codes: F21, F23, F30, G21 *We thank Pierre-Olivier Gourinchas, Galina Hale, Katheryn Russ, Tim Schmidt-Eisenlohr, Vania Stavrakeva, Frederic Boissay and two anonymous referees for their helpful comments. This paper was prepared for the DNB/IMF conference “International Banking: Microfoundations and Macroeonomic Implications”. **Deutsche Bundesbank, Wilhelm-Epstein Str. 14, 60431 Frankfurt am Main, Germany. E-Mail: Cornelia.Kerl@bundesbank.de. Theviewsinthispaperaresolelytheresponsibilityoftheauthor(s)andshould not be interpreted as reflecting the views of the the Deutsche Bundesbank, their staff, or of any other person associated with the Deutsche Bundesbank. ***The author is a staff economist in the Division of International Finance, Board of Governors of the Federal Reserve System, Washington, D.C. 20551 U.S.A. E-Mail: Friederike.Niepmann@frb.gov. The views in this paper are solely the responsibility of the author(s) and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of any other person associated with the Federal Reserve System.

1 Introduction Global banks are active around the world, lending to banks and firms abroad. While loans to foreign banks and loans to foreign firms were equally large 20 years ago, the largest share of foreign lending today goes to foreign non-banking firms.1 Figure 1 illustrates this by showing the composition of banks’ foreign activities according to data from the Bank for International Settlements (BIS). The top chart depicts bank claims of around 25 BIS reporting countries on foreign parties, which are split into claims on foreign banks, foreign firms and foreign governments. As the bottom chart highlights, claims on foreign firms as a share of total international claims have been increasing since 1999 and account for roughly 55 percent of international claims today. This paper provides a parsimonious model of banking across borders that can address these developments, showing how barriers to foreign bank operations affect the sectoral composition of bank activities. Based on detailed bank-level data from the Deutsche Bundesbank to support themodel, thepaperhighlightsthatdifferenttypesofbankactivitiesareinterconnectedandare often substitutes, for example, lending by banks on the interbank market and lending to foreign firms. When the impediments to foreign bank operations decline, bank lending to foreign firms rises, while lending to foreign banks falls.2 Theseresultsarerelevantforthecurrentdebatearoundglobalbanking. Sincethe2007/2008 financial crisis, when international bank flows collapsed, the foreign operations of banks have been under great scrutiny from policy makers, which have considered regulating foreign bank activities. To understand the implications of possible restrictions on global banking, it is key to have an integrated view of banks’ foreign operations. The model in this paper allows for this by explicitly incorporating an international interbank market, cross-border lending by banks to foreignfirmsaswellasentryofforeignbanksintolocalbankingmarketsintoasingletheoretical framework. This is in contrast to existing frameworks in the international banking literature, which focus on only one or two of these activities.3 The paper starts by presenting a closed economy model, in which the interbank market is a means to reallocate funds from the less efficient to the more efficient banks.4 Banks have to monitor firms when they extend a loan, which is costly. Because banks have equal amounts of deposits but face different monitoring costs, it is optimal that funds are redistributed from the high-cost to the low-cost banks. In the open economy model, banks can lend and borrow on the international interbank market and they can engage in cross-border lending to firms or establish foreign affiliates abroad for a fixed cost. A foreign affiliate allows the bank to 1Firms refer to non-banking firms as opposed to banks in this paper. 2The model provides a potential explanation for the patterns in figure 1 but the chart should be seen as mainlymotivational. Thepaperdoesnottestinhowfarthemodelgeneratestheobservedpatternsinthechart but instead uses German bank-level data to support the proposed theory. 3InBrunoandShin(2015),intra-bankandinterbanklendingareisomorphic. InNiepmann(2015),interbank funding, cross-border deposit taking and borrowing from foreign affiliates are isomorphic. Niepmann (2013) abstractsfrominterbanklending. IndeBlasandRuss(2013),cross-borderlendingandlendingthroughforeign affiliates are considered as separate scenarios. Corbae and D’Erasmo (2010) study banking industry dynamics without allowing for interbank lending. Building on the aforementioned work, Corbae and D’Erasmo (2014) introduce interbank lending into a closed economy without considering foreign bank operations. 4This in line with in the modeling approach in Boissay (2011), for example. 1

decrease variable transaction costs from operating abroad and to raise additional funds from foreign depositors. In equilibrium, banks lend to and borrow from each other as well as to and from foreign and domestic firms/depositors so that monitoring costs and efficiency losses due to cross-border frictions are minimized and the return on loans is maximized. We apply the model to data from the Deutsche Bundesbank with detailed information on German banks’ foreign operations and balance sheets. By taking a structural approach and matching the model to the data, we can study the quantitative responses of international bank activities to varying underlying structural parameters. In a first step, we exploit key structural equations of the model to learn about the barriers that banks face abroad. We derive simple equations that relate banks’ domestic and cross-border loans to their efficiency and bank entry barriers. With the German data at hand, these equations can be used to quantify the frictions. In particular, we obtain for each country in which German banks operate the variable costs of operating there as well as the fixed costs of lending cross-border to non-banking firms and of establishing affiliates in that market. The calculated cost parameters for various host countries are strongly correlated with proxies of countries’ openness. The model prediction that larger bankstendtobeborrowerswhilesmallerbankstendtobelendersontheinternationalinterbank market also holds in the data. Based on the parameters obtained from the previous exercise, we simulate a two-country version of the model to analyze and quantify how bank entry barriers affect the composition of foreign bank activities. The counterfactual analysis is for German bank operations in the US. The model matches the Bundesbank data well, in particular the observed loans of German banks to US firms as well as net interbank flows between the two countries. Different scenarios are compared to the calibrated baseline economy. We consider the case in which banks can lend and borrow on the international interbank market but not from foreign firms and depositors (scenario i). We also study a 10 percent reduction in the fixed cost of establishing a foreign affiliate in the US compared to the baseline economy (scenario ii). The counterfactual analysis shows how impediments to foreign bank operations affect the composition of banks’ foreign assets, their funding sources as well as the composition of crossborderbankflowsbetweencountries. Inthebaselineeconomy,consistentwiththedata,German banks’ operate both cross-border from home as well through affiliates in the US. To fund these activities, they use domestic deposits, foreign deposits and funds borrowed on the international interbank market. When cross-border frictions rise and eventually become prohibitively high in scenario (i), capital only flows between the two countries through the international interbank market. As a result, German banks extend loans only to US banks but not to US firms. These loans are exclusively funded with domestic deposits. Total loans to US firms decline as German banks redirect some of their lending back to domestic firms. Flows on the international interbankmarketreverseandGermanyturnsfrombeinganetborrowerfromUSbankstobeing a net lender. Effects are different when the cost of establishing a foreign affiliate in the US falls (scenario ii) compared to the baseline economy. German banks expand their lending to US non-banking firms, which they fund both through the interbank market and through raising more US deposits. When parent banks provide a large fraction of funding to their affiliates, both the 2

volume of net intrabank flows between German parent banks and their US affiliates as well as net interbank flows between Germany and the US rise (German banks continue to be net borrowers from US banks). Together the counterfactual analysis highlights how the barriers to foreign bank operations affect the quantity of domestic credit, who supplies credit (domestic versus foreign banks) and the sources of funding for domestic credit. Why does this matter? A growing number of empirical papers documents that banks do not uniformly adjust their activities in response to balance sheet shocks. Lending by foreign-owned banks in a country appears to be less stable than lending by domestically-owned banks.5 At the same time, there is evidence that funding provided on the interbank market is more volatile than funding provided by parent banks.6 Policies that alter the barriers to foreign bank operations may therefore affect the availability and volatility of domestic credit, with ultimate consequences for real activity.7 For example, higher barriers to foreign bank entry may make a country rely less on credit extended by foreign banks but, at the same time, domestic banks may increase their borrowing on international interbankmarkets. Itisnotclearwhetherdomesticcreditbecomeslessormoreproneto foreign shocks and the effects for domestic financial stability are ambiguous. This paper does not address such dynamic issues directly. However, by detailing theoretically the link between cross-border banking frictions and the composition of international bank flows, it provides a key building block that should be integrated into richer international macro models to study the implications of banking sector integration for financial stability.8 The quantitative results obtained from the structural exercise can also serve as inputs for future work in this direction. 2 A Model of International Banking The model presented in this papers is inspired by the theoretical framework in Niepmann (2013). In contrast to the aforementioned paper, it incorporates an international interbank marketasthebasisforstudyingtheimplicationsofimpedimentstoforeignentryonthesectoral composition of banks’ foreign activities. 5See de Haas and van Lelyveld (2010), Ongena et al. (2013), and de Haas and van Lelyveld (2014). Foreign bank ownership can also provide support in a domestic crisis as documented in Jeon et al. (2013), Popov and Udell (2012), and de Haas and van Lelyveld (2006). 6SeeSchnabl(2012),ReinhardtandRiddiough(2014),andMcCauleyetal.(2012). Additionalworksuggests that local lending by affiliates is more stable than cross-border lending by the parent banks. See Milesi-Ferretti and Tille (2011), Cetorelli and Goldberg (2011), de Haas and van Horen (2013), Kamil and Rai (2010), Duewel et al. (2011). 7Cutsinthesupplyofcreditbybankshavebeenshowntohaveadverseeffectsonproductionandemployment. See, e.g., Khwaja and Mian (2008), Rosengren and Peek (2000), and Chodorow-Reich (2014). 8For papers that introduce global banks in international macro, see, e.g., Kollmann (2013), Olivero (2010), Kollmann et al. (2011), and Greenwood et al. (2013). 3

2.1 Closed economy In the closed economy, there are a mass 𝑀 of bankers and 𝐾 units of depositor capital. Each banker has 𝑑 = 𝐾/𝑀 units of deposits and can lend the collected funds to the aggregate production sector, which yields an exogenous return 𝑅 > 1 per unit invested.9 Bankers incur costs from monitoring the firms they lend to. This cost differs across bankers, who draw an efficiency parameter 𝑎 from a distribution 𝑔(𝑎) with support [𝑎,𝑎] and mean 𝑎′. A higher value implies that the banker faces lower monitoring costs and is more efficient.10 Bankers face decreasing returns to scale, that is, total monitoring costs increase in the bank’s credit volume.11 The monitoring costs 𝑐 of a banker with efficiency 𝑎 are given by: 1 𝑐(𝑎) = ℎ(𝑧), (1) 𝑎 where ℎ(𝑧) is a continuous and twice differentiable function with ℎ′(𝑧) > 0 and ℎ′′(𝑧) > 0 and 𝑧 is the total capital lent to firms by the banker with efficiency 𝑎. Bankers can lend and borrow without costs from each other on the interbank market at the endogenous rate 𝑅𝐼. The profits of the banker with efficiency 𝑎 are therefore: 1 𝜋(𝑎) = 𝑅𝑧 − ℎ(𝑧)−𝑅𝐼(𝑧 −𝑑). (2) 𝑎 Each banker chooses 𝑧 to maximize profits. The first-order condition implies: ℎ′(𝑧) = 𝑎(𝑅−𝑅𝐼). (3) Given the assumed properties of ℎ(𝑧), there exists a unique solution to each banker’s lending volume 𝑧, which increases in the return to capital 𝑅 and the banker’s efficiency 𝑎 and decreases intheinterbanklendingrate𝑅𝐼. Intheremainderofthispaper,wewillassumethatℎ(𝑧) = 1𝑧2, 2 which delivers 𝑧 = 𝑎(𝑅−𝑅𝐼). Capital market clearing requires that banks invest the total depositor capital 𝐾 in the production sector: ∫︁ 𝑎 𝑀 𝑧(𝑎)𝑔(𝑎)𝑑𝑎 = 𝐾. (4) 𝑎 9The supply of domestic deposits is assumed to be fixed and the same for each bank. This assumption simplifies the model greatly because it eliminates any additional source of heterogeneity on the liability side of banks’balancesheets. Interbankborrowingandlendingissimplythegapbetweenbanks’optimalloanvolumes (which differ across banks) and deposits (which are the same for each bank) and is not an endogenous choice per se. To model the liability side of banks’ balance sheets more explicitly, one could assume, for example, that bankerscompetefordeposits,facingconvexcostsofraisingdeposits. Thedepositratewouldthenbeafunction of the interbank lending rate and the cost of raising deposits. The size of domestic deposits on banks’ balance sheets would vary across banks. As we show later based on German bank-level data, the model predictions regarding banks’ net interbank lending and borrowing hold in the data. Thus, the model captures key features of banks’ funding composition even without more sophisticated modeling of the deposit side. 10HeterogeneityinthecostoffinancialintermediationisalsomodeledindeBlasandRuss(2010)anddeBlas and Russ (2013). 11This could be rationalized as follows: As the size of a banker’s loan portfolio increases, the quality of the borrowers goes down, reflected in higher per unit monitoring costs. Alternatively, organizational complexity may increase with bank size and lead to higher operating costs. 4

