The Role of U.S. Monetary Policy in Global Banking Crises

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Role of U.S. Monetary Policy in Global Banking Crises Bora Durdu, Alex Martin, and Ilknur Zer 2019-039 Please cite this paper as: Durdu, C. Bora, Alex Martin, and Ilknur Zer (2019). “The Role of U.S. Monetary Policy in Global Banking Crises,” Finance and Economics Discussion Series 2019-039. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2019.039. 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.

The Role of U.S. Monetary Policy in Global Banking Crises∗ C. Bora Durdu Alex Martin Federal Reserve Board Federal Reserve Board Ilknur Zer Federal Reserve Board February 2019 Abstract We examine the role of U.S. monetary policy in banking crises globally by using a cross-country database spanning 1870-2010 across 69 countries. U.S. monetary policy tightening increases the probability of a banking crisis for those countries with direct linkages to the United States, either in the form of trade links or significant share of USD-denominated liabilities. Conversely, if a country is integrated globally, ratherthanhavingadirectexposure, theeffectisambiguous. Onepossible channel we identify is capital flows: if the correction in capital flows is disorderly (e.g., sudden stops), the probability of a banking crisis increases. These findings suggest that the effect of U.S. monetary policy in global banking crises is not uniform and largely dependent on the nature of linkages with the United States. JEL Classification: E44, E52, F42, G15 Keywords: bankingcrises,financialstability,monetarypolicyshocks,suddenstops ∗We would like to thank Luca Guerrieri and the seminar participants at the Federal Reserve Board, IMF 19th Jacques Polak Annual Research Conference, Central Bank of Turkey Conference on “Changing Global Economic Landscape” for helpful discussions. We are also thank to Michael Siemer for his contributions to an earlier version of this project. All remaining errors are exclusively our responsibility. Correspondence: bora.durdu@frb.gov, Alex.Martin@frb.gov, ilknur.zerboudet@frb.gov. The views expressed in this paper are those of the authors and should not be attributed to the Board of Governors of the Federal Reserve System or its staff.

1 Introduction Do U.S. monetary policy actions affect financial stability in foreign economies? Past experience suggests that they do, as in the early 1980s international debt crises preceded by the 1980-82 Volcker tightening and the Mexican Peso Crisis preceded by the 1994 Greenspan tightening. Additionally, during the last five years, since the current U.S. monetary policy easing cycle started, we have seen the “taper tantrum” in the spring of 2013, the strengthening of the dollar and the fall in commodity prices in mid-2014, the RMB shock in mid-2015, and the turbulence in Argentina and Turkey in mid-2018. Motivated by these observations on earlier international financial crises and the recent attention in academic and policy circles on the topic, we raise the following questions in this paper: What are the effects of U.S. monetary policy on foreign banking crises? Are there differential effects depending on country characteristics and the nature of linkages with the United States and the rest of the world? To answer these questions, we examine how the interaction of U.S. monetary policy with a country’s exposure to the United States and the rest of the world affects the probability of a banking crisis in those countries by using various macroeconomic, financial, and trade indicators. To this end, we construct a historical cross-country database covering 69 countries over the 1870-2010 period. We capture banking sector stress using the systemic banking crisis database of Reinhart and Rogoff (2009).1 On one hand, extant literature shows that local monetary policy decisions affect financial stability. Financial markets may react to monetary policy changes as they influence the pricing of risky assets, including equity and bonds. Bernanke and Kuttner (2005) and Gilchrist et al. (2015), among others, show that monetary policy decisions affect equity and corporate bond risk premiums, respectively. Moreover, an accommodative monetary policy can increase financial instability by leading to buildups of financial vulnerabilities, suchascreditboomsorexcessivefinancialleverage(AdrianandLiang,2018). Suchboomsaregenerallyassociatedwithover-optimisticinvestorsandlowerloanquality. When booms reverse, increasing number of defaults may stress the financial system and 1We check the sensitivity of our findings by employing other systemic crisis databases of Bordo et al. (2001); Laeven and Valencia (2012); Gourinchas and Obstfeld (2012); Schularick and Taylor (2012) . TheresultsdiscussedinSection4.4arequalitativelysimilartothoseusingReinhartandRogoff’s(2009) definition. 1

increase the likelihood of a banking crisis (See Schularick and Taylor, 2012; Baron and Xiong, 2017; Danielsson et al., 2018). However, this literature does not address how external monetary policy decisions (e.g., monetary policy decisions in an hegemon country) could affect domestic financial stability. On the other hand, in a recent influential paper, Rey (2015) argues that there is a global financial cycle, which is driven by U.S. monetary policy decisions. Gourinchas (2017) shows in the context of an estimated DSGE model that the degree of financial spillovers between the United States and emerging market economies matters for the transmission of U.S. monetary policy, a potential transmission mechanism of Rey’s (2015) global financial cycles. Jorda et al. (2018) outline another potential linkage by showing that U.S. monetary policy plays an important role in shaping risk appetite across global equity markets. Our paper fills the gap between the two aforementioned strands of literature. From a historical view, we study whether U.S. monetary policy is a uniform driver of financial vulnerabilities abroad or its effects are dependent on the country’s integration with the United States and the rest of the world. Examining logit regressions in our panel of countries, we find that U.S. monetary policy tightening has a significant and positive contemporaneous effect on the probability of a banking crisis for those countries with direct exposures to the United States. The impact is statistically significant and economically meaningful. A 1% tightening in U.S. monetary policy increases the default probability by about 1% to 7% for a given level of exposure to the U.S. However, if a county is integrated globally rather than having a primary direct exposure to the United States, U.S. monetary policy has an ambiguous effect on that country’s probability of crisis. The results are robust to alternative definitions of monetary policy stance and monetary policy shocks as well as country specific macroeconomic and financial indicators such as GDP growth, inflation, and institutional quality of a country. Since crises are rare events – a typical OECD country suffers a banking crisis once every 37 years on average according to the banking crisis database of Reinhart and Rogoff (2009) – focusing on long-time-series panel data helps to derive statistically meaningful relationships. However, using such historical data comes at the expense of limited data 2

availability. The biggest challenge for this time period is to proxy U.S. monetary policy decisions. In baseline specifications, we proxy U.S. monetary policy stance by using the changes in the U.S. 3-month Treasury rates. For the more recent period starting in 1990, we use monetary policy shocks based on three-month-ahead fed-funds futures rate (Gertler and Karadi, 2015) or six-month Euro-Dollar contracts (Rogers et al., 2014), and the FED Greenbook forecasts of output growth and inflation along with the fed-funds rates to estimate shocks (Romer and Romer, 2004), and we reach similar conclusions. Toidentifytheimplicationsofacountry’sbeingintegratedwiththeUnitedStatesand globally, we interact U.S. monetary policy proxies with measures of integration. In our baseline regressions, we use a country’s bilateral trade intensity with the United States as aproxyfordirecteconomicintegration. Wemeasuretheeconomicintegrationofacountry with the rest of the world by using a trade openness ratio (exports plus imports as a percentageofGDP).Inadditiontothosetwomeasures,wealsousethegravityinstrument oftradeintensityandopennessratioproposedbyFrankelandRomer(1999). Asdiscussed in Frankel and Romer (1999), the effects of trade on income or crises is expected to be endogenous and hence, one can question the causality. Gravity instruments, however, are derived via countries’ geographic characteristics. Such characteristics are expected to be correlated with trade as they have important effects on trade and are plausibly uncorrelated with other determinants of economic and financial stability measures. In our analysis of the more recent time period, in addition to the exposure measures introduced above, we use each country’s debt liabilities in USD (in net of assets) as a percent of GDP (Lane and Shambaugh, 2010; Benetrix et al., 2015), and Chinn-Ito’s capital account openness index. The former is used as another proxy to measure the direct spillovers from the United States, whereas the latter index measures each country’s capital account integration with the rest of the world. What are the channels through which U.S. monetary policy affects global banking crises? U.S. monetary policy may affect other countries through capital flows. Since U.S. monetary policy stance affects relative return on investment in foreign economies, it would also influence credit cycles and in turn financial sector leverage. Avdjiev et al. (2018) shows that capital flows are linked to boom-bust cycles and how such flows could in turn affect the banking sector. A negative monetary policy shock can lead to a credit 3

