How does the interaction of macroprudential and monetary policies affect cross-border bank lending?

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. How does the interaction of macroprudential and monetary policies affect cross-border bank lending? El˝od Tak´ats and Judit Temesvary 2019-045 Please cite this paper as: Tak´ats, El˝od, and Judit Temesvary (2019). “How does the interaction of macroprudential and monetary policies affect cross-border bank lending?,” Finance and Economics Discussion Series 2019-045. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2019.045. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) 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 Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

How does the interaction of macroprudential and monetary policies affect cross-border bank lending?1 Előd Takáts2 Bank for International Settlements Judit Temesvary3 Federal Reserve Board This draft: May 2019 Abstract: We combine a rarely accessed BIS database on bilateral cross-border lending flows with cross-country data on macroprudential regulations. We study the interaction between the monetary policy of major international currency issuers (USD, EUR and JPY) and macroprudential policies enacted in source (home) lending banking systems. We find significant interactions. Tighter macroprudential policy in a home country mitigates the impact on lending of monetary policy of a currency issuer. For instance, macroprudential tightening in the UK mitigates the negative impact of US monetary tightening on USD-denominated cross-border bank lending outflows from UK banks. Vice-versa, easier macroprudential policy amplifies impacts. The results are economically significant. Keywords: Monetary policy; macroprudential policy; cross-border claims; diff-in-diff analysis JEL codes: F34; F42; G21; G38 1 We are thankful for comments from Stijn Claessens, Bryan Hardy, Machiko Narita, Nikola Tarashev and Rebecca Zarutskie and from seminar participants at the Bank for International Settlements, the University of Basel and the 2019 Georgetown Center for Economic Research Biennial Conference. We are grateful for expert research assistance from Swapan-Kumar Pradhan. Temesvary is thankful for the financial support provided by the BIS in the context of the Central Bank Fellowship Program for the completion of this project. The views expressed in this paper are solely those of the authors, and do not necessarily reflect the views of the Bank for International Settlements or the Board of Governors of the Federal Reserve System. 2 Bank for International Settlements, Centralbahnplatz 2, Basel, CH-4002 Switzerland. Elod.takats@bis.org 3 Federal Reserve Board, 1801 K Street, Washington, DC 20006 USA. Judit.temesvary@frb.gov 1

1. Introduction Central banks and financial regulators use macroprudential tools increasingly frequently after the global financial crisis (IMF-FSB-BIS, 2016). However, our understanding of macroprudential policies, partly because of their short history, is imperfect. For instance, we do not yet understand well how macroprudential policy works together with monetary policy. Given that macroprudential and monetary policies are used in conjunction, understanding their interaction is critical (Yellen, 2010; Claessens, 2013; Praet, 2018). Yet, identifying the interaction between macroprudential and monetary policies is wrought with difficulties, precisely because they are used together: They respond to similar variables, such as credit growth, and often operate through similar channels, such as the cost of bank credit. This makes the identification of the policy interaction particularly challenging. We apply a novel identification strategy using international data to shed light on the interaction between macroprudential and monetary policies. Our identification relies on focusing on a monetary policy that is exogenous to the macroprudential policy, yet affect the same lending flows. We jointly examine (1) the currency-specific monetary transmission in international bank lending (“currency dimension of the bank lending channel” as detailed in Takats and Temesvary (2016)) and (2) source lending banking system-specific macroprudential policies, in driving cross-border bank lending. In our benchmark specifications, we apply a generalization of the Khwaja and Mian (2008) identification method, including country-time fixed effects to control for demand conditions in borrowers’ countries – thereby identifying the policy interaction. To see how we identify the policy interaction, consider an example of USD-denominated cross-border bank lending from UK banks. The currency dimension of the international bank lending channel posits that US monetary policy affects cross-border bank lending denominated in USD, even if the US is neither the source bank lending system nor the borrowers’ country (Takats and Temesvary, 2016). As an example of this channel, US monetary policy tightening would reduce UK-headquartered banks’ USD-denominated lending to Malaysia. At the same time UK macroprudential policies also affect this cross-border bank lending. For instance, macroprudential tightening, by making domestic bank lending relatively more expensive, may drive UK banks’ lending outward and thereby increase cross-border bank lending. In the context of this example, we investigate how UK macroprudential tools interact with US monetary policy in affecting USD-denominated cross-border bank lending outflows from the UK banking system. The assertion that US monetary policy is (almost fully) exogenous to UK macroprudential policy provides an identification which would be impossible to obtain in a single-country setup. 2

We construct a unique dataset from three sources to undertake the identification. We use the “Stage 1 Enhancements” to the Bank for International Settlements’ (BIS) International Banking Statistics. This dataset uniquely allows us to identify the currency dimension of the (international) bank lending channel, that is, monetary policy transmission through the currency denomination of cross-border bank lending (Takats and Temesvary, 2016). With the help of this dataset, we examine cross-border bank lending denominated in the three major internationally used currencies: the US dollar (USD), the euro (EUR) and the Japanese yen (JPY). We combine this data with two distinct macroprudential databases from the International Banking Research Network (IBRN) and the International Monetary Fund (Integrated Macroprudential Policy Database - iMaPP). Both databases contain country-specific measures of macroprudential policy actions. Having two distinct sources for macroprudential policies is critical to ensure robustness, because measuring macroprudential policies is still in its infancy. We conduct our analysis as follows. In the first step, we focus on the period of the effective (zero) lower bound, starting with 2012 Q2 and ending in 2014 Q4, the eve of the year of US monetary policy liftoff (Lhuissier et al, 2019). Given the binding effective lower bound, we use shadow interest rates from Krippner (2016) to capture the stance of post-crisis unconventional policy. For this period, we examine the interaction using regulatory measures from both the IBRN and IMF iMaPP databases. In the second step, we extend our analysis up until 2016 Q4 to study the policy interaction during and after US monetary policy liftoff. For this extended analysis only the IMF iMaPP database is available. Monetary tightening in a reserve currency reduces the availability of that currency across the globe. This reduced liquidity translates into lower cross-border lending flows in that currency. The international dimension of the bank lending channel of monetary policy posits that lending banking systems with more flexible liquidity access (with less funding frictions) are better able to buffer the negative lending impact of such a monetary tightening-induced liquidity crunch in a reserve currency (Takats and Temesvary, 2016). Insofar as tighter macroprudential policies ensure more stable and resilient funding access for financial institutions, we would then expect that lending banking systems with tighter macroprudential policies in place would be better equipped to buffer monetary tightening-induced effects on their cross-border lending flows. In other words, banking systems with tighter macroprudential rules would see a lower impact of monetary policy on their cross-border flows. Conversely, lending banking systems with lax macroprudential policies in place may be less resilient in buffering the impact of monetary policy. Hence, banking systems with lax macroprudential policies in place would see a stronger impact of monetary policy on their cross-border lending flows. Indeed, our results strongly confirm these hypotheses: We find consistent evidence that macroprudential measures enacted in source (lending) banking systems significantly interact with 3

changes in the monetary policy associated with the currency of lending. Furthermore, the interaction term is positive. This means that tighter macroprudential policy mitigates the lending impact of monetary policy (irrespective whether monetary policy tightens or eases) – whereas easier macroprudential policy amplifies the lending impact of monetary policy. Referring back to our earlier example, macroprudential tightening in the UK mitigates the negative impact of US monetary tightening on USD-denominated cross-border bank lending outflows from the UK banking system (say, to Malaysia). The macroprudential-monetary policy interaction that we identify is not only statistically but also economically significant. Given the nature of interactions quantifying the economic impact requires considering both policies simultaneously. As an example, following a 100 basis point monetary tightening over four quarters, cross-border lending outflows decline by around 10 percentage points more in a source banking system that relatively eases macroprudential tools (i.e. India in 2014 Q1) than from a source that relatively tightens such tools (i.e. Netherlands in 2014 Q1). A monetary tightening of 25 basis points over four quarters would still imply in such interaction a 2.5 percentage point decline in lending flows. This impact is substantial in magnitude, relative to the average quarterly growth in bilateral cross-border bank claims of 1.2 percent. Especially so, as this impact comes solely from the interaction, i.e. the effect that we observe in addition to the level effects of monetary and macroprudential policies. Our findings are robust to a range of alternative specifications. We find significant results in both our short and long sample, and both using the IBRN and iMaPP databases. For completeness, we also examine the potential interactions with macroprudential tools applied in borrowers’ countries. However, there we do not find consistently significant results. The results suggest that the interaction between monetary policy of a currency issuer and the macroprudential policy of major lending baking system jurisdictions materially affects the supply of cross-border bank lending. First, this policy interaction matters for central banks in the countries of borrowers to assess credit supply. Relating back to our earlier example, emerging market central banks would benefit from understanding early how the interaction of US monetary policy and UK macroprudential policy affects cross-border USD loan supply to their economies. An early recognition could help to execute the appropriate domestic macroprudential policy response in time for monetary transmission to take effect. Second, these interactions also matter for regulators of major international banks, when they calibrate macroprudential policies. In our example, understanding the policy interaction allows UK regulators to mitigate unintended policy externalities, which result from reserve-currency monetary policy actions. For instance, UK macroprudential easing can amplify the lending impact of US monetary policy. Furthermore, understanding this interaction also matters for 4

the central banks associated with the major international currencies. In our example, gauging policy interactions could aid US policymakers to more precisely assess potential spillbacks to the US. Paradoxically, to the degree such understanding becomes integrated to monetary and macroprudential policy setting over time, the policies themselves would ultimately become less exogenous. Last, but not least, the recognition of such positive interaction is also important when thinking about domestic application of monetary and macroprudential policies. While our quantitative results do not necessarily translate to domestic lending, the qualitative results suggest that central banks might want to think about potential interactions between domestic monetary and macroprudential policies. The paper proceeds as follows. In Section 2 we link our work to the related literature. In Section 3 we describe our data. We present the methodology in Section 4 and detail the results in Section 5. We discuss robustness in Section 6 and conclude with policy implications in Section 7. 2. Related literature Our research focuses on the interaction between macroprudential policies and monetary policies in an international bank lending setup. Hence, it builds on three strands of literature studying the drivers of international bank lending flows: (1) research on the impact of macroprudential policies (2) the relatively new research focusing specifically on the interaction between monetary and macroprudential policies, and (3) research on monetary policy spillovers. First, the research on macroprudential policies dates back to Crockett (2000) and Borio (2003) and is recently reviewed in detail by Claessens (2015) and Galati and Moessner (2018). Elliott et al (2013) provide of historical overview of such policies in the United States. The policy discussion, as shown for instance in the IMF-FSB-BIS (2016) publication, suggests that macroprudential policies might have an international dimension. From the perspective of borrowers’ countries, Houston et al (2012) shows that more strictly regulated jurisdictions receive less cross-border bank credit. Temesvary (2018) and Frame et al (2019) show that banks not only lend less to locations with stricter regulations, but they are also less likely to set up operations there. The body of research in the context of the IBRN’s 2016 project (summarized in Buch and Goldberg (2017)) also shows a wide range of evidence on regulatory impact on cross-border bank lending flows. Takats and Temesvary (2019) provide evidence that macroprudential rules can stabilize cross-border lending flows during times of severe financial stress, such as the taper tantrum. Second, the interaction of monetary and macroprudential policies became one critical focus for policymakers (Yellen, 2010; Claessens, 2013; Claessens and Valencia, 2013; Praet, 2018; Cecchetti 5

