Running for the Exit: International Bank Lending

During a Financial Crisis

Ralph De Haas and Neeltje Van Horen*

This version: November 2011

Abstract

The global financial crisis has reignited the debate about the risks of financial globalization, in particular the instability of cross-border lending. We use loan-level data on lending by the largest international banks to their various countries of operation to examine why banks reduced lending more to some countries than to others during the 2008-09 crisis. The use of country- and bank-fixed effects allows us to disentangle credit supply and demand and to control for general bank characteristics. We find that banks continued to lend more to countries that are geographically close, where they are integrated into a network of domestic co-lenders, and where they have more lending experience. This suggests that banks’ ability to generate information about new clients and to use and exploit information about existing borrowers contributes to lending stability during a crisis.

JEL codes: F36, F42, F52, G15, G21, G28

Keywords: Cross-border credit, bank-borrower closeness, syndicated lending

* De Haas is with the European Bank for Reconstruction and Development (EBRD) and Van Horen with De Nederlandsche Bank (DNB). The authors thank Jeromin Zettelmeyer for insightful discussions; Stephan Knobloch, Deimante Morkunaite, and Carly Petracco for excellent research assistance; and three anonymous referees, the Editor, Itai Agur, Robert Paul Berben, Erik Berglöf, Martin Brown, Giacinta Cestone, Stijn Claessens, Stefan Gerlach, Martin Goetz, Charles Goodhart, Graciela Kaminsky, Ayhan Kose, Luc Laeven, Eduardo Levy Yeyati, Steven Ongena, Andreas Pick, Sergio Schmukler, and participants at the Bank of Finland/CEPR conference on ‘Banking in Emerging Economies’, the 13th DNB Annual Research Conference, the Basel Committee Research Task Force Transmission Channels Workshop, the 10th Annual Darden International Finance Conference, the 4th Swiss Winter Conference on Financial Intermediation, the 8th Workshop in International Economics and Finance (Lima), the 17th Dubrovnik Economic Conference, the 2011 ESCB Day-Ahead Conference on Financial Markets Research, the 26th Congress of the European Economic Association, and seminars at the EBRD, the Bank of Finland, DNB, Tilburg University, Brunel University, the Federal Reserve Banks of Boston and Chicago, Cass Business School, and the IMF for useful comments. An earlier version of this paper was circulated under the title “Running for the Exit: International Banks and Crisis Transmission”. The views expressed in this paper are those of the authors and not necessarily of the institutions with which they are affiliated. E-mail: dehaasr@ebrd.com and n.van.horen@dnb.nl.

1

1. Introduction

In the wake of the 2008–09 crisis, the virtues and vices of financial globalization are being

re-evaluated. Financial linkages between countries, in particular in the form of bank lending,

have been singled out as a key channel of international crisis transmission. The IMF and the

G-20 have identified the volatility of cross-border capital flows as a priority for the reform of

the global financial system (IMF, 2010). A pertinent question that is high on the policy and

academic agenda is why cross-border bank lending to some countries is relatively stable

whereas it is more volatile in other cases. The recent crisis, which originated in the U.S. sub-

prime market but spilled over to much of the developed and developing world, provides for

an ideal testing ground to answer this question.

After the collapse of Lehman Brothers in September 2008, syndicated cross-border lending

declined on average by 58 per cent compared to pre-crisis levels (Dealogic Loan Analytics).

Figure 1 illustrates, however, that the magnitude of this decline differed substantially between

countries. While this partly reflects variation in the adjustment of credit demand, we also

expect that international banks differentiated their credit-supply response across countries.

[INSERT FIGURE 1 HERE]

We hypothesize that banks reduce cross-border lending less during a crisis when they are

better able to collect, use, and exploit borrower information. Banks’ ability to do so will

depend on how ‘close’ they are to borrowers and in this paper we employ four key

dimensions of bank-borrower closeness. To the extent that closeness attenuates agency

problems it will reduce the need for credit rationing during a crisis. Closeness may also

provide banks with local market power and this may further stabilize their lending. We use

unique data on lending by international banks to corporate borrowers in a large number of

countries to put this hypothesis to the test.

The use of micro data on a comprehensive group of banks and countries – i.e. a dataset that is

both deep and broad – allows us to make a significant contribution to the literature on the role

of international banks during the global financial crisis. A number of recent papers use data at

the level of national banking systems to study the contraction in cross-border credit. Although

their broad country coverage comes at the cost of a high aggregation level, these

contributions provide interesting insights. They find that cross-border lending declined more

in the case of banking sectors that were vulnerable to US dollar funding shocks (Cetorelli and

2

Goldberg, 2011), that displayed a low average level of profitability or high average expected

default frequency (McGuire and Tarashev, 2008) and that had a poor average stock-market

performance (Herrmann and Mihaljek, 2010). Takáts (2010) shows that such supply factors

drove the reduction in lending to emerging markets more than demand. Finally, Hoggarth,

Mahadeva and Martin (2010) argue on the basis of BIS data and information from market

participants that the cross-border credit crunch was concentrated in ‘non-core’ markets. The

authors hypothesize that banks mainly reduced credit to countries where they knew borrowers

less well. However, their aggregated data does not allow them to test this conjecture.

A second set of papers on the recent crisis uses bank-level data for a specific country or

region. Such detailed data allow for a neat identification of the transmission of shocks. Aiyar

(2011) finds that the reduction in British banks’ foreign funding after the Lehman Brothers

collapse caused a contraction in their domestic lending. This contraction was sharper for

foreign-owned than for domestic banks. Rose and Wieladek (2011) complement this analysis

by showing that in particular nationalized foreign banks operating in the United Kingdom

reduced their lending. Finally, Popov and Udell (2010) illustrate how balance-sheet problems

of Western European banking groups led to a reduction in the credit supply of their Eastern

European subsidiaries.

Taken together these contributions provide convincing evidence that after the Lehman

Brothers bankruptcy international banks transmitted funding shocks across borders. What

remains unclear is whether and, if so, why banks – given a certain funding shock – reduced

their lending more to some countries than to others. This question is not only interesting from

an academic perspective but also from the standpoint of policy makers who want to gauge

international banks’ commitment to their country during a crisis.

An empirical analysis to establish whether bank-borrower closeness can explain banks’

differentiated lending response requires detailed information. Data should cover lending to

various countries by individual banks (to exploit within-bank variation) as well as lending by

various banks to individual countries (to control for credit demand at the country level). The

data should ideally also contain the individual deals that underlie these credit flows, so that

information on borrowers and inter-bank cooperation can be exploited. We use data on cross-

border syndicated bank lending that fulfil these requirements.

Loan syndications – groups of financial institutions that jointly provide large loans to

corporate borrowers – are one of the main channels of cross-border debt finance to developed

3

countries and emerging markets. In 2007, international syndicated loans made up over 40 per

cent of all cross-border debt funding to U.S. borrowers and more than two-thirds of cross-

border flows to emerging markets.1 This paper concentrates on the 117 largest banks in the

cross-border syndications market which jointly have a market share of over 90 per cent. We

use data on individual syndicated loans to construct for each bank a snapshot of its pre-crisis

and post-crisis cross-border lending to each individual destination country.

We use regression techniques to explain the stability of cross-border lending through a set of

variables that measure how close banks are to corporate borrowers in various destination

countries. We control for changes in credit demand through country fixed effects – basically

analyzing why different banks change their lending to the same country differently (within-

country comparison). We also replicate this approach at the borrower level with firm fixed

effects. In both cases we also use bank fixed effects to analyze how a particular bank – given

a certain funding shock – changes its lending to different countries or firms differently

(within-bank comparison). This combination of fixed effects allows us to focus on variables

that are specific to bank-country pairs (or bank-firm pairs) and to empirically isolate the

impact of bank-borrower closeness on international lending stability.

Our analysis shows that bank-borrower closeness, measured along a number of dimensions, is

an important determinant of the resilience of cross-border lending. During the global

financial crisis banks continued to lend more to countries that were geographically close,

where they were integrated in a network of domestic co-lenders, and where they had built up

more lending experience. For emerging markets, countries with weak legal creditor

protection, and countries with less publicly available borrower information, we also find that

owning a subsidiary stabilizes a bank’s cross-border lending. That is, a local presence appears

to be more important in weaker institutional environments. We conclude that even in a ‘hard-

information’ setting such as the market for syndicated corporate loans, closeness and the

ability to generate and exploit proprietary information is important.

This paper not only contributes to the literature on the recent crisis but also complements

earlier studies on the determinants of international bank lending. Van Rijckeghem and Weder

(2001, 2003) find that international banks that are exposed to a financial shock, either in their

home market or in another country, reduce lending to third countries. Jeanneau and Micu

1 Cross-border debt funding is defined as the sum of international syndicated credit, international money market

instruments, and international bonds and notes (Bank for International Settlements, Tables 10, 14a, and 14b).

4

(2002) show that cross-border lending is determined by macroeconomic factors, such as the

business cycle and the monetary-policy stance, in both home and host country. Buch,

Carstensen and Schertler (2010) find that interest-rate differentials and energy prices also

influence international bank lending. Our paper goes beyond assessing the impact of

macroeconomic factors on international lending. We instead test for the importance of bank-

borrower closeness, a topic that hitherto has not been analyzed in an international context.

Our paper is also related to Schnabl (2011) and Chava and Purnanandam (2011) who focus

on the reduction in cross-border lending to Peruvian and U.S. banks, respectively, after the

1998 Russian default. Both papers find that this external shock forced banks to contract

domestic lending and that this credit crunch had a negative impact on corporate borrowers. In

a similar vein, Khwaja and Mian (2008) show that after unexpected nuclear tests in India and

Pakistan, Pakistani banks transmitted liquidity shocks to their corporate clients. We do not

identify shock transmission per se but instead analyze why, given a certain shock, banks

adjusted their credit supply differently to different countries.

Finally, we add to the literature on multinational banking. A number of papers demonstrate

that multinational-bank affiliates can act as shock transmitters. Peek and Rosengren (1997,

2000) show how the drop in Japanese stock prices in 1990 led Japanese bank branches in the

U.S. to reduce credit. Similarly, Imai and Takarabe (2011) find that Japanese banks

transmitted local real estate price shocks to other prefectures in Japan. De Haas and Van

Lelyveld (2010) show that lending by multinational bank subsidiaries depends on the

financial strength of the parent bank. Our paper is related to this literature as we compare

cross-border lending by banks with and without a subsidiary in a particular country. In doing

so we connect the literature on the stability of international and multinational bank lending.

The paper proceeds as follows. Section 2 reviews the literature on bank-borrower closeness

and derives our theoretical priors. Section 3 then explains our data and econometric

methodology, after which Section 4 describes our empirical findings, a set of robustness tests,

and a number of extensions. Section 5 concludes.

5

2. Bank-borrower closeness and cross-border lending stability: Theoretical predictions

In this paper we test whether bank-borrower closeness, and the associated ability of banks to

collect, use, and exploit information about borrowers, contributes to cross-border lending

stability during a financial crisis. There are two closely related reasons why we expect that

bank-borrower closeness may stabilize lending.

First, access to and use of borrower information helps a bank to reduce information

asymmetries and overcome debt-related agency problems. This in turn lessens the need for

credit rationing (Stiglitz and Weiss, 1981). Containing information asymmetries is especially

important during a financial crisis, when the proportion of firms with low net worth increases

(Ruckes, 2004). Low firm quality exacerbates adverse selection and moral hazard and banks

need to intensify their screening and monitoring (Allen, 1990; Diamond, 1984; Berger and

Udell, 2004). We therefore expect that banks are more likely to continue lending if they are

better able to generate information about borrowers.

Second, the proprietary information that banks generate through repeat lending to borrowers

may give them market power (Degryse and Ongena, 2005). Such market power can be used

to carve out local captive markets by creating adverse-selection problems for competitor

banks (Dell’Arriccia, 2001; Agarwal and Hauswald, 2010). We expect that during times of

distress banks will be more reluctant to withdraw from countries where they have built up

lending relationships and where they expect to be able to exploit their local market power.2

The literature suggests four concrete dimensions of bank-borrower closeness that proxy for

banks’ ability to collect, use and exploit borrower information. The first measure is the most

literal interpretation of closeness: the geographical distance between a bank and its

borrowers. Distant banks find it more difficult to screen and monitor borrowers (Jaffee and

Modigliani, 1971; Hauswald and Marquez, 2006). Agarwal and Hauswald (2010) find that

this mainly reflects distant banks’ inability to collect and use ‘soft’ information.3 To the

2 Relationship lending may also involve implicit long-term contracting in which banks accept low interest rates

in the short run in the expectation of more profitable lending in the long run (Sharpe, 1990). When banks have

invested in a borrower by initially lending at low interest rates, they will be less inclined to stop lending and

forego future rents. 3 Petersen and Rajan (2002) show that while distance matters for credit availability of small US firms, its role

has declined over time as more borrower information and better ways to process it have become available. In

line with geographical credit rationing, Portes, Rey, and Oh (2001); Buch (2005); and Giannetti and Yafeh

(2008) find a negative relationship between distance and international asset holdings, including bank loans.

6

extent that screening and monitoring become more important during a crisis, banks may

retrench more from distant countries. We therefore expect geographical distance to have a

negative impact on cross-border lending stability.

A second measure of closeness that we consider is the experience the bank has in lending to a

borrower.4 Through repeat lending banks reduce information asymmetries and build up

proprietary information about borrowers (Rajan, 1992; Boot, 2000; Boot and Thakor, 2000).

They can re-use this information when lending to the same borrower (Greenbaum and

Thakor, 1995) and the more experienced banks become, the more they rely on their

proprietary information (Agarwal and Hauswald, 2010). As experienced banks are better

informed about their borrowers, rely more on their own information, and can potentially

better exploit it, they may be less inclined to stop lending during a crisis.

As a third measure of closeness we use the presence of a local subsidiary. A presence on the

ground reduces information asymmetries as loan officers are closer to the borrower and better

placed to extract soft information (Mian, 2006). This may allow a bank to continue lending to

existing borrowers during uncertain times because screening and monitoring can be stepped

up quite easily. Local staff can also help a bank to generate (and subsequently monitor) new

cross-border deals. Finally, because soft information is not easily transferable across banks,

international banks with a local subsidiary may have greater local market power, which

provides the bank with an incentive to stay committed to the country.

