WPS6011 Policy Research Working Paper 6011 Channels of Transmission of the 2007/09 Global Crisis to International Bank Lending in Developing Countries Jonathon Adams-Kane Yueqing Jia Jamus Jerome Lim The World Bank Development Economics Prospects Group March 2012 Policy Research Working Paper 6011 Abstract During a financial crisis, credit provision by international the global financial crisis of 2007/09. It quantifies how banks may be stymied by three distinct, but related, changes in banks’ uncertainty about the value of their channels: changes in lending standards as a result of asset holdings, access to interbank liquidity, and internal increased economic uncertainty, changes in funding balance sheet considerations altered their supply of credit availability from interbank liquidity markets, and changes in the run-up, during, and in the immediate aftermath in solvency due to effects on bank balance sheets. This of the financial crisis, both in terms of their relative paper illuminates the manner by which each of these magnitudes, as well as the sensitivity of these magnitudes channels independently operated to affect developed- to the crisis. country bank lending in developing countries during This paper is a product of the Prospects Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at jadamskane@worldbank.org, yjia1@worldbank.org, and jlim@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Channels of Transmission of the 2007/09 Global Crisis to International Bank Lending in Developing Countries Jonathon Adams-Kane, Yueqing Jia, and Jamus Jerome Lim∗ Keywords: International bank lending, transmission channels, �nancial crisis JEL Classification: G21, G01, F34 Sector Board: EPOL ∗ The authors are with the Development Prospects Group at the World Bank. Their respective emails are: jadamskane@worldbank.org, yjia1@worldbank.org, and jlim@worldbank.org. We thank Mansoor Dailami and guidance and support. Financial support for this paper from the KCPII Window 2 Grant TF095266 “Analyzing the Impact of the Financial Crisis on International Bank Lending to Developing Countries� is gratefully acknowl- edged. The �ndings, interpretations, and conclusions expressed in this article are entirely those of the authors. They do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. 1 Introduction The economic environment that characterizes a �nancial crisis is one of fear, turmoil, and despair. For an international bank, these sentiments are experienced as changes in its lending standards as a result of heightened uncertainty over the value of its asset holdings, changes in its funding availability from interbank liquidity markets, and changes in its solvency due to balance sheet considerations. These three channels—of uncertainty, liquidity, and solvency— differentially affect the flow of credit provided by banks after the onset of a crisis. This paper seeks to illuminate the manner in which each of these channels independently operated to affect developed-country bank lending in developing countries in the run-up, during, and in the immediate aftermath of the global �nancial crisis of 2007/09. Gaining a better understanding of how different bank credit channels function during a �- nancial crisis is important, because the mitigation measures deployed by policymakers—whether before or after the fact—may differ according to the operative channel. For example, if liquidity access is the binding constraint, then the domestic central bank can relax its discount window, or engage in currency swap agreements with foreign central banks, in order to provide the nec- essary liquidity and alleviate the credit crunch. In contrast, if the problem is one of solvency, ex ante micro and macroprudential regulation may be more appropriate to limit the buildup of potentially nonperforming assets in the �rst place. Indeed, the issue of liquidity versus solvency is routinely discussed in the context of �nancial crisis management; however, relatively little of this debate has brought data to bear on the issue. While the speci�c conditions that govern the contraction of credit in any given bank are un- doubtedly unique, this does not mean that a more careful examination of the overall importance of the three distinct channels is unwarranted. After all, while the decision to wind down a bank does depend on its idiosyncratic circumstances, it is also valuable to identify the relative impor- tance of the different transmission channels at the economywide level. This paper contributes to that discussion by providing a systematic decomposition of these different channels. In a �nancial crisis, credit contraction occurs as a result of two distinct—but interrelated— factors. First, the crisis may bring about changes to the volatility of banks’ asset holdings, their access to liquidity, and their balance sheets, inducing them to withhold credit to borrowers. But a second factor is that the elasticity of credit provision to each of these considerations may also be different during a crisis. Thus, a bank may be forced to limit its lending due to, say, the greater difficulty that it faces in obtaining liquidity in primary markets (a level effect), but also due to a heightened sensitivity to liquidity scarcity under crisis conditions (an elasticity effect). Our �ndings suggest that during the crisis, bank liquidity problems and uncertainty were the main channels by which the crisis affected international lending from high income to developing countries, and that the solvency channel was relatively unimportant. Furthermore, the results suggest that banks’ sensitivity to these factors did not tend to change during the crisis; that is, the effect of the crisis on lending was essentially a normal reaction to changes in interbank liquidity and economic uncertainty, and that the impact on lending was due to abnormally large 2 shocks to liquidity and uncertainty rather than to any change in banks’ sensitivity to liquidity availability or change in risk. However, disaggregating lending into that by EU banks and that by U.S. banks yields a more nuanced message, with European banks becoming increasingly sensitive to market conditions during the most acute phase of the crisis, but this effect being offset, at least in part, by behavior of U.S. banks. The work here speaks to several areas of active research. One important strand of literature is work related to the impact that foreign bank presence has on credit availability, especially in developing countries. Foreign bank presence has been found to affect both overall (Dages, ınez Per´ & S´nchez 2005) Goldberg & Kinney 2000) and small business (Clarke, Cull, Mart´ ıa a lending in Latin America, as well as medium and large �rms located in India (Gormley 2010). Across the developing world, the entry of foreign banks has had measurable influence on credit access by domestic �rms, although the evidence favoring greater or lesser �nancing availability ınez Per´ 2006; Detragiache, Tressel & Gupta 2008). has been mixed (Clarke, Cull & Mart´ ıa While motivated by similar concerns, our work here focuses explicitly on crisis-related bank behavior, and explores the crisis effect in detail. The �ndings are also related to the studies concerned with the role that foreign banks play in domestic credit provision during times of �nancial stress. Foreign banks were more successful than domestic banks in maintaining low interest margins during the �nancial crisis in Malaysia (Detragiache & Gupta 2006), while foreign banks demonstrated no signi�cant difference in credit provision during the 1998 liquidity crunch in Pakistan (Khwaja & Mian 2008).1 Green�eld foreign banks were also able to refrain from contracting their credit base during crises in Central and Eastern Europe (while domestic banks were not) (de Haas & van Lelyveld 2006), and for emerging markets more generally, foreign banks tend to be less responsive to monetary shocks in the host country (Wu, Luca & Jeon 2011).2 Probably the paper closest in spirit to our work here is that of Cetorelli & Goldberg (2011). Like us, the authors are concerned with foreign liquidity provision during the recent crisis. However, given their focus on banks’ internal �nancing markets, their concerns with liquidity and solvency deal with their role as bank-speci�c controls—rather than as channels of credit contraction—and they do not address the uncertainty channel at all. Nevertheless, we regard their �ndings as an important complement to those that we present here.3 Our contribution to this literature is a more careful documentation of how the different channels affect credit provision, rather than a study of lending activity per se. A �nal group of papers is concerned with liquidity management by international banks, 1 Another notable feature of the Khwaja & Mian (2008) paper is that, in contrast to our work here, it decom- poses bank liquidity shocks into a bank lending channel and a �rm borrowing channel. While our focus is on alternate channels on the supply side of bank lending, we consider the demand side of the problem in greater detail in Section 5. 2 In a slight twist, Peek & Rosengren (1997) examine the lending behavior of Japanese banks in the United States when a stock market crisis was experienced in Japan. They �nd an economically and statistically signi�cant reduction in Japanese bank lending as a result of the shock. 3 In a compananion paper, Cetorelli & Goldberg (2011) also examine the liquidity channel in greater detail, decomposing it into direct cross-border lending by foreign banks and indirect local lending by foreign bank affiliates. 