et al, 2018) and therefore for economic research. Although earlier literature has addressed such interactions in the domestic context, to the best of our knowledge ours is the first paper to investigate such interaction in the global and cross-border bank lending context. Various models were proposed on how macroprudential policies could interact with monetary policy (Beau et al, 2011, 2012; De Paoli and Paustian, 2013; Brunnermeier and Sannikov, 2014; Smets, 2014; Darracq Paries et al, 2019, Coman and Lloyd, 2019). Broadly, macroprudential and monetary policies aim at different goals: financial stability and stable inflation (or business cycle), respectively. Following the Tinbergen principle the two tools may suffice to reach these two separate goals, but policymakers need to understand the interaction to fine tune the combined policy effects. Yet, the related empirical evidence remains scarce. Based on confidential credit registry data from Latin America Gambacorta and Murcia (2017) argue that macroprudential tools have a greater effect on credit growth when reinforced by the use of monetary policy moving in the same direction. Similarly, Bruno et al (2017) find evidence in the Asian context for the two policies reinforcing each other. Hills et al (2019) investigate this interaction through the external lending of UK banks. Third, there is also a fast-growing literature showing evidence for international monetary policy spillovers through cross-border bank lending (Cetorelli and Goldberg, 2012; Miranda-Agrippino and Rey, 2012; Forbes and Warnock, 2012; Temesvary et al, 2018). Furthermore, there are several papers showing, in line with our identification approach, that the currency denomination of bank lending acts as a separate dimension for the international bank lending channel (Alper et al, 2016; Ongena et al, 2015; Avdjiev and Takats, 2018; Avdjiev et al, 2016; Takats and Temesvary, 2016). Furthermore, our work also builds on research which argues that national borders and economically relevant decision-making units often diverge (see, for instance, Takats and Temesvary (2016) for a review). The discussion dates back to Fender and McGuire (2010) and Cecchetti et al (2010), who argue that the lending bank’s nationality tends to be more relevant than its residence in identifying the decision-making unit. Building on these findings, Avdjiev et al (2015) coin the term of the (absence of) triple coincidence in international finance. This term refers to the phenomenon that national borders, the conventional units of international economic analysis, often do not coincide with the economically relevant decision-making unit. Following these lessons, we focus on “lending banking systems” as opposed to “lending countries”, so that we can follow the decision-making unit as precisely as possible. 3. Data description 3.1 Data on macroprudential measures 6

Our data on country-level regulatory measures come from two sources: the macroprudential database employed by the 2016 IBRN project, also incorporating the 2013 Global Macro Prudential Instruments (GMPI) survey (Cerrutti et al, 2015; Correa et al, 2016; Avdjiev et al, 2017; Berrospide et al, 2017); and the IMF’s Integrated Macroprudential Policy Database (iMap). The IBRN database extends on a quarterly frequency up until 2014 Q4, and the IMF iMaPP database is available up to 2016 Q4. The panels in Table A1 summarize and describe these indices. In our investigation, we focus on strictly macroprudential tools. This distinction matters because both the IBRN and IMF iMaPP databases contain a mix of macroprudential and (micro)prudential measures. Most importantly, both databases contain information on minimum capital requirements. These capital requirements reflect more (micro)prudential considerations, In fact, they quite often reflect the adoption of the Basel III regulatory reform. Therefore, we exclude changes in minimum capital requirements, when we create an index of macroprudential tools. Importantly, we focus on the overall impact of macroprudential rules, rather than formulating hypotheses around specific tools and their impact on cross-border bank lending in our main analysis. Therefore, we construct macroprudential policy indices from both databases. In the construction, we follow similar steps as those taken in the IBRN database. The IBRN and IMF iMaPP regulatory databases describe quarterly changes in the stance of individual macroprudential tools, coded as 1 for tightening and -1 for easing. Our macroprudential index in each database is a country index by time t and country i, which equals 1 if the sum of changes in the individual policy tools listed in Table A1 is greater than or equal to 1, equals -1 if the sum of the instruments is less than or equal to -1, and is 0 otherwise. The two macroprudential databases show a similar, but not identical picture. The correlation across the macroprudential indexes constructed from the two databases is near 0.7. This underlines the importance of investigating interactions using both measures. 3.2 Data on bilateral cross-border bank claims Cross-border bank claims total around US$ 30 trillion globally. These claims include cross-border bank lending and other claims (such as securities holding). Bank for International Settlement’s International Banking Statistics (BIS IBS) provides detailed data about these cross-border claims along several dimensions. In order to study the interaction between home macroprudential tools and the monetary policy of the currency issuer we need to identify three dimensions of the cross-border bank claim data: 7

(A) the currency composition of cross-border claims; (B) the residence of the borrower and (C) the nationality of the lending banking system. The currency composition (A) is necessary to study the currency-specific monetary policy. The borrowers’ residence (B) is necessary to control for credit demand of the borrowers’ countries. The nationality of the lending banking system (C) is necessary to identify the home macroprudential agency whose policy we aim to follow. In our leading example these three dimensions enable to investigate how USD-denominated cross-border bank claims from UK-headquartered banks to Malaysia are affected by the interaction of (A) US monetary policy and (B) UK macroprudential policy while controlling for credit demand in Malaysia (C). In our analysis we use the Stage 1 enhancements of the BIS IBS, because this dataset uniquely allows us to use all three necessary dimensions of the underlying cross-border bank claims data (Table A2). Two main BIS IBS datasets cover cross-border claims: the consolidated and the locational data. The first, consolidated dataset groups claims according to the nationality of banks. It covers residence of borrower (B) and the nationality of the lending banking system (C), but not the currency composition (A). In our case, the consolidated dataset would not allow us to use the currency-specific monetary policy, i.e. to identify the currency dimension of the international bank lending channel. The second dataset, the locational banking statistics defines creditors and debtors according to their residence, consistently with national accounts and balance of payments principles. It has three main subsets: the residence-based, the nationality-based and the Enhanced Stage 1 data. The residence based data has information on the currency composition (A) and the residence of borrower (B), but not on the nationality of the lending banking system (C). This can be an issue with financial centers. For instance, a UK bank’s lending through its Hong Kong subsidiary to Malaysia, would be identified as two separate lending in residence based approach: one loan from the UK to Hong Kong and another one from Hong Kong to Malaysia. In our case, that would mask the impact of the home (i.e. the UK) macroprudential regulator’s impact on lending to Malaysia. In contrast, the nationalitybased data observes nationality of the lending bank (C) along with the currency denomination (A) – but not the residence of the borrower (B). In our case, not having access to the residence of borrower would preclude controlling for credit demand. Finally, the Enhanced Stage 1 dataset provides all three (A, B and C) dimensions. Therefore, it is the most suitable dataset to study our question. The Stage 1 Enhancement to the BIS IBS is available by quarterly frequency starting from 2012 Q2 onward both in stocks (levels) and in currency adjusted flows.4 The stocks and flows are also 4 The start of our sample is determined by data availability. However, 2012 Q2 is also the period that marks the start of the effective lower bound (ZLB) in the euro-area – thus allowing us to focus on a more uniform time frame during which each of our reserve currencies experienced a binding ZLB. 8

available by currency denomination, across the major international currencies.5 We focus on the three main currencies (USD, EUR and JPY) that are the most prevalent in cross-border lending. More precisely, we use quarterly changes in the natural logarithm of bilateral cross-border bank claim stocks denominated in these three currencies. When analyzing the Stage 1 enhanced dataset we use a large cross section that covers 27 lending banking systems and 50 borrowers’ countries.6 The Stage 1 enhanced IBS data is fairly representative, though not yet fully complete. On aggregate, information on the nationality of lending banks is available for more than 90% of global cross-border claims (Avdjiev and Takats, 2018). However, this ratio varies and tends to be higher for larger counterparty countries. Since smaller-scale lending flows can be very volatile, we winsorize the observations at the 5th and 95th percentile as is common in related work (Avdjiev and Takats, 2014; Takats and Temesvary, 2016; Avdjiev and Takats, 2018; Takats and Temesvary, 2019).7 3.3 Data on monetary policy stance Our benchmark sample focuses on the period of the binding effective zero lower bound, preceding the liftoff of US monetary policy from the zero lower bound (2012 Q2 – 2014 Q4). In this period, the major central banks, the Federal Reserve, the European Central Bank and the Bank of Japan relied on “unconventional” expansionary monetary policies. As a result, the short-term policy target interest rates set by these three central banks hit the effective lower bound in early 2009 (Figure 1, left panel). Therefore, we use the currency-specific short-term shadow interest rates (as described in Krippner (2013, 2015 and 2016)) to measure the change in monetary policy stance of the three major reserve currencies (Figure 1, right panel). By construction, the short-term shadow interest rates are not subject to the zero lower bound, and are therefore able to capture expansionary monetary policy actions by dipping into the negative range. 5 The flow is also adjusted for breaks in the series. 6 The 27 lending banking systems are Austria; Australia; Belgium; Brazil; Canada; Chinese Taipei; Denmark; Finland; France; Germany; Greece; India; Ireland; Italy; Japan; Korea; Luxembourg; Mexico; the Netherlands; Norway; Portugal; Spain; Sweden; Switzerland; Turkey; United Kingdom; United States. The 50 borrowing countries are Angola; Austria; Australia; Belgium; Brazil; Bulgaria; Canada; Chile; China; Chinese Taipei; Croatia; Cyprus; Czech Republic; Denmark; Finland; France; Germany; Greece; Hungary; Ireland; Israel; Italy; Japan; Korea; Liberia; Lithuania; Luxembourg; Malta; Marshall; Island; Mexico; Morocco; the Netherlands; New Zealand; Nigeria; Norway; Poland; Portugal; Romania; Russia; Slovakia; Slovenia; South Africa; Spain; Sweden; Switzerland; Turkey; Ukraine; United Kingdom; United States; Vietnam. 7 It is not unprecedented to observe several hundred percentage point changes across some very small bilateral claims even in response to small idiosyncratic shocks, such as a new FDI project. 9

Our larger sample extends through end-2016, including the post-liftoff period of conventional monetary policy actions. However, for consistency and comparability, we continue to use the Krippner shadow rates also in this extended sample. This is appropriate, as by construction the Krippner shadow short-term rates are identical with policy interest rates during conventional monetary policy periods. We define the change in the monetary policy stance as the quarterly change (from t–1 to t, in percentage points) in the short-term shadow interest rate that corresponds to the monetary conditions determined by the central bank that issues currency c. 3.4 Additional macro controls Whenever we do not rely on country*time fixed effects, we control for macroeconomic and financial effects on credit demand in borrowers’ countries and credit supply in source bank lending systems. To do so, we add (real) GDP growth and changes in domestic interest rates as controls in specifications where country*time fixed effects are not included. We also add quarterly changes in the exchange rate between the currencies of the source (home) and the borrowers’ country, to capture any additional valuation effects which may influence banks’ cross-border lending flows. We describe our model variables in detail in Table 1. 4. Estimation methodology We analyze how interactions between macroprudential tools and monetary policy affect bilateral quarterly cross-border lending flows in major currencies. 4.1 Identification The main identification issue we face is that the use of macroprudential tools can be endogenous to the use of monetary policy. In a domestic context, policy makers might observe overheating credit markets and react with either macroprudential or monetary tightening – or a combination of the two. In short, the use of the two policies are typically endogenous in a domestic context. Consequently, when we investigate interactions with source macroprudential tools, we need to focus on the effects of a monetary policy that is not linked to the source bank lending system. Similarly, when we extend the analysis to policy interactions with borrowers’ country macroprudential tools, then we need to examine a monetary policy that is unrelated to borrowers’ country regulatory policies. We achieve identification by focusing on the currency dimension of the bank lending channel. We use the result that the monetary policy of a currency issuer affects bank lending denominated in 10

