Geopolitical Risk and Global Banking

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1418 August 2025 Geopolitical Risk and Global Banking Friederike Niepmann, Leslie Sheng Shen Please cite this paper as: Niepmann, Friederike, and Leslie Sheng Shen (2025). “Geopolitical Risk and Global Banking,” International Finance Discussion Papers 1418. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2025.1418. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Geopolitical Risk and Global Banking∗ Friederike Niepmann† Leslie Sheng Shen‡ July 2025 Abstract How do banks respond to geopolitical risk, and is this response distinct from other macroeconomicrisks? UsingU.S.supervisorydataandnewgeopoliticalriskindices,we showthatbanksreducecross-borderlendingtocountrieswithelevatedgeopoliticalrisk but continue lending to those markets through foreign affiliates—unlike their response to other macro risks. Furthermore, banks reduce domestic lending when geopolitical risk rises abroad, especially when they operate foreign affiliates. A simple banking modelinwhichgeopoliticalshocksfeatureexpropriationriskcanexplainthesefindings: Foreign funding through affiliates limits downside losses, making affiliate divestment less attractive and amplifying domestic spillovers. Keywords: geopolitical risk, bank lending, credit risk, international spillovers JEL-Codes: F34, F36, G21 ∗Wethankconferenceandseminarparticipantsforhelpfulconversationsandfeedback,particularlyChristianBlickle,NickBloom,GiovanniDell’Ariccia,S¸ebnemKalemli-O¨zcan,CameliaMinoiu,AndreaPresbitero, Christopher Trebesch, and Andrei Zlate. We also thank Hannah Bensen, Isabel Culver, Caitlin Kawamura, JulianPerry,andDavidSchrammforfantasticresearchassistanceanddatasupport. Theviewsinthispaper are solely the responsibility of the authors and should not necessarily be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System, the Federal Reserve Bank of Boston, or any other person associated with the Federal Reserve System. †Board of Governors of the Federal Reserve System; email: friederike.niepmann@frb.gov. ‡Federal Reserve Bank of Boston; email: lesliesheng.shen@bos.frb.org.

1 Introduction Geopolitical risk has escalated in recent years, fueled by events such as Russia’s invasion of Ukraine, rising tensions between China and the West, and conflicts in the Middle East. The potentially adverse economic consequences of heightened geopolitical risk have become a top concern for policymakers and businesses.1 Yet the academic literature on this subject remains nascent. In particular, the financial and international mechanisms through which geopolitical risk affects economies are not well understood. This paper addresses this gap by analyzing how global banks respond to rising geopolitical risk and the resulting spillover effects. Operating across multiple jurisdictions, global banks are inherently exposed to a rangeofgeopoliticalshocks, which, unlikeotherformsofeconomicrisks, oftenentailuniquely catastrophic outcomes, such as expropriation. At the same time, global banks’ credit supply decisions have material effects on firm investment and employment (see, e.g., Peek and ¨ Rosengren 2000; Khwaja and Mian 2008; Schnabl 2012; Kalemli-Ozcan et al. 2013; Huber 2018). Giventheirglobalreach, thesebankscanserveascriticalconduitsforthepropagation of geopolitical risk, including to countries not directly involved in conflict. In this paper, we investigate how U.S. global banks manage geopolitical risk arising from exposure through their foreign operations, and how this behavior spills over to domestic credit supply. We do so by leveraging multiple sources of confidential supervisory data and both established and newly constructed geopolitical risk indices at the country and bank levels. We begin by establishing three key findings on how geopolitical risk shapes banks’ foreign operations. First, geopolitical risk in countries where banks operate increases the credit risk of those banks’ directly exposed loans and their overall balance sheets. Second, despite heightened credit risk, banks continue lending to countries with elevated geopolitical riskthroughtheirforeignaffiliates,evenastheyreducecross-borderlendingtothosemarkets. In other words, they maintain credit access through local operations while retreating from 1Geopoliticalriskhasbeenarecurringfocusofkeycentralbankpolicymeetingsandspeechessince2019. See, for example, the Federal Reserve’s FOMC meeting minutes and Christine Lagarde’s speech, “Central BanksinaFragmentingWorld,”fromApril17,2023. Ina2022speech,JPMorganCEOJamieDimonstated, “The most important [risk] is the geopolitics around Russia and Ukraine, America and China, relationships of the Western world. That to me would be far more concerning than whether there is a mild or slightly severe recession.” 1

direct cross-border operations.2 Third, this asymmetric response is specific to geopolitical risk, as banks do not adjust foreign operations in the same way to more traditional forms of country risk, such as macroeconomic or sovereign risk. We show that these findings can be explained through a stylized model in which differences in funding structures between cross-border and local operations shape banks’ credit allocation under geopolitical risk. In cross-border lending, banks raise funds domestically, leaving the parent fully liable in the event of loss. By contrast, local operations are at least partly funded through foreign deposits, which do not need to be repaid in the event of expropriation. Based on historical precedent, geopolitical risk heightens the likelihood of expropriation. Governments involved in geopolitical conflicts have, at times, seized foreign assets and extinguished corresponding local liabilities. Even when outright expropriation does not occur, incremental forms of government intervention—such as limits on profit repatriation or imposition of capital controls—can have similar effects. By funding foreign affiliates locally, banks reduce the volume of profits or principal that must be extracted across borders, thereby lowering their exposure to such intervention. As a result, banks with affiliate-based lending can partially offset asset losses through a reduction in liabilities, lowering the net loss from geopolitical risk. This asymmetry in expected losses creates a wedge in returns across modes of foreign operation, leading banks to reduce cross-border lending while maintaining local operations under geopolitical risk.3 Beyondshapingbanks’foreignoperations, geopoliticalriskabroadalsogeneratesspillover effects on domestic credit supply through capital requirements applied at the consolidated level, as predicted by the model and confirmed by our empirical analysis. We find that U.S. global banks reduce commercial and industrial (C&I) lending to domestic firms when 2Banks can extend credit to foreign borrowers either from an office outside the borrower’s country of residence (typically the banks headquarters country), resulting in cross-border claims, or from an office located within the borrower’s country, resulting in local claims. 3AswediscussinSection4,modelsthatdonotrelyonexpropriationriskcanalsogenerateanasymmetric response between cross-border and local lending under geopolitical risk. However, in such frameworks, it is more difficult to rationalize why geopolitical risk is distinct from other forms of country risk. We emphasize expropriation risk because it provides a novel explanation for the observed margin wedge and aligns with the prominence of expropriation-related actions during geopolitical conflicts—such as those seen in Russia’s invasionofUkraine. Thatsaid,whengeopoliticalriskemergesincountrieswhereexpropriationislesssalient, other mechanisms may also be at play, though such episodes may be more muted in geopolitical intensity. This is an area of ongoing investigation in our analysis. 2

geopolitical risk abroad rises. This effect is most pronounced when the risk originates in countries where the banks operate through local affiliates, underscoring how the structure of foreign operations—that is, local affiliate versus cross-border lending—shapes the transmission of geopolitical shocks. Our findings highlight the role internationally active banks play as conduits through which geopolitical instability spills over into domestic credit markets and, hence, the broader economy. We begin the analysis by compiling and constructing indices for country-specific geopolitical risk (CGPR) and bank-specific geopolitical risk (BGPR). For the former, we draw Caldara and Iacoviello (2022)’s index for 44 countries, which is based on a count of mentions of war and related terms in newspaper articles. To complement this measure, we construct a new CGPR index by applying textual analysis with similar terms to firms’ earnings-call transcripts, following the methodology outlined in Hassan et al. (2019, 2023). The earningscall-based index enables us to focus on the geopolitical risks most salient to firms’ perception and to distinguish between country-specific geopolitical risk arising from acts versus threats. Together, these measures provide a more comprehensive picture of country-specific geopolitical risk and enable us to assess the robustness of our findings across distinct sources of information. Compared with broader macro-level risk indicators—such as Hassan et al. (2019, 2023)’s country risk index (CRI) and Ahir et al. (2022)’s World Uncertainty Index— both the CGPR and BGPR indices reveal distinct patterns reflecting the realization and salience of geopolitical events and risks. Equipped with the CGPR indices, we construct BGPR indices that capture individual banks’ exposure to CGPR through their foreign operations. Specifically, we calculate BGPR by weighting each bank’s share of assets in a country by that countrys CGPR index, then summing across all foreign countries (excluding the United States). Data on banks’ foreign exposuresarederivedfromconfidentialFFIEC009reportssubmittedtotheFederalReserve. U.S. banks have substantial exposure to a wide range of countries, with significant variation across countries and over time within each bank. As a result, BGPR varies across banks and over time, providing the identifying variation we exploit to estimate the effects of geopolitical risk on bank behavior. Using the indices, we first examine the effects of geopolitical risk on banks’ credit risk, 3

using data from FR Y-14Q reports, which provide loan-level information on the amount and terms of C&I lending by all banks participating in Federal Reserve stress tests. Based on regressions at the bank-country-time level, we find that the probability of default on loans to a country—as assigned by the banks—increases with rising geopolitical risk in that country. To validate this result, we conduct an event study of two major geopolitical shocks: the Crimea conflict in 2013:Q4 and the Russia–Ukraine war in 2022:Q1. Consistent with our regression findings, we show that the sharp increase in geopolitical risk in Russia following these events led to a significantly greater rise in the default probabilities of loans to Russian borrowers relative to loans to borrowers from other countries. Building on these results, we assess whether the increase in credit risk is large enough to materially affect banks’ overall loan portfolios. Our bank-level analysis shows that as exposure to foreign geopolitical risk rises, theaggregateprobabilityofdefaultinU.S.banks’loanportfoliosincreasessignificantly. In other words, foreign geopolitical risk shocks materially elevate the overall credit risk that U.S. banks face. Next, we investigate how banks adjust their foreign lending in response to rising credit risk, using FFIEC 009 data on banks’ foreign claims by country. We find that U.S. banks’ responses vary systematically by the mode of foreign operation. Regressions at the bankcountry level show that banks reduce their cross-border lending to countries experiencing elevated geopolitical risk, while lending through local affiliates remains largely unchanged. That is, despite heightened credit risk, banks’ lending via foreign subsidiaries is remarkably persistent. This pattern aligns with anecdotal evidence from Russia following its invasion of Ukraine. More than three years after the initial invasion, Citigroup is still winding down its operations in Russia. Meanwhile, two other large internationally active banks, Raiffeisen Bank International (RBI) and UniCredit, continue to operate their Russian subsidiaries, despite mounting political and regulatory pressure to exit. Banks’ behaviors in response to geopolitical risk appear distinct from their reactions to other forms of country risk. We examine how banks adjust their cross-border and local exposurestoincreasesinbroadcountryrisk, usingmeasurescommonlyemployedintheliterature, including Hassan et al. (2023)’s CRI, Ahir et al. (2022)’s WUI, and sovereign credit default swap (CDS) spreads. The first two measures, constructed using a methodology similar to 4

our CGPR indices, capture broad perceptions of risk or uncertainty. Unlike geopolitical risk, which prompts banks to reduce cross-border lending while maintaining local operations, broad country and sovereign risk do not induce similarly asymmetric adjustments. This divergence underscores banks’ unique responses to geopolitical instability. To explain these empirical findings, we develop a stylized model in which a bank allocates investment between domestic and foreign markets, with foreign exposure taking one of two forms: cross-border lending or local affiliate operations. The key distinction is that affiliates raise funds through local deposits, which are not repaid if geopolitical risk materializes, as historical episodes of conflict often heighten the risk of expropriation. When this risk materializes fully, the foreign government may seize the bank’s local affiliate and extinguish its liabilities to local depositors. This asymmetry in liability structure alters banks’ incentives, making affiliate-based lending less exposed to net losses from geopolitical shocks.4 As a result, banks adjust cross-border and affiliate exposures differently in response to geopolitical risk—reducing the former while maintaining the latter. By contrast, broader economic risks, despite potentially generating losses, do not impair the enforceability of foreign liabilities and therefore lead to more uniform adjustments across modes of operation. The model also generates a new prediction about foreign operations: Banks that rely more heavily on foreign funding are less likely to divest from local investments in response to geopolitical risk. We confirm this empirically and further show that, unlike geopolitical risk, local funding positions do not significantly affect how banks adjust foreign exposures to macroeconomic and sovereign risks. In addition to shaping foreign exposures, the model has implications for domestic lending through spillover channels. It predicts that geopolitical risk abroad can tighten domestic credit supply, particularly for banks with affiliate operations in affected countries. To test this, we analyze the effect of geopolitical risk on banks’ domestic corporate loan origination using FR Y-14 data and our BGPR indices. We conduct the analysis at both the loan level, which enables us to control for potential demand-side responses by firms using firm-time fixed effects, and the bank level, to evaluate whether this effect is substantial enough to be 4Milder forms of government intervention—such as financial restrictions that complicate profit repatriation or claim recovery across borders—can also create asymmetry in liability structure and generate similar effects. 5

observed in aggregate. Both analyses show that U.S. banks originate fewer loans to domestic firms in response to an increase in BGPR. We further test the role of banks’ cross-border versus local exposure in driving these spillover effects. We decompose the BGPR indices into two components: one capturing geopolitical risk from countries where banks operate only cross-border, and another from countries where they maintain local affiliates. We find that the effects on domestic loan origination are significant only for BGPR stemming from countries where banks maintain branches or subsidiaries, confirming the model’s prediction and aligning with the earlier finding on the persistence of local claims. Additionally, we examine how banks’ capital positions influence spillover effects. Consistent with the model’s prediction, better-capitalized banks reduce domestic lending less in response to foreign geopolitical risk. Moreover, we find that these spillovers are triggered more by perceived threats than realized events, underscoring the role of uncertainty in the transmission channel and reinforcing the model framework. Because FR Y-14 data cover less than 15 years, we extend our analysis using confidential responses from the Senior Loan Officer Opinion Survey (SLOOS), available since 1990. This surveycapturesbanks’self-reportedchangesincreditstandards—tighteningorloosening—as well as shifts in credit demand. We find that increases in BGPR significantly tighten lending standards for domestic C&I loans—especially for banks with affiliate exposure abroad— further confirming the impact of geopolitical risk on U.S. credit supply. Ourfindingsshowthatgeopoliticalriskabroadcanreducedomesticcreditsupplythrough theglobaloperationsofinternationallyactivebanks. However, thisshouldnotbeinterpreted as evidence that global banking is inherently harmful. The other side of this dynamic is that international linkages allow domestic shocks to be absorbed through foreign operations, so shocks are naturally transmitted in both directions (Shen and Zhang 2024). Furthermore, the international banking literature highlights several benefits of cross-border banking. For instance, banks facilitate the efficient allocation of capital across countries (Niepmann 2015) and export advanced technologies to reduce the cost of financial services (Niepmann 2023). 6

Related Literature. A growing body of literature explores the economic and financial effects of geopolitical risk, following the seminal work of Caldara and Iacoviello (2022) who introduce the geopolitical risk index used in this paper. They show that heightened geopolitical risk reduces aggregate investment and employment. At the firm level, Wang et al. (2019) find that geopolitical risk lowers corporate investment. However, research on banks’ responses to geopolitical risk remains limited. The most closely related study, Pham et al. (2021), finds that Ukrainian banks operating in the conflict-affected regions after 2014 reducedlendingelsewhereinUkraine. DeHaasetal.(2025)findthatbanksreducecross-border lending in response to violent conflicts but increase lending to military-related sectors within the affected countries. Pradhan et al. (2025) also find a reduction in cross-border lending in responsetogeopoliticaltensionsbetweencountries, highlightinginteractioneffectswithmonetary policy. Other studies show that geopolitical risk constrains bank credit growth (Demir and Danisman 2021), weakens bank stability (Phan et al. 2022), and reduces profitability (Alsagr and Almazor 2020), primarily by curbing household lending. Other related work examines the effects of sanctions—a specific policy response to geopolitical events—on bank lending, including Mamonov et al. (2022), Drott et al. (2024), and Danisewicz et al. (2025). Efing et al. (2023)’s particularly relevant study finds that German banks reduced lending to sanctioned countries from domestic operations but not necessarily from foreign affiliates, especially those in jurisdictions with weak enforcement, suggesting an enforcement-avoidance mechanism under targeted sanctions. By contrast, we find no evidence that U.S. banks increased intragroup lending to affiliates in countries with heightened geopolitical risk. Instead, we highlight a distinct internal mechanism—rooted in funding structure and capital regulation—that explains the persistence of affiliate exposures. Thus, our results highlight a new and complementary channel through which geopolitical risk shapes global banking. Beyond banking, research on the economic effects of geopolitical power and risk has focused on the impact of geopolitical events—particularly the U.S.–China trade war—on global supply chains (see, e.g., Amiti et al., 2020, Fajgelbaum et al., 2020, Fajgelbaum et al., 2021, Alfaro and Chor, 2023). Clayton et al. (2023) develop a model explaining how geopolitical power and economic coercion shape global financial and real activity. 7

In addition to the literature on geopolitical risk, our paper contributes to research on the international transmission of shocks through global banks (see, e.g., Peek and Rosengren, 2000, Schnabl, 2012, Cetorelli and Goldberg, 2012a, Ivashina et al., 2015, Hale et al., 2020, Shen and Zhang, 2024). Methodologically, our approach is similar to that of Temesvary and Wei (2024), who show that U.S. banks with greater exposure to foreign markets affected by COVID-19 reduced domestic C&I lending more sharply. Related work also examines how different forms of global uncertainty influence credit supply. For instance, Correa et al. (2023) analyze how U.S. banks’ exposure to trade uncertainty through their borrowers influences bank lending, while Federico et al. (2025) show that trade shocks can trigger broad contractions in lending by raising non-performing loans. A relevant theme in this literature is that the mode of foreign operations influences the transmission of shocks. Fillat et al. (2023) find that shock transmission is stronger through branchesthansubsidiariesduetodifferencesinfundingstructures. Dell’AricciaandMarquez (2010) argue that higher expropriation risk makes subsidiaries less attractive in politically unstable countries. However, we find that the branch-versus-subsidiary distinction does not play a central role in shaping banks’ responses to geopolitical risk. Instead, we highlight the broader distinction between cross-border and local affiliate lending—whereby the latter encompasses both branches and subsidiaries—as the key margin along which banks adjust exposures, with important implications for spillovers.5 Our paper also contributes to the literature on risk and capital flows (see, e.g., Rey, 2016, ¨ Kalemli-Ozcan, 2019, Jiang et al., 2020, Akinci et al., 2022). Hassan et al. (2023) construct country risk measures from firms’ earnings call transcripts and show that heightened risk reduces capital flows. We build on this approach by applying similar textual analysis to developa newgeopoliticalrisk measure. Relatedworkexamineshow riskaffects cross-border bank lending (e.g., Correa et al., 2022, Bruno and Shin, 2015). Choi and Furceri (2019) find that rising country-level uncertainty reduces both cross-border lending and borrowing from affected countries. 5Several papers examine how banks’ responses to shocks differ depending on their mode of operating abroad and the importance of the lending market to banks. See, for example, Cetorelli and Goldberg (2012b), De Haas and Van Horen (2013), Claessens and Van Horen (2012), and Claessens and Van Horen (2015). Schnabl (2012) finds that the transmission of liquidity shocks from the parent bank is weaker to its foreign subsidiaries than through its direct cross-border lending to foreign banks. 8