Plugging in z, we obtain: ∫︁ 𝑎 𝐾 𝑎(𝑅−𝑅𝐼)𝑔(𝑎)𝑑𝑎 = . (5) 𝑀 𝑎 Solving for 𝑅𝐼 yields: 𝐾 1 𝐾 1 𝑅𝐼 = 𝑅− = 𝑅− , (6) 𝑀 ∫︀𝑎 𝑎𝑔(𝑎)𝑑𝑎 𝑀 𝑎′ 𝑎 where 𝑎′ = 𝐸(𝑎) = ∫︀𝑎 𝑎𝑔(𝑎)𝑑𝑎 reflects the average efficiency of bankers in the economy. Ex- 𝑎 pression (6) shows that the interbank lending rate in the economy is a function of the return on loans and the efficiency of the economy’s banking sector. The larger the return 𝑅 on loans and the lower the bankers’ average monitoring costs are, the higher is the equilibrium interbank lending rate 𝑅𝐼.12 This rate also depends on the number of bankers relative to deposits in the economy. The more bankers there are relative to deposits, the lower are the deposits that a single banker has, the tougher is the competition for funds and the higher is the interbank rate. 2.2 Open economy with international interbank lending In the open economy, there are 𝑁 countries. Countries differ in the return 𝑅 on loans, their size 𝐾, the mass of bankers 𝑀 they host, and in their banking sector efficiencies. Asthefirstscenario, weconsiderthecaseinwhichthereisaninternationalinterbankmarket thatallowsbankstolendandborrowacrossborders. Bankscannotlendtoforeignfirmsdirectly so that interbank lending is the only channel through which capital can be reallocated from one country to the other. An equilibrium requires that the international interbank market clears, that is, the capital lent to firms in all countries must equal the world capital endowment: ∑︁ 𝑁 ∑︁ 𝑁 ∫︁ 𝑎𝑖 𝐾 = 𝑀 𝑎 (𝑅 −𝑅𝐼)𝑔 (𝑎 )𝑑𝑎 . (7) 𝑖 𝑖 𝑖 𝑖 𝑖 𝑖 𝑖 𝑎 𝑖=1 𝑖=1 𝑖 Solving for 𝑅𝐼 delivers the following expression: ∑︀𝑁 𝑀 𝑅 𝑎′ − ∑︀𝑁 𝐾 𝑅𝐼 = 𝑖=1 𝑖 𝑖 𝑖 𝑖=1 𝑖 , (8) ∑︀𝑁 𝑀 𝑎′ 𝑖=1 𝑖 𝑖 where 𝑎′ = 𝐸(𝑎 ) = ∫︀𝑎𝑖𝑎 𝑔 (𝑎 )𝑑𝑎 . The interbank lending rate in the open economy thus 𝑖 𝑖 𝑎 𝑖 𝑖 𝑖 𝑖 𝑖 depends on relative country sizes and returns on loans as well as the average efficiency of banking sectors and the mass of bankers in each country. The same factors also determine the allocation of capital across countries. The capital flow into country 𝑖, denoted by 𝐾𝑋, is given 𝑖 12It is assumed that parameters are such that investment and financial intermediation are beneficial in the economy so that all funds are in fact invested in projects. This requires that monitoring costs are not too high so that 𝑅−1/𝑎′ >1. 5

by: ∑︀𝑁 (𝑅 −𝑅 )𝑀 𝑀 𝑎′𝑎′ +𝑀 𝑎′ ∑︀𝑁 𝐾 −𝐾 ∑︀𝑁 𝑀 𝑎′ 𝐾𝑋 = 𝑗̸=𝑖 𝑖 𝑗 𝑖 𝑗 𝑖 𝑗 𝑖 𝑖 𝑗̸=𝑖 𝑗 𝑖 𝑗̸=𝑖 𝑗 𝑗 . (9) 𝑖 ∑︀𝑁 𝑀 𝑎′ 𝑗=1 𝑗 𝑗 Assuming that 𝑁 = 2 and 𝑀 = 𝐾 , this expression reduces to: 𝑖 𝑖 ∑︀𝑁 (𝑅 −𝑅 )𝐾 𝐾 𝑎′𝑎′ + ∑︀𝑁 (𝑎′ −𝑎′)𝐾 𝐾 𝐾𝑋 = 𝑗̸=𝑖 𝑖 𝑗 𝑖 𝑗 𝑖 𝑗 𝑗̸=𝑖 𝑖 𝑗 𝑖 𝑗 . (10) 𝑖 ∑︀𝑁 𝐾 𝑎′ 𝑗=1 𝑗 𝑗 It shows that, in equilibrium, capital is allocated such that differences in monitoring costs and differences in returns are optimally traded off. The larger the return on loans is in a country relative to the other countries and the more efficient a country’s banks are relative to other banking sectors, the larger the capital flow is into that country. Every additional unit of capital thatisemployedinproductionincountry𝑖mustbeintermediatedbybanksincountry𝑖(foreign banks cannot lend to firms in country 𝑖). Thus a more efficient banking sector in country 𝑖 implies that the economic loss from monitoring an additional firm (or unit of capital) there is lower than in other countries. As a result, it is beneficial to employ more capital in production in country 𝑖, even when the return on loans is the same in all countries. 2.3 Open economy with cross-border lending and affiliate lending Inthesecondopeneconomyscenario, bankscanlendtofirmsabroad, inadditiontolendingand borrowing on the international interbank market. Lending to firms abroad is costly, however. Banks from country 𝑗 that lend cross-border to firms in country 𝑖 have to pay the fixed cost 𝑓𝑋 > 0. These costs can be interpreted as fixed costs associated with acquiring information 𝑖𝑗 about the business environment abroad, for example, or about establishing client relationships there. It is further assumed that banks can grow their balance sheet if they lend to firms abroad. Specifically, we model the profit function of a banker from country 𝑗 that lends to firms in country 𝑖 ∈ 𝑁 as follows: (︃ )︃ 𝑁 ∑︁ 1 𝜋 (𝑎 ) = 𝑅 𝑧 − ℎ(𝑧 )−𝑅𝐼𝑧 −𝑓𝑋 +𝑅𝐼𝑑, (11) 𝑗 𝑗 𝑖 𝑖𝑗 𝜑 (𝑎 )𝛿𝑋𝑎 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖=1 𝑖𝑗 𝑗 𝑖𝑗 𝑗 where 0 ≤ 𝛿𝑋 < 1, 𝑓𝑋 = 0, 𝛿𝑋 = 1 and 𝜑 (𝑎 ) = 1. In this formulation, banks’ monitoring 𝑖𝑗 𝑗𝑗 𝑗𝑗 𝑗𝑗 𝑗 costs at home and abroad are separable so that their decision to engage in cross-border lending is independent of lending at home and banks seek to replicate their business abroad. This assumption can be motivated by “love for variety” in loans, for example. If banks can offer differentiated loans, then each bank specializes in providing a particular type of loan or in lending to a particular type of firm/sector, and it is optimal that every bank operates in every country for sufficiently low fixed costs.13 13Analternativeinterpretationofourassumptionisthatbankswanttoinvestathomeandabroadinorderto 6

𝛿𝑋 reflects inversely the efficiency loss of a banker from country 𝑖 that lends to firms in 𝑖𝑗 country 𝑗 cross-border. This efficiency loss can be due to information frictions, since it may be harder for firms to access information about clients abroad. It can also reflect greater transaction costs.14 We allow the efficiency loss to differ across banks as some banks may have an advantage/disadvantage in operating in certain countries. Bankers draw a parameter 𝜑 (𝑎 ) 𝑖𝑗 𝑗 from a distribution 𝑚(𝜑) with mean 1 and choose to lend to firms in country 𝑖 if: 1 𝜋𝑋(𝑎 ) = 𝑅 𝑧𝑋 − ℎ(𝑧𝑋)−𝑅𝐼𝑧𝑋 −𝑓𝑋 ≥ 0, (12) 𝑖𝑗 𝑗 𝑖 𝑖𝑗 𝜑 (𝑎 )𝛿𝑋𝑎 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑗 𝑖𝑗 𝑗 where 𝑧𝑋 = 𝑎 𝜑 (𝑎 )𝛿𝑋(𝑅 − 𝑅𝐼). The banker who breaks even on the lending business in 𝑖𝑗 𝑗 𝑖𝑗 𝑗 𝑖𝑗 𝑖 country 𝑖 is then given by: 2𝑓𝑋 𝜑 (𝑎 )𝑎𝑋 = 𝑎˜𝑋 = 𝑖𝑗 . (13) 𝑖𝑗 𝑗 𝑖𝑗 𝑖𝑗 (𝑅 −𝑅𝐼)2𝛿𝑋 𝑖 𝑖𝑗 The lower the fixed cost 𝑓𝑋 and the efficiency loss from operating cross-border 𝛿𝑋 are, the 𝑖𝑗 𝑖𝑗 lower is the efficiency of the banker that makes zero profits on the cross-border operations in country 𝑖 𝑎˜𝑋. 𝑖𝑗 Instead of lending cross-border, banks can open up affiliates in the foreign country 𝑖 for a fixed cost 𝑓𝐹 > 𝑓𝑋 − 𝐾 /𝑀 max{𝑅 ,𝑅 ,...,𝑅 }.15 A foreign affiliate has the advantage 𝑖𝑗 𝑖𝑗 𝑖 𝑖 1 2 𝑁 of lowering banks’ monitoring costs when lending abroad. Banks still face a friction in the form of 𝛿𝐹 ≤ 1 but the friction is assumed to be lower than for cross-border operations, i.e. 1 ≥ 𝛿𝐹 > 𝛿𝑋. Having an affiliate also allows banks to raise foreign deposits.16 If a banker of 𝑖𝑗 𝑖𝑗 type 𝑎 establishes an affiliate in country 𝑖, his profits that come solely from operations in that 𝑗 country are: 1 𝜋𝐹(𝑎 ) = 𝑅 𝑧𝐹 − ℎ(𝑧𝐹)−𝑅𝐼𝑧𝐹 −𝑓𝐹 +𝑅𝐼𝑑 , (14) 𝑖𝑗 𝑗 𝑖 𝑖𝑗 𝜑 (𝑎 )𝑎 𝛿𝐹 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑗 𝑗 𝑖𝑗 where 𝑧𝐹 = 𝜑 (𝑎 )𝑎 𝛿𝐹(𝑅 − 𝑅𝐼). A banker chooses to open up an affiliate abroad if the 𝑖𝑗 𝑖𝑗 𝑗 𝑗 𝑖𝑗 𝑖 resulting profits are positive and higher than the profits from lending cross-border. The banker who is indifferent between cross-border lending and operating through an affiliate is found by settingprofitsundercross-borderlendingtomarket𝑖equaltoprofitswithanaffiliateincountry 𝑖: 2 1 𝜑 (𝑎 )𝑎𝐹 = 𝑎˜𝐹 = (𝑓𝐹 −𝑓𝑋 −𝑅𝐼𝑑 ). (15) 𝑖𝑗 𝑗 𝑖𝑗 𝑖𝑗 (𝑅 −𝑅𝐼)2𝛿𝐹 −𝛿𝑋 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖 𝑖𝑗 𝑖𝑗 diversify. If risk is reduced, banks may be able to increase their leverage and, thereby, the size of their balance sheets. Love for variety in loans is modeled in de Blas and Russ (2010). 14Thereisempiricalevidencethatinformationfrictionsanddistanceaffectbanks’foreignactivities. SeeBuch (2003), Focarelli and Pozzolo (2005) and Degryse and Ongena (2005). 15The condition insures that the fixed cost of establishing an affiliate 𝑓𝐹 net of the benefits from raising 𝑖𝑗 deposits in country 𝑖 (𝑅𝐼𝑑 ) is always larger than the fixed cost of cross-border lending. 𝑖 16An affiliate in our framework can be interpreted both as a branch and a subsidiary, although the interpretation as a subsidiary is preferred. Branches often facilitate lending to or borrowing from foreign banks or wholesale investors. In contrast, subsidiaries make it easier for banks to raise retail deposits in a foreign market. The model could distinguish between branch and subsidiary by assuming that a branch implies 𝛿𝑋 = 𝛿𝐹 𝑖𝑗 𝑖𝑗 whereas a subsidiary allows banks to compete for foreign deposits. 7