boom in foreign economies since it is likely that capital would flow out of the U.S. due to an increase in reach for yield incentives. During a credit boom, loan quality decreases (Greenwood and Hanson, 2013), which eventually increases the likelihood of a banking crisis. Schularick and Taylor (2012) and Baron and Xiong (2017) find that excessive lending adversely affects the likelihood of a banking crisis and bank equity crash risk, respectively. Hence, the findings of this literature suggest that a tight U.S. monetary policy could help rein in excesses and reduce the probability of a crisis (e.g., leaning-against-the wind channel). Another line of literature finds that a tightening U.S. monetary policy might increase vulnerabilities, especially in emerging economies, since it might lead to a sudden reversal of capital flows (see Neumeyer and Perri, 2005; Uribe and Yue, 2006, etc.). In our analysis of the drivers of capital flows and how U.S. monetary policy affects these flows, we ask which one of these effects is dominant. We find evidence that the latter effect dominates. In particular, an increase in U.S. monetary policy rates significantly reduces capital flows to foreign economies for those countries with direct exposure to the United States. When the adjustment becomes disorderly, the crisis probability indeed increases. By splitting our sample into emerging market and developed countries, we find that the increase in the probability of a crisis due to disorderly adjustments in capital flows or sudden stops is an emerging market phenomenon. This paper is related to three strands of the literature: The first strand is literature on the spillovers of monetary policy. Rey (2015) builds on the empirical framework in Bekaert et al. (2013) and shows that U.S. monetary policy is a key driver of stock market volatility as measured by the VIX, which, in turn, is an important driver of the global financial cycle. Bruno and Shin (2015) study the relationship between capital flows and monetary policy. Jorda et al. (2018) study the synchronization of global markets and its relationship to a global financial cycle. They find that global financial cycles are closely related to changes in risk premiums, with changes in U.S. monetary policy driving risk appetite and thus serving as a transmission mechanism. In this framework, the variability in exchange rate regimes among countries is one significant explanation of the differential effects of U.S. monetary policy on other countries. We contribute to this literature on global financial cycles by providing empirical evidence in support of the hypothesis that 4

the degree of financial spillovers matters. However, we rather argue that U.S. monetary policyaffectsglobalfinancialstabilityonlytotheextentthatforeigncountrieshavedirect exposure to the United States. Those countries with indirect exposure do not always face an increase in their financial stability risks. The second strand is literature on the determinants of financial crises. Prominent early examples include Demirguc-Kunt and Detragiache (1998) and Kaminsky and Reinhart (1999). Demirguc-Kunt and Detragiache (1998) consider the factors affecting the probability of a banking crisis for 65 countries for the period of 1980 to 1994. By constructingadatasetofbankingandcurrencycrisesspanning120years, Bordoetal.(2001) document that capital controls affect the probability of a crisis. Broner et al. (2013) look at the behavior of capital flows during business cycles and economic crises. Several authors have made use of the Reinhart and Rogoff (2009) database, focusing on banking crises and relevant variables affecting their likelihood. Finally, Danielsson et al. (2018) show that domestic risk appetite–proxied by low financial volatility–is an important predictor of banking crises. We contribute to this literature by providing evidence that U.S. monetary policy decisions play a significant role in foreign banking crises. Moreover, we identify the domestic factors that play a role in determining how U.S. monetary policy affects financial stability risks of foreign countries. The third strand is the literature on the role of integration in probability of crises. In this literature, there are two opposing views on the relationship between a country’s exposure to the world and whether such integration makes the country more or less prone to crises. On one hand, high integration may increase the probability of a crisis through propagation, as the country is more exposed to shocks from abroad, as in Stiglitz (2010). In a multi-country model, Azzimonti et al. (2014) argue that government debt increases with economic integration. Therefore, a policy implication of the model is that integration increases the vulnerability to a crisis. On the other hand, countries that are open to international financial and trade markets could be less vulnerable to shocks, per Ayhan et al. (2006); Cavallo and Frankel (2008). Cavallo and Frankel (2008) point out that a number of channels could reduce vulnerability to a crisis for countries with higher trade integration. First, countries that rely more on trade would be less prone to default, as they are heavily incentivized to maintain trade. Hence, international investors 5

would be less likely to pull out of countries with high trade integration. In addition, trade integration helps countries better absorb shocks. We contribute to this literature by distinguishing the integration with the center country and globally. Therestofthepaperisorganizedasfollows. Section2detailsthedataweuse. Section 3 describes the empirical methodology. Section 4 summarizes our results. Section 5 offers concluding remarks. 2 Data and Descriptive Analysis 2.1 Banking crises data For the analysis, we create an annual panel dataset on 69 countries spanning 1870-2010, as available. The sample includes 24 developed and 45 emerging countries (based on the IMF’s classification). Appendix B lists the countries included in our sample with their coverage. We base our analysis on the systemic banking crises of Reinhart and Rogoff (2009). A crisis is defined as an event with a closure, merger, or public takeover of one or more financial institutions or large scale government assistance of a systemically important financial institution. The unbalanced panel contains a binary indicator of whether a banking crisis starts in a given year and country and includes 239 distinct banking crises. Figure 1 plots the unconditional probability of banking crises for each country in our sample, defined as the number of crises divided by the available sample period. The figure also contains the unconditional probability of banking crises for the United States, for comparison purposes only, as the United States is not included in our sample. Within the developed countries, Italy has the highest annual crisis probability at 6.38%; New Zealand has the lowest, 0.96%. For emerging countries, the annual unconditional crisis probability ranges from 0% for Mauritius to 7.8% for Brazil. Given the differences in the probability of banking crises for emerging and developed countries, it is important to explore how our analysis on the likelihood of banking crises differs in these two groups of countries. 6

ytilibaborP alognA anitnegrA ailartsuA airtsuA muigleB aiviloB lizarB cilbupeR nacirfA lartneC adanaC dnalreztiwS elihC anihC eriovI'd etoC aibmoloC aciR atsoC ynamreG kramneD cilbupeR nacinimoD aireglA rodaucE .peR barA ,tpygE niapS dnalniF ecnarF modgniK detinU anahG eceerG alametauG sarudnoH yragnuH aisenodnI aidnI dnalerI dnalecI ylatI napaJ ayneK aeroK aknaL irS occoroM ocixeM ramnayM suitiruaM aisyalaM airegiN augaraciN sdnalrehteN yawroN dnalaeZ weN amanaP ureP senippilihP dnaloP lagutroP yaugaraP ainamoR aissuR eropagniS rodavlaS lE nedewS dnaliahT aisinuT yekruT nawiaT yaugurU setatS detinU aleuzeneV acirfA htuoS aibmaZ ewbabmiZ 8% Emerging Developed 6% 4% 2% 0% Figure 1: Unconditional annual probability of banking crises The figure presents the probability of banking crises for emerging and developed countries. For a given country, the probability of a banking crisis is calculated as the number of crises divided by the available sample period. 2.2 U.S. Monetary policy decisions We proxy U.S. monetary policy decisions as the change in short term interest rates from the Jorda et al. (2017) macrohistory database. In the more recent period, it is possible to disentanglemonetarypolicysurprisesfromexpectedchanges. Wepursuethisapproachas a robustness to our main findings in 4.3. In particular, we use surprise series constructed by Gertler and Karadi (2015), Romer and Romer (2004) and Rogers et al. (2014). Romer and Romer (2004) narratively identify changes in the federal funds rate targets surrounding FOMC meetings. By regressing these target changes on the current rate and the Greenbook forecasts for output growth and inflation in the following two quarters, they are able to separate the natural policy response of the economy from the exogenous monetary policy surprise. The residuals from this estimation can be used as a proxy for monetary policy shocks in regression analysis. Gertler and Karadi (2015) construct a measure of monetary policy surprise using the change in high-frequency interest-rate futures, limited to a 30-minute period surrounding the publication of a monetary policy 7

decision. They then compute a measure of monetary policy shocks by taking the monthly average of these monetary policy surprises. Rogers et al. (2014) use a similar method to Gertler and Karadi (2015), but applied to the eurodollar contracts, where monetary policy surprises are calculated using the fourth eurodollar futures contract in a more limited time period, defined as 15 minutes prior to an FOMC announcement to 1 hour and 45 minutes after. Table 1 Panel A presents the descriptive statistics for the monetary policy proxies. We find that all four measures of U.S. monetary policy proxies are significantly correlated with each other, with correlation ranging from 0.52 to 0.83. Of these, the Romer and Romer (2004) shocks have the most dispersion, while the Gertler and Karadi (2015) and Rogers et al. (2014) shocks have similar and relatively small standard deviations, in addition to similar means. This is not surprising given the similarity in the way they are constructed. 2.3 Exposure variables Weincludedifferentexposureproxiesthatcanbegroupedintodirect exposure andindirect exposure measures. Direct exposure measures include the ratio of a country’s trade with the United States to its total trade. One can think of the measure as the country’s trade intensity with the United States. Additionally, in the recent sample (post-1990s), we include a country’s debt liabilities in USD (% of GDP) as a proxy of direct exposure with the U.S.(Lane and Shambaugh, 2010; Benetrix et al., 2015). As indirect exposure measure, we calculate country’s total exports and imports as a share of GDP (trade openness) to proxy the economic globalization of a country with the rest of the world. In the recent sample (post-1970s), we also include the capital account openness index of Chinn and Ito (2006). Table 1, Panel B, columns I to IV list the summary statistics of the exposure variables for the whole sample as well as for the developed and emerging economies. Developed countries on average are more globally integrated than emerging countries, irrespective of the way we measure. In contrast, emerging economies, on average, hold higher U.S.- 8