While a local subsidiary reduces the physical distance between loan officer and firm, it also

creates a functional or hierarchical distance within the bank. Banks may experience

difficulties in efficiently passing along (soft) information from the subsidiary to headquarters

(Aghion and Tirole, 1997; Stein, 2002). Indeed, Liberti and Mian (2009) find that when the

hierarchical distance between the information-collecting agent and the manager that approves

a loan is large, less ‘soft’ or subjective and more ‘hard’ information is used. Alessandrini,

Presbitero, and Zazzaro (2009) show for Italy that a greater functional distance between loan

officers and headquarters adversely affects credit availability of local firms.5

4 Petersen and Rajan (1994) and Berger and Udell (1995) find that the duration of a bank-borrower relationship

impacts credit availability. 5 Cerqueiro, Degryse, and Ongena (2009) provide an excellent overview of the literature on the relationship

between distance, banks’ organisational structure, and the supply of bank lending.

7

If the incentives of local loan officers are not well aligned with those of the parent bank,

internal agency costs may hamper cross-border lending as well (Scharfstein and Stein, 2000).

Such costs increase with distance if parent banks find it more difficult to supervise

management in far-away places (Rajan, Servaes, and Zingales, 2000). Kanz (2010) provides

experimental evidence that shows that hierarchical distance discourages both the collection

and the use of soft information and that such within-bank agency conflicts can be mitigated

by performance pay that aligns the incentives of managers and local loan officers.

In all, our prior is that cross-border lending by a bank with a subsidiary in a destination

country will be more stable than lending by a bank without one. However, this will only be

the case if the positive effect of having local loan officers is not offset by higher within-bank

information and agency costs due to an increased functional distance.

Finally, we use the extent to which the bank cooperates with domestic banks as our fourth

measure of closeness. Banks may improve the information they have about new and existing

foreign borrowers by cooperating with local domestic banks that possess a comparative

advantage in reducing information asymmetries vis-à-vis local firms (Mian, 2006; Houston,

Itzkowitz, and Naranjo, 2007). Domestic banks are not only geographically closer to these

firms but also share a common language and culture. Moreover, they may have a more

intimate knowledge of local legal, accounting and other institutions and their impact on

firms.6 Information sharing with domestic banks may also reduce adverse-selection problems

(Pagano and Jappelli, 1993). By co-lending with domestic banks, international banks may

also gradually increase their own knowledge of local firms and thus find it easier to collect

and use information about the firms themselves. We therefore expect that international banks

that are well-integrated in a lending network of domestic banks are a more stable credit

source during a financial crisis.

Section 3 now describes our data and methodology to test whether closeness – measured as

distance, lending experience, subsidiary presence, and cooperation with domestic banks –

determined the reduction in the credit supply of individual banks to individual countries.

6 Empirical evidence backs up the assertion that co-operation with domestic banks reduces agency problems.

Nini (2004) finds that local bank participation leads to larger, longer, and cheaper syndicated loans. Esty (2004)

and Qian and Houston (2007) show that participation of domestic banks in lending syndicates and project-

finance loans, respectively, reduces interest rates.

8

3. Data and econometric methodology

3.1. Data

Our main data source is the Dealogic Loan Analytics database, which contains

comprehensive information on virtually all syndicated loans since the 1980s. We download

all syndicated loans to private borrowers worldwide during the period January 2000-

September 2009 and split each loan into the portions provided by the syndicate members.

Loan Analytics contains information on loan breakdown for about 25 per cent of all loans.

When arranging banks report a loan to Dealogic they fill out a form with basic deal

information as well as details about the loan distribution. Whereas banks have an incentive to

declare deals in order to let them count towards their league-table position, they do not have a

strong incentive to provide all details. This is left to the discretion of the bank. Since we are

concerned that ‘full-information deals’ differ from ‘limited-information’ deals we perform a

probit analysis to check whether the propensity of banks to provide loan-distribution

information is related to observable characteristics. We do not find systematic differences

between deals with and without information on loan allocation.7

For the 25 per cent of the loans for which we have full information we allocate the exact loan

portions to the individual lenders. For the other 75 per cent we have to use a rule to allocate

loan portions. For our baseline regressions we use the simplest rule possible: we divide the

loan equally among the syndicate members. Section 4.3 presents robustness tests that show

that our results also hold when we allocate the 75 per cent of the sample in other ways. For

our main sample period – July 2006-September 2009 – we split a total of 21,323 syndicated

loans into 131,113 loan portions.

We then use these loan portions to reconstruct for each bank the volume and country

distribution of its cross-border lending. We define cross-border lending as loans where the

nationality of the (parent) bank is different from the nationality of the borrower and where the

loan is provided by the parent (Citibank lending from the U.S. to a Polish firm) rather than by

7 We find some statistically weak evidence that ‘busy’ arrangers with a high loan flow are somewhat less

inclined to provide full information. The same holds for arrangers from individualist societies (Hofstede, 2001).

The analysis is available upon request from the authors.

9

a subsidiary (Citibank Poland participating in a loan to a Polish firm). Over 94 per cent of

cross-border lending is of the former type and therefore included in our dataset.8

Next, we identify all commercial, savings, cooperative, and investment banks that at the

group level provided at least 0.01 per cent of global syndicated cross-border lending and that

participated in at least twenty cross-border loans in 2006. This leaves us with 117 banks from

36 countries, both advanced countries (75 banks) and emerging markets (42 banks).9

Together these banks lent to borrowers in 59 countries and accounted for over 90 per cent of

all cross-border syndicated lending in 2006.

Appendix Tables A1 and A2 list all banks and destination countries in our sample. Table A1

also shows each bank’s country of incorporation and its absolute and relative share of the

market for cross-border syndicated lending. Most banks have a small pre-crisis market share,

but there are a few big players with a share of over 3 per cent: RBS/ABN Amro (7.7 per

cent), Deutsche Bank (6.1), BNP Paribas (4.7), Citigroup (4.7), Credit Suisse (4.5), Barclays

(4.2), Calyon (3.2), JPMorgan (3.1), Mitsubishi UFJ (3.1), and Commerzbank (3.0).10

For each bank we calculate cross-border lending volumes to individual destination countries

in the pre-crisis period (July 2006-June 2007) and the period after the Lehman Brothers

collapse (October 2008-September 2009). Note that we disregard the intermediate period July

2007-September 2008 that encompasses the early stage of the crisis. This allows us to make a

clean comparison between the most severe crisis period – the year after the unexpected

collapse of Lehman Brothers in September 2008 – and the pre-crisis period – the year before

the market for short-term asset-backed commercial paper began to dry up in July 2007.

Our first cross-sectional dependent variable is a dummy Sudden stop that is 1 for each bank-

country pair where a bank completely stopped lending during the crisis (but where it was

8 In Section 4.3 we present robustness tests that indicate that our results remain unchanged when we include

syndicated lending through bank subsidiaries. 9 We define emerging markets as all countries except high-income OECD countries. Although Slovenia and

South Korea were recently reclassified as high-income countries we still consider them to be emerging markets. 10 During our sample period RBS acquired part of ABN Amro; Bank of America acquired Merrill Lynch; and

Wells Fargo acquired Wachovia. We treat these merged banks as a single entity over our whole sample period.

We add the number of loans their respective parts provided during the pre-merger period and calculate other

bank-specific variables as weighted averages, using total assets of the pre-merger entities as weights.

10

active before).11 Our second dependent variable – Volume – is defined as the log difference of

(1 plus) the amount of cross-border lending by a bank to a country between the post-Lehman

Brothers and the pre-crisis period. To reduce the probability that our results are affected by

outliers we exclude observations above (below) the 99th (1st) percentile for Volume.

Table 1 shows that our dataset includes 1,913 different bank-country pairs. On average an

international bank was lending to firms in 16 different countries before the demise of Lehman

Brothers. Banks reduced their lending on average by 52 per cent during the crisis to a

destination country. The variable Sudden stop indicates that banks even completely stopped

lending to borrowers in 44 per cent of the countries where they were active before the crisis.

[INSERT TABLE 1 HERE]

As discussed in Section 2, we create four variables that measure different aspects of bank-

borrower closeness for all bank-country combinations (the Closeness variables in Table 1).

We first use the great circle distance formula to calculate the geographical Distance between

a bank’s headquarters and its various countries of operation as the number of kilometers (in

logs) between the capitals of both countries. The average distance to a foreign borrower is

4,731 km, but there exists considerable variation (the standard deviation is 3,765 km).

Second, we create a variable that measures a bank’s prior experience in syndicated bank

lending to a particular country. Experience is defined as the number of loans that a bank

provided since 2000 and that had matured by July 2006. The average number of prior loans is

36 and ranges between 0 and 2,277. We exclude loans that were still outstanding during the

pre-crisis year in order to avoid a mechanical correlation with our dependent variable.

Because the relationship between Experience and lending stability may be concave in nature,

we use the natural log.

Third, we link our banks to Bureau van Dijk’s BankScope database to collect detailed

information on their ownership structure. For each banking group we identify all majority-

owned foreign bank subsidiaries. We create a dummy variable Subsidiary that is one in each

country where a bank owns a subsidiary. A typical bank owns a subsidiary in three foreign

countries and in 18 per cent of our bank-country pairs a subsidiary is present.

11 Note that complete information is available to construct this variable. Even though we only have loan share

information for 25 per cent of the sample, we know the identity of all lenders in each syndicate.

11

Fourth, we count for each bank and in each of its countries of operation, the number of

different domestic banks with which it has cooperated in a syndicate since 2000. We divide

this number by the total number of domestic banks in a particular country to create the

variable Domestic lenders. On average a bank has worked with 16 different domestic banks

in a given country, which is 34 per cent of the average number of domestic lenders. Variation

is large, however, with some banks never cooperating with domestic banks whereas others

have cooperated at least once with each domestic bank.

Appendix Table A4 contains a pair-wise correlation matrix of the closeness variables. The

correlations are moderate, between -0.22 and 0.36, and we are therefore confident that the

variables provide sufficient independent information and that multicollinearity is not an issue.

3.2. Econometric methodology

To examine whether bank-borrower closeness determines lending stability, we use the

bankruptcy of Lehman Brothers as an exogenous event that triggered a sudden stop in cross-

border lending. We compare, in a cross-sectional setting, the lending volume in the year after

the Lehman-Brothers collapse to the lending volume in the year before the crisis. In doing so,

we control for all time-invariant characteristics of recipient countries that influence the level

of cross-border lending (such as the institutional environment and the level of economic

development) plus all time-invariant factors that affect the lending volume of bank i to

country j. This allows us to focus on testing for heterogeneity in cross-border bank lending

stability due to differences in banks’ closeness to borrowers in various destination countries.

We use country-fixed effects to focus on differences across banks within countries. A key

advantage of this approach is that it allows us to neatly control for changes in credit demand

at the country level. Here we follow Khwaja and Mian (2008) who control for credit demand

at the firm level by using firm-fixed effects in regressions on a dataset of firms that borrow

from multiple banks (see Schnabl (2011) for a similar application). Since our dataset contains

information on multiple banks that lend to the same country, we use country fixed effects to

rigorously control for credit demand. This is important because the crisis hit the real economy

of countries to a different extent and with a different lag. Firms’ credit demand to finance

working capital and investments was consequently affected to varying degrees. Cetorelli and

Goldberg (2011) follow a similar approach on the basis of country-level data on lending from

17 developed countries to 94 emerging markets.

12

Since banks are active in multiple countries, we also include bank fixed effects to control for

bank-specific factors that might affect changes in lending. Whereas Khwaja and Mian (2008)

include bank-level control variables alongside a bank-specific funding measure (their

variable of interest), we prefer bank fixed effects because the combination of bank and

country fixed effects allows us to rigorously test the closeness variables that link bank i to

country j. Summarizing, our cross-sectional baseline specification is:

ijjiijij CL ηϕεβ +++⋅=Δ ' (1)

where subscripts i and j denote banks and destination countries, respectively; β’ is a

coefficient vector; Cij is a matrix of closeness variables; εi and φj are vectors of bank- and

country-fixed effect coefficients, respectively; and ηij is the error term. ijLΔ is either Sudden

stop (a dummy that is 1 for each bank-country combination where bank lending came to a

complete halt during the crisis) or Volume (the log change in lending by bank i to country j).

Second, we estimate firm-level regressions on a sample of firms that borrowed from at least

two banks in our sample before the crisis and borrowed at least once after the Lehman

Brothers collapse. The dependent variable Termination is the probability that bank i, a

creditor of firm k before the crisis, ended the lending relationship after the default of Lehman

Brothers. We can now use firm level fixed effects to control for changes in credit demand:

ikkiikik CT ηϕεβ +++⋅= ' (2)

where subscripts i and k denote individual banks and firms, respectively; β’ is a coefficient

vector; Cik is a matrix of closeness variables; εi and φk are vectors of bank- and firm-fixed

effect coefficients, respectively; and ηik is the error term. Tik is a Termination dummy that is 1

if bank i stopped lending to firm k during the crisis and zero otherwise.

In all specifications, we control for Exposure which measures for each bank-country pair the

outstanding syndicated debt as a percentage of the bank’s total assets at the time of the

Lehman Brothers collapse. On average this country exposure was close to 0.5 per cent of the

bank’s balance sheet. We are agnostic about the impact of this variable on lending stability.

Banks may have adjusted lending the most where they had high pre-crisis exposures, for

13

instance because risk limits became more binding for such countries. On the other hand,

banks may have retrenched more from ‘marginal’ countries while staying put in core markets.

We use OLS to estimate the Volume regressions and a linear-probability model for the

Sudden stop and Termination regressions. Standard errors are heteroskedasticity robust and

clustered by bank. All our results are robust to clustering at the country level (see columns 5

and 10 in Table 5).

4. Empirical results

4.1. Baseline results: Country level

Table 2 presents the results of our country-level baseline specification. In columns 1 to 5 and

8 to 12 we first add the closeness variables one by one and then simultaneously while

including bank and destination-country fixed effects in each case. We explain about 45 per

cent of the cross-country variation in Sudden stop and 35 per cent of the change in Volume.

Our four closeness variables turn out to be strongly correlated with the stability in bank

lending during the crisis.

