International Financial Spillovers to Emerging Market Economies: How Important Are Economic Fundamentals?

K.7 International Financial Spillovers to Emerging Market Economies: How Important Are Economic Fundamentals? Ahmed Shaghil, Brahima Coulibaly, and Andrei Zlate Please cite paper as: Ahmed Shaghil, Brahima Coulibaly, and Andrei Zlate (2015). International Financial Spillovers to Emerging Market Economies: How Important Are Economic Fundamentals? International Finance Discussion Papers 1135. http://dx.doi.org/10.17016/IFDP.2015.1135 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1135 April 2015

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1135 April 2015 International Financial Spillovers to Emerging Market Economies: How Important Are Economic Fundamentals? Shaghil Ahmed, Brahima Coulibaly, Andrei Zlate* 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 References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.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.

International Financial Spillovers to Emerging Market Economies: How Important Are Economic Fundamentals? Shaghil Ahmed, Brahima Coulibaly, Andrei Zlate* April 2015 Abstract We assess the importance of economic fundamentals in the transmission of international shocks to financial markets in various emerging market economies (EMEs). Our analysis covers the socalled taper-tantrum episode of 2013 and six earlier episodes of severe EME-wide financial stress since the mid-1990s. Cross-country regressions lead us to the following results: (1) EMEs with relatively better economic fundamentals suffered less deterioration in financial markets during the 2013 taper-tantrum episode. (2) Differentiation among EMEs set in quite early and persisted throughout this episode. (3) Controlling for economic fundamentals, we also find that, during the taper tantrum, financial conditions deteriorated more in those EMEs that had earlier experienced larger private capital inflows and greater exchange rate appreciation. (4) For earlier episodes, we find little evidence of investor differentiation across EMEs being explained by differences in their relative vulnerabilities during EME crises of the 1990s and early 2000s. (5) That said, differentiation across EMEs based on fundamentals does not appear to be unique to the 2013 episode. Differences in economic fundamentals played a role in explaining the heterogeneous EME financial market responses during the global financial crisis of 2008, and the role of fundamentals appeared to progressively increase through the European crisis in 2011 and subsequently the 2013 taper tantrum. Keywords: Emerging market economies, financial spillovers, economic fundamentals, vulnerability, depreciation pressure, taper tantrum, financial stress. JEL classifications: E5, F3. *Shaghil Ahmed (shaghil.ahmed@frb.gov) and Brahima Coulibaly (brahima.coulibaly@frb.gov) are economists in the Division of International Finance, Board of Governors of the Federal Reserve System, Washington, D.C. 20551 U.S.A. Andrei Zlate (andrei.zlate@bos.frb.org) is a financial economist in the Department of Supervision, Regulation, and Credit at the Federal Reserve Bank of Boston, 600 Atlantic Avenue, Boston, MA 02210, U.S.A. We thank Steve Kamin, Prachi Mishra, Patrice Robitaille, as well as participants at the American Economic Association 2015 annual meetings and the Federal Reserve Board conference on “Spillovers from Accommodative Monetary Policies since the GFC” for very useful comments. We also thank Zina Saijid and Julio Ortiz for outstanding research assistantship. The views expressed in the paper are those of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or any other person associated with the Federal Reserve System.

1 Introduction Starting in May 2013, on news that the Federal Reserve could soon start tapering its large scale asset purchases (LSAPs), (cid:133)nancial conditions in emerging market economies (EMEs) deteriorated sharply. Investors withdrew capital, currencies depreciated, stock markets fell, and bond yields and premiums on credit default swaps rose. This so-called taper-tantrum episode sparked debate on how foreign economies may be a⁄ected(cid:151)and which economies, especially among the EMEs, may experience the most adverse e⁄ects(cid:151)once the process of U.S. monetary policy normalization begins. Indeed, one intriguing feature of the (cid:147)risk-o⁄(cid:148)episode during the summer and fall of 2013 was that it did not have a similar e⁄ect on all EMEs. While some countries experienced acute deteriorations in (cid:133)nancial conditions, others were much less a⁄ected. The varied experiences of the di⁄erent EMEs have spawned research on whether the heterogeneous response of EME (cid:133)nancial markets during the taper tantrum can be explained by di⁄erences in economic fundamentals across theseeconomies. (See, forexample: Prachietal., 2014; EichengreenandGupta, 2014; Aizenmanet al., 2014; and Sahay et al., 2014.) And, these experiences have also focused attention on whether, more broadly, the e⁄ects of U.S. monetary policy shocks on EMEs over a longer period have been related to these economies(cid:146)own vulnerabilities (Bowman et al., 2014). Inthispaper,weseektocontributetothissparsebutgrowingliterature. Speci(cid:133)cally,weaddress three questions: First, to what extent did investors di⁄erentiate among the EMEs based on their economic fundamentals during the (cid:147)risk-o⁄(cid:148)episode of 2013? Or, can the heterogeneous responses across EMEs be explained by other factors, such as whether those economies that initially received the heaviest capital in(cid:135)ows were also the ones from which investors receded the most during the episode, a possibility raised by some researchers such as Aizenman et al. (2014). Second, during the taper-tantrum stress episode, when exactly did di⁄erentiation across EMEs begin and how persistent was it? Did investors initially display herd behavior and pull back indiscriminately from all EMEs, and then only begin di⁄erentiating over time as the shock persisted? Or did they discriminate across EMEs from the early stages of the episode? Third, is the di⁄erentiation across EMEs by investors a relatively recent phenomenon, or does the application of our methodology suggest that investors have always distinguished among EMEs to some extent according to their economic fundamentals? 1

Research to date has not appeared to reach a consensus on the (cid:133)rst question(cid:151)the importance of fundamentals in explaining the heterogeneous EME responses. Using an event study approach, Prachi et al. (2014) analyze the reaction of (cid:133)nancial markets to the Federal Reserve(cid:146)s LSAPs tapering announcements in 2013 and 2014, including the episode that we examine in this study. They (cid:133)nd evidence of market di⁄erentiation among EMEs based on macroeconomic fundamentals. In addition, they (cid:133)nd that EMEs with deeper (cid:133)nancial markets and tighter macroprudential policy prior to the stress period experienced relatively less deterioration in (cid:133)nancial conditions. In contrast, Eichengreen and Gupta (2014) do not (cid:133)nd evidence that better macroeconomic fundamentals(cid:151)such as a lower budget de(cid:133)cit, lower public debt, a higher level of international reserves, and higher economic growth(cid:151)provided any insulation. They do (cid:133)nd, however, that EMEs whose exchange rates appreciated more earlier and whose current account de(cid:133)cits widened more experienced a larger e⁄ect. But they conclude that the heterogeneous reactions most importantly owedtothesizeofeachcountry(cid:146)s(cid:133)nancialmarket; largermarketsexperiencedmorepressure, asinvestors were better able to rebalance their portfolios in these EMEs with relatively large and liquid (cid:133)nancial markets. Similarly, Aizenman et al. (2014) (cid:133)nd no evidence that stronger macroeconomic fundamentals helped the EMEs weather the taper tantrum better, but their study focuses only on the very short-term responses of (cid:133)nancial indicators after taper news. Classifying EMEs into two categories ((cid:147)fragile(cid:148)and (cid:147)robust(cid:148)) based on their current account balances, levels of international reserves, and external debt, they (cid:133)nd that news of tapering from Fed Chairman Bernanke was associated with sharper deterioration of (cid:133)nancial conditions in the (cid:147)robust(cid:148)EMEs compared with the (cid:147)fragile(cid:148)ones in the very short term (24 hours following the announcement). The authors rationalize their results by conjecturing that EMEs with stronger fundamentals received more capital in(cid:135)ows during the expansionary phase of the Federal Reserve(cid:146)s conventional and unconventional monetary policy and accordingly experienced sharper capital out(cid:135)ows and deteriorations in (cid:133)nancial conditions on news of an impending tapering. This hypothesis is not explicitly tested in their study, and the authors acknowledge that, over a longer window than they considered in their event study, the fragile EMEs may have su⁄ered more than robust EMEs after the taper news. We approach the (cid:133)rst question in a somewhat di⁄erent manner from these previous studies. Rather than looking at market reactions on days when news may be coming in about the Federal Reserve(cid:146)s tapering of asset purchases as Prachi et al. (2014) do, we treat the taper tantrum as a single episode, de(cid:133)ned by the observed (cid:147)peak-to-trough(cid:148)behavior of (cid:133)nancial markets surrounding 2