𝑎˜𝐹 depends on the relative attractiveness of the two modes of entry. The lower 𝑓𝐹 and the 𝑖𝑗 𝑖𝑗 higher 𝛿𝐹 are compared to 𝛿𝑋 and 𝑓𝑋, the lower is 𝑎˜𝐹. 𝑎˜𝐹 is also a function of the interbank 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 rate. The higher the interbank rate 𝑅𝐼 is, the more attractive it is to raise additional deposits abroad and the more bankers establish an affiliate in country 𝑖, resulting in a lower 𝑎˜𝐹. 𝑖𝑗 The open economy model is closed by two equilibrium conditions. First, the capital invested in each country by all banks must equal the world capital endowment: 𝑁 𝑁 ∑︁ ∑︁ ˜ 𝐾 = 𝐾 , (16) 𝑖 𝑖 𝑖=1 𝑖=1 where 𝐾 ˜ = 𝑀 ∫︁ 𝑎𝑖 𝑧 𝑔 (𝑎 )𝑑𝑎 + ∑︁ 𝑁 𝑀 ∫︁ 𝑎˜𝐹 𝑖𝑗 𝑧𝑋𝑛 (𝑎˜ )𝑑𝑎˜ + ∑︁ 𝑁 𝑀 ∫︁ 𝑎˜𝑖𝑗 𝑧𝐹𝑛 (𝑎˜ )𝑑𝑎˜ . (17) 𝑖 𝑖 𝑖𝑖 𝑖 𝑖 𝑖 𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑎 𝑎˜𝑋 𝑎˜𝐹 𝑖 𝑗=1,𝑗̸=𝑖 𝑖𝑗 𝑗=1,𝑗̸=𝑖 𝑖𝑗 𝑛 (𝑎˜ ) denotes the joint distribution of 𝜑 (𝑎 )𝑎 and 𝑎˜ is its upper support. 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑗 𝑗 𝑖𝑗 Second, bankers with affiliates in a country compete for local deposits. Each banker obtains the capital stock divided by the mass of bankers that compete for the deposits. The following condition must hold for each country 𝑖: ∫︁ 𝑎𝑖 ∑︁ 𝑁 ∫︁ 𝑎˜𝑖𝑗 𝐾 = 𝑀 𝑑 𝑔 (𝑎 )𝑑𝑎 + 𝑀 𝑑 𝑛 (𝑎˜ )𝑑𝑎˜ . (18) 𝑖 𝑖 𝑖𝑖 𝑖 𝑖 𝑖 𝑗 𝑖𝑗 𝑗 𝑖𝑗 𝑖𝑗 𝑎 𝑎˜𝐹 𝑖 𝑗=1,𝑗̸=𝑖 𝑖𝑗 Because domestic bankers and foreign bankers with affiliates in market 𝑖 raise the same amount of deposits, implying 𝑑 = 𝑑 = 𝑑 . Solving for 𝑑 yields: 𝑖𝑖 𝑖𝑗 𝑖 𝑖 𝐾 𝑖 𝑑 = . (19) 𝑖 𝑀 + ∑︀𝑁 𝑀 ∫︀𝑎˜𝑖𝑗 𝑛 (𝑎˜ )𝑑𝑎˜ 𝑖 𝑗=1,𝑗̸=𝑖 𝑗 𝑎˜𝐹 𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 Proposition 1 There exists a unique solution to the open economy if 𝑓𝐹 > 𝑓𝑋 − 𝐾 /𝑀 𝑖𝑗 𝑖𝑗 𝑖 𝑖 max{𝑅 ,𝑅 ,...,𝑅 }. 1 2 𝑁 With cross-border and affiliate lending, the equilibrium interbank lending rate increases compared to the previous scenario with international interbank lending only. Because the more efficient banks extend loans to firms abroad in addition to lending domestically, their demand for interbank funds increases. For smaller banks to be willing to provide these funds, the interbank lending rate must go up. Moreover,thetradeoffbetweenallocatingcapitalefficientlyandminimizingmonitoringcosts is alleviated, since banks can now lend directly to foreign firms and the intermediation does not have to be done by the domestic banks. As a consequence, more capital flows into the country with the higher return to capital. We revisit the implications of openness to foreign bank entry for net capital flows and the composition of banks’ foreign activities in section 4. 8

3 The Barriers to German Banks’s Foreign Operations The model provides a structural framework that can be applied to the data and used to back out the parameters 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 as well as 𝑓𝐹. In this section, we describe our strategy to obtain 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 values for these parameters based on detailed bank-level data available at the Deutsche Bundesbank for all foreign countries in which German banks operate. Before using the computed values to calibrate the model (see section 4), we compare the frictions that German banks face abroad across host countries. As one might expect, these are strongly related to proxies of bank entry barriers, such as financial openness or bureaucratic quality. We also check whether the modeled relationship between bank size and net interbank lending is supported by the data. In line with the theory, larger banks are more likely to be borrowers on the interbank market and borrow more on net than smaller banks. 3.1 Structural equations For ease of notation, a bank is now denoted by subscript b. Subscript 𝑗, which is dropped where possible, stands for the bank’s home country (Germany in our application). Country 𝑖 stands for the foreign country the (German) bank is lending to. 𝑅 −𝑅𝐼 is referred to as the 𝑖 net interest margin prevailing in country 𝑖, the interest collected on loans net of (interbank) funding costs. Bank efficiency distribution With quadratic monitoring costs, a bank’s domestic lending is given by 𝑧 = 𝑎 (𝑅 −𝑅𝐼). The efficiency of bank 𝑏 relative to bank 𝑏′ is then equal to the 𝑏 𝑏 𝑗 ratio of the two banks’ domestic loans to firms: 𝑎 𝑧 𝑏′ 𝑏′ = . (20) 𝑎 𝑧 𝑏 𝑏 Denoting the bank with the largest domestic non-bank private sector lending by 𝑏 and normalizing this bank’s efficiency parameter so that 𝑎 = 𝑧 𝑏 , where 𝑅 −𝑅𝐼 is proxied by the 𝑏 𝑅𝑗−𝑅𝐼 𝑗 domestic net interest margin, we can obtain the efficiency parameter 𝑎 for each bank 𝑏. 𝑏 Distribution of 𝜑 The model also prescribes a relationship between a bank’s efficiency parameter 𝑎 and its cross-border lending and affiliate lending, respectively. Recall that 𝑧𝑋 = 𝑏 𝑏𝑖 𝑎 𝜑 𝛿𝑋(𝑅 − 𝑅𝐼) and 𝑧𝐹 = 𝑎 𝜑 (𝑅 − 𝑅𝐼), where 𝜑 is bank 𝑏’s idiosyncratic advantage of 𝑏 𝑏𝑖 𝑖 𝑖 𝑏𝑖 𝑏 𝑏𝑖 𝑖 𝑏𝑖 lending to country 𝑖. Dividing bank 𝑏’s cross-border lending to firms in country 𝑖 by bank 𝑏′’s cross-border lending to firms in country 𝑖 and solving for the ratio 𝜑 𝑏′𝑖, we have: 𝜑 𝑏𝑖 𝜑 𝑧𝑋 𝑎 𝑏′𝑖 = 𝑏′𝑖 𝑏 . (21) 𝜑 𝑧𝑋 𝑎′ 𝑏𝑖 𝑏𝑖 𝑏 Equivalently: 𝜑 𝑧𝐹 𝑎 𝑏′𝑖 = 𝑏′𝑖 𝑏 . (22) 𝜑 𝑧𝐹 𝑎′ 𝑏𝑖 𝑏𝑖 𝑏 9

∑︀ Exploiting 𝐸(𝜑 ) = 1 to normalize 𝜑 , that is, setting 𝜑 = 1, we obtain for each bank 𝑏 𝑏𝑖 𝑏𝑖 𝑏 𝑏𝑖 that lends to country 𝑖 a value for 𝜑 .17 𝑏𝑖 Solving for 𝛿𝑋 and 𝛿𝐹 The model implies the following ratio of bank 𝑏’s cross-border loans in country 𝑖 to its domestic loans: 𝑧𝑋 𝑎 𝜑 𝛿𝑋(𝑅 −𝑅𝐼) 𝑏𝑖 = 𝑏 𝑏𝑖 𝑖 𝑖 . (23) 𝑧 𝑎 (𝑅 −𝑅𝐼) 𝑏 𝑏 𝑗 Solving for 𝛿𝑋 yields: 𝑖 𝑧𝑋 (𝑅 −𝑅𝐼) 𝛿𝑋 = 𝑏𝑖 𝑗 . (24) 𝑖 𝑧 (𝑅 −𝑅𝐼)𝜑 𝑏 𝑖 𝑏𝑖 Havingdetermined𝜑 , wecancalculate𝛿𝑋 foreachcountry𝑖usinginformationoncross-border 𝑏𝑖 𝑖 loans 𝑧𝑋 and domestic loans 𝑧 from the data. 𝑏𝑖 𝑏 Equivalently, 𝛿𝐹 can be obtained through: 𝑖 𝑧𝐹 (𝑅 −𝑅𝐼) 𝛿𝐹 = 𝑏𝑖 𝑗 . (25) 𝑖 𝑧 (𝑅 −𝑅𝐼)𝜑 𝑏 𝑖 𝑏𝑖 Solving for 𝑓𝑋 and 𝑓𝐹 The model solution for the cross-border lending cutoff 𝑎˜𝑋, that is, 𝑖 the efficiency of the bank that breaks even on its cross-border lending to country 𝑖 is: 2𝑓𝑋 𝑎˜𝑋 = 𝑖 . (26) 𝑖 (𝑅 −𝑅𝐼)2𝛿𝑋 𝑖 𝑖 Solving for the fixed cost 𝑓𝑋 delivers: 𝑖 1 𝑓𝑋 = 𝑎˜𝑋(𝑅 −𝑅𝐼)2𝛿𝑋. (27) 𝑖 2 𝑖 𝑖 𝑖 With information on the net interest margin prevailing in country 𝑖 and the cutoff banker 𝑎˜𝑋 𝑖 from the bank-level data, 𝑓𝑋 can be obtained by applying the above formula. 𝑖 Similarly, the model provides an expression for the FDI cutoff 𝑎˜𝐹, i.e. the efficiency of the 𝑖 banker who is indifferent between lending cross-border or through an affiliate to market 𝑖: 2 1 𝑎˜𝐹 = (𝑓𝐹 −𝑓𝑋 −𝑅𝐼𝑑 ). (28) 𝑖 (𝑅 −𝑅𝐼)2𝛿𝐹 −𝛿𝑋 𝑖 𝑖 𝑖 𝑖 𝑖 𝑖 Solving for 𝑓𝐹 −𝑅𝐼𝑑 delivers: 𝑖 𝑖 1 𝑓𝐹 −𝑅𝐼𝑑 = 𝑎˜𝐹(𝑅 −𝑅𝐼)2(𝛿𝐹 −𝛿𝑋)+𝑓𝑋. (29) 𝑖 𝑖 2 𝑖 𝑖 𝑖 𝑖 𝑖 All the elements on the right hand side of the above expression are known from the data. So we can compute 𝑓𝐹 −𝑅𝐼𝑑 , the fixed cost of establishing a foreign affiliate net of the benefit 𝑖 𝑏𝑖 17Note that 𝜑 is normalized separately for banks with cross-border lending and banks with affiliate lending. 𝑏𝑖 10

from raising foreign deposits, for each country 𝑖. 3.2 Data To implement the described strategy, we draw on balance sheets and foreign positions reports which German banks file with the Deutsche Bundesbank. These statistics are unique in that they provide extremely detailed information on German banks’ domestic and foreign activities. In particular, one can observe each bank’s cross-border assets and liabilities as well as the positions of its foreign subsidiaries and branches by sector and country. Available to us for this paper is monthly information for 2005, which we average over 12 months.18 The sample includes roughly 2,000 German banks, covering essentially the entire German banking sector except a few foreign-owned banks. Almost all of the banks in our sample have some non-zero foreign position but only around 50 have affiliates abroad. German banks conduct operations in around 180 foreign countries. In addition to German bank-level data, we also use information on net interest margins across host countries from the World Bank’s Financial Development and Structure Database (see Beck et al. (2000)). Summary statistics of all data used in the analysis are presented in table 1. 3.3 Distribution of 𝑎 and 𝜑 Calculating 𝑎 and 𝜑 The efficiency parameter 𝑎 for each German bank is calculated based 𝑏 on equation (20). 𝑧 is proxied by the loans of German bank 𝑏 to the domestic non-bank private 𝑏 sector. Figure 2 shows the distribution of 𝑎 obtained from the German data. Because the 𝑏 relationship between 𝑎 and 𝑧 is linear, the distribution reflects essentially the size distribution 𝑏 𝑏 of German banks. The distribution resembles the Pareto distribution, a feature we exploit in the next section. We also obtain 𝜑 for each bank 𝑏 that lends to host country 𝑖. 𝜑 reflects bank 𝑏’s 𝑏𝑖 𝑏𝑖 idiosyncratic advantage in lending to country 𝑖 and indirectly indicates how well the size of a bank’s domestic lending predicts its foreign lending. In order to calculate 𝜑 , we divide 𝑏𝑖 German parent banks in three groups:19 (i) banks which do not have positions in country 𝑖, for which we cannot calculate 𝜑 (domestic banks); (ii) banks which do not lend through foreign 𝑏𝑖 affiliates to country 𝑖 but which extend loans cross-border to country 𝑖 (cross-border banks). For these banks we obtain 𝜑 by applying equation (21). (iii) Banks whose affiliates abroad 𝑏𝑖 lend to country 𝑖 (banks with FDI). In equation (22), 𝑧𝐹 is proxied by the sum of total affiliate 𝑏𝑖 lending to country 𝑖, no matter whether the affiliate is located in country 𝑖 or in a third country, plus cross-border lending by the parent to country 𝑖.20 18Research Data and Service Centre of the Deutsche Bundesbank, Monthly Balance Sheet Statistics and External Positions of Banks, 2005. 19The described grouping is based specifically on German banks’ positions vis-`a-vis the non-bank private sector in country 𝑖. 20For a discussion of the non-negligible role of German banks’ third-country affiliates, see Frey and Kerl (2015). 11