3 especially those based in the United States. One early study examined the determinants of the allocation of banking assets, and found that existing economic ties, level of development of the host economy, and domestic deposits were all correlated with greater asset holdings (Goldberg & Johnson 1990). However, asset holdings are not equivalent to liquidity exposures, and subsequent studies have attempted to make a more direct connection to credit provision. For example, Goldberg (2002) has found that U.S. bank lending to emerging markets is remarkably stable, and largely insulated from demand conditions in the host economies. While we are certainly interested in the factors affecting credit allocation by international banks, these are ultimately secondary to our central concern of the constraints such banks may face in different economic environments. Thus, non-crisis lending behavior in our study is mainly a benchmark against which to examine the more interesting (in our view) question of crisis lending behavior. The rest of this paper is organized as follows. In Section 2, we walk through the theoretical literature on each of the three transmission channels for foreign bank lending. The next section goes on to describe our main dataset and variables of interest (Section 3.1), as well as the econometric model that we employ (Section 3.2). Section 4 reports our main results, along with robustness checks. Section 5 explicitly incorporates the demand side of the bank loan market into our analysis. A �nal section concludes with some thoughts on policy implications. 2 Channels of Transmission for Financial Shocks There is a large literature that is concerned with alternative channels of monetary transmission in general, and effects of shocks on credit provision in particular. Credit frictions give rise to a wedge between the cost of funds raised externally and the opportunity cost of internal funds; this so-called “external �nance premium� in turn operates either by shifting the supply of banks’ intermediated credit (a liquidity effect) (Bernanke 1983), or by impacting borrowers’ balance sheets (a solvency effect) (Bernanke & Gertler 1989). Strictly speaking, the balance sheet channel of monetary transmission operates on the de- mand, rather than supply side, of the market for loanable funds. However, �nancial imper- fections, not unlike those operating in the commercial credit market, may also lead to credit rationing in the wholesale credit market (Freixas & Jorge 2008). Thus, keeping in mind that foreign banks essentially operate in capital markets as both demanders (in the global interbank market) as well as suppliers (in the domestic commercial loan market), we consider both of these channels as relevant when considering the case of international banking activity. Moreover, in the presence of informational asymmetries, interest rates alone may be in- sufficient as a mechanism for efficient allocation of credit, leading banks to engage in rationing behavior (Stiglitz & Weiss 1981). The increase in uncertainty due, in part, to such informational imperfections translates into increased variability in banks’ expected pro�ts during a �nancial crisis. To �x ideas, consider a simple (partial equilibrium) model of foreign bank activity in an 4 emerging market. In non-crisis periods, a given bank i will maximize expected pro�ts at time t by allocating its liabilities (deposits and borrowings) d toward loans l and investible assets a4 : T a max Et π (lit ) f (ai,t+1 ; lit , σ1 ) dai,t+1 , (1) t=0 a where f (a; σ1 ) is the density function of the (continuous) random variable a with support [a, a] and variance σ1 . This density function captures the distribution of total returns from assets given the state of the world 1, where a captures both period returns (such as dividend payouts or coupon payments), as well as the current market valuation of the asset. The bank is subject to an intertemporal balance sheet constraint ai,t+1 + mit = B (ait , mi,t−1 , −lit , dit ) , (2) where m is loanable cash, of which holdings are necessary for next-period loans in the form of a liquidity constraint mi,t−1 ≥ lit . (3) For simplicity, we limit our solution to the circumstance where the bank will otherwise wish to hold no spare cash, that is, when the liquidity constraint is binding.5 This allows us to simplify the problem by substituting (3) into (2), and maximizing (1) subject to ai,t+1 = B (ait , li,t+1 , dit ) . The Bellman value function for the problem is a V0 (ai,t+1 ) = max π (ait , li,t+1 ) + E1 V1 (ai,t+1 ) f (ai,t+1 ; li,t+2 , σ1 ) dai,t+1 . lit a This yields an Euler that governs the allocation of loans between t and t + 1 given by ∂B (ait , li,t+1 , dit ) ∂B (ai,t+1 , li,t+2 , di,t+1 ) π (lit ) · = π (li,t+1 ) · · ∂ait ∂ait a (4) ai,t+1 f (ai,t+1 ; σ1 ) dait+1 . a As is standard in �rst order conditions of this nature, (4) essentially says that bank i will 4 For the sake of simplicity, the model is stylized so that loans are regarded as nontradable assets with a �xed rate of return, whereas non-loan assets are tradable and thus subject to price and return volatility, and therefore subject to shocks to the variance of their return. In the context of the 2007/09 crisis, banks held a large portfolio of nontraditional investments, many in derivatives such as mortgage-backed securities (MBS) and collateralized debt obligations (CDO), often in special investment vehicles that the parent banks were ultimately liable for. 5 This would be the case if the returns on investible assets exceed that of loans, since in that case agents will never choose to hold cash in advance of the next period’s loans, as they would instead earn a higher return by placing the cash in assets. Whether this condition holds is, ultimately, an empirical question, which we abstract from in our very simple model of bank lending. 5 equate the marginal value of a loan made at time t to the expected value of the loan at time t + 1, taking into account the opportunity cost of making the loan as opposed to placing it in ∂B(ait ,li,t+1 ,dit ) ∂B(ai,t+1 ,li,t+2 ,di,t+1 ) the investible assets (the relative price of which is given by ∂ait / ∂ait ). Moreover, any given bank will face a transversality condition lim λT +1 aT +1 = 0, (5) T →∞ where λT +1 is the shadow price (multiplier) on the constraint (2). Now, consider the behavior of the same bank during a �nancial crisis. While a �nancial crisis is likely to result in a �rst-moment shock to asset valuations, it is also likely to incorporate a second-moment element, where uncertainty becomes more pervasive. Following Bloom (2009), we call this crisis-induced change in the variance σ an uncertainty shock, which leads to a revised density function given by f (a; σ2 ), where the variance of the function is now given by σ2 > σ1 .6 Repeating the exercise above yields the crisis-period analogue of (4): ∂B (ait , li,t+1 , dit ) ∂B (ai,t+1 , li,t+2 , di,t+1 ) π (lit ) · = π (li,t+1 ) · · ∂ait ∂ait a (6) ai,t+1 f (ai,t+1 ; σ2 ) dait+1 . a Taken together, (4), (6), and (5) characterize the three channels that impact a bank’s credit provision in and out of a �nancial crisis: its need for access to liquidity (the constraint (3) and ∂B(ait ,li,t+1 ,dit ) the presence of next-period loans in the current-period derivative in the Euler, ∂ait ); balance sheet solvency considerations (the constraints (2) and (5), and the inclusion of the function B (·)); and the impact of uncertainty (the presence of σ2 and σ1 ). The importance of the solvency channel has been empirically veri�ed for periods of tight money (Bernanke, Gertler & Gilchrist 1996), as well as, more speci�cally, in the context of �nancial crises (Duchin, Ozbas & Sensoy 2010). Likewise, the liquidity channel is an important conduit for monetary policy (Kashyap & Stein 2000), and was important in the propagation of the Great Depression (Bernanke 1983), and appears to be relevant for the recent �nancial crisis as well (Ivashina & Scharfstein 2010). Finally, Bloom (2009) has found that uncertainty shocks are crucial for understanding the dynamics of crises; the uncertainty channel has also been explored in the context of �nancial contagion by Kannan & Koehler-Geib (2011).7 These three channels all appear to have been important in the 2007/09 �nancial crisis. Widespread concerns about the ability of counterparties to make good on unsecured loans dis- 6 An alternative approach to capturing the role of risk is to directly introduce time-varying relative risk aversion into a utility function, as in Boschi & Goenka (2012). Crises are then modeled as shocks to the risk premia of portfolios held by international investors, which may give rise to contagion effects. Since this approach to modeling risk relies on propensities as opposed to asset volatility, it can have a material impact on the variables used to measure risk. We address this concern in our robustness section. 7 a o Fern´ndez-Villaverde, Guerr´n-Quintana, Rubio-Ram´ ırez & Uribe (2011) also explore the real impacts of what they call a volatility shock on small open economies, arguing that emerging economies often face time- varying volatility of real interest rates. 