theperiodwhenconcernsabouttaperingcametotheforefront. AndunlikeAizenmanetal.(2014), but consistent with Prachi et al. (2014), we exploit (cid:133)ner cross-section di⁄erences in vulnerabilities across the EMEs rather than divide the EMEs into two groups based on the behavior of only three variables. Using a sample of 20 emerging market economies and cross-section regression analysis, we assess the role of economic fundamentals in the heterogeneous cumulative performance of EME (cid:133)nancial markets over the whole episode from May 2013 through August 2013. We (cid:133)nd that EMEs that had relatively better fundamentals to begin with(cid:151)as measured by a host of individual variables capturing vulnerability as well as an aggregate index of relative vulnerabilitiesacrossEMEsthatweconstruct(cid:151)su⁄eredlessdeteriorationduringthetaper-tantrum episode as measured by a broad range of (cid:133)nancial variables, including exchange rate, depreciation pressure,andgovernmentbondyields,aswellasEMBIandCDSspreads. Ourresultsareconsistent withthoseinPrachietal.(2014),butcontrastwiththoseinAizenmanetal.(2014)andEichengreen andGupta(2014). Wealso(cid:133)ndthat(cid:133)nancialconditionsdeterioratedmoreinthoseEMEsthathad earlier experienced larger private capital in(cid:135)ows and exchange rate appreciations, consistent with the (cid:147)more-in-more-out(cid:148)hypothesis that EMEs that experienced larger in(cid:135)ows prior to the taper tantrum also experienced larger out(cid:135)ows once the episode began. Also, we (cid:133)nd some evidence that the structure of the (cid:133)nancial markets(cid:151)such as the size of the capital market, the level of foreign investor participation, the extent of capital account openness, or revisions to the growth outlook(cid:151)shaped the heterogeneous responses of some (cid:133)nancial indicators, as in Eichengreen and Gupta(2014). However,thestrengthofeconomicfundamentalsexplainsmuchmoreofthevariation in responses across EMEs compared with other factors, such as the previous runup in (cid:135)ows or the structure of (cid:133)nancial markets. Onthesecondquestion,relatedtothetimingandpersistenceofdi⁄erentiation,theconventional wisdom seems to be that early in the taper tantrum, investors did not di⁄erentiate much among EMEsaccordingtotheirvulnerabilities,butsuchdi⁄erencesemergedwhenconcernsabouttapering persisted. (See,forexample,Sahayetal.,2014). However,ourevidencesuggeststhatdi⁄erentiation according to relative vulnerabilities set in relatively early during the stress episode and persisted for some time, although the extent of di⁄erentiation was less pronounced in the (cid:133)rst few weeks. Turning to the third question of the extent to which investors also discriminated among EMEs inthepast, weuseourmethodologytoidentifysixpreviousepisodesofsevere(cid:133)nancialstressinthe EMEs over the past 20 years: the European sovereign debt crisis (2011), the global (cid:133)nancial crisis 3

(GFC, 2008-09), the Argentine crisis (2002), the Russian crisis (1998), the Asian crisis (1997-98), and the Mexican crisis (1994-95). For each of these episodes, we examine the role of economic fundamentals in driving any heterogeneous cumulative reaction of asset prices over the full episode across EMEs. We (cid:133)nd little evidence of di⁄erentiation based on economic fundamentals in the 1990s and the early 2000s, consistent with literature on herding behavior of investors toward EMEs and contagion during that period, such as Dornbusch et al. (2000) and Calvo and Mendoza (2001). However, the results also suggest that di⁄erentiation was not unique to the taper tantrum in 2013. In fact, it appears that fundamentals played a role in explaining the heterogeneous reaction of EME asset prices during the GFC in 2008, and that their role increased during the European sovereign crisis in 2011, and then increased further during the taper tantrum in mid-2013. These results that di⁄erentiation has occurred among EMEs since the GFC are consistent with those from a complementary approach taken in Bowman et al. (2014): identifying monetary policyrelated shocks to U.S. interest rates, rather than episodes of severe (cid:133)nancial stress as we do, they (cid:133)nd that the e⁄ect of such shocks on EME interest rates have been larger in the relatively more vulnerable EMEs in the recent past as well, and that there is no di⁄erence in this respect between the e⁄ects of conventional and unconventional U.S. monetary policy. Taken at face value, our results suggest that international investors may have moved from herd behaviorinthepast(cid:151)suchasduringthe1990sandtheearly2000s(cid:151)toprogressivelydi⁄erentiating more and more among EMEs according to their economic fundamentals through the GFC, the European debt crisis, and the taper tantrum. It remains an open question, however, whether our results are perhaps driven by di⁄erent sources of the shock. The stress episodes since the GFC, during which we (cid:133)nd evidence of di⁄erentiation among EMEs, emanated primarily from the advanced economies, whereas the crises of the 1990s and 2000s, during which we (cid:133)nd little evidence of di⁄erentiation, emanated to a much larger extent(cid:151)although not exclusively so(cid:151)from shocks originating in the EMEs themselves. The remainder of the study is organized as follows: In section 2, we describe the data and the estimation strategy. Then, section 3 presents our results of di⁄erentiation across EMEs during the 2013 taper-tantrum episode, and section 4 shows the results for di⁄erentiation in previous well-recognized EME (cid:133)nancial stress episodes. section 5 concludes. 4