that currency, irrespective of the source lending system or the borrower country. For instance, US monetary policy affects cross-border bank lending denominated in USD even from UK banks to Malaysia – although neither the UK nor Malaysia uses the dollar as its own currency. This channel of monetary policy transmission is typically exogenous to source (an also to borrowers’) countries. Of course to achieve clear identification, we need to exclude the US both as a source lending banking system and as a country of borrowers, when we investigate USD-denominated lending. Similarly, we exclude euro-area countries and Japan when we analyze EUR and JPY-denominated lending flows, respectively. 4.2 Panel regression setup Our dependent variable, Δclaims is the quarterly change in the log of bilateral claims between the source lending banking system i and borrowers’ country j, denominated in currency c. Our two main explanatory variables are (1) our IBRN and IMF iMaPP indices of applied macroprudential measures (macroprudential) in source bank lending system i as defined in Section 3.1 above, and (2) the change in monetary policy stance (monetary) associated with the major international currencies (USD, EUR, JPY) as measured by the Krippner (2016) shadow rates. Following the standards of the bank lending literature (Kashyap and Stein, 2000; Cetorelli and Goldberg, 2012) in accounting for potential persistence in lending flows, we consistently add the lagged dependent variable to the right-hand side. To strengthen identification, we restrict all our estimations to exclude both same country lending and own currency lending (in the terminology of Takats and Temesvary (2016)). These two sets of lender-borrower pairs could potentially confuse identification. First, same country lending (e.g. US-owned bank subsidiaries lending back to US-based borrowers) suffer from a more severe endogeneity of monetary and macroprudential policies. Second, as discussed in Section 4.1 on identification, own currency lending (e.g. German bank lending in EUR or US banks’ lending in USD) might confound the country and currency-specific impact of monetary policy. We use six equations throughout the paper. The first regression explains lending flows as a function of macroprudential policies in source bank lending system i ( ). In addition, we control for macroeconomic variables both in source bank len 𝛥𝛥 d 𝑚𝑚 in 𝑚𝑚 g 𝑚𝑚 s 𝑚𝑚 y 𝑚𝑚 s 𝑚𝑚 te 𝑚𝑚 m 𝑚𝑚𝑚𝑚 i 𝑚𝑚 ( 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖 ) and borrowers’ country j ( ). Furthermore we apply fixed effects for each source ba 𝛥𝛥 n 𝑚𝑚 k l 𝑚𝑚 e 𝑚𝑚 n 𝑚𝑚 d 𝑚𝑚 in𝑖𝑖𝑖𝑖g system ( ), borrowers𝛥𝛥’ 𝑚𝑚co𝑚𝑚u𝑚𝑚n𝑚𝑚t𝑚𝑚r𝑗𝑗y𝑖𝑖 ( ) and currency ( ) to capture any time-invariant level differenc𝐹𝐹es𝐹𝐹. 𝑖𝑖Finally, we apply time fixe𝐹𝐹d𝐹𝐹 e𝑗𝑗ffects for each qua𝐹𝐹r𝐹𝐹te𝑐𝑐r ( ) to control for unobserved global factors. Taken all the above together, Equation (1) is formally wr 𝐹𝐹 it 𝐹𝐹 te𝑖𝑖n as: 11

1. 4 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 = ∑𝑘𝑘=1(𝛼𝛼1𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖− 𝑘𝑘 + 𝛼𝛼2𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 +𝛼𝛼3𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑗𝑗𝑖𝑖−𝑘𝑘) + + 𝐹𝐹𝐹𝐹𝑖𝑖 +𝐹𝐹𝐹𝐹𝑗𝑗 +𝐹𝐹𝐹𝐹𝑐𝑐 +𝐹𝐹𝐹𝐹𝑖𝑖 +𝜀𝜀𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 In the second regression, we add monetary policy by currency issuer c ( ): 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖 4 2. 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 = ∑𝑘𝑘=1(𝛽𝛽1𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘+𝛽𝛽2𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖−𝑘𝑘 +𝛽𝛽3𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 + 𝛽𝛽4𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑗𝑗𝑖𝑖−𝑘𝑘)+ 𝐹𝐹𝐹𝐹𝑖𝑖 +𝐹𝐹𝐹𝐹𝑗𝑗 +𝐹𝐹𝐹𝐹𝑐𝑐 +𝐹𝐹𝐹𝐹𝑖𝑖 +𝜀𝜀𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 In the third regression, we add our main interest: the interaction between macroprudential and monetary policy ( ): 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖 ∗𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖 3. 4 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 = ∑𝑘𝑘=1(𝜸𝜸𝟏𝟏𝟏𝟏𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏∗𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏+ + 𝛾𝛾2𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 +𝛾𝛾3𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖−𝑘𝑘 + 𝛾𝛾4𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 + + 𝛾𝛾5𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑗𝑗𝑖𝑖−𝑘𝑘)+𝐹𝐹𝐹𝐹𝑗𝑗 +𝐹𝐹𝐹𝐹𝑖𝑖 +𝐹𝐹𝐹𝐹𝑐𝑐 +𝐹𝐹𝐹𝐹𝑖𝑖 +𝜀𝜀𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 While Equation (3) addresses the policy interaction, a potential identification question remains. Namely, the question is the extent to which the macro controls capture non-policy related changes in credit demand from the borrowers’ countries and credit supply from the source bank lending systems. Less than fully controlling for such confounding factors might result in omitted variable bias, which may, in turn, affect our interaction estimates. To address this potential omitted variable bias, we expand the logic outlined in Khwaja and Mian (2008) to a broader context by adding (1) country*time fixed effects for borrower’s country j and (2) currency*time fixed effects for currency c. The borrowers’ country-specific fixed effects allow us to control for any potential direct time-varying country-level credit demand shocks in the borrowers’ country. Similarly, the currency specific currency*time fixed effect controls for any shocks related to the use of that currency. Consequently, we drop the stand-alone macroprudential and macro terms for borrowers’ country j ( and ) and the monetary policy by currency issuer c ( ) th𝛥𝛥a𝑚𝑚t w𝑚𝑚o𝑚𝑚𝑚𝑚u𝑚𝑚ld𝑚𝑚 b𝑚𝑚e𝑚𝑚 𝑚𝑚su𝑚𝑚b𝑚𝑚s𝑚𝑚u𝑚𝑚𝑚𝑚m𝑚𝑚e𝑗𝑗𝑖𝑖d by o𝛥𝛥u𝑚𝑚r e𝑚𝑚x𝑚𝑚te𝑚𝑚n𝑚𝑚𝑗𝑗s𝑖𝑖ive fixed effects. The resulting Equation (4) is wri 𝛥𝛥 tt 𝑚𝑚 en 𝑚𝑚 a 𝑚𝑚 s 𝑚𝑚 : 𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖 12

4. 4 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 = ∑𝑘𝑘=1(𝜹𝜹𝟏𝟏𝟏𝟏𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏∗𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏+ + 𝛿𝛿2𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 +𝛿𝛿3𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘)+𝐹𝐹𝐹𝐹𝑗𝑗∗𝑖𝑖 +𝐹𝐹𝐹𝐹𝑖𝑖 +𝐹𝐹𝐹𝐹𝑐𝑐∗𝑖𝑖 + + 𝐹𝐹𝐹𝐹𝑖𝑖 +𝜀𝜀𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 Next, we add further fixed effects to address potential omitted variables on the credit supply side from source bank lending systems. In other words, we add country*time fixed effects for source bank lending system i so as to focus attention on the interaction of macroprudential and monetary policy. This fixed effect allows us to control for any potential direct time-varying source banking system-specific credit supply shocks. We drop the stand-alone terms for lending system i both for macroprudential policy ( ) and macro controls ( ) that would now be subsumed. The resulting 𝛥𝛥 E 𝑚𝑚 qu 𝑚𝑚 a 𝑚𝑚 t 𝑚𝑚 io 𝑚𝑚 n 𝑚𝑚 𝑚𝑚 (5 𝑚𝑚 ) 𝑚𝑚 i 𝑚𝑚 s 𝑚𝑚 w 𝑚𝑚 r 𝑚𝑚 i 𝑚𝑚 tt 𝑚𝑚 e𝑖𝑖𝑖𝑖n as: 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 5. 4 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 = ∑𝑘𝑘=1𝝀𝝀𝟏𝟏𝟏𝟏𝛥𝛥𝒎𝒎𝒎𝒎 𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏∗𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏+𝐹𝐹𝐹𝐹𝑖𝑖∗𝑖𝑖 +𝐹𝐹𝐹𝐹𝑗𝑗∗𝑖𝑖 + + 𝐹𝐹𝐹𝐹𝑐𝑐∗𝑖𝑖 +𝜀𝜀𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 Finally, we address the potential concern that some unobserved structural drivers embedded in the global cross-border bank lending system drive our result. Technically, we introduce a fixed effect for each lending-borrowing pair to assume such structural impact ( ). For instance, in our earlier example the UK-Malaysia link would receive a fixed effect. Given t𝐹𝐹ha𝐹𝐹t𝑖𝑖 ∗o𝑗𝑗ur identification relies much more on cross-sectional than on time-series variation, this constitutes a demanding specification. In order to avoid overloading the regression with fixed effects, we drop the country*time fixed effects for lending banking systems and borrowers’ countries here and reintroduce the macroeconomic controls ( and ). The resulting final Equation (6) is written as: 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑗𝑗𝑖𝑖 6. 4 𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑐𝑐𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 = ∑𝑘𝑘=1(𝜽𝜽𝟏𝟏𝟏𝟏𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏∗𝛥𝛥𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎𝒎−𝟏𝟏 + 𝜃𝜃2𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑖𝑖𝑖𝑖−𝑘𝑘 +𝜃𝜃3𝑘𝑘𝛥𝛥𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑗𝑗𝑖𝑖−𝑘𝑘)+𝐹𝐹𝐹𝐹𝑖𝑖∗𝑗𝑗 +𝐹𝐹𝐹𝐹𝑐𝑐∗𝑖𝑖 +𝜀𝜀𝑖𝑖𝑗𝑗𝑐𝑐𝑖𝑖 Importantly, while the extensive use of time-country and time-currency specific fixed effects identifies the policy interaction precisely, it also precludes us from being able to observe the impact of source (home) and borrowers’ country policy measures one by one. While Equation (1-3) and partly 13

