Foreign Effects of Higher U.S. Interest Rates

K.7 Foreign Effects of Higher U.S. Interest Rates Iacoviello, Matteo and Gaston Navarro Please cite paper as: Iacoviello, Matteo and Gaston Navarro (2018). Foreign Effects of Higher U.S. Interest Rates. International Finance Discussion Papers 1227. https://doi.org/10.17016/IFDP.2018.1227 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1227 May 2018

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1227 May 2018 Foreign Effects of Higher U.S. Interest Rates Matteo Iacoviello and Gaston Navarro NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Foreign Effects of Higher U.S. Interest Rates ∗ Matteo Iacoviello and Gaston Navarro † ‡ April 23, 2018 Abstract This paper analyzes the spillovers of higher U.S. interest rates on economic activity in a large panel of 50 advanced and emerging economies. We allow the response of GDP in each country to vary according to its exchange rate regime, trade openness, and a vulnerability index that includes current account, foreign reserves, inflation, and external debt. We document large heterogeneity in the response of advanced and emerging economies to U.S. interest rate surprises. In response to a U.S. monetary tightening, GDP in foreign economies drops about as much as it does in the United States, with a larger decline in emerging economies than in advanced economies. In advanced economies, trade openness with the United States and the exchange rate regime account for a large portion of the contraction in activity. In emerging economies, the responses do not depend on the exchange rate regime or trade openness, but are larger when vulnerability is high. Keywords: U.S. Monetary Policy; Foreign Spillovers; Local Projection; Macroeconomic Transmission; Panel Data. JEL Classification: F4, E5, C3 ∗We thank Joshua Herman and Andrew Kane for excellent research assistance. We also thank Jim Hamilton, Andrew Rose, Christopher Erceg, Giovanni Favara, Zheng Liu, Glenn Rudebusch, and Mark Spiegel for their comments and suggestions. The views expressed are those of the authors and not necessarily those of the Federal Reserve Board or the Federal Reserve System. †FederalReserveBoard,DivisionofInternationalFinance,20thandCSt. NW,WashingtonD.C.,20551,United States. Email: matteo.iacoviello@frb.gov ‡FederalReserveBoard,DivisionofInternationalFinance,20thandCSt. NW,WashingtonD.C.,20551,United States. Email: gaston.m.navarro@frb.gov 1

1 Introduction This paper presents new empirical evidence regarding the cyclical response of foreign economies to U.S. monetary shocks. We make use of a large dataset exploiting the time-series and crosssectional variation of foreign economies in their exchange rate regime, trade openness, and an index of their external vulnerability. Our goal is to gain some empirical sense of the differential importance of exchange rate channels, trade channels and broad “financial” channels in response tochangesinU.S.interestrates. Unlikepreviousstudiesthathavefocusedonlimitedtimeperiods, a few countries, or limited controls, we rely on a comprehensive dataset containing observations on quarterly GDP and time-varying country characteristics for 50 foreign economies for over 50 years. While data quality in international datasets varies systematically across countries and over time, we believe this is a reasonable price to pay for a dataset that, by exploiting nearly 10,000 observations, is about two orders of magnitude larger than the typical dataset used to study the domestic effects of U.S. monetary shocks. Figure 1 shows the federal funds rate from 1965 through 2016. The shaded areas denote periods of rising interest rates. Figure 2 zooms in on the six tightening episodes before the Global Financial Crisis, showing GDP growth in each episode relative to what one could have predicted using a simple forecasting model.1 The bars measure average growth surprises from the beginning of each episode until one year after its end. For instance, in panel 1, Mexico’s GDP growth from 1978:Q1 through 1982:Q2 was about 3 percentage points higher, on average, than what one could have predicted using data up to 1977:Q4.2 The non-uniform pattern of the bars across countries and episodes illustrates how the experience in the aftermath of U.S. monetary tightenings varies across foreign economies. The high interest rates of the late 1970s–early 1980s eventually led to lackluster growth in the United States andmostforeigneconomies(panel1). Thetighteningsofthe1980swerefollowedbyweakergrowth in many emerging market economies (panels 2 and 3), but the situation was reversed with the higher interest rates of the mid–1990s, which were followed by stronger growth across the board (panel 4). The higher interest rates of the late 1990s were followed by lower growth among some emerging economies (panel 5). Finally, the most recent tightening period was followed by an acceleration in growth across all global economies (panel 6). Averaging across episodes, growth in the 1The forecasting model is, for each country, a univariate autoregressive model for log GDP with four lags and a time trend. To avoid cluttering, some economies are grouped ex post into regional clusters, with a bar for the average GDP response across them. 2Foreachcountry,theregressionsstartin1960:Q1orlaterdependingondataavailability,andareestimatedusing the full sample. The forecasts are computed dynamically—using the coefficients estimated for the full sample— starting from the last observation before the monetary tightening. The dynamic forecasts do not use actual data but exploit the hindsight of knowing the estimated trend growth and AR coefficients for the full sample. 2

United States and advanced economies was slightly higher than forecasted (+0.2 and +0.3 percent, respectively), whereas growth in emerging economies was slightly lower (-0.4 percent) in the years after these episodes. Additionally, the dispersion across episodes for emerging economies was twiceaslargeasforadvancedeconomies(standarddeviationof2versusstandarddeviationof0.9). This large dispersion—across and between countries—suggests that not all tightenings are created equal. The nature of the tightening episode as well as country or region-specific characteristics could account for their heterogeneous responses. This is the perspective adopted here. In the first step, we extract interest rate surprises using quarterly data from 1965 through 2016 to isolate exogenous movements in U.S. interest rates that are unlikely to be correlated with either domestic or global economic conditions.3 In the second step, we study how the spillovers to foreign economies of interest rate surprises depend on three factors: (1) the exchange rate regime against the dollar, (2) trade openness with the United States, and (3) an index of external vulnerability. We use a panel for 50 advanced and emerging economies, and estimate spillovers using a local projections method (Jorda, 2005). The interest rate spillovers are allowed to differ over time according to these three factors, and across emerging and advanced economies. The paper’s main results are: 1. The foreign spillovers of higher U.S. interest rates are large, and on average nearly as large as the U.S. effects. A monetary policy-induced rise in U.S. rates of 100 basis points reduces GDP in advanced economies and in emerging economies by 0.5 and 0.8 percent, respectively, after three years. These magnitudes are in the same ballpark as the domestic effects of a U.S. monetary shock, which reduce U.S. GDP by about 0.7 percent after two years. 2. Inadvancedeconomies, higherU.S.interestratesaretransmittedthroughstandardexchange rate and trade channels. In particular, the responses within advanced economies are larger when a country’s currency is (de facto) pegged to the dollar, or when its trade volume with the United States is high. 3. In emerging economies, exchange rate and trade channels explain little of the differential GDP responses within economies. Instead, a vulnerability index that we interpret as capturingacountry’sfinancialfragilityexplainsasizablecomponentofdifferencesacrosseconomies, with GDP in more vulnerable economies falling much more in response to a U.S. monetary 3Mostofourfocusisoninterestrateincreasesdrivenbymonetarypolicyshocks. However,Section6.1discusses the effect of higher U.S. interest rates due to improved economic conditions. 3

tightening. This vulnerability index is constructed combining by current account, foreign reserves, inflation, and external debt. Our estimation methodology exploits both the between- and the within-country variation in a set of observables that are often viewed as important determinants of the foreign spillovers of U.S. interest rate changes. Several studies that have recently examined the international effects of U.S. monetary actions using vector autoregressions (VARs) or event studies have relied on the implicit assumption that many country characteristics that determine such effects are fixed across the sample.4 Such an assumption is invalidated by the data for virtually all the variables that we consider in our sample, with all our indicators exhibiting far more variation within than across country borders. For instance, in the 1960s and 1970s, Mexico had a lower level of trade openness with the United States than South Korea did, but Mexico’s trade exposure grew by a factor of four in the decades since the NAFTA trade agreement, while Korea’s openness remained constant. Similarly, several advanced economies were effectively pegged to the dollar before the collapse of the Bretton Woods system in 1971, and adopted a floating exchange rate regime afterwards. More recently, China abandoned its peg to the dollar in 2010, increasing its exchange rate flexibility. Studies that ignore time-variation in these country characteristics are likely to estimate the effects of interest rate changes with a large amount of noise. Section 2 reviews the theoretical underpinnings of the international transmission of interest rate shocks. Section 3 describes the data. Section 4 discusses the methodology and results of the effects of U.S. interest rates shocks. Section 5 extends our methodology to look at state-dependent effectsofinterestrateshocks. Section6containsrobustnessanalysis. Section7containsahistorical quantification of the effect of U.S. monetary shocks on foreign economies. Section 8 concludes. 2 Channels of International Interest Rate Transmission 2.1 The Channels Models of international interest rate transmission typically emphasize exchange rate channels, trade channels, and financial channels as key determinants of the response of foreign economies to changes in interest rates in another country.5 The first two channels are a staple of virtually 4For a list of papers that have examined the foreign effect of higher interest rates, see Kim (2001), Canova (2005), Dedola, Rivolta, and Stracca (2017), Ehrmann and Fratzscher (2005), Ma´ckowiak (2007), Di Giovanni and Shambaugh (2008), and Georgiadis (2016). See the Appendix. 5WeborrowthisclassificationfromAmmer,DePooter,Erceg,andKamin(2016). Blanchard,Das,andFaruqee (2010) discuss a similar set of channels in accounting for the impact of the Global Financial Crisis on emerging economies. See also Kim (2001). 4