2 Econometric speci(cid:133)cation and choice of variables For each EME stress episode identi(cid:133)ed, we regress performance in selected (cid:133)nancial markets across EMEsoverthedurationofthestressperiodonasetofvariablescapturingeconomicconditionsthat existed prior to the beginning of the stress episode and other control variables. Our cross-section regression is represented by the following equation: (cid:1)FinVar = c+ (cid:12) X +" (1) i;k i;j i;j i j X Note that each i denotes a particular country, and each k denotes a particular (cid:133)nancial variable with its performance over the stress episode represented by (cid:1). X are a set of explanatory i;j variables j speci(cid:133)c to country i measured in the year prior to the onset of the stress period, (cid:12)(cid:146)s are parameters to be estimated, and "(cid:146)s are error terms. Data availability across countries and time limits our sample to 20 EMEs.1 Note that the cross-section observations in each regression are the countries, and a separate regression is run for each dependent variable k and each sub-set of explanatory variables j. Among the dependent variables measuring (cid:133)nancial performance, we consider the percent change in the country(cid:146)s bilateral nominal exchange rate against the dollar, the change in the local currency bond yields on 10-year government bonds, the percent change in the stock market index, and the change in EMBI and CDS spreads between the peak and trough of each episode (for example, from April to August for the 2013 episode) using monthly data.2 Pressures on (cid:133)nancial markets may not necessarily be re(cid:135)ected in the exchange rate if the central bank intervenes in foreign exchange markets to prop up the currency or raises the policy rate to contain capital (cid:135)ight. For this reason, we also consider a depreciation pressure index that is a weighted average of the change in the exchange rate and the foreign exchange reserves, following Eichengreen et al. (1995), with the weights given by the inverse of the standard deviation of each (cid:133)nancial indicator measured in the cross section.3 1 Argentina, Brazil, China, Chile, Colombia, Indonesia, India, South Korea, Malaysia, Mexico, Peru, Paraguay, the Philippines, Pakistan, Russia, South Africa, Taiwan, Thailand, Turkey, and Uruguay. Although data for some countrieswereavailable,weexcludeeuro-areacountries,dollarizedeconomies,andtheEMEswithhardpegexchange rate regimes. 2 The monthly data re(cid:135)ect end-of-month values for exchange rates, foreign exchange reserves, and policy interest rates, and average monthly values for bond yields, stock market indices, and EMBI and CDS spreads. 3 Alternatively,we also constructed a depreciation pressureindex based on the changesin exchangerates,foreign exchange reserves, and the policy rates between April and August 2013. Since the two indexes behave similarly during thetaper-tantrum episode,ourresultsaremostly based on thestandard depreciation index based on changes 5

We consider several di⁄erent types of independent variables: those that might be capturing the macroeconomic fundamentals and policy choices of a country (category 1), those that might help identify how much capital might have been (cid:135)owing in prior to the episode so we can test the (cid:147)more-in-more out(cid:148)hypothesis (category 2), and those that might be capturing aspects of a country(cid:146)s (cid:133)nancial structure such as openness and (cid:133)nancial development (category 3), which are included as other control variables. Potentialcandidatesforcategory1includethefollowingsixvariablesre(cid:135)ectingthestrength ofmacroeconomicfundamentals: thecurrentaccountbalanceasapercentofGDP;foreignexchange reserves as a percent of GDP; short-term external debt as a percent of foreign exchange reserves; the gross government debt as a percent of GDP; the average annual in(cid:135)ation over the past three years; and the runup in bank credit to the private sector, measured as the change in the ratio of bank credit to GDP over the (cid:133)ve years prior. To mitigate concerns over endogeneity, each of these is taken as of the year prior to the one in which the episode occurs. For the 2013 event, we also use the revisions to the outlook for economic activity as an additional variable capturing fundamentals since analysts were, at least anecdotally, pointing to this as an important factor. The revisions to the outlook are measured as the change between January and May of 2013 in the outlook for annual average real GDP growth for 2013 provided by Consensus Economics. Finally, we account for the monetary policy and exchange rate frameworks through two indicator variables with the (cid:133)rst variable equal to 1 for countries operating under an in(cid:135)ation targeting regime and 0 if not, and the second variable equal to 1 for countries with (cid:135)oating exchange rate regimes based on the classi(cid:133)cation in the IMF(cid:146)s Annual Report on Exchange Rate Arrangements and Exchange Restrictions, and 0 otherwise. Given the relatively small cross-section, not all of these variables can be included simultaneously; we consider individual variables one at a time and sub-sets of them. We also compute an aggregate index of relative vulnerabilities across the cross-section using some ofthesevariablesinlightofthelimitednumberofobservations; thisvulnerabilityindexisdescribed in detail in section 3.2.2. For category 2, we include the following variables: the cumulative gross capital in(cid:135)ows over the past three years as a percent of GDP and the appreciation of the real e⁄ective exchange over the past three years. For category 3 variables, we consider the capitalization of the domestic equity market as a share of GDP; foreign participation in the domestic equity market, measured as the in exchange rates and reserves only. 6

share of foreign equity holdings in the domestic equity market; and capital account openness, as measured by the Chinn-Ito index, a de jure measure of (cid:133)nancial openness initially introduced in Chinn and Ito (2008) and subsequently updated by the authors through 2011. For each of these indicators, the data sources are presented in table 1. 3 Di⁄erentiation across EMEs during the 2013 taper-tantrum episode This section provides an overview of the taper-tantrum episode of 2013, presents new evidence on the drivers of EME di⁄erentiation for the entire duration of the episode, and discusses the timing and persistence of di⁄erentiation within sub-intervals of the episode itself. 3.1 Overview of the episode Market participants shifted their expectation for the Federal Reserve(cid:146)s asset purchases program following former Chairman Bernanke(cid:146)s testimony to the Congress on May 22. The testimony heightened perceptions that the Federal Reserve would soon begin tapering its LSAPs and thus lessentheamountofmonetarystimulusitwasputtingintotheeconomythroughitsunconventional monetary policies. This shift in market expectations led to sharp movements in U.S. and global (cid:133)nancial markets, including a large sell-o⁄of EME assets by international investors, causing large depreciations of currencies, increases in bond yields, EMBI spreads, and CDS spreads, as well as declines in equity markets. As shown in table 1, for the median EME in our sample, the dollar exchange rate depreciated 9 percent, sovereign bond yields rose almost 2 percentage points, the EMBI and CDS spreads each rose 0.5 percentage point, and the stock market fell 5 percent from April to August. In response, central banks in some EMEs intervened in the foreign exchange markets to curb depreciation pressures, which resulted in losses of foreign exchange reserves, and also, in some cases, raised policy rates to discourage capital (cid:135)ight. The stress episode persisted for much of the summer, until the Federal Reserve chose not to reduce the size of its asset purchases at the September 2013 Federal Open Market Committee meeting. Thus, we date the taper tantrum episode from May to August and compute the change in (cid:133)nancial indicators relative to April, consistent with the literature. Althoughmost(cid:133)nancialmarketsinEMEswerenegativelya⁄ectedduringthetapertantrum, there was wide dispersion in the (cid:133)nancial performance across EMEs, as shown by the histograms 7

in (cid:133)gure 1. As can be seen in panel 1, although most EME currencies depreciated between 5 and 10 percent, the extent of depreciation did vary quite a bit, with some hardly depreciating at all, while others depreciated more than 15 percent. The use of foreign exchange rate intervention and the response of policy rates also appeared to vary across EMEs (panels 2 and 3, respectively). The heterogeneous response is also apparent in the exchange rate pressure index that combines changes in the exchange rate and foreign exchange reserves (panel 4). Similarly, the increases in bond yields, EMBI spreads, and CDS spreads varied quite a bit across EMEs, as did the declines in equity markets (panels 6-9). With U.S. interest rates expected to rise, some e⁄ect on the EMEs was to be expected for several reasons. For example, rising U.S. rates encourage investors to shift their holdings toward U.S.assets. Atthesametime,investorsmaywithdrawfundsduringrisk-o⁄episodesespeciallyfrom those EMEs perceived to be more risky. Indeed, although the movements in most EME (cid:133)nancial markets were sizable, investors did not appear to withdraw funds indiscriminately from all EMEs en masse, as shown in (cid:133)gure 1. Therefore, we next turn to the question of whether the varied response of EME (cid:133)nancial markets during the taper tantrum episode was systematically related to their economic fundamentals. 3.2 Di⁄erentiation across EMEs during the 2013 stress episode as a whole Asa(cid:133)rstcuttoexaminingthisissue,wepresentsomescatterplotscapturingbivariaterelationships. Figure 2 plots exchange rate depreciations across EMEs during the episode (on the vertical axis) against six potential variables (on the horizontal axis in the di⁄erent panels) that investors may care about when assessing a country(cid:146)s macroeconomic fundamentals. The (cid:133)gure suggests that the EMEs whose currencies depreciated by more also had larger current account de(cid:133)cits as a share of GDP(panel1), lessreserves-to-GDP(panel2), moreshort-termexternaldebtasashareofreserves (panel 3), higher gross government debt ratios (panel 4), and higher average in(cid:135)ation over the past few years (panel 5). The only variable among the ones we show that did not seem to a⁄ect (cid:133)nancial market performance across EMEs is the runup in bank credit to the private sector as a ratio to GDP (panel 6). Figure3doesasimilarexerciseoflinkingindividualmacroeconomicvariablestotheperformance of bond markets, speci(cid:133)cally the rise in local currency bond yields during the taper tantrum. The results obtained are generally consistent with those found in (cid:133)gure 2. Bond yield increases were 8