Equation (4) provide some estimates for such level effects, these results should be treated cautiously due to the identification challenge that the less saturated specifications mentioned above face. In all estimations we apply two-way clustering of the standard errors across the source (lending) banking system and borrowers’ country dimensions. 5. Results Our estimates show consistent evidence that the monetary policy of major currency issuers and the macroprudential policies in source bank lending systems interact in a statistically and economically significant way. In our analysis, we start from relatively simple models and gradually develop more sophisticated estimates as we move from Equation (1) to (6) outlined in Section 4.2. We estimate our benchmark set of specifications first over the unconventional monetary policy period of 2012 Q2 – 2014 Q4 using both the IBRN (Table 2) and IMF iMaPP (Table 3) regulatory databases. We then extend our sample through end-2016, using the IMF iMaPP regulatory data (Table 4). We discuss economic significance and interpretation in separate subsections. 5.1 IBRN data (2012-2014) First, we investigate the policy interaction with the help of the IBRN macroprudential database for the 2012 Q2 – 2014 Q4 period (Table 2). Our first model estimates Equation (1), where only source bank lending system macroprudential policy is included – with currency-specific monetary policy and its interaction omitted (Model 1). We see that the coefficient on macroprudential tightening has a positive and significant coefficient. This is consistent with the assertion that tighter macroprudential regulation increases the costs of lending at the home jurisdiction and thereby makes lending abroad, everything else being constant, more attractive. We then estimate Equation (2), which also includes the impact of the cumulative shadow interest rates (Model 2). We find a negative, albeit insignificant coefficient for the level impact. This is consistent with the observation that tighter monetary policy of a currency issuer implies lower crossborder bank lending in that currency. Importantly, the estimates on the macroprudential coefficient remains significant and of similar size as in Model 1. Next, we turn to estimate our main interest by adding the interaction term between monetary and macroprudential policies to our regressions. Formally, we estimate Equation (3). The results show that the interaction is positive and statistically significant (Model 3). That is, macroprudential tightening in a source bank lending system significantly mitigates the negative impact of a monetary 14

tightening of the currency issuer on cross-border bank lending. (We discuss the interpretation of this interaction in more detail in Section 5.5.) As we discussed in the model setup, omitted variable bias might affect the results of Model (3). That is, there might be some uncontrolled demand or supply factors that could affect our interaction coefficient estimate. We address these concerns by applying a generalization of the Khwaja and Mian (2008)-style identification to address potential non-interaction related demand effects from borrowers’ countries (Equation 4). The interaction results stemming from this estimation remain significant and materially unchanged from our earlier estimates (Model 4). We then extend the Kwaja and Mian (2008)-type of identification from the demand side to the supply side, i.e. to the source bank lending systems. Formally, we estimate Equation (5). The results show that the interaction term remains highly significant and positive (Model 5). Notably, the R2 increases when we add the source*time fixed effects – suggesting that substantial variation in our estimations is due to variance across source lending systems. Finally, as a robustness check, we add source*borrower fixed effects to our estimation to estimate Equation (6). That is, we add a time-invariant fixed effect for each pair of source bank lending system i and borrowers’ country j in our specification. Though this is a very demanding control, the interaction coefficient estimate remains consistently significant (Model 6). Furthermore, its sign and size also remains in line with our earlier models. 5.2 IMF iMaPP data (2012-2014) In the next step, we use the IMF iMaPP data for the 2012 Q2 – 2014 Q4 period (Table 3). This setup allows us to broadly compare the IMF iMaPP estimates to the IBRN estimates.8 We run regressions from Equation (1) to Equation (6) exactly as for the IBRN dataset. A similar picture emerges as before: the interaction term is significant with a positive sign for Models 3, 4 and 6. However, the interaction coefficient estimate becomes insignificant for Model 5. In evaluating the Model 5 results, it is important to emphasize that this specification, which includes the most complete fixed effects on both the source and borrower sides, is extremely demanding of the data. 5.3 IMF iMaPP data (2012-2016) 8 The result are only broadly comparable, because – even though we made the time series consistent – the crosssection differs across the IBRN and IMF iMaPP databases. We address this point in Section 6.5, when we restrict the cross section to those observations which are covered by both the IBRN and the IMF iMaPP data. 15

In the next step, we use the IMF iMaPP data for the 2012 Q2 – 2016 Q4 period (Table 4). This utilizes the most recent regulatory data available. We run regressions from Equation (1) to Equation (6) exactly as before. The results are very close to the earlier findings, in particular to the short sample iMaPP results: the interaction term is significant with a positive sign for Models 3, 4 and 6, while the coefficient estimate remains insignificant in Model 5. Therefore, the results suggest that the statistically significant policy interaction was not only a feature of unconventional monetary policy regimes. Rather, these interaction effects also generalize to the post-US monetary policy liftoff period. 5.4 Economic significance The coefficient estimates on the interaction terms do not allow for straightforward translation to economic significance. The reason is that both macroprudential and monetary policy stances matter for characterizing the interaction effects. In addition, while we have an intuitive understanding of how significant a given monetary tightening is, it is less clear how to assess the size of change in macroprudential policies. Hence, we use percentile ranks to characterize the magnitude of the effects of macroprudential measures. We compare the interaction effect for a 100 basis points tightening in the shadow interest rates over the course of four quarters, evaluated in a source banking system with a substantial strengthening of regulatory policies (at the 99th percentile of macroprudential policy tightening) vs one with easing macroprudential rules (at the 1st percentile).9 The results show that the macroprudential-monetary interaction effects are economically significant (see bottom of Tables 2-4). For instance, our main model estimates show that tighter macroprudential policies in source bank lending systems (comparing the 1st and 99th percentile of macroprudential tightening) mitigate the decline induced by a 100 basis point four-quarter cumulated monetary tightening by around 20 percentage points (Models 3-6 in Table 2). These figures imply an around 5 percentage point mitigating effect for a more moderate, 25 basis point tightening. These estimates are even larger, at around 30 percentage points, when we use the IMF iMaPP data (Tables 3 and 4). 9 We examine economic significance over a wider interval so as to capture sufficient variation in the index, given the high concentration of both our IBRN and iMaPP regulatory indices at zero (Table 1). There are two features of our regulatory indices that contribute to their narrow spread. First, our indices cumulate multiple individual macroprudential tools – as such, our index can show a value of zero simply because one tool tightens while another one eases simultaneously. Second, our indices are measured at a quarterly – rather than annual – frequency, which carries the correspondingly higher probability that in any give quarter a given country may not see a macroprudential action. 16

The interaction is also economically significant when we consider somewhat smaller percentile differences across macroprudential policies.10 For instance, our main model estimates show that tighter macroprudential policies in source bank lending systems (i.e. comparing the 5th and 95th percentile of macroprudential tightening) mitigate the lending decline induced by a 100 basis point four-quarter cumulated monetary tightening by around 10 percentage points (based on Table 2 models). This, for example, is the comparison between India (as the 5th percentile) and the Netherlands (as the 95th percentile) in 2014 Q1. In sum, the estimated interactions are not only statistically, but also economically significant. 5.5 Interpretation of policy interaction effects across policy actions In this subsection, we detail the interpretation of the interaction of monetary and macroprudential policies. Doing so is instructive because there is no established literature or language on how to think about such interaction. We interpret the interaction effects for cases: the combination of monetary easing and tightening along with macroprudential easing and tightening, as Figure 2 illustrates. Let’s consider first monetary easing (left-hand column in Figure 2).11 Monetary easing by the currency issuer, in itself, increases cross-border bank lending denominated in that currency. Now, consider the case in which monetary easing coincides with macroprudential easing (top-left quadrant in Figure 2). In itself, macroprudential easing tends to reduce cross-border bank lending (as discussed in the Table 2 analysis). The interaction of these two policies is positive, consistent with the hypothesis that lending banking systems with more lax macroprudential regulations may be less resilient to funding shocks. Therefore, they would experience a stronger impact of expansionary monetary policy on their cross-border lending. That is, when macroprudential and monetary easing are combined, their interaction increases cross-border bank lending compared to the two standalone effects. Macroprudential easing, therefore, amplifies the impact of monetary easing. Next, consider the case when monetary easing is combined with macroprudential tightening (bottom-left quadrant). In this case, both policies have a positive standalone effects on cross-border bank lending outflows. Yet, their interaction, that is the positive interaction of a positive and negative variable, reduces lending outflows compared to the standalone effects. This positive coefficient is consistent with the hypothesis that tighter macroprudential policies make banking systems better 10 We do not display these values in our tables, because there is not sufficient variation in macroprudential policies in all setups. Of course, the mentioned estimations are available upon request with all details. 11 An alternative approach could be to consider the perspective of macroprudential policy first. We opted against this option, because macroprudential policies are much less examined and discussed than monetary policies. 17

equipped to buffer the impact of monetary policy on their cross-border lending flows. Macroprudential tightening, therefore, mitigates the impact of monetary easing. When we move to examine the interaction effects under monetary policy tightening we see symmetric impacts (middle column in Figure 2). Monetary tightening, in itself reduces cross-border bank lending. Macroprudential easing, in itself, also tends to reduce cross-border bank lending. Furthermore, their positive interaction further decreases cross-border bank lending compared to the two standalone effects (top-right uadrant). In other words, macroprudential easing amplifies the negative impact of monetary tightening. Finally, consider the case when monetary tightening is combined with macroprudential tightening (bottom-right quadrant in Figure 2). In this case, the policies work in the opposite direction: monetary tightening, in itself, reduces cross-border bank lending while macroprudential tightening, again in itself, increases it. Their positive interaction, however, increases cross-border bank lending compared to the two standalone effects In other words, macroprudential tightening mitigates the negative impact of monetary tightening. In tying the above discussion together, a clear picture emerges: tighter macroprudential policy mitigates the lending impact of monetary policy – whereas easier macroprudential policy amplifies the cross-border lending impact of monetary policy (right-hand column in Figure 2). 6. Alternative specifications and robustness checks 6.1 Borrowers’ country macroprudential tools For completeness, we also examine the role of macroprudential tools applied in borrowers’ countries. For instance, if the currency issuer tightens monetary policy, policymakers in borrowers’ countries might want to limit the subsequent contraction in cross-border lending inflows by loosening macroprudential policies in their economies. In Table 5, Columns 1-2, 3-4 and 5-6 repeat the Model 3- 4 specifications from Table 2, 3 and 4, respectively. We do not find consistently significant evidence of interactions between borrowers’ country macroprudential policies and the monetary policy of the currency of borrowing. This suggests that the monetary-borrower macroprudential interaction is insignificant, or at least much weaker than the monetary-source macroprudential interaction. This is in part because the lending impact of borrowers’ country macroprudential actions may depend on the type of action implemented: A tightening of macroprudential tools on banks’ clients (that is, on the credit demand side) in borrowers’ countries can further reduce credit inflows. This would amplify the 18

contractionary lending impact of monetary tightening of the currency issuer. Conversely, a tightening of macroprudential tools on resident lenders (on the credit supply side) in borrowers’ countries can lead borrowers to substitute cross-border for domestic credit. This would mitigate the impact of monetary tightening on inflows. Therefore, at this stage, we would not interpret our results in such a way as to exclude the possibility of policy interaction on the borrowers’ country side. 6.2 Source loan-to value ratio caps In the next step, we focus on a single macroprudential tool: limits on Loan-to-Value (LTV) ratios. While our initial hypothesis does not concern single tools (and rather focuses on the joint effect of macroprudential tools), the LTV is special for both economic and technical reasons. Economically, the LTV is often perceived to be very effective at constraining demand as it does not have to work through price signals (IMF-FSB-BIS, 2016). Furthermore, Alam et al (2019) show emerging evidence that LTV ratio has a significant lending impact. Technically, the LTV is also directly comparable across the IBRN and IMF iMaPP databases. Insofar as tightening in source LTV limits (a credit demand-side measure) reduces borrowers’ credit demand domestically, such tightening would push source banks’ lending “outward” into cross-border lending outflows. Hence, such tightening would mitigate the cross-border lending contraction resulting from monetary tightening by the currency issuer. In analyzing the LTV ratios, we replicate Equations (3) and (4) for both the short and long sample, and for both the IBRN and IMF iMaPP databases. That is, we replicate columns 3 and 4 of Tables 2, 3 and 4 for the LTV ratio (see Columns 1-2, 3-4 and 5-6 of Table 6, respectively). Consistent with our benchmark results, we find significantly positive interactions throughout. That is, tightening source LTV limits significantly mitigates the cross-border lending-reducing impact of a tighter monetary policy. 6.3 Source FX reserve requirements Given the significant results on the LTV ratio caps that operate on the credit demand side, we apply our analytic setup on a credit supply-side tool: reserve requirements on banks’ FX funds. There is some recent evidence that macroprudential FX regulations impact cross-border lending flows (when enacted on banks in borrowers’ countries; Ahnert et al, 2019), and, technically, this tool is also directly comparable across the IBRN and IMF iMaPP databases. In analyzing the FX reserve requirements, we follow similar steps as in the case of the LTV limits. We again replicate Equations (3) and (4) for both the short and long sample, and for both the IBRN 19