[INSERT TABLE 2 HERE]

First, we find a significant and robust negative effect of distance on lending stability. The

probability of a sudden stop increases with distance to the destination country and banks

continue to lend more during the crisis to borrowers that are geographically close. Distance

constraints do not only have a negative impact on the amount of cross-border lending, as

documented in earlier studies, but also on its stability. Based on the results in column 5, we

estimate that a one standard deviation increase in distance increases the probability of a

sudden stop by 5.4 percentage points. In addition, column 12 shows that a 10 percent increase

in distance aggravates the decline in cross-border credit by 2.6 per cent (column 12).12

Second, a bank’s previous experience with cross-border syndicated lending to a particular

country has a strong positive impact on its lending stability during the crisis. A one standard

deviation increase in experience reduces the probability of a sudden stop by 12 percentage

12 The dependent variable and the regressor are expressed in logs so the parameter can be interpreted as an elasticity.

14

points. Furthermore, a 10 per cent increase in experience reduces the decline in cross-border

lending during the crisis by 3.1 per cent.

Third, we find some statistically weak evidence that indicates that cross-border lending to

countries in which a bank owns a subsidiary declines less than lending to countries where that

same bank does not own an affiliate (columns 3 and 5). The results in column 5 indicate that

the probability of a full lending stop is almost 5 percentage points lower when a bank owns a

local affiliate, all else equal. Yet, this effect is only significant at the 10 per cent level and the

impact of having a subsidiary is even less precisely estimated in the Volume regressions. In

Table 6 in Section 4.3 we will show that the subsidiary effect is more robust in emerging

markets, countries with weak creditor protection, and countries with less publicly available

borrower information. In such countries a local presence may be more important.

Fourth, we find that international banks that cooperate with many domestic banks are more

stable sources of cross-border credit. Based on the results in column 5, the probability of a

full sudden stop is 10 percentage points less in case a bank’s level of cooperation with

domestic banks in a country equals the mean level, compared to countries where it had not

cooperated with domestic banks at all. We find no significantly positive impact of

cooperation with domestic lenders on lending volume during the crisis.

So far we have combined destination country fixed effects and bank fixed effects to focus on

variables at the level of bank-country pairs. Since these fixed effects capture (un)observed

characteristics of banks and destination countries, concerns about omitted variable bias

should be fairly limited. Any such bias must stem from omitted country-pair variables that

are correlated with both our closeness variables and lending stability. We now check whether

our results continue to hold when we control for specific country-pair characteristics or

country-pair fixed effects.

In columns 6 and 13 we first add two country-pair variables. Trade is (the log of 1 plus) total

export plus import volume in US dollars between the home country of bank i and destination

country j. Following Tinbergen (1962) a voluminous literature has developed on gravity

models that link trade flows to the distance between trading partners. Since deeper trade

integration may be associated with more stable financial integration, we expect a negative

relationship between Trade and the stability of cross-border lending. The coefficients have

the correct sign but are imprecisely estimated.

15

Our second country-pair variable is Bank FDI, which measures the number of banks from the

home country of bank i that own a subsidiary in country j (based on information from

Claessens and Van Horen, 2011). To the extent that financial FDI (foreign-bank ownership)

and financial trade (cross-border lending) are complements (Brainard, 1997) we expect a

stabilizing impact of bilateral FDI on cross-border bank lending. This is indeed the case.13

Importantly, a comparison with columns 5 and 12 shows that the coefficients for the

closeness variables Subsidiary, Distance, Domestic banks, and Experience remain virtually

unaffected when we control for the country-pair characteristics Trade and Bank FDI.

Next, we go one step further and include country-pair fixed effects in columns 7 and 14. We

now compare banks from the same home country that operate in the same destination

country. This prevents us from estimating the effect of geographical distance but still allows

us to estimate the other coefficients. Our results continue to hold in this very restricted

specification: lending stability is higher for more experienced banks and banks that cooperate

more with domestic banks. The coefficient for Subsidiary is imprecisely estimated due to

limited within-destination country variation among banks from the same home country.

In all, we find strong evidence in favor of our main hypothesis: bank-borrower closeness – as

measured along a number of dimensions – is a defining characteristic of cross-border lending

stability during financial-system stress.

4.2. Baseline results: Firm level

So far our identification strategy has relied on controlling for changes in credit demand

through country fixed effects. Although we follow Khwaja and Mian (2008) our approach is

coarser as we apply it at a higher aggregation level (bank-country pairs instead of bank-firm

pairs). If country fixed effects imperfectly control for changes in credit demand this could

bias our results if heterogeneity in credit demand at the firm or sector level would not be

orthogonal to our closeness variables. To get around this issue, we analyze whether our

closeness variables determine lending stability at the firm level as well.

We estimate regressions for a sample of firms that borrowed from at least two banks in our

dataset before the crisis and at least once after the default of Lehman Brothers. The dependent

variable is the probability that bank i, a creditor of firm k before the crisis, terminated the

13 We find similar results when we measure Bank FDI as the number of subsidiaries in country j owned by

banks from the home country of bank i as a percentage of all banks in country j.

16

lending relationship after the default of Lehman Brothers. We can now use fixed effects at the

firm level to control more precisely for changes in credit demand. We now also measure

Experience at the firm level as the (log of the) number of loans provided by bank i to firm k

since 2000 that had matured by July 2006.

We also include a dummy variable Arranger that is one if the bank acted at least once as an

arranger for the firm. A typical syndicate consists of two tiers: arrangers and participants. The

arrangers comprise the senior tier and negotiate the terms with the borrower. They then

allocate a substantial part of the loan to a junior tier, the participants, who assume a more

passive role. Arrangers usually act to some extent as delegated monitors for the participants

(Sufi, 2007). We therefore expect that banks that previously arranged a loan for a client are

particularly well-informed (are ‘closer’). Arrangers may also be more stable lenders because

they expect future fee income from ancillary business – such as re-arranging the syndicated

loan or arranging a future bond issue – as part of a longer-term relationship.

In columns 1 to 5 of Table 3 we again first add the closeness variables one by one and then

simultaneously. As a robustness check, we also re-estimate the simultaneous specification on

the basis of a logit model in column 6. The firm-level results confirm our earlier findings: the

probability that a bank terminates a lending relationship increases if the firm is more distant,

if the bank has not cooperated much with domestic banks, and if the bank does not have

substantial lending experience with a particular firm. As expected, we also find that banks

that performed a due diligence in the past as a syndication arranger are more likely to

continue to lend to a firm.

The economic impact of our closeness variables remains significant too. A one standard

deviation increase in distance increases the probability that a bank terminates a lending

relationship by 3 percentage points. Furthermore, a one standard deviation increase in

experience decreases the probability of termination by 8 percentage points. Finally,

increasing the cooperation level with domestic lenders from zero to the mean level reduces

the probability that a bank stops lending to a firm by 14 percentage points.14

[INSERT TABLE 3 HERE]

14 Based on column 5. Note that the summary statistics for some regressors in this firm-level sample deviate

slightly from those in Table 1. The standard deviation of experience at the firm level is 0.71 and the mean level

of cooperation with domestic lenders is 37.

17

4.3. Robustness

CLOSENESS AND LENDING DURING NORMAL TIMES: A PLACEBO TEST

We put forward the idea that bank-borrower closeness determined cross-border lending

stability during the crisis. The results in Table 2 and 3 are in line with this but it is possible

that the strong relationship we find is not specific to our sample period. To be clear, we do

not expect closeness to be irrelevant before the start of our sample. After all, there exists a

sizeable literature on closeness and the quantity of bank lending (we cite the main

contributions in Section 2). We expect instead that the importance of bank-borrower

closeness increased during the crisis and that closeness explains a substantial part of the

heterogeneity in the lending reversal.

To compare the importance of the closeness variables during our sample period with a non-

crisis period, we perform a placebo test on out-of-sample data over April 2004-June 2007.

This is the period that ends just before the market for asset-backed commercial paper started

to contract and the British bank Northern Rock experienced a run. We partition this ‘crisis-

free’ placebo period into three sub-periods (April 2004-March 2005; April 2005-June 2006;

and July 2006-June 2007) and calculate our dependent variables Sudden stop and Volume by

comparing the last and the first period, just like we do for our real sample period.

Next, we combine all observations for both the placebo and the real sample and re-run our

baseline regressions. We interact our closeness variables with a dummy that is one for

observations in the real sample and zero for placebo observations. If bank-borrower closeness

was more important during the crisis, we expect significant interaction terms with the same

sign as the closeness coefficients on their own.

Table 4 confirms that while our closeness variables where relevant in the placebo period, they

became much more significant during the real sample period. For instance, in column 1 the

marginal effect of distance on the probability of a sudden stop is 0.022 before the crisis

(significant at the 5 per cent level) whereas the total effect (base plus interaction) increases to

0.10 during the crisis (interaction term significant at the 1 per cent level). This means that the

impact of distance on the probability of a lending stop was five times higher in our real

sample than in the placebo period.

[INSERT TABLE 4 HERE]

18

MISCELLENEOUS ROBUSTNESS CHECKS

Table 5 presents a battery of robustness tests to check whether our results are sensitive to

changes in variable definitions or estimation techniques. For ease of comparison, columns 1

and 6 replicate our baseline results.

First, column 2 reports a logit regression instead of a linear probability OLS and shows that

this does not alter our results. Next, columns 3 and 8 show regression results where we

include syndicated lending by bank subsidiaries. So far we have focused on pure cross-border

lending: loans where the nationality of the (parent) bank is different from the nationality of

the borrower and where the loan is provided by the parent rather than by a subsidiary.

However, in about six per cent of all loans it is a local subsidiary rather than the parent bank

that holds the loan on its balance sheet.15 When we include these loan portions, our results

continue to hold and – not surprisingly – the impact of having a local subsidiary becomes

somewhat more significant.

Columns 4 and 9 show results where we exclude the quartile of bank-country pairs with the

smallest number of pre-crisis loans to check whether our results are driven by artificially

large changes in countries with only a few pre-crisis loans. This turns out not to be the case.

Next, in columns 5 and 10 we cluster our standard errors by country instead of bank and in

column 7 we do not exclude any observations. Our results do not appear to be sensitive to the

level of clustering or the way we handle outliers. In unreported regressions we also winsorize

by setting the upper and lower tail values equal to the values of the 1st and 99th percentile.

This does not materially change any of our results either.

Finally, as mentioned in Section 3.1, Loan Analytics only provides information on the loan

breakdown for about 25 per cent of all loans. So far we have used a rather simple rule to

distribute the other 75 per cent of the loans: we assumed that each lender provided the same

amount. To minimize the risk that we introduce a measurement error by choosing a particular

distribution rule, we recalculate our Volume variable using two alternative methods.16

15 Foreign subsidiaries typically do not participate in syndicated loans as the amounts involved are too large for

the their own balance sheet (in particular when the host-country regulator enforces large-exposure limits). Funds

are therefore provided directly by the bank’s headquarters. The subsidiary is often still involved by providing

the parent bank with local information, which is one of the closeness effects we analyze. 16 Simply using the 25 per cent of the loan population for which we have breakdown information would

introduce a measurement error in the dependent variables Volume and Sudden stop. When constructing the

lending flows from bank i to country j we would need to assume that the loan amount is zero for the loan shares

19

In column 11, we use the 25 per cent of our sample with full information to estimate a model

in which the loan amount of individual lenders is the dependent variable. As explanatory

variables we use the average loan amount (loan amount divided by the number of lenders), a

dummy that indicates whether a lender is an arranger or a participant, an interaction term

between this arranger dummy and a variable that measures whether the borrower is a repeat

borrower or not, an interaction term between the arranger dummy and a post-Lehman time

dummy, and a set of bank and country dummies. We then use the estimated coefficients to

predict the loan portion for those lenders for whom we do not know the actual amounts (we

replace negative predicted values with zero and predicted values exceeding the total loan

amount with this amount). Our results are robust to this alternative rule.

Finally, we apply an allocation rule where each lender receives just one per cent of the loan

amount with the exception of one randomly chosen lender that receives the rest of the loan

(column 12). Even in this case our results continue to hold.17 When we apply the various

rules to the 25 per cent of our sample for which we know the loan distribution, we find that

the average deviation from the true loan amount is 29 per cent for the simple rule, 47 per cent

for the model-based allocation, and 90 per cent for the extreme distribution rule.

[INSERT TABLE 5 HERE]

4.4. Extensions

BANK-BORROWER CLOSENESS AND INSTITUTIONAL QUALITY

If closeness allows banks to rely more on lending relationships and to generate proprietary

information about new borrowers, it may matter more in countries where less public borrower

information is available, where legal protection of creditors is below par, and where loan

contracts are costly to enforce (Jappelli, Pagano, and Bianco, 2005). In such institutionally-

that are missing, which obviously is incorrect. Because the availability of loan allocation information is more or

less random, this problem would extend to all bank-country pairs. 17 We perform three additional but unreported robustness tests. We first allocate half of each loan to the

arrangers and half to the participants (this reflects the average loan allocation across both lender types in our

full-information sample). Within both groups we divide the loan equally. The second rule is similar except that

we increase the arrangers’ share during the crisis to reflect that arranger shares are countercyclical (Ivashina and

Scharfstein, 2010). Third, we apply our basic rule to all loans, including those for which we have actual

allocation details. In each case our results remain quantitatively and qualitatively the same.

20

difficult countries, adequate screening and monitoring is particularly important to

complement scant public information and to prevent the need to seek legal recourse.

Indeed, the law and finance literature as developed by La Porta, Lopez-de-Silanes, Shleifer

and Vishny (1998) finds that better creditor protection during bankruptcy procedures makes

banks lend more. Qian and Strahan (2007) use data on syndicated loans and show that strong

creditor protection is also associated with longer tenors and lower interest rates. They find

that foreign banks are particularly sensitive to the legal environment.

To see whether our data support these ideas, we distinguish in various ways between

destination countries with ‘good’ and ‘bad’ institutions. We expect that bank-borrower

closeness and ‘inside’ information is more important in countries with a weak institutional

framework. In such countries trustworthy public or ‘outside’ borrower information is less

readily available and creditors are legally less well-protected and may be less successful in

recovering bad loans (see also Mian, 2006). We therefore expect a stronger relationship

between closeness and cross-border lending stability in such countries.18

In columns 1 and 2 of Table 6 we first make a rough distinction between developed countries

and emerging markets. In columns 3 and 4 we then distinguish between countries where the

quality and quantity of public creditor information is above and below the median of all

destination countries. In columns 5 and 6 we split the countries into those where creditor

protection is above and below the median protection level, whereas in columns 7 and 8 we

distinguish between countries with a legal system of English origin versus those with a

French, German, Nordic, or Socialist system. Creditor protection is strongest in English

systems (La Porta, López-de-Silanes, Shleifer, and Vishny, 1998). In each case we interact

the closeness variables with a dummy that is one for countries with below-median

institutional quality.