all general equilibrium, intertemporal models of macroeconomic policy transmission that merge Keynesian pricing assumptions and international market segmentation building on the Mundell- Fleming-Dornbusch framework.6 Financial channels have been emphasized in recent work that has studied the international implications of various types of credit market frictions.7 The exchange rate channel is predicated on the idea of demand substitution between domestic and foreign-produced goods, and implies that higher interest rates in the United States may lead to an expansion of activity abroad. Consider, for instance, an increase in interest rates in the United States. Via the uncovered interest parity condition, higher U.S. interest rates lead to an appreciation of the dollar. In turn, the stronger dollar moves the composition of world demand away from U.S. goods and towards goods produced in other countries. With flexible exchange rates, GDP in foreign economies should rise, boosted by cheaper exports. By contrast, a country that pegs its exchange rate to the dollar should experience an appreciation that lowers its GDP. The trade channel rests on the idea that higher U.S. interest rates reduce incomes and expenditures in the United States, thus leading to lower U.S. demand for both domestically produced and imported goods, and reducing activity and GDP abroad.8 Overall, the strength of this channel should depend on the share of exports and imports in economic activity (the trade exposure), especially with the United States. Financial channels capture the idea that higher U.S. interest rates can spillover to the price of various financial assets and liabilities held abroad, thus affecting activity in foreign countries even after controlling for exchange rate and trade channels. For instance, when domestic agents are credit constrained and hold dollar denominated debt, an increase in U.S. interest rates may lead to a deterioration of domestic balance sheets in the presence of flexible exchange rates.9 A common themebehindthefinancialchannelsisthatfrictionsthatpreventintertemporalsmoothingthrough foreign borrowing and lending may magnify the impact of foreign shocks for economies that are integrated with the world markets. These frictions can be exacerbated when the fundamentals of a country are weak. For instance, high inflation may create political instability and constrain domesticmonetaryandfiscalresponsestoadverseshocks. Similarly, alargecurrentaccountdeficit or low foreign reserves may put a country at risk of facing financial pressure from foreign lenders. 6See for instance the work of Obstfeld and Rogoff (1995) for a modern, micro-founded exposition of this framework. 7See for instance Aghion, Bacchetta, and Banerjee (2004) and Gertler, Gilchrist, and Natalucci (2007). 8See Erceg, Guerrieri, and Gust (2005) for a two-country DSGE model where demand shocks in one country yield positive output spillovers to another country via the trade balance channel. 9These“financialaccelerator”effectsmayworkevenwithfixedexchangerates. Whenacountrypegsitsexchange rate, the rise in domestic nominal interest rate which is required to maintain the peg may lead to a significant increase in the country’s real borrowing costs. In turn, the rise in borrowing costs may induce a contraction in output which is further magnified by asset price channels operating through the financial accelerator. 5

Recent work has also highlighted the importance of global factors that can propagate changes in one country’s monetary conditions to the rest of the world, especially when capital markets are highly integrated. Rey (2015) and Miranda-Agrippino and Rey (2017) show that changes in interestratesin“core”countriescantriggeraglobalfinancialcyclethat, regardlessoftheexchange rate regime, may generate positive global spillovers. Bruno and Shin (2015) find evidence of monetary policy spillovers on cross-border capital flows. This work highlights channels that seem to operate independently of, and above, the traditional exchange rate and trade channels. 2.2 Disentangling the Channels Is it possible to tell these channels apart? Without loss of generality, consider an increase in U.S. interest rates driven by an exogenous monetary shock. Iftheexchangeratechannelisimportant, theexchangerateregimeshouldexplainasubstantial portion of the cross-country variation in GDP response following an increase in U.S. interest rates. In particular, the traditional version of this channel predicts that a country that pegs its exchange rate to the dollar should experience a larger negative GDP response. If trade channels are important, trade intensity with the United States should matter for the cross-country GDP response to higher U.S. interest rates, even after controlling for the exchange rate response. In particular, this channel predicts that higher levels of trade with the United States will lead to a larger GDP contraction in response to an increase in U.S. interest rates, as the decrease in U.S. demand spills over to the exports of the largest U.S. trading partners. All other transmission mechanisms fall under the category of financial channels. By financial channels, we mean mechanisms that stem from the presence of various forms of market imperfections and that operate above and beyond the standard Mundell-Fleming-Dornbusch model. Suppose that we have already controlled for exchange rate regime and trade openness with the United States in assessing the foreign GDP response to U.S. interest rate shocks. We conjecture that, if additional financial variables can explain residual differences in how countries respond to U.S. interest rate changes, these additional variables are likely to capture the role of financial channels in international business cycles. To what extent can we measure the strength of financial channels in the international transmission of monetary policy? Our strategy is to construct a summary indicator of variables that have a high probability of signaling the weakness in the economic fundamentals of a country. For practical purposes, these variables must be readily available and be somewhat consistently defined across countries and over time. In our analysis, we focus on four variables: a country’s current account deficit, foreign reserves, inflation, and external debt. We combine these four variables 6

in a summary indicator which combines them using equal weights, and we label this summary indicator the vulnerability index. The above classification is obviously a simplification, and we illustrate potential pitfalls with one example. It is possible that the exchange rate channel matters but not through the standard dollar anchoring classification that we use. For instance, the exchange rate channel might be captured by trade invoicing, as discussed by Gopinath (2015).10 U.S. monetary policy might matter because exports and imports are priced in U.S. dollars regardless of the exchange rate regime. Channels of this kind—or broadly-based confidence channels based on the outsize role of U.S. monetary policy—could also capture residual differences in the effects of higher U.S. interest rates, but we do not control for them in our analysis. 3 The Data This Section describes the data used in our paper. Additional details on the sources are provided in the Appendix. 3.1 Data on GDP Our main focus is on the effects of changes in U.S. interest rates on foreign real GDP. To this end, we put together a novel dataset containing quarterly GDP data for 50 foreign economies (25 advanced and 25 emerging) plus the United States. The coverage, which varies across countries, spans from as early as 1965:Q1 to as late as 2016:Q2. Our benchmark analysis uses GDP data for the countries listed in Table 1. For some countries, we extend backward the original, publicly available quarterly GDP series using annual GDP data that are available from the World Bank’s World Development Indicators. To convert the annual data into a quarterly frequency, we use Denton’s proportional interpolation method (Chen, 2007). Foremergingeconomies,the“indicatorseries”usedforinterpolationisthepurchasingpowerparity (PPP) weighted GDP of the emerging economies for which quarterly GDP data are available. We adopt a similar procedure for advanced economies, where the interpolation method uses PPPweighted GDP of the other advanced economies (excluding the United States). 10Long-span information on trade invoicing is scant. Gopinath (2015)’s index of trade invoicing starts in 1999. 7