The relationship between domestic lending and foreign lending A bank’s domestic lending volume is, in general, a strong predictor of its foreign lending. To illustrate this, we rank banks according to their cross-border loans to country 𝑖 and correlate this ranking with the ranking based on their lending to German firms (𝑎 ). The distribution of rank correlation 𝑏 coefficients across host countries is shown in figure 3. The average rank correlation coefficient is 33 percent. Rank correlations are low or negative for countries that receive only few loans from German banks, for example, Guyana (correlation coefficient of -0.5) or Honduras (correlation coefficient of 0.1). For cross-border lending to firms in these countries, special knowledge about local markets or relationships may be highly relevant. Rank correlations are much higher for countrieslikeItaly(correlationcoefficientof0.5)ortheUS(correlationcoefficientof0.6), which are key destinations for German banks’ cross-border loans. Similar results emerge for banks’ foreign lending via affiliates. The average rank correlation across host countries is slightly higher at 54 percent.21 Figure 4 highlights the role played by 𝜑 in the model, which captures the idiosyncratic 𝑏𝑖 advantage/disadvantage that bank 𝑏 faces when lending to country 𝑖.22 Banks’ cross-border lending to the non-bank private sector 𝑧𝑋 is plotted against the efficiency parameter 𝑎 in 𝑏𝑖 𝑏 the chart (scattered dots). If banks did not have any idiosyncratic advantage/disadvantage in lending to country 𝑖, 𝜑 would be equal to 1 for each bank and all dots would lie on the straight 𝑏𝑖 line depicted in the chart. Applying the strategy described in the previous section, 𝜑 is set for 𝑏𝑖 each bank to the value that makes the relationship between 𝜑 𝑎 and 𝑧𝑋 linear. The horizontal 𝑏𝑖 𝑏 𝑏𝑖 arrows in the chart reflect the adjustment through 𝜑 . 𝑏𝑖 3.4 Comparing 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 − 𝑅𝐼𝑑 across host countries Calculating 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 − 𝑅𝐼𝑑 The parameters 𝛿𝑋 and 𝛿𝐹 reflect inversely the 𝑖 𝑖 𝑖 efficiency loss that banks encounter when operating in country 𝑖 cross-border and through a foreign affiliate, respectively. The two parameters affect the intensive margin, that is, the volume of bank lending abroad. They are calculated based on equations (24) and (25). Tocomputethefixedcost𝑓𝑋 ofcross-borderlendingandthefixedcost𝑓𝐹 ofestablishingan 𝑖 𝑖 affiliate in country 𝑖 , we follow equations (27) and (29). For the calculation of the fixed costs, which affect the extensive margin of banks’ foreign operations, we use a narrower definition to group banks into domestic, cross-border and FDI with respect to country 𝑖. Banks are only classified as having FDI if they have an affiliate in country 𝑖.23 The cross-border lending cutoff 𝑎˜𝑋 corresponds to the lowest value of 𝑎˜ = 𝜑 𝑎 observed within the group of cross-border 𝑖 𝑖 𝑏𝑖 𝑏 banks. Equivalently, the FDI cutoff 𝑎˜𝐹 is set to the lowest value of 𝑎˜ = 𝜑 𝑎 observed within 𝑖 𝑖 𝑏𝑖 𝑏 the group of banks with FDI. 21These findings are consistent with Buch et al. (2011) and Niepmann (2013) who show that bank efficiency predicts the intensive margin (and extensive margin) of banks’ foreign activities. 22We use artificially constructed data in this graph as original bank-level data cannot be shown due to confidentiality. 23The set of banks with local affiliate lending to country 𝑖 is smaller than the set of banks with local and/or affiliate lending channelled from third countries to country 𝑖. 𝜑 is renormalized after banks are regrouped so 𝑏𝑖 that the mean of 𝜑 within each group is 1. 𝑏𝑖 12

Comparing 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 − 𝑅𝐼𝑑 across host countries The model puts some restrictions on the parameters 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 − 𝑅𝐼𝑑 . For example, the model requires 𝑖 𝑖 𝑖 𝑖 𝑖 𝛿𝑋 < 𝛿𝐹 ≤ 1 ∀ 𝑖. All values obtained for 𝛿𝑋 and 𝛿𝐹 are in fact below 1 and 𝛿𝑋 < 𝛿𝐹 holds for 𝑖 𝑖 𝑖 𝑖 𝑖 𝑖 all countries in our sample. Moreover, the fixed cost of cross-border lending 𝑓𝑋 is smaller than 𝑖 the fixed cost of establishing an affiliate net of the benefits from raising additional funding in market 𝑖 (𝑓𝐹 − 𝑅𝐼𝑑 ), consistent with the model assumptions. Table 2 shows the computed 𝑖 𝑖 values for 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 −𝑅𝐼𝑑 for the different host countries in the sample. All values are expressed relative to the values obtained for the US for better illustration. Countries are sorted according to 𝛿𝑋. Overall, German banks encounter relatively low frictions when lending to geographically close areas as well as European financial centers such as Luxembourg, Switzerland and Ireland. Beyond this general pattern, the efficiency loss from lending cross-border (inversely related to 𝛿𝑋) is small for the US compared to other countries, which is likely related to the country’s large and competitive corporate loan market that is fairly open to foreign lenders. The fixed cost of establishing an affiliate is not only low in geographically close markets but also in more distant financial centers such as Hong Kong or Singapore. This fits well with these countries’ policies to attract foreign banks, for example, through favorable tax conditions. The efficiency loss from lending via affiliates (inversely related to 𝛿 ) is particularly small in Central and 𝐹 Eastern European countries such as Bulgaria and Romania. This finding matches the fact that several German banks operate large subsidiaries in these markets which supply loans not only locally but also across borders to the whole region (see Frey and Kerl (2015)). Regressing 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 − 𝑅𝐼𝑑 on host country characteristics To assess further whether the computed measures of the barriers to German banks’ foreign operations are sensible, we conduct a simple cross-sectional regression analysis. We regress 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑖 𝑖 𝑖 𝑓𝐹−𝑅𝐼𝑑 onseveralcountrycharacteristics.24 Inparticular, weincludeasexplanatoryvariables 𝑖 𝑖 host country GDP, distance to Germany and three proxies for bank entry barriers following Niepmann (2013): the Chinn and Ito index, which measure de jure financial openness (see Chinn (2008)), as well as measures of a country’s bureaucratic quality and property rights protection from the World Bank.25 Controlling for GDP is essential since a country’s size is a strong predictor of foreign bank activity (see, for example, Aviat and Coeurdacier (2007)) and differences in size across host markets are not automatically accounted for in the calculation of bank entry barriers. Summary statistics for all variables used in the empirical analysis are given in table 1. Regression results are presented in tables 3 and 4. To get a better sense of how relevant the different variables are for predicting 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 −𝑅𝐼𝑑, we calculate semipartial correlations between explanatory variables and dependent variables. Semipartial correlations measure the decrease in R-squared that results when the respective variable is omitted from the regression. They are presented in the bottom rows of tables 3 and 4. Overall, the efficiency loss from lending abroad and associated fixed costs tend to be lower 24The three highest and lowest values of each dependent variable are excluded from the sample. This essentially means excluding countries in which only very few German banks operate, that is, countries for which the calculated values of 𝛿𝑋, 𝑓𝑋, 𝛿𝐹 and 𝑓𝐹 −𝑅𝐼𝑑 are based on only a few observations. 25See the data appendix for data sources and further details on the different variables. 13

in countries with a better bureaucracy, better property rights protection and greater financial openness. The predictive power of these variables is particularly strong for the efficiency loss associated with cross-border and affiliate lending (inversely related to 𝛿𝑋 and 𝛿𝐹, respectively). Distance is linked to the efficiency loss from cross-border lending (see table 3, columns 1 to 3) butisnotcorrelatedwiththeefficiencylossfromaffiliatelendingorthefixedcostofestablishing an affiliate in country 𝑖 (see table 4). This indicates that local affiliates eliminate frictions that arise from distance. The fixed cost of cross-border lending and establishing an affiliate are significantly influenced by GDP, however, in opposite directions (see columns (4) to (6) in tables 3 and 4). While the fixed cost of cross-border lending 𝑓𝑋 is lower in countries with a bigger GDP, the fixed cost of FDI is higher. The positive effect of GDP on 𝑓𝐹 is somewhat surprising and may reflect tougher competition in bigger markets. 3.5 Interbank lending in the model and the data The model yields predictions regarding each bank’s interbank claims as a function of its efficiency. In the closed economy model and the open economy with interbank lending only, we have: 𝑧 −𝑑 = 𝑎 (𝑅−𝑅𝐼)−𝑑 , (30) 𝑏 𝑏 𝑏 𝑏 where 𝑑 = 𝑑 ∀ 𝑏. The equation implies that a bank’s net interbank borrowing is an increasing 𝑏 functionofitsefficiency. Thisrelationshipstemsfromtheassumptionthateachbankisendowed with the same amount of depositor capital but faces different monitoring costs. In the open economy model with cross-border and affiliate lending, the statement remains true under the sufficient condition that 𝑧𝐹 −𝑑 ≥ 0, in other words, the amount of additional deposits that 𝑏𝑖 𝑏𝑖 banks with affiliates can raise abroad is not larger than the funds that these banks want to lend to non-banking firms abroad.26 Figure 5 plots the relationship of net interbank claims and bank efficiency 𝑎 in the German 𝑏 data obtained from a simple cross-sectional regression. Net interbank claims are computed as the difference between a bank’s consolidated lending to foreign and domestic banks and its borrowing from these entities. Consistent with the model, there is a clear negative relationship between the two. Larger banks (higher 𝑎 ) are more likely to be net borrowers on the interbank 𝑏 market and borrow more on net than smaller banks. This finding is in line with descriptions of the structure of interbank markets in Stigum (1990) and Craig and von Peter (2014). 4 The Composition of International Bank Flows In this section, we calibrate the model to the data and undertake a couterfactual analysis to study how the composition of international bank flows changes with impediments to foreign bank operations. We do not have data at hand that would allow us to sensibly simulate the N-country model, so we work in the following with a two-country version. Specifically, we use 26Thesufficientconditionguaranteesthatabank’sdemandforinterbankfundsdoesnotdeclinewhenitopens up an affiliate in a foreign country. 14

the previously calculated parameters and detailed information on German bank positions in the US to simulate the model and bank flows between these two countries. 4.1 Parameters used for simulation The parameter values used for the simulation are summarized in table 5. As figure 2 showed, the distribution of the efficiency parameter 𝑎 among German banks resembles the Pareto 𝑏 distribution. We assume a truncated Pareto distribution and estimate the shape parameter based on the computed 𝑎 ’s for German banks with Maximum Likelihood. The upper limit of 𝑏 the truncated Pareto is set equal to the 99th percentile of the observed distribution of 𝑎 .27 𝑏 The bank efficiency distribution for the USA is assumed to have the same shape parameter and the same upper support as for German banks.28 The lower support of each country’s bank efficiency distribution is chosen so that each country’s net interest margin under autarky (𝑅 −𝑅𝐼) matches the 1998 value of the net interest margin, which is the earliest value available 𝑖 from Beck et al. (2000).29 This implies assuming that German banks are, on average, more efficient than US banks (average efficiency of 7.33e+07 for German banks versus 1.58e+07 for US banks). The return on loans in Germany is set to 10%. The return on loans in the US is determined by the difference in net interest margins between the US and Germany in 2005 from the Financial Structure Database, which implies 𝑅 = 𝑅 +0.0252. 𝑈𝑆𝐴 𝐺𝐸𝑅 𝑎˜ = 𝜑 𝑎 is also assumed to be Pareto distributed. The shape parameter of that 𝑏𝑈𝑆 𝑏𝑈𝑆 𝑏 distribution is estimated based on the computed values for 𝑎˜ . The upper limit corresponds 𝑏𝑈𝑆 to the highest value of 𝑎˜ in the data. The lower limit of the distribution is not observed 𝑏𝑈𝑆 because 𝜑 can only be computed for banks that extend loans to US firms. We therefore 𝑏𝑈𝑆 choose the lower limit of the distribution so that the share of German banks that lend to US firms equals the empirical counterpart (61.85%), where the cross-border cutoff 𝑎˜𝑋 is set equal 𝑏𝑈𝑆 to the lowest value of 𝑎˜ observed among banks with cross-border loans in the US. 𝑏𝑈𝑆 The number (mass 𝑀 ) of German banks is taken from the German bank-level data 𝐺𝐸𝑅 and is 1,995. The number (mass 𝑀 ) of US banks in the year 2005 comes from Janicki and 𝑈𝑆 Prescott (2006), which is 6,500. The capital endowment 𝐾 of German banks is set equal to 𝐺𝐸𝑅 the total deposits collected from German residents. US banks hold twice as many deposits as German banks, so we set 𝐾 = 2×𝐾 .30 𝛿𝑋 , 𝑓𝑋 , and 𝛿𝐹 are based on the computations 𝑈𝑆 𝐺𝐸𝑅 𝑈𝑆 𝑈𝑆 𝑈𝑆 described in the previous section. The fixed cost 𝑓𝐹 of establishing an affiliate in the US is chosen so that German banks’ total loans to the US non-bank private sector match those in 27We exploit the fact that when a Pareto distribution is truncted, the resulting distribution is also Pareto distributed and has the same shape parameter. 28Bremus et al. (2013) find similar Pareto shape parameters for Germany and the US. 29Ideally, we would like to go back further in time. However, the large rise in BIS foreign claims took place after 1998 as indicated by figure 1 so using 1998 values for autarky net interest margins seems sensible. 30See Board of Governors of the Federal Reserve Systems, http://www.federalreserve.gov/econresdata/releases/statisticsdata.htm, table H.8 - Assets and Liabilities of Commercial Banks in the U.S., and the Deutsche Bundesbank, Monthly Report, http://www.bundesbank.de/Navigation/EN/Publications/Monthly reports/monthly reports.html, table IV.2 Banks - Principal Assets and Liabilities of Banks (MFIs) in Germany, by Category of Banks. 15

the data (=C226 bn). Because we do not have detailed data on US banks’ positions in Germany, we assume that US banks lend to and borrow only from domestic firms and German banks but they do not lend to German non-banking firms or raise deposits in Germany. This implicitly means setting the entry costs for US banks prohibitively high.31 4.2 Comparing the model to the data Column (3) of table 6 presents key characteristics of the simulated open economy. Column (5) of the same table shows the corresponding moments from the data. Overall, the model matches the data well. Recall that 𝑓𝐹 was chosen so that German banks’ total loans to the 𝑈𝑆 US non-bank private sector from the model and the data coincide. The share of banks that engage in cross-border lending in the model is 62.68% compared to 61.25% in the data. The share of banks with affiliates is lower than in the data (0.026% in the model versus 0.6% in the data), while total local lending is higher (C=223 bn in the model versus C=199 bn in the data). Net lending by German banks to US banks is negative both in the model and in the data so German banks are net borrowers from US banks on the international interbank market. The model generates net interbank lending of =C-77 bn compared to =C-69 bn in the data. 4.3 Counterfactual analysis We are interested in studying how US openness to German bank operations affects the two economies. Based on the baseline simulation, we analyze the following three scenarios: (i) autarky, (ii) the case in which banks can only lend and borrow on the international interbank market and (iii), a reduction in the cost of establishing affiliates in the US faced by German banks of 10 percent compared to the baseline calibration. Columns (1), (2), and (4) of table 6 show model outcomes for these scenarios. Autarky Under autarky, each banking sector only lends to domestic firms and intermediates domestic funds. The net interest margin in Germany and the US are 1.34 % and 3.81%, respectively, and correspond to the 1998 values in the Financial Structure Database.32 Net capital flows and interbank flows are zero. International interbank market When interbank markets integrate, interbank rates equilibrate. Notethatbecausebankscanonlylendtoforeignbanksbutnottoforeignfirmsdirectly, each additional unit invested in one of the two countries must be intermediated by domestic banks. While the return on loans is higher in the US than in Germany, US banks incur higher monitoring costs than German banks. Whether capital flows in equilibrium to Germany or the 31Thisisnotanunreasonableassumption. AccordingtotheBISstatistics,theclaimsofGermanbanksonthe US non-bank private sector in 2005 were roughly 20 times higher than the claims of US banks on the German non-bank private sector. 32Note that the lower limits of the German and US bank efficiency distributions were set so that the model matches these values in autarky. 16