2 U.S. Banks’ Exposure to Geopolitical Risk 2.1 U.S. Banks’ Foreign Operations U.S. banks are exposed to geopolitical risk abroad through their foreign operations. To understand the extent of this exposure, we examine data from the FFIEC 009 report, which provides detailed information on U.S. banks’ foreign assets and liabilities by country.6 The FFIEC 009 reporters consist of U.S. banks, bank holding companies (BHCs), and intermediate holding companies (IHCs) holding $30 million or more in claims on residents of foreign countries. We focus on reporters whose ultimate parent bank is in the United States, relying on information from the National Information Center to identify each reporter’s ultimate parent bank and its location. Our sample runs from 1986:Q1 to 2022:Q4 and consists of 67 banks in an average period. Figure 1 illustrates the size, mode, and geographical distribution of U.S. banks’ foreign operations. Panel (a) of Figure 1 shows that the share of U.S. banks’ foreign assets in total assets averages about 20 percent over the sample period. Larger banks tend to be the most internationally active (Buch et al., 2011; Niepmann, 2023), contributing disproportionately to this aggregate share. Panel (b) illustrates the mode of U.S. banks’ foreign operations. It displays the share of foreign exposures held in foreign offices (either branches or subsidiaries), referred to as “local exposures.” The remaining share, known as “cross-border exposures,” represents the share of foreign exposures whereby the U.S. parent offices lend directly to foreign residents.7 The figure shows that approximately half of U.S. banks’ operations are conducted through offices abroad, while the other half comprises cross-border operations. The share of foreign operations conducted through local operations increased up to the Global Financial Crisis and declined to about 45 percent in the subsequent years. Panels (c) through (f) of Figure 1 provide snapshots of the geographical distribution 6Inthispaper, weusetheterms‘foreignclaims,’ ‘foreignexposures,’ and‘foreignassets’interchangeably. 7To be more precise, cross-border exposures are claims held by bank offices that are outside the country of residence of its counterparty. For example, U.S. Bank A generates a cross-border claim on Mexico when itextendsaloanfromitsU.S.officetoaMexicanresident. Localexposuresareclaimsextendedbyabank’s local offices, whether subsidiary or branch, in a foreign country to residents of that country. For example, BankAgeneratesalocalclaimonRussiawhenitlendstoaRussianresidentthroughitsRussiansubsidiary. 9

Figure 1: U.S. Banks’ Foreign Operations (a) Foreign Exposures as a Share of Total Assets .4 .3 .2 .1 0 stessA latoT ni serusopxE ngieroF fo erahS (b) Local Exposures as a Share of Foreign Exposures .8 .6 .4 .2 0 1990q1 2000q1 2010q1 2020q1 )erahS lacoL( seciffO ngieroF ni dleH erahS 1990q1 2000q1 2010q1 2020q1 (c) Distribution of Foreign Exposure by Region, 2010:Q4 6 4 2 0 ytisneD lenreK (d) Distribution of Foreign Exposure by Region, 2019:Q4 Europe 6 Asia Latin America Rest of World 4 2 0 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Share of Foreign Claims in Region ytisneD lenreK Europe Asia Latin America Rest of World 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Share of Foreign Claims in Region (e) Distribution of Foreign Claims by Country for Selected Banks, 2010:Q4 25 20 15 10 5 0 Bank of Citigroup Goldman State NY Mellon Sachs Street stessA latoT fo tnecreP (f) Distribution of Foreign Claims by Country for Selected Banks, 2019:Q4 25 20 N Au e s th tr e a r l l i a a nds 15 France United Kingdom India Germany Japan 10 China 5 0 Bank of Citigroup Goldman State NY Mellon Sachs Street stessA latoT fo tnecreP Australia Canada France United Kingdom South Korea Germany Japan Belgium Mexico Note: Panel(a)ofthefigureshowsU.S.banks’averageforeignexposuresasashareoftotalassetsfrom1990:Q1to2021:Q4. Panel (b) shows U.S. banks’ local exposures, or exposures through foreign offices, as a share of their total foreign exposures. Panels(c)and(d)illustratethekerneldensityoftheshareofforeignoperationsinfourregions—Europe,Asia,LatinAmerica, andtherestoftheworld—in2010:Q4and2019:Q4,respectively,acrossU.S.banks. Panel(e)and(f)illustratethetopcountries byforeignclaimssize(expressedasashareoftotalassets)in2010:Q4and2019:Q4,respectively,forfourselectedU.S.banks. Data source(s): FFIEC 009, FR Y9-C, and Call Reports for Panels (a)–(d); public version of FFIEC 009/009a for Panels (e)–(f). 10

of U.S. banks’ foreign operations around the world. Panels (c) and (d) display the kernel density of the share of foreign operations across four regions—Europe, Asia, Latin America, and the rest of the world—in 2010:Q4 and 2019:Q4, respectively, across U.S. banks. Across all regions, there is significant heterogeneity in the extent of exposure among banks. For example, in 2010:Q4, roughly the same number of banks had nearly zero exposure as had 60 percent of their total exposure to Europe. Moreover, this degree of heterogeneity changes over time. By 2019:Q4, fewer banks had more than 60 percent of their exposure in Europe. Panels (e) and (f) provide more granular snapshots of the geographical distribution of foreign claims for selected banks, displaying their top five countries of exposure in 2010:Q4 and 2019:Q4, using the public version of the FFIEC 009/009a data.8 These snapshots reveal substantial variation across banks in both the geographical composition and the magnitude of their foreign exposure. Moreover, both the origins and magnitudes of exposure shift over time within individual banks, reflecting the fluid nature of foreign banking operations. Overall, Figure 1 demonstrates that U.S. banks have substantial exposure to a diverse range of countries worldwide, with a significant portion of this exposure stemming from their operations within these countries. These foreign operations expose them to global geopolitical risks. Moreover, since the origin and magnitude of these exposures vary markedly among banks, there is considerable variation in their exposure to geopolitical risk, and this variation also changes over time with each bank. These cross-sectional and time-series variations in foreign exposure are incorporated into the bank-specific measures of geopolitical risk we subsequently construct and play a key role in the identification strategy we apply in the empirical analysis. 2.2 Constructing and Dissecting Geopolitical Risk Indices Constructing the BGPR index. To measure the extent of U.S. banks’ exposure to geopolitical risk through their foreign operations, we construct a bank-specific geopolitical risk index. This BGPR index captures the geopolitical risk each bank faces based on the 8The public version of the FFIEC 009/009a data provides information on material foreign country exposures,definedasexposuresexceeding1percentoftotalassetsor20percentofcapital,whicheverislower,for U.S. banks filing the FFIEC 009 report. Reporting institutions must also disclose a list of countries where their lending exposures exceed 0.75 percent of total assets or 15 percent of total capital, whichever is lower. 11

geography of its foreign lending activities. For each bank b and quarter t, we calculate the index by weighting the geopolitical risk of country c (CGPR) by the share of the bank’s total assets exposed to that country. We then sum the weighted CGPR indices over all countries. Specifically, we compute: (cid:88) BGPR = ω CGPR , (1) bt bct−1 ct c where (cid:32) (cid:33) 4 1 (cid:88) exp bct−i ω = , bct−1 (cid:80) 4 asset i=1 c bct−i and exp denotes bank b’s total exposure in country c, encompassing both cross-border and bc local claims that the bank has toward the residents of the respective country. As defined in Equation (1), the BGPR index is more sensitive to changes in geopolitical risk in country c when bank b has a larger operation in that country.9 CGPR indices. A key component of the BGPR index is CGPR, for which we use two measures. The first is from Caldara and Iacoviello (2022), who construct a country-specific geopolitical risk index for 44 countries (including the United States). We use the authors’ recent CGPR index, which is based on ten newspapers and begins in 1985, rather than the “historical” index, which is based on three newspapers and available from 1900 onward. This set of indices captures perceptions of geopolitical risk from media coverage, reflecting how geopolitical events are reported and emphasized across different news sources over time. We denote Caldara and Iacoviello (2022)’s CGPR index as CGPRN. WeconstructasecondmeasureofCGPRtocapturefirms’perceptionsofgeopoliticalrisk, building on Hassan et al. (2019, 2023)’s natural language processing method. This approach uses the NL Analytics platform, developed by the authors’ team, to apply textual analysis to nearly 400,000 earnings-call transcripts from about 14,000 public companies worldwide, startingin2002. AcrucialstepinconstructingtheCGPRindexinvolvesidentifyinginstances in which conference call discussions focus on geopolitical risk in particular countries. To do 9We have also used variants of this index to assess the robustness of our results. We alter the way of computing the weights (ω ) by normalizing the exposure of bank b in a country by total foreign claims bct (instead of total assets) and by using one-quarter lagged exposure shares as weights (instead of averaging bank exposure shares over the previous four quarters). When normalizing by total foreign claims, we use exposure to all 43 foreign countries for which Caldara and Iacoviello (2022)’s CGPR index is available. 12

this, we compile a dictionary of words associated with geopolitical threats and actions, as well as a database of terms identifying the 43 foreign countries of interest, primarily major cities. To count toward our measure of geopolitical risk for a given country, words from both sets must appear in the same sentence. The dictionary of geopolitical risk-related words is extracted from Caldara and Iacoviello (2022) to allow for a close alignment with CGPRN. Appendix Table A.1 lists the search query for geopolitical risk, which is organized into eight categories. Following Caldara and Iacoviello (2022), each category includes a search query consisting of two sets of words: The first set contains topic words (e.g., “war,” “military,” “terrorist”), and the second set contains “threat” words for five categories and “act” words for three categories. Specifically, we construct the CGPR index based on earnings-call transcripts, denoted as CGPRT, as follows: 1 (cid:88) GPRCount CGPRT = fct , ct F N ct ft f where GPRCount denotes the number of geopolitical risk-related sentences in the tranfct script of firm f pertaining to country c at time t, N denotes the total number of sentences ft in the earnings-call transcript of firm f at time t, and F denotes the number of firms in ct country c at time t. The index is designed to be flexible, enabling closer examinations of various dimensions of geopolitical risk for a given country. For instance, we decompose the index into two components: geopolitical risk arising from threats (CGPRT(Threat)) and from ct acts (CGPRT(Act)). We also construct a sub-index focused on the geopolitical risk perceived ct by financial firms (CGPRTfin). ct We construct BGPR indices using both CGPRN and CGPRT. The index based on CGPRN serves as our baseline measure of geopolitical risk due to its longer sample period starting in 1985. The index based on CGPRT is used to assess the robustness of our results and to further explore how the components of geopolitical risk drive these results, utilizing the various sub-indices of CGPRT that we construct. Panel(a)ofFigure2showsthetwoCGPRindices, aggregatedtothegloballevel(GGPR) and normalized by their respective standard deviations within the sample, from 2002:Q1 to 2023:Q4. GGPRN (top) and GGPRT (bottom) both spike around the onset of three major 13

geopolitical events: the Iraq War in 2003:Q1, the Russia–Ukraine War in 2022:Q1, and the Israel–Hamas War in 2023:Q4. We compare these geopolitical risk indices to two wellknown risk indices: Hassan et al. (2023)’s CRI and Ahir et al. (2022)’s WUI. The CRI is a measure of broad risk perception constructed using the same data and methodology as our CGPRT index; the WUI is a measure of uncertainty constructed by counting the frequency of synonyms for risk or uncertainty using the country reports of the Economist Intelligence Unit. As shown in Panel (b) of Figure 2, both the CRI and WUI spike primarily during periods of significant economic uncertainty, including the height of the Global Financial Crisis around 2008:Q4, the peak of the European sovereign debt crisis in 2011, and the onset of COVID-19 in 2022:Q1. The correlations between the GGPR indices and these two broad risk indices are either low or negative, suggesting that the geopolitical risk captured by CGPRN and CGPRT is a distinct form of risk. We also compare the CGPR indices to other risk indices at the country level. Appendix Figure A.1 shows these indices for three countries: Poland (Panel (a)), the United Kingdom (Panel (b)), and South Korea (Panel (c)). Charts in the left panel illustrate CGPRN (top), CGPRT (middle), andCGPRT(Fin) (bottom), whiletherightpaneldisplaysthreebroadrisk indices for these countries: CRI, WUI, and 5-year sovereign CDS spreads. Similarly to the aggregated global indices, the CGPR indices show sharp increases around significant adverse geopolitical events, including the Russia–Ukraine War that started in 2022 for Poland, a series of terrorist incidents in London in 2005 and 2007 for the United Kingdom, and periods of heightened geopolitical tensions in South Korea due to North Korea’s withdrawal from the Nuclear Nonproliferation Treaty in 2003 and missile tests in 2017. Notably, many of these events are specific to the respective country rather than global (e.g., the CGPR indices for South Korea did not spike with the outbreak of the Russia–Ukraine War). By contrast, the broad risk indices for these countries spike primarily during major economic crises, many of which are global. These examples further highlight that our geopolitical risk indices capture a distinct form of risk. Based on Equation (1), we construct BGPR indices using CGPRN and CGPRT, producing BGPRN and BGPRT, respectively. Appendix Figure A.2 illustrates these two indices 14

Figure 2: Global Geopolitical Risk and Other Risk Indices Iraq War Russia-Ukraine War Israel-Hamas War )dts( NRPGG 6 4 2 0 2- 4- (a) GGPR Indices Global Financial Covid Crisis European debt crisis & stock fall )dts( IRC 6 4 2 0 2- 4- (b) Other Risk Indices Iraq War Russia-Ukraine War Israel-Hamas War )dts( TRPGG 6 4 2 0 2- 4- European debt crisis Covid US fiscal cliff 2000q1 2005q1 2010q1 2015q1 2020q1 2025q1 )dts( IUW 6 4 2 0 2- 4- 2000q1 2005q1 2010q1 2015q1 2020q1 2025q1 Note: Panel(a)showstwoglobalgeopoliticalrisk(GGPR)indices,whichareaggregatedfromcountry-specificgeopoliticalrisk (CGPR) indices, covering the period from 2002:Q1 to 2023:Q4. The top chart displays GGPR from Caldara and Iacoviello (2022)(GGPRN),andthebottomchartdisplaysGGPRconstructedbyapplyingtextualanalysistoearnings-calltranscripts usingtheNLAnalyticsplatform(GGPRT). Panel(b)showstheaggregatedcountryriskindex(CRI)byHassanetal.(2023) (top), and the World Uncertainty Index (WUI) by Ahir et al. (2022) (bottom). All the indices are standardized by their respectivestandarddeviationswithinthesample. 15

at the 25th, 50th, and 75th percentiles over time. The differences among these percentiles reveal significant variation in the level of the index across banks, driven by the heterogeneity in the geography of U.S. banks’ foreign operations. Furthermore, these cross-sectional differences evolve substantially over time across banks. 2.3 Additional Data Sources Given that the goal of our analysis is to understand the effect of geopolitical risk on U.S. banks’ foreign and domestic operations, we construct variables that capture the outcomes of interest. To do this, we utilize a variety of regulatory datasets collected by the Federal Reserve. Bank foreign exposure by country. We use the FFIEC 009 data, which were also used to construct our geopolitical risk indices, to capture the margins of foreign exposure adjustment in response to geopolitical risk. These margins of adjustment include exposure through cross-border and local claims. Loan-level data. For more granular information on U.S. banks’ foreign and domestic operations, we use quarterly loan-level data from the FR Y-14 reports. These reports have beenfiledconfidentiallybyallBHCsparticipatinginofficialFederalReservebankstresstests since late 2012. The participating institutions report detailed information on individual C&I loans exceeding $1 million, including the borrower’s name, country, and industry, as well as the loan amount, origination date, and the probability of default assigned by the bank.10 The probability of default information allows us to study how geopolitical risk affects U.S. banks’ assessment of credit risk for exposed loans. Additionally, the loan origination data enables us to analyze the transmission of geopolitical risk to domestic lending. Bank lending standards. We use data from the Federal Reserve’s Senior Loan Officer Opinion Survey to construct additional outcome variables related to U.S. banks’ lending 10Notably, this dataset includes loans extended through banks’ foreign offices, including foreign subsidiaries. However, we cannot distinguish between loans held by the parent bank and those held by foreign subsidiaries. As a result, we are unable to separate loan exposures into cross-border and local exposures in this dataset. 16

standards. In the quarterly survey, the Federal Reserve asks banks about changes in their lending standards and the demand for credit over the previous three months. The aggregate results are published on the Federal Reserve’s website, while bank-level responses from 1990 onward are available to researchers in the Federal Reserve System. Banks’ responses are recorded on a scale of one to five. Following standard practice in the literature, we transform these responses into three outcome categories: 1 = loosening, 0 = unchanged, and -1 = tightening. To map SLOOS reporters with corresponding FFIEC 009 reporters, we identify whether a SLOOS-reporting entity is a subsidiary of a BHC that reports the FFIEC 009. If so, we aggregate the responses of all loan officers within that BHC. We focus on lending standards for C&I loans to large and medium-sized enterprises, in line with the predominant loan composition in the FR Y-14 data. Bank balance sheet information. We supplement our database with quarterly balance sheet data from FR Y-9C and Call Reports, which provide detailed information on the income statements and balance sheets of all U.S. banks. Using these data, we construct a set of bank-level control variables for our regressions, including a bank’s Tier 1 capital ratio and liquid-asset ratio.11 Macro, financial, and other data. In addition to bank-level information, we construct country-level macro and financial variables from a variety of data sources for use as control variables. This includes countries’ stock price indices and exchange rates from Bloomberg, sovereign CDS spreads from IHS Markit, and sanction status from the Global Sanctions Database. 3 Geopolitical Risk & U.S. Banks’ Foreign Operations In this section, we examine how geopolitical risk abroad affects banks’ foreign exposures and how they adjust in response. We present three key findings: (i) Geopolitical risk increases 11Theliquid-assetratioiscalculatedas(CashandBalancesDuefromDepositoryInstitutions+Availablefor-sale Debt Securities + Held-to-maturity Securities at Amortized Cost) / Total Assets. 17

the credit risk of U.S. banks with foreign operations; (ii) these banks continue to lend to high-risk countries, despite rising credit risk, through their branches and subsidiaries, while reducing cross-border lending to these countries; and (iii) banks do not adjust their foreign exposures in a similarly asymmetric way in response to other types of risk. 3.1 Geopolitical Risk and Credit Risk When a country’s geopolitical risk increases, the credit risk associated with banks’ claims on that country is likely to rise as well. In response, banks are expected to assign a higher probability of default to their exposures to borrowers from that country. We begin our analysis by testing this conjecture, using data from the FR Y-14 reports for the sample period 2013:Q1 through 2022:Q4. Bank-country level evidence. We first conduct the analysis at the bank-country level. Using the quarterly FR Y-14 data, we compute the average probability of default (PD) of C&Iloanstocountrycheldbybankbattimet. ThePDsareweightedbyloansize, usingthe committed loan amounts. To isolate changes in the probability of default for existing loans— rather than shifts driven by banks originating safer loans—we exclude loans originated in quarter t. Withtheweighted-averagePDvariable, westudytherelationshipbetweenCGPRindices and credit risk at the bank-country-time level using the specification: ln(PD ) = βCGPR +α +α +(cid:15) , (2) bct ct bt bc bct where PD denotes the weighted average probability of default assigned by bank b to loans bct to residents of country c at time t, CGPR denotes CGPRN or CGPRT, and α and α bt bc denote bank-time and bank-country fixed effects, respectively. Standard errors are clustered at the country-time level. Columns (1) and (2) of Table 1 present the results. Banks assign higher probabilities of default to existing loans made to borrowers in countries with increasing geopolitical risk, as measured by either CGPRN or CGPRT. A one-standard-deviation increase in CGPR 18