3.2 Control Variables: The Exchange Rate Regime, Trade Openness, and Vulnerability Index. Ouranalysisalsofocusesonhowspecificvariablesacrosscountriesaffectthespilloversfrominterest rate changes to GDP outcomes. To this end, we compile data on the exchange rate regime against the dollar, trade openness with the United States, and other variables for all the countries in the dataset. We use these data to construct indexes of (1) exchange rate exposure, (2) trade exposure, and (3) external vulnerability. 1. For the exchange rate regime, we draw on the narrative analysis of Ilzetzki, Reinhart, and Rogoff (2017) and our own analysis of the literature to construct an index ranging from 0 to 1 for each country and period. We classify a country as 0 if it maintains a flexible exchange rate against the U.S. dollar, 1/2 if it maintains an exchange rate band, and 1 if it pegs against the dollar. In sum, the index takes on higher values the “more” a country pegs its exchange rate to the dollar. 2. For each country, we measure its trade openness with the United States by taking the sum of exports to, and imports from, the United States, divided by GDP. 3. Our external vulnerability index is an equally-weighted average of four indicators that we use to measure the financial “health” of a country:11 (a) Inflation,measuredineachcountrybytheyear-on-yearchangeintheheadlineconsumer price index; (b) Current account deficit, expressed as a share of GDP; (c) External debt less foreign exchange reserves, expressed as a share of GDP; (d) Foreign exchange reserves, expressed as a share of GDP. 4 Average Spillovers of Higher U.S. Interest Rates In this section, we estimate the foreign and domestic spillovers of higher U.S. interest rates. We consider higher rates as a scenario in which the policy rate is higher than what could have been predicted usingan estimated feedback rule.12 In this section, we estimatethe averageinternational 11Some of these indicators are not available early in the sample, as shown in Table 1. To avoid dropping observations relative to our benchmark analysis, we fill in the missing observations using backward extrapolation. For instance, we assume that the current account position of a country in 1965-1969 is equal to its 1970 value. Repeating this analysis without filled-in observations yields nearly identical results to those reported in the paper. 12We also analyzed the effects of an alternative scenario in which monetary policy endogenously responds to improved domestic conditions. The results of this alternative scenario are discussed in Section 6. 8

spillover of higher rates, while Section 5 discusses how this spillovers may depend on the economy’s exposure to exchange rate, trade, and financial vulnerability channels. 4.1 Identification of U.S. Monetary Shocks We identify U.S. monetary shocks by regressing the federal funds rate on a set of controls, and use the residuals as the identified shocks. In particular, we estimate shocks u as the residual in t following regression: r = θ +θ Z +u (1) t 0 1 t t where r is the federal funds rate. The set of controls Z includes contemporaneous and lagged t t values of inflation, log U.S. GDP, corporate spreads, log foreign GDP, as well as lagged values of the federal funds rate and a quadratic time trend.13 Because we include current macroeconomic variables as controls, our shock identification is analogous to a Cholesky identification in a VAR that orders the federal funds rate last, as done by Christiano, Eichenbaum, and Evans (2005) and others.14 We use quarterly data from 1965:Q1 to 2016:Q2, and replace the federal funds rate with the Wu-Xia shadow rate from 2009 to 2015 to account for the zero lower bound and for the stimulus to the economy provided by the unconventional monetary policy actions that followed the Great Recession.15 Figure 3 plots the identified monetary shocks. The largest contractionary shocks are in the early 1980s during the Volcker tightening period, and in 2008 at the onset of the zero-lower-bound era. In recent years, the identified shocks point to a tightening of policy in 2013, around the period of the taper tantrum, as well as to an easing in 2014 and 2015. 4.2 Estimation of the Foreign Effects With the identified monetary shocks at hand, we compute the dynamic responses of foreign and U.S. GDP using the local projection method developed by Jorda (2005). This method allows us to computetheresponseofvariablestoshocksatdifferenthorizonswithoutimposingmanystructural 13We use four lags for all variables. Inflation is measured as the four-quarter change in the GDP deflator. Corporate spreads correspond to the difference between the Moody’s seasoned Baa corporate bond yield and the 10-YearTreasurynoteyieldatconstantmaturity. WeconstructanindexofforeignGDPbycumulatingtheaverage of quarter-on-quarter GDP growth for the countries in the sample. Each quarter, the weights are based on each country’s GDP in constant US dollars from the World Bank World Development Indicators (if data for a country are not available, its weight is set at zero, and the weights of other countries are changed accordingly). 14Our results below are robust to using the monetary shocks measure constructed by Romer and Romer (2004). See Section 6. 15See Wu and Xia (2016) for details. 9

restrictions. This flexibility can be easily extended to estimate state-dependent responses, which eases comparison with the findings of the next section, where we compute responses as a function of the economy’s exposure to interest rate shocks.16 For computing the response of U.S. GDP, we estimate the following equation: y = α +β u +A Z +(cid:15) for h = 0,1,2,...,H (2) t+h h h t h t t+h where y is U.S. GDP in quarter t + h, u is the monetary shock, and Z is a set of controls. t+h t t A plot of β is the dynamic response of U.S. output to an innovation in u . We also estimate h t { } equation (2) using the federal funds rate as y to compute its response to the identified shock. t+h In both cases, the set of controls Z includes four lags of y and a quadratic time trend. t t We take advantage of the panel dimension when computing the foreign GDP response to the monetary shock. In particular, we estimate a version of (2) as follows: y = α +β u +A Z +(cid:15) for h = 0,1,2,...,H (3) i,t+h i,h h t h,i i,t i,t+h where y is the GDP of country i in quarter t + h, and α is a country-specific fixed effect. i,t+h i,h Notice that we project all countries on the same shock u . Accordingly, β measures the average t h { } responseofoutputacrosscountriestoaninnovationinu . ControlsZ includefourlagsofcountry t i,t i’s GDP, as well as a linear and a quadratic trend.17 WeareinterestedindocumentinghowresponsestohigherU.S.ratesmaydifferacrossadvanced and emerging economies. To this end, we estimate equation (4) separately for advanced and emerging economies. 4.3 Results: U.S. Monetary Policy Matters Figure 4 shows the response of U.S. GDP, the federal funds rate, and foreign GDP to a monetary shock. The shaded areas denote 68 percent confidence intervals and are based on robust standard errors that account for serial correlation (in the case of the U.S. responses) and for clusters by time and country (in the case of the foreign responses).18 A shock that increases the federal funds rate by 1 percentage point induces a lasting decline in U.S. GDP, which contracts output by 0.7 16See, for instance, Auerbach and Gorodnichenko (2013) for a recent example of state-dependent multipliers estimation using Jorda (2005)’s local projections method. 17We let the coefficients on the controls Z be country-specific. Assuming common coefficients across countries i,t makes foreign responses to U.S. monetary shocks marginally larger than in the specification presented here. 18We calculate the confidence bands using the Driscoll and Kraay (1998) standard errors that already allow arbitrary correlations of the error term across countries and time. 10

percent after two years and recovers thereafter. The magnitude and duration of the U.S. output response to a monetary shock is largely in line with previous findings in the literature (Ramey, 2016). The dynamic response of GDP in advanced foreign economies follows a similar profile to the U.S. one, but is smaller and more delayed, with GDP dropping by about 0.5 percent three years after the shock. The GDP response of emerging economies is as delayed as that of the advanced economies, but eventually as large as the one in the United States, with GDP falling 0.7 percent fouryearsaftertheshock. Alltold, theresultshighlighthowemergingeconomiesaremoreexposed than advanced economies to higher U.S. interest rates. 5 Foreign Effect of Higher U.S. Interest Rates: Disentangling the Channels of Transmission We turn now to estimating how a country’s dynamic response to a monetary shock depends on exchange rate, trade, and financial channels. 5.1 Methodology Consider a set of variables v that measure the exposure of an economy to higher U.S. interest ∈ V rates, and let higher values of v represent higher exposure. To estimate how exposure affects the economy’s response to a monetary shock, we extend the specification in equation (3) so that the identified shock interacts with the measures of exposure. In particular, we estimate the following equation: y = α +β u + (cid:88) βv (cid:0) ev u (cid:1)⊥ +A Z +(cid:15) for h = 0,1,2,...,H, (4) i,t+h i,h h t h i,t−1 t h,i i,t i,t+h v∈V where ev is the exposure index for variable v. The interaction term (cid:0) ev u (cid:1)⊥ is constructed so i,t i,t−1 t that β captures the response to a shock when the exposure measures are at their median values, h and βv represents the marginal response to the shock when exposure ev is high. h i,t−1 We construct the interaction term (cid:0) ev u (cid:1)⊥ in five steps. First, we standardize each exposure i,t−1 t variablev bysubtractingitsmeananddividingbyitsvariance. 19 Second, weconstructalogistic i,t exp vs transformation of the standardized variable (vs ) as (cid:96)v = { i,t} . Third, we re-center (cid:96)v in i,t i,t 1+exp vs i,t { i,t} 19The standardization is a simple device to put all variables on equal footing, and follows the lead of many, including Auerbach and Gorodnichenko (2013) and Herrera and Garcia (1999). 11