greater in economies with greater current account de(cid:133)cits, less reserves, higher short-term debt, and higher average in(cid:135)ation. However, the gross government debt position and, again, the runup in credit to the private sector were not correlated with changes in bond yields. To further understand the drivers of the heterogeneous performance of EME (cid:133)nancial markets, we estimate equation (1) using explanatory variables from the three categories described in section 2, that is, (cid:133)rst, those describing the strength of macroeconomic fundamentals, the growth outlook, and the policy framework; second, those relevant for the (cid:147)more-in-more-out(cid:148)hypothesis; and third, those describing the structure of (cid:133)nancial markets. 3.2.1 The role of macroeconomic fundamentals We (cid:133)nd that macroeconomic fundamentals were important drivers of the heterogeneous performance of EME (cid:133)nancial markets during the taper tantrum episode, as was suggested by the scatter plots shown earlier. In tables 2 through 4, four of the six variables considered to describe the strength of macroeconomic fundamentals (that is, current account de(cid:133)cits, short-term external debt, government debt, and credit growth) are statistically signi(cid:133)cant in explaining the behavior of EME (cid:133)nancial indicators. As shown in table 2 (column 1), the EMEs with larger current account de(cid:133)cits and higher gross government debt su⁄ered more severe currency depreciations relative to the dollar. All else equal, a 1percentage-pointhighercurrentaccountde(cid:133)cit(orstockofgovernmentdebt)relativetoGDPwas associated with 0.8 percentage point (or 0.2 percentage point) greater depreciation. Similarly, in table3(columns1and7), theEMEswithgreatershort-termexternaldebtsu⁄eredlargerincreases in bond yields, while the EMEs with faster previous runups in bank credit to the private sector su⁄ered greater declines in equity prices during the episode; an additional 10 percentage points in short-term external debt relative to reserves implied 20 basis points larger increases in bond yields and an additional 1 percentage-point previous increase in bank credit relative to GDP led to 0.4 percentage point greater declines in equity prices during the event. Unsurprisingly, notallofthesixmacroeconomicvariablesarestatisticallysigni(cid:133)cantinexplaining the behavior of each (cid:133)nancial indicator. Some macroeconomic characteristics should be more relevant for certain types of (cid:133)nancial assets than others. In addition, while the coe¢ cients tend to have the expected signs, their statistical signi(cid:133)cance is likely constrained by the limited degrees of freedominmultivariatecross-sectionalregressionswitharelativelysmallsampleandbysomelikely 9

interdependence between the variables. For example, although the scatter plots in (cid:133)gure 2 suggest that (cid:133)ve of the six macroeconomic variables a⁄ected the extent of exchange rate depreciation during the stress episode, only two of these variables (the current account and government debt) are statistically signi(cid:133)cant in the multivariate regression in table 2 (column 1). Also, while the current account de(cid:133)cit is statistically signi(cid:133)cant in bivariate regressions for the increase in EMBI and CDS spreads (not shown), it is not statistically signi(cid:133)cant in the multivariate regressions in table 4 (columns 1 and 7). However,giventhateachofthesixmacroeconomicvariableshasaroleinexplainingsomeaspect of (cid:133)nancial performance(cid:151)in either bivariate or multivariate regressions(cid:151)in what follows we will consider them jointly in our baseline assessment of the link between macroeconomic fundamentals and di⁄erentiation across EMEs (cid:133)nancial markets during the taper-tantrum episode. 3.2.2 The vulnerability index To better understand the determinants of (cid:133)nancial market performance, we build an index of EME vulnerabilities that summarizes the relative strength of EMEs(cid:146)macroeconomic fundamentals based on the six variables discussed in section 2, given their importance in shaping the (cid:133)nancial market responses. The vulnerability index has the advantage of aggregating the information embedded in multiple macroeconomic variables while addressing the problem of limited degrees of freedom in our multivariate regressions. Toconstructtheindex, we(cid:133)rstranktheEMEsrelativetoeachotheraccordingtoeachvariable, from lowest vulnerability to highest vulnerability. We then take the average ranking of a country across the six variables to obtain the value of the index for each EME, with higher values representing higher vulnerability. It is important to note that this index represents relative vulnerabilities across EMEs at a point in time and not the absolute levels of vulnerabilities. Simple estimates from univariate regressions show positive and statistically signi(cid:133)cant relationships between the vulnerability index as the sole explanatory variable and, alternatively, as dependent variable, the exchange rate depreciation (table 2, column 2), changes in the depreciation pressure index (table 2, column 8), and the increase in local currency bond yields (table 3, column 2). Notably, there are also positive and statistically signi(cid:133)cant links between the vulnerability index and the increases in the EMBI spreads and CDS spreads (table 4, columns 2 and 8), even though the individual explanatory variables that comprise the index were not statistically signi(cid:133)- 10

cant as determinants of (cid:133)nancial performance. However, we do not (cid:133)nd a statistically signi(cid:133)cant link between the vulnerability index and changes in stock prices (table 3, column 8). 3.2.3 The role of other variables While using the EME vulnerability index to summarize the strength of macroeconomic fundamentals, we also explore the role of additional variables in explaining the heterogeneous response of EME (cid:133)nancial markets during the taper-tantrum episode.4 First, we (cid:133)nd some evidence that the performance of EME (cid:133)nancial markets was linked to changes in EMEs(cid:146)growth outlook prior to the episode, which recall is measured as revisions to the Consensusgrowthoutlookfor2013donebetweenJanuaryandMayofthesameyear. However, this evidence is limited in that it applies only to bond yields and EMBI spreads (see column 3 of tables 3 and 4). In addition, we (cid:133)nd very limited evidence that the EMEs with in(cid:135)ation targeting regimes or(cid:135)oatingexchangerateregimessu⁄eredlessdeteriorationin(cid:133)nancialmarketsthanotherwise(see tables 2 through 4, columns 3 and 9). Second, controlling for the in(cid:135)uence of macroeconomic fundamentals, we also (cid:133)nd evidence to support the (cid:147)more-in-more-out(cid:148)hypothesis for exchange rates and bond yields(cid:150)that is, the EMEs that received more capital in(cid:135)ows or experienced more exchange rate appreciation prior to the episode su⁄ered greater deteriorations in (cid:133)nancial conditions during the episode itself. As shown in table 2, countries that received more private gross in(cid:135)ows (column 4) or had more real e⁄ective appreciation previously (column 5) also experienced more depreciation during the episode. In table 3(column5),countriesthatexperiencedmoreappreciationpreviouslyalsosu⁄eredgreaterincreases in bond yields during the episode.5 All in all, the results con(cid:133)rm that, although the volatility of capital in(cid:135)ows during boom-bust cycles may have weighed on the EME (cid:133)nancial markets during the taper tantrum, those countries with stronger fundamentals were better prepared to sustain the (cid:133)nancial turbulence brought by the episode. Third, while controlling for macroeconomic fundamentals, the structure of (cid:133)nancial markets seems to matter. Those EMEs with greater capital account openness su⁄ered larger increases in bond yields (table 3, column 6) and also greater declines in the stock market index during the 4 For stock market indexes, for which the vulnerability index is not statistically signi(cid:133)cant, we use the runup in bank credit instead of the vulnerability index as a control when testing the relevance of the additional variables examined in this section (table 3, columns 9-12). 5 In table 3, column 5, when using the vulnerability index instead of short-term external debt, the coe¢ cient estimate for real e⁄ective appreciation maintains its sign but loses statistical signi(cid:133)cance. 11