Table 1: Geopolitical Risk and Credit Risk Bank-country Level Bank Level ln(PD ) (1) (2) (3) (4) bct/bt CGPRN 0.100∗∗ ct (0.040) CGPRT 0.076∗∗ ct (0.032) BGPRN 0.134∗∗∗ bt (0.024) BGPRT 0.215∗∗∗ bt (0.042) Bank-country FE Yes Yes No No Bank-time FE Yes Yes No No Bank FE No No Yes Yes Time FE No No Yes Yes Observations 9588 8890 411 411 R2 0.680 0.679 0.871 0.871 Note: Thistablereportsregressionswithlogaverageweightedprobabilityofdefault(PD)asthedependentvariableusingdata fromFRY-14forthesampleperiod2013:Q1through2022:Q4. Columns(1)and(2)reportresultsfromregressionsatthebankcountry-timelevelbasedonEquation(2). CGPRN denotesthe(recent)country-specificgeopoliticalriskindexfromCaldara andIacoviello(2022). CGPRT denotesthecountry-specificgeopoliticalriskindexconstructedbasedonearningscalltranscripts usingtheNLAnalyticsplatform. Columns(3)and(4)reportresultsfromregressionsatthebank-timelevelbasedonEquation (4). BGPRN andBGPRT denotethebank-specificgeopoliticalriskindicesbasedonCGPRN andCGPRT,respectively. All thegeopoliticalriskindicesarestandardizedbytheirrespectivestandarddeviationswithinthesample. Standarderrors,shown inparentheses,areclusteredatthecountryandtimelevelinColumns(1)and(2)andthebankandtimelevelinColumns(3) and(4). *p<.1;**p<.05;***p<.01. 19

raises the weighted average probabilities of default of these loans by 8 to 10 percent. These results support the conjecture that banks perceive higher credit risk in loans to borrowers from countries facing rising geopolitical risk. Event study. To further investigate how banks adjust their assigned probabilities of defaultinresponsetoincreasinggeopoliticalrisk,weconductaneventstudyfocusedonRussia’s annexation of Crimea in 2013:Q4 and its invasion of Ukraine in 2022:Q1. These two major geopolitical shocks provide a natural setting to analyze how banks reassess the credit risk of their outstanding exposures to Russia relative to other countries. Specifically, we run the regression: (cid:88) (cid:88) ln(PD ) = δ Dk + δ Dk ×R +θ +γ +(cid:15) , (3) bct 0k t 1k t c bc bt bct k≥−m k≥−m where PD denotes the average probability of default of loans of bank b in country c at time bct t, Dk denotes dummy variables that take the value 1 if the geopolitical risk shock occurred t k quarters following the event and 0 otherwise, R denotes dummy variables that take the c value 1 if the borrower country is Russia and 0 otherwise, θ denotes bank-country dummies, bc and γ denotes bank-time dummies.12 The coefficients δ capture the differential effect of bt 1k the two Russia-related geopolitical risk shocks on the average probability of default of loans to Russia compared with loans to other countries in the k quarters following the shocks. For this analysis, we restrict the loan sample to all ongoing loans by U.S. banks that have foreign claims on Russia. Figure 3 plots the coefficients δ from Equation (3). It shows that the credit risk of 1k the loans to Russian borrowers increased significantly more than that of loans to borrowers from all other countries in response to the two adverse geopolitical risk shocks. While credit risk did not significantly change across countries on average in the post-shock period, we observe a sharp increase in the average probability of default of outstanding loans to Russian borrowers in the quarter immediately following the shock, and this effect persists for several additional quarters. The magnitude of the increase three quarters after the shock is about 12WealsorantheregressionwithR takingthevalue1iftheborrowercountryiseitherRussiaorUkraine. c The results remain largely unchanged, primarily because U.S. banks have limited exposure to Ukraine. 20

two standard deviations of the average probability of default measure, or 20 basis points. This result further confirms that banks attribute greater credit risk to their exposures to borrowers from countries facing escalating geopolitical risk. Figure 3: Geopolitical Risk and Credit Risk: Russia-Ukraine Conflicts 4 3 2 1 0 -1 tneciffeoC detamitsE -4 -3 -2 -1 0 1 2 3 4 Quarters since GPR shock Note: The figure illustrates the effect of geopolitical risk shocks from the Crimea conflict in 2013:Q4 and the Russia-Ukraine war in 2022:Q1 on the log average probability of default of loans to Russian borrowers relative to loans to borrowers in other countries. Itplotsthecoefficientsδ fromEquation(3). Standarderrors,showninparentheses,areclusteredatthecountry- 1k timelevel. Datasource: FRY-14. Aggregate bank-level evidence. Given the bank-country-level and event study evidence, a key question is whether the increases in credit risk following adverse geopolitical risk shocks are substantial enough to materially affect banks’ aggregate loan portfolios. To address this, we assess whether an increase in BGPR predicts a rise in the probability of default of a bank’s aggregate C&I loan portfolio. Specifically, we compute the weightedaverage probability of default for each bank b’s entire C&I loan portfolio in quarter t. We then regress the measure (in log) on the BGPR indices, controlling for bank characteristics, bank fixed effects, and time fixed effects: ln(PD ) = βBGPR +γX +α +α +(cid:15) , (4) bt bt bt b t bt where BGPR denotes BGPRN or BGPRT, and X denotes bank-level control variables bt bt bt bt including a bank’s lagged Tier 1 capital ratio and liquid-asset ratio. Columns (3) and (4) of Table 1 report the results. An increase in BGPR, as measured 21

by either BGPRN or BGPRT, significantly increases the aggregate probability of default of bank loans. A one-standard-deviation increase in BGPR raises the probability of default of a bank’s C&I loan portfolio by 13 to 22 percent. Taken together, the evidence at the bank-country level, from specific events, and at the bank level shows robustly that banks assign a higher probability of default to their exposures to borrowers from countries experiencing increasing geopolitical risk, and that the increase in credit risk is substantial enough to materially affect banks’ aggregate loan portfolios. 3.2 Geopolitical Risk and Banks’ Foreign Operations How do banks respond to the increased riskiness of their loan portfolios as a result of rising geopolitical risk? Do they de-risk? We investigate how banks adjust their foreign exposures in response to increasing geopolitical risk in the countries where they operate, using the FFEIC 009 data for the sample period 1986:Q1 through 2022:Q4. Specifically, we run the following regression: ln(exp ) = β CGPR +β CGPR +β X +β X +α +α +(cid:15) , (5) bct 1 ct 2 ct−1 2 ct 3 ct−1 bt bc bct whereexp representsameasureofbankb’sexposuretocountrycinquartert, andCGPR bct ct stands for CGPRN or CGPRT. We include both the contemporaneous and one-quarter lagged values of CGPR.13 X captures country-level macro control variables, including ct the log of the exchange rate of country c’s currency vis-a`-vis the U.S. dollar, the log of country c’s main stock price index, and an indicator variable equal to 1 if the country faces any sanctions from the United States. We also control for bank-time fixed effects (α ) to bt account for changes in banks’ foreign exposures common to all countries, and bank-country fixed effects (α ) to account for level differences in exposures of banks across countries. bc Standard errors are clustered by country and time. Table 2 reports the results with CGPRN as the main regressor. Columns (1) and (2) present results from regressions with banks’ log total foreign exposures as the dependent 13Coefficients for additional lags of CGPR are not statistically significant. 22

variable. Columns (3) and (4) and Columns (5) and (6) are based on log cross-border and localexposuresasthedependentvariables, respectively. AsdescribedinSection2, bankscan extend credit to foreign borrowers through two modes of operation: from an office outside the borrower’s country of residence, resulting in cross-border claims, or from an office located in the borrower’s country, resulting in local claims. The odd-numbered columns show the baseline results, and the even-numbered columns add country-level macro controls. The results show that while banks reduce their total exposure to countries experiencing increasing geopolitical risk, their reallocation behavior varies significantly depending on their mode of operation in the affected country. While banks reduce cross-border exposures to countries facing escalating geopolitical risk, their operations through local offices in those countriesremainlargelyunchanged.14 Aone-standard-deviationincreaseinCGPRN reduces cross-border exposure by 6 percent (Column 4). By contrast, the corresponding coefficients for local claims are small and not statistically significant (Column 6).15 The results are quantitatively and qualitatively similar with CGPRT as the main regressor, as shown in Appendix Table B.1. Additional evidence. The distinction between cross-border retrenchment and local operation persistence is evident in banks’ responses to Russia’s 2022 invasion of Ukraine. At the time of the invasion in February 2022, several large global banks were running significant operations in Russia, including operations through local subsidiaries. UniCredit, RBI, Societe Generale, and Citigroup were among those with the largest exposures. Yet despite 14The effect becomes even stronger when earlier years are excluded from the sample. After 1999, the negativeimpactofgeopoliticalriskoncross-borderclaimsisbothlargerinmagnitudeandmorestatistically significant, driven primarily by stronger effects on claims in emerging markets. 15Appendix Table B.2 further separates local claims exposures into those denominated in local currency and in foreign currency (primarily U.S. dollars) to examine whether they respond differently to geopolitical risk. When geopolitical risk rises, the local currency typically depreciates, reducing the U.S. dollar value of local currency-denominated claims without necessarily affecting banks’ local operations. The results align withthisexpectation: Localclaimsinforeigncurrencyshownosignificantresponsetogeopoliticalrisk,while there is some evidence that local currency-denominated claims decline, likely due to exchange rate effects. We also examine how the mode of banks’ local operations in foreign countries (branch versus subsidiary) influences their response to rising geopolitical risk. We find that banks with a higher share of assets in subsidiaries, relative to branches, reduce local claims less but cut cross-border claims more. However, further analysis of how geopolitical risk affects the size of branch versus subsidiary assets suggests that this distinction does not play a central role in shaping banks’ responses to geopolitical risk. In addition, we find no evidence that banks respond by increasing intragroup lending to affiliates in countries with heightened geopolitical risk. 23

Table 2: Response of Banks’ Foreign Operations to Geopolitical Risk Total Cross-border Local ln(exp ) (1) (2) (3) (4) (5) (6) bct CGPRN -0.018∗∗ -0.022∗∗∗ -0.026∗∗∗ -0.031∗∗∗ 0.011 0.010 ct (0.007) (0.008) (0.008) (0.008) (0.015) (0.015) CGPRN -0.010 -0.010 -0.014 -0.013 0.012 0.009 ct−1 (0.008) (0.008) (0.009) (0.009) (0.014) (0.014) 1(Sanction) 0.007 -0.020 -0.009 t (0.017) (0.018) (0.027) ln(Exch.Rate) -0.002 0.004 -0.187∗ t (0.025) (0.025) (0.109) ln(StockIndex) -0.125∗∗∗ -0.117∗∗ -0.113 t (0.046) (0.046) (0.088) ln(Exch.Rate) -0.064∗∗ -0.068∗∗ 0.129 t−1 (0.032) (0.032) (0.106) ln(StockIndex) 0.152∗∗∗ 0.146∗∗∗ 0.213∗∗ t−1 (0.049) (0.049) (0.086) Bank-country FE Yes Yes Yes Yes Yes Yes Bank-time FE Yes Yes Yes Yes Yes Yes Observations 137312 108303 135803 106891 34801 31039 R2 0.894 0.906 0.875 0.887 0.878 0.885 Note: This table reports results from regressions at the bank-country-time level based on Equation (5) using the FFEIC 009 dataforthesampleperiod1986:Q1through2022:Q4. CGPRN denotesthe(recent)country-specificgeopoliticalriskindexfrom CaldaraandIacoviello(2022). ThedependentvariableisthelogtotalforeignclaimsinColumns(1)and(2),logcross-border claimsinColumns(3)and(4),andloglocalclaimsinColumns(5)and(6). Columns(1),(3),and(5)showthebaselineresults for each dependent variable. Columns (2), (4), and (6) add country-level macro controls, including a country’s log exchange rate vis-`a-vis the U.S. dollar, log domestic stock price index, and an indicator variable equal to 1 if the country faces any sanctions from the United States. All regressions include bank-country and bank-time fixed effects. CGPRN is standardized byitsrespectivestandarddeviationwithinthesample. Standarderrors,showninparentheses,areclusteredatthecountryand timelevel. *p<.1;**p<.05;***p<.01. 24

the geopolitical turmoil, most continued operating their local affiliates.16 UniCredit, RBI, and Citigroup have deliberately reduced their cross-border operations with Russia while continuing to operate their Russian subsidiaries, consistent with the empirical evidence presented earlier. UniCredit’s and RBI’s 2022:Q2 earnings presentations explicitly describe this strategy: Both banks emphasize efforts to reduce cross-border exposures through early repayment and proactive client management, while maintaining locally funded subsidiary operations. As UniCredit’s CEO stated, “Our Russia exposure has been reduced further at minimum cost. [...] Net cross-border exposures were reduced...mainly as a result of proactive discussions with clients producing early repayment at nominal value. The [Russian] subsidiary is robust and performing well.” Despite mounting regulatory and political pressure—including a 2024 ECB directive requiring banks to present plans for exiting or reducing their Russian operations—both UniCredit and RBI have continued maintaining their subsidiaries.17 These developments underscore that even under acute geopolitical stress, globalbankstendtopreserveaffiliate-basedlendingwhileretreatingfromcross-border exposures.18 This pattern—the retrenchment of cross-border lending alongside the persistence of local operations in response to geopolitical risk—also emerges in aggregate data. We track the evolution of cross-border and local claims on Russia following three major geopolitical events: the conflict with Georgia in 2008:Q3, the annexation of Crimea in 2013:Q4, and 16An exception is Societe Generale, which was the only major global bank to fully exit Russia soon after the invasion. Before the war, the bank derived approximately 3 percent of its net income from Russian operations. In April 2022, it sold its Russian subsidiary, Rosbank, to a business group linked to a Russian oligarch, incurring a $3.3 billion loss. By acting quickly, Societe Generale completed the sale before the oligarch in question was sanctioned by the European Union. 17Following the ECB directive, UniCredit took legal action while RBI halted brokerage account openings at its Russian subsidiary. One point to note is that while these global banks are reportedly still seeking opportunities to sell their Russian subsidiaries, any sale now requires approval from the Russian president and is likely to come at a hefty cost, further complicating their potential exit strategies. 18Citigroup has adopted a more phased strategy, allowing business to run off while selling individual portfolios. For more information on the post-invasion operations of global banks in Russia, see articles including “Why Are Raiffeisen and Unicredit still in Russia?,” Euromoney, October 4, 2022; “Western BanksStruggletoExitRussiaafterPutinIntervention,”FinancialTimes,January16,2023; and“Citigroup Expects $190 mln of Costs Tied to Russia Wind-down,” Reuters, February 27, 2023. For a summary article on global banks’ operations in Russia since the outbreak of the Russia–Ukraine War, see “European Banks Still in Russia: Should They Stay or Should They Go?” The Banker, March 17, 2023. Related information can also be found in the JPMorgan report titled “Global Banks: Russian Risk Assessment” from January 22, 2022, and in banks’ quarterly earnings presentations and annual filings (see, e.g., Citigroup’s 2022 10-K filing with the U.S. Securities and Exchange Commission.) 25

the invasion of Ukraine in 2022:Q1. Panel (a) of Appendix Figure B.3 presents the claims by the U.S. banking sector on Russia, and Panel (b) presents those for all BIS-reporting banking sectors. Notably, while both local and cross-border claims on Russia declined after these geopolitical shocks, local exposures fell significantly less, in percentage terms, than cross-border exposures. In sum, our regression results indicate that while banks primarily reduce cross-border exposures to countries facing heightened geopolitical risk, they largely maintain existing loans within their local operations despite the rising credit risk. This persistence aligns with anecdotal evidence on banks’ responses to Russias 2022 invasion of Ukraine and patterns observed in the raw data. 3.3 Geopolitical Risk and Other Economic Risks Do banks adjust their foreign operations similarly to other forms of country risk? Or is geopolitical risk distinct? We explore these questions by examining how banks adjust their cross-border and local exposures in response to other types of risks. We run Equation (5) using broad country-specific risk indices (instead of CGPR) as the main regressor, replacing CGPRwithHassanetal.(2023)’sCRI,Ahiretal.(2022)’sWUI,andsovereignCDSspreads. Table 3 reports the regression results. Columns (1)–(2), (3)–(4), and (5)–(6) correspond to specifications using CRI, WUI, and CDS spreads as the key regressors, respectively. Odd-numbered columns use log cross-border claims as the dependent variable, while evennumbered columns use log local claims. The results for CRI suggest a positive relationship with cross-border and local claims, though the effect of country risk on cross-border claims is not statistically significant. For WUI, the coefficients on both cross-border and local claims are small and statistically insignificant, suggesting that foreign exposures exhibit little sensitivity to broad country-level uncertainty. The results for CDS spreads show a negative relationship with cross-border and local claims, though only the effect on local claims is marginally significant, while the effect on cross-border claims remains insignificant.19 Overall, the results suggest that country risk, uncertainty, and sovereign credit risk do 19Results remain consistent when alternative risk variables are included in log form. 26

not have strong or consistent effects on banks’ cross-border and local exposures, in contrast to the clear and asymmetric response observed with geopolitical risk. While banks reduce cross-border exposures but maintain local exposures in response to geopolitical risk, their adjustments to other types of risk do not follow this pattern. Instead, the effects of country risk,uncertainty,andsovereigncreditriskoncross-borderandlocalclaimsappearweakerand less systematic, with no clear distinction in how banks adjust these two types of exposures. Discussion. Our findings indicate that banks respond differently to geopolitical risk than to other financial and economic risks, reinforcing the idea that geopolitical risk is a distinct category of risk. One possible reason for this distinction is that geopolitical risk often entails expropriation risk. Throughout history, geopolitical conflicts have led to the seizure of foreign bank assets, making expropriation a uniquely catastrophic feature of geopolitical risk. Notable examples include the 1917 Russian Bolshevik Revolution, during which the new government nationalized the financial system, expropriating all foreign-owned banks. During World War II, Germany expropriated foreign-owned banks, including Austria’s Kreditanstalt; and Japan took control of Allied-associated banks operating in occupied territories across Southeast Asia. In 1957, following the 1956 Suez Crisis, Egypt nationalized British and French banks in retaliation for military intervention. In 1960, after the Cuban Revolution, the government nationalized all U.S. banks, seizing the assets of Citibank, Chase Manhattan, and First National City Bank. From 2008 to 2010, Venezuela, under Hugo Cha´vez, nationalized Banco de Venezuela, previously owned by Spain’s Santander, as part of broader policies to expand state economic and industrial control. From 2023 to 2025, Russia seized assets from U.S. and European banks, including JPMorgan and Deutsche Bank, through a series of court rulings and state actions. While framed partly as legal responses to Western sanctions, these moves also reflected broader geopolitical retaliation, underscoring the rising legal and political risks foreign banks face in conflict-affected jurisdictions. Beyond outright seizure, geopolitical risk may also involve softer forms of state intervention that nonetheless pose major challenges for foreign banks. These include capital controls, profit repatriation limits, sudden regulatory shifts, asset freezes, and windfall taxes targeting 27

foreign institutions. Though less extreme than nationalization, such actions can trap capital, reduce operational flexibility, and generate legal and compliance uncertainty—risks that are difficult to anticipate or hedge. Because all forms of expropriation risk are typically accompanied by erosion of the rule of law and erratic policy shifts, geopolitical risk becomes far harder for banks to manage using standard risk-assessment tools. This sets it apart from conventional financial or economic risks, which are generally more predictable and manageable within established institutional frameworks. While expropriation risk may be a distinctively catastrophic feature of geopolitical risk, it remains an open question whether it helps explain the asymmetric response of banks’ cross-border and local exposures. In the next section, we formalize expropriation risk within a global banking framework and show that it can account for the divergence in how banks adjust their foreign exposures under geopolitical risk. 4 A Model of Global Banking under Geopolitical Risk In this section, we present a stylized model to rationalize the empirical facts established in the previous section and generate testable qualitative predictions on the transmission of geopolitical risk to domestic credit through global banks for the subsequent analysis. The model examines banks’ choices to operate abroad via cross-border lending or local affiliates, as well as their domestic operations, and analyzes how credit allocation across these channels responds to heightened foreign geopolitical risk. We focus on expropriation risk as a distinctive feature of geopolitical risk: As detailed in Section 3.3, both historical and recent geopolitical conflicts are often accompanied by the seizure of foreign bank assets or by other forms of state intervention, making expropriation risk a salient element in shaping global banks’ responses to geopolitical shocks. 4.1 Setup The framework consists of three periods and a global bank that makes investment decisions. At t = 0, the bank decides how much to invest abroad and at home. It can invest a fixed 28