(cid:96)v −(cid:96)v terms of the distance between its 50th and its 95th percentile: ev = i,t 50, where (cid:96)v corresponds i,t (cid:96)v −(cid:96)v p 95 50 (cid:0) (cid:1) to the pth percentile of (cid:96)v . Fourth, we construct the interaction term ev ur . Finally, we i,t i,t−1 t (cid:0) (cid:1) orthogonalize ev ur using a recursive procedure. For the first exposure variable v , we regress i,t−1 t 1 (cid:0) ev1 ur (cid:1) on [u , Z ] and obtain the residual (cid:0) ev1 ur (cid:1)⊥ . For the second variable v , we regress i,t−1 t t i,t i,t−1 t 2 (cid:104) (cid:105) (cid:0) ev2 ur (cid:1) on u , Z , (cid:0) ev1 ur (cid:1)⊥ and obtain the residual (cid:0) ev2 ur (cid:1)⊥ . We proceed in a similar i,t−1 t t i,t i,t−1 t i,t−1 t vein with the other exposure measures.20 The standardization step makes all the exposure variables comparable. The logistic transformation maps variables to the unit interval which allows us to consider them in distributional/probabilistic terms.21 The re-centering step allows us to interpret the coefficients as deviations from median levels of exposure. In particular, β is the response to the shock when all h exposure indexes are at their median value, and β +βv is the response when the exposure index h h ev is at the 95th percentile of its distribution. i,t TheorthogonalizationstepeasesinterpretationandcomparisonwithSection4.3. Inparticular, because all the interaction terms are orthogonal to the shockur, the β estimated in equation (4) is t h identical to the one from equation (3). Thus, we keep considering β as the average response to h { } the shock. Furthermore, because each additional exposure measure is orthogonal to the previous ones, we can interpret βv as the marginal effect of variable v on the pass-through of the monetary h shock to foreign GDP when v moves from the 50th to the 95th percentile of its distribution. 5.2 Exposure Variables In practice, we consider three measures of exposure that capture the three channels discussed in Section 2. 1. Exchange Rate Channel: We construct a variable measuring the degree to which a country’s currencyispeggedtothedollar. Thevariableequals0whenacountryhasaflexibleexchange rate against the dollar, 0.5 if the country pegs against the dollar within a somewhat large band (+/ 5 percent), and 1 if the country is closely pegged to the dollar (including a − +/ 2 percent band). We consider countries with a higher degree of anchoring to the dollar − as more exposed to U.S. monetary shocks, as higher U.S. rates would induce an appreciation of the dollar—and thus, the domestic currency—which depresses GDP by making imports 20More generally: for the nth exposure variable v , we regress (cid:0) evn ur(cid:1) on n i,t−1 t (cid:104) u , Z , (cid:0) ev1 ur(cid:1)⊥ , (cid:0) ev2 ur(cid:1)⊥ ,..., (cid:0) evn−1ur(cid:1)⊥ (cid:105) and obtain the residual (cid:0) evn ur(cid:1)⊥ . This procet i,t i,t−1 t i,t−1 t i,t−1 t i,t−1 t dure is known as regression by successive (Gram-Schmidt) orthogonalization. See for instance Balli and Sørensen (2013) for an application to regressions with interaction effects. 21The logistic transformation is a simple manner to estimate the state-dependent effect of shocks that has been extensively used in recent work. See Auerbach and Gorodnichenko (2017) and Ramey (2016). 12

cheaper and exports more expensive. The median observation in our sample for advanced economies is a flexible exchange regime, which applies to 80 percent of the country-quarter observations. Instead, the median for emerging economies is a system with a close anchor to the dollar, which applies to 55 percent of the observations.22 2. Trade Channel: We measure the amount of trade with the United States (exports plus imports) as a fraction of the country’s GDP. Note that the median amount of trade with the United States is about 3.5 percent of GDP for advanced economies (such as the United Kingdom in the 2000s), and around 10 percent of GDP for emerging economies (such as Chile in the 2000s). 3. Financial Channel: We construct a vulnerability index as an equally-weighted average of the following four variables: current account deficit, foreign reserves (entering with a negative sign), inflation, and external debt.23 A large current account deficit may limit the willingness of foreign lenders to extend credit, or may even trigger sharp capital outflows, especially in the presence of high interest rates abroad. Additionally, evidence from Claessens, Dell’Ariccia, Igan, and Laeven (2010) indicates that large current account deficits raise the incidence and severity of a crisis. Both credit risk agencies and international organizations frequently consider foreign reserves and external debt in assessing the external vulnerability of a country.. See for instance Santacreu (2015). Additionally, there is evidence that both variables are important in capturing the sensitivity of an economy to adverse shocks. For instance, Frankel and Saravelos (2012) suggest that central bank reserves are one of the leading indicators in explaining crisis incidence across different countries. Lane and Milesi-Ferretti (2017) indicate that excessive reliance on debt finance may increase a country’s actual and perceived vulnerability. Although not a direct measure of financial channels, we also include inflation—measured by the annual change in the consumer price index—in our vulnerability index. High inflation may indicate structural problems in a government’s finances, or could generate political instability which in turn acts as an amplifier of the effects of higher U.S. interest rate. High inflation may also increase the sensitivity of a country’s borrowing costs to changing interest rates. For instance, Cantor and Packer (1996) find that inflation is a significant determinant of sovereign ratings. 22Ilzetzkietal.(2017)alsonotethat,bytheirclassification,theU.S.dollarscoresbyawidemarginastheworld’s dominant anchor currency. 23Asanalternativetoanequally-weightedaverage,wealsoconsideredthefirstprincipalcomponent. Theresults were qualitatively and quantitatively similar to those presented here. 13

For each variable, we take a three-year moving average and truncate observations on both sides at a 5 percent threshold in order to remove outliers and to guard against extreme measurement error—to us, it seems immaterial whether a country has a 100 or a 1,000 percent inflation rate. The three exposure measures are constructed separately for advanced and emerging economies. Table 2 presents the summary statistics for the exposure variables in our analysis. The vulnerability index is constructed so that it takes on high values when foreign reserves are low, and when inflation, external debt, and the current account deficit are high. To give a visual impression of the evolution of these indicators, Figure 5 plots the recent evolution of the three exposure measures for a selected sample of countries.24 The figure showcases the evolution of our exposure measure over time and across countries, which allows us to measure the heterogeneous effects of U.S. interest rates. The top left panel shows how Canada, Japan, and the United Kingdom have at some point in the past abandoned their peg to the dollar.25 Canada, for instance, was closely pegged to the dollar until 2002, kept a managed floating regime between 2002 and 2010, and moved to a floating exchange regime thereafter. The orthogonalization procedure merits some discussion. This procedure is a convenient method to illustrate the marginal effect of each exposure variable after controlling for the others. However, it also implies that the particular ordering of the exposure measures matters. We choose the ordering in a way that conforms closely to the historical evolution of the channels. The exchange rate channel is perhaps the most intuitive and natural, and we order it first. The trade channel matters over and above the exchange rate channel, and we order it second. Finally, the financial channel is a residual channel that captures forces that operate beyond the standard channels, and we order it last. That said, there is little correlation in the data across our exposure measures. Therefore, we experimented with different orderings and found very similar quantitative results. 5.3 Results: Exposure Matters Figure 6 shows the foreign GDP response to a monetary shock, as well as the marginal effects of varying each exposure measure from its median value to its value at the 95th percentile. Theleftcolumnshowshowtheexchangeratechannelaffectstheresponsesofforeigneconomies. For advanced economies, moving from the median—corresponding to a flexible exchange rate regime vis-a`-vis the dollar—to the high end of the distribution—corresponding to a dollar peg— 24In particular, we plot the logistic transformation of the original exposure variables after the second step, that is after truncation and before re-centering. 25See Ilzetzki et al. (2017), which we draw on for our classification. 14