US thus depends on the relative magnitudes of differences in the return on loans and differences in average bank efficiencies across countries. These are reflected implicitly in autarky interbank rates. The autarky interbank rate is higher in the US than in Germany (8.66% versus 8.71%), implying that the higher return on loans in the US outweighs the higher monitoring costs. In equilibrium, C=33.75 bn flow through the interbank market from Germany to the US and the interbank rate equilibrates at 8.68%. Accordingly, net interest margins in the two countries move in opposite directions. The net interest margin falls in Germany to 1.32% and rises in the US to 3.84%. Baselinecalibration Whenthebarrierstointernationalbankoperationsarenotprohibitively high and 𝛿𝑋, 𝛿𝐹, 𝑓𝐹 and 𝑓𝑋 take the calibrated values, the capital flow into the US rises to C=182.82 bn. German banks find it profitable to replicate their business in the US and lend to US firms directly. This brings monitoring costs in the global economy down so that more capital can flow to the high return location. As German banks expand their balance sheets, their demand for interbank funds rises. As a result, the interbank lending rate increases and net interest margins fall in both countries to 1.21% and 3.73%, respectively. The German banking sector expands in size, while the US banking sector, which does not operate abroad, contracts since German banks take over part of the intermediation business from US banks. Importantly, the direction of the net interbank flow reverses. To finance their lending to US firms, German banks take deposits in the US and, in addition, borrow C=76.85 bn from US banks on the interbank market. 10% fall in 𝑓𝐹 Column (4) shows the implications of a 10 percent fall in the cost 𝑓𝐹 that GermanbanksfacewhenestablishinganaffiliateintheUS.Asthebarrierstoforeignbankentry go down, banks switch from cross-border lending to lending through local affiliates and increase their loan volumes to US firms. This, in turn, raises the demand for interbank funding so that the interbank lending rate rises further. The net interest margin falls to 1.1% in Germany and 3.62% in the US. With higher interbank lending rates, it is less profitable for German banks to lend cross-border to US firms so that the share of banks that engage in cross-border lending goes down to 62.04% and the volume of cross-border loans falls from =C42 bn to C=39.8 bn. Overall total lending to US firms goes up to C=536 bn. Most of this increase is funded through the interbank market, leading to a total of C=188.27 bn in net borrowing by German banks from US banks. Comparing the composition of German banks’ foreign lending, funding sources and gross capital flows The proposed model makes predictions regarding the composition of German bank loans in the US, the funding sources for these loans as well as the volume of capital flows as a function of frictions to foreign bank operations. The three charts in figure 6 further illustrate this. The top chart shows the composition of German banks’ assets in the US when (i) only interbank lending is possible (bar on the left), (ii) in the baseline economy (bar in the middle), and (iii) in the scenario with a 10% lower fixed cost 𝑓𝐹 (bar on the right). While German bank loans in the US are entirely loans to US banks under scenario (i), they 17

consist of loans to US non-banking firms by parent banks and local affiliates in the other two scenarios. This highlights that lending to non-banking firms and interbank lending can be seen as substitutes. As German banks are able to lend directly to US firms, they switch away from lending to US banks.33 A reduction in the barriers to establishing affiliates in the US increases local lending (by affiliates) while cross-border lending (by the parent) falls. The next chart shows how German banks’ assets in the US are funded. In the scenario with interbank lending only, lending to US banks is funded entirely through German deposits. In the baseline economy, in contrast, funding comes from US depositors, German depositors as well as US banks. Interbank funding and funding through US deposits becomes more important relative to German depositor funding as the barriers to foreign bank entry decline.34 The bottom chart illustrates the composition of bank flows between Germany and the US. With interbank lending only, all capital that flows between the US and Germany goes through the interbank market. In the baseline economy, there are more types of cross-border bank flows and total flows are higher. Under the assumption that US affiliates of German banks use US deposits to fund local lending, this type of lending and borrowing does not create cross-border bank flows. However, German banks extend loans to US non-banking firms cross-border. At the same time, parent banks partly fund the operations of their local affiliates, which leads to so called intrabank flows between German banks and their US affiliates.35 As the costs of establishing affiliates in the US fall, cross-border lending flows go down, while the volumes of net intrabank and interbank flows increase.36 The effect of changes in the return on loans As a final exercise, we study a 25 basis point reduction in the return on loans in Germany across scenarios. Table 7 shows the effects. In all three scenarios, capital is reallocated from German to US firms, although to varying degrees. The effect on net interbank lending does not go in the same direction. In scenario (i), in which banks only borrow and lend on the international interbank market, a lower return on loans in Germany leads to higher net interbank flows from German banks to US banks. This is because an expansion in credit in the US can only be achieved through domestic banks, who consequently borrow from German banks. In scenarios (ii) and (iii) in contrast, the lower return on loans in Germany causes an increase in interbank borrowing by German banks. Because these are more efficient and can lend directly to US firms, they expand their business in the US and borrow more from US banks. 33Similar to figure 1, the share of German claims on US banks relative to total German claims in the US has fallen steadily since the late 1990s. The difference between German claims on US banks and US claims on German banks in the BIS data (a proxy for net interbank lending between Germany and the US) became smaller over the period from 1999 to 2010, with the difference turning negative for the first time at the end of 2004. Based on the model, these patterns are consistent with a reduction of the barriers to German bank operations in the US. 34The model does not quantitatively match well the amount of US deposits that German banks take because competition for deposits is not modeled in great detail. However, the model is useful for its qualitative implications regarding bank liabilities and funding composition. 35Inthechart,itisassumedthattheaffiliatesofGermanbanksdonotraisefundingontheinterbankmarket fromUSbanksbutthatthefundstofillthegapbetweenaffiliateloansanddepositsareprovidedbytheparent. 36Note that while the model pins down gross cross-border loans and gross local loans to US non-banking firms, only net interbank and intrabank flows are determined. 18

4.4 Interbank lending and lending to firms as substitutes The model highlights that interbank lending and lending to non-banking firms abroad can be seen as substitutes. If the barriers to foreign bank entry are prohibitively high, all bank flows are channeled through the international interbank market. As soon as banks are able to lend directly to firms, they will do so; lending by local affiliates and intra-bank lending replaces interbank lending and the direction of net flows on the international interbank market might actuallyreverse. Lowerbarrierstobankentryalsoimplythatbankscantakedepositsinforeign countries, which can reduce lending and borrowing on the interbank market. The data supports the model view that interbank lending and lending to the private sector are substitutes. In table 8, host countries in the German data are grouped into three similarly large bins according to their openness to foreign bank entry (high, middle, low openness) based on the three proxies employed before: the Chinn-Ito-Index, bureaucratic quality and property rights protection.37 The table compares the mean value of claims on banks relative to claims on the non-bank private sector (computed separately for each bank per country) across groups of countries. It shows that German banks hold on average significantly fewer claims on banks relative to claims on the non-bank private sector in countries with high levels of openness compared to countries with low levels of openness consistent with the model.38 5 Conclusions This paper proposes a parsimonious model of banking across borders to study the composition of international bank activities, which take the form of international interbank lending, cross-border lending to firms and lending through affiliates in foreign markets. The structural approach allows us to analyze both qualitatively and quantitatively how banks’ foreign asset positions, their funding sources as well as inter- and intrabank flows change in response to altering cross-border banking frictions. We emphasize two key implications presented in our analysis. First, different types of bank activities are interconnected and are often substitutes (for example, lending to non-banking firms and lending to banks). They thus have to be studied jointly and not in isolation. This is an insight that can be useful especially for future empirical work.39 Second, policies that affect cross-border frictions alter the composition of banks’ foreign activities and, thereby, how much credit is supplied by whom and how it is funded. The model in this paper is static and is reduced to the minimal ingredients, but it alludes to key issues that should be studied in richer models in the future. As the literature has shown, different types of international bank flows respond differently to foreign and domestic shocks. For example, 37Bins do not contain exactly the same number of countries since we allocated all countries with the same level of openness to the same bin. 38A similar pattern is also observed in the BIS data. The share of claims on banks compared to claims on the non-bank private sector held by BIS reporting countries is higher in less developed countries. 39Empirical studies tend to compare the responses of different types of international bank flows to shocks without taking into account that these flows are determined simultaneously. 19

interbank lending appears to be less stable than cross-border or affiliate lending.40 The model thus indicates that policies that affect entry barriers have consequences for domestic financial stability and ultimately, real economic activity. More research is needed to better understand banks’ global business models and margins of adjustments to balance sheet shocks. At the same time, global banks’ profit maximizing behavior and margins of adjustments should be integrated in international macro models to better understand the consequences of banking sector integration for financial stability and welfare. This paper delivers one building block to achieve this. 40For references, see footnotes 5 and 6. 20

References Aviat, Antonin and Nicolas Coeurdacier, “The geography of trade in goods and asset holdings,” Journal of International Economics, March 2007, 71 (1), 22–51. Beck, Thorsten, Asli Demirguc-Kunt, and Ross Levine, “A New Database on the Structure and Development of the Financial Sector,” World Bank Economic Review, 2000, 14 (3), 597–605. Boissay, Frederic, “Financial imbalances and financial fragility,” Working Paper Series 1317, European Central Bank 2011. Bremus, Franziska, Claudia Buch, Katheryn Russ, and Monika Schnitzer, “Big Banks and Macroeconomic Outcomes: Theory and Cross-Country Evidence of Granularity,” NBER Working Papers 19093, National Bureau of Economic Research, Inc May 2013. Bruno, Valentina and Hyun Song Shin, “Cross-Border Banking and Global Liquidity,” Review of Economic Studies, 2015, 82 (2), 535–564. Buch, Claudia M., “Information or Regulation: What Drives the International Activities of Commercial Banks,” Journal of Money, Credit and Banking, 2003, 35 (6), 851–869. , Catherine Tahmee Koch, and Michael Koetter, “Size, productivity and international banking,” Journal of International Economics, 2011, 85 (2), 329–334. Cetorelli, Nicola and Linda S. Goldberg, “Global Banks and International Shock Transmission: Evidence from the Crisis,” IMF Economic Review, 2011, 59 (1), 41–76. Chinn, Menzie, “A New Measure of Financial Openness,” Journal of Comparative Policy Analysis, 2008, 10. Chodorow-Reich, Gabriel, “The Employment Effects of Credit Market Disruptions: FirmlevelEvidencefromthe2008-09FinancialCrisis,” The Quarterly Journal of Economics, 2014, 129 (1), 1–59. Cihak, Martin, Aslı Demirgu¨¸c-Kunt, Erik Feyen, and Ross Levine, “Benchmarking financial development around the world,” World Bank Policy Research Working Paper, 2012, 6175, 1–58. Corbae, Dean and Pablo D’Erasmo, “A Quantitative Model of Banking Industry Dynamics,” 2010 Meeting Papers 268, Society for Economic Dynamics 2010. and , “Capital requirements in a quantitative model of banking industry dynamics,” Working Papers 14-13, Federal Reserve Bank of Philadelphia 2014. Craig, Ben R. and Goetz von Peter, “Interbank tiering and money center banks,” Journal of Financial Intermediation, 2014, 23 (3), 322–347. 21

de Blas, Beatriz and Katheryn Niles Russ, “FDI in the Banking Sector,” NBER Working Papers 16029, National Bureau of Economic Research 2010. and , “All banks great, small, and global: Loan pricing and foreign competition,” International Review of Economics & Finance, 2013, 26 (C), 4–24. de Haas, Ralph and Iman van Lelyveld, “Foreign banks and credit stability in Central and Eastern Europe. A panel data analysis,” Journal of Banking & Finance, July 2006, 30 (7), 1927–1952. and , “Internal capital markets and lending by multinational bank subsidiaries,” Journal of Financial Intermediation, January 2010, 19 (1), 1–25. and , “Multinational Banks and the Global Financial Crisis: Weathering the Perfect Storm?,” Journal of Money, Credit and Banking, 02 2014, 46 (s1), 333–364. and Neeltje van Horen, “Running for the Exit? International Bank Lending During a Financial Crisis,” Review of Financial Studies, 2013, 26 (1), 244–285. de Sousa, Jose, Thierry Mayer, and Soledad Zignago, “Market access in global and regional trade,” Regional Science and Urban Economics, 2012, 42 (6), 1037–1052. Degryse, Hans and Steven Ongena, “Distance, Lending Relationships, and Competition,” The Journal of Finance, 2005, 60 (1), 231–266. Duewel, Cornelia, Rainer Frey, and Alexander Lipponer, “Cross-border bank lending, risk aversion and the financial crisis,” Discussion Paper Series 1: Economic Studies 2011,29, Deutsche Bundesbank, Research Centre 2011. Focarelli, Dario and Alberto Franco Pozzolo, “Where Do Banks Expand Abroad? An Empirical Analysis,” Journal of Business, November 2005, 78 (6), 2435–2464. Frey, Rainer and Cornelia Kerl, “Multinational banks in the crisis: Foreign affiliate lending as a mirror of funding pressure and competition on the internal capital market,” Journal of Banking & Finance, 2015, 50, 52–68. Greenwood, Jeremy, Juan M. Sanchez, and Cheng Wang, “Quantifying the impact of financial development on economic development,” Review of Economic Dynamics, 2013, 16 (1), 194 – 215. Special issue: Misallocation and Productivity. Head, Keith, Thierry Mayer, and John Ries, “The erosion of colonial trade linkages after independence,” Journal of International Economics, May 2010, 81 (1), 1–14. Janicki, Hubert and Edward Prescott, “Changes in the size distribution of U.S. banks: 1960-2005,” Economic Quarterly, 2006, (Fall), 291–316. Jeon, Bang Nam, Maria Pia Olivero, and Ji Wu, “Multinational banking and the international transmission of financial shocks: Evidence from foreign bank subsidiaries,” Journal of Banking & Finance, 2013, 37 (3), 952–972. 22