Table 3: Other Country Risks and Banks’ Foreign Operations (1) (2) (3) (4) (5) (6) ln(exp ) Cross-border Local Cross-border Local Cross-border Local bct CRI -0.004 0.021 ct (0.017) (0.017) CRI 0.008 0.036∗∗ ct−1 (0.016) (0.018) WUI 0.004 0.003 ct (0.005) (0.007) WUI -0.007 0.004 ct−1 (0.005) (0.007) CDS -0.013 -0.028∗ ct (0.009) (0.016) CDS -0.004 -0.022 ct−1 (0.012) (0.014) Bank-country FE Yes Yes Yes Yes Yes Yes Bank-time FE Yes Yes Yes Yes Yes Yes Observations 53655 18940 127821 33810 60464 19961 R2 0.917 0.904 0.876 0.877 0.914 0.902 Note: Thistablereportsresultsfromregressionsatthebank-country-timelevelbasedonEquation(5)withalternativecountryspecific risk indices as the main regressor (instead of CGPR). The alternative indices include CRI by Hassan et al. (2023) (Columns(1)and(2)), WUIbyAhiretal.(2022)(Columns(3)and(4)), andsovereignCDSspreads(Columns(5)and(6)). The dependent variable is the log cross-border claims in Columns (1), (3), and (5), and log local claims in Columns (2), (4), and (6). All regressions include bank-country and country-time fixed effects. All the risk indices are standardized by their respectivestandarddeviationswithinthesample. Standarderrors,showninparentheses,areclusteredatthecountryandtime level. *p<.1;**p<.05;***p<.01. 29

amount L∗ abroad for two periods and a variable amount L domestically for one period, with the option to reinvest in domestic assets at t = 1. The return on the foreign two-period investment is uncertain. At t = 0, the probability of success is high (pG) with probability (1−φ) and low (pB) with probability φ. These good (G) and bad (B) states correspond to states of low and high geopolitical risk, respectively. At t = 1, the bank learns whether geopolitical risk is high or low, which determines the probability of success of its foreign investment: If geopolitical risk is high, the probability of success is low and p = pB; if geopolitical risk is low, the probability of success is high and p = pG. At t = 2, geopolitical risk either materializes or does not. If geopolitical risk does not materialize, the foreign investment succeeds and pays R∗. If geopolitical risk materializes, it leads to expropriation by the foreign government: The government seizes the investment, resulting in a zero payoff. For simplicity, we do not model domestic geopolitical risk. Domestic investment is assumed to be risk-free, yielding a guaranteed return of R at both t = 1 and t = 2. The bank has an initial equity endowment E at t = 0 and is subject to a leverage 1 constraint that closely follows the formulation of minimum regulatory capital ratios under Basel III. Specifically, the bank’s equity-to-risk-weighted assets ratio must remain above a constant threshold µ: E 1 ≥ µ, (6) L +L∗α(φ,pG,pB) 1 where α(φ,pG,pB) > 1 is the risk weight on the foreign investment L∗, which decreases with φ, pG, and pB. The effect of heightened geopolitical risk abroad on capital constraints in the model maps actualregulatorypractice. AsshowninSection3.1, geopoliticalriskincreasestheprobability of default on loans extended to borrowers in affected countries. Since default probability directly influences the risk weight assigned to loans, rising geopolitical risk results in higher capital requirements for foreign exposures.20 By contrast, the risk weight on the domestic, 20Note that this increase in risk-weighted assets applies to both modes of foreign exposure—cross-border lending and exposures held through foreign affiliates—because material foreign branches and subsidiaries are consolidated with the parent bank’s balance sheet for capital regulation purposes. For U.S. banks, an increaseinrisk-weightedassetsmayalsoresultinhigherprojectedlossesunderregulatorystresstests,further increasing the parent bank’s capital requirements. 30

risk-free investment is set to 1. We assume L∗ < E1 ensuring that the fixed foreign investment L∗ does not exceed the αµ bank’s total lending capacity given the risk weights on foreign assets and allowing room for domestic investment. Additionally, we assume that foreign investment is preferable to investing solely in the domestic asset, which holds if (1−φ)pGR∗ is sufficiently high. Because L∗ and E are fixed, the equity constraint pins down L : 1 1 E −µL∗α(φ,pG,pB) 1 L = . 1 µ To finance its investments, the bank borrows D = L + L∗ − E from depositors at an 1 1 1 exogenous interest rate i < R.21 The funding is for one period but can be rolled over at t = 1 at the same rate. At t = 1, the bank learns the probability of success of its foreign investment and may choose to liquidate early, recovering δL∗, where δ < 1. This option allows the bank to withdraw from foreign operations in response to rising geopolitical risk, albeit at a cost. While early liquidation results in a direct loss, it eliminates risk exposure and reduces risk-weighted assets, thereby enhancing the bank’s lending capacity. Two modes of foreign operations. The bank can choose between two modes of foreign operation: cross-border investment (X), whereby it lends directly from its home country, or local investment (A), whereby it lends through a locally established affiliate in the foreign country. Note that establishing a local affiliate incurs a non-pecuniary fixed cost κ > 0.22 Whenconductingcross-borderoperations, thebankraisesfundingdomestically. Bycontrast, when operating from a local affiliate, it raises funding D∗ in the foreign market, where t D∗ < D , while borrowing the remainder D −D∗ at home.23 We assume that the foreign t t t t and domestic interest rates on deposits are the same. The key distinction between the two modes—aside from the fixed cost κ—is that foreign deposits, unlike domestic deposits, are not repaid if the geopolitical risk materializes at 21While we refer to the bank’s external liabilities as “deposits,” these can represent any form of debt, including wholesale funding. 22This fixed cost is consistent with the literature, such as Niepmann (2023), and helps explain why banks may prefer cross-border operations over establishing a foreign affiliate. 23D∗isassumedtobeexogenoustokeepthemodelsimple. Alternatively,foreignfundingcouldbemodeled t as proportionate to the amount of foreign lending. 31

t = 2. The rationale is that when geopolitical conflict leads to expropriation, the foreign government seizes the bank’s local affiliate, and the bank is no longer obligated to repay foreign depositors. As a result, the expected profits from operating a local affiliate at t = 1 exceed those from cross-border investment by (1−p)D∗i. 2 This assumption that local liabilities are repaid under expropriation is consistent with how market participants and financial institutions assess risk exposure in the context of geopolitical crises. Analysts and bank disclosures have highlighted the importance of funding segmentation between parent banks and their foreign affiliates. For example, at the onset of the Russia-Ukraine war, UniCredit was a net borrower from its Russian subsidiary, which limited the bank’s losses in a worst-case expropriation scenario because it would not be expected to repay the intragroup loan.24 Similarly, RBI’s 2022:Q1 financial report emphasizes the strategic importance of local self-funding in managing geopolitical risk: “Naturally, we did not foresee a military conflict such as the one we are currently witnessing. We have however...ensured that [RBI’s subsidiaries] are self-financing, allowing only a restricted amount of cross-border financing.” The assumption of full expropriation is intended to capture a uniquely salient feature of geopolitical risk and to illustrate its most stark implications; however, the core mechanism remainsvalidundersofterformsofexpropriationrisk. Forexample,insteadofassumingzerorecovery expropriation when geopolitical risk materializes at t = 2, it is sufficient to assume that repatriating profits becomes costly. In such cases, raising local funding still reduces the parent’s exposure to these frictions by allowing the affiliate to meet its obligations locally, preserving the incentive to operate through affiliates rather than via cross-border lending. Thus, the mechanism whereby local funding mitigates parent-level downside risk does not rely on the strict assumption of zero-recovery expropriation. Additional mechanisms may be relevant, particularly in settings where outright expropriation is viewed as unlikely. For example, one could assume that interest rates abroad are lower, incentivizing banks to establish affiliates to access cheaper local funding at a fixed cost, while liquidation costs are higher for investments made through affiliates than for cross-border lending. While many of the model’s predictions continue to hold under this 24See JPMorgan’s report “Global Banks: Russian risk assessment” from January 25, 2022. 32

alternative, some do not.25 Nonetheless, these additional features may help capture banks’ incentives in countries facing elevated geopolitical tensions but maintaining strong legal institutions. We focus on expropriation risk given its historical and empirical prominence in major geopolitical episodes, while acknowledging that other mechanisms may also influence banks’ decisions across a broader range of geopolitical environments. 4.2 Foreign Operations under Geopolitical Risk Having established the key differences between the two modes of foreign operation, we now solve the model to analyze how the bank adjusts its cross-border and affiliate investments in response to heightened geopolitical risk, explaining the empirical findings presented in the previous section. Under liquidation, profits realized at t = 2 are the same across both modes. This follows because δ is identical in both cases, and investments in the domestic asset at t = 0 and t = 1 are the same. Specifically, πX,L = πA,L = RLL −iDL, where LL denotes the investment in 2 2 2 2 2 the domestic asset at t = 1 under liquidation (L). Investment decisions remain unchanged because they are governed by the leverage constraint, which is independent of D (and D∗). t t Whentheforeigninvestmentcontinues(C),thebanksexpectedprofitsundercross-border investment are: πX,C = pR∗L∗ +LCR−DCi. (7) 2 2 2 The banks expected profits when it continues operating through a local affiliate are: πA,C = pR∗L∗ +LCR−DCi+(1−p)D∗i > πX,C. (8) 2 2 2 2 2 Note the superscript associated with p is suppressed because the formulas hold for both the good and bad states of the world. Equations (7) and (8) highlight a key implication of the model: Because local deposits 25In particular, Proposition 1(b) would no longer hold under the alternative model setup. When the advantage of affiliate lending is driven by interest rate differentials and liquidation costs rather than expropriation risk, the wedge between affiliate and cross-border responses no longer varies with p, making it harder to reconcile with the empirical observation that banks respond differently to geopolitical risk than to other forms of country risk. 33

raised by foreign affiliates do not have to be repaid if the foreign government expropriates the affiliate, the bank has a stronger incentive to liquidate cross-border investment than investment through a foreign affiliate amid heightened geopolitical risk. ˆ PROPOSITION 1. Let δ denote the threshold value of δ at which the bank is indifferent between liquidating or continuing its foreign investment at t = 1. (a) Since πA,C > πX,C and πX,L = πA,L, it follows that δ ˆA > δ ˆX. In other words, the 2 2 2 2 threshold δ required for liquidation is higher when the bank operates through a foreign affiliate than when it invests cross-border. (b) The difference in liquidation thresholds, ∆δ ˆ = δ ˆA−δ ˆX, increases as p decreases. That is, the lower the probability of success p, the larger the difference between the two liquidation thresholds. (c) The difference in liquidation thresholds, ∆δ ˆ = δ ˆA−δ ˆX, increases as D∗ increases. That 2 is, the more funding the bank raises in the foreign market, the larger the difference between the two liquidation thresholds. Proof. See Appendix C.. Proposition 1 shows that, for the same liquidation cost δ, banks are less likely to liquidate investments in a foreign affiliate than in cross-border operations. The model thus explains the empirical finding from Section 3.2 that banks reduce exposures primarily through crossborder lending, while maintaining affiliate-based lending when geopolitical risk rises.26 Furthermore, Proposition 1(b) stipulates that as geopolitical risk increases (reflected in a lower p), the divergence between liquidation decisions for cross-border and affiliate investments becomes more pronounced, which helps explain the empirical finding from Section 3.3 that geopolitical risk is distinct from other types of risk. Expropriation risk plays a uniquely catastrophic role in shaping how banks adjust their cross-border and local operations. When sovereign or economic risk rises, banks may incur losses, but operations typically continue and obligations to foreign creditors remain. As a result, banks’ responses to sovereign and economic risk tend to be more symmetric across cross-border and local operations, unlike the asymmetric adjustment observed under geopolitical risk. 26Asdiscussed,whileitisplausiblethatliquidatinglocalaffiliateoperationsismorecostlythanliquidating cross-borderactivities,themodelgeneratesahigherlikelihoodofcross-borderactivitiesbeingliquidatedeven without this assumption. 34

Empirical validation. We further validate the model by empirically testing Proposition 1(c), which predicts that the more funding a bank raises in the foreign market, the less it divests from local investments in that market in response to geopolitical risk. To test this, we gather data on local liabilities from FFIEC 009 and augment Equation (5) with interaction terms between CGPR and banks’ lagged local liability position, measured as four-quarter moving averages (in log). The coefficient on this interaction term estimates the extent to which a larger local funding position influences the sensitivity of foreign exposure to geopolitical risk. Panel (a) of Table 4 presents the results. Columns (1) and (2) present results from regressions with banks’ log total foreign exposures as the dependent variable, and Columns (3) and (4) and Columns (5) and (6) are based on log local and cross-border exposures as the dependent variables, respectively, with the even-numbered columns including macro control variables. With total foreign exposures as the dependent variable, our coefficients of interest on the interaction terms are positive and significant, indicating that banks with larger local funding positions are less likely to reduce their overall foreign exposures in response to heightened geopolitical risk. This effect is primarily driven by local exposures, as shown in Columns (3) and (4), where the coefficients on the interaction terms remain positive and statistically significant. By contrast, the coefficients in Columns (5) and (6), where cross-border exposures are the dependent variable, are not statistically different from zero, suggesting that the mitigating effect of local funding applies specifically to local investments rather than cross-border positions. All these findings support the model’s prediction. Panel (b) of Table 4 further examines whether local funding positions influence banks’ foreign lending responses to other types of risk, as measured by CRI, WUI, and sovereign CDS spreads. The results show that, unlike in the case of geopolitical risk, local funding positionsdonotsignificantlyaffecthowbanksadjusttheirforeignexposureswhenfacedwith these alternative risks. This finding reinforces the model’s prediction that geopolitical risk, particularly due to expropriation concerns, uniquely alters banks’ foreign lending behavior. It also highlights that the ability to default on foreign liabilities plays a central role in banks’ responses to geopolitical risk but is less relevant when responding to other macroeconomic 35

or financial risks.27 4.3 Spillovers of Geopolitical Risk into Domestic Operations Next, we use the model to analyze the implications of rising geopolitical risk abroad for domestic lending. The bank’s equity position and the riskiness of its investments determine its domestic lending at t = 1. When the bank liquidates its foreign investment, its equity is given by EL = δL∗ + R L − D i, where R L − D i captures earnings from domestic 2 1 1 1 1 1 1 investmentatt = 1. Ifthebankdoesnotliquidate, itsequityisEC = L∗+R L −D i, which 2 1 1 1 satisfies EC > EL, indicating that liquidation results in a lower equity position. Although 2 2 liquidation reduces the bank’s equity, it also frees up leverage capacity, as the risk weight on domestic investment is 1, whereas the risk weight on the riskier foreign investment is higher. As a result, domestic lending following liquidation is given by: δL∗ +R L −D i LL = 1 1 1 . 2 µ If geopolitical risk turns out to be high at t = 1 and the bank does not liquidate, the bank’s borrowing capacity shrinks relative to the good state of the world due to an increase in foreign risk-weighted assets L∗α(p): L∗ +R L −D i−µL∗α(p) LC = 1 1 1 . 2 µ The effects of geopolitical risk on domestic lending are summarized in the following proposition: PROPOSITION 2. (a) LG,C > LB,C. Domestic lending under continuation is higher in 2 2 the good state of the world with low geopolitical risk than in the bad state with high geopolitical risk. (b) LL > LB,C if δ > 1 − α(p)µ. Domestic lending is higher when the bank liquidates 2 2 its foreign investment at t = 1 than when it continues its foreign operation, provided that the reduction in borrowing capacity from higher foreign risk-weighted assets due 27We also test the robustness of the results in Table 4 using an alternative measure of local liabilities: for each bank, we calculate its local liabilities from each foreign country as a share of its total lending to that country. Appendix Tables B.3 and B.4 present the results for geopolitical risk and other risks, respectively, which are qualitatively similar to those in Table 4. 36

Table 4: Banks’ Foreign Response to Risk by Ex Ante Local Liabilities (a) Geopolitical Risk TotalExp. Local Cross-border ln(exp ) (1) (2) (3) (4) (5) (6) bct CGPRN -0.049∗∗∗ -0.050∗∗∗ -0.067∗∗∗ -0.066∗∗∗ -0.074∗∗∗ -0.071∗∗∗ ct (0.019) (0.017) (0.022) (0.022) (0.015) (0.013) CGPRN ×ln(LL) 0.004∗∗ 0.004∗∗ 0.008∗∗ 0.008∗∗ 0.002 0.002 ct bct−1 (0.002) (0.002) (0.004) (0.004) (0.002) (0.002) CGPRN -0.018 -0.019 -0.034 -0.034 -0.027∗ -0.023 ct−1 (0.016) (0.015) (0.026) (0.026) (0.015) (0.015) CGPRN ×ln(LL) 0.002 0.002 0.005 0.005 -0.001 -0.001 ct−1 bct−2 (0.002) (0.002) (0.005) (0.005) (0.002) (0.002) MacroControls No Yes No Yes No Yes Bank-countryFE Yes Yes Yes Yes Yes Yes Bank-timeFE Yes Yes Yes Yes Yes Yes Observations 16829 16107 15870 15208 16040 15374 R2 0.956 0.958 0.919 0.922 0.938 0.938 (b) Other Risks CRI WUI CDS (1) (2) (3) (4) (5) (6) ln(exp ) Local Cross-border Local Cross-border Local Cross-border bct CRIt -0.025 -0.019 (0.033) (0.035) CRIt ×ln(LL) bct−1 0.002 -0.003 (0.004) (0.004) CRIt−1 -0.010 -0.059∗ (0.032) (0.033) CRIt−1 ×ln(LL) bct−2 0.004 0.005 (0.004) (0.004) WUIt -0.004 0.030∗∗ (0.015) (0.012) WUIt ×ln(LL) bct−1 -0.000 -0.006∗∗∗ (0.002) (0.002) WUIt−1 0.021 0.002 (0.015) (0.013) WUIt−1 ×ln(LL) bct−2 -0.002 -0.003 (0.002) (0.002) ln(CDS)t 0.004 -0.067 (0.086) (0.096) ln(CDS)t ×ln(LL) bct−1 -0.004 0.007 (0.012) (0.007) ln(CDS)t−1 -0.167∗ 0.083 (0.087) (0.086) ln(CDS)t−1 ×ln(LL) bct−2 0.008 0.008 (0.012) (0.007) MacroControls Yes Yes Yes Yes Yes Yes Bank-countryFE Yes Yes Yes Yes Yes Yes Bank-timeFE Yes Yes Yes Yes Yes Yes Observations 12631 12521 14490 14347 13982 13803 R2 0.943 0.922 0.940 0.922 0.941 0.922 Note: This table reports results from regressions at the bank-country-time level based on an augmented version of Equation (5),usingtheFFEIC009dataforthesampleperiod1986:Q1through2022:Q4. Panel(a)usesCGPRN,the(recent)countryspecific geopolitical risk index from Caldara and Iacoviello (2022), along with ln(LL) , the log of local liabilities received bct−1 by bank b from country c, calculated as a four-quarter moving average from t−4 to t−1, and their interactions as the main regressors. ThedependentvariableisthelogtotalforeignclaimsinColumns(1)and(2),loglocalclaimsinColumns(3)and (4),andlogcross-borderclaimsinColumns(5)and(6). Columns(1),(3),and(5)showthebaselineresultsforeachdependent variable. Columns(2),(4),and(6)addcountry-levelmacrocontrols,includingacountry’slogexchangeratevis-a`-vistheU.S. dollar,logdomesticstockpriceindex,andanindicatorvariableequalto1ifthecountryfacesanysanctionsfromtheUnited States. Panel (b) replaces CGPRN with alternative country-specific risk indices, ln(LL) , and their interactions as the bct−1 main regressors. The alternative indices include Hassan et al. (2023)’s CRI (Columns (1) and (2)), Ahir et al. (2022)’s WUI (Columns (3) and (4)), and log sovereign CDS spreads (Columns (5) and (6)). The dependent variable is log local claims in Columns(1),(3),and(5),andlogcross-borderclaimsinColumns(2),(4),and(6). Allregressionsincludebank-countryand country-timefixedeffects. Allriskindicesarestandardizedbytheirrespectivestandarddeviationwithinthesample. Standard errors,showninparentheses,areclusteredatthecountryand3t7imelevel. *p<.1;**p<.05;***p<.01.

to geopolitical risk exceeds the combined effect of the equity loss and the decrease in risk-weighted assets under liquidation. (c) L > LB,C if (R1−1)L1−(i−1)D1 < (α(pB)−α(φ,pB,pG))L∗. LG,C > L always holds. In 1 2 µ 2 1 other words, domestic lending contracts at t = 1 in the bad state of the world relative to t = 0 if the positive effect of increased equity from domestic investment realized in t = 1 on leverage is sufficiently small relative to the increase in foreign risk-weighted assets. Domestic lending always expands in the good state of the world. Proof. See Appendix C.. Proposition2highlightsthatheightenedgeopoliticalriskabroadreducesdomesticlending whenbanksdonotdivest,creatingspillovereffectsfromforeigngeopoliticalriskintodomestic credit supply. Whether domestic lending is higher under liquidation or continuation of the foreign investment depends on the cost of liquidation. When banks liquidate foreign investment, they free up lending capacity due to lower risk weights and, at the same time, reallocate lending capacity from the foreign to the home country. As long as the liquidation cost is relatively low and banks can recover sufficient capital, the negative spillover effects on domestic credit supply will be limited. As a result, these spillover effects tend to be smaller under liquidation than under continuation. Since banks with foreign affiliates are less likely to liquidate, spillover effects tend to be stronger for banks operating through affiliates than through cross-border lending. Furthermore, whengeopoliticalriskincreases, domesticlendingwilldeclinerelativetothe previous period—unless banks generate sufficient domestic profits to counteract the negative spillover effects. Lower capital requirements can also help mitigate these spillovers. Banks typically hold capital buffers above the regulatory minimum, providing some flexibility to absorb shocks without immediately constraining lending. Instead of depending on regulatory interventiontoeasecapitalrequirements,banksmaychoosetodrawdowntheirexcessbuffers to sustain domestic lending in the face of heightened geopolitical risk. Model predictions. From our theoretical framework, we derive the following testable hypotheses on the spillover of geopolitical risk into domestic lending through global banks: 1. Banks exposed to heightened geopolitical risk in their foreign operations reduce domestic lending more significantly. 38