6 rupted interbank credit markets, leading to sharp spikes in the spread between the interbank lending rate and the overnight index swap (Libor-OIS), especially in the aftermath of the Lehman collapse in September 2008 (Figure 1(a)). This led to widespread difficulties in ob- taining basic rollover credit, and likely played a signi�cant role in the contraction of credit. As the crisis wore on, moreover, spreads on credit default swaps (CDS) for bank bonds began to widen considerably, suggesting increasing concerns over (bank) credit impairment as a result of worsening balance sheets (Figure 1(b)). The degraded balance sheets, in turn, would likely have led banks to limit their lending to only the most creditworthy borrowers. Finally, the volatility of asset returns increased substantially during the crisis period; the implied volatility of the S&P 500, for example, rose in the �rst phase of the crisis, and jumped sharply in the second (Figure 1(c)). bps bps 600 350 U.S. banks 5-year CDS spread 300 500 250 400 200 300 150 200 100 100 50 LIBOR-OIS spread 0 0 Jan-07 Jul-07 Jan-08 Jul-08 Jan-09 Jul-09 Jan-07 Jul-07 Jan-08 Jul-08 Jan-09 Jul-09 Source: Bloomberg Source: Datastream (a) LIBOR-OIS spread, 2007–10 (b) CDS spreads on credit sector, 2007–10 90 80 70 60 50 VIX implied volatility 40 30 20 10 0 Jan-07 Jul-07 Jan-08 Jul-08 Jan-09 Jul-09 Source: Datastream (c) VIX market volatility index, 2007–10 Figure 1: Three-month LIBOR-OIS and TED spreads (top panel), �ve-year CDS spreads on U.S. banking sector index (middle panel), and VIX implied volatility of S&P 500 index options (bottom panel), January 2007 through December 2010. The charts show, for both the initial (light shaded area) and acute (dark shaded area) phases of the crisis, that there was severe difficulty in obtaining liquidity (interest rate spreads), signi�cant concerns over bank solvency (credit default swap spreads), and that the volatility of returns on asset markets was elevated relative to non-crisis periods (market volatility index). 7 3 Data and Methodology 3.1 Data sources and description The dataset used for this paper was compiled from a variety of sources, the full details of which are described in the technical annex. Here, we limit our discussion to the main variables of interest. The main dependent variable that we consider is quarterly total foreign bank claims for BIS- reporting banks with lending to developing countries. The claims detail the foreign exposure, on a worldwide consolidated basis with inter-office positions netted out, of up to 5,615 BIS- reporting international banks with foreign exposure to as many as 138 developing economies per quarter. This coverage within the foreign claims universe is nearly 100 percent, and the foreign claims include cross-border claims in all currencies, along with local claims of foreign affiliates in both foreign currencies and local currency. In our robustness checks, we also consider the exposures of just the 150 U.S. banks with foreign exposure, along with those of the 4,488 BIS-reporting banks based in Europe.8 Overall, claims appear to have responded with some lag from the start of the crisis in 2007Q3 (peaks were attained about year later), and troughed at the start of 2009 (Figure 2). $ Trillions 4.0 Foreign claims, other 3.5 Foreign claims, US banks Foreign claims, EU banks 3.0 2.5 2.0 1.5 1.0 0.5 0.0 2004Q1 2005Q1 2006Q1 2007Q1 2008Q1 2009Q1 2010Q1 2011Q1 Figure 2: Total cross-border foreign claims by BIS-reporting international banks to developing countries, disaggregated by exposure into EU, U.S., and other. Aggregate foreign claims rose steadily from the beginning of the sample period and peaked around 2008Q2, before assuming troughs in 2009Q1. Claims at the end of the sample period exceed the pre-crisis peaks. Our main independent variables seek to capture the different channels of interest. To capture liquidity effects, our benchmark analysis relies on the Libor-OIS spread.9 Since the spread represents the difference between (risky) interbank borrowing costs and a (low-risk) 8 European countries with BIS-reporting banks are primarily, although neither exclusively nor exhaustively, members of the euro area: they comprise the EU-15 countries and Norway. 9 We utilize a spread because rates alone—for example, the Libor overnight rate—are a reflection of not only liquidity availability but also the cost of capital, which is netted out when we take the spread between two rates of equivalent tenor. 8 derivative based on expectations of such costs, it captures the compensation demanded by investors to insure against default. However, the Libor-OIS spread also captures an element of liquidity. Since an OIS contract does not require the exchange of loan principal, but only the interest component, this premium can also be viewed as an inducement to forgo liquid assets; such premia may be especially important in the environment of a �nancial crisis, where liquidity is very scarce. Moreover, it appears that liquidity may in fact be the larger component of Libor-OIS spreads (Schwarz 2010). Consequently, while the Libor-OIS rate does in fact comprise risk premia for both (inter- bank) credit risk and liquidity risk, in our application it nevertheless serves as a reasonable proxy for the availability of liquidity, as adjudged by the ability of a bank to secure short-term liquidity from the wholesale market. Another way to interpret this choice is that, when thinking about the liquidity channel, we are concerned not primarily with market liquidity—which is cer- tainly well measured by more traditional spreads, such as those between Libor and certi�cates of deposit (CD), or that between bid and ask prices within a given securities market—but with funding liquidity for banks, which is well referenced by the Libor-OIS spread. To measure solvency, our benchmark speci�cation uses a synthetic global bank CDS, which is a foreign claims-weighted average of the 5-year (the most liquid term) banking sector CDS indexes for U.S. and European banks, which are in turn averages of the mid-spread CDS rate for each constituent bank in the respective index.10 Since contingent payment on a CDS is typically triggered by credit events such as, inter alia, bankruptcy and restructuring, CDS spreads for the banking sector serve as important indicators of market expectations of bank solvency. While the accuracy of such spreads in predicting the actual strength of banks’ balance sheets has been certainly called into question by the �nancial crisis, they nevertheless remain valuable forward- looking signals of potential solvency issues (as opposed to accounting-based indicators, which are typically lagging). For the uncertainty channel, we construct an index of asset price volatility by doing principal component analysis of the VIX (an index of the volatility of the S&P 500) and the GARCH(1,1) conditional variance of �rst differences in prices in 7 other asset markets, and averaging the �rst two components (since those two have eigenvalues greater than 1), weighted by their proportions of the cumulative variance of the components. The conditional variance of a return on an asset captures uncertainty of its future path, given past observations, and is thus an appropriate measure for testing the effect of volatility on forward looking lending behavior. The assets included in the index are debt (the TED spread), equity (the VIX), foreign exchange (the exchange rate of the dollar to the euro, Japanese yen, and pound sterling), and commodities (agricultural, energy, and industrial metals prices) (the complete list is reproduced in the data annex). The working dataset for the benchmark is quarterly, beginning in 2004Q1 and ending 2011Q1, the last quarter where the European sovereign debt crisis remained largely contained 10 Since banking sector CDS were only widely traded beginning in 2004, using this measure limits the sample coverage to begin in 2004Q1. 9 to Greece.11 with the crisis period de�ned as encompassing 2007Q3 through 2009Q2, inclusive. This brackets the genesis of the crisis to the Quant event (Khandani & Lo 2011) and widespread central bank intervention starting in early August 2007, and the spread of the crisis to the real economy in the summer of 2009, accompanied by the signi�cant easing in tensions in �nancial markets. The crisis period comprises about thirty percent of the sample, which offers sufficient variation for reasonable statistical inference. 3.2 Econometric approach Our benchmark speci�cation used to test the relative importance of (4), (6), and (5) is an econometric model that embeds the measures described in Section 3.1 for all three channels into a single equation for foreign claims by international banks: F Cit = F Ci,t−1 + αi + β1 Lt + β2 St + β3 Ut + γDt + Γ Xit + crisist + it , (7) where F Cit (F Ci,t−1 ) is total (nominal) foreign claims by all international banks in developing country i at time t (t − 1), Lt , St , and Ut are measures of the liquidity, solvency, and uncertainty channels, respectively, Dt is a proxy variable for loan demand conditions in high income countries at time t, X is a vector of country-speci�c controls, and crisist is an indicator variable that takes on unity during the crisis, and zero otherwise. αi is a country �xed effect and it ∼ N 0, σ 2 is an i.i.d. disturbance term. The lagged dependent variable is included to account for possible persistence in the claims series, especially given the quarterly nature of the data.12 In our benchmark, X includes, for country i, its output, output growth, and inflation rate. Our benchmark relies on �xed effects estimators, with standard errors corrected for het- eroskedasticity and autocorrelation.13 In the sections that follow, we explore alternative esti- mation methodologies, designed to address considerations about demand-side factors as well as possible endogeneity among the different channels. 4 Results and Findings 4.1 Baseline results Our benchmark results are reported in Table 1. In column (B2 ), we regress foreign claims on its one-period lag, controls, a crisis dummy, and the three transmission channels as measured by the Libor-OIS spread (liquidity), global banking sector CDS (solvency), and the asset volatility 11 We also experimented with the sample period in two ways. First, we considered a restricted sample ending in 2010Q4. Second, we also included an additional crisis dummy for the European debt crisis that began in 2010Q1. Neither change affected the qualitative �ndings we report here in a signi�cant fashion, and these additional results are available on request. 