denominated debt and have higher trade intensity with the U.S. compared to developed economies. Both trade intensity and trade openness can affect the macroeconomic outlook and financial stability of a country, suggesting a possible endogeneity problem. To address this issue, we use gravity estimates to construct instrumental variables for trade intensity and openness, following the methodology first introduced by Frankel and Romer (1999). The gravity estimates would serve as a robust instrument as argued by the authors. To this end, we instrument a country’s bilateral trade by means of its distance (to its partners), population, common language, land-border, land-area, landlocked status, and their colonial relationship. Gravity estimates are expected to be good instrumental variables becausetheyarebasedonvariablesthatareplausiblyexogenousandyethighlycorrelated with a country’s overall trade. To estimate gravity instruments for the trade intensity of a country with the U.S., for each year t, we first run the following regressions: log(T /T ) = c+β logdist +β pop +β comlang +β border i,US i 1 i,US 2 US 3 i,US 4 i,US + β areap +β landlocked +β colony +ε (1) 5 i,US 6 i 7 i,US i,US T is the total trade of country i with the United States, T is the total trade with the i,US i whole trade partners. pop is the population of the United States, logdist is the log US i,US of the weighted-distance between the economic centers of the two countries, comlang i,US is a dummy variable that takes value 1 if i and the United States share the same common language, and is 0 otherwise; border is a dummy variable that takes value 1 if the i,US two countries share a border and is 0 otherwise; areap is the log of the product of i,US the areas (in km2) of countries i and U.S., landlock equals to 1 if i is landlocked (i.e., i entirely enclosed by land), and 0 otherwise, and finally colony takes the value 1 if i,US the country has ever had a colonial link with the United States and 0 otherwise. The intensity gravity estimates are the exponential of the fitted values of (1). 9

Similarly, following Cavallo and Frankel (2008), we run the following regression to estimate the gravity of trade openness: log(T /GDP ) = c+β logdistwi,j +β pop +β comlang +β border i,j i 1 2 j 3 i,j 4 i,j + β areap +β landlocked +β colony +ε (2) 5 i,j 6 i,j 7 i,j i,j where, T is the bilateral trade value between countries i and j and GDP is the real i,j i GDP level of country i. landlock equals to 2 if both i and j are landlocked (i.e., i,j entirely enclosed by land), 1 if either i or j are landlocked, and 0 otherwise. The rest of the variables are constructed as above. The gravity estimates (or predicted trade to GDP ratios used in the regressions) are then calculated as the exponential of the fitted values, summing across bilateral trading partners j. For the sake of brevity, we do not present the estimates for the gravity equations (1) and (2). 2.4 Control variables While testing the effects of monetary policy decisions on crises, we include a number of variables known to be predictors of crises as control variables. We first include per-capita gross domestic product growth. Second, inflation affects the likelihood of a financial crisis (see e.g., Demirguc-Kunt and Detragiache, 1998). We calculate inflation as the annual percentage change in the consumer priceindex. Lastly, to control forinstitutionalquality, whichcanaffectpoliticalandmacroeconomicstability(see, e.g., CerraandSaxena, 2008), we use the POLCOMP variable from the Polity IV Project database as a proxy for institutional quality.2Details of the variables constructed can be found in appendix A. Table 1, Panel B, Columns V through X detail selected descriptive statistics for the control variables. Most notably, developed countries have much higher institutional qualityandmuchlowerinflationthantheiremergingcounterparts. Inaddition, thevariability 2Local monetary policy decisions and changes in the exchange rates are also expected to affect the economicandfinancialconditions. However, historicalcoverageforbothseriesformanyofthecountries are poor. When we include changes in the short-term interest rates and exchange rate in our baseline specification, the sample size shrinks by three quarters, hence we do not include these local variables in our baseline regressions and instead present them as part of robustness analysis in section 4.4. 10

ofGDPgrowthandthevariabilityofinflationarealsolowfordevelopedcountriesrelative to emerging ones. 3 Empirical Methodology We hypothesize that U.S. monetary policy affects a country’s financial system only to the extent that the country has direct linkages with the United States. To test this hypothesis, for country i and year t, we estimate the following logit-panel regressions: logit(C ) = β Exposure +β Exposure ×MP +β Exposure ×MP i,t 1 i,t 2 i,t t 3 i,t−1 t−1 + γ ×X +γ ×X +η +ν +ε , (3) 1 i,t 2 i,t−1 i t i,t where logit(C) = log(C/(1−C)) is the log of the odds ratio of the binary banking crisis indicator C .3 Exposure is the measure of a country’s exposure to the U.S. and the i,t world, introduced in Section (2), and MP is the change in the U.S. monetary policy decisions, defined as the change in U.S. 3-month Treasury yields. X are the control variables, namely: inflation rate, GDP growth rate, and political competition. η and ν i t are cross-sectional and time-series fixed effects, respectively. Throughout the analysis, we dually cluster standard errors both at the country and year levels to address possible time-series and cross-country correlation of residuals. 4 Results This section first establishes how U.S. monetary policy affects banking crises and how the nature of linkages affects these results. Subsequently, we provide some evidence on the transmissionmechanisms, andexploretherobustnessofourresultsinvariousdimensions. 3Using an interaction term is also necessary from an econometric standpoint. U.S. monetary policy shocks do not vary by country, so including is as a stand-alone variable in a panel regression would be akin to adding time-series fixed effects. 11

4.1 Effects of U.S. Monetary Policy on Banking Crises Table 2 shows our baseline panel-logit regression results with the historical data. Column I shows the effect of the interaction of monetary policy with direct exposure measures of tradeintensity. U.S.monetarypolicytighteninghasapositiveandstatisticallysignificant effect on the probability of a banking crisis for those countries that have direct trade linkages with the United States. Column II uses the gravity instrument for the trade intensity to correct potential endogenity that occurs when using trade intensity as the dependentvariable. ThecontemporaneousinteractiontermforU.S.monetarypolicywith the exposure variable remains positive and becomes more significant both statistically and economically. The estimated marginal effects (MEs) show that the impact of U.S. monetarypolicyontheprobabilityofacrisisiseconomicallymeaningful: a1%tightening in monetary policy increases the probability of a crisis by 1.0-6.8% for a given level of direct exposure to the United States. As we formally explore below, when U.S. monetary policy tightens, foreign countries could experience capital outflows, leading to an adjustment in external accounts and domestic vulnerabilities. If this correction is sudden, and sizable, it might lead to a sudden stop. Indeed, the countries that have direct exposure to the U.S. appear to be more prone sudden stops, hence for these countries, the probability of a banking crisis increases. Columns III and IV explore the role of U.S. monetary policy for those countries that are open and globally integrated but do not necessarily have direct, primary exposure to the United States. We measure such integration using the trade openness indicator (in column III) and its gravity instrument (in column IV). For those countries without direct exposure to the United States, the role of U.S. monetary policy is ambiguous. In column III, the coefficient for the contemporaneous interaction term is negative and statistically significant, suggesting that the contemporaneous rate changes for these countries might decrease the probability of a banking crisis. Our interpretation of this result is that, for these countries, openness helps with diversification, and these countries might be the immediate beneficiaries of the funds flowing from those other countries which have direct exposure with the United States. In addition, even if these countries are not the direct beneficiaries of capital flowing out of countries that have direct exposure to the 12

United States, a more orderly reversal of capital flows might help correct imbalances that might have accumulated in the run-up period. With an orderly correction of imbalances, the probability of a banking crisis drops, as indicated by the negative marginal effects. However, the impact diminishes when we correct for the endogeneity. Our control variables have expected signs. Higher GDP growth has a negative coefficient, suggesting higher growth reduces the probability of a banking crisis. Higher institutional quality of a country (POLCOMP) lowers the probability of a banking crisis (albeit not significant). It could be that governance is better for countries with better quality scores, where it is more difficult for politicians to distort bank lending decisions. 4.2 Transmission Mechanisms Our results so far show that U.S. monetary policy rates affect the likelihood of banking crises in foreign countries, to the extent that these countries have direct links to the U.S. economy. In this section, we explore possible mechanisms as to why U.S. monetary policy may lead to financial instability abroad. U.S. monetary policy may affect other countries through capital flows, credit growth, and bank leverages. Since U.S. monetary policy stance affects relative return on investment in foreign economies, it may affect credit flows across countries. A loosening stance of U.S. monetary policy can lead to a credit boom in foreign economies since it is likely that capital would flow out of the United States due to an increase in reach-for-yield incentives. With large capital inflows to these countries, the quality of loans becomes poor (Greenwood and Hanson, 2013), which eventually increases the likelihood of a banking crisis. Caballero and Simsek (2018a,b) show that asymmetric capital flows can be destabilizing in these reach-for-yield scenarios. Schularick and Taylor (2012) and Baron and Xiong (2017) find that excessive lending adversely affects the likelihood of a banking crisis and the bank equity crash risk, respectively. Conversely, a tightening of the stance of U.S. monetary policy can lead to reversal of capital flows. If the correction in capital flows becomes sudden and disorderly, it can lead to an increase in the probability of a banking crisis (see Neumeyer and Perri, 2005; Uribe and Yue, 2006, among others). 13