2. The reduction in domestic lending is more pronounced when geopolitical risk rises in markets where banks operate through affiliates. 3. Spillover effects are larger for banks with lower capital ratios and profitability. 5 Transmission of Geopolitical Risk to Domestic Credit Guided by the model predictions from the previous section, we test the spillover effects of geopolitical risk on domestic lending through global banks. Our main part of this analysis examines how U.S. banks’ exposure to foreign geopolitical risk, as measured by the BGPR indices, affects their loan origination to U.S. firms, using FR Y-14 data. 5.1 Geopolitical Risk and Domestic Loan Origination Loan-level analysis. To test the prediction that banks exposed to heightened geopolitical risk in their foreign operations reduce domestic lending more (Prediction 1 from Section 4), we first estimate the following specification at the loan level using the FR Y-14 data for the period 2013:Q1 through 2022:Q4: ln(orig ) = βBGPR +δZ +δX +γ +α +(cid:15) , (9) bit bt bt bit it b bit where orig denotes the amount of loan origination by bank b to domestic firm i at time t, bit BGPR denotes BGPRN or BGPRT, Z denotes bank-level controls including liquid-asset bt bt bt bt ratio and Tier 1 capital ratio, X denotes loan-level controls including maturity and interest bit rate, γ denotes firm-time fixed effects, and α denotes bank fixed effects.28 The regression it b sample is restricted to loans by U.S.-headquartered banks to U.S. firms. Our coefficient of interest, β, measures the extent to which banks that experienced a greater increase in geopolitical risk through their foreign exposures, as captured by the BGPRindices,adjustedtheirloanoriginationtodomesticfirms,conditioningonthespecified 28We include only contemporaneous BGPR in loan-level regressions, as identification comes from crossbank variation in current risk exposure for the same firm at the same time. Lagged BGPR does not add meaningful additional identifying variation and may confound interpretation. By contrast, the subsequent bank-levelregressionscaptureaggregatelendingdynamics,inwhichresponsestogeopoliticalriskmayunfold over time. We therefore include both contemporaneous and lagged GPR to allow for delayed adjustments. 39

controls and fixed effects. As described in Section 2, the BGPR indices contain considerable variation, both across banks and over time, due to differences in the geographical origin and magnitude of their exposures, both of which fluctuate over time. Our estimation relies exclusively on cross-bank within-firm variation for identification, given the inclusion of firmtime fixed effects. This alleviates concerns about confounding factors from the demand side, such as changes in credit demand by firms in response to geopolitical risk. Panel(a)ofTable5reportstheresults. Columns(1)through(4)presentsestimatesusing BGPRN as the main regressor, while Columns (5) through (8) use BGPRT. Columns (1) and (5) include bank and firm-time fixed effects and incorporate both bank- and loan-level controls. The remaining columns further include alternative risk controls including bankspecific risk indices based on CRI (Columns (2) and (6)), WUI (Columns (3) and (7)), and sovereign CDS spread (Columns (4) and (8)), which are constructed following Equation (1). TheresultsshowthatU.S.bankssignificantlyreduceloanoriginationtodomesticfirmsin response to an increase in BGPR, whether measured by BGPRN or BGPRT. The inclusion of firm-time fixed effects indicates that changes in credit demand are not a significant confounding factor. The coefficients remain stable when alternative risk controls are included, indicating that the effect of geopolitical risk on loan origination is not confounded by broader measures of financial and economic risk. This finding is consistent with our illustrations and results from Sections 2 and 3.3, which highlight that geopolitical risk is distinct from other types of risk. The consistency of these estimates across the two measures and various model specifications further reinforces the robustness of the results, confirming that the impact of geopolitical risk on lending is not driven by firm-level credit demand shocks but rather by banks’ adjustments in credit supply. Based on the estimates in Columns (1) and (5), a one-standard-deviation increase in BGPR reduces U.S. banks’ loan origination to U.S. firms by 8 to 9 percent. Bank-level analysis. In addition to the loan-level analysis, which allows us to control for potential demand-side responses by firms and isolate the supply effect, we conduct a bank-level analysis to assess whether this effect is substantial enough to be observed at the 40

Table 5: Geopolitical Risk and U.S. Domestic Loan Origination (a) Loan Level BGPRN BGPRT ln(orig ) (1) (2) (3) (4) (5) (6) (7) (8) bit BGPRN -0.087∗∗∗ -0.061∗∗ -0.089∗∗∗ -0.087∗∗∗ bt (0.027) (0.029) (0.027) (0.027) BGPRT -0.081∗∗∗ -0.061∗∗∗ -0.083∗∗∗ -0.081∗∗∗ bt (0.020) (0.022) (0.020) (0.020) BCRI 0.072∗∗ 0.069∗∗ bt (0.032) (0.032) BWUI -0.044 -0.047 bt (0.030) (0.030) BCDS 0.001 0.005 bt (0.024) (0.024) Bank Controls Yes Yes Yes Yes Yes Yes Yes Yes Loan Controls No Yes Yes No Yes Yes Yes Yes Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Firm-time FE Yes Yes Yes Yes Yes Yes Yes Yes N 175943 175943 175943 175943 175943 175943 175943 175943 R2 0.617 0.617 0.617 0.617 0.617 0.617 0.617 0.617 (b) Bank Level and BGPRN BGPRT ln(orig ) (1) (2) (3) (4) (5) (6) (7) (8) bt BGPRN -0.073 -0.095 -0.072 -0.078 bt (0.062) (0.071) (0.062) (0.063) BGPRN -0.177∗∗ -0.185∗∗ -0.160∗∗ -0.185∗∗ bt−1 (0.074) (0.073) (0.066) (0.072) BGPRT -0.045 -0.066 -0.042 -0.053 bt (0.069) (0.073) (0.068) (0.070) BGPRT -0.175∗∗ -0.172∗∗ -0.163∗∗ -0.163∗∗ bt−1 (0.070) (0.068) (0.068) (0.073) Bank Controls Yes Yes Yes Yes Yes Yes Yes Yes AltRisk Controls No CRI WUI CDS No CRI WUI CDS Bank FE Yes Yes Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes N 475 475 475 475 475 475 475 475 R2 0.955 0.955 0.956 0.955 0.956 0.957 0.957 0.956 Note: This table reports results with log loan origination amount (orig) as the dependent variable, using FR Y-14 data from 2013:Q1through2022:Q4. Panel(a)reportsresultsfromloan-levelregressionsbasedonEquation(9). Panel(b)reportsresults from bank-level regressions based on Equation (10). BGPRN denotes the bank-specific geopolitical risk index, constructed fromCGPRN orthe(recent)country-specificgeopoliticalriskindexfromCaldaraandIacoviello(2022)accordingtoEquation (1). BGPRT denotesthebank-specificgeopoliticalriskindexderivedfromCGPRT,whichisbasedonearningscalltranscripts processedthroughtheNLAnalyticsplatform,capturinggeopoliticalriskperceptionbyfirmsworldwide. Bankcontrolsinclude Tier 1 capital ratio, liquid-asset ratio as well as their lagged versions in bank-level regressions. Loan controls include interest rate and maturity. Alternative risk controls include bank-specific risk indices based on the country risk index (BCRI) by Hassan et al. (2023) and the World Uncertainty Index (BWUI) by Ahir et al. (2022), and sovereign CDS spread (BCDS), as well as their lagged versions in bank-level regressions. All4th1e geopolitical risk indices are standardized by their respective standard deviations within the sample. Standard errors, shown in parentheses, are clustered at the bank and time level for loan-levelregressionsandatthebanklevelforbank-levelregressions. *p<.1;**p<.05;***p<.01.

aggregate level. We apply the following specification: ln(orig ) = β BGPR +β BGPR +δZ +γ +α +(cid:15) , (10) bt 1 bt 2 bt−1 bt−1 t b bit where orig denotes the total amount of loan origination by bank b at time t, BGPR bt bt denotes BGPRN or BGPRT, and the lagged BGPR indices are included to capture any bt bt persistent effects. Z denotes bank-level controls including contemporaneous and lagged bt liquid-asset ratio and Tier 1 capital ratio, γ denotes time fixed effects, and α denotes bank t b fixed effects. The coefficients of interest, β and β , capture the total spillover effects of 1 2 foreign geopolitical risk on U.S. banks’ domestic loan origination on average. Panel (b) of Table 5 reports the results. As in Panel (a), Columns (1) through (4) present estimates using BGPRN as the main regressor, while Columns (5) through (8) use BGPRT. Columns (1) and (4) include bank and time fixed effects as well as bank-level controls, while the remaining columns further add alternative risk controls, including bank-specific risk indices based on the CRI, WUI, and sovereign CDS spreads. The coefficients on both BGPRN and BGPRT are negative, significant, and of similar magnitude, indicating a strong relationship between foreign geopolitical risk and domestic credit supply at the bank-level. Based on the estimates in Columns (1) and (5), a onestandard-deviation increase in BGPR reduces U.S. banks’ loan origination to U.S. firms by 22 to 25 percent on average. This indicates that the spillover effects of foreign geopolitical risk on domestic credit markets through global banks are substantial enough to be observed at the aggregate level. Taken together, the loan- and bank-level results confirm Prediction 1 from the theoretical framework: Banks exposed to heightened geopolitical risk in their foreign operations reduce domestic lending more. This finding underscores the spillover effects of foreign geopolitical shocks, demonstrating that banks do not simply adjust their foreign operations in response to geopolitical risk but also contract their domestic credit supply. 42

5.2 Role of Local versus Cross-border Foreign Exposures Next, we test Prediction 2 from the model, which states that the reduction in domestic lending is more pronounced when geopolitical risk rises in markets where banks operate through affiliates. To analyze this, we estimate Equations (9) and (10) using BGPR indices decomposed into two separate components to distinguish between exposure from local claims and cross-border claims: (cid:88) BGPR (1(Cross-border)) = 1(Cross-border) ×ω CGPR , (11a) bt bct−1 bct−1 ct c (cid:88) BGPR (1(Local)) = 1(Local) ×ω CGPR , (11b) bt bct−1 bct−1 ct c where 1(Cross-border) denotes a dummy variable equal to 1 if bank b has no local claims bct on country c at time t and 0 otherwise, and 1(Local) is a dummy variable equal to 1 if bct bank b has non-zero local claims on country c at time t and 0 otherwise. All other variables are consistently defined with Equation (1). The theory presented in Section 4 suggests that, as long as the hit to equity from liquidation is limited, spillovers from geopolitical risk into domestic lending should be smaller under cross-border lending than under affiliate lending. This is because liquidating crossborder claims frees domestic lending capacity, allowing banks to reallocate credit back to the home country. By contrast, continued local lending raises risk-weighted assets as geopolitical risk increases, tightening capital constraints. Therefore, we expect the coefficients on BGPRN1(Local) to be negative and significant, whereas those on BGPRN1(Cross-border) bt bt should be smaller, if significant at all. Table 6 presents the results with BGPRN as the main regressor, with Panel (a) displaying the loan-level results and Panel (b) displaying the bank-level results. Columns (1) and (2) include BGPRN(1(Local)) as the regressor, without and with bank-level controls, bt respectively; Columns (3) and (4) include BGPRN(1(Cross-border)) as the regressor; and bt Columns (5) and (6) include both as regressors. As shown in the first two columns, the coefficients on BGPRN(1(Local)) are negative and significant, indicating that geopolitical risk, bt through banks’ local exposure, plays a significant role in reducing domestic loan origination 43

and driving the spillover effects. By contrast, the coefficients on BGPRN(1(Cross-border)) bt are not statistically significant, suggesting that geopolitical risk transmits to domestic credit supply primarily through local affiliate exposure rather than cross-border operations. When both indices are included in the regression, the coefficient on BGPRN(1(Local)) continues bt to be negative and significant, confirming the role of foreign exposure through local claims in driving the spillover effects. These results hold at both the loan and bank levels. Appendix Table B.5 presents the results with BGPRT as the main regressor, and all the results are quantitatively and qualitatively similar. Overall, these results provide strong evidence that global banks with local affiliate exposure react more significantly to geopolitical shocks abroad, leading to a greater contraction in domestic lending. This finding aligns with the model’s Prediction 2, confirming that spillover effects are stronger when geopolitical risk increases in markets where banks have local affiliates. By contrast, banks with predominantly cross-border operations adjust their foreign exposures more quickly and to a greater extent, allowing them to absorb geopolitical shocks with less impact on their domestic lending activity. The distinction between affiliatebased and cross-border exposure highlights the role of global banks’ corporate structures in shaping their responses to geopolitical risk and influencing its transmission to the domestic economy. 5.3 Additional Results In the following section, we conduct additional analyses to complement the main findings on thespillovereffectsofgeopoliticalriskondomesticlendingthroughglobalbanksandtoassess robustness. First, we test Prediction 3 from the model, which examines the role of capital constraints. Second, we investigate whether the threat or the realization of geopolitical risk is the primary driver of spillover effects. Third, we analyze how banks’ exposure to geopolitical risk influences their lending standards for domestic loans, leveraging SLOOS data, which covers a broader set of banks and extend to the 1990s. Role of capital constraints and bank profitability. Prediction 3 from the model stipulates that spillover effects are larger for banks with lower capital ratios and profits. To 44

Table 6: Geopolitical Risk Transmission: Cross-border vs. Local Exposure, BGPRN (a) Loan Level ln(orig ) (1) (2) (3) (4) (5) (6) bit BGPRN(1(Local)) -0.060∗∗ -0.062∗∗ -0.060∗∗ -0.060∗∗ bt (0.026) (0.026) (0.027) (0.027) BGPRN(1(Cross-border)) -0.021 -0.037 -0.010 -0.023 bt (0.044) (0.046) (0.045) (0.046) Bank Controls No Yes No Yes No Yes Bank FE Yes Yes Yes Yes Yes Yes Firm-time FE Yes Yes Yes Yes Yes Yes Observations 205642 199753 205642 199753 205642 199753 R2 0.594 0.592 0.594 0.592 0.594 0.592 (b) Bank Level ln(orig ) (1) (2) (3) (4) (5) (6) bt BGPRN(1(Local)) -0.061 -0.075 -0.069 -0.082 bt (0.061) (0.060) (0.061) (0.060) BGPRN (1(Local)) -0.168∗∗ -0.165∗∗ -0.169∗∗ -0.167∗∗ bt−1 (0.076) (0.075) (0.075) (0.074) BGPRN(1(Cross-border)) -0.175 -0.159 -0.179 -0.160 bt (0.229) (0.237) (0.234) (0.242) BGPRN (1(Cross-border)) -0.108 -0.148 -0.198 -0.238 bt−1 (0.265) (0.276) (0.288) (0.298) Bank Controls No Yes No Yes No Yes Bank FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Observations 475 461 475 461 475 461 R2 0.954 0.955 0.952 0.953 0.954 0.955 Note: Thistablereportsresultsfromregressionswithlogloanoriginationamount(orig)asthedependentvariableusingdata fromFRY-14forthesampleperiod2013:Q1through2022:Q4. Panel(a)reportsresultsfromregressionsattheloanlevelbased on Equation (9), using BGPRN(1(Local)) and BGPRN (1(Cross-border)), which are constructed based on Equation (11). bt bt−1 Panel(b)reportsresultsfromregressionsatthebanklevelbasedonEquation(10). Bank-levelcontrolsincludecontemporaneous Tier 1 capital ratio and liquid-asset ratio as well as their lagged versions in bank-level regressions. All the geopolitical risk indicesarestandardizedbytheirrespectivestandarddeviationswithinthesample. Standarderrors,showninparentheses,are clusteredatthebankandtimelevel. *p<.1;**p<.05;***p<.01. 45

test this, we estimate bank-level regressions with domestic loan origination as the dependent variable and the BGPR indices, along with the interaction of the BGPR indices with either a bank’s lagged Tier 1 capital ratio or a bank’s lagged return on average assets (ROAA), as the key regressors. If capital constraints or profitability influence the spillover effect of geopolitical risk on domestic loan origination, the coefficients on the interactions should be positive, indicating that banks with stronger capital or profitability positions reduce loan origination less in response to increasing geopolitical risk abroad. TheresultsarereportedinPanel(a)ofTable7withColumns(1)and(2)presentingthose for capital ratios using BGPRN and BGPRT as the regressor, respectively, and Columns (3) and (4) presenting the results for bank profitability. The coefficients on the interaction terms are positive, supporting the role of capital constraints and bank profitability in mitigating the spillover effects of geopolitical risk. Geopolitical risk: threat versus act. Next, we examine the different dimensions of geopolitical risk to assess whether spillover effects are driven more by the threat or the realization of geopolitical risk. This analysis differentiates the impact of anticipated versus actual geopolitical disruptions and evaluates the validity of the model setup in Section 4, in which geopolitical risk is primarily modeled as arising from the threat rather than its realization. As described in Section 2, BGPRT is designed to be flexible, enabling decomposition into different components. We construct five subindices of BGPRT. BGPRT(Threat) is constructed using the component of CGPR that captures firms’ perceptions of the threats of geopolitical risk, while BGPRT(Act) isolates their perceptions of geopolitical risk arising from realized events (e.g., attacks and wars). Additionally, BGPRTfin reflects perceptions of geopolitical risk specifically by financial firms, with BGPRTfin(Threat) and BGPRTfin(Act) representing the corresponding subcomponents for threats and acts, respectively. We estimate the impact of each subindex of geopolitical risk on U.S. banks’ loan origination to domestic firms using Equation (9) for loan-level regressions and Equation (10) for bank-level regressions. Panel (b) of Table 7 presents the results from the loan-level regressions. Columns(1)through(5)correspondtoregressionsusingBGPRT(Threat), BGPRT(Act), 46

Table 7: Role of Capital Constraints, Profits, and Type of Geopolitical Risk (a) Capital Constraints and Bank Profitability ln(orig ) (1) (2) (3) (4) bt BGPRN -0.824∗∗ -0.100 bt (0.342) (0.096) BGPRT -0.284 -0.274∗∗∗ bt (0.237) (0.079) BGPRN x Capital 0.050∗∗ bt bt−1 (0.021) BGPRT x Capital 0.011 bt bt−1 (0.015) BGPRN x ROAA 0.010 bt bt−1 (0.036) BGPRT x ROAA 0.155∗∗∗ bt bt−1 (0.040) Bank Control Yes Yes Yes Yes Bank FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes Observations 477 477 477 477 R2 0.952 0.952 0.952 0.953 (b) Geopolitical Risk Threat versus Act (Loan Level) ln(orig ) (1) (2) (3) (4) bit BGPRT(Threat) -0.075∗∗∗ bt (0.021) BGPRT(Act) -0.048∗ bt (0.025) BGPRTfin(Threat) -0.061∗∗∗ bt (0.021) BGPRTfin(Act) -0.026 bt (0.019) Bank Controls Yes Yes Yes Yes Loan Controls Yes Yes Yes Yes Bank FE Yes Yes Yes Yes Firm-time FE Yes Yes Yes Yes Observations 171380 171380 171380 171380 R2 0.615 0.615 0.615 0.615 Note: Thistablereportsregressionresultswithlogloanoriginationamount(orig)asthedependentvariableusingdatafromFR Y-14forthesampleperiod2013:Q1through2022:Q4. Columns(1)and(2)ofPanel(a)includeBGPRN orBGPRT,lagged Tier1capitalratio,andtheirrespectiveinteractionsaskeyregressorsinbank-levelregressions. Columns(3)and(4)includes BGPRN or BGPRT, lagged return on average assets (ROAA), and their respective interactions as key regressors. BGPRN denotes the bank-specific geopolitical risk index, constructed from CGPRN or the (recent) country-specific geopolitical risk indexfromCaldaraandIacoviello(2022)accordingtoEquation(1). BGPRT denotesthebank-specificgeopoliticalriskindex derived from CGPRT, which is based on earnings call transcripts processed through the NL Analytics platform, capturing geopoliticalriskperceptionbyfirmsworldwide. Bankcontrolincludeslaggedliquid-assetratioforColumns(1)and(2)and,in addition,laggedTier1capitalratioinColumns(3)and(4). Panel(b)reportsresultsfromloan-levelregressionswithsubindices ofBGPRT asthemainregressors. BGPRT(Threat) capturesfirms’perceptionsofgeopoliticalriskthreats,andBGPRT(Act) capturestheirperceptionsofgeopoliticalriskstemmingfromacts. BGPRTfin(Threat) andBGPRTfin(Act) capturefinancial firms’perceptionsofgeopoliticalriskstemmingfromthreatsandacts,respectively. BankcontrolsincludeTier1capitalratio 47 and liquid-asset ratio. Loan controls include interest rate and maturity. All the geopolitical risk indices are standardized by their respective standard deviations within the sample. Standard errors, shown in parentheses, are clustered at the bank and timelevel. *p<.1;**p<.05;***p<.01.