Finally, we compare the importance of closeness for lending to first-time versus repeat

borrowers (columns 9 and 10). If loans to repeat borrowers are plagued by fewer agency

problems, we expect closeness to have less impact on the reduction in cross-border lending.19

We define first-time borrowers as firms that had never borrowed from bank i before the 18 Brown and Zehnder (2007) present experimental evidence that shows that private, bilateral relationship

banking and public credit reporting, such as through credit registries, may be substitutes. 19 Prior loans and the associated borrower reputation can attenuate information asymmetries (Diamond, 1991

and Gorton and Pennachi, 1995). In line with this, De Haas and Van Horen (2010) find that arrangers of

syndicated loans need to retain less of repeat loans.

21

collapse of Lehman Brothers, whereas repeat borrowers are firms to whom bank i had lent at

least once. In these regressions we include two observations for each bank i-country j pair:

one where the dependent variable (Sudden stop or Volume) captures lending to first-time

borrowers in country j and a second one for lending to repeat borrowers. We interact the

closeness variables with a dummy that is one if the observation concerns lending to first-time

borrowers.

Overall, Table 6 provides some evidence that our closeness variables are more important in

institutionally-difficult countries. This effect is strongest for Subsidiary which has a

stabilizing impact in emerging markets and in countries where public credit information is

not widely available, where creditor protection is limited, and where the legal system has a

non-English origin. Prior Experience in a country is somewhat more important in emerging

markets and when lending to first-time borrowers.20 These results thus complement those of

Qian and Strahan (2007) by showing that better creditor protection is not only associated with

cheaper and longer-maturity loans but also with more stable lending during a crisis.

While these results consistently point in the same direction – closeness matters more if the

institutional framework is less developed – they are not equally strong for all variables. For

instance, Distance reduces lending stability and cooperation with Domestic banks increases

lending stability regardless of how bank-friendly the institutional framework is.

[INSERT TABLE 6 HERE]

DISTANCE: GEOGRAPHY, CULTURE, OR INSTITUTIONS?

We find robust evidence for a strong negative relationship between geographical distance and

cross-border lending stability. However, the collection and transmission of borrower

information from a destination country to a bank’s headquarters may also be impaired by

cultural and institutional differences. For instance, notwithstanding a long geographical

distance, Spanish banks may have continued to lend to Mexican firms during the crisis 20 The finding that closeness is important for lending to both first-time and repeat borrowers shows that our

baseline results do not merely reflect a shift to repeat borrowers (who may be concentrated in ‘closer’

countries). Interestingly, the percentage of first-time borrowers did not decline much after the Lehman Brothers

collapse (from 42.4 to 41.0 per cent) possibly because concerns about high corporate leverage made repeat

lending less attractive.

22

because the cultural and historical ties between both countries made Spanish banks more at

ease in dealing with Mexican firms than with, say, Turkish firms (which are closer in

geographical than cultural terms).

Similarly, banks may feel more confident – in particular during a crisis – when lending to

firms in countries where the institutional and legal environment resembles that in their home

country. Table 6 provided some evidence that bank-borrower closeness matters more when

institutions are weaker in the destination country. In this sub-section we analyze whether

institutional differences between home and destination countries have a direct, negative

impact on lending stability.

To look into the importance of geographical, cultural, and institutional distance, we create a

number of non-geographical distance measures: a dummy variable that indicates whether the

home and destination country share a common language; the difference between an index of

creditor protection in the home and the host country; the difference between an index of

credit-information availability in the home and the host country; and a dummy variable that

indicates whether the origin of the legal system differs.

In Table 7 we add these alternative distance measures to our baseline specification. It

becomes clear that geographical distance is a very robust determinant of lending stability.

When we add Common language we find that this cultural proxy has a significant positive

impact on Volume (and a negative one on Sudden stop) but this effect does not survive in the

combined regressions in columns 6 and 12. Even when we add this variable, the coefficient

of Distance remains highly significant and is only marginally reduced in size.

The institutional-distance variables Dif. creditor protection and Dif. legal origin also have an

independent impact on lending stability but hardly affect the coefficient of Distance in a

bilateral ‘horse race’. Their coefficients also remain significant in the combined regressions

(with the exception of the coefficient of Dif. legal origin in column 12). Cross-border lending

is less stable if the bank’s home country and the destination country have different legal

origins and, more specifically, when banks are protected less in the destination country

compared to what they are used to in their home country. We do not find an impact of the

Dif. credit information variable. We conclude that cross-border lending is more stable when

banks and borrowers are geographically and institutionally closer.

[INSERT TABLE 7 HERE]

23

CLOSENESS AND THE CRISIS IMPACT ON LOAN SPREADS AND MATURITIES

Bank-borrower closeness may not only affect how banks adjust the amount of cross-border

bank lending during a crisis, but also how they change the spreads they charge and the

maturities they are willing to extent.

On the one hand, spreads may increase less when banks are closer to their borrowers. This

would indicate that borrowers profited from the relationship with their bank (see also Berger,

Miller, Petersen, Rajan, and Stein, 2005). On the other hand, the literature on geographical

credit rationing discussed in Section 2 sketches a less benign scenario in which close banks

exploit proprietary information (see also Schenone, 2010). Banks can extract rents if nearby

borrowers face lower transportation costs (Lederer and Hurter, 1986) or if distant competitor

banks face higher monitoring costs that they need to pass on to borrowers (Sussman and

Zeira, 1995 and Almazan, 2002).21 Spreads may also be positively related to the duration of

lending relationships as banks become more informed over time (Degryse and Van Cayseele,

2000). These issues may have been particularly relevant during the crisis when many lenders

left the syndications market and the remaining banks could exploit ‘captive’ borrowers.

As regards maturities, we expect a negative relationship between bank-borrower closeness

and the change in maturities during the crisis. Shorter maturities may reduce agency problems

(Barnea, Haugen and Senbet, 1980; Rey and Stiglitz, 1993) and banks may therefore have

reduced maturities the most for borrowers they were less close to.

Our data show that the average spread increased from 127 basis points in the pre-crisis period

to 260 basis points in the post Lehman period, while average maturity declined from 54

months to 37 months. However, as with the amount of loans, there is substantial variation

across lenders and destination countries. To analyze whether bank-borrower closeness can

explain this variation, we estimate regressions where our dependent variable is either the log

change in the average Spread charged by bank i to borrowers in country j or the log change in

21 Degryse and Ongena (2005) find evidence for transportation-cost induced spatial price discrimination: loan

rates are lower if the bank-firm distance is higher and the bank-competitor bank distance is lower. Hauswald and

Marquez (2006) present a theoretical framework in which there exists a positive (negative) relationship between

distance and spreads charged by transactional (relationship) banks.

24

the average Maturity of loans by bank i to country j.22 Spread is measured over Libor or

another relevant reference rate.

Our results in Table 8 indicate that loan spreads increased more during the crisis if a bank

owned a subsidiary in the destination country, if that destination country was geographically

close, if the bank had cooperated more with domestic banks, and if the bank had more

lending experience. When we add these explanatory variables at the same time (column 5)

only the impact of Distance remains statistically significant. These results are in line with

exacerbated spatial price discrimination during the crisis.

We also find that banks reduced loan maturities less for close borrowers and when lending to

countries where the bank had more experience and had cooperated more with domestic

banks. This is in line with theories that stress that shorter maturities reduce agency problems.

When we add these explanatory variables at the same time, only the impact of cooperation

with domestic lenders remains significant.

[INSERT TABLE 8 HERE]

4.5. Discussion and interpretation

FIXED LENDING COSTS

We interpret our results as indicating that banks’ ability to collect, use, and exploit

information about existing and new clients affects the stability of their cross-border lending.

However, one may argue that our results could also be explained by country-specific fixed

lending costs. After the Lehman Brothers collapse banks may have decided that it was not

worthwhile to incur such costs in countries where they only grant a few loans. Such countries

could be those that are geographically distant or where the bank had only built up limited

lending experience. While fixed lending costs are likely to have had some impact on lending

stability during the crisis, we think there are a number of reasons why such costs are unlikely

to drive our results.

First, throughout this paper we control for Exposure which measures for each bank-country

pair (or bank-firm pair) the amount of outstanding loans as a percentage of the bank’s total

assets before the crisis. So we already control for the level of activity in each destination 22 Our country fixed effects control for changing country risk and we thus focus on how different banks behave

differently within a country.

25

country. For instance, the robust negative impact of Distance on bank-lending stability

indicates that a bank that operates in two countries with the same amount of pre-crisis

business reduced lending more to the distant country.

Second, fixed costs that are sunk, such as the entry costs in Melitz (2003), cannot explain our

findings since we use a dataset of banks that had already entered their various destination

countries. Moreover, entry costs are much lower for syndicated lending than for exports of

manufactured goods (as in Melitz, 2003). In the case of exports of physical products (such as

cars) firms need to adjust products to local taste, set up distribution channels, and adjust

marketing. In contrast, the global market for syndicated loans is homogeneous and fixed

entry costs are low.

Entry costs may be high, however, when the bank sets up a subsidiary upon entry. Whereas

the costs of creating or buying a subsidiary will be sunk, there will still be substantial and

recurring fixed costs, for instance related to paying staff. However, if such fixed costs would

be important, we would expect our results to be particularly strong for our Subsidiary

variable. In contrast, the impact of this variable on lending stability is relatively weak.

Third, fixed costs that are prospective rather than sunk – such as costs to deal with local

lawyers and regulators – are also unlikely to explain our findings. In the case of syndicated

lending such prospective fixed costs are small compared to the per-unit (i.e. per loan) costs of

screening and therefore unlikely to affect a bank’s lending decision in a major way. That

being said, to the extent that such fixed costs matter, we expect that they matter more in

countries where they are relatively high. Indeed, in Table 7 we find some evidence for this.

All else equal, banks reduced lending more to countries where the legal system and

supervisory regime are different compared to their home country. Countries with very

different institutions are arguable more costly to operate in. Importantly, our closeness

variables continue to hold even when we control for such costly institutional differences.

Fourth, one may argue that (prospective and fixed) re-entry costs drive our results. In that

case, however, we would expect banks to mainly reduce or temporarily stop lending to

countries where re-entry is cheap. Arguably, this is the case in nearby countries, countries

where the bank has more experience, and countries where the bank has built up a network

with domestic banks. So, if re-entry costs are important, we would expect banks to continue

to lend in particular to remote countries and countries where they had cooperated less with

domestic banks, as re-entering these countries will be costly. Yet, we find the opposite.

26

ARRANGERS VERSUS PARTICIPANTS

As explained in Section 4.1, a typical syndicate consists of two tiers of lenders: arrangers and

participants. The former take the lead in negotiating the lending terms with the borrower and

in the due diligence, whereas the participants have a more passive role. In this paper, we do

not distinguish banks according to their role in the syndicate. We estimate the impact of

bank-borrower closeness on lending stability regardless of whether the underlying loans were

provided by arrangers or participant banks. However, one may argue that bank-borrower

closeness may be particularly important for arrangers, who take the lead in screening and

monitoring borrowers. In that case, one should allow the impact of our closeness variables to

vary over the two types of lenders.23 Our firm-level results in Table 3 provide some evidence

that the role of the lender might be important: all else equal, banks that previously acted as an

arranger were more likely to continue to lend to a firm during the crisis.

To look into this issue in more detail, we rerun our baseline regressions with a variable

Arranger that measures for each bank-country pair the percentage of pre-crisis loans where

the bank acted as an arranger rather than as a participant. We then interact our closeness

variables with Arranger to allow for a differential impact. We find no significant differences

between both tiers of syndicate members.24 Even for participant banks it matters whether they

are experienced in a country, whether the country is distant or not, and whether they are

integrated into a network of domestic banks.

This absence of a differential effect of our closeness variables on lending stability for the two

types of lenders justifies the aggregated approach we take in this paper. A potential

explanation is that while participants ‘outsource’ part of the screening and monitoring to

arrangers, in particular the administrative aspects, they likely perform an independent due

diligence as well. Sufi (2007, p. 642) provides evidence from practitioner interviews to this

effect: “Two main factors drive participation in syndicates, specifically, the quality of the

firm and how well the participant bank “knows” the firm. The measure of information

asymmetry I construct therefore attempts to capture how well participating banks know the

firm absent any information relayed by the lead arranger”. Indeed, Sufi then shows how

information asymmetries between the participant bank and the firm influences whether a loan

is provided or not. When public borrower information is limited “participant lenders on 23 Note that if closeness to borrowers would be inconsequential for participant banks, this would work against

finding impacts of our closeness variables on overall lending stability. 24 These results are available upon request from the authors.

27

syndicates are “closer” to the borrowing firm, both geographically and in terms of previous

relationships.” (p. 631).25

CROSS-BORDER VERSUS DOMESTIC SYNDICATED LENDING

Our results indicate that international banks were more inclined to keep lending to some

countries than to others. This shows that not only demand but also supply shocks have caused

the substantial variation we document in the reduction in cross-border lending after the

collapse of Lehman Brothers (Figure 1).

While a full analysis of the real impact of the crunch in cross-border lending on destination

countries is beyond the scope of this paper, Figure 2 illustrates that most countries were

unable to offset the decline in cross-border lending through increasing domestic syndicated

lending. The left-hand pane shows that there were only a few countries – such as China and

Japan – where increased lending by (often state-owned) banks more than compensated for the

severe drop in cross-border inflows. The right-hand pane shows that most countries

experienced a decline in total syndicated lending very similar to the decline in cross-border

syndicated lending (observations on the 45º line). Domestic lending was unable to cushion

much of the decline in credit from abroad. Only a few countries partially counterbalanced

reduced inflows with increased domestic lending.

This imperfect substitutability between cross-border and domestic syndicated loans implies

that the results we document in this paper are likely to have had severe consequences for the

total lending supply in the destination countries and therefore to have had a negative impact

on these countries’ real economies.

[INSERT FIGURE 2 HERE]

5. Conclusions

An increasing body of evidence indicates that international banks played a pivotal role in

transmitting the shock of the Lehman Brothers collapse across borders, thus spreading the

financial crisis throughout the global financial system. Yet, what remains unclear is whether

25 In line with this, Nini (2004) shows that the presence of local participants increases loan size and maturity and

conjectures that this is because they do a better job in screening and monitoring local borrowers.