Kamil, Herman and Kulwant Rai, “TheGlobalCreditCrunchandForeignBanks’Lending to Emerging Markets: Why Did Latin America Fare Better?,” IMF Working Papers 10/102, International Monetary Fund April 2010. Khwaja, Asim Ijaz and Atif Rehman Mian, “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market,” American Economic Review, 2008, 98 (4), 1413–42. Kollmann, Robert, “Global Banks, Financial Shocks, and International Business Cycles: Evidence from an Estimated Model,” Journal of Money, Credit and Banking, 2013, 45 (s2), 159–195. , Zeno Enders, and Gernot J. Mueller, “Global banking and international business cycles,” European Economic Review, 2011, 55 (3), 407 – 426. Special Issue: Advances in International Macroeconomics: Lessons from the Crisis. McCauley, Robert, Patrick McGuire, and Goetz von Peter, “After the global financial crisis: From international to multinational banking?,” Journal of Economics and Business, 2012, 64 (1), 7 – 23. Milesi-Ferretti, Gian-Maria and Cedric Tille, “The great retrenchment: international capital flows during the global financial crisis,” Economic Policy, 04 2011, 26 (66), 289–346. Niepmann, Friederike, “Banking across borders with heterogeneous banks,” Staff Reports 609, Federal Reserve Bank of New York 2013. , “Banking across borders,” Journal of International Economics, 2015, 96 (2), 244 – 265. Olivero, Maria Pia, “Market power in banking, countercyclical margins and the international transmission of business cycles,” Journal of International Economics, March 2010, 80 (2), 292–301. Ongena, Steven, Jose Luis Peydro, and Neeltje van Horen, “Shocks Abroad, Pain at Home? Bank-Firm Level Evidence on the International Transmission of Financial Shocks,” DNB Working Papers 385, Netherlands Central Bank, Research Department July 2013. Popov, Alexander and Gregory F. Udell, “Cross-border banking, credit access, and the financial crisis,” Journal of International Economics, 2012, 87 (1), 147–161. Reinhardt, Dennis and Steven J. Riddiough, “The two faces of cross-border banking flows: an investigation into the links between global risk, arms-length funding and internal capital markets,” Working Paper 498, Bank of England April 2014. Rosengren, Eric S. and Joe Peek, “Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States,” American Economic Review, March 2000, 90 (1), 30–45. 23

Schnabl, Philipp, “The International Transmission of Bank Liquidity Shocks: Evidence from an Emerging Market,” The Journal of Finance, 2012, 67 (3), 897–932. Stigum, Marcia, The Money Market, Dow Jones-Irwin, Homewood, IL, 1990. 24

A Data Appendix Data from the Deutsche Bundesbank: The main data source for the empirical analysis in this paper are the External Positions Reports and Balance Sheet Statistics that German banks file with the Deutsche Bundesbank on a monthly basis. These reports contain information the positions of parent banks, their branches and subsidiaries by country and sector. Banklevel data is confidential but available for research purposes on the premises of the Deutsche Bundesbank. All data used in this paper is for 2005. Our sample excludes foreign owned banks but comprises all domestically-owned banks with a German banking license. Banks fall into one of the following categories: commercial banks, Landesbanken, savings banks, regional institutions of credit cooperatives, credit cooperatives, building credit societies, savings and loan associations, and banks with special functions. We work with claims on the foreign nonbank private sector and claims on the foreign banking sector, excluding claims on foreign central banks. Claims represent accounts receivable and do not include securities holdings. When consolidating the parent bank and its affiliates, intragroup exposures are netted out by declaring liabilities that affiliates have on the German banking sector as representing parent bank funding. Net interest margins: Information on net interest margins by country for 1998 and 2005 is from the World Bank’s Financial Development and Structure Dataset. Descriptions of this dataset can be found in Beck et al. (2000) and Cihak et al. (2012). The net interest margin for the Netherlands Antilles is proxied by the value for Cura¸cao. The net interest margin for Serbia and Montenegro is the average between the net interest margins of the two countries. Chinn-Ito openness index: Capital account openness is proxied by the Chinn & Ito Index documented in Chinn (2008). It is a de jure measure of openness, which increases with greater capital account openness of a country. Bureaucratic quality: Bureaucratic quality is from the International Country Risk Guide provided by the PRS Group.41 A high value of the index means that obstacles to conduct business stemming from bureaucracy are low, as bureaucracy “has the strength and expertise to govern without drastic changes in policy or interruptions in government services”. Property rights protection: Information on property rights protection comes from the Heritage Foundation.42 The index increases with greater protection of private property by a country’s laws and the enforcement of those laws. Other country-level variables: GDP in current U.S. dollars is from the World Development Indicators. Distance from Germany to foreign countries comes from a dataset provided by CEPII (see de Sousa et al. (2012) and Head et al. (2010)). 41See http://www.prsgroup.com/about-us/our-two-methodologies/icrg 42See http://www.heritage.org/index/property-rights. 25

B Proof of Proposition 1 Proof. When 𝑅𝐼 = max{𝑅 ,𝑅 ,...,𝑅 }, RHS of equation 17 is equal to zero. If 𝑅𝐼 = 0, then 1 2 𝑁 𝑅𝐻𝑆 > ∑︀𝑁 𝐾 because monitoring is assumed to be beneficial, which implies 𝐾 𝑎′𝑅 > 𝐾 . 𝑖=1 𝑖 𝑖 𝑖 𝑖 𝑖 RHS of equation 17 is strictly decreasing in 𝑅𝐼 on the interval 𝑅𝐼 ∈ [0,max{𝑅 ,𝑅 ,...,𝑅 }]. 1 2 𝑁 To see this, note that: 𝜕𝑅𝐻𝑆 ∫︁ 𝑎𝑖 𝜕𝑧 𝑖𝑖 = 𝑀 𝑔 (𝑎 )𝑑𝑎 + (B.1) 𝜕𝑅𝐼 𝑖 𝜕𝑅𝐼 𝑖 𝑖 𝑖 𝑎 𝑖 (︃ )︃ + ∑︁ 𝑁 𝑀 ∫︁ 𝑎˜𝐹 𝑖𝑗 𝜕𝑧 𝑖 𝑋 𝑗 𝑛 (𝑎˜ )𝑑𝑎˜ + 𝜕𝑎˜𝐹 𝑖𝑗 𝑧𝑋(𝑎˜𝐹)− 𝜕𝑎˜𝑋 𝑖𝑗 𝑧𝑋(𝑎˜𝑋) 𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑖𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑎˜𝑋 𝑗=1,𝑗̸=𝑖 𝑖𝑗 (︃ )︃ ∑︁ 𝑁 ∫︁ 𝑎˜𝑖𝑗 𝜕𝑧𝐹 𝜕𝑎˜𝐹 + 𝑀 𝑖𝑗 𝑛 (𝑎˜ )𝑑𝑎˜ − 𝑖𝑗 𝑧𝐹(𝑎˜𝐹) 𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑎˜𝐹 𝑗=1,𝑗̸=𝑖 𝑖𝑗 (︃ )︃ ∫︁ 𝑎𝑖 𝜕𝑧 𝑖𝑖 ∑︁ 𝑁 ∫︁ 𝑎˜𝐹 𝑖𝑗 𝜕𝑧 𝑖 𝑋 𝑗 ∫︁ 𝑎˜𝑖𝑗 𝜕𝑧 𝑖 𝐹 𝑗 = 𝑀 𝑔 (𝑎 )𝑑𝑎 + 𝑀 𝑛 (𝑎˜ )𝑑𝑎˜ + 𝑛 (𝑎˜ )𝑑𝑎˜ + 𝑖 𝜕𝑅𝐼 𝑖 𝑖 𝑖 𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑎 𝑎˜𝑋 𝑎˜𝐹 𝑖 𝑗=1,𝑗̸=𝑖 𝑖𝑗 𝑖𝑗 ⏟ ⏞ ⏟ ⏞ <0 ≤0 (︃ )︃ + ∑︁ 𝑁 𝑀 𝜕𝑎˜𝐹 𝑖𝑗 (︀ 𝑧𝑋(𝑎˜𝐹)−𝑧𝐹(𝑎˜𝐹) )︀ − 𝜕𝑎˜𝑋 𝑖𝑗 𝑧𝑋(𝑎˜𝑋). 𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑖𝑗 𝜕𝑅𝐼 𝑖𝑗 𝑗=1,𝑗̸=𝑖 ⏟ ⏞ ≥0 𝑧𝑋(𝑎˜ ) < 𝑧𝐹(𝑎˜ ) because 𝛿𝑋 < 𝛿𝐹 ⇒ 𝑧𝑋(𝑎˜𝐹) − 𝑧𝐹(𝑎˜𝐹) < 0. Under the assumption that 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑓𝐹 > 𝑓𝑋 − 𝐾𝑖 max{𝑅 ,𝑅 ,...,𝑅 }, 𝜕𝑎˜𝐹 𝑖𝑗 ≥ 0. This implies that 𝜕𝑅𝐻𝑆 < 0. With RHS of 𝑖𝑗 𝑖𝑗 𝑀𝑖 1 2 𝑁 𝜕𝑅𝐼 𝜕𝑅𝐼 equation 17 being strictly decreasing in 𝑅𝐼, it follows that RHS of equation 17 cuts LHS of equation 17 once from above on the intervale 𝑅𝐼 ∈]0,max{𝑅 ,𝑅 ,...,𝑅 }]. 1 2 𝑁 26

scitsitats yrammuS :1 elbaT ataD .sisylana noisserger eht ni desu seirtnuoc ngierof dna sknab namreG fo elpmas eht rof scitsitats yrammus stneserp elbat sihT dna knabsednuB ehcstueD eht htiw elfi sknab namreG taht stroper erusopxe ngierof dna scitsitats teehs ecnalab ylhtnom morf emoc tonnac selbairav level-knab fo amixaM dna aminiM .detats esiwrehto sselnu 000,1=C ni detroper si ataD .shtnom 21 revo degareva era yrammuS .setailffia evitcepser rieht dna ’sknab neewteb snoitisop edulcxe snoitisop knabretni teN .ytilaitnedfinoc ot eud detroper eb .snoitavresbo orez-non rof era selbairav cfiiceps-yrtnuoc-knab rof scitsitats .veD .dtS naeM .sbO SELBAIRAV CIFICEPS-KNAB 781,640,5 348,450,1 599,1 rotces etavirp knab-non citsemod no smialC 192,873,6 129,741- 591,1 sknab no smialc teN .veD .dtS naeM sbO SELBAIRAV CIFICEPS-YRTNUOC DNA -KNAB 839,75 260,3 702,42 rotces etavirp knab-non ngierof no smialc redrob-ssorc knab tneraP 124,387,4 631,114 395,1 rotces etavirp knab-non ngierof no smialc )redrob-ssorc+lacol( etailffiA 365,973,3 426,372 395,1 rotces etavirp knab-non ngierof no smialc etailffia lacoL 084,87 474,1 508,6 rotces etavirp knab-non no smialc ot sknab no smialc fo oitaR xaM niM .veD .dtS naeM sbO SELBAIRAV CIFICEPS-YRTNUOC 861.0 200.0 130.0 940.0 751 nigram tseretni teN 057.428,81 425.371 389.675,3 679.756,5 651 )5002( ogangiZ dna reyaM ni sa erusaem ecnatsiD 31+E062.1 80+E016.4 21+E031.1 11+E057.2 351 )000,1 $ tnerruc ni( PDG 235.2 218.1- 116.1 585.0 261 xedni ssennepo otI-nnihC 4 0 231.1 121.2 731 ytilauq citarcuaeruB 09 01 484.32 192.64 151 noitcetorp sthgir ytreporP 27