To examine if U.S. monetary policy tightening could cause a reversal of capital flows in foreign economies, we run the following regression ∆CF = β Exposure +β Exposure ×MP +β Exposure ×MP i,t 1 i,t 2 i,t t 3 i,t−1 t−1 + γ ∆CF +γ ×X +γ ×X +η +ν +ε , (4) 1 i,t−1 2 i,t 3 i,t−1 i t i,t where, ∆CF isthechangeintotalportfolioinvestmentflows(%ofGDP)forcountryiin i,t yeart. Weincludeallofthecontrolvariablesintroducedin(3). Inaddition, wecontrolfor thechangeindomesticinterestratestoaccountforthelocalmonetarypolicydecisions. In addition, as the sample coverage starts in the 1970’s, we can use two additional exposure measures, which are not available at our original historical sample period. The first one is the difference between the debt liabilities and debt assets denominated in U.S. dollar, and the second one is the capital account openness. We consider countries with dollar-denominated liabilities as having direct exposures to U.S. monetary policy, since changes in U.S. monetary policy directly affect the debt servicing costs. As thoroughly discussed in the literature (See Calvo, 2002; Choi and Cook, 2004; Mendoza, 2002, among others), liability dollarization is a significant source of financial stability risk. When dollar denominated liabilities are financed by income derived in local currency, any changes in exchange rate fluctuations could make the debt servicing cost higher. When U.S. monetary policy tightens, the cost of holding dollardenominated debt for foreign economies rises through two channels. First, the rate at which the borrowers roll over their debt would be higher. Second, a higher U.S. monetary policy rate would drive up the value of the dollar relative to other currencies. We consider countrieswithmoreopencapitalaccountsasbeinggloballyintegratedbutnotnecessarily having direct exposure with the United States. Therefore, we consider capital account openness as an indirect exposure measure. Table 3 shows that the interaction term is negative and significant for the direct exposure measures but insignificant for indirect exposure measures. That is, a positive shock to, or a tightening stance in, the U.S. monetary policy is followed by reduction in capital flows to these countries only if the country has direct economic exposure to the U.S.However, ifthecountryisgloballyintegrated, thentheeffectofU.S.monetarypolicy is ambiguous. This finding suggests that when U.S. monetary policy tightens, countries 14

with direct exposure to the United States will face capital outflows. If the outflows are disorderly (e.g., sudden stops), the probability of a banking crisis increases. 4.3 Monetary policy surprises and crises The literature has offered various proxies for monetary policy shocks. To explore the effects of policy surprises, rather than changes in the decisions, in Table 4, we use the monetary policy surprise series constructed by Gertler and Karadi (2015), Romer and Romer (2004) and Rogers et al. (2014). These indicators of monetary policy surprises, however, are available only for the more recent period, hence we have to restrict our sample in these regressions to the 1990–2010 period. In these regressions, we can use two additional exposure variables: dollar denominated liabilities, as a proxy for direct exposure, and Chinn-Ito’s capital account openness indicator, as a proxy for indirect exposure. Table 4 shows that our main findings in the historical sample hold. U.S. monetary policy shocks increase the probability of a banking crisis for countries with direct trade links with the United States. (columns I to VI) or countries that hold more dollardenominated liabilities (columns VII to IX). For countries integrated globally, results are again ambiguous but statistically stronger compared to using monetary policy stances. U.S. monetary policy shocks decrease the probability of a banking crisis for countries that have strong trade links globally, even controlled for the endogeneity (columns X to XV). Whereas, we failed to document a robust relationship between the policy surprises and banking crises probability using the Chinn-Ito index as the exposure variable (columns XIX to XXI). To sum up, our results in this section reinforces our earlier finding that U.S. monetary policy shocks lead to a contemporaneous increase in probability of a banking crisis only for those countries with a direct exposure to the U.S. For other countries with indirect exposure, the effect of monetary policy shocks is ambiguous. The results in these cases point to a reduction in probability of a banking crisis with some exposure measures and point to an ambiguous effect with some other exposure measures. 15

4.4 Robustness We examine the robustness of our findings in two main dimensions. First, we look at subsamples. Second, we examine alternative econometric specifications and alternative data. In the interest of space, we exclusively present robustness results for our regression results for the gravity instrument for trade intensity, e.g., our main direct exposure variable shown in column II of Table 2, and leave robustness for all the other columns to an online appendix. Table 5 shows our results with different sub-samples. In particular, we look at post- WWII period (column II), a sample that excludes the Great Depression and the Great Recession, and both WWI and WWII periods (column III), a sample with emerging markets only (column IV), a sample with developed countries only (column V), a sample controlling for countries that anchor their exchange rates to the U.S. dollar (column VI). In all these sub-samples, the interaction variable for U.S. monetary policy and exposure variable remains positive and statistically significant, with the exception of the sample for developed countries. This finding suggests that the effect of U.S. monetary policy on the probability of a banking crisis (due to increased risk of a sudden stop) is mainly an emerging market phenomena. It is also worth highlighting the results with exchange rate anchors, countries who directly anchor their exchange rate to the U.S. dollar. In this exercise, we add the contemporaneous and lagged interaction of U.S. monetary policy with a dummy for exchange rate anchor countries, which takes a value of 1 if the country anchors its exchange rate to the U.S. dollar and 0 otherwise. We find that U.S. monetary policy has a positive effect on the probability of a banking crisis for those countries with direct exposure to the United States, regardless of their anchor policy. However, the impact is economically more meaningful for the countries that anchor their exchange rate to the dollar. In Table 6, we present additional robustness analyses with alternative econometric specifications and alternative data. In columns II and III, we examine the robustness of our findings with the use of simple OLS and probit regressions, respectively. In column IV, we investigate whether our results are sensitive to the inclusion of local monetary policy changes and changes in the exchange rates. Local monetary policy decisions, not 16

only the U.S. monetary policy ones, are expected to affect the economic conditions. We proxy local monetary policy changes as the changes in the short-term local interest rates. In column V, we test the sensitivity of our findings by considering alternative crisis databases. Our motivation in doing this to see if our results are sensitive to the critiques raised in the literature (see, e.g., Romer and Romer, 2015) regarding Reinhart and Rogoff (2009). Inparticular, followingDanielssonetal.(2018)wemergedthedatabasesofBordo et al. (2001); Laeven and Valencia (2012); Gourinchas and Obstfeld (2012); Schularick andTaylor(2012)withthatofReinhartandRogoff(2009)forbankingbyusingconsistent definitions of crises and then use it as the dependent variable. Finally, we re-estimate the baseline equation with non-winsorized variables (column VI). Overall, we find that the results are qualitatively similar under the various robustness checks. There are small changes in different specifications, but the main conclusions hold. 5 Conclusion In this paper we examine the role of U.S. monetary policy in global financial stability. We find that positive U.S. monetary policy shocks leads to an increase in the probability of a banking crisis for those countries with direct linkages to the U.S., either in the form of trade links or significant share of USD-denominated liabilities. However, if a country is integrated globally, rather than having a direct exposure to the U.S., the effects of U.S. monetary policy shocks are ambiguous. 17