more than doubles the drop in GDP following an adverse U.S. monetary shock. The response among the “high-peg” countries is mostly representative of the early part of the sample, when a large fraction of advanced economies were de facto pegged to the dollar. By contrast, the response of emerging economies is less sensitive to whether they peg to the dollar or not. We illustrate this point in the bottom left panel of Figure 6. Of note, for emerging economies, median and high responses both identify countries that are anchored to the dollar. Nevertheless, the response of countries that are not pegged (shown by the black “low exposure” line) exhibits a similar pattern, with a delayed decline in GDP which bottoms out three years after the monetary shock. Themiddlecolumnshowstheroleofthetradechannel. Foradvancedeconomies,tradeintensity with the United States is an important determinant of the spillovers of U.S. monetary shocks. For instance, moving from the U.K.’s median trade openness to Canada’s high trade openness (see Figure 5) doubles the negative response. For emerging economies, however, trade intensity with the United States matters little. Moving from Korea’s current trade exposure with the United States—a value close to the median—to Mexico’s trade exposure with the United States—a value at the upper end of the distribution—increases the GDP decline only marginally. The right column shows the importance of the financial channels. In both advanced and emerging economies, a high value of the vulnerability index increases the spillovers. This effect is particularly pronounced for emerging economies, where moving from a median to a high level of vulnerability more than doubles the GDP response. Taken at face value, the traditional Mundell-Fleming-Dornbusch view of foreign spillovers is consistent with the response of advanced economies. However, such a view appears at odds with the response of emerging economies, where exchange rate and trade exposure to the United States matteronlylittle. Bycontrast, thefinancialchannelsseemveryimportantforemergingeconomies, much more so than for advanced ones. To shed further light into the contribution of the subcomponents of the vulnerability index to foreign spillovers, Figure 7 illustrates the individual contribution of the four indicators, when they are increased from their median value to their 95th percentile of the distribution. The indicators have little explanatory power for the responses of advanced foreign economies, although a higher current account deficit and higher inflation are associated with a slightly larger GDP decline following a contractionary U.S. monetary policy shock. By contrast, in emerging economies all fourindicators—inflationinparticular—haveexplanatorypowerinenhancingtheresponseofGDP to a U.S. shock. We next provide additional evidence for the channels by investigating how other foreign variables respond to a U.S. monetary shock. These exercises are shown in Figures 8 and 9 for foreign 15

real exchange rate indexes and foreign short-term interest rates, respectively.26 In advanced economies (top panels of Figures 8 and 9), the exchange rate and the interest rate responses follow textbook predictions. The exchange rate appreciates for countries that peg to the dollar, while it depreciates for the (majority of) countries that maintain a flexible exchange rate regime. Peggers increase their interest rate almost one-for-one with the U.S. rate, which leads to an overall appreciation of their currencies. For peggers, the large increase in interest rates causes a large decline in GDP. In experiments not reported here, we also found that real exports drop more in countries that peg against the dollar and in countries that trade relatively more with the United States. In emerging economies (bottom panels of Figures 8 and 9), the real exchange rate appreciates, and policy rates increase: although the peak increase of policy rates is about 50 basis points, policy rates increase much more persistently than they do in the United States. These effects occur regardless of the exchange rate regime. It is perhaps puzzling that the results for emerging economies suggest a significant appreciation of their real exchange rate in response to a U.S. monetary tightening. To us, this puzzling result follows from the persistent increase in domestic interest rates in emerging economies. 6 Robustness This section focuses on studying how the results regarding the foreign effects of an interest rate increase vary when we consider alternative sources of interest rate increases, alternative samples, or alternative monetary shocks. 6.1 Demand Shocks Figure 10 shows the impulse responses when the source of higher interest rates is a U.S. demand rather than a U.S. monetary shock. We compute the aggregate demand shock as the residual of a U.S. log GDP equation against a quadratic time trend, own lags, as well as lagged values of inflation, corporate spreads, log foreign GDP, and federal funds rate. The demand shock is better understood as any combination of supply and demand factors that increases U.S. GDP within the quarter after controlling for past domestic and foreign activity. U.S. GDP and U.S. interest rates (not plotted) increase by 1 percent and by 0.8 percentage points, respectively, before gradually returning to the baseline. The increase in the U.S. interest rate is in line with what one could 26Note that here we plot trade-weighted real exchange rates (with higher values meaning appreciation), which can move even if a country pegs against the dollar. 16

expect from an endogenous response in monetary policy (as would be implied, for instance, by a Taylor rule). When the source of higher interest rates is a U.S. demand shock, the initial foreign response is positive, although the “foreign multiplier” is smaller for emerging than for advanced economies. In emerging economies, the positive spillovers of a positive demand shock are quickly offset by the negative spillovers of higher U.S. interest rates, and GDP falls below baseline after about one year. 6.2 Alternative Samples and Alternative Monetary Shocks We now explore the robustness of the foreign effects of monetary policy shocks around our benchmark specification, which we use as a reference point. Figure 11 shows the results when we allow the foreign effects of U.S. monetary shocks to differ between the pre-1985 and post-1985 period.27 We choose 1985 as the breakpoint following a large literature that dates the mid-1980s as the beginning of the Great Moderation in the United States.28 We find more uncertain effects of monetary shocks for the United States in the post- 1985 period, as shown by the larger confidence intervals around the point estimates. The results for advanced and emerging economies portray a similar picture: GDP initially increases in both blocs, before falling substantially below baseline after two to three years. Importantly, in both subsamples the maximum GDP decline is larger in emerging than in advanced economies, in line with the evidence for the full sample. Additionally, the larger uncertainty around the estimates in the later sample echoes several studies that find that after the 1980s the effects of monetary policy shocks have become more uncertain and harder to interpret (see for instance Ramey (2016)). It is interesting to compare the subsample results with the implications of our full-sample estimates that allow for time-varying exposure measures. Specifically, we compute the impulse responses by subsample by setting the exposure indexes of advanced and emerging economies to their average values in the two subsamples. The results using the median “exposure by period” are shown by the brown lines in Figure 11. According to this alternative set of estimates, the effects of monetary shocks on advanced and emerging economies are slightly smaller in the second half of the sample, mostly because advanced economies have moved on average towards a “more flexible” exchange rate regime, and because emerging economies have become “less vulnerable” in the second part of the sample. However, caution must be used in comparing the two sets of 27Note that we re-estimate the monetary policy rule that we use to extract the monetary shocks across the two different subsamples. 28See for instance McConnell and Perez-Quiros (2000) and Iacoviello, Schiantarelli, and Schuh (2011). 17

estimates. When we split our sample, we are allowing for changes both in the monetary policy rule and in the effects of deviations from that rule. By contrast, when we only vary the exposure by period, we implicitly keep the systematic component of U.S. monetary policy unchanged, thus ignoring the effects of any change in the monetary policy rule itself. Additional robustness exercises are shown in Figure 12. In the top panel, we show the results when we replace the monetary shocks identified using the benchmark specification with the updated Romer and Romer (2004) shocks as constructed by Ramey (2016) for the period from 1969 to 2007 (we use quarterly averages of the original monthly values).29 The results are very similar across exercises, showing that our baseline findings are robust to alternative methods of identifying monetary policy surprises. In the middle panel, we return to our benchmark specification but truncate the sample in 2007:Q4, in order to limit ourselves to the pre-zero lower bound period. The results excluding the zero lower bound period are similar to the benchmark results. In the lower panel, we change the quarterly interpolation method for the observations on GDP that are available at an annual frequency only. In particular, we retain Denton’s interpolation method, but assume that log GDP follows a linear trend within the quarters of the year (subject to the constraint that the sum of quarterly GDP equals annual GDP). As the panel shows, the results barely change. 7 The Historical Contribution of U.S. Interest Rates to Foreign Activity Up to now, we have focused on the question of understanding the nature of foreign spillovers of U.S. monetary shocks. A related question is: How have U.S. monetary policy shocks contributed, historically, to fluctuations in activity in foreign economies? Figure 13 presents the historical contribution of the estimated U.S. monetary shocks to GDP in someselectedeconomies, basedonthecoefficientestimatesofequation(4), andstartingin1975(to avoid cluttering). The bars measuring the median contribution are common across all economies in the advanced bloc, as well as across all economies in the emerging bloc, and illustrate the contribution of U.S. monetary surprises to GDP growth in these economies over the sample. The marginal contribution of the exchange rate, trade, and financial channels varies across economies 29Rudebusch(1998)arguesthatVAR-basedmeasuresofmonetaryshocksmakelittlesense, becausetheyappear at odds with narrative evidence on the nature of the Federal Reserve’s reaction function and because they show little correlation across specifications. 18

and over time, reflecting changes in exposure. For instance, a comparison in the top row between Canada and Japan illustrates the somewhat larger role of U.S. monetary shocks to business cycles in Canada because of Canada’s large trade exposure with the United States. By contrast, in the bottom panel, much of the disparity between Mexico and Korea reflects differences in their vulnerability indexes. For instance, the positive contribution of expansionary monetary policy shocks around 2014, in the aftermath of the taper tantrum, benefits Mexico more than it benefits Korea, reflecting Mexico’s larger values of the vulnerability index. 8 Conclusions Our results shed light on the relative importance of the exchange rate, trade, and financial channels in propagating the effects of U.S. interest rate shocks around the world. The traditional Mundell-Fleming-Dornbusch view of foreign spillovers is consistent with the response of advanced economies. However, such a view appears at odds with the response of emerging economies, where trade and exchange rate exposure to the United States do not seem to matter. By contrast, the financial channels are very important for emerging economies, in addition to having a non-negligible effect on advanced economies. Our findings also highlight both the bright and the dark side of foreign responses to U.S. interest rate increases. On the dark side, these responses seem to be large, to the point that they suggest that foreign economies—especially vulnerable, emerging economies—may react to U.S. monetary shocks more so than the U.S. economy itself. On the bright side, they illustrate how countries that succeed in keeping their financial house in order can weather foreign shocks relatively better than their vulnerable counterparts. 19