Table 2: German banks’ frictions of operating abroad by country This table reports the values of 𝛿𝑋, 𝛿𝐹, 𝑓𝑋 and 𝑓𝐹 −𝑅𝐼𝑑 for various host countries 𝑖 obtained from applying 𝑖 𝑖 𝑖 𝑖 𝑖 the strategy described in section 3 to bank-level data from the Deutsche Bundesbank for the year 2005. For better illustration, all parameter values were divided by the respective values for the United States. 𝛿𝑋 [𝛿𝐹] is 𝑖 𝑖 aninversemeasureoftheefficiencylossthatabankerincurswhenlendingcross-border[throughalocalaffiliate] tofirmsincountry𝑖. 𝑓𝑋 representsthefixedcostofcross-borderlending; 𝑓𝐹−𝑅𝐼𝑑 standsforthefixedcostof 𝑖 𝑖 𝑖 establishinganaffiliatenetofthebenefitsfromraisingdepositsincountry𝑖. Notallvaluesthatwerecomputed can be shown due to confidentiality. Country 𝛿𝑋 𝑓𝑋 𝛿𝐹 𝑓𝐹 −𝑅𝐼𝑑 Country𝑖 𝛿𝑋 𝑓𝑋 𝛿𝐹 𝑓𝐹 −𝑅𝐼𝑑 Luxembourg 4.02602 0.350 0.3321 0.00011 Korea,Rep. 0.01979 1.188 0.0060 Netherlands 3.13475 0.117 0.4332 0.00895 Mauritius 0.01794 1.745 0.0015 Poland 2.57348 2.498 0.0288 0.00128 Ukraine 0.01713 2.395 0.0012 France 2.33543 0.442 0.3230 0.00162 Namibia 0.01652 4.374 Uruguay 2.26136 2.287 Cameroon 0.01608 24.948 Switzerland 1.61588 0.388 0.3097 0.00007 BruneiDarussalam 0.01532 12.082 Denmark 1.56711 0.810 0.0336 Aruba 0.01497 3.431 Norway 1.26851 0.858 Colombia 0.01489 1.952 UnitedStates 1 1 1 1 Morocco 0.01327 2.318 Austria 0.87210 0.885 0.1263 0.00033 Mali 0.01294 3.677 UnitedKingdom 0.76153 1.075 0.9515 0.02374 Ethiopia 0.01260 2.324 Belgium 0.73548 0.481 0.1141 0.00048 Brazil 0.01235 3.318 0.0054 0.01251 Malta 0.70782 1.614 Tanzania 0.01184 4.776 Belize 0.49601 4228.198 Indonesia 0.01148 3.075 0.0049 Cyprus 0.46795 0.537 Jamaica 0.01147 4.974 Panama 0.46341 1.136 0.3888 UnitedArabEmirates 0.01129 1.408 0.0136 Sweden 0.38843 0.448 0.0610 SriLanka 0.01073 3.058 0.0005 India 0.32409 2.237 0.0305 Mongolia 0.01013 11.670 Japan 0.28145 0.686 0.4200 0.00685 Pakistan 0.01007 3.037 0.0009 Iceland 0.27593 0.099 Benin 0.00996 0.161 Spain 0.27519 1.227 0.0879 0.04086 Djibouti 0.00965 3.318 Canada 0.25783 0.905 0.0470 Philippines 0.00929 3.112 0.0026 Gabon 0.22419 1.959 Albania 0.00906 8.291 CzechRepublic 0.19853 1.603 0.0186 0.00036 Ecuador 0.00892 3.774 Greece 0.19764 2.605 0.0182 0.01046 ElSalvador 0.00882 3.831 Cuba 0.18892 0.989 Guatemala 0.00857 3.102 Slovenia 0.14518 1.708 SerbiaandMontenegrob 0.00825 4.064 0.0010 Lebanon 0.14262 1.760 Romania 0.00814 2.737 0.0032 Macedonia,FYR 0.14149 3.310 Bolivia 0.00800 3.081 Nicaragua 0.14111 4.271 Azerbaijan 0.00785 3.281 Ireland 0.13535 0.483 0.2083 0.00029 Vietnam 0.00705 1.402 0.0003 Argentina 0.13302 1.204 0.4370 Georgia 0.00632 8.049 SaudiArabia 0.12559 2.303 Paraguay 0.00622 5.901 Myanmar 0.12456 0.589 Mozambique 0.00587 4.411 Chile 0.12099 2.621 0.3575 BurkinaFaso 0.00532 3.361 Hungary 0.09467 2.895 0.0158 0.00853 Uganda 0.00522 8.344 Togo 0.09246 1.433 Malawi 0.00518 7.940 Latvia 0.09143 1.981 0.0105 Coted’Ivoire 0.00505 3.762 Algeria 0.08428 2.582 Libya 0.00469 0.063 Turkey 0.08181 0.577 0.0535 DominicanRepublic 0.00438 7.236 RussianFederation 0.08166 2.285 0.0214 0.08666 Uzbekistan 0.00387 2.282 Finland 0.07953 0.359 0.0577 SyrianArabRepublic 0.00371 2.674 Egypt,ArabRep. 0.07877 1.074 Kenya 0.00366 6.820 China 0.07633 1.256 0.0138 0.01059 Yemen,Rep. 0.00341 7.184 Angola 0.07436 16.130 Zambia 0.00330 7.911 Mexico 0.06837 0.699 1.8875 Turkmenistan 0.00319 1.292 Armenia 0.06643 13.070 PapuaNewGuinea 0.00318 5.808 Italy 0.06570 1.483 0.1033 0.14500 Sudan 0.00289 7.019 Australia 0.06497 1.411 0.0557 Grenada 0.00278 167.594 Kazakhstan 0.06462 3.050 Guinea 0.00266 13.331 Neth. Antillesa 0.06339 4.761 0.0148 Nepal 0.00256 2.784 Qatar 0.06290 1.516 LaoPDR 0.00248 2.021 HongKongSAR,China 0.05817 1.406 0.0022 0.00003 Oman 0.00247 2.394 Estonia 0.05286 0.880 0.0030 Mauritania 0.00244 4.553 Malaysia 0.05175 1.322 0.0044 0.00373 Botswana 0.00223 3.830 Thailand 0.05121 1.944 0.0030 TrinidadandTobago 0.00191 3.228 Jordan 0.05042 2.085 Gambia,The 0.00186 6.366 SouthAfrica 0.05030 1.768 0.0035 Lesotho 0.00154 4.106 SlovakRepublic 0.04944 1.628 0.0032 Chad 0.00117 4.715 Bulgaria 0.04792 3.107 0.0020 SierraLeone 0.00116 9.079 BosniaandHerzegovina 0.04576 2.879 0.0007 Senegal 0.00086 3.958 Tunisia 0.04429 1.847 Peru 0.00076 3.309 CostaRica 0.04042 4.764 Cambodia 0.00062 3.019 Israel 0.03929 0.623 Iraq 0.00055 6.416 Croatia 0.03754 1.562 0.0054 Tajikistan 0.00050 9.191 Nigeria 0.03703 3.535 Niger 0.00046 8.089 NewZealand 0.03577 1.041 0.0285 Madagascar 0.00042 5.662 Singapore 0.03559 1.353 0.0600 0.00760 Afghanistan 0.00034 6.154 Venezuela,RB 0.03397 4.731 Haiti 0.00026 14.466 Lithuania 0.03294 2.415 0.0208 Moldova 0.00025 8.120 Bahrain 0.03141 1.045 CentralAfricanRepublic 0.00020 5.748 Bangladesh 0.02892 3.449 KyrgyzRepublic 0.00018 3.610 Belarus 0.02728 1.427 Rwanda 0.00009 70.442 Kuwait 0.02527 0.374 Guyana 0.00007 5.781 Honduras 0.02348 9.790 Swaziland 0.00006 4.123 Ghana 0.02335 10.873 Burundi 0.00003 13.557 Portugal 0.02036 1.457 0.0301 0.00319 28

snoitcirf gnidnel redrob-ssorc rof stluser noissergeR :3 elbaT redrob-ssorc dnel taht sreknab fo ssol ycneicffie eht ylesrevni stcefler hcihw( 𝑋𝛿 gnisserger morf stluser stroper elbat eht fo trap pot ehT 𝑖 yrotanalpxe eht no sliateD .scitsiretcarahc 𝑖-yrtnuoc no )𝑖 yrtnuoc ot gnidnel redrob-ssorc fo tsoc dexfi eht( 𝑋𝑓 dna )𝑖 yrtnuoc ot 𝑖 laitrapimesfostluserswohselbatehtfotrapmottobehT .sesehtnerapnierasrorredradnatstsuboR .xidneppaatadehtnieraselbairav * ,50.0<p ** ,10.0<p *** .noisserger eht morf elbairav evitcepser a gnittimo morf 2𝑅 ni noitcuder eht erutpac hcihw ,snoitalerroc .1.0<p )6( )5( )4( )3( )2( )1( 𝑋𝑓 𝑋𝑓 𝑋𝑓 𝑋𝛿 𝑋𝛿 𝑋𝛿 SELBAIRAV *223000.0 ***923000.0 992000.0 **50-e09.8- **401000.0- **301000.0ecnatsiDnl )371000.0( )521000.0( )891000.0( )50-e81.4( )50-e82.4( )50-e90.4( ***428000.0- ***665000.0- ***649000.0- 50-e47.1 50-e23.2 ***50-e51.3 PDGnl )052000.0( )421000.0( )542000.0( )50-e72.1( )50-e16.1( )50-e90.1( *519000.0- **50-e65.3 xedni ssennepo otI-nnihC )405000.0( )50-e94.1( *544000.0- ***50-e00.7 ytilauq ycarcuaeruB )532000.0( )50-e42.2( 50-e60.3- ***60-e66.4 tnemecrofne sthgir ytreporP )50-e40.2( )60-e93.1( ***1220.0 ***7410.0 ***8420.0 622000.0 772000.0 291000.0 tnatsnoC )11700.0( )68200.0( )70700.0( )592000.0( )053000.0( )413000.0( 931 421 341 931 321 341 snoitavresbO 330.0 413.0 760.0 082.0 232.0 522.0 derauqs-R detsujdA STNEICIFFEOC NOITALERROC LAITRAPIMES DERAUQS 9000.0 5210.0 7000.0 ***2850.0 ***970.0 ***5080.0 ecnatsiDnl *8320.0 ***7411.0 **8430.0 9900.0 9210.0 ***4830.0 PDGnl *3020.0 **7030.0 xedni ssennepo otI-nnihC *8020.0 **2630.0 ytilauq ycarcuaeruB 8300.0 ***680.0 tnemecrofne sthgir ytreporP 29

snoitcirf gnidnel etailffia rof stluser noissergeR :4 elbaT setailffia aiv dnel taht sreknab fo ssol ycneicffie eht ylesrevni stcefler hcihw( 𝐹𝛿 gnisserger morf stluser stroper elbat eht fo trap pot ehT 𝑖 no )ereht stisoped gnisiar morf stfieneb eht fo ten 𝑖 yrtnuoc ni etailffia na gnihsilbatse fo tsoc dexfi eht( 𝑑𝐼𝑅− 𝐹𝑓 dna )𝑖 yrtnuoc ot 𝑖𝑏 𝑖 .sesehtnerap ni era srorre dradnats tsuboR .xidneppa atad eht ni era selbairav yrotanalpxe eht no sliateD .scitsiretcarahc 𝑖-yrtnuoc evitcepser a gnittimo morf 2𝑅 ni noitcuder eht erutpac hcihw ,snoitalerroc laitrapimes fo stluser swohs elbat eht fo trap mottob ehT .1.0<p * ,50.0<p ** ,10.0<p *** .noisserger eht morf elbairav )6( )5( )4( )3( )2( )1( 𝑑𝐼𝑅− 𝐹𝑓 𝑑𝐼𝑅− 𝐹𝑓 𝑑𝐼𝑅− 𝐹𝑓 𝐹𝛿 𝐹𝛿 𝐹𝛿 SELBAIRAV 9.584- 0.964- 4.291 56100.0- 04100.0- 101000.0ecnatsiDnl )380,1( )405,1( )190,1( )81200.0( )73200.0( )53200.0( **517,2 *458,2 *180,2 57100.0 44100.0 *52300.0 PDGnl )023,1( )835,1( )861,1( )10200.0( )52200.0( )77100.0( 37.49- **28300.0 xedni ssennepo otI-nnihC )150,1( )45100.0( *588,3- ***87500.0 ytilauq ycarcuaeruB )320,2( )09100.0( *1.421- 441000.0 tnemecrofne sthgir ytreporP )59.16( )401000.0( *053,45- *903,45- *243,05- 8030.0- 6230.0- *1970.0tnatsnoC )957,72( )175,82( )178,82( )8250.0( )4650.0( )2540.0( 84 64 74 84 74 74 snoitavresbO 670.0 290.0 110.0 440.0 680.0 501.0 derauqs-R detsujdA STNEICIFFEOC NOITALERROC LAITRAPIMES DERAUQS 6200.0 4200.0 3000.0 4510.0 1110.0 1000.0 ecnatsiDnl **611.0 **9801.0 *7470.0 1220.0 1310.0 **1180.0 PDGnl 1000.0 **7290.0 xedni ssennepo otI-nnihC *2970.0 **3680.0 ytilauq ycarcuaeruB *2650.0 7630.0 tnemecrofne sthgir ytreporP 30