References Adrian, T. and N. Liang (2018). Monetary policy, financial conditions, and financial stability. International Journal of Central Banking 14, 73–131. Avdjiev, S., B. Hardy, S. Kalemli-Ozcan, and L. Serven (2018). Gross capital flows by banks, corporates and sovereigns. Mimeo, University of Maryland. Ayhan, K. M., P. E. S., and M. E. Terrones (2006). How do trade and financial integration affect the relationship between growth and volatility. Journal of International Economics 69(1), 176–202. Azzimonti, M., E. de Francisco, and V. Quadrini (2014). Financial globalization, inequality, and the rising public debt. American Economic Review 104, 2267–2302. Baron, M. and W. Xiong (2017). Credit expansion and neglected crash risk. Quarterly Journal of Economics 132, 713–764. Bekaert, G., M. Hoerova, and M. L. Duca (2013). Risk, uncertainty and monetary policy. Journal of Monetary Economics 60(7), 771 – 788. Benetrix, A. S., J. C. Shambaugh, and P. R. Lane (2015). Financial exchange rates and international currency exposures. American Economic Review 100, 518–540. Bernanke, B. S. and K. N. Kuttner (2005). What explains the stock market’s reaction to federal reserve policy? Journal of Finance 60, 1221–1257. Bordo, M., B. Eichengreen, D. Klingebiel, S. M. Martinez-Peria, and A. K. Rose (2001). Is the crisis problem growing more severe? Economic Policy 24, 51–82. Broner, F., T.Didier, A.Erce, andS.L.Schmukler(2013). Grosscapitalflows: Dynamics and crises. Journal of Monetary Economics 60(1), 113–133. Bruno, V. and H. S. Shin (2015). Cross-Border Banking and Global Liquidity. Review of Economic Studies 82(2), 535–564. Caballero, R. J. and A. Simsek (2018a, December). A model of fickle capital flows and retrenchment. Working Paper 22751, National Bureau of Economic Research. 18

Caballero, R. J. and A. Simsek (2018b). Reach for yield and fickle capital flows. AEA Papers and Proceedings 108, 493–498. Calvo, G. (2002). On dollarization. Economics of Transition 10, 393–403. Cavallo, E. and J. Frankel (2008). Does openness to trade make countries less vulnerable to sudden stops? using gravity to establish causality. Journal of International Money and Finance 27, 1530–1452. Chinn,M.D.andH.Ito(2006). Whatmattersforfinancialdevelopment? capitalcontrols, institutions, and interactions. Journal of Development Economics 81, 163–192. Choi, W. and D. Cook (2004). Liability dollarization and the bank balance sheet channel. Journal of International Economics 64, 247–275. Danielsson, J., M. Valenzuela, and I. Zer (2018). Learning from history: Volatility and financial crises. Review of Financial Studies forthcoming. Demirguc-Kunt, A. and E. Detragiache (1998, March). The determinants of banking crises in developing and developed countries. IMF Staff Paper. Frankel, J. A. and D. H. Romer (1999, June). Does trade cause growth? American Economic Review 89(3), 379–399. Gertler, M. and P. Karadi (2015). Monetary policy surprises, credit costs, and economic activity. American Economic Journal: Macroeconomics 7, 44–76. Gilchrist, S., D. Lopez-Salido, and E. Zakrajsek (2015). Monetary policy and real borrowing costs at the zero lower bound. American Economic Journal: Macroeconomics 7, 77—-109. Gourinchas, P.-O. (2017). Monetary policy transmission in emerging markets: An application to chile. Gourinchas, P.-O. and M. Obstfeld (2012). Stories of the twentieth century for the twenty-first. American Economic Journal: Macroeconomics 8, S85–S118. Greenwood, R. and S. G. Hanson (2013). Issuer quality and corporate bond returns. Review of Financial Studies 26, 1483–1525. 19

Ilzetzki, E., C. M. Reinhart, and K. S. Rogoff (2017). Exchange arrangements entering the 21st century: Which anchor will hold? (23134). Jorda, O., M. Schularick, and A. M. Taylor (2017). Macrofinancial history and the new business cycle facts. NBER Macroeconomics Annual 2016 31. Jorda, O., M. Schularick, A. M. Taylor, and F. Ward (2018, June). Global financial cycles and risk premiums. (24677). Kaminsky, G. and C. Reinhart (1999). The twin crises: The causes of banking and balance-of-payments problems. American Economic Review 89, 473–500. Laeven, L. and F. Valencia (2012). Systemic banking crises database : An update. IMF Working Paper No. 12/163. Lane, P. R. and J. C. Shambaugh (2010). Financial exchange rates and international currency exposures. American Economic Review 100, 518–540. Mayer, T. and S. Zignago (2011). Notes on cepii’s distances measures: The geodist database. Technical Report 2011-25, CEPII. Mendoza, E.G.(2002). Credit, prices, andcrashes: Businesscycleswithasuddenstop. In Preventing Currency Crises in Emerging Markets, pp. 103–148. University of Chicago Press. Neumeyer, P. A. and F. Perri (2005). Business cycles in emerging economies: the role of interest rates. Journal of Monetary Economics 52, 345–380. Reinhart, C. M. and K. S. Rogoff (2009). This Time is Different: Eight Centuries of Financial Folly. Princeton University Press. Rey, H. (2015, May). Dilemma not trilemma: The global financial cycle and monetary policy independence. Working Paper 21162, National Bureau of Economic Research. Rogers, J. H., C. Scotti, and J. H. Wright (2014). Evaluating asset-market effects of unconventional monetary policy: a multi-country review. Economic Policy 29, 749– 799. 20

Romer, C. D. and D. H. Romer (2004). A new measure of monetary shocks: Derivation and implications. American Economic Review 94, 1055–1084. Romer, C. D. and D. H. Romer (2015). New evidence on the impact of financial crises in advanced countries. (21021). Schularick, M. and A. Taylor (2012). Credit booms gone bust: Monetary policy, leverage cycles, and financial crises, 1870–2008. American Economic Review 102, 1029–1061. Stiglitz, J. E. (2010). Risk and global economic architecture: Why full financial integration may be undesirable. American Economic Review 100(2), 388–392. Uribe, M. and V. Yue (2006). Country spreads and emerging countries: Who drives whom? Journal of International Economics 69(1), 6–36. 21

6 Appendix A: Definition of variables 6.1 Monetary policy shocks • MP: US monetary policy change, defined as the change in US short-term interest rates from the Jorda et al. (2017) macrohistory database. • GK: US monetary policy shocks introduced in Gertler and Karadi (2015), and defined as the surprises in the three months ahead federal funds rate futures. • RSW: US monetary policy shocks introduced in Rogers et al. (2014), and constructed through the surprises on the six-month Euro-Dollar contracts. • RR: US monetary policy shocks introduced in Romer and Romer (2004). The authors use the FED Greenbook forecasts of output growth and inflation along with the fed-funds rates to estimate shocks. 6.2 Exposure variables • UStradeIntensity: Trade intensity to U.S., calculated as total trade to US divided by total trades of the country. Data is from COW trade project. • Gravity–UStradeIntensity: The instrument of trade intensity, introduced in (1). • Debt_in_USD: Debt liabilities minus debt assets in USD (% of GDP) in log terms, constructed by using data from the IMF’s Coordinated Portfolio Investment Survey (CPIS) and the BIS locational banking statistics as detailed by Lane and Shambaugh (2010); Benetrix et al. (2015). • EconInteg: Economic integration, calculated as a country’s total exports and imports as a % of GDP (trade openness). Trade data is from COW trade project and GDP data is from Maddison project. • Gravity–EconInteg: The instrument of trade openness, introduced in (2). • KAOPEN: The Chinn-Ito financial openness index. It measures a country’s degree ofcapitalaccountopenness, introducedinChinnandIto(2006). KAOPENisbased 22

on the binary dummy variables that codify the tabulation of restrictions on crossborder financial transactions reported in the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER). 6.3 Capital flows • ∆CF: The change in total portfolio flows as a percentage of the local country’s GDP, taken from the IMF Balance of Payments statistics (BPM5). 6.4 Control variables • GDPgrowth: Real GDP per capita growth rate. Data from the Maddison project. • INF: Inflation rate calculated as the annual percentage change of the CPI index. Data from the Global Financial Data. • POLCOMP: Political competition as a proxy for institutional quality. Data is from the Polity IV Project database. POLCOMP is the combination of the degree of institutionalization or regulation of political competition and the extent of government restriction on political competition. The higher the value of the POLCOMP, the better the institution quality of a given country. • ∆ INT_RATES: Change in local 3-month Treasury yields. Used as a proxy for the local monetary policy surprises. Data from the Global Financial Data. • ∆XR: The change in the exchange rate of the local currency to the dollar, from Global Financial Data. • ANCHOR: A dummy variable equal to 1 if a country’s currency is pegged to the U.S. Dollar in year t and 0 otherwise. Data from Ilzetzki et al. (2017). • Gravity variables – areap is the log of the product of the areas in km2 of two countries. Data is from the GeoDist database–CEPII (Mayer and Zignago, 2011) – T is the bilateral trade value between countries i and j. Data is from the i,j COW project 23

– pop is the population of a country. Data from the Maddison project – distw is the bilateral distances between the biggest cities of two countries, those inter-city distances being weighted by the share of the city in the overall country’s population (see Mayer and Zignago, 2011, for details). – areap is the log of the product of the areas (in squared kilometers) of countries i and U.S. – comlang is equals to 1 if the countries share the same official language and 0 otherwise – border if equals to 1 if the countries share a border and 0 otherwise – landlocked equals to 1 if the local country is landlocked (i.e., entirely enclosed by land) and 0 otherwise – colony equals 1 if the countries have ever had a colonial link with the U.S 24