and IMF iMaPP databases. That is, we replicate columns 3 and 4 of Table 2, 3 and 4 focusing on FX reserve requirements as the macroprudential tool of interest (see Columns 1-2, 3-4 and 5-6 of Table A3, respectively). Our results show no significant interaction between monetary policy and source lending system FX reserve requirements in driving cross-border lending flows. Importantly, the results do not imply that source FX reserve requirements would not work as macroprudential tools. Rather, they merely suggest that FX reserves requirements do not interact with the monetary policy of the currency issuer in affecting cross border bank lending. Yet, the results, in particular when we combine them with those on LTV ratios, might suggest that not all macroprudential policies imposed on source banking systems are equally effective in mitigating monetary policy effects on cross-border lending flows. 6.4 Level of initial macroprudential stringency As described above, both the IBRN and IMF iMaPP macroprudential databases provide information on changes in regulatory stringency over time, but not on the level of the policy stance. While focusing on changes in regulatory stringency, as we do, is consistent with the approach taken in the vast literature on the lending impact of policies, a concern remains on potential non-linearity. Thus, the level might be relevant in conjunction with the change for macroprudential policies. To address this feature, we use the historic macroprudential changes to create a proxy for the level of macroprudential stance by country. We define a new level variable (Level of Initial Macroprudential Stringency) as the cumulative sum of regulatory changes in each source banking system from 2000 Q1 to 2012 Q1. We define this Level of Initial Macroprudential Stringency both for the IBRN and the IMF iMaPP databases. Naturally, this variable should only be seen as a proxy for the unobserved macroprudential stance and be interpreted cautiously. To examine the impact of this Level of Initial Macroprudential Stringency variable, we horserace its interaction impact with our standard interaction measure (Table A4). More formally, we interact this Level of Initial Macroprudential Stringency with our standard change in Source Regulatory Stringency measure and horserace this interaction with the monetary – macroprudential interaction that has been our main focus. The results confirm that the significance of our monetary – macroprudential change interaction results remains generally robust to controlling for cross-sectional differences across countries in the level of macroprudential stringency. 6.5 Common IBRN – IMF iMaPP Sample 20

We address a potential concern about the implication that the differing cross-section coverage of the IBRN and IMF iMaPP databases may have for our results. While in Section 5.2 we have already estimated our interaction results on the same time series for the two databases, we have not yet addressed the potential impact of cross-section heterogeneity across the two databases. We reestimate Equations (3), (4) and (6) from Tables 2 and Table 3, restricting the estimation sample to a common set of observations for each model (Table A5). The significance of the policy interaction term remains highly consistent with our main findings. 6.6 Interaction term implied model restriction The standard estimation technique implies that all four possible interactions have the same sign and size. However, potentially the four possible interactions (as described in Figure 2 and Subsection 5.5) may differ in size, or at least in size. To control for such implicit model restriction, we separately estimate all four coefficients, i.e. we estimate an interaction coefficient for each quadrant of Figure 2, and test their statistical equivalence. The standard Wald tests cannot refuse the null hypothesis that the interaction coefficient estimates across all four cases are equal, or even that they are pairwise equal. This provides further evidence that our interaction model is well specified. 6.7 Endogeneity of macroprudential policies to monetary policy The main advantage of our identification strategy is that macroprudential policies enacted in source lending systems are almost fully exogenous to the monetary policy of the issuers of the three reserve currencies – thereby avoiding the endogeneity pitfall of studying policy interaction effects in a domestic setting. We further ensure the clarity of our identification strategy by excluding “same country lending” and “own currency lending” from our specifications. However, a potential concern that may remain is the extent to which the macroprudential policies of a reserve currency issuer may be endogenous to the monetary policies of other reserve currency issuers. To address this issue, we exclude completely the US, the euro area and Japan, all three home regions of the issuers of the reserve currencies, from our analysis. Given the limits imposed by the resultant substantial reduction in the cross-section of our dataset, we focus on the long IMF iMaPP (2012-2016) series (i.e. Table 4) in this exercise. Our findings remain robust to this exclusion throughout. In the interest of space, we do not show this table, but make the results available by request. 21

6.8 Foreign currency (FX)-based macroprudential tools An additional noteworthy delineation is the extent to which macroprudential tools applied to banks’ domestic vs. cross-border lending may operate differently. An intuitive way to address such potential differences is to examine macroprudential tools on FX lending separately – as cross-border lending flows are overwhelmingly denominated in non-domestic currencies in non-reserve currency issuing source lending systems. The IMF iMaPP database provides additional information on macroprudential tools imposed on FX lending flows – which tend to be credit supply-side measures. We re-estimate our benchmark specifications using the long IMF iMaPP (2012-2016) dataset, i.e. Table 4, using a newly created macroprudential index encompassing only FX-related macroprudential tools. Our benchmark results remain robust to the use of this new FX lending-focused macroprudential index throughout. In the interest of space, we do not include this table, but make the results available by request. 6.9 Additional Robustness Checks Our benchmark results are also robust to (i) using the IBRN pre-defined macroprudential index construction (which includes minimum capital requirements), (ii) excluding the euro area or emerging market borrowers, and (iii) excluding interoffice claims (claims between parent banks and their subsidiaries) or all interbank lending. In the interest of space, we do not include these tables, but make the results available by request. 7. Conclusion In this paper, we use a novel identification strategy on a unique dataset to examine the interaction between monetary policy and macroprudential policy in cross-border bank lending. We combine the new BIS Stage 1 enhanced banking statistics on bilateral cross-border lending flows with the IBRN’s macroprudential database and the IMF’s iMaPP database. We find statistically significant evidence for a positive interaction. This means that tighter macroprudential policy mitigates the lending impact of monetary policy – whereas easier macroprudential policy amplifies the impact of monetary policy. This is consistent with the hypothesis that tighter macroprudential policies make lending banking systems better equipped to buffer monetary policy-induced effects on their cross-border lending flows. The results are robust to numerous alternative specifications, and are also significant economically. 22

The policy interaction results are economically important from three distinct perspectives. First, they are relevant in those countries where cross-border bank lending plays a major role in credit supply. Central bankers can study the policy interaction effects to understand what the changes in monetary policy of major currency issuers and macroprudential changes by the regulators of major lending banking systems imply for their economy. For instance, referring back to our example of UK bank lending to Malaysia in USD, interaction effects between US monetary policy (affecting the USD) and the macroprudential policy of the UK affect USD-denominated lending inflows, and are thus economically important for borrowers in Malaysia. Second, the results are important for regulators of major international banks to assess the impact of their macroprudential regulation on lending outflows. For instance, UK policymakers might want to consider the regulatory policy interaction with US monetary policy, and its impact on USD-denominated lending outflows, when formulating macroprudential tools in the UK. Third, the results are also economically significant from the perspective of major currency issuer countries, as our findings allow a more precise gauging of potential spillback effects into these economies and externalities for emerging markets. The results are also economically significant indirectly. The international identification highlights meaningful interactions between monetary and macroprudential policies on cross-border bank lending. This suggests that there may be a meaningful interaction between these policies in the domestic setting as well – a strand of research which we hope our results will motivate. Finally, we also hope that our research provides a stepping stone for future economic research to better understand the interaction of regulatory and monetary policies in various contexts. 23

References Ahnert, Toni, Kristin Forbes, Christian Friedrich and Dennis Reinhardt (2019). “Macroprudential FX regulation: Shifting the snowbanks of FX vulnerability,“ working paper. Alam, Zohair, Adrian Alter, Jesse Eisemann, Gaston Gelos, Heedon Kang, Machiko Narita, Erlend Nier and Naixi Wang (2019). “Digging deeper – Evidence on the effect of macroprudential policies from a new database,” IMF Working Paper 19/66. Alper, Koray, Fatih Altunok, Tanju Capacioglu and Steven Ongena (2016). “The effect of US unconventional monetary policy on cross-border bank loans: Evidence from an emerging market,” working Paper. Avdjiev, Stefan and Elod Takats (2014). “Cross-border bank lending during the taper tantrum: the role of emerging market fundamentals,” BIS Quarterly Review (September). Avdjiev, Stefan, Robert N McCauley and Hyun Song Shin (2015). “Breaking free of the triple coincidence in international finance,” BIS Working Paper 524. Avdjiev, Stefan, Agne Subelyte and Elod Takats (2016). “The ECB's QE and euro cross-border bank lending,” BIS Quarterly Review (September) Avdjiev, Stefan, Catherine Koch, Patrick McGuire and Goetz von Peter (2017). “International prudential policy spillovers: A global perspective,” International Journal of Central Banking 13(1), pp. 5-33. Avdjiev, Stefan and Elod Takats (2018). “Monetary policy spillovers and currency networks in crossborder bank lending: Lessons from the 2013 Fed Taper Tantrum,” Review of Finance, 10 October Beau, Denis, Laurent Clerc and Benoît Mojon (2011). “Macro-prudential Policy and the Conduct of Monetary Policy,” Banque de France Occasional Paper 8 (Paris: BdF) Beau, Denis, Laurent Clerc and Benoît Mojon (2012). “Macro-prudential Policy and the Conduct of Monetary Policy,” Banque de France Working Paper 390 Borio, Claudio (2003), “Towards a macroprudential framework for financial supervision and regulation?” BIS Working Paper 128. Berrospide, Jose, Ricardo Correa, Linda Goldberg and Friederike Niepmann (2017). “International banks and cross-border effects of regulation: Lessons from the United States,” International Journal of Central Banking 13(s1), pp. 435-476. Brunnermeier, Markus and Yuliy Sannikov (2014). “A Macroeconomic Model with a Financial Sector,” American Economic Review 104(2), pp 379-421. 24

Bruno, Valentina and Hyun Song Shin (2015). “Cross-border banking and global liquidity,” Review of Economic Studies 82, pp. 535–564. Bruno, Valentina and Hyun Song Shin (2015). “Capital flows and the risk-taking channel of monetary policy,” Journal of Monetary Economics 71, pp. 119–132. Bruno, V, I Shim and H S Shin (2017). “Comparative assessment of macroprudential policies,” Journal of Financial Stability 28, pp. 183–202. Cerrutti, Eugenio, Stijn Claessens and Luc Laeven (2015). “The use and effectiveness of macroprudential policies: New evidence,” Journal of Financial Stability 28, pp. 203-224. Cetorelli, Nicola and Linda S. Goldberg (2012). “Banking globalization and monetary transmission,” The Journal of Finance 67(5), pp. 1811–1843. Cecchetti, Stephen G, Ingo Fender and Patrick McGuire (2010). “Towards a global risk map,” BIS Working Paper 309. Cecchetti Stephen G, Tommaso Mancini-Griffoli, Machiko Narita and Ratna Sahay (2018). “U.S. or Domestic Monetary Policy: Which Matters More for Financial Stability?” paper prepared for the IMF’s Nineteenth Jacques Polak Annual Research Conference “International Spillovers and Cooperation,” November 1-2, 2018, Washington D.C. Claessens, Stijn (2013). “Interactions between Monetary and Macroprudential Policies in an Interconnected World,” IEA-BCU Roundtable on “Capital Flows, Capital Controls and Monetary Policy” Montevideo, December 7th and 8th, 2013. Claessens, Stijn (2015). “An overview of macroprudential policy tools,” Annual Review of Financial Economics 7(1), pp. 397-422. Claessens, Stijn and Fabian Valencia (2013). “The interaction between monetary and macroprudential policies,” VOX EU, 14 March. Coman, Andra, and Simon P Lloyd (2019). “In the Face of Spillovers: Prudential Policies in Emerging Markets,” Bank of England mimeo (March) Crockett, Andrew (2000). “Marrying the micro- and macroprudential dimensions of financial stability,” BIS Speeches (21 September). Darracq Paries, Matthieu, Christoffer Kok, and Elena Rancoita (2019). “Macroprudential policy in a monetary union with cross-border banking,” ECB Working Paper 2260 (March). De Paoli, Bianca and Matthias Paustian (2013). “Coordinating Monetary and Macroprudential Policies,” FRB of New York Staff Report 653. Elliott, Douglas J, Greg Feldberg and Andreas Lehnert (2013). “The History of Cyclical Macroprudential Policy in the United States,” Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board 2013-29. 25