episode (table 3, column 12), results that lend indirect support to the (cid:147)more-in-more-out(cid:148)hypothesis. Also, the EMEs with greater market capitalization su⁄ered more exchange rate depreciation (table 2, column 6), while those with greater foreign participation in equity markets su⁄ered larger increases in EMBI spreads during the episode (table 4, column 6).6 Notwithstanding the role of these other factors, we (cid:133)nd our relative vulnerability index across the EMEs to be a very robust summary statistic explaining heterogeneous responses across EMEs during the taper tantrum. In the univariate speci(cid:133)cations in which the vulnerability index is statistically signi(cid:133)cant (see tables 2 through 4, columns 2 and 8), the vulnerability index explains more of the variation of the dependent variable than any of the other explanatory variables taken individually. For example, the vulnerability index explains 65 percent of the variation in exchange ratedepreciationsduringthetapertantrumepisode(table2,column2). Incontrast,thecumulated gross capital in(cid:135)ows alone (not shown) explain 29 percent of the variation, foreign participation explains only 14 percent, and the growth forecast revision explains a mere 2 percent.7 3.3 Timing and persistence of di⁄erentiation across EMEs within the 2013 episode We next explore the timing and persistence of di⁄erentiation among EMEs during the episode of (cid:133)nancialstressthatstartedinMay2013. Tothisend,weusetwomeasuresof(cid:133)nancialperformance, one cumulative and another one incremental. First, for the cumulative measure, we keep the basis of comparison (cid:133)xed in April 2013 and take windows of varying lengths (in months) up to December (thatis,windowsthatendsuccessivelybetweenMayandDecember2013). Wecomputethe(cid:133)nancial performance of EMEs for each of these windows and explore the link with the vulnerability index describedearlier. Second, fortheincrementalmeasureof(cid:133)nancialperformance, wetakethemonthto-month changes in EME (cid:133)nancial indicators for the entire duration of the stress episode (that is, the (cid:133)rst window is April to May of 2013; the last window is November to December 2013). The results suggest that the di⁄erentiation across EMEs started relatively early in the stress periodandpersistedevenaftertheepisodeended,althoughthedegreeofdi⁄erentiationeasedinthe 6 In table 4, column 6, when using the vulnerability index instead of the current account, the coe¢ cient estimate for foreign participation maintains its sign but loses statistical signi(cid:133)cance. 7 In general, the vulnerability index also explains more of the variation in dependent variables than each of the six variables used to construct the index. The exception applies to the change in bond yields as the dependent variable (table 3, column 2), for which the vulnerability index explains a bit less than either the current account or the short-term external debt taken individually. 12

latermonthsoftheepisode. Intable5(column1ofeitherthetoporbottompanel),theresultsshow a positive and statistically signi(cid:133)cant link between the depreciation pressure index (the dependent variable) and the vulnerability index (the explanatory variable) even over the end-April to end- May period, suggesting that di⁄erentiation set in early when the taper-tantrum episode began in late May. Computing the depreciation pressure cumulated over windows of di⁄erent lengths in the top panel, the di⁄erentiation among EMEs according to the vulnerability index persisted throughout the stress period. For the incremental depreciation pressure in the bottom panel, most of the di⁄erentiation occurred in May, June, and August (columns 1, 2, and 4). However, some of the di⁄erentiation was reversed later in October (column 6), when the coe¢ cient switches signs, suggesting that the most vulnerable EMEs retraced some of their previous depreciations or declines in reserves as stress in global (cid:133)nancial markets eased. Similarly, as shown in table 6, there was a positive and statistically signi(cid:133)cant link between the cumulative increase in local currency bond yields (the dependent variable) and the vulnerability index starting in July 2013, and the link persisted until at least December (top panel). For the month-to-month increase in yields (bottom panel), most of the di⁄erentiation occurred in June, July, and August (columns 2, 3, and 4), with a renewed bout in December (column 8). Insum,theseresultsareconsistentwiththeinterpretationthatinvestorsadjustedtheirportfolio exposure by di⁄erentiating against the relatively more vulnerable EMEs and did not fully reverse this adjustment several months after the stress episode ended. 4 Di⁄erentiation across EMEs during past stress episodes In this section, we assess whether the di⁄erentiation across EMEs according to macroeconomic fundamentals was a development speci(cid:133)c to the 2013 taper tantrum episode or, whether any di⁄erentiation across EMEs during past episodes of (cid:133)nancial stress is also explained well by di⁄erences in their vulnerabilities. Toaddressthisquestion, we(cid:133)rstdevelopamethodologytoidentifyhistoricaleventsof(cid:133)nancial stress going back to the 1990s. Second, taking the identi(cid:133)ed start and end dates for each historical event as given, we construct the EME vulnerability index for the year prior to each event (using the methodology in section 3.2.2), and also construct measures of (cid:133)nancial performance during the event using the month prior to the start date as the basis of comparison. Finally, we examine how 13

the link between macroeconomic fundamentals and (cid:133)nancial performance has evolved over time. 4.1 Identi(cid:133)cation of past stress episodes Toidentifystressevents,welookforoutsizedmovementsinthreebroadindicatorsthatcharacterize (cid:133)nancial markets in the EMEs. Speci(cid:133)cally, we search for: (1) unusually large increases in VIX, whichservesasaproxyfor perceivedriskandglobalriskaversion; (2)unusuallylargedepreciations in an aggregate index of EME currencies against the dollar; and (3) unusually large declines in the MSCI equity index for the EMEs.8 Generally, we de(cid:133)ne an episode of (cid:133)nancial stress occurring when at least two of the three indicators display unusually large movements over the same period. Figure 4 shows the three aggregate (cid:133)nancial indicators for the EMEs at the weekly frequency starting in 1990. To identify unusually large movements in each of the three indicators along with the corresponding peak and trough dates, we use the following algorithm. First, for the VIX, we compute its deviations from a Hodrick-Prescott trend (cid:133)tted over the interval from 1990 to 2013. To identify possible episodes and their duration, we select all consecutive observations in which the VIXwasatleasttwostandarddeviationsabovethetrend; whentheseobservationsareimmediately precededorfollowedbyvaluesthatwereatleastonestandarddeviationabovethetrend,weinclude these other observations in the event as well. The (cid:133)rst and last dates of each VIX-determined event are considered as intervals for the episodes. Second, for the aggregate index of EME currencies, we compute the percent change in the index relative to the maximum value recorded over the previous six months, and select instances when the depreciation was 5 percent or more. We select the trough as the end date (that is, the week of maximum depreciation relative to the six-month maximum), and the maximum as the start date for each possible event. Third, for the MSCI stock market index, we also compute the percent change in the index relative to the maximum recorded over the previous six months, and focus on instances when the decline was 10 percent or more. As in the case of exchange rates, we select the trough as the end date (that is, the week of maximum decline), and the maximum as the start date for each possible event. Finally, we consider the dates of events indicated by each of the three measures, and focus on instances in which at least two of the three indicators point to overlapping events. 8 Speci(cid:133)cally,weusetheFederalReserveBoard(cid:146)sU.S.trade-weightedaggregatenominalindexofthedollaragainst a number of EMEs. 14