ledom eht etalumis ot desu sretemaraP :5 elbaT .noitpircseddeliatedarof1.4noitceseeS .esacynamreG-SUehtotledomehtetarbilacotdesuseulavretemarapehtstneserpelbatsihT .atad level-knab eht fo ytilaitnedfinoc eht evreserp ot detroper ton era 𝑎˜ dna 𝑎 , 𝑎 , 𝑎 , 𝑎 rof seulaV 𝑅𝐸𝐺,𝑆𝑈 𝑆𝑈 𝑆𝑈 𝑅𝐸𝐺 𝑅𝐸𝐺 eulaV ecruoS 862.0 atad level-knab morf detamitse 𝑎 rof oteraP fo retemarap epahs 𝑅𝐸𝐺 ykratua ni nigram tni ten 8991 hctam ot 𝑎 𝑅𝐸𝐺 atad morf 𝑎 𝑅𝐸𝐺 2251.0 atad level-knab morf detamitse 𝑎˜ rof oteraP fo retemarap epahs 𝑅𝐸𝐺,𝑆𝑈 26 𝑋𝑎˜ ffotuc devresbo nevig sknab redrob-ssorc fo erahs hctam ot 𝑎˜ 𝑆𝑈 𝑅𝐸𝐺,𝑆𝑈 atad morf 𝑎˜ 𝑅𝐸𝐺,𝑆𝑈 862.0 ynamreG rof sa emas 𝑎 rof oteraP fo retemarap epahs 𝑆𝑈 ykratua ni nigram tseretni ten 8991 hctam ot 𝑎 𝑆𝑈 ynamreG rof sa emas 𝑎 𝑆𝑈 5991 atad morf 𝑀 sknab namreG fo ssam 𝑅𝐸𝐺 0056 )6002( ttocserP dna ikcinaJ 𝑀 sknab SU fo ssam 𝑆𝑈 nb 69.1=C atad teehs ecnalab namreG morf 𝐾 stisoped namreG 𝑅𝐸𝐺 nb 2.93=C stisoped namreG × 2 𝐾 stisoped SU 𝑆𝑈 1.1 srohtua yb tes 𝑅 snaol namreG no nruter ssorg 𝑅𝐸𝐺 2520.0+ 𝑅 5002 ni snigram tni ten ni ecnereffid hctam ot 𝑅 snaol SU no nruter ssorg 𝑅𝐸𝐺 𝑆𝑈 4366000.0 3 noitces morf 𝑋𝑓 gnidnel redrob-ssorc fo tsoc dexfi 𝑖 4538000.0 3 noitces morf 𝑋𝛿 57021.0 3 noitces morf 𝐹𝛿 6+e533.8 smrfi SU ot sknab namreG yb snaol hctam ot 𝐹𝑓 SU ni etailffia gnihsilbatse fo tsoc dexfi 𝑖 31

stnemom atad dna soiranecs tnereffid rof semoctuo ledoM :6 elbaT muirbiliuqe eht sedivorp )3( nmuloC .stnemom atad emos htiw rehtegot soiranecs tnereffid fo semoctuo ledom eht stneserp elbat sihT eht morf atad level-knab tfi ot nesohc erew sretemaraP .5 elbat ni deliated snoitpmussa retemarap eht rednu ledom eht fo seulav seulav muirbiliuqe swohs )1( nmuloC .)5( nmuloc ni nwohs era stnemom atad gnidnopserroC .5002 raey eht rof knabsednuB ehcstueD ngierof ot ylno tub yltcerid smrfi gniknab-non ngierof ot dnel tonnac sknab hcihw ni oiranecs eht stcefler )2( nmuloC .ykratua rednu rucni sknab namreG taht 𝐹𝑓 tsoc dexfi eht morf tpecxe )3( nmuloc sa seulav retemarap emas eht no desab si )4( nmuloC .sknab 𝑆𝑈 .)3( nmuloc ni naht rewol tnecrep 01 si hcihw ,SU eht ni etailffia na pu gninepo nehw ataD 𝐹𝑓 ni llaf %01 ymonoce enilesaB gnidnel knabretnI ykratuA retemaraP )5( )4( )3( )2( )1( nb 662=C nb 635=C nb 662=C nb 57.23=C 0 smrfi SU ot sknab namreG yb snaol latot nb 8.66=C nb 8.93=C nb 99.24=C 0 0 rotces .virp knab-non SU ot snaol redrob-ssorc nb991=C nb 694=C nb 322=C 0 0 rotces .virp knab-non SU ot setailffia yb snaol %52.16 %40.26 %86.26 0 0 .ces .virp knab-non SU ot snaol htiw sknab namreG fo erahs 6.0 %260.0 %620.0 0 0 .lffia htiw sknab namreG fo erahs nb 96-=C nb 72.881-=C nb 58.67-=C nb 57.23=C 0 sknab SU ot gnidnel knabretni ten rt 1.2=C rt 16.1=C rt 77.1=C rt 39.1=C rt 69.1=C smrfi namreG ot gnidnel latot rt 62.4=C rt 11.4=C rt 59.3=C rt 29.3=C smrfi SU ot gnidnel latot rt 73.2=C rt 51.2=C rt 40.2=C rt 39.1=C rt 69.1=C smrfi ot gnidnel latot ’sknab namreG rt 37.3=C rt 48.3=C rt 59.3=C rt 29.3=C smrfi ot gnidnel latot ’sknab SU nb 46.743=C nb 28.881=C nb 57.23=C 0 wofl latipac ten - %09.8 %97.8 %86.8 %66.8 etar gnidnel knabretni namreG - %09.8 %97.8 %86.8 %17.8 etar gnidnel knabretni SU %59.0 %1.1 %12.1 %23.1 %43.1 nigram tseretni ten namreG %74.3 %26.3 %37.3 %48.3 %18.3 nigram tseretni ten SU 32

snaol no nruter eht ni segnahc fo tceffE :7 elbaT rof stceffe eht swohs )1( nmuloC .ynamreG ni snaol no nruter eht ni noitcuder tniop-sisab-52 a fo stceffe eht sezirammus elbat sihT nmuloc ni detneserp stceffE .sknab ngierof ot ylno tub yltcerid smrfi gniknab-non ngierof ot dnel tonnac sknab hcihw ni oiranecs eht nehw stsoc rewol tnecrep 01 rucni sknab namreG hcihw ni oiranecs eht no desab si )3( nmuloC .ymonoce enilesab eht rof era )2( .noitarbilac enilesab eht ni naht SU eht ni etailffia na pu gninepo 𝐹𝑓 rewol %01 ymonoce enilesaB gnidnel knabretnI )3( )2( )1( )%40.63( nb 037=C )%63.37( nb 164=C )%0( 0 smrfi SU ot sknab namreG yb snaol latot )%3.1-( nb 82.93=C )%52.1-( nb 4.24=C )%0( 0 rotces .virp knab-non SU ot snaol redrob-ssorc )%30.93( nb 096=C )%57.78( rt 914=C )%0( 0 rotces .virp knab-non SU ot setailffia yb snaol )%14.61-( rt 53.1=C )%99.41-( rt 15.1=C )%48.7-( rt 87.1=C smrfi namreG ot gnidnel latot )%02.6( rt 35.4=C )%64.6( rt 73.4=C )%28.3 ( rt 1.4=C smrfi SU ot gnidnel latot )%13.3-( rt 80.2=C )%64.3-( rt 79.1=C )%48.7-( rt 87.1=C smrfi ot gnidnel latot ’sknab namreG )%19.1( rt8.3=C )%38.1( rt 19.3=C )%28.3( rt 1.4=C smrfi ot gnidnel latot ’sknab SU )%01.67( nb 216=C )%46.041( nb 454=C )%42.164( nb 481=C wofl latipac ten )%9.73-( nb 711-=C )%10.29-( nb 1.6-=C )%42.164( nb 481=C sknab SU ot gnidnel knabretni ten 33

Table 8: Sectoral composition of foreign bank positions by openness of destination country This table tests whether there are systematic differences in the composition of German banks’ positions across countries with different degrees of openness. We split countries into three similarly large groups (high, middle, low openness) according to their Chinn-Ito Index, bureaucratic quality and property rights protection. We then compare the log ratio of banks’ average claims on banks over their average claims on the non-bank private sector 𝑙𝑛(𝐶𝑙𝑎𝑖𝑚𝑠𝐵𝑣𝑃) across groups (high versus low openness). Calculations are based on data from the Deutsche Bundesbank for 2005. Standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. High Low Difference Chinn-Ito openness index Range of index [2.262;2.532] [-1.812;-1.131] No. of countries in group 54 60 Obs. (No. of banks in group) 5756 296 Mean of ln(ClaimsBvP) -0.386 0.516 -0.902*** St. Err. (0.039) (0.206) (0.178) Bureaucratic quality Range of index [3;4] [0;1.54] No. of countries in group 45 41 Obs. (No. of banks in group) 6005 158 Mean of ln(ClaimsBvP) -0.368 0.004 -0.372* St. Err. (0.039) (0.301) (0.247) Property rights protection Range of index [70;90] [10;30] No. of countries in group 37 73 Obs. (No. of banks in group) 5878 418 Mean of ln(ClaimsBvP) -0.322 -0.027 -0.294** St. Err. (0.04) (0.184) (0.156) 34

Figure 1: The sectoral composition of international bank claims held by BIS reporting countries, 1999-2013 The upper chart shows the evolution of total international bank claims of BIS reporting countries over time split by sector. The lower chart depicts the share of claims on the foreign non-bank private sector and the share of claims on the foreign banking sector in total international claims. The data source are the BIS Consolidated Statistics. Claims are on an immediate borrower basis and exclude local claims. The sectoral composition of international claims 25 20 15 10 5 0 1999q2 2000q3 2001q3 2002q3 2003q3 2004q3 2005q3 2006q3 2007q3 2008q3 2009q3 2010q3 2011q3 2012q3 2013q3 snoilliM unallocated non‐bank private sector public sector banking sector Shares in total international claims by sector 0.6 0.5 0.4 0.3 0.2 share of claims in the banking sector share of claims in the non‐bank private sector 0.1 0 35

Figure 2: The distribution of bank efficiency 𝑎 This graph shows the distribution of the bank-specific efficiency measure 𝑎. The smallest and largest 5th percentiles are not displayed. The calculation of 𝑎 is based on bank-level data (in C=1,000) from the Deutsche Bundesbank for 2005. 36

Figure 3: Bank efficiency 𝑎 as a predictor of cross-border lending This chart shows the distribution of rank correlation coefficients across host countries between the following two rankings: each bank’s rank based on the size of its cross-border lending to country 𝑖 and its rank based on the size of its domestic lending. The rank correlation coefficient for each country indicates how well a bank’s domestic lending predicts its cross-border lending to the respective country. The line that overlays the histogram depicts the corresponding kernel density estimate. Some countries are listed exemplarily across the distribution near the histogram bin into which their correlation coefficient falls. Data used in the calculations are from the Deutsche Bundesbank and for the year 2005. 37

Figure 4: The relationship between domestic and foreign lending volumes This chart plots banks’ cross-border lending to the non-bank private sector of country 𝑖 as a function of bank efficiency 𝑎 using an artificial dataset (original bank-level data cannot be 𝑏 shown due to confidentiality). Arrows depict the effect from multiplying 𝑎 with 𝜑 . 𝑏 𝑏𝑖 38

Figure 5: Bank efficiency 𝑎 as a predictor of net interbank lending This chart plots the estimated relationship between German banks’ net interbank claims and bank efficiency 𝑎 obtained from a simple OLS regression together with the 95th percent confi- 𝑏 dence interval. The 5th and the 95th percentiles of the bank efficiency distribution are excluded from the estimation. Net interbank claims are in C=1,000. Data used in the calculations are from the Deutsche Bundesbank and for the year 2005. 39

Figure 6: Counterfactual analysis This table illustrates the composition of German banks’ foreign assets, their funding sources, and bank flows between Germany and the US in the model. Each of the three charts in the figure shows outcomes for three different scenarios. The bars on the left capture scenario (i) in which German and US banks can only lend and borrow from each other on the international interbank market. The bars in the middle show outcomes for scenario(ii),whichcorrespondstothebaselineeconomythatiscalibratedbasedontheparameterspresentedin table5. ThebarsontherighthandsideassumethatthefixedcostofestablishinganaffiliateintheUS𝑓𝐹 is10 percentlowerthaninthebaselineeconomy(scenarioiii). ThetopchartshowsthecompositionofGermanbank loans in the US, which are split into loans to US banks, loans to US firms extended by German parent banks and loans to US firms extended by local affiliates of US parent banks. The chart in the middle summarizes how German banks’ US assets are funded. Funding comes either from German depositors, US depositors or US banks. The bottom chart illustratesthe composition ofinternational bankflowsbetweenGermanyand theUS. Banks flows consist of intrabank flows between German parent banks and their US affiliates, interbank flows between German and US banks as well as cross-border flows from parent banks to US non-banking firms. Composition of German banks' assets in the US 100% 90% 80% 70% 60% 50% 100% 83.8% 92.6% 40% 30% 20% 10% 16.2% 0% 7.4% (i) interbank market only (ii) baseline economy (iii) 10% decrease in entry barriers f^F net loans to US banks cross‐border loans to US firms loans by affill. to US firms Funding sources of German banks' assets in the US 100% 90% 80% 6 7 0 0 % % 71.0% 64.8% 50% 100.0% 40% 30% 20% 28.9% 35.1% 10% 0% 0.12% 0.14% (i) interbank market only (ii) baseline economy (iii) 10% decrease in entry barriers f^F net borrowing from US banks German deposits US deposits Bank flows between Germany and US in Euro bn 600 39.80 500 400 42.99 300 535.91 200 100 265.68 32.75 0 ‐76.85 ‐100 ‐188.27 ‐200 (i) interbank market only (ii) baseline economy (iii) 10% decrease in entry barriers f^F cross‐border loans by German banks to US non‐banking firms interbank lending from German banks to US banks Intrabank lendng from German parent banks to US affiliates 40