BGPRTfin, BGPRTfin(Threat), and BGPRTfin(Act) as the main regressors, respectively. The results indicate that the effect of BGPR on domestic loan origination is primarily driven by perceived threats of geopolitical risk (Columns (1) and (3)) rather than the realization of specific events (Columns (2) and (4)). This underscores the role of uncertainty in generating the spillover effects of geopolitical risk through banks. Appendix Table B.6 presents the results from the bank-level regressions, which closely mirror those from the loan-level analysis, further supporting these findings. Overall, the results show that the threat of geopolitical risk has a stronger influence on lending decisions than realized shocks. Banks preemptively adjust exposures to mitigate potential losses, validating the model framework outlined in Section 4. Domestic lending standards. To supplement our main analysis on loan origination, we examine the spillover effects of geopolitical risk on U.S. banks’ domestic lending standards, which have predictive power for loan origination (Niepmann and Schmidt-Eisenlohr, 2023).29 We use survey data from the SLOOS, which, compared with the FR Y-14 used in the loan originationanalysis, hastheadvantageofcoveringalargersetofbanksandextendingfurther back in time, starting in 1990.30 To measure lending standards, we analyze each bank’s response to the survey question on whether the bank tightened or loosened credit standards for C&I loans to large and medium-sized enterprises, where higher values indicate greater loosening. As is standard in the literature, we code responses as 1 for loosening, 0 for no change, and –1 for tightening. We regress this variable on the contemporaneous and lagged quarterly change in BGPR, controlling for bank fixed effects as well as macro and bank-level conditions. Following common practice in the literature (e.g., Bassett et al., 2014), we include the first lag of the dependent variable to account for the persistence in SLOOS responses. The baseline regression equation is specified as follows: ls = β ls +β ∆log(BGPR )+β ∆log(BGPR )+γ ∆X +γ ∆X (12) bt 0 bt−1 1 bt 2 bt−1 1 t 2 t−1 +δ Z +δ Z +α +(cid:15) , 1 bt 2 bt−1 b bt 29See, e.g., Table A.6 in Niepmann and Schmidt-Eisenlohr (2023). 30The Federal Reserve surveys as many as 80 domestic banks each quarter. 48

where ls represents bank bs response to the SLOOS survey question on lending standards in bt quarter t, and BGPR denotes the BGPR indices. The macroeconomic controls, X , include bt t thetwo-yearTreasuryyield, theslopeoftheyieldcurve(10y–2y), theCBOEVolatilityIndex (VIX), the S&P 500 index, and U.S. industrial production. The BGPR index, VIX, S&P 500 index, and industrial production enter as quarterly log changes, while other variables, except the lagged dependent variable, enter as simple changes. The regression also includes bank fixed effects (α ) and controls for changes in loan demand, based on banks’ response b to the SLOOS survey question on loan demand, as well as their lagged Tier 1 capital ratio and liquid-asset ratio (Z ).31 bt Panel(a)ofTable8presentsthebaselineresultsfortheperiod1990:Q2through2022:Q2.32 Columns (1) through (3) use BGPRN as the main regressor, while Columns (4) through (6) use BGPRT. Columns (1) and (4) include bank fixed effects. Columns (2) and (5) add macroeconomic controls, and Columns (3) and (6) further incorporate bank-level controls, including banks’ responses to changes in credit demand, as well as their Tier 1 capital and liquid-asset ratios. AcrossColumns(1)through(3), thecoefficientsonBGPRN arenegativeandstatistically significant, oftenatthe1percentlevel, indicatingthatincreasedexposuretogeopoliticalrisk, as measured by BGPRN, leads to a significant tightening of lending standards for domestic loans. Regarding magnitude, a one-standard-deviation increase in BGPR leads to 2 percent of banks shifting from maintaining unchanged lending standards to tightening them within the same quarter, with an additional 4 percent tightening in the following quarter (Column 3). The results for BGPRT in Columns (4) through (6) are consistent with these findings, reinforcing the conclusion that geopolitical risk affects banks’ lending standards. Overall, 31We do not include time fixed effects in this regression because their inclusion, along with bank fixed effects, would leave the regressions reliant solely on cross-sectional variation to identify the effects of BGPR on credit supply. However, the SLOOS outcome variable is inherently limited to three discrete values— tightening, loosening, or no change in credit standards. This constraint means that when two banks experience different levels of increasing exposure to GPR but both tighten credit standards to some extent, the outcome variable still takes the same value (–1) for both. In other words, the coarseness of the outcome variablemakesitdifficulttopreciselycapturevariationinbankbehaviorusingapurelycross-sectionalidentification strategy. Unsurprisingly, when time fixed effects are included in the regression, the coefficients associated with BGPR are insignificant. 32The sample period varies slightly across specifications depending on data availability when control variables are included. 49

these results align with the loan origination results in Section 5.1, providing further support for Prediction 1 from the model. Parallel to the analysis in Section 5.2, which tests Prediction 2 from the model, we investigate whether the effect of BGPR on bank lending standards is driven by exposure through local claims versus cross-border claims. Panel (b) of Table 8 presents the results, confirming that the tightening effect of BGPR on domestic lending standards is primarily driven by banks’ foreign local exposures. This finding aligns with our proposed mechanism, confirms Prediction 2 from the model, and mirrors the corresponding results on loan origination. Following the earlier analysis, we investigate how different dimensions of geopolitical risk influence banks’ domestic lending conditions. Appendix Table B.7 reports results using BGPRT(Threat), BGPRT(Act), BGPRTfin(Threat), and BGPRTfin(Act) to capture banks’ exposure to geopolitical risk. These findings are consistent with our earlier results based on the FR Y-14 data, further confirming that banks respond more strongly to geopolitical risk stemming from perceived threats rather than realized acts. While we focus primarily on C&I loans, Appendix Table B.8 shows that banks also tighten lending standards on commercial real estate loans in response to geopolitical risk. This finding provides additional evidence that banks contract their domestic credit supply when foreign geopolitical risk increases. Notably, the U.S. commercial real estate sector is less directly affected by geopolitical risk compared with industries such as trade and manufacturing, which are more exposed to risks abroad. Therefore, our findings on spillover effects are unlikely to be driven by credit demand responses. 6 Conclusion This paper studies the impact of geopolitical risk on banks’ foreign operations and the resulting spillover effects on domestic credit supply. Using a combination of established and newly constructed geopolitical risk indices and multiple supervisory data covering U.S. bank lending activities spanning nearly four decades, we find that geopolitical risk significantly increases these banks’ credit risk. Despite this heightened risk, banks continue lending through their foreign branches and subsidiaries while scaling back cross-border lending. This 50

Table 8: Geopolitical Risk and Domestic Lending Standards (a) Baseline ls (1) (2) (3) (4) (5) (6) bt ∆log(BGPRN) -0.023∗∗∗ -0.015∗∗ -0.023∗∗ bt (0.008) (0.007) (0.011) ∆log(BGPRN ) -0.019∗∗ -0.014∗ -0.037∗∗∗ bt−1 (0.008) (0.008) (0.012) ∆log(BGPRT) -0.008 -0.032∗∗∗ -0.034∗∗∗ bt (0.011) (0.011) (0.012) ∆log(BGPRT ) -0.005 -0.014 -0.011 bt−1 (0.010) (0.010) (0.010) Macro Controls No Yes Yes No Yes Yes Bank Controls No No Yes No No Yes Bank FE Yes Yes Yes Yes Yes Yes Observations 3099 3050 2095 1486 1486 1476 R2 0.235 0.294 0.331 0.258 0.339 0.352 (b) Role of Local versus Cross-border Foreign Exposures ls (1) (2) (3) (4) (5) (6) bt ∆log(BGPRN (1(Local))) -0.027∗∗ -0.021∗ bt (0.011) (0.011) ∆log(BGPRN (1(Local))) -0.031∗∗∗ -0.025∗∗ bt−1 (0.012) (0.012) ∆log(BGPRN (1(Cross-border))) -0.020∗∗ -0.011 bt (0.008) (0.009) ∆log(BGPRN (1(Cross-border))) -0.025∗∗ -0.013 bt−1 (0.010) (0.011) ∆log(BGPRT (1(Local))) -0.038∗∗∗ -0.039∗∗∗ bt (0.013) (0.015) ∆log(BGPRT (1(Local))) -0.010 -0.010 bt−1 (0.013) (0.015) ∆log(BGPRT (1(Cross-border))) -0.004 0.011 bt (0.011) (0.013) ∆log(BGPRT (1(Cross-border))) -0.017∗ -0.014 bt−1 (0.010) (0.012) Macro Controls Yes Yes Yes Yes Yes Yes Bank Controls Yes Yes Yes Yes Yes Yes Bank FE Yes Yes Yes Yes Yes Yes Observations 1303 2067 1275 1019 1264 808 R2 0.340 0.330 0.339 0.341 0.338 0.323 Note: Thistablereportsbank-levelregressionresults,wherethedependentvariableisbanks’responsetotheSLOOSsurveyquestionontightening, maintaining,orlooseningcreditstandardsforC&Iloanstolargeandmedium-sizedfirms,usingasamplespanningfrom1990:Q2through2022:Q2. Panel(a)reportsresultsbasedonEquation(12),whereBGPRN (Columns(1)through(3))isthebank-specificgeopoliticalriskindexconstructed fromCGPRN,thecountry-specificgeopoliticalriskindexfromCaldaraandIacoviello(2022),usingEquation(1). BGPRT (Columns(4)through (6))isthebank-specificgeopoliticalriskindexderivedfromCGPRT, whichcapturesfirmsgeopoliticalriskperceptionsbasedonearnings-call transcriptsprocessedthroughtheNLAnalyticsplatform. Columns(2)and(5)addmacroeconomiccontrols,including(log)changesinthetwo-year Treasuryyield,theyieldcurveslope(10y–2y),theCBOEVolatilityIndex(VIX),theS&P500index,andU.S.industrialproduction. Columns (3) and (6) further control for loan demand, as well as banks’ liquid-asset and Tier 1 capital ratios. In Panel (b), BGPRb N t (1(Local)) and BGPRb N t (1(Cross-border))areconstructedfollowingEquation(11). Allspecificationsincludebankfixedeffects,macroeconomiccontrols,banklevelcontrols,andthelaggeddependentvariableasaregressor. Forbothpanels,thegeopoliticalriskindicesarestandardizedbytheirrespective standard deviations within the sample. Standard errors, shown in parentheses, are clustered at the bank and time level. *p < .1; **p < .05; ***p<.01. 51

asymmetric response is unique to geopolitical risk, as banks do not adjust their foreign operations in the same way in response to other types of risk. We develop a stylized model to explain these findings, emphasizing how banks’ funding structures and expropriation risk drive their responses to geopolitical risk. The model highlights that foreign affiliates rely on local funding, which helps contain losses in the event of expropriation under geopolitical shocks. By contrast, cross-border lending is funded domestically and remains more directly exposed to geopolitical risk. This distinction in net exposure explains why banks reduce cross-border exposure while maintaining affiliate-based lending. These forces generate significant spillover effects on domestic credit supply. We show that U.S. banks facing geopolitical risk abroad reduce lending to domestic firms, with the effect strongest when the risk originates in countries where banks operate through local affiliates. This underscores the importance of banks’ operational structures in shaping how geopolitical shocks are transmitted to the domestic economy. Our findings reveal the potential real and distributional consequences of geopolitical risk transmitted through global banks. Constrained firms may respond to reduced credit supply by cutting investment and employment (see, e.g., Chodorow-Reich, 2014, Alfaro et al., 2021). At the same time, credit reallocation can generate amplification effects: Firms with better credit access may shift to smaller domestic lenders, crowding out more marginal borrowers such as small- and medium-sized enterprises. In this way, geopolitical shocks may propagate through the domestic credit system not only via direct exposure but also through general equilibrium effects in financial intermediation—an important area for future research. 52

References Ahir, Hites, Nicholas Bloom, and Davide Furceri, “The World Uncertainty Index,” Technical Report, National bureau of economic research 2022. Akinci, Ozge, Sebnem Kalemli-O¨zcan, and Albert Queralto, “Uncertainty Shocks, Capital Flows, and International Risk Spillovers,” Technical Report, National Bureau of Economic Research 2022. Alfaro, Laura and Davin Chor, “Global Supply Chains: The Looming ‘Great Reallocation’,” 2023. Jackson Hole Economic Policy Symposium. , Manuel Garca-Santana, and Enrique Moral-Benito, “On the Direct and Indirect Real Effects of Credit Supply Shocks,” Journal of Financial Economics, 2021, 139 (3), 895–921. Alsagr, Naif and Stefan F. Van Hemmen Almazor, “Oil Rent, Geopolitical Risk and Banking Sector Performance,” International Journal of Energy Economics and Policy, Aug. 2020, 10 (5), 305314. Amiti, Mary, Sang Hoon Kong, and David Weinstein, “The Effect of the US-China Trade war on US Investment,” Technical Report, National Bureau of Economic Research 2020. Bassett, William F, Mary Beth Chosak, John C Driscoll, and Egon Zakrajˇsek, “Changes in Bank Lending Standards and the Macroeconomy,” Journal of Monetary Economics, 2014, 62, 23–40. Bruno, Valentina and Hyun Song Shin, “Cross-border Banking and Global Liquidity,” The Review of Economic Studies, 2015, 82 (2), 535–564. Buch, Claudia, Cathrine T. Koch, and Michael Koetter, “Size, Productivity, and International Banking,” Journal of International Economics, 2011, 85 (2), 329–334. Caldara, Dario and Matteo Iacoviello, “Measuring Geopolitical Risk,” American Economic Review, 2022, 112 (4), 1194–1225. 53

Cetorelli, Nicola and Linda S Goldberg, “Banking Globalization and Monetary Transmission,” The Journal of Finance, 2012, 67 (5), 1811–1843. and Linda S. Goldberg,“LiquidityManagementofU.S.GlobalBanks: InternalCapital Markets in the Great Recession,” Journal of International Economics, 2012, 88 (2), 299– 311. Chodorow-Reich, Gabriel, “The Employment Effects of Credit Market Disruptions: Firm-level Evidence from the 2008–9 Financial Crisis,” The Quarterly Journal of Economics, 2014, 129 (1), 1–59. Choi, Sangyup and Davide Furceri, “Uncertainty and Cross-border Banking Flows,” Journal of International Money and Finance, 2019, 93, 260–274. Claessens, Stijn and Neeltje Van Horen, “Being a Foreigner Among Domestic Banks: Asset or Liability?,” Journal of Banking & Finance, 2012, 36 (5), 1276–1290. and , “The Impact of the Global Financial Crisis on Banking Globalization,” IMF Economic Review, 2015, 63, 868–918. Clayton, Christopher, Matteo Maggiori, and Jesse Schreger, “A Framework for Geopolitics and Economics,” 2023. Working Paper. Correa, Ricardo, Julian Di Giovanni, Linda S. Goldberg, and Camelia Minoiu, “Trade Uncertainty and U.S. Bank Lending,” Staff Reports 1076, New York, NY 2023. , Teodora Paligorova, Horacio Sapriza, and Andrei Zlate, “Cross-border Bank Flows and Monetary Policy,” The Review of Financial Studies, 2022, 35 (1), 438–481. Danisewicz, Piotr, Min Park, Klaus Schaeck, and Yitao Zheng, “Global Bank Lending During Political Conflicts,” Available at SSRN, 2025. Dell’Ariccia, Giovanni and Robert Marquez, “Risk and the Corporate Structure of Banks,” The Journal of Finance, 2010, 65 (3), 1075–1096. 54

Demir, Ender and Gamze Ozturk Danisman, “The Impact of Economic Uncertainty and Geopolitical Risks on Bank Credit,” The North American Journal of Economics and Finance, 2021, 57, 101444. Drott, Constantin, Stefan Goldbach, and Volker Nitsch, “The Effects of Sanctions on Russian Banks in TARGET2 Transactions Data,” Journal of Economic Behavior & Organization, 2024, 219, 38–51. Efing, Matthias, Stefan Goldbach, and Volker Nitsch, “Freeze! Financial Sanctions and Bank Responses,” The Review of Financial Studies, 05 2023, 36 (11), 4417–4459. Fajgelbaum, Pablo D, Pinelopi K Goldberg, Patrick J Kennedy, and Amit K Khandelwal,“TheReturntoProtectionism,”The Quarterly Journal of Economics, 2020, 135 (1), 1–55. Fajgelbaum, Pablo, Pinelopi K Goldberg, Patrick J Kennedy, Amit Khandelwal, and Daria Taglioni, “The US-China Trade War and Global Reallocations,” 2021. National Bureau of Economic Research Working Paper. Federico, Stefano, Fadi Hassan, and Veronica Rappoport, “Trade Shocks and Credit Reallocation,” American Economic Review, April 2025, 115 (4), 114269. Fillat, Jos L., Stefania Garetto, and Arthur V. Corea-Smith, “Global banking and theinternationaltransmissionofshocks: Aquantitativeanalysis,”Journal of International Economics, 2023, 145, 103808. Haas, Ralph De and Neeltje Van Horen, “Running for the Exit? International Bank Lending During a Financial Crisis,” The Review of Financial Studies, 2013, 26 (1), 244– 285. , Mikhail Mamonov, Iliriana Shala, and Alexander Popov, “Violent Conflict and Cross-Border Lending,” Available at SSRN, 2025. Hale, Galina, Tu¨mer Kapan, and Camelia Minoiu, “Shock Transmission through Cross-border Bank Lending: Credit and Real Effects,” The Review of Financial Studies, 2020, 33 (10), 4839–4882. 55

Hassan, Tarek A, Stephan Hollander, Laurence Van Lent, and Ahmed Tahoun, “Firm-level Political Risk: Measurement and Effects,” The Quarterly Journal of Economics, 2019, 134 (4), 2135–2202. Hassan, Tarek Alexander, Jesse Schreger, Markus Schwedeler, and Ahmed Tahoun, “Sources and Transmission of Country Risk,” The Review of Economic Studies, 2023. Forthcoming. Huber, Kilian, “Disentangling the Effects of a Banking Crisis: Evidence from German Firms and Counties,” American Economic Review, 2018, 108 (3), 868–898. Ivashina, V., D. S. Scharfstein, and J. C Stein, “Dollar Funding and the Lending Behavior of Global Banks,” Quarterly Journal of Economics, 2015, 130 (3), 1241–1281. Jiang, Zhengyang, Arvind Krishnamurthy, and Hanno Lustig, “Dollar Safety and the Global Financial Cycle,” Technical Report, National Bureau of Economic Research 2020. Kalemli-O¨zcan, Sebnem, “U.S. Monetary Policy and International Risk Spillovers,” Proceedings of the Jackson Hole Symposium, 2019. , Elias Papaioannou, and Fabrizio Perri, “Global Banks and Crisis Transmission,” Journal of international Economics, 2013, 89 (2), 495–510. Khwaja, Asim Ijaz and Atif Mian, “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market,” American Economic Review, 2008, 98 (4), 1413– 1442. Mamonov, Mikhail, Anna Pestova, and Steven Ongena, “How Banks Anticipate and Propagate Global Financial Sanctions,” Swiss Finance Institute Research Paper, 2022, (23-59). Niepmann, Friederike, “Banking Across Borders,” Journal of International Economics, 2015, 96 (2), 244–265. 56