12 A Fisher-type test for unit roots (Maddala & Wu 1999) fails to reject the null of stationarity for all series in the panel in levels (χ2 = 240.9, p = 0.71), but does so in �rst differences (χ2 = 1288.2, p = 0.00). 13 A Hausman test further suggests that �xed, rather than random, effects should be applied (χ2 = 470.06, p = 0.00). 10 index (uncertainty). By way of comparison, the equivalent speci�cation, which pools crisis and non-crisis periods, is repeated in column (B1 ), but without the indicator variable for the crisis period. It is clear that controlling for the crisis period results in important differences. Interestingly, the coefficient for the crisis variable in column (B2 ) is positive and statistically signi�cant, despite that lending to developing countries dipped during the crisis, while the coefficient on the Libor-OIS spread turns from positive to negative, suggesting that funding liquidity problems tend to be negatively associated with cross-border lending, but that the effect is partially offset during the crisis either by convexity in the negative relationship, or by countervening effects of unobservables. The results from this speci�cation also suggest that lending to developing countries is affected by �nancial shocks primarily through the liquidity and uncertainty channels: the coefficients for the liquidity and uncertainty channels are both negative and statistically signi�cant at the conventional levels. The economic effects of marginal changes in liquidity and uncertainty are economically insigni�cant: a 1 percent increase in the Libor-OIS spread is associated with about a 0.025 percent decrease in lending, and the decline associated with a 1 percent increase in volatility is about 0.06 percent. However, during the crisis these measures increased by several hundred percent, and changes of this magnitude are associated with economically signi�cant reductions in cross-border lending. The solvency channel, on the other hand, does not appear to be important. The ultimate impact of any single channel may not be contemporaneous, but rather unfold over time. To uncover the possible effects of delayed transmission, we specify an autoregressive distributed lag model in column (B3 ), with one-, two-, and three-quarter lags on the liquidity measure.14 Allowing for delayed effects of liquidity problems gives a statistically signi�cant coefficient of about the same magnitude, -0.022, on the three-quarter lag. This speci�cation may suffer in efficiency due to collinearity between the various lags, but it is nonetheless suggestive that the transmission of shocks to cross-border bank lending is not instantaneous. In the �nal column, (B4 ), we explore the sensitivity of the coefficients β1 − β3 obtained from the earlier two speci�cations by interacting the contemporaneous measures of Libor-OIS, global bank CDS, and asset volatility with the crisis variable. This allows lending to not just react to changes in the levels of these variables—which did change markedly during the crisis period—but it also allows the elasticity of lending with respect to each of these variables to change under crisis conditions. This speci�cation yields a striking result. The coefficients on the liquidity and uncertainty measures are essentially unchanged, and the interaction terms appear to be insigni�cant. This suggests that the impact of liquidity problems and uncertainty on bank lending during the crisis was due entirely to changes in these variables during the crisis, and not to changes in sensitivity 14 This speci�cation was selected using the best-�t speci�cation among 53 = 125 models, for up to four lags per channel, according to Akaike and Bayesian information criteria. The relevant statistics are detailed in the technical annex. 11 Table 1: Benchmark regressions for transmission channels asso- ciated with foreign claims by international banks in developing countries, 2004Q1–2011Q1 † B1 B2 B3 B4 Lagged foreign claims 0.772 0.772 0.781 0.771 (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ Libor-OIS 0.030 -0.025 -0.019 -0.030 (0.01)∗∗∗ (0.01)∗ (0.02) (0.02)∗ 1st lag 0.010 (0.01) 2nd lag 0.017 (0.01) 3rd lag -0.022 (0.01)∗∗ Libor-OIS × crisis -0.000 (0.04) Global bank CDS -0.014 -0.001 -0.004 0.002 (0.01)∗∗ (0.01) (0.01) (0.01) Global bank CDS× crisis -0.012 (0.01) Asset volatility -0.062 -0.060 -0.067 -0.068 (0.02)∗∗ (0.02)∗∗ (0.02)∗∗∗ (0.04)∗ Asset volatility × crisis 0.018 (0.06) Loan demand 0.009 -0.002 -0.006 -0.017 (0.03) (0.03) (0.04) (0.04) Inflation -0.019 -0.022 -0.022 -0.022 (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.17) Growth -0.131 -0.168 -0.159 -0.173 (0.15) (0.15) (0.15) (0.19) GDP 0.308 0.289 0.289 0.282 (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ Crisis 0.108 0.078 0.153 (0.02)∗∗∗ (0.03)∗∗∗ (0.12) Adj. R2 0.770 0.771 0.776 0.762 R2 (within) 0.771 0.772 0.777 0.772 Estimator FE FE FE FE N 3,435 3,435 3,429 3,435 † All variables are in log form. Heteroskedasticity and autocorrelation- robust standard errors are reported in parentheses, with the exception of speci�cation (B4 ), where errors are bootstrapped. A constant term was included in the regressions, but not reported. ∗ indicates signi�cance at 10 percent level, ∗∗ indicates signi�cance at 5 percent level, and ∗∗∗ indicates signi�cance at 1 percent level. 12 to these variables. In the case of the uncertainty measure, this result sheds light on the question of whether risk aversion itself changed during the crisis, or whether market participants, facing a shock to uncertainty and risk, took actions to reduce their risk exposure; our results support the latter of these two competing views. Finally, the coefficients of some of the control variables appear to be consistent with a priori theory. Bank lending to developing countries is greater for larger economies, and smaller for economies with higher inflation rates, perhaps due to greater ex ante uncertainty of real returns on loans, or because inflation proxies for a less favorable policy environment more broadly.15 There does not appear to be a signi�cant association of lending with growth, nor with loan demand from high income markets (as proxied by demand for loans faced by U.S. banks16 ). 4.2 Robustness of the benchmark The results are generally robust to alternative measures of the three channels, as shown in Table 2. In column (R1 ), the Libor-CD spread is substituted for Libor-OIS as a measure of interbank liquidity; (R2 ) includes both Libor-CD and Libor-OIS. In (R3 ), loss allowances as a fraction of outstanding loans substitute for CDS as a measure of bank balance sheet problems. In (R4 ), the index of conditional volatility of asset prices used to capture the uncertainty channel in the benchmark is replaced with an index of the unconditional standard deviations of the same set of asset prices. In (R5 ), the volatility measure is replaced with measure of changes in risk aversion of bankers; in (R6 ), both this risk aversion measure and the volatility measure are included. In (R7 ), growth and inflation are measured on a quarter-on-quarter basis instead of year-on-year. Speci�cation (R8 ) adds depreciation of the local currency against the U.S. dollar to the set of control variables. As discussed in Section 3.1, the term “liquidity� can have more than one meaning. Since we are concerned with the availability of funding liquidity to banks, as opposed to the degree of liquidity of a market which is better captured by traditional measures such as the spread between on-the-run and off-the-run securities or the bid-ask spread, the Libor-OIS spread is our preferred measure. However, disentangling a pure liquidity component from a counterparty risk component is not straightforward, since availability of credit depends crucially on credit risk, and more than one approach is worth considering. One alternative measure in the literature has been the spread between Libor and the rate on CDs, since purchasers of CDs do not necessarily face the liquidity constraints that banks do during a �nancial crisis (Taylor & Williams 2009). Furthermore, since CDs are typically only insured up to a limit, CD rates may rise along with 15 We use quarter-on-quarter (QoQ) inflation and growth data to better capture sharp, short-run changes that may be important during a crisis, as well as for consistency with the overall frequency of the dataset. However, QoQ data for inflation may result in extreme outliers, which we ameliorate by taking logarithms. Dropping these cases would potentially bias the results, since we lose valuable information from those observations. In any case, the results reported here remain robust to either arti�cially limiting the inflation rate in these cases to 1,000 percent, or dropping these observations entirely. 16 The measure of loan demand is from the Fed’s Senior Loan Officer Survey. The ECB collects comparable data on loan demand faced by European banks, but we use the U.S. measure since U.S. banks lend less in developing countries. 