Fender, Ingo and Patrick McGuire (2010). “Bank structure, funding risk and the transmission of shocks across countries: concepts and measurement,” BIS Quarterly Review (September), pp. 63–79. Forbes, Kristin J. and Francis E. Warnock (2012). “Debt and equity-led capital flow episodes,” NBER Working Paper Series 18329. Gambacorta, Leonardo and Andres Murcia (2017). “The Impact of Macroprudential Policies and Their Interaction with Monetary Policy: An Empirical Analysis Using Credit Registry Data,” BIS Working Paper 636 (June). Galati, Gabriele and Richhild Moessner (2011). “Macroprudential policy – a literature review,” BIS Working Paper 337. Galati, Gabriele and Richhild Moessner (2018). “What do we know about the effects of macroprudential policy?” Economica 85, pp. 735-70. Hills, Robert, Simon P Lloyd, Dennis Reinhardt and Rhiannon Sowerbutts (2019). “The Interaction Between Prudential Policy and Monetary Policy: Evidence from UK Banks' External Lending,” Bank of England mimeo (February). IMF-FSB-BIS (2016). “Elements of Effective Macroprudential Policies - Lessons from International Experience”. Kashyap, Anil K., and Jeremy C. Stein (2000). “What do a million observations on banks say about the transmission of monetary policy?” American Economic Review 90(3), pp. 407-428. Khwaja, Asim and Atif Mian (2008). “Tracing the Impact of Bank Liquidity Shocks: Evidence from an Emerging Market,” American Economic Review 98(4), pp. 1413-1442. Krippner, Leo (2013). "Measuring the stance of monetary policy in zero lower bound environments," Economics Letters 118(1), pp. 135–138. Krippner, Leo (2015). “A comment on Wu and Xia (2015), and the case for two-factor shadow short rates,” Centre for Applied Macroeconomic Analysis Working Paper 48/2014. Krippner, Leo (2016). “Documentation for Measures of Monetary Policy,” Reserve Bank of New Zealand Working Paper. Lhuissier, Stéphane, Benoit Mojon and Juan Rubio-Ramirez (2019). “Does the liquidity trap exist?” mimeo. Miranda-Agrippino, Silvia, and Hélène Rey (2012). “World asset markets and the global financial cycle,” National Bureau of Economic Research Working Paper 21722. Praet, Peter (2018). “The interaction between monetary policy and macroprudential policy,” speech at the Money, Macro and Finance Research Group Conference on the Resilience of the Global Financial Architecture, London, 27 September 2018 26

Smets, Frank (2014). “Financial Stability and Monetary Policy: How Closely Linked?,” International Journal of Central Banking 10(2), pp. 263-299. Takats, Elod and Judit Temesvary (2016). “The currency dimension of the bank lending channel in international monetary transmission,” BIS Working Paper 600 (December); also in Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board 2017-001. Takats, Elod and Judit Temesvary (2019). “Can macroprudential measures make cross-border lending more resilient? Lessons from the taper tantrum,” International Journal of Central Banking 15(1), pp. 62-105. Temesvary, Judit (2018). "The Role of Regulatory Arbitrage in U.S. Banks' International Flows: Banklevel Evidence," Economic Inquiry 56(4), pp. 2077-2098. Temesvary, Judit, Steven Ongena and Ann Owen (2018). “A global lending channel unplugged? Does US monetary policy affect cross-border and affiliate lending by global US banks?” Journal of International Economics 112, pp. 50-69. Yellen, Janet (2010). “Macroprudential Supervision and Monetary Policy in the Post-crisis World,” speech at the Annual Meeting of the National Association for Business Economics, Denver, Colorado, October 11 27

Table 1: Summary Statistics [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] Mean S.D. Min p1 p5 p25 p50 p75 p95 p99 Max N Panel A: IBRN Database 2012 Q2 -2014 Q4 Dependent variable: Total Currency-specific Cross- 1.24 32.98 -86.88 -86.88 -57.61 -10.80 -0.03 12.49 66.54 93.51 93.51 8,155 border Lending Flows Regulatory measures: Source PruC6 Macropru 0.06 0.29 -1 -1 0 0 0 0 1 1 1 8,155 Stringency Source Loan-to-Value Cap 0.05 0.21 0 0 0 0 0 0 0 1 1 3,518 Source FX Reserve 0.01 0.13 -1 0 0 0 0 0 0 1 1 8,155 Borrower PruC6 Macropru 0.05 0.26 -1 0 0 0 0 0 1 1 1 8,155 Stringency Macro controls: ∆ Source Policy Interest Rate -0.02 0.37 -1.25 -0.50 -0.25 -0.10 0 0 0.15 1 5.5 8,155 Source Real GDP Growth 1.57 1.90 -5.96 -2.69 -1.30 0.39 1.61 2.45 5.01 7.06 7.99 8,155 Borrower Real GDP Growth 1.76 2.28 -14.78 -3.75 -1.45 0.31 1.71 2.86 6.03 7.50 7.90 8,155 ∆ Source - Borrower Exchange -0.67 9.37 -33.30 -25.58 -17.28 -4.70 0 3.16 13.72 26.24 67.95 8,155 Rate Panel B: IMF iMapp Database 2012 Q2 - 2014 Q4 Dependent variable: Total Currency-specific Cross- 1.19 49.84 -103.60 -103.60 -103.60 -18.36 -0.03 18.91 105.00 105.00 105.00 6,304 border Lending Flows Regulatory measures: Source PruC6 Macropru 0.00 0.19 -1 -1 0 0 0 0 0 1 1 6,304 Stringency Source Loan-to-Value Cap 0.00 0.14 -1 0 0 0 0 0 0 1 1 6,304 Source FX Reserve 0.00 0.06 -1 0 0 0 0 0 0 0 0 6,304 Borrower PruC6 Macropru 0.02 0.20 -1 0 0 0 0 0 0 1 1 6,304 Stringency Macro controls: ∆ Source Policy Interest Rate -0.02 0.45 -1.25 -1.00 -0.25 -0.10 0 0 0.15 1 5.5 6,304 Source Real GDP Growth 1.97 1.73 -3.59 -2.02 -0.91 1.11 2.05 2.71 5.08 7.06 7.99 6,304 Borrower Real GDP Growth 1.75 2.23 -14.78 -5.31 -1.45 0.42 1.73 2.82 6.02 7.40 7.90 6,304 ∆ Source - Borrower Exchange -0.23 10.12 -33.30 -25.58 -17.38 -4.98 0 4.22 16.35 27.57 67.95 6,304 Rate Panel C: IMF iMapp Database 2012 Q2 - 2016 Q4 Dependent variable: Total Currency-specific Cross- 0.34 49.31 -103.60 -103.60 -103.60 -18.97 -0.19 18.63 105.00 105.00 105.00 10,794 border Lending Flows Regulatory measures: Source PruC6 Macropru 0.01 0.22 -1 -1 0 0 0 0 0 1 1 9,967 Stringency Source Loan-to-Value Cap 0.00 0.14 -1 0 0 0 0 0 0 0 1 9,967 Source FX Reserve 0.00 0.05 -1 0 0 0 0 0 0 0 0 9,967 Borrower PruC6 Macropru 0.03 0.22 -1 -1 0 0 0 0 0 1 1 9,967 Stringency Macro controls: ∆ Source Policy Interest Rate -0.03 0.37 -1.25 -0.70 -0.38 -0.05 0 0 0.15 1 5.5 10,794 Source Real GDP Growth 1.93 1.83 -5.40 -2.89 -0.9128 1.06 1.94 2.68 5.32 7.60 8.30 9,954 Borrower Real GDP Growth 1.78 2.32 -17.16 -5.40 -1.45 0.57 1.73 2.85 6.15 7.40 7.90 9,891 ∆ Source - Borrower Exchange 0.65 11.54 -48.30 -26.31 -17.66 -4.98 0 6.09 20.18 38.92 76.08 9,878 Rate

Table 2: Main specifications: Source Macroprudential Stringency - IBRN Database; 2012 Q2 - 2014 Q4 Model [1] [2] [3] [4] [5] [6] Σ∆ Source Macropru Stringency {t-1 to t-4} 9.939 10.08 5.609 5.787 6 [4.796]** [5.069]** [2.782]** [6.401] [5.628] Σ∆ Shadow Interest Rate {t-1 to t-4} -3.319 -4.342 [2.25] [2.304*] Σ∆ Source Macropru Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 9.791 10.733 6.755 10.09 [4.672]** [5.453]** [2.310]*** [6.134]* Σ∆ Borrower Macropru Stringency {t-1 to t-4} -1.181 [5.296] Σ∆ Shadow Interest Rate* Σ∆ Source Macropru Stringency {t-1 to t-4}* Σ∆ Borrower Macropru Stringency {t-1 to t-4} 0.634 [17.43] Constant 2.854 1.792 1.579 2.64 1.765 -0.735 [2.276] [1.962] [2.162] [2.752] [0.393]*** [1.981] Source Macro Controls Yes Yes Yes Yes n/p Yes Borrower Macro Controls Yes Yes Yes n/p n/p Yes Source Fixed Effects Yes Yes Yes Yes -- -- Time Fixed Effects Yes Yes Yes -- -- -- Borrower Fixed Effects Yes Yes Yes -- -- -- Currency Fixed Effects Yes Yes Yes -- -- -- Source * Borrower Fixed Effects No No No No No Yes Borrower * Time Fixed Effects No No No Yes Yes No Source * Time Fixed Effects No No No No Yes No Currency * Time Fixed Effects No No No Yes Yes Yes R - squared 0.07 0.07 0.06 0.10 0.15 0.07 Number of Observations 8,155 8,155 9,173 9,173 9,173 9,173 Economicsignificance:Difference(inpercentagepoints)intheimpactofa100 basispoint changein theshort-term shadowinterest rateassociated with the currency of lending, originating from a source lending system with easing macroprudential rules (at the 1st percentile of the Source Macropru Stringency index) vs a source banking system with tightening macroprudential rules (at the 99th percentile). 19.58 21.47 13.51 20.18 [9.344]** [10.91]** [4.620]*** [12.27]* Thedependent variableis thequarterly changein thebilateral cross-border lending flows (to both bank and non-bank borrowers)from asource lending system I to a borrower country j, denominated in one of the three reserve currencies (USD, EUR, JPY). The coefficients shown are cumulativeover thepreceding four quarters.SourceMacropru Stringenc2y9 is an index ofseveral macroprudential tools enacted at thelevel ofthe sourcebanking system (as shown in TableA1)that weconstruct from theIBRN database over the 2012 Q2 -2014 Q4 period. Two-way clustered standard errors are shown in parentheses; *** p<0.01, ** p<0.05, * p<0.1