, “Banking Across Borders with Heterogeneous Banks,” Journal of International Economics, 2023, 142 (C), S002219962300034X. and Tim Schmidt-Eisenlohr, “Institutional Investors, the Dollar, and U.S. Credit Conditions,” Journal of Financial Economics, 2023, 147 (1), 198–220. Peek, Joe and Eric S Rosengren, “Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States,” American Economic Review, 2000, 90 (1), 30–45. Pham, Tho, Oleksandr Talavera, and Andriy Tsapin, “Shock Contagion, Asset Quality and Lending Behaviour: The Case of War in Eastern Ukraine,” Kyklos, 2021, 74 (2), 243–269. Phan, Dinh Hoang Bach, Vuong Thao Tran, and Bernard Njindan Iyke, “Geopolitical Risk and Bank Stability,” Finance Research Letters, 2022, 46, 102453. Pradhan, Swapan-Kumar, ViktorsStebunovs, ElodTakats, and Judit Temesvary, “Geopoliticsmeetsmonetarypolicy: decodingtheirimpactoncross-borderbanklending,” BIS Working Papers 1247, Bank for International Settlements 2025. Rey, H´el`ene, “International Channels of Transmission of Monetary Policy and the Mundellian Trilemma,” IMF Economic Review, 2016, 64 (1), 6–35. Schnabl, Philipp, “The International Transmission of Bank Liquidity Shocks: Evidence from an Emerging Market,” The Journal of Finance, 2012, 67 (3), 897–932. Shen, Leslie Sheng and Tony Zhang, “Risk Sharing and Amplification in the Global Financial Network,” SSRN 4032741, 2024. Temesvary, Judit and Andrew Wei, “Domestic Lending and the Pandemic: How does Banks’ Exposure to COVID-19 Abroad Affect Their Lending in the United States?,” Journal of International Money and Finance, 2024, 143, 103054. Wang, Xinjie, Yangru Wu, and Weike Xu,“GeopoliticalRiskandInvestment,”Journal of Money, Credit and Banking, Forthcoming, 2019. 57

Online Appendix Geopolitical Risk and Global Banking Friederike Niepmann and Leslie Sheng Shen July 2025 A. Additional Materials on Geopolitical Risk Indices This section provides additional details on the geopolitical risk indices constructed in the paper. Appendix Table A.1 lists the search query used to construct the earnings-call transcriptbased country-specific geopolitical risk index, CGPRT. This measure applies the natural languageprocessingmethodfromHassanetal.(2019,2023)usingtheNLAnalyticsplatform. The dictionary of geopolitical risk-related terms is drawn from Caldara and Iacoviello (2022), ensuring close alignment with the newspaper-based index, CGPRN. Appendix Figure A.1 presents the geopolitical and alternative risk indices for three countries: Poland (Panel (a)), the United Kingdom (Panel (b)), and South Korea (Panel (c)). The left panel of each chart shows CGPRN (top), CGPRT (middle), and CGPRT(Fin) (bottom), the latter focusing on geopolitical risk as perceived by financial firms. The right panel displays three broader risk measures for each country: the Country Risk Index (CRI) from Hassan et al. (2023), the World Uncertainty Index (WUI) from Ahir et al. (2022), and fiveyear sovereign CDS spreads. The geopolitical risk indices spike around major geopolitical events, while the broader risk indices respond primarily to large economic shocks. Appendix Figure A.2 plots the 25th, 50th, and 75th percentiles of BGPRN and BGPRT over time. The dispersion across percentiles highlights substantial cross-bank variation in exposure to geopolitical risk, reflecting differences in the geographic composition of U.S. banks’ foreign operations. Moreover, these cross-sectional differences change meaningfully over time, indicating that banks geopolitical risk exposures are both heterogeneous and dynamic. 58

Table A.1: Search Query for CGPR Index Based on Earnings-Call Transcripts Panel A. Search Categories and Search Queries Category Searchqueries Threats 1. Warthreats WarwordsANDthreatwords 2. Peacethreats PeacewordsANDpeacedisruptionwords 3. Militarybuildup MilitarywordsANDbuildupwords 4. Nuclearthreats NuclearbigramsANDthreatwords 5. Terroristthreats TerroristwordsANDthreatwords Acts 6. Beginningofwar WarwordsANDwarbeginwords 7. Escalationofwar ActorswordsANDactorsfightwords 8. Terroristacts TerroristwordsANDterrorismactwords Panel B. Search Words Topicsets Phrases Warwords warORconflictORhostilitiesORrevolution*ORinsurrectionORuprising ORrevoltORcoupORgeopolitical Peacewords peaceORtruceORarmisticeORtreatyORparley Militarywords militaryORtroopsORmissile*ORarmsORweapon*ORbomb*ORwarhead* Nuclearbigrams “nuclear war*” OR “atomic war*” OR “nuclear missile*” OR “nuclear bomb*” OR “atomic bomb*” OR “h-bomb*” OR “hydrogen bomb*” OR “nucleartest”OR“nuclearweapon*” Terrorismwords terror*ORguerrilla*ORhostage* Actorswords allies*ORenemy*ORinsurgent*ORfoe*ORarmyORnavyORaerialOR troopsORrebels Threat/actsets Phrases Threatwords threat*ORwarn*ORfear*ORrisk*ORconcern*ORdanger*ORdoubt* OR crisis OR trouble* OR dispute* OR tension* OR imminent* OR inevitableORfootingORmenace*ORbrinkORscareORperil* Peacedisruptionwords threat*ORmenace*ORreject*ORperil*ORboycott*ORdisrupt* Buildupwords buildup*ORbuild-up*ORsanction*ORblockade*ORembargoORquarantineORultimatumORmobilize* Warbeginwords begin*ORstart*ORdeclare*ORbegunORbeganORoutbreakORbroke outORbreakoutORproclamationORlaunch* Actorfightwords advance* OR attack* OR strike* OR drive* OR shell* OR offensive OR invasionORinvade*ORclash*ORraid*ORlaunch* Terrorismactwords attackORactORbomb*ORkill*ORstrike*ORhijack* Panel C. Excluded words Exclusionwords movie*ORfilm*ORmuseum*ORanniversary*ORobituary*ORmemorial* ORartsORbookORbooksORmemoir*ORpricewarORgameORstory OR history OR veteran* OR tribute* OR sport OR music OR racing OR cancerORrealestateORmafiaORtrialORtax Note: This table lists the search query used to construct the country-specific geopolitical risk index based on earnings-call transcripts(CGPRT). ThequeryisbasedonCaldaraandIacoviello(2022)’swithslightmodification. Thetruncationcharacter (*)denotesasearchincludingallpossibleendingsofaword,(e.g.,“threat*”includes“threat”or“threats”or“threatening”). 59

Figure A.1: Country-specific Geopolitical Risk and Other Risk Indices (a) Poland Russia-Ukraine War )dts( NRPGC 01 5 0 5- )dts( IRC 01 5 0 5- Russia-Ukraine War )dts( TRPGC 01 5 0 5- )dts( IUW 01 5 0 5- Russia-Ukraine War )dts( )nif(TRPGC 01 5 0 5- Height of European Global Sovereign Debt Financial Crisis Crisis 2002q1 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1 2020q1 2022q1 2024q1 )dts( SDC 01 5 0 5- 2002q1 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1 2020q1 2022q1 2024q1 (b) United Kingdom War on Terror— Russia-Ukraine Invasion War of Iraq London Bombing )dts( NRPGC 8 6 4 2 0 2- Brexit Vote Global Financial Crisis )dts( IRC 8 6 4 2 0 2- London Bombing London Terror Threat )dts( TRPGC 8 6 4 2 0 2- Global Brexit Ratification Financial Vote of Brexit Crisis )dts( IUW 8 6 4 2 0 2- Bombings at British London targets in Terror Threat Turkey )dts( )nif(TRPGC 8 6 4 2 0 2- Fi G n C l a r o i n s b c i a s i l al E h u i r c g o f r h i i p g s e e u i s s a r t e n a u s n s n d s o e i v n r m e e c p l e r e e l a o 1 ig s y 9 n e m 9 4 d o e f e n b t t 2002q1 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1 2020q1 2022q1 2024q1 )dts( SDC 8 6 4 2 0 2- 2002q1 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1 2020q1 2022q1 2024q1 60

(continued) (c) South Korea North Korea withdraws from North Korea nonproliferation begins treaty missile tests )dts( NRPGC 6 4 2 0 2- C Le o h lla m p a s n e Covid )dts( IRC 6 4 2 0 2n w o i N t n h o p d r t r t r r o h a e l i w a f K e t s o y r r a f e r t o a io m n laun n c N u h o c e r l s t e h a m r K i o t s e s re s il t a e s and W p a o v li e ti c o a f l d p o r m ot e e s s t t ic )dts( TRPGC 6 4 2 0 2- Beginning of Global Financial Crisis )dts( IUW 6 4 2 0 2- North Korea launches missile and nuclear tests )dts( )nif(TRPGC 6 4 2 0 2- Lehman Collapse 2002q1 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1 2020q1 2022q1 2024q1 )dts( SDC 6 4 2 0 2- 2002q1 2004q1 2006q1 2008q1 2010q1 2012q1 2014q1 2016q1 2018q1 2020q1 2022q1 2024q1 Note: Panels(a),(b),and(c)illustratethecountry-specificgeopoliticalrisk(CGPR)indicesandotherriskindicesforPoland, theUnitedKingdom,andSouthKorea,respectively,coveringtheperiodfrom2002:Q1to2023:Q4. Ineachpanel,theleftcharts, from top to bottom, display CGPR from Caldara and Iacoviello (2022) (CGPRN), CGPR constructed by applying textual analysistoearnings-calltranscriptsusingtheNLAnalyticsplatform(CGPRT),andasub-indexofCGPRT constructedbased solely on earnings-call transcripts of financial firms (CGPRT(fin). The right charts display the country risk index (CRI) by Hassan et al. (2023) (top), the World Uncertainty Index (WUI) by Ahir et al. (2022) (middle), and the five-year CDS spread (bottom)fortherespectivecountries. Allindicesarestandardizedbytheirrespectivestandarddeviationswithinthesample. 61

Figure A.2: Bank-specific Geopolitical Risk Indices (a) BGPRN .2 .1 0 -.1 -.2 )dts( NRPGB 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 2015q1 2020q1 25th percentile 50th percentile 75th percentile (b) BGPRT .3 .2 .1 0 -.1 )dts( TRPGB 2001q1 2006q3 2012q1 2017q3 2023q1 25th percentile 50th percentile 75th percentile Note: Panels (a) and (b) show the bank-specific geopolitical risk (BGPR) indices constructed based on Equation (1) using CGPRN andCGPRT,respectively,overtheperiodsof1985:Q1through2023:Q4and2002:Q1through2023:Q4. Seethenotes under Appendix Figure A.1 for sources and definitions of the CGPR indices. Each panel illustrates the BGPR indices at the 25th,50th,and75thpercentile. Datasources: FFIEC009,FRY-9C,andCallReports. 62

B. Supplementary Results This section presents additional regression results and supporting evidence that complement the main findings in the paper. Appendix Table B.1 reports results from Equation (5), using CGPRT as the main regressor. Similarly to the results with CGPRN in Table 2, banks reduce cross-border exposures to countries experiencing rising geopolitical risk (Columns (3) and (4)), while their operations through local offices in those countries remain largely unchanged (Columns (5) and (6)). Appendix Table B.2 reports regressions based on Equation (5), where the dependent variable is either banks’ local claims in foreign currency (primarily U.S. dollars) (Columns (1) and (2)) or in local currency (Columns (3) and (4)). The results show no significant response of foreign currency claims to geopolitical risk, while local currency-denominated claims exhibit some decline. When geopolitical risk rises, the local currency typically depreciates, reducing the U.S. dollar value of local currency claims without necessarily affecting the underlying local operations. Thus, the observed decline is likely driven by exchange rate effects. Appendix Figure B.3 shows the evolution of cross-border and local claims on Russia following three major geopolitical events: the conflict with Georgia in 2008:Q3, the annexation of Crimea in 2013:Q4, and the invasion of Ukraine in 2022:Q1. Panel (a) presents claims by the U.S. banking sector, while Panel (b) shows claims by all BIS-reporting banking sectors. In both cases, cross-border and local claims declined after each shock, but the decline in local exposures was significantly smaller than cross-border exposures. Appendix Tables B.3 and B.4 parallel Panels (a) and (b) of Table 4, replacing log local claims with a bank’s share of local liabilities in total assets as the key interaction variable. Specifically, CGPRN and CGPRN are interacted with local liability shares at time t−1 t t−1 and t − 2, respectively. The results are qualitatively similar. Appendix Table B.3 shows that banks with higher local funding shares are less likely to reduce overall foreign exposures in response to geopolitical risk. This effect is driven by local exposures (Columns (3) and (4)), while the coefficients for cross-border exposures (Columns (5) and (6)) are statistically indistinguishable from zero. By contrast, Appendix Table B.4 shows that local funding shares do not significantly affect how banks adjust foreign exposures in response to other types of risk, including the country risk index (CRI) by Hassan et al. (2023), the World Uncertainty Index (WUI) by Ahir et al. (2022), and sovereign CDS spreads. These results reinforce those in Table 4 and further support the models Proposition 1(c). Appendix Table B.5 parallels Table 6, using CGPRT instead of CGPRN to construct the BGPR measures in Equation (11). Panel (a) reports loan-level results; Panel (b), bank-level results. As in Table 6, the coefficients on BGPRT1(Local) are negative and significant, indicating that geopolitical risk transmits to domestic credit supply through banks’ local foreign exposures. By contrast, the coefficients on BGPRT1(Cross-border) are not statistically significant, suggesting that spillovers operate primarily through affiliates. When both indices are included, the local interaction remains significant. The results are consistent with the model’s prediction that cross-border claims are more easily liquidated, freeing lending capacity, while affiliate exposures tighten capital constraints. Appendix Table B.6 presents bank-level estimates of the impact of geopolitical risk— 63

measured using subindices of BGPRT that distinguish between threats and acts—on U.S. banks’ loan origination to domestic firms using the FR Y-14 data. Consistent with Panel (b) of Table 7, which reports loan-level results, perceived threats of geopolitical risk have a stronger effect on loan origination decisions than realized shocks. Appendix Table B.6 presents regression results on the spillover effects of geopolitical risk—measured using subindices of BGPRT—on U.S. banks domestic C&I lending standards, using SLOOS data. Similarly to the findings on loan origination, perceived threats of geopolitical risk have a stronger influence on lending standards than realized events. Appendix Table B.8 reports regression results based on Equation (12), where the dependent variable captures banks’ responses to whether they tightened or loosened lending standards for commercial real estate (CRE) loans. The results show that banks also tighten CRE lending standards in response to geopolitical risk, reinforcing the finding that foreign geopolitical shocks lead to a contraction in domestic credit supply. Notably, the U.S. CRE sector is less directly exposed to geopolitical risk than sectors like trade and manufacturing. Therefore, the observed spillovers are unlikely to be driven by credit demand responses. Table B.1: Response of Banks’ Foreign Operations to Geopolitical Risk, CGPRT Total Cross-border Local ln(exp ) (1) (2) (3) (4) (5) (6) bct CGPRT -0.016∗ -0.016∗ -0.023∗∗ -0.023∗∗ -0.015 -0.014 ct (0.009) (0.009) (0.012) (0.011) (0.018) (0.018) CGPRT -0.000 -0.001 -0.004 -0.004 -0.010 -0.011 ct−1 (0.009) (0.008) (0.011) (0.010) (0.026) (0.026) 1(Sanction) -0.120∗∗∗ -0.140∗∗∗ -0.246∗∗∗ t (0.031) (0.034) (0.052) ln(Exch.Rate) -0.009 -0.004 -0.163∗∗∗ t (0.008) (0.011) (0.056) ln(StockIndex) 0.081 0.155∗ 0.200 t (0.071) (0.081) (0.161) ln(Exch.Rate) 0.009 0.011 -0.016 t−1 (0.009) (0.012) (0.058) ln(StockIndex) -0.120∗ -0.176∗∗ -0.198 t−1 (0.063) (0.072) (0.154) Bank-country FE Yes Yes Yes Yes Yes Yes Bank-time FE Yes Yes Yes Yes Yes Yes Observations 35515 33501 34813 32826 11587 11094 R2 0.947 0.949 0.936 0.937 0.938 0.942 Note: Thistablereportsresultsfromregressionsatthebank-country-timelevelbasedonEquation(5)usingtheFFEIC009data coveringthesampleperiod2013:Q1through2022:Q4. CGPRT denotesthecountry-specificgeopoliticalriskindexconstructed based on earnings-call transcripts using the NL Analytics platform. The dependent variable is the log total foreign claims in Columns(1)through(3),logcross-borderclaimsinColumns(4)through(6),andloglocalclaimsinColumns(7)through(9). Columns (1), (4), and (7) show the baseline results for each dependent variable. Columns (2), (5), and (8) add country-level macrocontrols,includingacountry’slogexchangeratevis-`a-vistheU.S.dollar,logdomesticstockpriceindex,andanindicator variablethattakesthevalue1ifthecountryfacesanysanctionsfromtheUnitedStates. Allregressionsincludebank-country and country-time fixed effects. CGPRT is standardized by its respective standard deviation within the sample. Standard errors,showninparentheses,areclusteredatthecountryandtimelevel. *p<.1;**p<.05;***p<.01. 64

Table B.2: Response of Banks’ Local Claims to Geopolitical Risk, Local versus Foreign Currency Claims, CGPRN Foreign currency Local currency ln(exp ) (1) (2) (3) (4) bct CGPRN -0.027 -0.030 -0.032∗ -0.030∗ ct (0.026) (0.028) (0.018) (0.017) CGPRN 0.052 0.047 -0.020 -0.019 ct (0.036) (0.038) (0.017) (0.016) 1(Sanction) 0.289∗∗∗ -0.078∗∗ t (0.061) (0.039) ln(Exch.Rate) -1.076∗∗∗ -0.158∗∗∗ t (0.385) (0.056) ln(StockIndex) -0.067 0.031 t (0.240) (0.148) ln(Exch.Rate) 0.625 -0.028 t−1 (0.392) (0.055) ln(StockIndex) 0.035 -0.060 t−1 (0.234) (0.145) Bank-country FE Yes Yes Yes Yes Bank-time FE Yes Yes Yes Yes Observations 8038 7709 18947 18059 R2 0.887 0.888 0.903 0.907 Note: This table reports results from regressions at the bank-country-time level based on Equation (5) using the FFEIC 009 data for the sample period 1986:Q1 through 2022:Q4. CGPRN denotes the (recent) country-specific geopolitical risk index fromCaldaraandIacoviello(2022). ThedependentvariableistheloglocalclaimsinforeigncurrencyinColumns(1)and(2) andloglocalclaimsinlocalcurrencyinColumns(3)and(4). Columns(1)and(3)showthebaselineresultsforeachdependent variable. Columns (2) and (4) add country-level macro controls, including a country’s log exchange rate vis-a`-vis the U.S. dollar,logdomesticstockpriceindex,andanindicatorvariableequalto1ifthecountryfacesanysanctionsfromtheUnited States. Allregressionsincludebank-countryandcountry-timefixedeffects. CGPRN isstandardizedbyitsrespectivestandard deviation within the sample. Standard errors, shown in parentheses, are clustered at the country and time level. *p < .1; **p<.05;***p<.01. 65

Figure B.3: Banks’ Cross-border and Local Exposures to Russia (b) U.S. Banks’ Claims on Russia 30 Georgia Crimea Ukraine 20 10 0 noilliB DSU ni aissuR no smialC 2001q1 2006q3 2012q1 2017q3 2023q1 Cross-border Local (a) All Banking Sectors’ Claims on Russia 150 Georgia Crimea Ukraine 100 50 0 noilliB DSU ni aissuR no smialC 2001q1 2006q3 2012q1 2017q3 2023q1 Cross-border Local Note: Thefigureillustratescross-borderclaims(blue)andlocalclaims(red)onRussiabytheU.S.bankingsectorinPanel(a) andallBIS-reportingbankingsectorsinPanel(b). Theverticallinesdenotethreegeopoliticalevents: Russia’sconflictwith Georgiain2008:Q3,Russia’sannexationofCrimeain2013:Q4,andRussia’sinvasionofUkrainein2022:Q1. Datasources: BISConsolidatedBankingStatisticsandFFIEC009. 66