References Aghion, P., Bacchetta, P., Banerjee, A., 2004. A corporate balance-sheet approach to currency crises. Journal of Economic theory 119 (1), 6–30. Ammer, J., De Pooter, M., Erceg, C. J., Kamin, S. B., 2016. International spillovers of monetary policy. Tech. rep., Board of Governors of the Federal Reserve System (US). Auerbach, A., Gorodnichenko, Y., 2017. Fiscal stimulus and fiscal sustainability. Tech. rep. Auerbach, A. J., Gorodnichenko, Y., 2013. Fiscal multipliers in recession and expansion. Fiscal Policy after the Financial Crisis, 63–98. Balli, H. O., Sørensen, B. E., 2013. Interaction effects in econometrics. Empirical Economics, 1–21. Blanchard, O. J., Das, M., Faruqee, H., 2010. The initial impact of the crisis on emerging market countries. Brookings papers on economic activity 2010 (1), 263–307. Bruno, V., Shin, H. S., 2015. Capital flows and the risk-taking channel of monetary policy. Journal of Monetary Economics 71, 119–132. Canova, F., 2005. The transmission of us shocks to latin america. Journal of Applied Econometrics 20 (2), 229–251. URL http://dx.doi.org/10.1002/jae.837 Cantor, R., Packer, F., 1996. Determinants and impact of sovereign credit ratings. The Journal of Fixed Income 6 (3), 76–91. Chen, B., 2007. An empirical comparison of methods for temporal distribution and interpolation at the national accounts. Bureau of Economic Analysis. Christiano, L. J., Eichenbaum, M., Evans, C. L., 2005. Nominal rigidities and the dynamic effects of a shock to monetary policy. Journal of Political Economy 113 (1), 1–45. Claessens, S., Dell’Ariccia, G., Igan, D., Laeven, L., 2010. Cross-country experiences and policy implications from the global financial crisis. Economic Policy 25 (62), 267–293. URL http://www.jstor.org/stable/40603207 Darvas, Z., 2012. Real effective exchange rates for 178 countries: a new database. 20

Dedola, L., Rivolta, G., Stracca, L., 2017. If the fed sneezes, who catches a cold? Journal of International Economics 108 (1), S23 – S41, 39th Annual NBER International Seminar on Macroeconomics. URL http://www.sciencedirect.com/science/article/pii/S0022199617300041 Di Giovanni, J., Shambaugh, J. C., 2008. The impact of foreign interest rates on the economy: The role of the exchange rate regime. Journal of International economics 74 (2), 341–361. Driscoll, J. C., Kraay, A. C., 1998. Consistent covariance matrix estimation with spatially dependent panel data. The Review of Economics and Statistics 80 (4), 549–560. Ehrmann, M., Fratzscher, M., 2005. Equal size, equal role? interest rate interdependence between the euro area and the united states. The Economic Journal 115 (506), 928–948. Erceg, C., Guerrieri, L., Gust, C., 2005. Sigma: A new open economy model for policy analysis. International Journal of Central Banking. Frankel, J., Saravelos, G., 2012. Can leading indicators assess country vulnerability? evidence from the 2008–09 global financial crisis. Journal of International Economics 87 (2), 216–231. Georgiadis, G., 2016. Determinants of global spillovers from us monetary policy. Journal of International Money and Finance 67, 41–61. Gertler, M., Gilchrist, S., Natalucci, F. M., 2007. External constraints on monetary policy and the financial accelerator. Journal of Money, Credit and Banking 39 (2-3), 295–330. Gopinath, G., 2015. The international price system. Tech. rep., National Bureau of Economic Research. Herrera, S., Garcia, C., 1999. User’s Guide to an Early Warning System for Macroeconomic Vulnerability in Latin American Countries. Vol. 2233. World Bank Publications. Iacoviello, M., Schiantarelli, F., Schuh, S., 2011. Input and output inventories in general equilibrium. International Economic Review 52 (4), 1179–1213. Ilzetzki,E.,Reinhart,C.M.,Rogoff,K.S.,2017.Exchangearrangementsenteringthe21stcentury: Which anchor will hold? Tech. rep., National Bureau of Economic Research. Jorda, O., 2005. Estimation and inference of impulse responses by local projections. American Economic Review 95 (1), 161–182. 21

Kim, S., 2001. International transmission of us monetary policy shocks: Evidence from var’s. Journal of Monetary Economics 48 (2), 339–372. Lane, P. R., Milesi-Ferretti, G. M. M., 2017. International financial integration in the aftermath of the global financial crisis. Ma´ckowiak, B., 2007. External shocks, us monetary policy and macroeconomic fluctuations in emerging markets. Journal of Monetary Economics 54 (8), 2512–2520. McConnell, M. M., Perez-Quiros, G., 2000. Output fluctuations in the united states: What has changed since the early 1980’s? American Economic Review 90 (5), 1464–1476. URL http://www.aeaweb.org/articles?id=10.1257/aer.90.5.1464 Miranda-Agrippino, S., Rey, H., 2017. U.s. monetary policy and the global financial crisis. Tech. rep. Obstfeld, M., Rogoff, K., 1995. Exchange rate dynamics redux. Journal of political economy 103 (3), 624–660. Ramey, V. A., 2016. Macroeconomic shocks and their propagation. Handbook of Macroeconomics 2, 71–162. Rey, H., 2015. Dilemma not trilemma: the global financial cycle and monetary policy independence. Tech. rep., National Bureau of Economic Research. Romer, C. D., Romer, D. H., 2004. A new measure of monetary shocks: Derivation and implications. American Economic Review 94 (4), 1055–1084. Rudebusch, G. D., 1998. Do measures of monetary policy in a var make sense? International Economic Review 39 (4), 907–931. URL http://www.jstor.org/stable/2527344 Santacreu, A. M., 2015. The economic fundamentals of emerging market volatility. Economic Synopses 2015. Wu, J. C., Xia, F. D., 2016. Measuring the macroeconomic impact of monetary policy at the zero lower bound. Journal of Money, Credit, and Banking 48 (2-3), 253–291. 22

Table 1: Data Availability GDP Dollar Peg Trade U.S. Curr.Acct. Reserves Inflation Ext.Debt Country first firstq last first last first last first last first last first last first last Argentina 1965 1993 2016 1965 2016 1971 2016 1970 2016 1970 2016 1965 2016 1970 2016 Australia 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Austria 1965 1970 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Belgium 1965 1970 2016 1965 2016 1965 2016 1994 2016 1970 2016 1965 2016 1970 2016 Botswana 1965 1994 2016 1965 2016 1974 2016 1974 2016 1975 2016 1965 2016 1975 2016 Brazil 1965 1990 2016 1965 2016 1982 2016 1970 2016 1970 2016 1965 2016 1970 2016 Canada 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Chile 1965 1996 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 China 1965 1992 2016 1965 2016 1972 2016 1981 2016 1976 2016 1965 2016 1980 2016 Colombia 1965 2000 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Czech Republic 1990 1996 2016 1965 2016 1993 2016 1992 2016 1992 2016 1971 2016 1992 2016 Denmark 1965 1966 2016 1965 2016 1965 2016 1970 2016 1970 2016 1967 2016 1970 2016 Ecuador 1965 1990 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 El Salvador 1965 1990 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Finland 1965 1970 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 France 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Germany 1970 1970 2016 1965 2016 1970 2016 1970 2016 1970 2016 1965 2016 1970 2016 Greece 1965 1970 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Hong Kong 1965 1990 2016 1965 2016 1965 2016 1997 2016 1970 2016 1965 2016 1978 2016 Hungary 1991 1995 2016 1965 2016 1991 2016 1991 2016 1991 2016 1967 2016 1991 2016 Iceland 1965 1997 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 India 1965 1996 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Indonesia 1965 1984 2016 1965 2016 1967 2016 1970 2016 1970 2016 1965 2016 1970 2016 Ireland 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Israel 1965 1995 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Italy 1965 1970 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Japan 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Jordan 1975 1992 2016 1965 2016 1975 2016 1975 2016 1975 2016 1970 2016 1975 2016 Korea 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Luxembourg 1965 1965 2016 1965 2016 1997 2016 1970 2016 1983 2016 1965 2016 1989 2016 Malaysia 1965 1991 2016 1965 2016 1966 2016 1970 2016 1970 2016 1965 2016 1970 2016 Mexico 1965 1980 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Netherlands 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 New Zealand 1965 1965 2016 1965 2016 1965 2016 1977 2016 1977 2016 1965 2016 1977 2016 Norway 1965 1970 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Peru 1965 1980 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Philippines 1965 1981 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Poland 1990 1995 2016 1965 2016 1990 2016 1990 2016 1990 2016 1971 2016 1990 2016 Portugal 1965 1965 2016 1965 2016 1965 2016 1971 2016 1970 2016 1965 2016 1971 2016 Singapore 1965 1975 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 South Africa 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Spain 1965 1970 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Sweden 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Switzerland 1965 1965 2016 1965 2016 1965 2016 1980 2016 1980 2016 1965 2016 1980 2016 Taiwan 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1976 2016 Thailand 1965 1993 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Turkey 1965 1987 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 United Kingdom 1965 1965 2016 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 United States 1965 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Venezuela 1965 1997 2015 1965 2016 1965 2016 1970 2016 1970 2016 1965 2016 1970 2016 Note: Datacoverageforeachofthevariablesincludedinthepanel. TheGDPdataspantheperiodbetweencolumns“first”and“last.” Forsome countries,weextendbackwardtheoriginalquarterlyGDPseries—availablestartingintheyearlistedincolumn“firstq”—usingannualGDPdatathat areavailablefromtheWorldBank’sWorldDevelopmentIndicators. Toconverttheannualdataintoaquarterlyfrequency,weuseDenton’sproportional interpolationmethod(Chen,2007). 23