7 Appendix B: Sample details Table B1: This table lists the countries in our sample and sample coverage, divided into panels by IMF Classification. Panel A: Developed Countries Country Coverage Country Coverage Country Coverage Australia 1901-2010 France 1870-2010 Norway 1870-2010 Austria 1870-2010 Greece 1870-2010 New Zealand 1907-2010 Belgium 1870-2010 Ireland 1922-2010 Portugal 1870-2010 Canada 1870-2010 Iceland 1918-2010 Singapore 1965-2010 Switzerland 1870-2010 Italy 1870-2010 Spain 1870-2010 Germany 1870-2010 Japan 1870-2010 Sweden 1870-2010 Denmark 1870-2010 Korea 1945-2010 Taiwan 1945-2010 Finland 1917-2010 Netherlands 1870-2010 United Kingdom 1870-2010 Panel B: Emerging Countries Country Coverage Country Coverage Country Coverage Algeria 1962-2010 Guatemala 1870-2010 Philippines 1946-2010 Angola 1975-2010 Honduras 1870-2010 Poland 1918-2010 Argentina 1870-2010 Hungary 1918-2010 Paraguay 1870-2010 Bolivia 1870-2010 Indonesia 1949-2010 Romania 1878-2010 Brazil 1870-2010 India 1947-2010 Russia 1870-2010 Central African Republic 1960-2010 Kenya 1963-2010 El Salvador 1870-2010 Chile 1870-2010 Morocco 1956-2010 South Africa 1910-2010 China 1870-2010 Mexico 1870-2010 Sri Lanka 1948-2010 Cote d’Ivoire 1960-2010 Myanmar 1948-2010 Thailand 1870-2010 Colombia 1870-2010 Mauritius 1968-2010 Tunisia 1956-2010 Costa Rica 1870-2010 Malaysia 1963-2010 Turkey 1870-2010 Dominican Republic 1870-2010 Nigeria 1960-2010 Uruguay 1870-2010 Ecuador 1870-2010 Nicaragua 1870-2010 Venezuela 1870-2010 Egypt, Arab Rep. 1870-2010 Panama 1903-2010 Zambia 1966-2010 Ghana 1957-2010 Peru 1870-2010 Zimbabwe 1965-2010 25

scitsitatS evitpircseD detceleS :1 elbaT A lenaP VI III II I RR WSR KG PM skcohS PM 030.0- 090.0- 851.0- 000.0 egarevA 201.1 532.0 982.0 310.0 .veD .dtS 0372 0712 0161 0048 .sbO B lenaP X XI IIIV IIV IV V VI III II I RX∆ setaRtnI∆ FNI PMOCLOP htworgPDG swolFpaC nepOAK DSU_ni_tbeD getnInocE ytisnetnIedartSU 481.1 582.0- 860.8 281.6 020.0 680.0 042.0 176.2 258.1 791.0 egarevA elohW 465.7 943.4 018.41 933.3 150.0 171.1 395.1 379.0 550.3 291.0 .veD .dtS 5238 6202 7706 2247 3607 808 1282 1011 0975 4316 .sbO 744.0 741.0- 084.4 938.7 220.0 690.0 152.1 952.2 594.2 721.0 egarevA depoleveD 878.4 687.1 778.8 171.3 840.0 899.0 633.1 349.0 010.4 721.0 .veD .dtS 3923 9111 5192 7872 1503 065 3001 433 8242 2442 .sbO 766.1 654.0- 573.11 781.5 910.0 360.0 713.0- 158.2 783.1 442.0 egarevA gnigremE 168.8 781.6 060.81 820.3 450.0 194.1 244.1 139.0 689.1 212.0 .veD .dtS 2305 709 2613 5364 2104 842 8181 767 2633 2963 .sbO egnahc ycilop yratenom sthgilhgih A lenaP .snoisserger eht ni esu ew atad eht rof snoitavresbo fo rebmun dna ,noitaived dradnats ,egareva eht swohs elbat sihT fo erusaem eht si KG ,setar sdnuf laredef htnom-3 ni egnahc eht si PM .yrtnuoc yb yrav ton od dna ycilop yratenom .S.U tneserper hcihw ,serusaem esirprus ro rallodorue htruof eht ni esirprus ycilop yratenom fo erusaem eht si WSR ,)5102( idaraK dna reltreG ni sa setar sdnuf laredef htnom-eerht ni esirprus yratenom selbairav rof scitsitats evitpircsed sevig B lenaP .)5102( remoR dna remoR ni sa esirprus yratenom fo erusaem eht si RR dna ,)4102( .la te sregoR ni sa tcartnoc .A xidneppA ni nevig era snoitinfied elbairaV .yrtnuoc yb yrav od hcihw ,slortnoc dna erusopxe fo serusaem sa desu 26

Table 2: Role of US Monetary Policy in Financial Crises: Historical Sample Y :C I II III IV i,t Exp: UStradeIntensity Gravity–UStradeIntensity EconInteg Gravity Exp 0.33 -3.45** -0.26 -4.91* i,t (0.329) (1.351) (0.259) (2.816) (Exp*MP) 23.68** 162.88*** -29.61*** -96.94 i,t (11.930) (55.724) (8.766) (93.510) (Exp*MP) -10.66 -4.12 15.14 39.64 i,t−1 (11.747) (56.419) (11.181) (63.851) GDPgrowth -10.84*** -9.63*** -9.86*** -9.48*** i,t (2.682) (2.705) (2.659) (2.832) POLCOMP -0.07 -0.00 -0.05 0.01 i,t (0.075) (0.091) (0.073) (0.092) INF -0.01 -0.00 -0.00 -0.00 i,t (0.010) (0.009) (0.009) (0.009) GDPgrowth 0.62 0.08 0.26 0.15 i,t−1 (2.306) (2.141) (2.023) (2.238) POLCOMP -0.00 -0.06 -0.01 -0.06 i,t−1 (0.065) (0.078) (0.067) (0.079) INF 0.01 0.01 0.01 0.00 i,t−1 (0.010) (0.010) (0.010) (0.009) Obs. 1,496 1,541 1,555 1,541 Pseudo R2 0.168 0.203 0.185 0.191 MFX (Exp*MP) 0.998 6.825 -1.216 -4.692 i,t (Exp*MP) -0.450 -0.173 0.539 2.223 i,t−1 Thetableshowstheestimatedcoefficientsofthepanel-logitregressionsintroducedin(3). Thedependent variable is a dummy variable that equals to 1 at the beginning year of a systemic banking crises, defined in Reinhart and Rogoff (2009). MP is the U.S. monetary policy decisions, defined as the change in US 3-month Treasury yields. The exposure variable used (Exp) is listed at the column header. UStradeIntensity is a country’s total trade to U.S. divided by its total trades. Gravity–UStradeIntensity is the instrument of trade intensity, introduced in (1). EconInteg is economic integration proxied by the trade openness(exports+importsasaratioofGDP),Gravityistheinstrumentedtrademeasureasintroduced in (2). GDPgrowth is the GDP growth rate, POLCOMP is the degree of political competition, and INF is the annual inflation rate. All of the specifications include country and year fixed effects, where the estimated coefficients are omitted for the sake of brevity. The panel covers 69 countries and spans 1870–2010. The standard errors, reported in parentheses, are robust and dually clustered at the year and country level. Estimated marginal effects of the interaction term and lagged interaction term are reported in the last two rows. 27