Based on this methodology, we identify 13 episodes of (cid:133)nancial stress in the EMEs going back to 1990, which are illustrated by the shaded areas in (cid:133)gure 4 (either shaded grey or blue). Out of these, we judgmentally choose a sub-set of seven events (that is, those in blue) that are associated withwell-knownepisodesof(cid:133)nancialstress, including: (1)theMexicancrisis, fromSeptember1994 to March 1995; (2) the Asian crisis, from July 1997 to January 1998; (3) the Russian crisis, from August to November 1998; (4) the aftermath of the Argentine crisis, from April to October 2002; (5) the GFC, from September 2008 to February 2009; (6) the European sovereign debt crisis, from July to December of 2011; and (7) the 2013 taper tantrum.9 4.2 Did economic fundamentals matter during past stress episodes? We estimate univariate regressions based on equation (1) for the seven historical episodes during which EME (cid:133)nancial conditions deteriorated signi(cid:133)cantly. The sole explanatory variable, the vulnerabilityindex,iscomputedbasedonthesixmacroeconomicvariablesdiscussedinsection2,using valuesfortheyearprecedingeachstressepisode. Thedependentvariable, thedepreciationpressure index, is based on changes in exchange rates and losses in foreign exchange reserves measured from the month prior to the start date to the end month of each event. Given the shifts in exchange rate regimes over time, it is important that we use the depreciation pressure index rather than just the exchange rate depreciation to measure (cid:133)nancial performance during the 1990s. For instance, the EMEs with heavily managed exchange rate regimes may have experienced less depreciation during a stress episode (that is, barring a devaluation), which would have falsely indicated resilience of their currencies; the depreciation pressure index solves this problem by also taking into account the loss in reserves required to maintain the managed (cid:135)oat during a crisis episode. Finally, for each episode, we exclude the EMEs at the epicenter of each crisis given their status as outliers (that is, Mexico in 1994, Indonesia, Malaysia, the Philippines, South Korea, and Thailand in 1997, Russia in 1998, and Argentina and Uruguay in 2001). The results for the depreciation pressure index are presented in (cid:133)gure 5 and table 7 (top panel). Figure 5 shows scatter plots between the vulnerability index (on the horizontal axis) and the depreciation pressure index (on the vertical axis) for each of the seven historical episodes along with the regression lines, while table 7 provides more detailed results for the corresponding univariate re- 9 We include the taper-tantrum episode again with this endogenous dating of the episode, which picks out May to August as the dates for the episode, consistent with what we chose in section 3.1. 15

gressions. From (cid:133)gure 5, it appears that while there was some heterogeneity in the (cid:133)nancial market performances during the EME crises of the 1990s and the early 2000s, the ability of macroeconomic fundamentals (as measured by our relative vulnerability index) to explain this heterogeneity was weak (panel 1 thourgh 4). However, judging from the increasingly close alignment between vulnerability and depreciation pressure, di⁄erentiation appears to begin with the GFC in 2008-09, then strengthens during the European sovereign debt crisis in 2011, and strengthens even more during the 2013 taper tantrum (panel 5 through 7). The results in table 7 (top panel) con(cid:133)rm this: the vulnerability index is not statistically signi(cid:133)cant and the R-squared values are very low for the Mexican crisis in 1994 through 1995 (column 1), the Asian crisis in 1997 (column 2), the Russian crisis in 1998 (column 3), and the Argentine crisis in 2002 (column 4). However, the vulnerability index became signi(cid:133)cant during the GFC in 2008 (column 4), and the coe¢ cient and R-squared values increase for the European sovereign crisis in 2011 (column 6), and even more for the taper tantrum in 2013 (column 7). Regarding the link between the vulnerability index and changes in stock market prices during each stress episode ((cid:133)gure 6), we see little evidence of di⁄erentiation during the past episodes of (cid:133)nancialstress,justaswefoundnoevidenceforthesummerof2013. Thecorrespondingcoe¢ cients intable7(thebottompanel)aregenerallynotstatisticallysigni(cid:133)cant. Finally,wecouldnotperform asimilarexercisefortheincreaseinbondyields,EMBIspreads,andCDSspreadsduetothelimited availability of historical data. In sum, it is noteworthy that macroeconomic fundamentals have gained increasing importance in recent years in explaining di⁄erences in (cid:133)nancial pressures across EMEs during episodes of heightened (cid:133)nancial stress. However, more research is needed to fully understand the factors explaining this apparent shift in behavior. There are at least two main hypotheses. On one hand, the improvement in macroeconomic fundamentals (to varying degrees across EMEs) combined with a better knowledge (perhaps made easier by technological advances and improvements in data quality) of individual EMEs(cid:146)characteristics could have changed the extent to which investors view EMEs as representing a single asset class. On the other hand, the shifting nature of the sources of the shocks that triggered stress events could also be a potential contributing factor to the rise in di⁄erentiation over time. While the (cid:133)nancial crises before the GFC(cid:151)when we (cid:133)nd little evidenceofanydi⁄erentiationbyinvestorsbeingrelatedtoeconomicfundamentals(cid:151)hadoriginated to an important extent in the EMEs themselves, the latest three episodes of (cid:133)nancial stress(cid:151)when 16

di⁄erentiation by investors could be explained well by di⁄erences in economic fundamentals(cid:151)were more obviously triggered by events originating in the advanced economies.10 As such, it is an issue forfutureresearchwhethertheEMEs(cid:146)remotenessfromtheoriginsofthe(cid:133)nancialstresslendsitself to investors discriminating more across EMEs according to their economic fundamentals. 5 Conclusion The taper-tantrum episode was triggered by a shift in May 2013 in market perceptions regarding the prospects for LSAPs by the Federal Reserve. It is important to understand the implications for the global economy, and particularly for the EMEs, that arise from evolving expectations of monetary policy actions by the Federal Reserve and other major advanced-economy central banks. In this study, we documented the deterioration in the (cid:133)nancial conditions of EMEs during the 2013 taper-tantrum episode. One intriguing aspect of this episode is that the response of EME (cid:133)nancial markets was heterogeneous. Our study explored the role of country-speci(cid:133)c characteristics, especially countries(cid:146)macroeconomic fundamentals, in explaining the heterogeneous response. Looking at performance across the whole episode, our results indicate that there was di⁄erentiation by investors across EMEs explained in large part by variation in the strength of their economic fundamentals, and that the di⁄erentiation began early in the episode and persisted throughout it. This result holds whether we use di⁄erent individual variables to measure economic fundamentals, or aggregate them to come up with a single index or relative vulnerabilities across EMEs. Taken at face value, these results suggest that policies to further strengthen economic fundamentals could go a long way to help the EMEs mitigate any disruptive e⁄ects from eventual normalization of monetary policy in the advanced economies. While the strength of economic fundamentals was the most important factor in(cid:135)uencing the deteriorationofEME(cid:133)nancialmarketsduringthetaper-tantrumepisode,we(cid:133)ndthatotherfactors were at play as well. Controlling for economic fundamentals, (cid:133)nancial market performance during the taper tantrum appeared to be weaker in those EMEs that had earlier received the largest gross private in(cid:135)ows of capital and had the greatest currency appreciations. For the performance of bond 10 Of course, as some observers have noted, even in the well-known EME crises such as those in the 1990s, shocks outsideoftheEMEsthemselvescouldsometimesactasoneofthetriggerpointsforthestartofthecrisisorformaking itworse. Oftenthesecrisesinvolvedheightenedcountry-speci(cid:133)cdomesticvulnerabilities,includingover-borrowingin anenvironmentof(cid:133)xedexchangeregimes,and,inthissetting,externaldevelopments,suchasariseinglobalinterest rates, contributed to their crises. 17