Table 2: Summary Statistics for the Exposure Measures Advanced Economies Emerging Economies Exposure Variables 5% Median 95% 5% Median 95% Exchange Rate Regime versus Dollar 0 0 1 0 0.85 1 Trade Openness with the U.S., % 1.3 3.5 17.8 1.9 9.8 34.4 Inflation Rate 0.6 3.4 18.3 0.6 7.5 88.2 Current Account Deficit, % of GDP -6.9 0.3 4.9 -8.5 0.5 4.4 Foreign Reserves, % of GDP 0.4 2.3 16.5 0.4 5.1 66.1 External Debt minus Reserves, % of GDP 2.1 29.9 361.6 -31.4 11.4 75.0 Note: All variables computed as 12-quarters moving averages. The exchange rate regime ranges from zero (flexible exchange rate vis-`a-vis the dollar) to one (fixed regime). Trade openness is the sum of nominal merchandise imports and nominal merchandise exports, divided by nominal GDP. The vulnerability index is an equally-weighted average of a logistic transformation of a country’s inflation, current account deficit, foreign reserves (with a negative sign), and external debt. 24

02 51 01 5 0 1965q1 1975q1 1985q1 1995q1 2005q1 2015q1 Figure 1: U.S. Interest Rates Note: Shaded Areas denote interest rate tightenings. 1 Figure 1: The federal funds rate (FFR) from 1965:Q1 through 2016:Q2 Note: The shaded areas denotes periods of interest rate tightenings. A quarter t contains a tightening if it satisfies any of the following criteria: (1) the FFR does not fall in t and rises by at least 20 and 40 basis points in quarters t 1 and t 2, respectively; (2) the FFR does − − not fall by more than 30, 20, and 10 basis points in t, t 1, and t 2, does not fall in t+1, and rises by at least 20 and 30 basis points in − − t+2 and t+3; (3) the FFR rises by at least 100 and 200 basis points in t 3 and t 2, and rises by at least 100 basis points in t+2. − − 25

1. 1978q1 - 1981q2 Tightening 2. 1983q3 - 1984q3 Tightening 3. 1987q2 - 1989q2 Tightening Mexico Brazil Japan M.East & Africa Canada Asia ex China Japan Japan Europe ex U.K. Asia ex China U.S.A. U.K. Brazil Oceania U.S.A. Europe ex U.K. China Mexico Canada Europe ex U.K. Canada Oceania U.K. Oceania U.K. Mexico M.East & Africa Oth.Lat.Am. M.East & Africa Oth.Lat.Am. U.S.A. Oth.Lat.Am. China China Asia ex China Brazil -4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4 4. 1993q4 - 1995q2 Tightening 5. 1999q3 - 2000q3 Tightening 6. 2004q2 - 2006q3 Tightening Asia ex China Europe ex U.K. China Brazil U.K. Oth.Lat.Am. Japan Canada Europe ex U.K. U.S.A. U.S.A. U.K. China Japan Brazil Oth.Lat.Am. Oceania M.East & Africa M.East & Africa Brazil Mexico Europe ex U.K. Mexico Oceania U.K. Oth.Lat.Am. Asia ex China Oceania China U.S.A. Canada Asia ex China Canada Mexico M.East & Africa Japan -4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4 Figure 2: Foreign GDP Growth Relative to Forecast After U.S. Interest Rate Increases Note : Annual GDP growth relative to ARIMA model in the aftermath of U.S. monetary policy tightenings. 2 Figure 2: Foreign GDP Growth Relative to Forecast After U.S. Interest Rate Increases Note: Annual GDP growth surprises (actual minus forecast) in each region relative to ARIMA model in the aftermath of selected U.S. monetary policy tightenings. The bars measure average growth surprises from the beginning of each episode until one year after its end. 26

4 2 0 2 − 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 2010 2014 stnioP egatnecreP Identified U.S. Monetary Shocks Figure 3: Identified U.S. Monetary Shock 3 Figure 3: Identified Monetary Shocks Note: The shocks are calculated as the residuals of a regression of the federal funds rate on contemporaneous and lagged values of inflation, log U.S. GDP, corporate spreads, log foreign GDP, as well as lagged values of the federal funds rate and a quadratic time trend. 27

Response to Monetary Shocks US GDP Fed Funds Rate AFE GDP EME GDP 1 1 1 1 0 0 0 0 1 1 1 1 − − − − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Quarters Figure 4: IRF Response to Monetary Shocks Note: Impulse Response to a U.S. Monetary Shock in the Benchmark Specification. 4 Figure 4: Responses to a Monetary Shock Note: Impulse response to a U.S. monetary shock in the benchmark specification. AFE denotes advanced foreign economies, EME denotes emerging market economies. GDP is in percent deviation from baseline. Federal funds rate is in percentage points. The shaded areas denote 68 percent confidence intervals. 28

AFE Exposure Indexes Dollar Peg Vulnerability Index Trade with U.S. 1 0.75 0.5 0.25 Canada Japan United Kingdom 0 1965 1975 1985 1995 2005 2015 1965 1975 1985 1995 2005 2015 1965 1975 1985 1995 2005 2015 EME Exposure Indexes Dollar Peg Vulnerability Index Trade with U.S. 1 Mexico Korea China 0.75 0.5 0.25 0 1965 1975 1985 1995 2005 2015 1965 1975 1985 1995 2005 2015 1965 1975 1985 1995 2005 2015 Evolution of the Exposure Indexes for six countries Figure 5: Note: The indexes are constructed separately for Advanced and for Emerging Economies. The Vulnerability Index is the first principal component of Inflation, minus GDP growth, and current account deficit. 5 Evolution of the Exposure Indexes for Six Countries Figure 5: Note: The indexes are constructed separately for advanced economies (AFEs) and emerging economies (EMEs). The vulnerability index is described in Table 2. 29

AFE GDP Response by Index Dollar Peg Trade with U.S. Vulnerability Index 0 0.5 − 1 − Median 1.5 High − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters EME GDP Response by Index Dollar Peg Trade with U.S. Vulnerability Index 0 0.5 − 1 − 1.5 Low Exposure − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters GDP Response % to a Monetary Shocks by Index Figure 6: Note: Matteo will add something here 6 GDP Response (in Percent) to Monetary Shock by Index Figure 6: Note: The “median” response is the GDP response (in percent) of an economy with values for each index equal to the median value, as reported in Table 2. The “high” response is the response of an economy with values for each index equal to the 95th percentile, as reported in Table 2. The shaded areas denote 68 percent confidence intervals. 30

AFE GDP Response by Index Dollar Peg Trade with U.S. Current Account (-) Reserves (-) Inflation External Debt 0 0.5 − 1 Median − 1.5 High − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Quarters Quarters Quarters EME GDP Response by Index Dollar Peg Trade with U.S. Current Account (-) Reserves (-) Inflation External Debt 0 0.5 − 1 Median − 1.5 High − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Quarters Quarters Quarters GDP Response % to a Monetary Shocks by Index Figure 7: Note: Matteo will add something here 7 Figure 7: GDP Response (in Percent) to Monetary Shock for Each Component of the Index. Note: The shaded areas denote 68 percent confidence intervals. 31