seirtnuoC ngieroF ot swolF latipaC ni yciloP yratenoM .S.U fo eloR ehT :3 elbaT IV V VI III II I FC∆: Y t,i nepOAK ytivarG getnInocE DSU_ni_tbeD ytisnetnIedartSU–ytivarG ytisnetnIedartSU :pxE 60.0 52.0- 01.0- 51.0- 21.0 15.0pxE t,i )701.0( )509.0( )761.0( )241.0( )442.1( )664.0( 61.3- 35.8- 21.0- *86.9- ***69.82- **26.5- )PM*pxE( t,i )318.2( )822.22( )801.3( )966.4( )974.5( )691.2( 92.4 49.62- 74.1 95.1- 15.61- 95.0 )PM*pxE( 1−t,i )763.3( )852.22( )836.1( )269.4( )381.71( )074.1( ***34.0- ***44.0- ***44.0- ***05.0- ***64.0- ***64.0- FC∆ 1−t,i )270.0( )160.0( )170.0( )080.0( )760.0( )570.0( 36.3 56.2 03.3 *15.6 69.2 68.3 htworgPDG t,i )095.2( )785.2( )399.2( )396.3( )408.2( )609.2( **81.0- **71.0- **71.0- 20.0- **81.0- *81.0- PMOCLOP t,i )270.0( )770.0( )770.0( )491.0( )970.0( )590.0( 20.0 20.0 20.0 *40.0- 20.0 20.0 FNI t,i )810.0( )020.0( )810.0( )220.0( )810.0( )220.0( 20.0- 20.0- 30.0- 00.0- 10.0- 20.0setaRtnI∆ t,I )920.0( )030.0( )820.0( )720.0( )920.0( )230.0( ***11.31 ***61.21 ***59.31 ***46.41 ***84.21 ***14.31 htworgPDG 1−t,i )650.3( )968.2( )121.3( )610.3( )706.3( )560.3( **81.0 ***32.0 ***02.0 90.0- **22.0 ***02.0 PMOCLOP 1−t,i )670.0( )570.0( )760.0( )752.0( )580.0( )760.0( 10.0- 10.0- 10.0- **50.0 10.0- 10.0- FNI 1−t,i )910.0( )220.0( )910.0( )420.0( )020.0( )820.0( 10.0- 10.0- 10.0- 10.0- 00.0- 10.0setaRtnI∆ 1−t,i )510.0( )410.0( )310.0( )020.0( )410.0( )710.0( 364 464 954 922 464 644 .sbO 253.0 553.0 053.0 444.0 563.0 463.0 2R egatnecrepasaswofloiloftroplatotniegnahcehtsielbairavtnednepedehT .)4(nidecudortnisnoissergertigol-lenapehtfostneicffieocdetamitseswohselbatehT dna ,redaeh nmuloc eht ta detsil si )pxE( desu elbairav erusopxe ehT .)5MPB( scitsitats stnemyaP fo ecnalaB FMI eht morf nekat ,PDG s’yrtnuoc lacol eht fo mret trohs s’yrtnuoc a ni egnahc eht si hcihw ,setaRtnI∆ .elbaliava sa ,0102 ot 1591 morf snaps dna seirtnuoc 96 sedulcni elpmaS .2 elbaT ni decudortni sa era .stceffe dexfi raey dna yrtnuoc edulcni snoitacfiiceps llA .)2( elbaT ni decudortni era selbairav lortnoc rehto eht fo llA .ataD laicnaniF labolG morf setar tseretni .level yrtnuoc dna raey eht ta deretsulc yllaud dna tsubor era ,sesehtnerap ni detroper ,srorre dradnats ehT 28

sesirprus PM :sesirC laicnaniF ni yciloP yratenoM SU fo eloR :4 elbaT XI IIIV IIV IV V VI III II I C: Y t,i DSU_ni_tbeD ytisnetnIedartSU–ytivarG ytisnetnIedartSU :pxE RR WSR KG RR WSR KG RR WSR KG :PM *74.1 **76.1 *66.1 **51.2- 71.0- 72.0 21.3 **69.31 **39.31 pxE t,i )518.0( )828.0( )088.0( )310.1( )042.1( )042.1( )975.3( )165.6( )295.5( 00.0- 82.1 **37.0 ***00.1 **66.2 05.1 **40.1 ***68.41 ***15.8 )PM*pxE( t,i )052.0( )448.0( )223.0( )683.0( )321.1( )353.1( )774.0( )768.4( )042.2( *47.2 83.1- 13.1- 73.0- 97.0 30.0 47.0- ***41.6- ***11.6- )PM*pxE( 1−t,i )925.1( )901.2( )323.1( )605.0( )260.1( )701.1( )126.0( )942.2( )747.1( 26.21- 63.31- 02.31- ***33.22- ***48.91- ***33.02- *92.21- *89.51- 53.31htworgPDG t,i )114.51( )608.9( )748.9( )452.6( )961.3( )979.2( )448.6( )852.8( )537.8( **07.0- 94.0- 94.0- **55.0- ***26.0- ***76.0- ***07.0- ***61.1- ***52.1- PMOCLOP t,i )833.0( )473.0( )663.0( )032.0( )361.0( )721.0( )932.0( )843.0( )742.0( 40.0- 30.0- 30.0- 10.0- 20.0- 20.0- 20.0- 30.0- 30.0- FNI t,i )460.0( )740.0( )840.0( )840.0( )740.0( )940.0( )840.0( )940.0( )550.0( **02.0 *11.0 **11.0 40.0 30.0 30.0 50.0 30.0 40.0 setaRtnI∆ t,I )980.0( )850.0( )350.0( )801.0( )701.0( )501.0( )490.0( )690.0( )490.0( 70.33- *07.03- *49.92- 96.4- 67.6- 23.6- 23.8- 80.7- 78.5htworgPDG 1−t,i )795.12( )917.71( )545.71( )280.01( )050.01( )878.9( )862.9( )316.31( )889.21( 65.0 64.0 74.0 ***67.0 ***67.0 ***27.0 42.0 14.0 34.0 PMOCLOP 1−t,i )283.0( )334.0( )364.0( )682.0( )691.0( )422.0( )112.0( )733.0( )672.0( 30.0 30.0 30.0 20.0 20.0 20.0 20.0 20.0 30.0 FNI 1−t,i )190.0( )240.0( )240.0( )050.0( )740.0( )840.0( )050.0( )950.0( )060.0( 01.0- **11.0- **11.0- *60.0- *60.0- *60.0- 40.0- 70.0- *80.0setaRtnI∆ 1−t,I )221.0( )940.0( )840.0( )430.0( )230.0( )230.0( )630.0( )550.0( )740.0( 641 641 641 022 022 022 532 532 532 snoitavresbO 134.0 904.0 904.0 703.0 392.0 482.0 962.0 893.0 183.0 2R oduesP XFM 100.0- 640.0 530.0 950.0 211.0 401.0 760.0 481.0 102.0 )PM*pxE( t,i 060.0 260.0- 360.0- 220.0- 440.0 200.0 840.0- 460.0- 541.0- )PM*pxE( 1−t,i elbairav ymmud a si elbairav tnedneped ehT .0991 retfa elpmas eht gnisu ,)3( ni decudortni snoisserger tigol-lenap eht fo stneicffieoc detamitse swohs elbat ehT eht ta detsil si )pxE( desu elbairav erusopxe ehT .)9002( ffogoR dna trahnieR ni denfied ,sesirc gniknab cimetsys a fo raey gninnigeb eht ta 1 ot slauqe taht .la te sregoR ;)5102( idaraK dna reltreG yb senfied skcohs eht era RR dna ,WSR ,KG .skcohs ycilop yratenom eht fo noitinfied eht sa llew sa redaeh nmuloc dna enaL ni decudortni ,)PDG fo %( DSU ni stessa tbed sunim seitilibail tbed ’seirtnuoc si DSU_ni_tbeD .ylevitcepser ,)4002( remoR dna remoR ;)4102( fo llA .seirtnuoc 96 sedulcni elpmaS .2 elbaT ni decudortni era selbairav eht fo tser ehT .xedni ssennepo xednI otI-nnihC eht si NEPOAK .)0102( hguabmahS yrtnuoc dna raey eht ta deretsulc yllaud dna tsubor era ,sesehtnerap ni detroper ,srorre dradnats ehT .stceffe dexfi raey dna yrtnuoc edulcni snoitacfiiceps eht .swor owt tsal eht ni detroper era mret noitcaretni deggal eht dna mret noitcaretni eht fo stceffe lanigram detamitsE .level 29