13 Table 2: Robustness regressions for transmission channels associated with foreign claims by international banks in developing countries, 2004Q1–2011Q1 † R1 R2 R3 R4 R5 R6 R7 R8 Lagged foreign claims 0.772 0.771 0.772 0.772 0.770 0.771 0.769 0.772 (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ Libor-OIS -0.027 -0.019 -0.026 -0.047 -0.028 -0.023 -0.028 (0.01)∗ (0.02) (0.01)∗∗ (0.01)∗∗∗ (0.01)∗ (0.01)∗ (0.01)∗∗ Libor-CD -0.003 -0.006 (0.01) (0.01) Global bank CDS -0.007 -0.000 0.002 0.002 -0.004 -0.001 -0.000 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Loss allowances -0.012 (0.01) Asset volatility -0.082 -0.071 -0.062 -0.083 -0.057 -0.061 (0.03)∗∗ (0.03)∗∗ (0.02)∗∗∗ (0.03)∗∗∗ (0.02)∗∗ (0.02)∗∗ Asset volatility -0.085 (unconditional) (0.03)∗∗ Risk aversion 0.025 0.154 (0.07) (0.09)∗ Loan demand 0.003 -0.015 -0.005 0.004 0.015 0.025 0.004 -0.001 (0.04) (0.04) (0.03) (0.03) (0.04) (0.04) (0.03) (0.03) 14 Inflation -0.021 -0.022 -0.021 -0.022 -0.022 -0.023 -0.056 (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ Inflation, YoY -0.033 (0.01)∗∗∗ Growth -0.154 -0.168 -0.176 -0.158 -0.137 -0.177 -0.184 (0.15) (0.15) (0.15) (0.15) (0.15) (0.16) (0.16) Growth, YoY -0.224 (0.37) GDP 0.298 0.286 0.295 0.284 0.282 0.279 0.289 0.290 (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ Depreciation 0.020 (0.00)∗∗∗ Crisis 0.077 0.113 0.095 0.113 0.108 0.097 0.105 0.113 (0.02)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗ Adj. R2 0.771 0.771 0.771 0.771 0.771 0.771 0.770 0.772 R2 (within) 0.772 0.772 0.772 0.772 0.771 0.772 0.771 0.773 Estimator FE FE FE FE FE FE FE FE N 3,435 3,435 3,435 3,435 3,435 3,435 3,401 3,399 † All variables are in log form. Heteroskedasticity and autocorrelation-robust standard errors are reported in parentheses. A constant term was included in the regressions, but not reported. ∗ indicates signi�cance at 10 percent level, ∗∗ indicates signi�cance at 5 percent level, and ∗∗∗ indicates signi�cance at 1 percent level. Libor due to concerns about the credit worthiness of banks, so the Libor-CD spread is viewed by some as a measure of pure liquidity scarcity. We view the premise that purchasers of CDs are not liquidity constrained during a crisis as questionable, but Libor-CD is nonetheless a clear candidate for a robustness check of our main liquidity variable. When Libor-CD is included in lieu of Libor-OIS (column R1 ), the coefficient is close to zero and statistically insigni�cant. When both Libor-CD and Libor-OIS are included (column R2 ), Libor-OIS retains its sign, magnitude, and signi�cance. Our preferred measure of balance sheet problems is an index of CDS spreads, which has the advantages of being forward-looking, and of being based on rates faced by a wide sample of banks in both the U.S. and Europe. However, more direct measures of solvency issues can be found in bank �nancial data. As a robustness check, loss allowances as a fraction of outstanding loans, as reported by U.S. banks to the FDIC, are included in place of the CDS index (column R3 ). The coefficient is insigni�cant, as is that on CDS in the benchmark estimates that do not include lagged channel variables. Although not reported, similar results are found for other variables reported to the FDIC: measures of noncurrent loans and net charge-offs as fractions of outstanding loans, and the fraction of lenders which are unpro�table. These results are broadly supportive of the idea that balance sheet problems have little, if any, contemporaneous impact on cross-border bank lending. In the benchmark estimates, we rely on a GARCH (1,1) conditional volatility index as a measure of uncertainty, for reasons given above. Here, we consider the robustness of this measure by instead using an unconditional measure of volatility: an index of the rolling standard deviations of prices in the same 7 asset markets, along with VIX, used to compute the conditional volatility measure.17 Results with this alternative measure of volatility are reported in column (R4 ), and mirror the benchmark results, with a signi�cant negative coee�cient on the alternative volatility measure. As discussed in Section 2, the uncertainty channel variable is intended to measure shocks to uncertainty, as distinct from shocks to risk aversion. However, the two concepts are interrelated, as the effect of uncertainty on behavior depends on the degree of risk aversion; and the degree of perceived risk aversion, as well as the effect of a given degree of risk aversion on behavior, may depend on the level of uncertainty. Furthermore, there is a question of whether risk aversion itself changed during the crisis, or whether behavior to reduce risk exposure during the crisis was predominantly a reaction to heightened risk and uncertainty for a given degree of risk aversion. Results with a variable measuring U.S. loan officers’ risk aversion are reported in columns (R5 ) and (R6 ).18 While the coefficient is insigni�cant when risk aversion is included in place of 17 The results are robust to the method of aggregation. While we report here the weighted average of the �rst two principal components, using a simple average of the 8 series yields qualitatively similar results. 18 The risk aversion variable is constructed from a question on the Fed’s Senior Loan Officer Survey that asks loan officers who reported that they tightened or eased credit standards or terms on C&I loans what factors were relevant to that decision, with reduced or increased tolerance for risk given as an option for the reason for tightening or easing, respectively. Banks’ responses are averaged, with each response that a change in risk tolerance was “very important� given twice the weight of a response that it was “important.� 15 volatility, it is notable that it is positive and signi�cant when volatility is also included in the speci�cation. One explanation is that, while heightened uncertainty may cause banks to reduce their lending both at home and abroad, shocks to risk aversion for a given level of uncertainty cause banks to substitute towards lending in relatively stable markets. During the crisis, which is mainly when increases in risk aversion were reported, this means lending more in developing countries, ceteris paribus. It is also worth noting that when both volatility and risk aversion are included, the coefficient on volatility remains negative and highly signi�cant, suggesting that the two variables’ relationships with lending are distinct and that our benchmark measure is not primarily picking up shocks to risk aversion as opposed to uncertainty. While we prefer quarter-on-quarter measures of growth and inflation because they capture sharp short-term shocks to output and prices during a crisis, they are also sensitive to seasonal variation, so we consider year-on-year measures as alternate controls, with the results reported in column (R7 ). Also, it has been argued that investment can be sensitive to exchange rate variations (Campa & Goldberg 1999), and so loan demand by banks may fluctuate accordingly. We allow for this possibility in column (R8 ), where we include the depreciation of the local a currency vis-`-vis the U.S. dollar as an additional control. The main results are robust to these checks. 4.3 Does foreign bank lending differ by source and destination? Foreign claims by high income country banks on counterparties in developing countries consist mainly of loans by European banks, but the U.S. was at the center of the crisis, and there is no reason to think that the benchmark results apply equally well to European and U.S banks. Separating foreign claims into those by U.S. banks and those by EU banks yields a number of insights; the results are given in Table 3. Columns (S1 ) and (S5 ) replicate the benchmark regression (B2 ) for U.S. and EU banks’ claims, respectively, but with two crisis dummies for each, with the crisis divided into pre- and post-Lehman Brothers collapse for the U.S., and pre- and post-Vienna Initiative for the EU, since these events are widely regarded to mark relevant shifts in the crisis for these two markets. Due to small period length for both the latter periods, we regress the interaction terms for each channel separately; these are reported in columns (S2 )–(S4 ) for U.S. banks, and (S6 )–(S8 ) for EU banks. One key result is that the solvency channel is signi�cant, but with a positive coefficient for U.S. banks, and negative for EU banks, which helps to explain why it is insigni�cant for total foreign claims in the benchmark estimates. One possible explanation for why solvency considerations seem to have afflicted European bank lending (insofar as exposures to developing countries are concerned), but not U.S. lending, is that the Federal Reserve intervened far more aggressively than the European Central Bank (ECB). Indeed, anecdotal evidence is generally supportive of this difference: While the U.S. Federal Reserve introduced the Troubled Asset Relief Program (TARP) to shore up bank balance sheets, along with a large number of new credit lines—such as the Term Auction Facility (TAF) and the Commercial Paper Lending 16 Table 3: Regressions for transmission channels associated with foreign claims by EU and US banks in developing countries, 2004Q1–2011Q1 † S1 S2 S3 S4 S5 S6 S7 S8 US Banks EU Banks Lagged foreign 0.600 0.600 0.600 0.599 0.752 0.751 0.751 0.751 claims (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ Libor-OIS/ -0.110 -0.080 -0.058 -0.073 0.042 0.072 0.072 0.072 Euribor-Eonia (0.04)∗∗∗ (0.05)∗ (0.05) (0.05) (0.02)** (0.03)*** (0.03)** (0.03)** Libor-OIS -0.180 -0.196 -0.187 × pre-Lehman (0.20) (0.19) (0.19) Libor-OIS -0.153 × post-Lehman (0.12) Euribor-Eonia -0.014 -0.014 -0.014 × pre-Vienna (0.05) (0.05) (0.05) Euribor-Eonia -0.142 × post-Vienna (0.05)∗∗∗ US/EU bank CDS 0.066 0.056 0.046 0.053 -0.017 -0.031 -0.031 -0.031 (0.03)∗∗ (0.03)∗∗ (0.03)∗ (0.03)∗∗ (0.01)∗ (0.01)∗∗∗ (0.01)∗∗ (0.01)∗∗ US bank CDS -0.340 -0.420 -0.364 × pre-Lehman (0.12) (0.12) (0.12) US bank CDS 1.101 × post-Lehman (0.56)∗ EU bank CDS 0.098 0.098 0.098 × pre-Vienna (0.04)∗∗ (0.04)∗∗∗ (0.04)∗∗ EU bank CDS -0.706 × post-Vienna (0.26)∗∗∗ Asset volatility -0.003 0.114 0.191 0.137 -0.113 -0.134 -0.134 -0.134 (0.06) (0.12) (0.10)∗ (0.11) (0.03)∗∗∗ (0.06)∗∗ (0.06)∗∗ (0.05)∗∗ Asset volatility -0.340 -0.420 -0.364 × pre-Lehman (0.20)∗ (0.18)∗∗ (0.20)∗ Asset volatility -0.219 × post-Lehman (0.14) Asset volatility -0.067 -0.067 -0.067 × pre-Vienna (0.11) (0.10) (0.11) Asset volatility -0.214 × post-Vienna (0.07)∗∗∗ Loan demand, -0.174 -0.140 -0.094 -0.131 0.159 0.219 0.219 0.219 US/EU banks (0.12) (0.12) (0.12) (0.14) (0.06)∗∗∗ (0.08)∗∗∗ (0.08)∗∗∗ (0.09)∗∗ Inflation 0.109 0.109 0.109 0.109 -0.025 -0.028 -0.028 -0.028 (0.01)∗∗∗ (0.20) (0.25) (0.22) (0.00)∗∗∗ (0.10) (0.10) (0.12) Growth -0.237 -0.308 -0.267 -0.306 -0.404 -0.443 -0.443 -0.443 (0.40) (0.51) (0.44) (0.51) (0.20)∗ (0.26)∗ (0.26)∗ (0.27) GDP 0.275 0.269 0.275 0.269 0.214 0.187 0.187 0.187 (0.09)∗∗∗ (0.09)∗∗∗ (0.10)∗∗∗ (0.10)∗∗∗ (0.04)∗∗∗ (0.05)∗∗∗ (0.05)∗∗∗ (0.05)∗∗∗ pre-Lehman 0.121 0.474 0.511 0.488 (0.06)∗ (0.45) (0.43) (0.43) post-Lehman 0.153 0.744 -5.969 0.255 (0.10) (0.48) (3.12)∗ (0.14)∗ pre-Vienna 0.085 -0.226 -0.226 -0.226 (0.03)∗∗∗ (0.18) (0.17) (0.17) post-Vienna 0.044 0.651 3.842 0.229 (0.03)∗ (0.20)∗∗∗ (1.38)∗∗∗ (0.06)∗∗∗ Adj. R2 0.437 0.416 0.416 0.416 0.702 0.691 0.691 0.691 R2 (within) 0.439 0.439 0.440 0.440 0.703 0.704 0.704 0.704 Estimator FE FE FE FE FE FE FE FE N 3,435 3,435 3,435 3,435 3,435 3,435 3,435 3,435 † All variables are in log form. Heteroskedasticity and autocorrelation-robust standard errors are reported in parentheses. A constant term was included in the regressions, but not reported. Indicator variables for Lehman and Vienna separate the crisis variable at 2008Q4 and 2009Q1, respectively. ∗ indicates signi�cance at 10 percent level, ∗∗ indicates signi�cance at 5 percent level, and ∗∗∗ indicates signi�cance at 1 percent level. 