28

and, if so, why banks – given a certain funding shock – reduced their cross-border lending

much more to some countries than to others.

To answer this question we construct a detailed dataset on international bank lending and

analyze to what extent various dimensions of bank-borrower closeness determine cross-

border lending stability. We combine bank fixed effects with country or firm fixed effects to

rigorously control for credit demand and other confounding factors.

In line with our priors, we find a strong and robust negative effect of geographical distance

on lending stability during the crisis. Banks remained more committed to countries in which

they had built up pre-crisis lending experience and where they were well-integrated into a

network of domestic co-lenders. For emerging markets, countries with weak creditor

protection, and countries with scarce public borrower information, we also find that owning a

local subsidiary stabilized a bank’s cross-border lending.

These findings paint a more nuanced picture than the black-and-white dichotomy of

transaction-based lending by large banks versus relationship lending by small banks. We

show that even in a sample of the largest international banks that provide loans to large

companies, bank-borrower closeness and access to ‘soft’ information – gathered through

repeat lending and co-lending with domestic banks – is important.

We also find that banks that are further from their customers are less reliable lenders during a

crisis. This finding is in line with studies that document a persistent regional segmentation of

the syndicated loan market. Carey and Nini (2004) find that borrowers tend to issue

syndicated loans in their own regional market (Europe, US or Asia) and that banks overweigh

their portfolios in favor of their regional market. Houston, Itzkowitz and Naranjo (2007)

show that information costs prevent domestic borrowers from issuing syndicated loans

abroad. Our results suggest that geographical segmentation may gradually be overcome if

banks build up lending relationships with local borrowers and banks.

Our results clearly bear on the policy debate on financial globalization. On the upside,

financial integration can help increase the quality and quantity of financial services available

to firms and households, especially in emerging markets. This may stimulate domestic

demand, eventually reduce global imbalances (Pongsaparn and Unteroberdoerster, 2011), and

allow countries to specialize more in line with their comparative advantage (Kose, Prasad,

and Terrones, 2003). Yet, shocks at the core of the global financial system may make cross-

border lending fickle and sudden-stops often have devastating consequences for local

29

economies. Can countries reap the benefits of financial integration while reducing concerns

about negative side effects? Our results provide a number of insights that may help answer

this question.

First, we interpret our findings to indicate that financial integration is a gradual process.

Through fostering (repeat) lending relationships with domestic borrowers and banks,

international banks become more deeply embedded in a local economy. We find that such

entrenched banks are less likely to ‘run’ during a crisis. This may partly explain why periods

of rapid lending inflows often turn to busts: banks have not had the time to get closer to local

borrowers, to forge durable lending relationships, and to collect proprietary information.

Macro-prudential policies to manage lending inflows may therefore make sense if they allow

for a more gradual deepening of lending relationships.

Second, we find that international banks with a local presence on the ground tend to be more

stable lenders in emerging markets and other countries with relatively weak institutional

environments. For emerging markets that are considering to open up their banking system

this implies that stimulating banks to ‘set up shop’ may kill two birds with one stone. Not

only do foreign-bank subsidiaries provide for a relatively stable credit source themselves, but

their presence may also stabilize cross-border debt flows.

Finally, our findings suggest that local financial development remains important and funding

cannot be completely ‘outsourced’ to foreign lenders. Domestic banks that are close to local

borrowers may continue to have a comparative advantage in screening and monitoring such

borrowers. Their ability to generate local information may, however, usefully be leveraged by

co-lending with international banks.

References

Agarwal, S. and R. Hauswald (2010), Distance and private information in lending, Review of

Financial Studies, 23(7), 2757-2788.

Aghion, P. and J. Tirole (1997), Formal and real authority in organisations, Journal of

Political Economy, 105, 1-29.

Aiyar, S. (2011), How did the crisis in international funding markets affect bank lending?

Balance sheet evidence from the UK, Working Paper No 424, Bank of England,

London.

30

Alessandrini, P., A.F. Presbitero, and A. Zazzaro (2009), Banks, distances, and firms’

financing constraints, Review of Finance, 13, 261-307.

Allen, F. (1990), The market for information and the origin of financial intermediation,

Journal of Financial Intermediation, 1, 3-30.

Almazan, A. (2002), A model of competition in banking: Bank capital versus expertise,

Journal of Financial Intermediation, 11, 87-121.

Barnea, A., R. Haugen, and L. Senbet (1980), A rationale for debt maturity structure and call

provisions in the agency theoretic framework, Journal of Finance, 35, 1223-1234.

Berger, A. and G. Udell (1995), Relationship lending and lines of credit in small firm finance,

Journal of Business, 68, 351-381.

Berger, A. and G. Udell (2004), The institutional memory hypothesis and the procyclicality

of bank lending behavior, Journal of Financial Intermediation, 13, 458-495.

Berger, A., N. Miller, M. Petersen, R. Rajan, and J. Stein (2005), Does function follow

organizational form? Evidence from the lending practices of large and small banks,

Journal of Financial Economics, 76(2), 237-269.

Boot, A.W.A. (2000), Relationship banking: What do we know?, Journal of Financial

Intermediation, 9, 7-25.

Boot, A.W.A. and A.V. Thakor (2000), Can relationship banking survive competition?,

Journal of Finance, LV(2), 679-713.

Brainard, S. (1997), An empirical assessment of the proximity concentration trade-off

between multinationals sales and trade, American Economic Review, 8, 520-544.

Brown, M. and C. Zehnder (2007), Credit reporting, relationship banking, and loan

repayment, Journal of Money, Credit, and Banking, 39(8), 1883-1918.

Buch, C.M. (2005), Distance and international banking, Review of International Economics,

13, 787-804.

Buch, C.M., K. Carstensen and A. Schertler (2010), Macroeconomic shocks and banks’

foreign assets, Journal of Money, Credit, and Banking, 42, 171-188.

Carey, M. and G. Nini (2007), Is the corporate loan market globally integrated? A pricing

puzzle, Journal of Finance, 62, 2969-3007.

Cerqueiro, G.M., H.A. Degryse, and S. Ongena (2009), Distance, bank organizational

structure and credit, In: P. Alessandrini, M. Fratianni, and A. Zazzaro (Eds.), The

Changing Geography of Banking and Finance, 54-74, Heidelberg: Springer-Verlag.

Cetorelli, N. and L.S. Goldberg (2011), Global banks and international shock transmission:

Evidence from the crisis, IMF Economic Review, 59(1), 41-76.

31

Chava, S. and A. Purnanandam (2011), The effect of banking crisis on bank-dependent

borrowers, Journal of Financial Economics, 99, 116-135.

Claessens, S. and N. Van Horen (2011), Trends in foreign banking. A database on foreign

bank ownership, mimeo.

Degryse, H. and P. van Cayseele (2000), Relationship lending within a bank-based system:

Evidence from European small business data, Journal of Financial Intermediation, 9,

90-109.

Degryse, H. and S. Ongena (2005), Distance, lending relationships, and competition, Journal

of Finance, 60, 231-266.

Dell’Ariccia, G. (2001), Asymmetric information and the structure of the banking industry,

European Economic Review, 45, 1957–1980.

De Haas, R. and N. Van Horen (2010), The crisis as a wake-up call. Do banks tighten

screening and monitoring standards during a financial crisis?, EBRD Working Paper

No. 117, European Bank for Reconstruction and Development, London.

De Haas, R. and I. Van Lelyveld (2010), Internal capital markets and lending by

multinational bank subsidiaries, Journal of Financial Intermediation, 19, 1-25.

Diamond, D.W. (1984), Financial intermediation and delegated monitoring, Review of

Economic Studies, 51(3), 393-414.

Diamond, D.W. (1991), Monitoring and reputation: The choice between bank loans and

directly placed debt, Journal of Political Economy, 99, 689-721.

Esty, B. (2004), When do foreign banks finance domestic projects? New evidence on the

importance of legal and financial systems, mimeo, Harvard Business School.

Giannetti, M. and Y. Yafeh (2008), Do cultural differences between contracting parties

matter? Evidence from syndicated bank loans, mimeo.

Gorton, G. and G. Pennachi (1995), Banks and loan sales: Marketing non-marketable assets,

Journal of Monetary Economics, 35, 389-411.

Greenbaum, S.I. and A.V. Thakor (1995), Contemporary Financial Intermediation, Dryden

Press, New York.

Hauswald, R. and R. Marquez (2006), Competition and strategic information acquisition in

credit markets, Review of Financial Studies, 19(3), 967-1000.

Herrmann, S. and D. Mihaljek (2010), The determinants of cross-border bank flows to EMs –

New empirical evidence on the spread of financial crises, mimeo.

Hofstede, G. (2001), Culture's consequences: Comparing values, behaviors, institutions, and

organizations across nations, 2nd ed., Thousand Oaks, CA: SAGE Publications.

32

Hoggarth, G., L. Mahadeva and J. Martin (2010), Understanding international bank capital

flows during the recent financial crisis, Financial Stability Paper No. 8, Bank of

England, London.

Houston, J., J. Itzkowitz, and A. Naranjo (2007), Borrowing beyond borders: The geography

and pricing of syndicated bank loans, mimeo.

Hauswald, R. and R. Marquez (2006), Competition and strategic information acquisition in

credit markets, Review of Financial Studies, 19, 967-1000.

Imai, M. and S. Takarabe (2011), Bank integration and transmission of financial shocks:

Evidence from Japan, American Economic Journal: Macroeconomics, 3, 155-183.

International Monetary Fund (2010), The fund’s role regarding cross-border capital flows,

Report prepared by the Strategy, Policy, and Review Department and the Legal

Department, November 15th.

Ivashina, V. and D. Scharfstein (2010), Loan syndication and credit cycles, American

Economic Review: Papers and Proceedings, 100(2), 1-8 .

Jaffee, D.M. and F. Modigliani (1971), A theory and test of credit rationing, American

Economic Review, 59, 850-872.

Jappelli, T. and M. Pagano (2002), Information sharing, lending and defaults: Cross-country

evidence, Journal of Banking and Finance, 26, 2017-2045.

Kanz, M. (2010), Bank structure and entrepreneurial finance: Experimental evidence from

small business loans in India, Harvard University, mimeo.

Khwaja, A.I. and A. Mian (2008), Tracing the impact of bank liquidity shocks, American

Economic Review, 98, 1413-1442.

Kose, M.A., E.S. Prasad, and M.E. Terrones (2003), How does globalization affect the

synchronization of business cycles?, American Economic Review, 93(2), 57-62.

Jappelli, T., M. Pagano, and M. Bianco (2005), Courts and banks: Effects of judicial

enforcement on credit markets, Journal of Money, Credit, and Banking, 37, 223-44.

Jeanneau, S. and M. Micu (2002), Determinants of international bank lending to emerging

market countries, BIS Working Paper No. 112, Bank for International Settlements,

Basel.

La Porta, R., F. López-de-Silanes, A. Shleifer and R.W. Vishny (1998), Law and finance,

Journal of Political Economy, 106(6), 1113-55.

Lederer, P. and A.P. Hurter (1986), Competition of firms: Discriminatory pricing and

location, Econometrica, 54, 623-640.

33

Levine, R. (1999), Law, finance, and economic growth, Journal of Financial Intermediation,

8, 8-35.

Liberti, J.M. and A.R. Mian (2009), Estimating the effect of hierarchies on information use,

Review of Financial Studies, 22(10), 4057-4090.

McGuire, P. and N. Tarashev (2008), Bank health and lending to EMs, BIS Quarterly Review,

December, 67-80.

Melitz, M.J. (2003), The impact of trade on intra-industry reallocations and aggregate

industry productivity, Econometrica, 71(6), 1695-1725.

Mian, A. (2006), Distance constraints: The limits of foreign lending in poor economies,

Journal of Finance, 61, 1465-1505.

Pagano, M. and T. Jappelli (1993), Information sharing in credit markets, Journal of Finance,

43, 1693-781.

Peek, J. and E. Rosengren (1997), The international transmission of financial shocks: The

case of Japan, American Economic Review, 87, 495-505.

Peek, J. and E. Rosengren (2000), Collateral damage: Effects of the Japanese bank crisis on

the United States, American Economic Review, 90, 30-45.

Petersen, M.A. and R.G. Rajan (1994), The benefits of firm-creditor relationships: Evidence

from small business data, Journal of Finance, 49, 3-37.

Petersen, M.A. and R.G. Rajan (2002), Does distance still matter? The information revolution

in small business lending, Journal of Finance, 57, 2533-2570.

Pongsaparn, R. and O. Unteroberdoerster (2011), Financial integration and rebalancing in

Asia, IMF Working Paper No. 243, International Monetary Fund, Washington, D.C.

Popov, A. and G. Udell (2010), Cross-border banking and the international transmission of

financial distress during the crisis of 2007-2008, ECB Working Paper No. 1203,

European Central Bank, Frankfurt.

Portes, R., H. Rey, and Y. Oh (2001), Information and capital flows: The determinants of

transactions in financial assets, European Economic Review, 45, 783-796.

Qian, J. and P.E. Strahan (2007), How laws and institutions shape financial contracts: The

case of bank loans, Journal of Finance, LXII(6), 2803-2834.

Rajan, R.G. (1992), Insiders and outsiders: The choice between informed and arm’s-length

debt, Journal of Finance, 47, 1367-400.

Rajan, R.G., Servaes, H., and L. Zingales (2000), The cost of diversity: The diversification

discount and inefficient investments, Journal of Finance, 55, 35-80.

34

Rey, P. and J.E. Stiglitz (1993), Short-term contracts as a monitoring device, NBER Working

Paper No. 4514, National Bureau of Economic Research, Cambridge MA.

Rose, A.K. and T. Wieladek (2011), Financial protectionism: the first tests, External MPV

Unit Discussion Paper No. 32, Bank of England, London.

Ruckes, M. (2004), Bank competition and credit standards, Review of Financial Studies, 17,

1073-1102.

Scharfstein, D.S. and J.C. Stein (2000), The dark side of internal capital markets: Divisional

rent-seeking and inefficient investment, Journal of Finance, 55, 2537-2564.

Schenone, C. (2010), Lending relationships and information rents: Do banks exploit their

information advantages?, Review of Financial Studies, 23(3). 1149-1199.