AFE Exchange Rate Response by Index Dollar Peg Trade with U.S. Vulnerability Index Median 2 High 0 2 − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters EME Exchange Rate Response by Index Dollar Peg Trade with U.S. Vulnerability Index 2 0 Low Exposure 2 − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Real Exchange Rate Response % to a Monetary Shocks by Index Figure 9: Note: Response of Foreign Real Exchange Rate Indexes to a 100 basis points increase in U.S. interest rates. Higher values indicate an appreciation of the real exchange rate. 9 Exchange Rate Response (in Percent) to Monetary Shock by Index Figure 8: Note: The “median” response is the response of the real exchange rate for an economy with values for each index equal to the median value, as reported in Table 2. The “high” response is the response of an economy with values for each index equal to the 95th percentile, as reported in Table 2. Higher values indicate an appreciation. The shaded areas denote 68 percent confidence intervals. 32

AFE Interest Rate Response by Index Dollar Peg Trade with U.S. Vulnerability Index 1.5 Median High 0.75 0 0.75 − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters EME Interest Rate Response by Index Dollar Peg Trade with U.S. Vulnerability Index 1.5 0.75 0 Low Exposure 0.75 − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Interest Rate Response (Percentage Points) to a Monetary Shocks by Index Figure 10: Note: Response of foreign interest rates to a 100 basis points surprise increase in the U.S. Interest Rate. 10 Interest Rate Response (Percentage Points) to Monetary Shock by Index Figure 9: Note: The “median” response is the short-term interest rate response of an economy with values for each index equal to the median value, as reported in Table 2. The “high” response is the response of an economy with values for each index equal to the 95th percentile, as reported in Table 2. The shaded areas denote 68 percent confidence intervals and are based on Newey-West standard errors that account for serial correlation. 33

AFE GDP Response by Index (Demand Shock) Dollar Peg Trade with U.S. Vulnerability Index 1 1 1 0 0 0 Median 1 High 1 1 − − − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters EME GDP Response by Index (Demand Shock) Dollar Peg Trade with U.S. Vulnerability Index 1 1 1 0 0 0 1 1 1 − − − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters GDP Response % to a Demand Shocks by Index Figure 11: Note: Matteo will add something here 11 GDP Response (in Percent) to a Demand Shock by Index Figure 10: Note: The “Median” response is the GDP response of an economy with values for each index equal to the median value, as reported in Table 2. The “High” response is the response of an economy with values for each index equal to the 95th percentile, as reported in Table 2. The shaded areas denote 68 percent confidence intervals and are based on Newey-West standard errors that account for serial correlation. 34

GDP Response period 1965-1985 US AFE EME 1 0 1 − exposure by period 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters GDP Response period 1986-2016 US AFE EME 1 0 1 − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters GDP Response % to a Monetary Shocks by Index Figure 9: Note: Taylor Rule computed by sample 9 GDP Response (in Percent) to a Monetary Shock across Subsamples Figure 11: Note: The samples cover the periods 1965:Q1–1985:Q4 and 1986:Q1-2016:Q2 respectively. The shaded areas denote 68 percent confidence intervals. 35

Impulse Responses: Romer & Romer Shocks US GDP Fed Funds Rate AFE GDP EME GDP 1 1 1 Benchmark 1 0 0 0 0 1 1 1 1 − − − − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Quarters Impulse Responses: Excluding ZLB period US GDP Fed Funds Rate AFE GDP EME GDP 1 1 1 1 0 0 0 0 1 1 1 1 − − − − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Quarters Impulse Responses: Linearly Interpolated GDP Data US GDP Fed Funds Rate AFE GDP EME GDP 1 1 1 1 0 0 0 0 1 1 1 1 − − − − 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20 Quarters Quarters Quarters Quarters Figure 12: Impulse Responses to Monetary Shocks: Robustness Note: Matteo will add something right here 12 Figure 12: Impulse Responses to Monetary Shocks: Robustness Note: GDP responses are in percent from baseline. The fed funds rate response is in percentage points. The shaded areas denote 68 percent confidence intervals. 36

Canada Japan 8 8 6 6 e e g g n 4 n 4 a a h h c c t 2 t 2 n n e e c 0 c 0 r r e e p p r-2 r-2 e GDP e t t r r a-4 Median Effect a-4 u u q Dollar Peg q - - 4-6 Trade w U.S. 4-6 Vulnerability Index -8 -8 1975 1980 1985 1990 1995 2000 2005 2010 2015 1975 1980 1985 1990 1995 2000 2005 2010 2015 Mexico Korea 15 15 e e g10 g10 n n a a h h c c t t n 5 n 5 e e c c r r e e p p r e 0 r e 0 t t r r a a u u q q - - 4 -5 4 -5 1975 1980 1985 1990 1995 2000 2005 2010 2015 1975 1980 1985 1990 1995 2000 2005 2010 2015 Figure 13: Historical Contribution of U.S. Monetary Policy Shocks Note: The green bar is the “median” effect that is common to all advanced economies or emerging economies. The blue, cyan, and red bars are the marginal effects of the exchange rate, trade, and vulnerability channels, respectively. 37

A Related Empirical Literature Several papers have examined the global implications of changes in U.S. interest rates. Examples include the following: 1. Kim (2001) uses structural VARs to measure the effects of U.S. monetary shocks for six advanced economies. He finds that U.S. monetary tightenings lead to a decrease in activity abroad, which appears mostly driven by an increase in the world real interest rate rather than by trade channels. 2. Canova (2005), using few years of data and multiple VARs, estimates the effects of U.S. monetary policy shocks on emerging markets in Latin America. He finds that a U.S. monetary policy shock affects the interest rates in Latin America quickly and strongly. 3. Dedola et al. (2017) find that countries with lower capital mobility and a floating exchange rate regime are better insulated from the financial repercussions of U.S. monetary policy. 4. Ehrmann and Fratzscher (2005) study the transmission of U.S. monetary policy abroad but focus on financial variables. 5. Ma´ckowiak (2007) uses estimated structural VARs and finds that output in a typical emerging market economy responds to U.S. monetary policy shocks by more than the output in the United States itself. 6. DiGiovanniandShambaugh(2008)findthathighforeigninterestrateshaveacontractionary effect on real GDP in the domestic economy, but that this effect is centered on countries with fixed exchange rates. 7. Georgiadis (2016)’s recent study is closely related to ours. He first projects estimated U.S. monetarypolicyshocksfromaglobalVARmodelagainstGDPinalargenumberofcountries. Hethenestimatesthedeterminantsofspilloversusingindicatorsthatareassumedtobefixed in a given country over time. This assumption may lead to spurious results as a country’s position might change over time. He finds that the magnitude of spillovers depends on the receiving country’s trade and financial integration, de jure financial openness, exchange rate regime, financial market development, labor market rigidities, industry structure, and participation in global value chains. 38

B Data Sources GDP. We collect data from the country’s national statistical offices or the central bank • through Haver (databases G10+ and EMERGE). GDP in each country is real GDP, constructed using either chain-weighting or dividing nominal GDP by its deflator. Inflation. The source is a country’s national statistical office via Haver. • Current account, external debt (“Debt Liabilities, Stock”), and foreign reserves • (“FX Reserves minus Gold”). The source is the External Wealth of Nations Mark II database (Lane and Milesi-Ferretti, 2017). All variables are in current U.S. dollars and are divided by GDP in current dollars. Interest rates. Figure 9 shows the response of foreign short-term nominal interest rates to • a U.S. monetary shock. We obtain foreign interest rates via either Haver or the OECD. Our interest rate measure is one of the following: (1) the central bank policy rate from the IMF InternationalFinancialStatistics(IFS);(2)theTreasurybillrate(IFS);(3)the discount rate (IFS); (4) the short-term interest rate (OECD); (5) the overnight interest rate (OECD); (6) The lending Rate (IFS). We use measure (1), and move to measure (2) if (1) is not available, to (3) if (1) and (2) are not available, and so on. This procedure allows us to assemble data for 48 of the 50 countries in our panel. We drop observations where the interest rate exceeds 50 percent. The procedure yields a panel of 9,018 country-quarter observations. Exchange rates. Real effective exchange rates are taken from the dataset described in • Darvas (2012). We drop the observations where the year-on-year change in the real exchange rate is larger than 50 percent in absolute value. Most of the data begin in 1970. The sample contains 8,116 country-quarter observations. Trade with the United States. These data are the merchandise trade data from the IMF • Direction of Trade Statistics. 39