17 Facility (CPLF)—to enhance liquidity during the height of the crisis, European banks only had the standard discount window of the ECB (along with currency swap agreements, although these likewise had to be intermediated by the ECB). For the U.S., CDS interacted with the second phase of the crisis is also positive and signi�cant, meaning that the positive association increased during this phase, which is consistent with the policy response explanation. For the EU, the analagous interaction term is negative and signi�cant, meaning that EU bank lending became more sensitive to solvency problems during the acute second phase of the crisis as well. The results on liquidity also add some nuance to the benchmark estimates. For U.S. banks, Libor-OIS is associated with decreased lending to developing countries (in speci�cations (S1 ) and (S2 )), and the sensitivity to Libor-OIS does not change during either phase of the crisis. For EU banks, on the other hand, the Euribor-Eonia spread is associated with increased lending during normal times, but decreased lending during the second phase of the crisis. This can be seen in the overall foreign claims patterns in Figure 2. As for the uncertainty channel, U.S. lending appears to have been negatively associated with volatility during the �rst phase of the crisis only, and insensitive to it at other times. EU lending is negatively associated with volatility in good times and bad, but was especially so during the second phase of the crisis, despite the efforts of the Vienna Initiative to prop up credit provision to Eastern Europe. 5 What is the Role of the Interest Rate? Up till this stage, our examination of the three main transmission channels has taken the form of Equation (7), which deliberately suppresses any discussion of internal demand conditions in developing countries. To the extent that we incorporated demand considerations, this was through a measure—changes in demand conditions reported by the Federal Reserve—which proxies for external demand conditions, mainly from high-income countries. While we certainly recognize that (7) does not fully account for internal loan demand, our preference for this approach—at least at the outset—was conditioned by a desire to focus which transmission channels affect the supply side, along with the desire to introduce interaction effects between the crisis and the main coefficients of interest (β1 , β2 , and β3 ). In this section, we substitute the generic measure of demand conditions with country-speci�c interest rates. Accordingly, (7) can be re-expressed as F Cit = F Ci,t−1 + αi + β1 Lt + β2 St + β3 Ut + γ rit + Γ Xit + crisist + it , (8) where rit is the domestic lending rate in country i at time t, which now more fully approxi- mates the potential effect of local demand for bank loans. Of course, since interest rates are codetermined by bank lending as well as saver deposits, r is endogenous. Consequently, we re- quire an instrument to identify the loan demand function, and (8) is estimated via instrumental variables (IV). For this purpose, we rely on country demographics; in particular, we exploit the 18 (plausibly) exogenous variation offered by the aged dependency ratio. Increases in this ratio are associated with reduced saving, as domestic agents draw down on the stock of national saving. Since local banks typically have access to only the domestic deposit base, the interest rates they charge will rise, which in turn reduces the demand for their loans as �rms substitute toward foreign lenders with lower interest rates on offer. In equilibrium, the observed domestic lending rate will, mutatis mutandis, increase in tandem with the dependency ratio. There are two additional empirical issues that we need to consider in our estimation of (8). First, what does a national-level interest rate mean in the context of multiple distinct foreign banks? Arguably, interest rates should be bank- and borrower-speci�c, but this is precluded by the aggregate nature of the data. Fortunately, even if our measure of interest rates does turn out to be noisy, one of the advantages of IV estimates is that, conditional on a reliable instrument, the approach corrects for measurement error in the instrumented variable. We accordingly proceed with the use of the quarterly average national lending rate as a measure of our domestic lending rate. A second issue concerns the inclusion (or not) of our original measure of demand conditions, D. In theory, these two demand forces are distinct, and should both be included in (8). However, since our proxy measure of loan demand in high-income countries may also contain a component of developing-country loan demand,19 we limit any conflation between the two by specifying (8) with only the interest rate variable. The estimation results are reported in Table 4. In column (I1 ), we replicate the �xed-effects model in (B2 ), but replace loan demand with the lending rate. The coefficient on the interest rate is positive (as expected a priori ), but statistically indistinguishable from zero; this is also the case for the IV speci�cations that follow. The coefficients on the three channels are also the same as before (negative), but again are statistically insigni�cant. Due to the endogeneity problems inherent in the estimate, however, we regard these results are merely illustrative, and proceed to a discussion of the IV estimates. Columns (I2 )–(I4 ) report IV estimates that correspond to the benchmark speci�cations (B2 )–(B4 ). The instruments satisfy the relevance condition—Kleibergen-Papp LM tests are statistically signi�cant—and the (insigni�cant) Hansen J statistic indicates that the instruments are valid. While F statistics (not reported) imply that the �rst-stage �t is reasonable, the corresponding Kleibergen-Papp F statistics are relatively low (F ∈ [4.41, 4.94], equivalent Stock- Yogo critical values place this range at around 25 percent relative bias), which suggests that the speci�cations could be weakly identi�ed. We recognize this possible weak instrument problem, and, consequently, our interpretation of the results here proceed with greater caution. While the results are largely similar to those in Table 1, the asset volatility channel does not appear to be operative any longer in any of our speci�cations. We conjecture that the inclusion of the interest rate could dilute the impact of the asset volatility channel, since interest rates 19 The question on changes in demand conditions posed in Senior Loan Officer Survey does not distinguish between borrower nationality. While cross-border exposures of U.S. banks are mainly to high-income countries, the response may nevertheless incorporate some consideration of demand conditions in developing countries. 19 Table 4: Regressions for transmission channels associated with foreign claims by EU and US banks in developing countries, 2004Q1–2011Q1 † I1 I2 I3 I4 Lagged foreign claims 0.788 0.796 0.797 0.799 (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.02)∗∗∗ Libor-OIS -0.040 -0.035 -0.027 -0.029 (0.01)∗∗∗ (0.01)∗∗ (0.02) (0.02) 1st lag 0.005 (0.01) 2nd lag 0.015 (0.02) 3Rd lag -0.017 (0.01) Libor-OIS × crisis -0.006 (0.05) Global bank CDS 0.007 -0.000 -0.003 0.008 (0.01) (0.01) (0.01) (0.01) Global bank CDS× crisis -0.008 (0.02) Asset volatility -0.024 -0.029 -0.039 0.029 (0.02) (0.03) (0.03) (0.06) Asset volatility × crisis -0.048 (0.07) Lending rate 0.110 0.641 0.873 -1.520 (0.11) (1.82) (1.73) (2.44) Inflation -0.007 -0.061 -0.085 0.157 (0.02) (0.19) (0.18) (0.31) Growth -0.075 -0.045 -0.044 -0.140 (0.15) (0.13) (0.14) (0.19) GDP 0.261 0.286 0.301 0.174 (0.04)∗∗∗ (0.11)∗∗∗ (0.10)∗∗∗ (0.12) Crisis 0.106 0.104 0.081 0.175 (0.02)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.14) Adj. R2 0.801 0.793 0.792 0.788 Kleibergen-Papp rk 12.915∗∗ 11.812∗∗ 11.917∗∗ Hansen J 2.343 1.999 0.209 Estimator FE IV IV IV Instruments 13 19 16 N 2,849 2,713 2,713 2,782 † All variables are in log form. Heteroskedasticity and autocorrelation- robust standard errors are reported in parentheses, with the exception of speci�cation (I4 ), where errors are bootstrapped. A constant term was included in the regressions, but not reported. The excluded instruments are the aged dependency ratio and its lags for four quarters. ∗ indicates signi�cance at 10 percent level, ∗∗ indicates signi�cance at 5 percent level, and ∗∗∗ indicates signi�cance at 1 percent level. 