Table 3: Main specifications: Source Macroprudential Stringency - IMF iMapp Database; 2012 Q2 - 2014 Q4 Model [1] [2] [3] [4] [5] [6] Σ∆ Source Macropru Stringency {t-1 to t-4} -3.396 -2.586 -13.781 -15.378 -33.45 [11.43] [12] [14.752] [12.194] [12.36]*** Σ∆ Shadow Interest Rate {t-1 to t-4} -15.51 -3.104 [9.757] [5.509] Σ∆ Source Macropru Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 25.2 20.51 -107.485 16.17 [10.13]** [8.481]** [69.491] [7.896]** Σ∆ Borrower Macropru Stringency {t-1 to t-4} -14.96 [21.2] Σ∆ Shadow Interest Rate* Σ∆ Source Macropru Stringency {t-1 to t-4}* Σ∆ Borrower Macropru Stringency {t-1 to t-4} -82.13 [59.71] Constant 4.589 -7.883 -6.838 -7.536 1.037 -2.51 [13.84] [17.2] [5.693] [3.529] [2.496] [5.854] Source Macro Controls Yes Yes Yes Yes n/p Yes Borrower Macro Controls Yes Yes Yes n/p n/p Yes Source Fixed Effects Yes Yes Yes Yes -- -- Time Fixed Effects Yes Yes Yes -- -- -- Borrower Fixed Effects Yes Yes Yes -- -- -- Currency Fixed Effects Yes Yes Yes -- -- -- Source * Borrower Fixed Effects No No No No No Yes Borrower * Time Fixed Effects No No No Yes Yes No Source * Time Fixed Effects No No No No Yes No Currency * Time Fixed Effects No No No Yes Yes Yes R - squared 0.08 0.08 0.08 0.13 0.18 0.08 Number of Observations 6,304 5,393 5,440 5,440 5,440 5,393 Economic significance: Difference (in percentage points) in the impact of a 100 basis point change in the short-term shadow interest rate associatedwiththecurrency oflending,originatingfrom asourcelendingsystem witheasingmacroprudentialrules (atthe1stpercentileofthe Source Macropru Stringency index) vs a source banking system with tightening macroprudential rules (at the 99th percentile). 50.4 41.02 -214.97 32.35 [22.26]** [16.96]** [138.98] [15.79]** Thedependent variableis thequarterly changeinthebilateralcross-borderlendingflows(tobothbank andnon-bank borrowers)from asource lending system I to a borrower country j, denominated in one of the three reserve currencies (USD, EUR, JPY). The coefficients shown are cumulativeovertheprecedingfourquarters.SourceMacropruStringencyis anindexofseveralmacroprudentialtools enactedat thelevelofthe source banking system (as shown in Table A1) that we construct from30t he IMF iMap databaseover the2012 Q2 - 2014 Q4 period.Two-way clustered standard errors are shown in parentheses; *** p<0.01, ** p<0.05, * p<0.1

Table 4: Main specifications: Source Macroprudential Stringency - IMF iMapp Database; 2012 Q2 - 2016 Q4 Model [1] [2] [3] [4] [5] [6] Σ∆ Source Macropru Stringency {t-1 to t-4} -9.988 -10.33 -6.176 -4.954 -6.467 [8.763] [9.463] [6.692] [5.879] [11.12] Σ∆ Shadow Interest Rate {t-1 to t-4} 1.916 2.789 [5.736] [7.066] Σ∆ Source Macropru Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 15.253 14.02 -11.601 14.72 [6]*** [6.10]** [19.485] [6.694]** Σ∆ Borrower Macropru Stringency {t-1 to t-4} -9.533 [11.94] Σ∆ Shadow Interest Rate* Σ∆ Source Macropru Stringency {t-1 to t-4}* Σ∆ Borrower Macropru Stringency {t-1 to t-4} -28.84 [29.1] Constant 11.31 9.94 12.682 -2.65 -3.085 6.386 [8.583] [10.26] [5.805]** [1.540]* [1.903] [3.092]** Source Macro Controls Yes Yes Yes Yes n/p Yes Borrower Macro Controls Yes Yes Yes n/p n/p Yes Source Fixed Effects Yes Yes Yes Yes -- -- Time Fixed Effects Yes Yes Yes -- -- -- Borrower Fixed Effects Yes Yes Yes -- -- -- Currency Fixed Effects Yes Yes Yes -- -- -- Source * Borrower Fixed Effects No No No No No Yes Borrower * Time Fixed Effects No No No Yes Yes No Source * Time Fixed Effects No No No No Yes No Currency * Time Fixed Effects No No No Yes Yes Yes R - squared 0.08 0.08 0.08 0.13 0.18 0.08 Number of Observations 10,794 9,875 10,076 10,076 9,875 9,887 Economic significance: Difference (in percentage points) in the impact of a 100 basis point change in the short-term shadow interest rate associated with the currencyoflending,originatingfrom asourcelendingsystem witheasing macroprudentialrules (atthe1stpercentileoftheSourceMacropru Stringencyindex)vsa source banking system with tightening macroprudential rules (at the 99th percentile). 30.51 28.04 -23.201 29.44 [12]*** [12.20]** [38.97] [13.39]** Thedependentvariableisthequarterly changein thebilateral cross-borderlending flows(to bothbank andnon-bank borrowers)from asourcelendingsystem I to aborrower country j, denominated in oneofthethreereservecurrencies (USD,EUR,JPY).Thecoefficients shown arecumulativeover thepreceding fourquarters. SourceMacropruStringencyisanindexofseveralmacroprudential toolsenacted3a1t thelevelofthesourcebankingsystem (asshown inTableA1)that weconstruct from the IMF iMap database over the 2012 Q2 - 2016 Q4 period. Two-way clustered standard errors are shown in parentheses; *** p<0.01, ** p<0.05, * p<0.1

Table 5: Selected specifications: Borrower Macroprudential Stringency Model [1] [2] [3] [4] [5] [6] Database IBRN IMF iMap IMF iMap 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - Time period 2014 Q4 2014 Q4 2014 Q4 2014 Q4 2016 Q4 2016 Q4 Σ∆ Borrower Macropru Stringency {t-1 to t-4} -0.019 -0.578 -15.05 -15.17 -11.07 -10.42 [2.914] [5.000] [14.05] [13.36] [11.43] [11.93] Σ∆ Shadow Interest Rate {t-1 to t-4} -3.692 -6.466 2.145 [2.610] [4.315] [3.567] Σ∆ Borrower Macropru Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 8.326 9.788 -8.77 -9.929 4.36 5.303 [4.513]* [6.042]* [14.51] [14.80] [12.30] [10.67] Constant 0.21 0.788 -7.372 -2.893 11.26 5.33 [2.386] [0.964] [4.526] [2.782] [3.660]*** [1.804]*** Source Macro Controls Yes n/p Yes n/p Yes n/p Borrower Macro Controls Yes Yes Yes Yes Yes Yes Source Fixed Effects Yes -- Yes -- Yes -- Time Fixed Effects Yes -- Yes -- Yes -- Borrower Fixed Effects Yes Yes No Yes No Yes Currency Fixed Effects Yes -- Yes -- Yes -- Source * Borrower Fixed Effects No No No No No Yes Borrower * Time Fixed Effects No No No No No No Source * Time Fixed Effects No Yes No Yes No Yes Currency * Time Fixed Effects No Yes No Yes No Yes R - squared 0.06 0.12 0.08 0.12 0.08 0.12 Number of Observations 9,173 9,173 5,440 5,440 10,076 10,089 Economicsignificance:Difference(inpercentagepoints)intheimpactofa100 basispoint changein theshort-term shadowinterest rateassociated withthe currencyoflending,toborrowersinaBorrowercountrywitheasingmacroprudentialrules (atthe1stpercentileoftheBorrowerMacropru Stringencyindex) vs a Borrower country with tightening macroprudential rules (at the 99th percentile). 24.98 29.36 -17.54 -19.86 8.719 10.61 [13.54]* [18.13]* [29.02] [29.60] [24.59] [21.33] The dependent variable is the quarterly change in the bilateral cross-border lending flows (to both bank and non-bank borrowers) from a sourcelending system I to a borrower country j, denominated in one of the three reserve currencies (USD, EUR, JPY). The coefficients shown are cumulative over the preceding four quarters. Borrower Macropru Stringency is an index of several macroprudential tools enacted at the level of the Borrower country of borrowers (as shown in TableA1)that weconstruct from theIBRN database(Models 1-2)and theIMF iMapp database (Models 3-6) over the timeperiod indicated at thetopofeachcolumn.Models1-2 replicatetheModel3-4 spec3i2fi cationsfrom Table2.Models3-4 replicatetheModel3-4 specificationsfrom Table3 and Models 5-6 replicate theModel 3-4 specifications from Table 4.Two-way clustered standard errors are shown in parentheses; *** p<0.01,** p<0.05, * p<0.1

Table 6: Selected specifications: Source Loan-to-Value Cap Stringency Model [1] [2] [3] [4] [5] [6] Database IBRN IMF iMapp IMF iMapp 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - Time period 2014 Q4 2014 Q4 2014 Q4 2014 Q4 2016 Q4 2016 Q4 Σ∆ Source Loan-to-Value Cap Stringency {t-1 to t-4} 17.16 18.43 -7.698 -14 0.787 -1.2 [6.405]*** [8.363]** [11.69] [9.242] [18.81] [19.43] Σ∆ Shadow Interest Rate {t-1 to t-4} -1.079 -22.14 -1.389 [5.808] [4.571]*** [2.15] Σ∆ Source Loan-to-Value Cap Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 15.27 16.23 49.56 44.12 33.48 21.77 [8.548]* [7.533]** [12.46]*** [15.56]*** [18.88]* [26.57] Constant 6.506 -11.51 -7.055 -7.923 12.33 -3.509 [5.567] [2.629]*** [2.13]*** [2.748]*** [1.227]*** [1.368]** Source Macro Controls Yes Yes Yes Yes Yes Yes Borrower Macro Controls Yes n/p Yes n/p Yes n/p Source Fixed Effects Yes Yes Yes Yes Yes Yes Time Fixed Effects Yes -- Yes -- Yes -- Borrower Fixed Effects Yes -- Yes -- Yes -- Currency Fixed Effects Yes -- Yes -- Yes -- Source * Borrower Fixed Effects No No No No No No Borrower * Time Fixed Effects No Yes No Yes No Yes Source * Time Fixed Effects No No No No No No Currency * Time Fixed Effects No Yes No Yes No Yes R - squared 0.07 0.15 0.08 0.14 0.07 0.13 Number of Observations 3,796 3,785 5,440 5,440 10,076 10,076 Economicsignificance:Difference(inpercentagepoints)intheimpactofa100basispointchangeintheshort-termshadowinterestrateassociatedwith thecurrency of lending, originating from a source lending system with easing Loan-to-Valuecap rules (at the1st percentileof theSource Loan-to-Valuecap index)vs asource banking system with tightening Loan-to-Value cap rules (at the 99th percentile). 15.27 16.23 99.12 88.25 66.97 43.53 [8.548]* [7.533]** [24.93]*** [31.12]*** [37.76]* [53.14] Thedependentvariableisthequarterlychangeinthebilateralcross-borderlendingflows(tobothbankandnon-bankborrowers)fromasourcelending systemI to a borrower country j, denominated in one of the three reserve currencies (USD, EUR,JPY). Thecoefficients shown are cumulativeover thepreceding four quarters. Source Loan-to-Value Cap Stringency captures limits imposed on Loan-to-Value ratios at the level of the source banking system (as shown in Table A1) that we construct from the IBRN database (Models 1-2) and the IMF iMapp database(Models 3-6)over thetime period indicated at the top of each column. Models 1-2 33 replicate the Model 3-4 specifications from Table 2. Models 3-4 replicate the Model 3-4 specifications from Table 3 and Models 5-6 replicate the Model 3-4 specifications from Table 4. Two-way clustered standard errors are shown in parentheses; *** p<0.01, ** p<0.05, * p<0.1