Schnabl, P. (2011), Financial globalization and the transmission of bank liquidity shocks:

Evidence from an emerging market, Journal of Finance, forthcoming.

Sharpe, S. (1990), Asymmetric information, bank lending and implicit contracts: A stylized

model of customer relationships, Journal of Finance, 45, 1069–1087.

Stein, J.C. (2002), Information production and capital allocation: Decentralized versus

hierarchical firms, Journal of Finance, LVII(5), 1891-1921.

Stiglitz, J. and A. Weiss (1981), Credit rationing in markets with imperfect information,

American Economic Review, 71, 393-410.

Sussman, O. and J. Zeira (1995), Banking and Development, CEPR Discussion Paper No.

1127, London.

Takáts, E. (2010), Was it credit supply? Cross-border bank lending to emerging market

economies during the financial crisis, BIS Quarterly Review, June, 49-56.

Tinbergen, J. (1962), Shaping the World Economy, The Twentieth Century Fund, New York.

Van Rijckeghem, C. and B. Weder (2001), Sources of contagion: Finance or trade?, Journal

of International Economics, 54, 293-308.

Van Rijckeghem, C. and B. Weder (2003), Spillovers through banking centres: A panel data

analysis of bank flows, Journal of International Money and Finance, 22(4), 483-509.

35

Figure 1 Change in cross-border lending after the Lehman Brothers collapse

This figure shows the distribution across destination countries of the change in cross-border syndicated lending inflows after the collapseof Lehman Brothers compared to the pre-crisis period. The pre-crisis period is July 2006-June 2007 and the post-Lehman Brothers periodis October 2008-September 2009. Each bar indicates the number of destination countries that experienced a change in bank lendingwithin the percentage bracket on the horizontal axis. For instance, there were 11 countries to which cross-border syndicated bank lendingdeclined by between 25 and 50 per cent while there were only 5 countries that experienced an increase in cross-border syndicatedlending by between 25-50 per cent.

3

1412

11

6

2

5

1 1

4

02468

10121416

Number of destination countries

lending stop

0-25% 25-50% 75-99% 50-75%

decrease in lending

> 100%0-25% 25-50% 75-99% 50-75%

increase in lending

36

This figure compares the change in cross-border syndicated lending to a country (horizontal axis) with the change in total syndicatedlending (cross-border plus domestic syndicated lending) in that country. Lending change is the percentage change in lending in the post-Lehman Brothers period compared to the pre-crisis period. The pre-crisis period is July 2006-June 2007 and the post-Lehman Brothersperiod is October 2008-September 2009. The left-hand pane shows all 59 destination countries in our dataset whereas the right-hand panezooms in on those countries that experienced a decline in both cross-border and total syndicated lending. Countries that experienced apercentage change in domestic lending that was exactly equal to the percentage change in cross-border lending are on the 45º line.Countries where domestic lending shrank faster (slower) than cross-border lending are to the right (left) of this line.

QAT

AUT

MYS

GBRESP

TWN

ITAVNM

AZE

IRL

PRT

KOR

NZL

HKGUSA

ARG

CZE

CHESVN

CAN

TUR

EGYGRC

AUSARE

NORDEURUS

NGASWE

BGRUKR

UKROMN

SAU

CHLTHA

HUN

INDBRAKAZ

FRABEL

ZAF

NLD

-100%

-80%

-60%

-40%

-20%

0%

-100% -80% -60% -40% -20% 0%-100%

-50%

0%

50%

100%

150%

200%

-100% -50% 0% 50% 100% 150% 200%

Tot

al L

endi

ng IND

KWTPER

DNKMEX

FIN ROU

JPN

LUX

HRV

CHN

PHL

POL

Cross-border lending

Figure 2A comparison of cross-border and total syndicated lending

37

Unit Obs Mean Median St Dev Min MaxDependent variablesSudden stop Dummy 1,913 0.44 0.00 0.50 0 1Change in cross-border lending (volume) Log change 1,913 -2.46 -2.20 2.30 -8.25 5.25

Closeness variablesDistance Km (in logs) 1,913 7.99 8.15 1.11 4.63 9.61Experience No. loans (in logs) 1,913 2.49 2.48 1.34 0 7.73Subsidiary Dummy 1,913 0.18 0.00 0.39 0 1Domestic lenders Share 1,913 0.34 0.30 0.25 0 1

OtherExposure Share 1,913 0.01 0.00 0.01 0 0.07

Table 1 Summary statistics

This table shows summary statistics for our main variables. Table A3 in the Appendix contains information on all variabledefinitions, the units and period of measurement, and the data sources.

38

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]Distance 0.099*** 0.049*** 0.045*** -0.406*** -0.262*** -0.239***

(0.000) (0.002) (0.003) (0.000) (0.000) (0.001)Experience -0.133*** -0.087*** -0.085*** -0.078*** 0.438*** 0.310*** 0.296*** 0.252**

(0.000) (0.000) (0.000) (0.001) (0.000) (0.000) (0.000) (0.014)Subsidiary -0.147*** -0.044* -0.041* -0.035 0.506*** 0.180 0.160 0.164

(0.000) (0.083) (0.099) (0.155) (0.000) (0.145) (0.209) (0.260)Domestic lenders -0.547*** -0.280*** -0.282*** -0.235** 1.428*** 0.371 0.389 0.603

(0.000) (0.000) (0.000) (0.015) (0.000) (0.234) (0.206) (0.134)Trade -0.007 0.068

(0.625) (0.270)Bank FDI -0.048* 0.292**

(0.089) (0.027)Exposure -1.293 0.782 -2.187 -0.732 1.718*** 1.925*** 1.838** -6.291 -11.592*** -1.983 -5.143 -15.078*** -16.346*** -14.504***

(0.383) (0.249) (0.172) (0.563) (0.009) (0.003) (0.033) (0.173) (0.000) (0.700) (0.276) (0.000) (0.000) (0.000)Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesDestination country FE Yes Yes Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes NoCountry-pair FE No No No No No No Yes No No No No No No YesObservations 1,913 1,913 1,913 1,913 1,913 1,913 1,599 1,875 1,875 1,875 1,875 1,875 1,875 1,564R-squared 0.407 0.432 0.396 0.421 0.447 0.448 0.689 0.338 0.345 0.326 0.333 0.355 0.357 0.642

Table 2

Sudden stop Volume

Closeness and cross-border bank lending stability - Baseline resultsThis table shows estimations to explain the decline in cross-border lending from bank i to destination country j after the Lehman Brothers default. Table A3 in the Appendix contains definitions of allvariables. We use a linear probability model and an OLS model for the Sudden stop and Volume regressions, respectively. All specifications include bank and destination country fixed effects, with theexception of those in columns [7] and [14] which contain bank and country-pair fixed effects. Standard errors are heteroskedasticity robust and clustered by bank. Coefficients are marginal effects. Robustp-values appear in parentheses and ***, **, * correspond to the one, five and ten per cent level of significance, respectively.

39

[1] [2] [3] [4] [5] [6]Distance 0.050*** 0.025* 0.056**

(0.000) (0.080) (0.023)Experience with firm -0.134*** -0.113*** -0.202***

(0.000) (0.000) (0.000)Subsidiary -0.010 0.025 0.027

(0.699) (0.369) (0.543)Domestic lenders -0.448*** -0.390*** -0.590***

(0.000) (0.000) (0.000)Arranger -0.189*** -0.119*** -0.199*** -0.172*** -0.105*** -0.160***

(0.000) (0.000) (0.000) (0.000) (0.001) (0.001)Exposure -0.780 -1.057 -1.284* 0.284 0.444 0.701

(0.295) (0.142) (0.087) (0.712) (0.565) (0.603)Estimation method OLS OLS OLS OLS OLS LogitObservations 2,224 2,224 2,224 2,201 2,201 1,710(Pseudo) R-squared 0.448 0.451 0.444 0.457 0.464 0.290

Table 3Closeness and cross-border bank lending stability

Termination

Firm-level evidenceThis table shows firm-level estimations to explain the variable Termination , the probability that bank i discontinued lending to firm k after the Lehman Brothers default. The sample includes all firms thatborrowed from at least two different lenders in our sample during the pre-crisis period. All specificationsinclude bank and firm fixed effects. Table A3 in the Appendix contains definitions of all variables.Standard errors are heteroskedasticity robust and clustered by bank. Coefficients are marginal effects.Columns [1] to [5] show linear probability OLS regressions and column [6] a logit regression. Robust p-values appear in parentheses and ***, **, * correspond to the one, five and ten per cent level ofsignificance, respectively.

40

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10]Distance 0.022** 0.002 -0.185*** -0.116**

(0.027) (0.836) (0.000) (0.030)Distance * Original 0.078*** 0.046*** -0.221*** -0.144*

(0.000) (0.005) (0.004) (0.078)Experience -0.049*** -0.033*** 0.186*** 0.055

(0.000) (0.003) (0.001) (0.371)Experience * Original -0.084*** -0.054*** 0.252*** 0.252***

(0.000) (0.002) (0.001) (0.004)Subsidiary -0.058*** -0.027* 0.174* -0.005

(0.000) (0.077) (0.082) (0.960)Subsidiary * Original -0.093*** -0.021 0.363** 0.218

(0.003) (0.506) (0.027) (0.194)Domestic lenders -0.193*** -0.118** 1.123*** 0.919***

(0.000) (0.030) (0.000) (0.001)Domestic * Original -0.354*** -0.161** 0.304 -0.554

(0.000) (0.045) (0.368) (0.147)Exposure -1.452** 0.226 -1.290* -0.752 0.578 0.452 -3.751 2.429 -2.441 -5.937

(0.031) (0.723) (0.055) (0.257) (0.353) (0.925) (0.437) (0.627) (0.601) (0.211)Exposure * Original 0.158 0.555 -0.869 0.020 1.148 -6.743 -7.842 -4.594 -2.701 -9.204

(0.893) (0.552) (0.478) (0.985) (0.210) (0.265) (0.168) (0.471) (0.651) (0.103)Observations 3,313 3,313 3,313 3,313 3,313 3,252 3,252 3,252 3,252 3,252R-squared 0.485 0.504 0.478 0.496 0.515 0.547 0.551 0.541 0.546 0.557

Table 4

Volume

Closeness and cross-border bank lending during normal times: a placebo test

Sudden stop

This table shows estimations to analyze to what extent the impact of closeness on changes in cross-border lending was stronger in or unique to thepost-Lehman Brothers period. The sample includes observations from our original baseline period as well as a non-crisis comparison ('placebo')period that preceded the original baseline period. The original baseline period comprises three sub-periods: pre-crisis (July 06-June 07), anintermediate period (July 07-September 08), and the post-Lehman period (October 08-September 09). The placebo period includes the sub-periodsApril 04-March 05; April 05-June 06; and July 06-June 07. The dependent variables are Sudden stop and Volume. Table A3 in the Appendixcontains definitions of all variables. All specifications include destination country and bank fixed effects. We use a linear probability model and anOLS model for the Sudden stop and Volume regressions, respectively. Standard errors are heteroskedasticity robust and clustered by bank. Robust p-values appear in parentheses and ***, **, * correspond to the one, five and ten per cent level of significance, respectively.

41

Base Logit Incl local lending

subs

> 25 pct number pre-crisis loans

Cluster country

Base All obs Incl local lending

subs

> 25 pct number pre-crisis loans

Cluster country

Model Extreme rule

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]Distance 0.049*** 0.095*** 0.042*** 0.062*** 0.049*** -0.262*** -0.244*** -0.202*** -0.308*** -0.262*** -0.202*** -0.254***

(0.002) 0.001 (0.001) (0.000) (0.003) (0.000) (0.000) (0.001) (0.000) (0.000) (0.005) (0.000)Experience -0.087*** -0.139*** -0.090*** -0.073*** -0.087*** 0.310*** 0.328*** 0.317*** 0.355*** 0.310*** 0.184*** 0.309***

(0.000) 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.007) (0.000)Subsidiary -0.044* -0.105** -0.055** -0.008 -0.044* 0.180 0.253* 0.247* -0.005 0.180 0.162 0.153

(0.083) 0.022 (0.034) (0.777) (0.071) (0.145) (0.067) (0.057) (0.971) (0.143) (0.295) (0.249)Domestic lenders -0.280*** -0.500*** -0.291*** -0.270*** -0.280*** 0.371 0.364 0.369 0.476 0.371 -0.258 0.330

(0.000) 0.000 (0.000) (0.000) (0.000) (0.234) (0.185) (0.170) (0.137) (0.304) (0.416) (0.241)Exposure 1.718*** -0.760 1.965*** 0.632 1.718*** -15.078*** -15.288*** -13.784*** -12.565*** -15.078*** -6.761* -14.783***

(0.009) 0.835 (0.003) (0.518) (0.006) (0.000) (0.000) (0.000) (0.007) (0.000) (0.097) (0.000)Observations 1,913 1,786 1,918 1,338 1,913 1,875 1,913 1,880 1,313 1,875 1,875 1,823(Pseudo-) R-squared 0.447 0.398 0.451 0.470 0.447 0.355 0.348 0.356 0.435 0.355 0.263 0.328

Table 5

Sudden stop Volume

Closeness and cross-border bank lending stability - Robustness checksThis table shows robustness tests for our baseline regressions to explain the decline in cross-border lending from bank i to destination country j after the Lehman Brothers default. Table A3in the Appendix provides variable definitions. Columns [1] and [6] replicate the baseline results from Table 2. Column [2] shows a logit regression instead of a linear probability OLS.Columns [3] and [8] include syndicated lending through local subsidiaries. Columns [4] and [9] exclude the quartile of bank-country pairs with the smallest number of pre-crisis loans. Incolumns [5] and [10] standard errors are clustered by country instead of by bank. Column [7] shows a regression on the basis of the full sample without excluding or winsorizing outliers.Column [11] shows a regression where the allocation of the loans over the syndicate members is based on predicted values from a regression model. Column [12] shows a regression wherethe dependent variable is based on an allocation rule where one randomly chosen lender receives (almost) all of the loan whereas the other lenders each receive only 1 per cent of the loan. Allspecifications include destination country and bank fixed effects. Coefficients are marginal effects. Robust p-values appear in parentheses and ***, **, * correspond to the one, five and tenper cent level of significance, respectively.