20 are themselves subject to signi�cant variation, especially during periods of crisis. The coefficient on the liquidity channel retains its sign and statistical signi�cance in the �rst two of the four speci�cations, and approaches statistical signi�cance in the �nal two (p = 0.134 and p = 0.159, respectively). Its magnitude is of the same order of signi�cance as the benchmark: the range of coefficients is between -0.02 and -0.04. Overall, we regard the �ndings reported in this section as broadly corroborative of the benchmark results. More importantly, given the insigni�cance of the interest rate even when it is instrumented, we conclude that concerns about endogeneity arising from its omission from the benchmark speci�cations are not generally well founded. 6 Conclusion In this paper, we have examined the transmission channels for �nancial shocks that affect international bank lending to developing countries, using the 2007/09 crisis as a case study. Our main �nding is that cross-border lending by international banks contracted primarily due to liquidity difficulties faced by these banks, and to heightened uncertainty. Another key message is that the reduction in aggregate lending to developing countries by high income banks during the crisis was not the result of changes in their sensitivity to liquidity shortages or changes in their tolerance of risk, but instead was a normal reaction to an abnormally large shock to liquidity and uncertainty. The main shortcomings of our work here stem from our reliance on aggregate cross-border lending data. While such data encompasses much greater coverage of the total volume of lending to the developing world, our inferences were largely limited to broad averages—and, in Subsec- tion 4.3, U.S. and EU exposures—which limits our ability to tease out idiosyncratic elements that may be of interest, such as whether solvency considerations were altogether unimportant, or whether they were more important for certain classes of �nancial institutions, such as those facing more severe balance-sheet difficulties. Future research along the lines of the work con- sidered here can attempt to explore these distinctions in greater detail. Our work here suggests that, during a �nancial crisis, policymakers in developing countries that host international banks may wish to supplement liquidity provision by these banks’ home nations with additional monetary support, rather than limiting their credit window to domes- tic banks alone. This will limit the deleterious effects of credit contraction during the crisis, especially if such international banks play a signi�cant role in their economies. Even though such banks are ultimately foreign-owned, policymakers are nevertheless invested in their success, insofar as their countries’ continued viability may be directly related to those of the banks. References Bernanke, Ben S. (1983). “Nonmonetary Effects of the Financial Crisis in Propagation of the Great Depression�. American Economic Review 73(3) (June): 257–276. URL http://www.jstor.org/stable/1808111 21 Bernanke, Ben S. & Mark L. Gertler (1989). “Agency Costs, Net Worth, and Business Fluctuations�. American Economic Review 79(1) (March): 14–31 Bernanke, Ben S., Mark L. Gertler & Simon Gilchrist (1996). “The Financial Accelerator and the Flight to Quality�. Review of Economics and Statistics 78(1) (February): 1–15 Bloom, Nicholas (2009). “The Impact of Uncertainty Shocks�. Econometrica 77(3) (May): 623–685 Boschi, Melisso & Aditya Goenka (2012). “Relative Risk Aversion and the Transmission of Financial Crises�. Journal of Economic Dynamics and Control 36(1) (January): 85–99 Campa, Jose Manuel & Linda S Goldberg (1999). “Investment, Pass-Through, and Exchange Rates: A Cross- Country Comparison�. International Economic Review 40(2) (May): 287–314 Cetorelli, Nicola & Linda S. Goldberg (2011). “Liquidity Management of U.S. Global Banks: Internal Capital Markets in the Great Recession�. NBER Working Paper 17355, National Bureau of Economic Research ıa Clarke, George R.G., Robert J. Cull & Mar´ Soledad Mart´ ıa ınez Per´ (2006). “Foreign Bank Participation and Access to Credit Across Firms in Developing Countries�. Journal of Comparative Economics 34(4) (December): 774–795 ıa ınez Per´ & Susana M. S´nchez (2005). “Bank Lending Clarke, George R.G., Robert J. Cull, Mar´ Soledad Mart´ ıa a to Small Businesses in Latin America: Does Bank Origin Matter?� Journal of Money, Credit and Banking 37(1) (February): 83–118 Dages, B. Gerard, Linda S. Goldberg & Daniel Kinney (2000). “Foreign and Domestic Bank Participation in Emerging Markets: Lessons from Mexico and Argentina�. Economic Policy Review 6(3) (September): 17–36 de Haas, Ralph & Iman van Lelyveld (2006). “Foreign Banks and Credit Stability in Central and Eastern Europe. A Panel Data Analysis�. Journal of Banking and Finance 30(7) (July): 1927–1952 Detragiache, Enrica & Poonam Gupta (2006). “Foreign Banks in Emerging Market Crises: Evidence from Malaysia�. Journal of Financial Stability 2(3) (October): 217–242 Detragiache, Enrica, Thierry Tressel & Poonam Gupta (2008). “Foreign Banks in Poor Countries: Theory and Evidence�. Journal of Finance 63(5) (October): 2123–2160 Duchin, Ran, Oguzhan Ozbas & Berk A. Sensoy (2010). “Costly External Finance, Corporate Investment, and the Subprime Mortgage Credit Crisis�. Journal of Financial Economics 97(3) (September): 418–435 a u o Fern´ndez-Villaverde, Jes´s, Pablo Guerr´n-Quintana, Juan F. Rubio-Ram´ırez & Martin Uribe (2011). “Risk Matters: The Real Effects of Volatility Shocks�. American Economic Review 101(6) (October): 2530–2561 e Freixas, Xavier & Jos´ M.P. Jorge (2008). “The Role of Interbank Markets in Monetary Policy: A Model with Rationing�. Journal of Money, Credit and Banking 40(6) (September): 1151–1176 Goldberg, Lawrence G. & Denise M. Johnson (1990). “The Determinants of U.S. Banking Activity Abroad�. Journal of International Money and Finance 9(2) (June): 123–137 Goldberg, Linda S. (2002). “When Is U.S. Bank Lending to Emerging Markets Volatile?� In Sebastian Edwards & Jeffrey A. Frankel (editors), Preventing Currency Crises in Emerging Markets, NBER Chapters, pp. 171–196. University of Chicago Press Gormley, Todd A. (2010). “The Impact of Foreign Bank Entry in Emerging Markets: Evidence from India�. Journal of Financial Intermediation 19(1) (January): 26–51 22 Ivashina, Victoria & David S. Scharfstein (2010). “Bank Lending During the Financial Crisis of 2008�. Journal of Financial Economics 97(3) (September): 319–338 Kannan, Prakash & Friederike Koehler-Geib (2011). “Uncertainty and Contagion�. In Robert W. Kolb (editor), Financial Contagion: The Viral Threat to the Wealth of Nations, pp. 21–28. Hoboken, NJ: John Wiley & Sons Kashyap, Anil K. & Jeremy C. Stein (2000). “What Do a Million Observations on Banks Say about the Trans- mission of Monetary Policy?� American Economic Review 90(3) (June): 407–428 Khandani, Amir E. & Andrew W. Lo (2011). “What Happened to the Quants in August 2007? Evidence from Factors and Transactions Data�. Journal of Financial Markets 14(1) (February): 1–46 Khwaja, Asim Ijaz & Atif R. Mian (2008). “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market�. American Economic Review 98(4) (September): 1413–1442 Maddala, G. S. & Shaowen Wu (1999). “A Comparative Study of Unit Root Tests with Panel Data and a New Simple Test�. Oxford Bulletin of Economics and Statistics 61(S1) (November): 631–652 Peek, Joe & Eric S. Rosengren (1997). “The International Transmission of Financial Shocks: The Case of Japan�. American Economic Review 87(4) (September): 495–505 Schwarz, Krista (2010). “Mind the Gap: Distentangling Credit and Liquidity in Risk Spreads�. Working paper, Philadelphia, PA: The Wharton School Stiglitz, Joseph E. & Andrew Weiss (1981). “Credit Rationing in Markets with Imperfect Information�. American Economic Review 71(3) (June): 393–410 Taylor, John B. & John C. Williams (2009). “A Black Swan in the Money Market�. American Economic Journal: Macroeconomics 1(1) (January): 58–83 Wu, Ji, Alina C. Luca & Bang Nam Jeon (2011). “Foreign Bank Penetration and the Lending Channel in Emerging Economies: Evidence from Bank-Level Panel Data�. Journal of International Money and Finance 30(6) (October): 1128–1156 23 Technical Appendix Table A.1: Summary statistics for main variables in benchmark speci�- cation, by crisis period† N Mean Std. Dev. Min Max Non-crisis Foreign claims 2,445 17,629 52,040 3 593,879 Libor-OIS 2,445 11.2 4.6 6.8 22.6 Global banking CDS 2,445 66.0 83.4 8.9 279.4 Asset volatility 2,445 -0.1 0.2 -0.4 0.4 Loan demand 2,445 0.1 0.2 -0.3 0.4 Inflation* 2,438 7.8 22.8 -35.9 941.1 Growth 2,445 5.7 5.4 -37.5 63.8 GDP 2,445 51,824 126,587 106.8 1,148,256 Crisis Foreign claims 990 22,263 58,476 4 377,268 Libor-OIS 990 88.7 49.7 42.7 211.5 Global banking CDS 990 128.9 64.9 33.7 229.7 Asset volatility 990 1.4 1.1 0.3 4.0 Loan demand 990 -0.3 0.2 -0.6 -0.1 Inflation* 986 9.9 13.3 -29.4 122.9 Growth 990 3.1 6.3 -40.5 67.5 GDP 990 58,897 137,326 157.2 1,197,708 † Summary statistics are provided for the sample between 2001Q4 and 2011Q1. The crisis period was de�ned as the period between 2007Q3 and 2009Q2. * The inflation statistics reported here suppress the outlier cases where QoQ inflation exceeds 1,000 percent. 