).tnoC( sesirprus PM :sesirC laicnaniF no yciloP yratenoM SU fo eloR :4 elbaT IXX XX XIX VX VIX IIIX IIX IX X C: Y t,i nepOAK ytivarG getnI nocE :pxE RR WSR KG RR WSR KG RR WSR KG :PM 10.0 45.0- 90.0 *05.21- ***88.61- ***75.91- 94.0 **73.1 **12.2 pxE t,i )238.0( )078.0( )868.0( )505.6( )963.6( )748.6( )487.1( )156.0( )078.0( 17.0- *33.2- 42.0- 34.2- **22.33- ***75.12- *59.0- *62.3- *42.1- )PM*pxE( t,i )635.0( )652.1( )667.0( )934.4( )377.41( )481.8( )015.0( )147.1( )756.0( 02.0 14.1- 51.0 87.1 **88.22 ***55.91 45.0 81.2 09.1 )PM*pxE( 1−t,i )405.0( )775.1( )058.0( )276.2( )419.01( )186.6( )945.0( )275.3( )010.2( ***68.11- ***79.31- ***62.31- ***80.61- ***59.31- ***96.21- **86.01- ***48.11- ***28.31htworgPDG t,i )453.3( )698.3( )514.3( )951.5( )788.3( )281.3( )288.4( )630.4( )888.3( **66.0- **56.0- ***97.0- **25.0- **66.0- ***76.0- *77.0- ***19.0- ***39.0- PMOCLOP t,i )092.0( )862.0( )203.0( )722.0( )362.0( )062.0( )624.0( )443.0( )243.0( 30.0- 20.0- 20.0- 20.0- 10.0- 10.0- 30.0- 30.0- 20.0- FNI t,i )640.0( )540.0( )840.0( )340.0( )040.0( )040.0( )940.0( )740.0( )940.0( 80.0 70.0 50.0 70.0 70.0 70.0 70.0 90.0 80.0 setaRtnI∆ t,I )280.0( )180.0( )390.0( )790.0( )301.0( )001.0( )490.0( )101.0( )601.0( 18.9- 61.9- 89.7- 11.9- 98.8- 49.8- 29.21- 53.61- 44.61htworgPDG 1−t,i )412.9( )928.8( )488.8( )926.01( )929.01( )554.01( )385.9( )907.11( )377.01( 93.0 92.0 13.0 **07.0 **07.0 **96.0 14.0 24.0 63.0 PMOCLOP 1−t,i )062.0( )332.0( )902.0( )582.0( )143.0( )923.0( )063.0( )262.0( )132.0( 00.0 00.0- 10.0 20.0 10.0 10.0 10.0 10.0 10.0 FNI 1−t,i )440.0( )340.0( )050.0( )540.0( )840.0( )940.0( )540.0( )240.0( )440.0( *40.0- 20.0- 30.0- 50.0- 50.0- 50.0- 20.0- 30.0- 20.0setaRtnI∆ 1−t,I )220.0( )510.0( )120.0( )330.0( )840.0( )540.0( )230.0( )820.0( )720.0( 232 232 232 022 022 022 532 532 532 snoitavresbO 542.0 642.0 812.0 092.0 353.0 543.0 982.0 782.0 962.0 2R oduesP XFM 250.0- 190.0- 310.0- 051.0- 734.1- 451.1- 050.0- 741.0- 780.0- )PM*pxE( t,i 020.0 400.0- 310.0 010.0 579.0 510.1 310.0 971.0 301.0 )PM*pxE( 1−t,i .egap suoiverp eht morf deunitnoc si elbat sihT 30

sisylanA elpmasbuS :ssentsuboR :5 elbaT IV V VI III II I C: Y t,i ycnerruC rohcnA rof lortnoC seimonocE depoleveD seimonocE gnigremE sesirC rojaM tuohtiW 0102-6491 elpmaS elohW ytisnetnIedartSU–ytivarG :pxE 35.1- ***41.4- **51.5- **38.3- 25.0- **54.3pxE t,i )218.2( )503.1( )953.2( )194.1( )907.0( )153.1( **24.461 20.501 ***87.342 **46.251 **88.83 ***88.261 )PM*pxE( t,i )209.67( )228.09( )108.58( )832.16( )933.71( )427.55( 85.76- 28.13- 16.52 84.34- 34.31- 21.4- )PM*pxE( 1−t,i )391.95( )658.29( )896.67( )933.75( )302.31( )914.65( ***18.81- 34.4- ***63.21- ***48.01- ***34.71- ***36.9htworgPDG t,i )927.3( )441.5( )844.3( )449.2( )808.3( )507.2( 60.0- 61.0 40.0- 10.0 40.0- 00.0- PMOCLOP t,i )411.0( )892.0( )390.0( )490.0( )590.0( )190.0( 10.0- 30.0- 10.0- 00.0- 20.0- 00.0- FNI t,i )310.0( )920.0( )310.0( )010.0( )210.0( )900.0( 43.4 92.4 10.1- 29.0- 23.3 80.0 htworgPDG 1−t,i )292.5( )488.4( )457.2( )053.2( )235.4( )141.2( 00.0- 52.0- 40.0- 50.0- 20.0- 60.0- PMOCLOP 1−t,i )201.0( )913.0( )370.0( )580.0( )680.0( )870.0( 20.0 00.0- 00.0 10.0 20.0 10.0 FNI 1−t,i )210.0( )720.0( )310.0( )010.0( )310.0( )010.0( 96.0 ROHCNA t,I )154.0( *02.04 )PM*ROHCNA( t,I )515.32( 69.7 )PM*ROHCNA( 1−t,I )945.22( 315,1 026 439 578,1 145,1 860,2 snoitavresbO 102.0 442.0 591.0 061.0 681.0 302.0 2R denfiedsaeraselbairaV .2elbaTnidenfiedsaytisnetnIedartSU–ytivarG .)3(nisecudortnisnoissergertigol-lenapehtfostneicffieocdetamitseehtswohselbatehT deggal dna noitcaretni eht dna t raey nevig a ni ralloD .S.U eht ot deggep si yrtnuoc a fi 1 ot lauqe si hcihw ,rohcnA fo noitidda eht htiw ,)4( dna )2( selbaT ni .IIWW-tsopdetamitseledomehtrofstluserstneserpIInmuloC.2elbaTnidetneserpsaemasehteraelpmaselohwdelebalstluseR.PMdnarohcnAfonoitcaretni tneserp V dna VI snmuloC .sraw dlrow owt dna ,noisseceR taerG eht ,noisserpeD taerG eht ot gnitad snoitavresbo eht tuohtiw ledom eht setamitse III nmuloC .slortnoc rohcnA lanoitidda eht htiw tub elpmas elohw eht rof detamitse si IV nmuloC .ylevitcepser ,seimonoce gnigreme dna gnipoleved rof setamitse 31

snoitacfiiceps evitanretlA :ssentsuboR :6 elbaT IV V VI III II I C: Y t,i noitazirosniw tuohtiW sesirCKB degreM dedda RX & etaRtnI tiborP SLO enilesaB ytisnetnIedartSU–ytivarG :pxE **94.3- **38.2- 96.3- ***29.1- *40.0- **54.3pxE t,i )663.1( )051.1( )483.2( )725.0( )520.0( )153.1( ***13.361 **58.141 **93.291 ***52.57 **38.4 ***88.261 )PM*pxE( t,i )009.65( )361.65( )065.39( )809.32( )619.1( )427.55( 42.0 80.21- 89.59- 42.01- 34.0 21.4- )PM*pxE( 1−t,i )769.65( )351.55( )353.27( )736.22( )654.2( )914.65( ***95.8- ***57.9- ***77.31- ***15.5- ***53.0- ***36.9htworgPDG t,i )207.2( )185.2( )141.5( )580.1( )590.0( )507.2( 00.0- 00.0- 01.0 00.0- 00.0 00.0- PMOCLOP t,i )190.0( )380.0( )842.0( )050.0( )400.0( )190.0( 10.0- 00.0- 30.0- 00.0- 00.0 00.0- FNI t,i )700.0( )900.0( )720.0( )400.0( )000.0( )900.0( 90.0- 84.1- 23.5- 02.0- 00.0 80.0 htworgPDG 1−t,i )190.2( )980.2( )358.7( )690.1( )760.0( )141.2( 60.0- 80.0- 11.0- 30.0- 00.0- 60.0- PMOCLOP 1−t,i )870.0( )270.0( )832.0( )240.0( )400.0( )870.0( 10.0 10.0 30.0 00.0 00.0 10.0 FNI 1−t,i )800.0( )900.0( )520.0( )500.0( )000.0( )010.0( 70.0 setaRtnI∆ t,I )940.0( 30.0setaRtnI∆ 1−t,I )730.0( 30.0 setaRhcxE∆ t,I )430.0( 20.0 setaRhcxE∆ 1−t,I )240.0( 860,2 031,2 955 860,2 152,4 860,2 .sbO 002.0 402.0 712.0 502.0 451.0 302.0 derauqs-R ni detneserp sa emas eht era enilesaB delebal stluseR .)3( ot snoitacfiidom fo smret ni snoisserger tigol-lenap eht fo stneicffieoc detamitse eht swohs elbat ehT ,setaRtnI∆ edulcni ew ,VI nmuloc nI .tigol fo daetsni ,ylevitcepser ,tiborp dna SLO htiw detamitse sa stluser eht wohs snmuloc tiborP dna SLO ehT .2 elbaT ,ataD laicnaniF labolG morf ,rallod eht ot ycnerruc lacol eht fo etar egnahcxe eht ni egnahc eht ,setaRhcxE∆ dna ,sdleiy yrusaerT htnom-3 lacol ni egnahc eht ;)1002( .la te odroB fo sesabatad sesirc cimetsys eht gnigrem yb demrof sesirc gniknab fo noitinfied evitanretla na esu ew ,V nmuloc nI .selbairav lortnoc sa gniylppatuohtiwdetamitseledomniamruorofstluserwohsoslaeW .)2102(rolyaTdnakciraluhcS;)2102(dleftsbOdnasahcniruoG;)2102(aicnelaVdnaneveaL .IV nmuloc ni noitazirosniw 32