42

X →Sudden stop Volume Sudden stop Volume Sudden stop Volume Sudden stop Volume Sudden stop Volume

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10]Distance 0.042*** -0.245*** 0.048*** -0.306*** 0.044*** -0.298*** 0.049** -0.293*** 0.059*** -0.255***

(0.009) (0.002) (0.001) (0.000) (0.009) (0.000) (0.012) (0.002) (0.000) (0.000)Distance * X 0.021 -0.004 -0.001 0.110 0.014 0.068 0.002 0.046 0.008 -0.049

(0.402) (0.973) (0.960) (0.298) (0.517) (0.481) (0.935) (0.697) (0.570) (0.515)Experience -0.070*** 0.319*** -0.084*** 0.396*** -0.083*** 0.343*** -0.060*** 0.259** -0.080*** 0.361***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.005) (0.011) (0.000) (0.000)Experience * X -0.041* -0.067 -0.020 -0.137 -0.015 -0.052 -0.054** 0.130 -0.041*** -0.005

(0.072) (0.532) (0.351) (0.173) (0.536) (0.613) (0.035) (0.247) (0.002) (0.943)Subsidiary 0.003 0.060 -0.007 0.002 -0.007 0.021 0.017 -0.138 -0.033 0.067

(0.937) (0.737) (0.814) (0.989) (0.809) (0.887) (0.687) (0.512) (0.262) (0.645)Subsidiary * X -0.098** 0.239 -0.110** 0.507** -0.093** 0.382 -0.100* 0.494* 0.014 -0.100

(0.050) (0.341) (0.039) (0.050) (0.047) (0.130) (0.060) (0.058) (0.727) (0.629)Domestic lenders -0.358*** 1.141** -0.205** -0.363 -0.255** 0.045 -0.135 -0.356 -0.156** 0.030

(0.001) (0.033) (0.019) (0.352) (0.013) (0.907) (0.393) (0.622) (0.015) (0.919)Domestic lenders * X 0.143 -0.970 -0.122 1.278** -0.014 0.608 -0.124 0.662 -0.029 0.191

(0.272) (0.119) (0.292) (0.016) (0.906) (0.222) (0.459) (0.387) (0.659) (0.549)Exposure 1.381** -14.895*** 1.380** -14.000*** 1.598** -13.813*** 0.925 -10.142** 0.707 -10.686***

(0.032) (0.000) (0.045) (0.000) (0.012) (0.000) (0.221) (0.014) (0.381) (0.003)Exposure * X 0.299 -9.950 0.193 -2.609 -0.830 -9.344** -0.521 -2.881 0.615 1.568

(0.783) (0.105) (0.815) (0.505) (0.314) (0.020) (0.545) (0.422) (0.460) (0.684)Observations 1,913 1,875 1,913 1,875 1,913 1,875 1,810 1,776 2,567 2,517R-squared 0.449 0.357 0.450 0.359 0.449 0.357 0.449 0.359 0.423 0.312

Table 6Closeness, institutional quality, and cross-border bank lending stability

First-time borrowersLegal originEmerging market Credit information Creditor protection

This table shows estimations to explain the decline in cross-border lending from bank i to destination country j after the Lehman Brothers default. Table A3 in the Appendix containsthe variable definitions. Columns [1] and [2] test whether closeness is more important in emerging markets than in advanced countries. Emerging markets are all countries except high-income OECD countries. Although Slovenia and South-Korea were recently reclassified as high-income countries we still consider them as emerging markets. Columns [3] and [4] testwhether closeness is more important in countries where the public availability of credit information is less than in the median country in our sample. Columns [5] and [6] test whethercloseness is more important in countries where the quality of creditor protection is below the median quality in our country sample. Columns [7] and [8] test whether closeness is moreimportant in countries with a legal system of non-English origin. Columns [9] and [10] test whether closeness is more important when lending to first-time compared to repeatborrowers. Here the data contain two observations for each bank i -country j pair: one where the dependent variable is a dummy that indicates whether there was a sudden stop inlending from bank i to first-time borrowers in country j and a similar one for lending to repeat borrowers. All specifications include destination country and bank fixed effects. We use alinear probability and an OLS model for the Sudden stop and Volume regressions, respectively. Standard errors are heteroskedasticity robust and clustered by bank. Robust p-valuesappear in parentheses and ***, **, * correspond to the one, five and ten per cent level of significance, respectively.

43

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]Distance 0.049*** 0.041*** 0.049*** 0.046*** 0.049*** 0.042*** -0.262*** -0.228*** -0.262*** -0.250*** -0.262*** -0.234***

(0.002) (0.010) (0.001) (0.003) (0.002) (0.008) (0.000) (0.001) (0.000) (0.000) (0.000) (0.001)Common language -0.100*** -0.060 0.429*** 0.250

(0.004) (0.142) (0.010) (0.226)Dif creditor protection 0.147*** 0.040** -0.362*** -0.277***

(0.000) (0.039) (0.000) (0.000)Dif credit information 0.017 0.042 0.055 -0.125

(0.767) (0.315) (0.742) (0.308)Dif legal origin 0.070*** 0.048* -0.303*** -0.213

(0.003) (0.079) (0.010) (0.143)Experience -0.087*** -0.087*** -0.087*** -0.087*** -0.086*** -0.087*** 0.310*** 0.307*** 0.310*** 0.306*** 0.310*** 0.305***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Subsidiary -0.044* -0.038 -0.044* -0.044* -0.039 -0.044* 0.180 0.155 0.180 0.161 0.180 0.152

(0.083) (0.130) (0.083) (0.083) (0.119) (0.083) (0.145) (0.214) (0.145) (0.199) (0.145) (0.226)Domestic lenders -0.280*** -0.278*** -0.280*** -0.280*** -0.267*** -0.280*** 0.371 0.362 0.365 0.318 0.371 0.329

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.234) (0.245) (0.240) (0.304) (0.234) (0.286)Exposure 1.718*** 1.909*** 1.718*** 1.718*** 1.859*** 1.718*** -15.078*** -15.848*** -15.075*** -15.670*** -15.078*** -15.943***

(0.009) (0.003) (0.009) (0.009) (0.004) (0.009) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)Observations 1,913 1,913 1,913 1,913 1,913 1,913 1,875 1,875 1,875 1,875 1,875 1,875R-squared 0.447 0.449 0.447 0.447 0.450 0.447 0.355 0.357 0.355 0.357 0.355 0.358

Sudden stop Volume

Table 7Distance and cross-border bank lending stability

This table uses different distance measures to explain the decline in cross-border lending from bank i to destination country j after the Lehman Brothers default. The dependent variablesare Sudden stop and Volume . Table A3 in the Appendix contains definitions of all variables. All specifications include destination country and bank fixed effects. We use a linearprobability model and an OLS model for the Sudden stop and Volume regressions, respectively. Standard errors are heteroskedasticity robust and clustered by bank. Robust p-valuesappear in parentheses and ***, **, * correspond to the one, five and ten per cent level of significance, respectively.

44

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10]Distance -0.279*** -0.197** -0.047** -0.020

(0.000) (0.019) (0.025) (0.379)Experience 0.288*** 0.179 0.074** 0.028

(0.001) (0.136) (0.013) (0.383)Subsidiary 0.315** 0.167 0.006 -0.053

(0.030) (0.267) (0.902) (0.364)Domestic lenders 0.923** 0.167 0.480*** 0.421***

(0.014) (0.735) (0.001) (0.009)Exposure 7.461* 1.366 9.446** 7.408 -0.293 -1.221 -3.319* -0.433 -2.963* -3.645**

(0.089) (0.770) (0.047) (0.152) (0.947) (0.398) (0.051) (0.760) (0.052) (0.028)Observations 956 956 956 956 956 1,043 1,043 1,043 1,043 1,043R-squared 0.440 0.439 0.434 0.436 0.444 0.335 0.338 0.332 0.343 0.345

MaturitySpread

Table 8Closeness and the change in spreads and maturities of cross-border loans

after the Lehman Brothers collapseThis table shows estimations to explain the change in the spreads and maturity of cross-border lending from bank i to destination country j after theLehman Brothers default. Table A3 in the Appendix contains definitions of all variables. All specifications include destination country and bank fixedeffects. We use an OLS model with heteroskedasticity robust standard errors that are clustered by bank. Coefficients are marginal effects. Robust p-valuesappear in parentheses and ***, **, * correspond to the one, five and ten per cent level of significance, respectively.

45

Market share (ppts.)

Name Pre- crisis

Post- Leh man

Pre-crisis Post-Leh man

Pre- crisis

Post- Leh man

Pre- crisis

Australia ANZ 37 43 6,597 4,916 91 73 0.37Australia Commonwealth Bank of Australia 32 24 5,215 2,530 59 28 0.29Australia National Australia Bank 50 12 10,207 752 79 15 0.57Australia Westpac 30 20 5,156 2,118 61 34 0.29Austria BAWAGPSK 90 100 380 236 17 3 0.02Austria Erste Group Bank AG 93 83 5,099 1,063 185 16 0.28Austria Hypo Alpe-Adria-Bank 100 50 1,253 67 18 1 0.07Austria Oesterreichische Volksbanken AG 98 88 1,379 323 31 6 0.08Austria RZB 89 55 9,540 1,957 312 37 0.53Bahrain Arab Banking Corp - BSC 96 41 1,684 195 28 5 0.09Bahrain Gulf International Bank BSC 100 100 2,185 75 32 1 0.12Belgium Dexia 63 74 7,276 2,573 58 23 0.40Belgium Fortis 80 74 32,985 10,363 473 118 1.83Belgium KBC 86 81 15,402 4,045 219 54 0.85Canada BMO Capital Markets 50 41 12,762 6,876 206 131 0.71Canada CIBC World Markets 46 10 6,898 904 70 16 0.38Canada RBC Capital Markets 52 57 14,099 10,962 140 97 0.78Canada Scotia Capital 74 68 31,210 20,390 279 178 1.73Canada TD Securities Inc 33 38 5,819 6,640 63 71 0.32China Agricultural Bank of China 67 3 574 65 17 1 0.03China Bank of China Ltd 79 63 8,647 5,555 155 54 0.48China Bank of Communications Co Ltd 79 7 1,478 893 36 13 0.08China China Construction Bank Corp 79 5 2,163 635 58 14 0.12China China Merchants Securities Co Ltd 13 3 50 397 6 9 0.00China CITIC Group 36 5 322 568 19 7 0.02China Industrial & Commercial Bank of China 69 14 2,365 2,042 54 29 0.13Denmark Danske Bank 79 43 12,023 2,430 107 22 0.67Egypt National Bank of Egypt 79 59 547 86 47 1 0.03France Banque Federative du Credit Mutuel 53 73 11,008 7,384 102 48 0.61France BNP Paribas 80 85 85,012 49,411 744 432 4.71France Calyon 67 74 57,164 31,448 501 283 3.17France CASDEN Banque Populaire 20 25 497 190 23 4 0.03France Natixis 65 73 28,675 12,453 335 145 1.59France SG Corporate & Investment Banking 76 82 46,586 30,521 445 264 2.58Germany Commerzbank Group 71 68 54,574 19,022 661 143 3.02Germany Deutsche Bank 91 88 110,371 41,424 530 268 6.11Germany DZ Bank 73 52 11,401 4,548 194 54 0.63Germany NordLB 70 62 5,409 1,805 95 22 0.30Germany WGZ 46 11 1,102 71 61 3 0.06Greece Alpha Bank 65 100 1,692 39 84 1 0.09Greece National Bank of Greece 46 95 996 662 79 21 0.06Hong Kong Bank of East Asia 54 67 657 670 28 15 0.04India ICICI Bank 53 57 1,052 511 34 5 0.06

Appendix Table A1List of international lenders

Share of cross-border in total

lending (percent)

Volume of cross-border lending

(USD m)

Number of cross-border

loans

This table lists all 117 banks in our sample, ordered by country of incorporation. Pre-crisis refers to the period July 2006 to June 2007 and post-Lehman to the period October 2008-September 2009. Share of cross-border in total lending measures the volume of cross-border syndicatedlending of the bank divided by the total volume of syndicated lending by that bank (in per cent). Volume of cross-border lending measures the totalvolume of cross-border syndicated lending by the bank in US dollar millions. Number of cross-border loans measures the number of cross-bordersyndications the bank took part in. Market share measures the market share of the bank in 2006 in the global market for cross-border syndicatedlending (in percentage points).

46

Market share (ppts.)