Table A1: Construction of Macroprudential Indices Panel A: IBRN Macroprudential Subcategories sscb_res Change in sector specific capital buffer: Real estate credit. Requires banks to finance a larger fraction of these exposures with capital. sscb_cons Change in sector specific capital buffer: Consumer credit Requires banks to finance a larger fraction of these exposures with capital. sscb_oth Change in sector specific capital buffer: Other sectors. Requires banks to finance a larger fraction of these exposures with capital. Concrat Change in concentration limit. Limits banks' exposures to specific borrowers or sectors. Ibex Change in interbank exposure limit. Limits banks exposures to other banks. ltv_cap Change in the loan-to-value ratio cap. Limits on loans to residential borrowers. rr_foreign Change in reserve requirements on foreign currency-denominated accounts. rr_local Change in reserve requirements on local currency-denominated accounts. Panel B: IMF iMapp Macroprudential Subcategories CCB Changes in countercyclical capital buffers based on various private sector credit exposures. LCG Changes in limits and penalties on banks' household-sector and corporate-sector credit growth. LTV Changes in limits to the loan-to-value ratios, including thosetargeted at housing, automobile and commercial real estate loans. RR Changes in ieserve requirements (domestic or foreign currency) for macroprudential purposes. Table A2: Characterization of the BIS IBS Stage 1 Enhanced Banking Statistics Currency composition Residence of borrower (B) Nationality of lending bank (C) (A) Consolidated Data No Yes Yes Locational Data by Residence Yes Yes No by Nationality Yes No Yes Stage 1 data Yes Yes Yes 34

Table A3: Selected specifications: Source FX Reserve Requirement Stringency Model [1] [2] [3] [4] [5] [6] Database IBRN IMF iMapp IMF iMapp 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - Time period 2014 Q4 2014 Q4 2014 Q4 2014 Q4 2016 Q4 2016 Q4 Σ∆ Source FX Reserve Requirement Stringency {t-1 to t-4} 15.8 14.48 89.34 77.5 141.2 132.5 [6.254]** [4.687]*** [64.12] [58.48] [19.95]*** [26.85]*** Σ∆ Shadow Interest Rate {t-1 to t-4} -3.577 -22.13 -4.1 [2.551] [4.07]*** [1.363]*** Σ∆ Source FX Reserve Requirement Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} -2.367 1.709 -2.846 10.93 -0.464 37.16 [3.155] [5.495] [48.65] [46.77] [44.69] [42.45] Constant 0.325 2.022 -7.381 -8.457 10.08 -2.909 [2.478] [3.153] [2.703]*** [3.769]** [0.593]*** [1.285]** Source Macro Controls Yes n/p Yes n/p Yes n/p Borrower Macro Controls Yes Yes Yes Yes Yes Yes Source Fixed Effects Yes -- Yes -- Yes -- Time Fixed Effects Yes -- Yes -- Yes -- Borrower Fixed Effects Yes Yes No Yes No Yes Currency Fixed Effects Yes -- Yes -- Yes -- Source * Borrower Fixed Effects No No No No No Yes Borrower * Time Fixed Effects No No No No No No Source * Time Fixed Effects No Yes No Yes No Yes Currency * Time Fixed Effects No Yes No Yes No Yes R - squared 0.06 0.10 0.08 0.14 0.08 0.14 Number of Observations 9,173 9,173 5,440 5,440 10,076 10,076 Economicsignificance: Difference(in percentagepoints)in theimpact ofa100basis pointchangeintheshort-termshadowinterestrateassociatedwith the currency oflending, originatingfrom asource lendingsystem with easing FX reserve requirements (at theminimum level of thesource FX reserve requirement index) vs a source banking system with tightening FX reserve requirements (at the maximum level). -4.734 3.418 -5.693 21.86 -0.928 74.33 [6.311] [10.99] [97.31] [93.53] [89.38] [84.9] Thedependent variableis thequarterly changein thebilateral cross-border lendingflows (toboth bank and non-bankborrowers)fromasourcelending system I to aborrower country j, denominated in oneof thethree reservecurrencies (USD,EUR, JPY).The coefficients shown arecumulative over the precedingfourquarters.SourceFXReserveRequirementStringencycapturesFXreserverequirementsatthelevelofthesourcebankingsystem (asshown in TableA1)that weconstruct from theIBRNdatabase(Models 1-2)and theIMF iMappdatabase(Models3-6)overthetimeperiod indicatedat thetop of each column. Models 1-2 replicate the Model 3-4 specifications from T3a5b le 2. Models 3-4 replicate the Model 3-4 specifications from Table 3 and Models 5-6 replicate the Model 3-4 specifications from Table 4. Two-way clustered standard errors are shown in parentheses; *** p<0.01, ** p<0.05, *

Table A4: Selected specifications: Role of Initial Macropru Stringency Model [1] [2] [3] [4] [5] [6] Database IBRN IMF iMapp IMF iMapp 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - 2012 Q2 - Time period 2014 Q4 2014 Q4 2014 Q4 2014 Q4 2016 Q4 2016 Q4 Σ∆ Source Macropru Stringency {t-1 to t-4} 14.37 15.04 -47.04 -41.01 -9.051 -8.552 [7.95]* [7.215]** [39.29] [39.98] [10.98] [10.89] Σ∆ Shadow Interest Rate {t-1 to t-4} -4.602 -26.79 -4.198 [2.628]* [6.008]*** [4.629] Σ∆ Source Macropru Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 10.37 11.38 -5.801 -3.611 14.2 13.75 [6.082]* [5.556]** [20.95] [20.15] [7.606]* [7.844]* Σ∆ Source Macropru Stringency * Level of Initial Macropru Stringency -1.635 -1.662 7.567 5.808 0.445 0.642 [1.238] [1.366] [9.922] [10.52] [2.318] [2.281] Constant 0.883 2.587 -5.952 -7.743 11.89 -2.627 [2.541] [2.855] [5.569] [3.31]** [6.756]* [2.868] Source Macro Controls Yes Yes Yes Yes Yes Yes Borrower Macro Controls Yes n/p Yes n/p Yes n/p Source Fixed Effects Yes Yes Yes Yes Yes Yes Time Fixed Effects Yes -- Yes -- Yes -- Borrower Fixed Effects Yes -- Yes -- Yes -- Currency Fixed Effects Yes -- Yes -- Yes -- Source * Borrower Fixed Effects No No No No No No Borrower * Time Fixed Effects No Yes No Yes No Yes Source * Time Fixed Effects No No No No No No Currency * Time Fixed Effects No Yes No Yes No Yes R - squared 0.06 0.10 0.08 0.14 0.08 0.14 Number of Observations 9,173 9,173 5,440 5,440 10,076 10,076 Economic significance: Difference (in percentage points) in the impact of a 100 basis point change in the short-term shadow interest rate associated with the currency of lending, originating from a source lending system with easing Loan-to-Valuecap rules (at the1st percentileof theSource Loan-to-Valuecap index) vs a source banking system with tightening Loan-to-Value cap rules (at the 99th percentile). 20.74 22.76 -11.6 -7.223 28.41 27.5 [12.16]* [11.11]** [41.89] [40.31] [15.21]* [15.69]* The dependent variable is the quarterly change inthe bilateralcross-border lendingflows (toboth bank and non-bank borrowers) from a sourcelending system I to a borrower country j, denominated in one of the three reserve currencies (USD, EUR, JPY). The coefficients shown are cumulative over the precedingfourquarters.LevelofInitialMacropruStringency capturesregulatory actions(∆SourceMacropruStringency)cumulatedoverthefull20002012 period for each source country. Models 1-2 replicate the Model 3-4 specifi3ca6t ions from Table 2. Models 3-4 replicatethe Model3-4 specifications from Table 3 and Models 5-6 replicate the Model 3-4 specifications from Table 4. Two-way clustered standard errors are shown in parentheses; *** p<0.01, **

Table A5: Selected specifications: Source Macroprudential Stringency - IBRN and IMF iMapp Common Sample; 2012 Q2 - 2014 Q4 Model [1] [2] [3] [4] [5] [6] Database IBRN IMF iMapp Σ∆ Source Macropru Stringency {t-1 to t-4} 24.25 23.62 23.52 -13.71 -12.19 -36.92 [7.916]*** [12.86]* [15.08] [13.03] [15.91] [29.86] Σ∆ Shadow Interest Rate {t-1 to t-4} 5.686 -25.92 [5.854] [10.61]** Σ∆ Source Macropru Stringency * Σ∆ Shadow Interest Rate {t-1 to t-4} 24.78 26.29 26.28 31.66 34.74 7.64 [7.782]*** [10.73]** [10.21]*** [12.76]** [7.471]*** [9.817] Σ∆ Borrower Macropru Stringency {t-1 to t-4} -9.557 -25.15 [9.157] [20.7] Σ∆ Shadow Interest Rate* Σ∆ Source Macropru Stringency {t-1 to t-4}* Σ∆ Borrower Macropru Stringency {t-1 to t-4} 17.53 88.11 [45.67] [54.88] Constant 3.23 0.859 -0.504 -7.004 -2.291 -5.423 [4.571] [9.275] [5.175] [9.666] [4.015] [9.676] Source Macro Controls Yes Yes Yes Yes Yes Yes Borrower Macro Controls Yes n/p Yes n/p Yes Yes Source Fixed Effects Yes Yes -- Yes Yes -- Time Fixed Effects Yes -- -- -- Yes -- Borrower Fixed Effects Yes -- -- -- Yes -- Currency Fixed Effects Yes -- -- -- Yes -- Source * Borrower Fixed Effects No No Yes No No Yes Borrower * Time Fixed Effects No Yes No Yes No No Source * Time Fixed Effects No No No No No No Currency * Time Fixed Effects No Yes Yes Yes No Yes R - squared 0.11 0.18 0.11 0.10 0.19 0.12 Number of Observations 2,787 2,784 2,787 2,787 2,784 2,787 Economicsignificance:Difference(inpercentagepoints)intheimpactofa100basispointchangeintheshort-termshadowinterestrateassociatedwith the currency of lending, originating from a source lending system with easing source macroprudential stringency (at the 1st percentile of the Source Macropru Stringency) vs a source banking system with tightening macroprudential rules (at the 99th percentile). 49.56 52.58 52.56 63.33 69.48 15.28 [15.56]*** [21.46]** [20.42]*** [25.52]** [14.94]*** [19.63] Thedependentvariableisthequarterlychangeinthebilateralcross-borderlendingflows(tobothbankandnon-bankborrowers)fromasourcelending systemItoaborrowercountryj,denominatedinoneofthethreereservecurrencies(USD,EUR,JPY).This tableshows resultsof estimationson adata samplethatwasconstructedsothateachdatapointispresentinboththeIBRNandIMFiMappestimationssamples(fromTables2and3,respectively). The coefficients shown are cumulative over the preceding four quarters. Borrower Macropru Stringency is an index of several macroprudential tools enactedattheleveloftheBorrowercountryof borrowers(as shownin TableA1) thatwe constructfrom theIBRN database(Models 1-3)and theIMF 37 iMap database (Models 4-6) over the 2012 Q2 - 2014 Q4 period. Models 1-3 replicate the Model 3-4 and 6 specifications from Table 2. Models 4-6 replicate the Model 3-4 and 6 specifications from Table 3. Two-way clustered standard errors are shown in parentheses; *** p<0.01, ** p<0.05, * p<0.1

Short –term policy and shadow interest rates In per cent Figure 1 Target rates Shadow rates Sources: Krippner (2016); national data. Figure 2: Policy interactions How do monetary and Monetary easing Monetarytightening Macroprudentialpolicy macroprudential policy (inlending currency) (inlending currency) impact on monetary policy interactions impact crossborder bank lending? Macroprudentialeasing Amplify (positive) Amplify (negative) Amplify (in source bank lending system) Macroprudentialeasing Macroprudentialeasing Macroprudentialeasing strengthens the positive strengthens the negative strengthens the impact of impact of monetary easing impact of monetary monetary policy tightening Macroprudentialtightening Mitigate (negative) Mitigate (positive) Mitigate (in source bank lending system) Macroprudentialtightening Macroprudentialtightening Macroprudentialtightening weakens the positive impact of weakens the negative impact weakens the impact of monetary tightening of monetary tightening monetary policy 38