24 Table A.2: De�nitions and sources of variables Variable De�nition and construction Data source(s) Foreign claims Total consolidated foreign claims of BIS reporting banks on up to 138 developing countries BIS Libor-OIS spread Quarterly average of daily 3-month Libor and overnight index swap rate differential Bloomberg Euribor-Eonia spread Quarterly average of daily 3-month Euribor and Euro overnight index swap rate differential Bloomberg Libor-CD spread Quarterly average of daily 3-month LIBOR and certi�cate of deposit rate differential Thomson Datastream Global bank CDS Synthetic claims-weighted index of U.S. and European bank CDS credit default swap indices Thomson Datastream Loss allowances Allowances for losses as fraction of outstanding loans, banks with foreign offices FDIC Qtr Bank Pro�le Asset volatility Predicted common factor of 1-month GARCH(1,1) conditional volatility of 8 assets† Datastream and GEM 25 Asset volatility (uncond) Predicted common factor of 1-month rolling coefficient of variation of 8 assets† Datastream and GEM Risk aversion Net tightening of lending due to change in risk aversion (range: -1 to 1) Fed SLOS/ECB BLS Loan demand Strength of demand change for C&I loans by borrowers from U.S./EU banks (range: -1 to 1) Fed SLOS/ECB BLS GDP Gross domestic product, in current U.S. dollars World Bank GEM & WDI GDP growth Year-on-year percentage change of real gross domestic product World Bank GEM & WDI Inflation Year-on-year percentage change of CPI IMF IFS Depreciation Quarterly average of change in market rate of local currency to U.S. dollars IMF IFS Lending rate Quarterly average of national lending rate IMF IFS † These comprised the VIX and the computed volatility for 7 additional constituent assets: exchange rates for the USD/EUR, USD/JPY, USD/GBP, price indexes for agriculture, energy, and industrial metals, and the TED spread. Table A.3: Information criterion statistics for autoregressive distributed lag model† AIC BIC Lt Lt Ut Ut−1 Ut−2 Ut−3 Ut−4 Ut Ut−1 Ut−2 Ut−3 Ut−4 St 146.73 146.73 133.07 38.86∗∗ 43.51 220.01 220.01 212.45 124.34∗∗ 135.08 St St−1 146.73 146.73 133.07 38.86∗∗ 43.51 220.01 220.01 212.45 124.34∗∗ 135.08 St−1 St−2 175.63 175.63 177.48 88.72 93.44 254.55 254.55 262.47 179.76 190.54 St−2 St−3 100.71 100.71 102.53 102.35 107.15 185.15 185.15 193.01 198.86 209.68 St−3 St−4 82.19 82.19 82.86 82.56 84.55 172.05 172.05 178.72 184.41 192.39 St−4 Lt−1 Lt−1 Ut Ut−1 Ut−2 Ut−3 Ut−4 Ut Ut−1 Ut−2 Ut−3 Ut−4 St 146.73 146.73 133.07 38.86∗∗ 43.51 220.01 220.01 212.45 124.34∗∗ 135.08 St St−1 146.73 146.73 133.07 38.86∗∗ 43.51 220.01 220.01 212.45 124.34∗∗ 135.08 St−1 St−2 175.63 175.63 177.48 88.72 93.44 254.55 254.55 262.47 179.76 190.54 St−2 St−3 100.71 100.71 102.53 102.35 107.15 185.15 185.15 193.01 198.86 209.68 St−3 St−4 82.19 82.19 82.86 82.56 84.55 172.05 172.05 178.72 184.41 192.39 St−4 Lt−2 Lt−2 Ut Ut−1 Ut−2 Ut−3 Ut−4 Ut Ut−1 Ut−2 Ut−3 Ut−4 St 126.56 126.56 126.43 35.36 40.03 205.95 205.95 211.92 126.94 137.70 St St−1 126.56 126.56 126.43 35.36 40.03 205.95 205.95 211.92 126.94 137.70 St−1 St−2 172.37 172.37 173.16 86.22 90.97 257.36 257.36 264.22 183.33 194.14 St−2 St−3 96.75 96.75 97.30 99.20 102.99 187.23 187.23 193.81 201.74 211.55 St−3 St−4 77.62 77.62 74.83 76.43 75.86 173.48 173.48 176.68 184.28 189.69 St−4 Lt−3 Lt−3 Ut Ut−1 Ut−2 Ut−3 Ut−4 Ut Ut−1 Ut−2 Ut−3 Ut−4 St 33.41∗∗∗ 33.41∗∗∗ 33.64∗ 34.63 36.00 118.89∗∗∗ 118.89∗∗∗ 125.23∗ 132.32 139.78 St St−1 33.41∗∗∗ 33.41∗∗∗ 33.64∗ 34.63 36.00 118.89∗∗∗ 118.89∗∗∗ 125.23∗ 132.32 139.78 St−1 St−2 80.55 80.55 82.48 82.71 84.68 171.59 171.59 179.59 185.88 193.91 St−2 St−3 96.68 96.68 98.26 99.15 99.57 193.19 193.19 200.80 207.72 214.16 St−3 St−4 78.81 78.81 76.81 78.38 76.22 180.67 180.67 184.65 192.21 196.05 St−4 Lt−4 Lt−4 Ut Ut−1 Ut−2 Ut−3 Ut−4 Ut Ut−1 Ut−2 Ut−3 Ut−4 St 35.54 35.54 34.13 36.05 25.53 127.11 127.11 131.81 139.83 135.41 St St−1 35.54 35.54 34.13 36.05 25.53 127.11 127.11 131.81 139.83 135.41 St−1 St−2 77.10 77.10 78.89 80.78 69.22 174.19 174.19 182.05 190.01 184.52 St−2 St−3 96.93 96.93 98.55 100.47 89.14 199.45 199.45 207.10 215.06 209.75 St−3 St−4 79.18 79.18 78.15 80.10 72.43 187.02 187.02 191.98 199.93 198.24 St−4 † Notes: Akaike (AIC) and Bayesian (BIC) information criteria reported for models with up to four lags for the liquidity (L), solvency (S), and uncertainty (U ) channels for the benchmark model. ∗ indicates third-lowest statistic, ∗∗ indicates second-lowest statistic, and ∗∗∗ indicates lowest statistic, for the corresponding criterion. Multiple asterisks indicate ties. 26 The following tables report, without comment, additional robustness regressions for the au- toregressive distributed lag (Table A.4) and interaction (Table A.5) speci�cations analogous to the robustness checks reported in the main text (Table 2). Note that the ADL(3,0,0) speci�- cation reported in these additional robustness checks were chosen for comparability with the benchmark, and may not reflect the best model selected according to information criteria. The qualitative results are, in the main, unaffected by the alternative speci�cations, although in some cases (notably for the CDS variable), additional coefficients were rendered signi�cant. The total effect in these cases yielded results that were consistent with those reported in the text. 27 Table A.4: Additional robustness regressions for transmission channels associated with foreign claims by international banks in developing countries, 2004Q1–2011Q1 † AR1 AR2 AR3 AR4 AR5 AR6 AR7 AR8 Lagged foreign claims 0.780 0.781 0.782 0.779 0.780 0.782 (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ Libor-OIS -0.017 -0.018 -0.022 -0.052 -0.012 -0.018 -0.022 (0.02) (0.02) (0.02) (0.02)∗∗∗ (0.02) (0.02) (0.02) 1st lag 0.006 0.013 0.016 0.018 -0.004 0.013 0.011 (0.01) (0.01) (0.01) (0.01) (0.02) (0.01) (0.01) 2nd lag 0.032 0.014 0.013 0.002 0.028 0.012 0.017 (0.01)∗∗ (0.01) (0.01) (0.01) (0.01)∗ (0.01) (0.01) 3rd lag -0.034 -0.017 -0.022 -0.018 -0.039 -0.017 -0.023 (0.01)∗∗∗ (0.01) (0.01)∗∗ (0.01) (0.01)∗∗∗ (0.01) (0.01)∗∗ Libor-CD -0.051 -0.009 (0.02)∗∗ (0.02) 1st lag 0.012 (0.01) 2nd lag 0.013 (0.01) 3rd lag -0.003 (0.01) Global bank CDS -0.007 -0.001 -0.001 0.004 -0.002 -0.001 -0.003 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Loss allowances -0.002 (0.02) Asset volatility -0.053 -0.116 -0.061 -0.118 -0.055 -0.068 (0.04) (0.04)∗∗∗ (0.02)∗∗ (0.03)∗∗∗ (0.03)∗∗ (0.02)∗∗∗ 28 Asset volatility -0.096 (unconditional) (0.04)∗∗∗ Risk aversion 0.072 0.304 (0.09) (0.11)∗∗∗ Loan demand -0.016 -0.062 0.015 0.005 0.013 0.001 0.021 -0.004 (0.04) (0.06) (0.04) (0.04) (0.04) (0.04) (0.05) (0.04) Inflation -0.023 -0.022 -0.023 -0.022 -0.024 -0.025 -0.059 (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗ Inflation, YoY -0.034 (0.01)∗∗∗ Growth -0.180 -0.162 -0.235 -0.148 -0.128 -0.191 -0.176 (0.15) (0.15) (0.16) (0.15) (0.15) (0.16) (0.16) Growth, YoY -0.403 (0.40) GDP 0.278 0.276 0.321 0.285 0.269 0.248 0.324 0.289 (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.05)∗∗∗ (0.04)∗∗∗ Depreciation 0.022 (0.00)∗∗∗ Crisis 0.075 0.080 0.074 0.083 0.083 0.041 0.075 0.082 (0.02)∗∗∗ (0.03)∗∗∗ (0.03)∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03) (0.03)∗∗∗ (0.03)∗∗∗ 2 Adj. R 0.776 0.776 0.715 0.776 0.775 0.776 0.714 0.777 R2 (within) 0.776 0.777 0.716 0.777 0.776 0.777 0.716 0.778 Estimator FE FE FE FE FE FE FE FE N 3,429 3,429 3,054 3,429 3,429 3,429 3,054 3,393 † All variables are in log form. Heteroskedasticity and autocorrelation-robust standard errors are reported in parentheses. A constant term was included in the regressions, but not reported. ∗ indicates signi�cance at 10 percent level, ∗∗ indicates signi�cance at 5 percent level, and ∗∗∗ indicates signi�cance at 1 percent level. Table A.5: Additional robustness regressions for transmission channels associated with for- eign claims by international banks in developing countries, 2004Q1–2011Q1 † BR1 BR2 BR3 BR4 BR5 BR6 BR7 BR8 Lagged foreign claims 0.772 0.771 0.772 0.771 0.769 0.770 0.769 0.772 (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ (0.03)∗∗∗ Libor-OIS -0.036 -0.022 -0.028 -0.024 -0.043 -0.027 -0.030 (0.02) (0.02) (0.02) (0.02) (0.02)∗∗ (0.02) (0.02)∗ Libor-OIS × crisis 0.011 0.005 -0.003 -0.075 0.036 -0.001 -0.004 (0.05) (0.04) (0.04) (0.04)∗ (0.04) (0.04) (0.04) Libor-CD 0.015 0.042 (0.06) (0.08) Libor-CD × crisis -0.020 -0.048 (0.06) (0.07) Global bank CDS -0.007 0.003 0.004 0.005 0.001 0.002 0.002 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Global bank CDS × crisis -0.002 -0.010 -0.010 -0.035 -0.016 -0.013 -0.010 (0.02) (0.01) (0.01) (0.02) (0.01) (0.01) (0.01) Loss allowances -0.006 (0.01) Loss allowances × crisis -0.024 (0.03) Asset volatility -0.058 -0.067 -0.069 -0.104 -0.062 -0.063 (0.04) (0.05) (0.04)∗ (0.05)∗∗ (0.04) (0.05) Asset volatility × crisis -0.035 -0.002 0.014 0.004 0.016 0.013 (0.04) (0.07) (0.06) (0.06) (0.07) (0.06) Asset volatility -0.089 (unconditional) (0.07) Asset vol (uncond) × crisis 0.017 29 (0.10) Risk aversion -0.092 0.187 (0.11) (0.10)∗ Risk aversion × crisis 0.380 (0.32) Loan demand 0.001 -0.013 -0.021 -0.009 -0.091 0.027 -0.012 -0.015 (0.05) (0.06) (0.04) (0.04) (0.08) (0.05) (0.04) (0.04) Inflation -0.021 -0.022 -0.022 -0.022 -0.024 -0.023 -0.057 (0.16) (0.15) (0.16) (0.16) (0.16) (0.18) (0.15) Inflation, YoY -0.034 (0.30) Growth -0.166 -0.176 -0.179 -0.165 -0.159 -0.179 -0.190 (0.18) (0.19) (0.16) (0.18) (0.20) (0.15) (0.18) Growth, YoY -0.241 (0.38) GDP 0.294 0.287 0.288 0.277 0.260 0.274 0.281 0.283 (0.04)∗∗∗ (0.05)∗∗∗ (0.04)∗∗∗ (0.05)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗ (0.05)∗∗∗ (0.04)∗∗∗ Depreciation 0.020 (0.09) Crisis 0.167 0.267 0.081 0.155 0.390 0.042 0.159 0.168 (0.16) (0.19) (0.12) (0.10) (0.12)∗∗∗ (0.15) (0.14) (0.14) Adj. R2 0.762 0.762 0.762 0.762 0.762 0.763 0.761 0.763 R2 (within) 0.772 0.772 0.772 0.772 0.772 0.772 0.771 0.773 Estimator FE FE FE FE FE FE FE FE N 3,435 3,435 3,435 3,435 3,435 3,435 3,401 3,399 † All variables are in log form. Bootstrapped standard errors are reported in parentheses. A constant term was included in the regressions, but not reported. ∗ indicates signi�cance at 10 percent level, ∗∗ indicates signi�cance at 5 percent level, and ∗∗∗ indicates signi�cance at 1 percent level.