yields and credit spreads, revisions to the growth outlook also seemed to matter. We further explored whether any heterogeneous responses of EME (cid:133)nancial markets during past episodes of severe EME-wide (cid:133)nancial stress over the past 20 years could also be explained by di⁄erences in economic fundamentals across EMEs. For EME (cid:133)nancial crises of the 1990s and early2000s, we(cid:133)ndlittleevidencethatinvestorsdiscriminatedacrossEMEssigni(cid:133)cantlyaccording to the strengths of their fundamentals. However, our results do suggest that di⁄erentiation among EMEsbasedoneconomicfundamentalshasoccurredsincethemid-2000s,beginningwiththeglobal (cid:133)nancial crisis in 2008 and progressively increasing over time through the European debt crisis of 2011 and through the 2013 taper-tantrum episode. The interpretation of the results that di⁄erentiation by investors across EMEs can be explained well by di⁄erences in economic fundamentals after the mid-2000s but not before is not clear-cut, however. One interpretation could be that, prior to the early 2000s, the EMEs were viewed as a single asset class by international investors due to a combination of still-developing policy frameworks and less knowledge of characteristics of and di⁄erences across individual EMEs. As policy frameworks matured (to varying degrees across EMEs) and as advancement of technology facilitated timely and less costly access to information and data about particular EMEs, it became natural for investors not to more or less put all EMEs in one basket, but to look more closely at di⁄erences in their relative vulnerabilities. But another interpretation may lie in the idea that the factors that have led to (cid:133)nancial stresses in the EMEs have been di⁄erent since the mid-2000s compared with the period before. In particular, there is a sense that the origins of the shocks causing periods of severe EME-wide (cid:133)nancial stresses since the mid-2000s have been further removed from the EMEs themselves compared with the EME crises of the 1990s and early 2000s. The source of the shock may matter in how much investors di⁄erentiate across EMEs based on their economic fundamentals. 18

References Aizenman Joshua, Mahir Binici, and Michael M. Hutchison, 2014. "The Transmission of Federal Reserve Tapering News to Emerging Financial Markets." Cambridge, Mass: NBER Working Paper No. 19980. Bowman, David, Juan Miguel Londono Yarce, and Horacio Sapriza, 2014. (cid:147)U.S. Unconventional Monetary Policy and Transmission to Emerging Market Economies,(cid:148)International Finance Discussion Paper No. 1109, Board of Governors of the Federal Reserve System, June. Calvo, Guillermo, and Enrique Mendoza, (2001). (cid:147)Rational Herd Behavior and the Globalization of the Securities Market.(cid:148)Journal of International Economics. Chinn, Menzie and Hiro, Ito (2006). "A New Measure of Financial Openness," Journal of Comparative Policy Analysis 10(3). September 2008: pp. 307-20. Dornbusch, Rudiger, Yung Chul Park, and Stijn Claessens, (2000). (cid:147)Contagion: Understanding How It Spreads.(cid:148)The World Bank Research Observer, vol. 15, no. 2, pp. 177-97. Eichengreen Barry and Poonam Gupta, (2014). "Tapering Talk: The Impact of Expectations of Reduced Federal Reserve Security Purchases on Emerging Markets" Policy Research Working Paper 6754. Washington: World Bank, January. Eichengreen, Barry, Andrew Rose and Charles Wyplosz, 1995. "Exchange Market Mayhem: The Antecedents and Aftermath of Speculative Attacks," Economic Policy, vol 10, issue 21, pp. 249- 312. Ilzetzki, Ethan and Reinhart, Carmen and Rogo⁄, Kenneth (2011). "Exchange Rate Arrangements Entering the 21st Century: Which Anchor Will Hold?" mimeo, University of Maryland and Harvard University. Prachi Mishra, Kenji Moriyama, Papa N(cid:146)Diaye, and Lam Nguyen, (2014). "Impact of Fed Tapering Announcements on Emerging Markets." IMF Working Paper 14/109. Washington: International Monetary Fund, June. 19

Sahay, Ratna, Vivek Arora, Thanos Arvanitis, Hamid Faruqee, Papa N(cid:146)Diaye, Tommaso Mancini- Gri⁄oli, and an IMF Team, 2014. (cid:147)Emerging Market Volatility: Lessons from the Taper Tantrum,(cid:148)IMF Sta⁄Discussion Note SDN 14/09, September 2014. 20

Table 1: Summary statistics for the 2013 taper-tantrum episode Variable: Obs Mean Median St.Dev. Min Max Source Dependent variables: Exchange rate depreciation (%) 20 9.4 8.7 6.2 -0.8 22.8 IMF's IFS database Depreciation pressure index 20 1.8 1.6 1.2 -0.2 4.2 Authors' calculations Depreciation pressure index 2 20 1.8 1.5 1.6 -0.3 5.5 Authors' calculations Change in local currency bond yields (ppt) 14 1.2 1.9 0.8 0.3 2.7 Bloomberg Change in stock market index (%) 17 -4.6 -5.1 7.3 -15.0 9.7 Bloomberg Change in EMBI spreads (ppt) 15 0.5 0.5 0.3 -0.1 1.1 JP Morgan's EMBI Global database Change in CDS spreads (ppt) 14 0.5 0.4 0.3 0.0 1.0 Markit Memo: Change in reserves (%) 20 -3.1 -2.6 7.3 -25.0 12.5 IMF's IFS database Change in policy rates (ppt) 20 0.0 0.0 0.6 -1.4 1.5 Bloomberg, Haver Macro fundamentals and policy variables: Current account/GDP 2012 20 -0.6 -1.7 4.4 -6.2 10.7 IMF's WEO database Reserves/GDP 2012 20 24.8 17.4 18.7 4.5 84.8 Haver, IMF's IFS database Short-term ext. debt/reserves 2012 20 37.5 35.4 19.6 12.1 87.5 Joint External Hub Database (BIS-IMF-OECD-WB) Gov debt/GDP 2012 20 39.3 40.7 17.8 12.0 68.2 IMF's Historical Debt and WEO databases Inflation, annual, 2010-12 average 20 5.3 4.4 2.9 1.4 11.8 Haver Bank credit/GDP 5-year change, 2012 20 9.7 7.6 11.3 -11.7 26.2 IMF's IFS database Vulnerability index 2012 20 23.0 23.5 6.7 11.8 36.0 Authors' calculations Growth forecast 2013 revision, Consensus 20 0.1 -0.1 0.6 -0.6 2.2 Consensus growth forecast Dummy, inflation targeter 19 0.6 1.0 0.5 0.0 1.0 IMF's Exchange Rate Classification Dummy, XR floater 19 0.1 0.0 0.3 0.0 1.0 IMF's Exchange Rate Classification More-in-more-out variables: Gross inflows/GDP, cumul. 2010-12 19 3.4 2.4 2.9 -0.3 8.5 Haver REER appreciation, average 2009-12 20 2.8 2.5 2.7 -2.0 8.3 Federal Reserve Board Financial market structure: Market cap/GDP 2011 19 55.2 46.0 39.6 0.0 137.0 WB's WDI database Foreign participation/market cap 2011 18 13.8 14.2 6.8 3.4 24.5 IMF's IFS database, WB's WDI database Capital account openess 2011 19 0.0 -0.1 1.2 -1.2 2.4 Chinn-Ito index database Note: Changes in dependent variables are computed from April to August 2013. Exchange rates are expressed as units of local currency to the dollar, so that positive changes indicate depreciation. The depreciation pressure index is based on currency depreciation and losses in foreign exchange reserves, with larger values indicating stronger depreciation pressure. Alternatively, the depreciation pressure index 2 is based on currency depreciation, losses in foreign exchange reserves, and policy rate increases, with larger values indicating stronger depreciation pressure. The sample includes Argentina, Brazil, China, Chile, Colombia, Indonesia, India, South Korea, Malaysia, Mexico, Peru, Paraguay, the Philippines, Pakistan, Russia, South Africa, Taiwan, Thailand, Turkey, and Uruguay. 21