Name Pre- crisis

Post- Leh man

Pre-crisis Post-Leh man

Pre- crisis

Post- Leh man

Pre- crisis

India SBI Capital Markets Ltd 54 72 2,135 1,754 78 28 0.12Ireland Allied Irish Banks plc 97 92 16,820 2,567 203 49 0.93Ireland Bank of Ireland 94 96 18,206 4,590 193 58 1.01Israel Bank Hapoalim BM 100 100 2,064 106 62 2 0.11Israel Bank Leumi Le-Israel BM 49 6 672 28 14 1 0.04Israel Israel Discount Bank Ltd 87 52 755 250 21 7 0.04Italy Gruppo Banco Popolare di Verona e Novara 40 1 1,779 16 35 1 0.10Italy Intesa Sanpaolo 69 74 19,101 13,930 259 100 1.06Italy Monte dei Paschi 39 16 1,930 554 56 13 0.11Italy UniCredit Group 70 57 36,206 9,451 456 97 2.00Japan Mitsubishi UFJ Financial Group 48 27 56,086 36,810 599 321 3.11Japan Mizuho 52 18 51,672 16,305 540 147 2.86Japan Nomura 70 58 11,474 976 44 6 0.64Japan Norinchukin Bank Ltd 22 5 1,943 549 20 8 0.11Japan Sumitomo Mitsui Financial Group, Inc 44 16 39,685 15,419 449 172 2.20Jordan Arab Bank Group 100 100 2,822 870 53 9 0.16Luxembourg BCEE 68 16 642 48 30 1 0.04Macao Tai Fung Bank Ltd 100 100 206 55 16 3 0.01Malaysia CIMB Group 22 65 335 112 21 3 0.02Malaysia Maybank Investment Bank Bhd 79 57 1,320 325 35 9 0.07Netherlands ING 86 81 45,900 17,530 468 163 2.54Netherlands NIBC Bank 84 44 3,366 396 35 8 0.19Netherlands Rabobank 79 78 17,532 9,109 231 120 0.97Norway DnB NOR Bank ASA 70 66 11,327 3,455 112 37 0.63Oman Bank Muscat SAOG 73 100 613 10 26 1 0.03Portugal Banco BPI 96 55 2,439 306 21 3 0.14Portugal Banco Espirito Santo de Investimento 92 65 5,938 738 59 15 0.33Portugal Caixa Geral de Depositos SA - CGD 74 89 5,528 2,874 87 19 0.31Qatar Commercial Bank of Qatar QSC 66 0 400 0 12 0 0.02Qatar Doha Bank QSC 74 23 202 67 18 3 0.01Qatar Qatar National Bank 66 2 1,190 10 23 1 0.07Singapore DBS 97 93 8,417 4,407 147 88 0.47Singapore Oversea-Chinese Banking Corp Ltd 84 73 2,188 908 65 25 0.12Singapore UOB 87 83 7,158 1,311 97 27 0.40South Africa Standard Bank 69 92 2,308 1,592 98 14 0.13Spain BBVA 69 58 23,120 13,884 224 103 1.28Spain Banco Santander SA 46 53 19,671 15,381 153 88 1.09Spain Caja Madrid 45 45 8,294 3,392 44 16 0.46Sweden Nordea Bank AB 73 81 14,440 8,232 140 61 0.80Sweden SEB 67 73 8,384 4,816 70 38 0.46Sweden Svenska Handelsbanken AB 82 89 8,031 4,460 54 29 0.44Sweden Swedbank Markets 65 77 2,112 1,398 44 8 0.12Switzerland Credit Suisse 95 88 81,554 26,640 412 149 4.52Switzerland UBS 93 85 47,557 21,462 330 159 2.63Taiwan Bank of Taiwan 54 40 1,646 738 70 17 0.09Taiwan Cathay United Bank Co Ltd 37 8 713 72 28 7 0.04Taiwan Chang Hwa Commercial Bank Ltd 68 50 2,898 1,384 63 28 0.16Taiwan Chinatrust Commercial Bank 19 36 404 577 18 21 0.02Taiwan First Commercial Bank Co Ltd 58 66 1,905 2,226 64 21 0.11Taiwan Fubon Financial Holding Co Ltd 30 30 1,139 627 27 9 0.06Taiwan Hua Nan Commercial Bank Ltd 34 29 874 381 44 10 0.05Taiwan Shanghai Commercial & Savings Bank 30 7 223 18 25 2 0.01

Appendix Table A1- cont'dShare of cross-border in total

lending (percent)

Volume of cross-border lending

(USD m)

Number of cross-border

loans

47

Market share (ppts.)

Name Pre- crisis

Post- Leh man

Pre-crisis Post-Leh man

Pre- crisis

Post- Leh man

Pre- crisis

Taiwan Mega International Commercial Bank 59 54 2,417 1,504 92 30 0.13Taiwan Taiwan Cooperative Bank 23 11 337 164 28 10 0.02Thailand Bangkok Bank 70 17 610 70 36 6 0.03Turkey Turkiye Garanti Bankasi 100 100 945 68 45 2 0.05UAE Emirates NBD PJSC 48 19 1,426 257 56 2 0.08UAE Mashreqbank PSC 74 3 1,407 14 48 1 0.08UK RBS / ABN AMRO 73 79 139,710 53,457 1,014 407 7.74UK Barclays Capital 54 62 76,230 31,854 438 220 4.22UK HSBC 70 79 51,119 36,294 588 347 2.83UK Lloyds Banking Group 54 60 30,000 12,200 305 103 1.66UK NM Rothschild 88 100 2,581 11 28 1 0.14UK Standard Chartered Bank 79 91 15,361 9,274 299 148 0.85US Bank of America - Merrill Lynch 16 11 36,096 9,881 232 112 2.00US Bank of New York Mellon Corp 5 5 1,549 695 52 16 0.09US Citi 41 32 84,845 27,222 560 171 4.70US Comerica Bank 10 8 1,397 618 28 14 0.08US GE Capital Markets 19 22 8,001 3,248 74 23 0.44US Goldman Sachs 29 16 23,971 4,053 73 18 1.33US JPMorgan 21 18 55,034 18,313 275 114 3.05US Morgan Stanley 42 23 27,756 4,935 89 35 1.54US PNC Bank 42 26 14,706 5,834 294 120 0.81US Wells-Wachovia 6 4 7,692 2,532 138 41 0.43

Share of cross-border in total

lending (percent)

Volume of cross-border lending

(USD m)

Number of cross-border

loans

Appendix Table A1- cont'd

48

Country

Pre-crisisPost- Leh man

Pre-crisis Post-Leh man

Pre-crisis Post-Leh man

Pre-crisis Post-Leh man

Argentina 432 623 3 4 4 13 10 10Australia 41,608 19,628 92 80 375 275 40 47Austria 5,554 337 8 4 39 14 21 7Azerbaijan 758 307 8 3 60 12 13 8Belgium 35,552 7,785 33 14 229 64 45 30Brazil 30,849 3,050 32 16 191 51 32 24Bulgaria 400 48 4 2 35 2 10 1Canada 58,932 29,215 181 145 594 380 45 54Chile 4,485 862 16 5 117 14 24 11China 13,046 3,764 74 41 470 126 50 37Croatia 453 790 2 5 16 18 13 11Czech Republic 2,574 900 12 3 49 7 12 5Denmark 29,203 17,220 32 9 192 33 42 23Egypt, Arab Rep. 2,081 1,360 8 5 39 31 23 20Finland 9,352 8,269 19 16 106 68 30 26France 113,117 26,183 182 47 821 160 62 39Germany 134,283 51,265 132 31 788 197 64 49Greece 8,260 1,186 31 4 114 14 29 12Hong Kong, China 11,346 5,113 68 29 368 146 53 40Hungary 6,209 527 10 2 70 16 20 14Iceland 6,998 5,146 17 1 202 11 38 10India 17,348 2,712 85 22 682 53 67 26Indonesia 2,705 4,105 28 19 115 55 29 23Ireland 6,799 4,427 19 20 100 39 24 15Italy 33,901 27,830 79 61 287 175 41 36Japan 20,024 10,698 94 31 208 113 32 28Kazakhstan 11,137 673 31 3 345 16 60 15Korea, Rep. 10,097 4,948 51 27 241 109 50 30Kuwait 2,479 3,311 13 7 83 19 40 10Latvia 2,064 12 122 32 Luxembourg 28,037 44,588 16 10 161 108 46 38Malaysia 4,494 1,217 21 10 68 18 22 10Mexico 16,176 7,061 35 18 213 109 34 32Netherlands 76,906 14,375 68 27 465 153 61 48New Zealand 8,585 5,380 44 32 115 92 12 20Nigeria 2,452 811 7 7 34 12 8 6Norway 16,858 5,222 65 24 260 53 45 18Oman 591 5 19 19 Peru 1,096 680 5 4 46 8 7 7Philippines 693 1,432 9 7 66 41 20 19Poland 2,588 2,663 11 5 58 26 20 14

Appendix Table A2Overview of destination countries

Volume of cross-border lending

(USD m)

Number of cross-border loans

Number of active banks

Number of cross-border loan portions

This table lists all 60 destination countries in our sample. Pre-crisis refers to the period July 2006 to June 2007 and post-Lehman to theperiod October 2008-October 2009. Volume of cross-border lending measures the total volume of cross-border syndicated lending tothe country by the banks in our sample in US dollar millions. Number of cross-border loans measures the number of cross-borderloans to the country in which at least one of the banks in our sample was active. Number of cross-border loan portions measures thetotal number of individual loan portions provided by the banks in our sample to the country (e.g. one loan with 5 lenders of which 3foreign lenders implies three loan portions). Number of active banks measures the number of different banks that were at least 3 timesactive as cross-border lenders in the country in the pre-crisis period.

This table lists all 59 destination countries in our sample. Pre-crisis refers to July 2006-June 2007 and post-Lehman to October 2008-September 2009. Volume of cross-border lending measures the total volume of cross-border syndicated lending to the country by thebanks in our sample in US dollar millions. Number of cross-border loans measures the number of cross-border loans to the country inwhich at least one of the banks in our sample was active. Number of cross-border loan portions measures the total number ofindividual loan portions provided by the banks in our sample to the country (e.g. one loan with 5 lenders of which 3 foreign lendersimplies three loan portions). Number of active banks measures the number of different banks that were at least 3 times active as cross-border lenders in the country in the pre-crisis period.

49

Country

Pre-crisisPost- Leh man

Pre-crisis Post-Leh man

Pre-crisis Post-Leh man

Pre-crisis Post-Leh man

Portugal 3,931 2,548 7 5 71 27 25 16Qatar 6,963 4,470 14 7 106 37 24 19Romania 1,753 709 16 4 87 16 23 12Russian Federation 64,166 12,236 128 20 1,098 120 76 33Saudi Arabia 9,799 7 71 28 Slovenia 2,352 1,487 9 6 79 38 22 19South Africa 11,700 2,993 16 7 138 36 31 29Spain 96,521 27,389 94 54 440 220 45 36Sweden 26,519 4,895 38 11 190 30 41 15Switzerland 42,966 19,022 36 15 324 154 55 46Taiwan, China 4,321 1,120 35 47 121 73 24 18Thailand 2,087 315 13 5 53 20 27 15Turkey 24,626 8,880 54 18 665 227 69 49Ukraine 4,464 236 42 4 273 10 37 7United Arab Emirates 24,837 3,963 41 7 319 22 53 16United Kingdom 167,532 46,627 256 82 1,048 392 71 72United States 530,075 302,361 1,627 917 4,457 2,744 81 82Vietnam 897 465 7 5 14 14 5 13

Appendix Table A2 - cont'dVolume of cross-border lending

(USD m)

Number of cross-border loans

Number of cross-border loan portions

Number of active banks

50

Variable nameMeasurement

period Unit Description Source

Data on bank-destination country pairs (# banks = 117 and # of destination countries is 59)Sudden stop Jul 06-Sept 09 0/1 Dummy that is 1 if bank i stopped lending to country j in the post-Lehman Brothers period Loan AnalyticsVolume Jul 06-Sept 09 Log change Log change in the amount of cross-border lending by bank i to country j post-Lehman Brothers compared to pre-crisis period Loan AnalyticsDistance - Log km Distance in km between bank i and country j according to the great circle distance formula (in log) CIA World Factbook 2005Subsidiary End 07 0/1 Dummy variable that is 1 if bank i majority owns a bank subsidiary in country j BankScope

Domestic lenders Jan 00-Jul 07 ShareNumber of different domestic lenders (banks, insurance companies, etc.) in country j with whom bank i has cooperated in a syndicate between 2000 and July2007 as a proportion of all domestic lenders Loan Analytics

Experience Jan 00-Jul 06 Log no. Number of loans provided by bank i to country j since 2000 that had matured by July 2006 Loan AnalyticsExposure Sept 08 Share Outstanding loan volume by bank i to country j as a proportion of total assets of bank i at the time of the Lehman Brothers collapse Loan AnalyticsCommon language 2006 0/1 Dummy variable that is 1 if the home country of bank i and country j share the same language CIA World Factbook 2005Trade 2006 USD billion Log (1+ export plus import between the home country of bank i and country j ) UN ComtradeBank FDI 2006 No. Number of banks from the home country of bank i that own a subsidiary in country j Claessens and Van Horen (2011)

Credit information 2006 0/1Dummy variable that is 1 if the credit information index in destination country j is below the median level of all destination countries. The credit informationindex captures rules affecting scope, access, and quality of credit information. Doing Business

Dif credit information 2006 pointsDifference between the credit information index in the home country of bank i and the index in destination country j . A positive difference indicates that theaccessibility of creditor information is better at home than in the destination country. Doing Business

Creditor protection 2006 0/1

Dummy variable that is 1 if the strength of creditor protection index in destination country j is below the median level of all destination countries. A score of one is assigned for each of the following rights of secured lenders: (1) restrictions for a debtor to file for reorganization; (2) secured creditors are able to seize collateral after the reorganization petition is approved, i.e. no "automatic stay" or "asset freeze"; (3) secured creditors are paid first out of the proceeds of liquidating a bankrupt firm; (4) management does not retain administration of its property pending the resolution of the reorganization. The index ranges from 0 (weak creditor rights) to 4 (strong creditor rights).

La Porta et al. (1998)

Dif creditor protection 2006 pointsDifference between the creditor protection index in the home country of bank i and the index in destination country j . A positive difference indicates that creditor protection is stronger at home than in the destination country. La Porta et al. (1998)

Legal origin 2006 0/1 Dummy variable that is 1 if the legal origin of the bankruptcy law in destination country j is not English but either French, German, Nordic or Socialist. La Porta et al. (1998)Dif legal origin 2006 0/1 Dummy variable that is 1 if the legal origin of the bankruptcy law in the home country of bank i is different from the legal origin in destination country j . La Porta et al. (1998)Termination Jul 06-Sept 09 0/1 Dummy that is 1 if bank i stopped lending to firm k in the post-Lehman Brothers period Loan AnalyticsExperience with firm Jan 00-Jul 06 Log no. Number of loans provided by bank i to firm k since 2000 that had matured by July 2006 Loan AnalyticsArranger Jan 00-Jul 07 0/1 Dummy variable that is 1 if bank i acted at least once as a mandated lead arranger for firm k Loan AnalyticsSpread Jul 06-Sept 09 % Change in the average all-in spread charged by bank i to borrowers in country j post-Lehman Brothers compared to pre-crisis period Loan AnalyticsMaturity Jul 06-Sept 09 % Change in the average maturity of loans by bank i to borrowers in country j post-Lehman Brothers compared to pre-crisis period Loan Analytics

Appendix Table A3Variable definitions and sources

This table presents definitions and sources of all variables used in the paper. Pre-crisis refers to July 2006-June 2007 and post-Lehman Brothers to October 2008-September 2009. Loan Analytics is Dealogic's Loan Analytics database on syndicated loans. BankScope isBureau van Dijk's database of bank balance sheet and income statement data. IFS are the International Financial Statistics provided by the International Monetary Fund. Doing Business is the World Bank Doing Business Survey (2008).

51

Subsidiary Distance Domestic lendersDistance -0.109

(0.000)

Domestic lenders 0.217 -0.222(0.000) (0.000)

Experience 0.357 -0.027 0.202(0.000) (0.222) (0.000)

Appendix Table A4Correlation matrix of closeness variables

This table provides a matrix of the pair-wise correlation coefficients between ourcloseness variables. P-values in parentheses.