Table B.3: Response of Banks’ Foreign Operations to Geopolitical Risk, by Ex Ante Local Liability Share Total Exp. Local Cross-border ln(exp ) (1) (2) (3) (4) (5) (6) bct CGPRN -0.018∗∗ -0.021∗∗ 0.003 0.001 -0.027∗∗∗ -0.030∗∗∗ ct (0.009) (0.010) (0.015) (0.016) (0.010) (0.010) CGPRN × LLShr 0.003 0.001 0.013 0.015 -0.013 -0.013 ct bct−1 (0.005) (0.005) (0.011) (0.011) (0.009) (0.009) CGPRN -0.014 -0.019∗ 0.004 0.001 -0.019∗ -0.023∗∗ ct−1 (0.009) (0.010) (0.014) (0.015) (0.010) (0.012) CGPRN × LLShr 0.015∗∗∗ 0.014∗∗∗ 0.026∗∗ 0.027∗∗ -0.005 -0.004 ct−1 bct−2 (0.006) (0.006) (0.012) (0.012) (0.008) (0.009) LLShr -0.014∗∗ -0.016∗∗ -0.021 -0.024∗ -0.022∗∗ -0.022∗∗ bct−1 (0.007) (0.007) (0.013) (0.013) (0.010) (0.011) LLShr 0.017∗∗∗ 0.016∗∗ 0.032∗∗ 0.037∗∗ 0.010 0.009 bct−2 (0.007) (0.007) (0.016) (0.016) (0.010) (0.010) Macro Controls No Yes No Yes No Yes Bank-country Yes Yes Yes Yes Yes Yes Bank-time FE Yes Yes Yes Yes Yes Yes Observations 94336 77649 30303 27420 93173 76556 R2 0.911 0.919 0.886 0.894 0.891 0.900 Note: This table reports results from regressions at the bank-country-time level based on an augmented version of Equation (5)usingtheFFEIC009dataforthesampleperiod1986:Q1through2022:Q4. CGPRN denotesthe(recent)country-specific geopolitical risk index from Caldara and Iacoviello (2022). LLShr denotes the local liabilities for bank b from country c as bct−1 a share of its total lending to that country, calculated as a four-quarter moving average from t−4 to t−1. The dependent variable is the log total foreign claims in Columns (1) and (2), log local claims in Columns (3) and (4), and log cross-border claimsinColumns(5)and(6). Columns(1),(3),and(5)showthebaselineresultsforeachdependentvariable. Columns(2), (4), and (6) add country-level macro controls, including a country’s log exchange rate vis-`a-vis the U.S. dollar, log domestic stockpriceindex,andanindicatorvariableequalto1ifthecountryfacesanysanctionsfromtheUnitedStates. Allregressions includebank-countryandcountry-timefixedeffects. CGPRN isstandardizedbyitsrespectivestandarddeviationwithinthe sample. Standarderrors,showninparentheses,areclusteredatthecountryandtimelevel. *p<.1;**p<.05;***p<.01. 67

Table B.4: Other Risks and Banks’ Foreign Operations, by Ex Ante Local Liability Position CRI WUI CDS (1) (2) (3) (4) (5) (6) (7) (8) (9) ln(exp ) Tot LC XB Tot LC XB Tot LC XB bct CRIt 0.000 0.022 -0.014 (0.014) (0.018) (0.016) CRIt ×LLS bc h t r −1 0.004 -0.003 0.002 (0.007) (0.012) (0.008) CRIt−1 -0.001 0.029 -0.006 (0.014) (0.019) (0.015) CRIt−1 ×LLS bc h t r −2 0.011 0.016 0.005 (0.007) (0.014) (0.007) WUIt 0.010∗∗ 0.003 0.007 (0.005) (0.008) (0.005) WUIt ×LLS bc h t r −1 -0.002 0.007 -0.006∗∗ (0.002) (0.006) (0.003) WUIt−1 -0.001 0.004 -0.004 (0.005) (0.008) (0.005) WUIt−1 ×LLS bc h t r −2 0.005∗∗ 0.002 0.001 (0.002) (0.006) (0.003) ln(CDS)t 0.030 -0.015 0.035 (0.041) (0.049) (0.051) ln(CDS)t ×LLS bc h t r −1 0.008 0.027∗ -0.003 (0.011) (0.015) (0.012) ln(CDS)t−1 0.015 -0.197∗∗∗ 0.048 (0.038) (0.049) (0.046) ln(CDS)t−1 ×LLS bc h t r −2 -0.003 -0.002 -0.000 (0.010) (0.016) (0.011) LLShr -0.016 -0.020∗ -0.018 -0.015∗∗ -0.031∗∗ -0.018∗ -0.044 -0.108∗∗ -0.010 bct−1 (0.010) (0.012) (0.012) (0.006) (0.015) (0.010) (0.034) (0.055) (0.039) LLShr -0.002 0.007 0.006 0.013∗ 0.038∗∗ 0.011 0.025 0.029 0.013 bct−2 (0.010) (0.014) (0.011) (0.007) (0.018) (0.009) (0.032) (0.062) (0.034) MacroControls Yes Yes Yes Yes Yes Yes Yes Yes Yes Bank-countryFE Yes Yes Yes Yes Yes Yes Yes Yes Yes Bank-timeFE Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 13183 12631 12521 15171 14490 14347 14654 13982 13803 R2 0.960 0.943 0.922 0.959 0.940 0.922 0.961 0.941 0.922 Note: This table reports results from regressions at the bank-country-time level based on an augmented version of Equation (5)withalternativecountry-specificriskindicesasthemainregressor(insteadofCGPR). ThealternativeindicesincludeCRI by Hassan et al. (2023) (Columns (1) through (3)), WUI by Ahir et al. (2022) (Columns (4) through (6)), and log sovereign CDSspreads(Columns(7)through(9)). LLShr denotesthelocalliabilitiesforbankbfromcountrycasashareofitstotal bct−1 lending to that country, calculated as a four-quarter moving average from t−4 to t−1. The dependent variable is the log total foreign claims in Columns (1) through (3), log local claims in Columns (4) through (6), and log cross-border claims in Columns(7)through(9). Allregressionsincludecountry-levelmacrocontrols,includingacountry’slogexchangeratevis-a`-vis theU.S.dollar,logdomesticstockpriceindex,andanindicatorvariableequalto1ifthecountryfacesanysanctionsfromthe UnitedStates,aswellasbank-countryandcountry-timefixedeffects. Standarderrors,showninparentheses,areclusteredat thecountryandtimelevel. *p<.1;**p<.05;***p<.01. 68

Table B.5: Geopolitical Risk Transmission: Cross-border versus Local Exposure, BGPRT (a) Loan Level ln(orig ) (1) (2) (3) (4) (5) (6) bit BGPRT (1(Local)) -0.059∗∗∗ -0.053∗∗ -0.064∗∗∗ -0.057∗∗∗ bt (0.020) (0.021) (0.020) (0.020) BGPRT (1(Cross-border)) -0.051 -0.050 0.263 0.228 bt (0.347) (0.366) (0.342) (0.351) Bank Controls No Yes No Yes No Yes Bank FE Yes Yes Yes Yes Yes Yes Firm-time FE Yes Yes Yes Yes Yes Yes Observations 205642 199753 205642 199753 205642 199753 R2 0.594 0.592 0.594 0.592 0.594 0.592 (b) Bank Level ln(orig ) (1) (2) (3) (4) (5) (6) bt BGPRT(1(Local)) -0.035 -0.036 -0.031 -0.032 bt (0.060) (0.060) (0.057) (0.057) BGPRT (1(Local)) -0.144∗∗ -0.149∗∗ -0.156∗∗∗ -0.159∗∗∗ bt−1 (0.059) (0.059) (0.058) (0.059) BGPRT(1(Cross-border)) -0.822 -0.769 -1.358 -1.309 bt (0.868) (0.857) (0.911) (0.893) BGPRT (1(Cross-border)) 0.565 0.616 0.944 1.015 bt−1 (0.776) (0.780) (0.880) (0.871) Bank Controls No Yes No Yes No Yes Bank FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Observations 475 475 475 475 475 475 R2 0.955 0.956 0.952 0.952 0.955 0.956 Note: Thistablereportsresultsfromregressionswithlogloanoriginationamount(orig)asthedependentvariableusingdata from FR Y-14 for the sample period 2013:Q1 through 2022:Q4. Panel (a) reports results from regressions at the loan level basedonEquation(9),usingamodifiedBGPRconstructedusingEquation(11). Panel(b)reportsresultsfromregressionsat the bank-time level based on Equation (10). Bank controls include lagged Tier 1 capital ratio and liquid-asset ratio. All the geopoliticalriskindicesarestandardizedbytheirrespectivestandarddeviationswithinthesample. Standarderrors,shownin parentheses,areclusteredatthebankandtimelevel. *p<.1;**p<.05;***p<.01. 69

Table B.6: Geopolitical Risk and Domestic Loan Origination: Threat vs. Act (Bank Level) orig (1) (2) (3) (4) bt BGPRT(Threat) -0.049 bt (0.069) BGPRT(Threat) -0.171∗∗ bt−1 (0.069) BGPRT(Act) 0.012 bt (0.038) BGPRT(Act) -0.045 bt−1 (0.039) BGPRTfin(Threat) -0.069 bt (0.067) BGPRTfin(Threat) -0.150∗∗ bt−1 (0.067) BGPRTfin(Act) -0.025 bt (0.035) BGPRTfin(Act) -0.035 bt−1 (0.033) Bank Controls Yes Yes Yes Yes Bank FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes Observations 475 475 475 475 R2 0.956 0.952 0.956 0.953 Note: Thistablereportsresultsfrombank-levelregressionswithlogloanoriginationamount(orig)asthedependentvariable using data from FR Y-14 for the sample period 2013:Q1 through 2022:Q4 based on Equation (10). The main regressors are subindices of BGPRT, or bank-specific geopolitical risk index based on CGPRT, which is constructed with earnings-call transcripts using the NL Analytics platform and captures geopolitical risk perceptions by firms worldwide. BGPRT(Threat) capturesfirms’perceptionsofgeopoliticalriskthreats,andBGPRT(Act)capturestheirperceptionsofgeopoliticalriskstemming from acts. BGPRTfin(Threat) and BGPRTfin(Act) capture financial firms’ perceptions of geopolitical risk stemming from threatsandacts,respectively. BankcontrolsincludelaggedTier1capitalratioandliquid-assetratio. Allthegeopoliticalrisk indicesarestandardizedbytheirrespectivestandarddeviationswithinthesample. Standarderrors,showninparentheses,are clusteredatthebankandtimelevel. *p<.1;**p<.05;***p<.01. 70

Table B.7: Geopolitical Risk and Lending Standards, Threats versus Acts ls (1) (2) (3) (4) bt ∆log(BGPR T(Threat) ) -0.036∗∗∗ bt (0.012) T(Threat) ∆log(BGPR ) -0.011 bt−1 (0.010) T(Act) ∆log(BGPR ) -0.002 bt (0.013) T(Act) ∆log(BGPR ) 0.011 bt−1 (0.012) ∆log(BGPR Tfin(Threat) ) -0.025∗∗ bt (0.011) Tfin(Threat) ∆log(BGPR ) -0.013 bt−1 (0.011) Tfin(Act) ∆log(BGPR ) -0.101 bt (0.089) Tfin(Act) ∆log(BGPR ) 0.056 bt−1 (0.065) Bank FE Yes Yes Yes Yes Macro Controls Yes Yes Yes Yes Bank Controls Yes Yes Yes Yes Observations 1466 1211 1430 144 R2 0.353 0.369 0.347 0.450 Note: Thistablereportsbank-levelregressionresultsbasedonEquation(12),inwhichthedependentvariable is banks’ response to the SLOOS survey question on tightening, maintaining, or loosening credit standards for C&I loans to large and medium-sized firms, using a sample spanning from 1990:Q2 through 2022:Q2. Each column corresponds to a subindex of BGPRT as the main regressor, where BGPRT is bank-specific geopolitical risk index based on CGPRT, which is constructed with earnings-call transcripts using the NL Analytics platform and captures geopolitical risk perceptions by firms worldwide. BGPRT(Threat) captures firms’ perceptions of geopolitical risk threats, and BGPRT(Act) captures their perceptions of geopolitical risk stemming from acts. Similarly, BGPRTfin(Threat) and BGPRTfin(Act) represent the corresponding subcomponents for threats and acts, respectively, when the firm sample is restricted to financial firms. All specifications include bank fixed effects, macroeconomic controls, bank-level controls, and the lagged dependent variable. The geopolitical risk indices are standardized by their respective standard deviations withinthesample. Standarderrors, showninparentheses, areclusteredatthebankandtimelevel. *p<.1; **p<.05; ***p<.01. 71

Table B.8: Geopolitical Risk and Lending Standards on Commercial Real Estate Loans ls (1) (2) (3) (4) (5) (6) bt ∆log(BGPRN) -0.002 0.000 -0.001 bt (0.017) (0.017) (0.017) ∆log(BGPRN ) -0.045∗∗∗ -0.040∗∗ -0.040∗∗ bt−1 (0.017) (0.016) (0.016) ∆log(BGPRT) -0.026 -0.041∗ -0.038∗ bt (0.020) (0.021) (0.020) ∆log(BGPRT ) -0.043∗∗ -0.046∗∗∗ -0.042∗∗ bt−1 (0.017) (0.017) (0.017) Bank FE Yes Yes Yes Yes Yes Yes Macro Controls No Yes Yes No Yes Yes Bank Controls No No Yes No No Yes Observations 1156 1156 1152 704 704 704 R2 0.246 0.298 0.325 0.250 0.305 0.357 Note: This table reports bank-level regression results based on Equation (12), in which the dependent variable is banks’ response to the SLOOS survey question on tightening, maintaining, or loosening credit standards for commercial real estates loans. The main regressor BGPRN (Columns (1) through (3)) is the bank-specific geopolitical risk index constructed from CGPRN, the country-specific geopolitical risk index from Caldara and Iacoviello (2022), using Equation (1). BGPRT (Columns (4) through (6)) is the bank-specificgeopoliticalriskindexderivedfromCGPRT,whichcapturesfirmsgeopoliticalriskperceptions based on earnings-call transcripts processed through the NL Analytics platform. Columns (2) and (5) add macroeconomic controls, including (log) changes in the two-year Treasury yield, the yield curve slope (10y– 2y),theCBOEVolatilityIndex(VIX),theS&P500index,andU.S.industrialproduction. Columns(3)and (6) further control for loan demand, as well as banks liquid-asset and Tier 1 capital ratios. The geopolitical risk indices are standardized by their respective standard deviations within the sample. Standard errors, shown in parentheses, are clustered at the bank and time level. *p<.1; **p<.05; ***p<.01. 72

C. Model: Proofs and Parameter Restrictions C.1 Proofs This section contains the proofs of the propositions stated in Section 4. Proposition 1: Proof. (1) Note that δ ˆA is the solution to πA,C = πL and δ ˆX is the solution to πX,C = πL. 2 2 2 2 Because πX,C < πA,C and ∂π 2 L > 0, δ ˆA > δ ˆX. 2 2 ∂δ (2) Note that ∆δ ˆ = δ ˆ A − δ ˆ X increases with πA,C − πX,C. πA,C − πX,C = (1 − p)D∗i, 2 2 2 2 2 and ∂(πA,C −πX,C) 2 2 = −iD∗ < 0. (C.1) ∂p 2 Because πA,C −πX,C decreases in p, ∆δ ˆ decreases in p. 2 2 (3) ∂(πA,C −πX,C) 2 2 = i(1−p) > 0. (C.2) ∂D∗ 2 Because πA,C −πX,C increases in D∗, ∆δ ˆ increases in D∗. 2 2 2 2 Proposition 2: Proof. (1) L∗ +R L −D i−µL∗α(pG) L∗ +R L −D i−µL∗α(pB) LG,C = 1 1 1 > LB,C = 1 1 1 (C.3) 2 µ 2 µ because pG > pB and α(pG) < α(pB). (2) δL∗ +R L −D i L∗ +R L −D i−µL∗α(pB) LL = 1 1 1 > LB,C = 1 1 1 . (C.4) 2 µ 2 µ Solving for δ delivers δ > (1−α(p)µ). (3) E −µL∗α(φ,pG,pB) L∗ +R L −D i−µL∗α(pB) L = 1 > LB,C = 1 1 1 . (C.5) 1 µ 2 µ Rearranging delivers: (R −1)L −(i−1)D 1 1 1 < (α(pB)−α(φ,pB,pG))L∗. (C.6) µ Because α(pG)−α(φ,pB,pG) < 0 and (R1−1)L1−(i−1)D1 > 0, LG,C > L . µ 2 1 73

C.2 Parameter Restrictions We outline the parameter assumptions needed for a model solution in which the bank optimally invests both domestically and internationally at t = 0 and, when geopolitical risk is high at t = 1, liquidates its cross-border investment but retains its affiliate lending. Profits with liquidation at t = 2 are given by: (cid:18) (cid:19) R−i R−i πL = ( +i) ( +i)E +((δ −i)−(R−i)α(φ,pG,pB))L∗ < πD. (C.7) 2 µ µ 1 2 Second-period profits without liquidation under the cross-border mode are given by: πX,C = pR∗L∗ +LCR−DCi. (C.8) 2 2 2 Plugging in LC = E 2 C −L∗α(p), DC = LC−(L R−D i) and EC = E +(R−1)L +(1−i)D , 2 µ 2 2 1 1 2 1 1 1 we obtain: (cid:18) (cid:19) (cid:18) (cid:19) R−i 1 R−i πX,C = pR∗L∗+(R−i)L +i +(R−i)L∗( (1−i)−α(p))−L∗i2+ +i E i. 2 1 µ µ µ 1 (C.9) Second-period profits without liquidation under the affiliate mode are given by: (cid:18) (cid:19) (cid:18) (cid:19) R−i 1 R−i πA,C = pR∗L∗+(R−i)L +i +(R−i)L∗( (1−i)−α(p))−L∗i2+ +i E i 2 1 µ µ µ 1 +(1−p)D∗i. (C.10) 2 By setting πX,C = πL and πA,C = πL, we can get δ ˆ X and δ ˆ A. 2 2 2 2 −Rαµ+R+R∗µp+iα(p)µ−i ˆ δX = . (C.11) R+iµ−i From πA,C = πL, we obtain: 2 2 −Rαµ+R+R∗µp+iα(p)µ−i+(1−p) µ D∗i δ ˆ A = L∗ 2 . (C.12) R+iµ−i ˆ ˆ ˆ Assume that min{δA,B,δX,G,1} > δ > δX,B. Then the bank does not liquidate the foreign investment in the good state of the world, while the bank liquidates the foreign investment under the cross-border mode in the bad state of the world but not under the affiliate mode. At t = 0, banks chose the investment that maximizes their expected (second-period) profits. The domestic asset invested for two periods delivers the following profits: (cid:18) R−i (cid:19)2 πD = +i E (C.13) 1 µ ˆ ˆ ˆ ˆ Assuming δA,B > δ > δX,B and δA,G > δX,G > δ, expected profits under cross-border 74

investment are: πX = (1−φ)πX,C,G +φπL. (C.14) 2 2 And profits with a foreign affiliates are: πA = (1−φ)πA,C,G +φπA,C,B −κ. (C.15) 2 2 Since πL < πD even for δ = 1, πD < πX implies πX,C,G > πL, hence δX ˆ ,G > 1. In other 2 words, if investing both at home and abroad yields a higher expected return than investing solely in the domestic asset, and given that δ < 1, the cross-border investment is never ˆ ˆ liquidated in the good state. Furthermore, since δX,G < δA,G, the same holds for the affiliate mode in the good state. In addition to the assumptions on δ, we therefore require parameters such that πD < πX, meaning that πX,C,G needs to be sufficiently high, since πX,C,B < πL follows from the 2 2 2 assumption on δ. This condition can be achieved by setting (1 − φ)pGR∗—the expected return in the good state of the world —sufficiently high. If κ = 0, we know that πA > πX. Hence, we additionally require κ to be sufficiently small to satisfy πD < πA. 75