Table 2: Determinants of exchange rate depreciation and the depreciation pressure index (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Exchange rate depreciation, April-August 2013 (%) Depreciation pressure index, April-August 2013 Macro CA/GDP 2012 -0.82** -0.084 fundamentals (0.37) (0.082) and policy Reserves/GDP 2012 0.039 -0.0051 (0.089) (0.020) Short-term ext. debt/reserves 2012 0.069 0.0056 (0.062) (0.014) Gov debt/GDP 2012 0.16** 0.023 (0.061) (0.014) Inflation, average 2010-12 0.13 0.15 (0.44) (0.099) Bank credit/GDP 5-year change, 2012 0.059 0.0030 (0.088) (0.020) Vulnerability index 2012 0.74*** 0.75*** 0.66*** 0.76*** 0.78*** 0.14*** 0.14*** 0.14*** 0.14*** 0.14*** (0.13) (0.18) (0.12) (0.12) (0.14) (0.027) (0.037) (0.031) (0.028) (0.037) Growth forecast 2013 revision 1.03 0.11 (1.66) (0.35) Dummy, inflation targeter 0.97 -0.036 (2.12) (0.45) Dummy, XR floater -0.98 -0.089 (3.68) (0.78) More-in- Gross inflows/GDP, cumul. 2010-12 0.84*** 0.043 more-out (0.26) (0.068) REER appreciation, average 2009-12 0.55* 0.0079 (0.30) (0.069) Financial Market cap/GDP, 2011 0.048* 0.0037 structure (0.024) (0.0062) Foreign participation/market cap, 2011 0.13 0.014 (0.12) (0.032) Capital account openess -0.066 -0.20 (0.75) (0.19) Constant -2.39 -7.76** -8.50* -8.65*** -9.67*** -13.5*** -0.022 -1.52** -1.42 -1.48* -1.55** -1.91* (4.58) (3.12) (4.52) (2.87) (3.13) (4.00) (1.02) (0.65) (0.95) (0.76) (0.71) (1.03) Observations 20 20 19 19 20 18 20 20 19 19 20 18 R-squared 0.72 0.64 0.61 0.76 0.70 0.75 0.65 0.61 0.56 0.57 0.61 0.65 Note: Exchange rates are expressed as units of local currency to the dollar, so that positive changes in exchange rates indicate depreciation. The depreciation pressure index is based on currency depreciation and losses in foreign exchange reserves, with larger values indicating stronger depreciation pressure. Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. 22

Table 3: Determinants of the increase in bond yields and the decline in stock market prices (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Increase in bond yields, April-August 2013 (ppt) Stock market price increase, April-August 2013 (%) Macro CA/GDP 2012 -0.051 0.59 fundamentals (0.071) (0.60) and policy Reserves/GDP 2012 -0.016 -0.046 (0.016) (0.15) Short-term ext. debt/reserves 2012 0.021* 0.033*** -0.077 (0.011) (0.0077) (0.093) Gov debt/GDP 2012 -0.015 0.20* (0.011) (0.098) Inflation, average 2010-12 -0.13 0.36 (0.092) (0.75) Bank credit/GDP 5-year change, 2012 0.023 -0.41** -0.35** -0.32* -0.35* -0.39** (0.016) (0.16) (0.15) (0.18) (0.17) (0.16) Vulnerability index 2012 0.072** 0.074** 0.076** 0.086** -0.074 (0.024) (0.027) (0.030) (0.027) (0.27) Growth forecast 2013 revision -0.94* -4.38 (0.47) (4.24) Dummy, inflation targeter 0.52 -8.26** (0.36) (3.35) Dummy, XR floater -0.0094 4.04 (0.51) (4.66) More-in- Gross inflows/GDP, cumul. 2010-12 -0.042 -0.077 more-out (0.076) (0.67) REER appreciation, average 2009-12 0.14** -0.66 (0.063) (0.67) Financial Market cap/GDP, 2011 0.0034 0.0058 structure (0.0064) (0.046) Foreign participation/market cap, 2011 -0.052 0.0012 (0.033) (0.25) Capital account openess 0.58** -3.72** (0.24) (1.45) Constant 1.84** -0.42 -0.81 -0.34 -0.33 0.12 -5.58 -2.90 3.87 -1.23 0.81 -1.51 (0.73) (0.56) (0.63) (0.69) (0.39) (0.86) (7.34) (6.30) (3.31) (3.33) (3.19) (5.27) Observations 14 14 13 13 14 13 17 17 16 16 17 16 R-squared 0.76 0.44 0.68 0.40 0.64 0.68 0.61 0.01 0.53 0.21 0.24 0.53 Note: Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. 23

Table 4: Determinants of the increase in EMBI spreads and CDS spreads (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Increase in EMBI spreads, April-August 2013 (ppt) Increase in CDS spreads, April-August 2013 (ppt) Macro CA/GDP 2012 -0.031 -0.061** -0.034 fundamentals (0.041) (0.021) (0.023) and policy Reserves/GDP 2012 -0.0048 0.0022 (0.011) (0.0088) Short-term ext. debt/reserves 2012 -0.00018 0.00026 (0.0064) (0.0039) Gov debt/GDP 2012 0.0053 0.0023 (0.0047) (0.0040) Inflation, average 2010-12 0.033 0.079 (0.044) (0.046) Bank credit/GDP 5-year change, 2012 -0.0054 0.00075 (0.0081) (0.0074) Vulnerability index 2012 0.034*** 0.036*** 0.033** 0.035** 0.032*** 0.035** 0.030*** 0.032*** 0.031*** (0.011) (0.010) (0.012) (0.012) (0.0077) (0.011) (0.0079) (0.0081) (0.0094) Growth forecast 2013 revision -0.20* 0.14 (0.10) (0.17) Dummy, inflation targeter -0.17 (0.13) Dummy, XR floater 0.047 (0.18) More-in- Gross inflows/GDP, cumul. 2010-12 0.016 0.019 more-out (0.028) (0.018) REER appreciation, average 2009-12 0.0091 -0.0055 (0.027) (0.022) Financial Market cap/GDP, 2011 0.00044 -0.00085 structure (0.0020) (0.0017) Foreign participation/market cap, 2011 0.022* -0.0043 (0.012) (0.0086) Capital account openess -0.060 -0.039 (0.073) (0.050) Constant 0.27 -0.31 -0.32 -0.34 -0.35 0.10 -0.015 -0.20 -0.15 -0.22 -0.18 -0.061 (0.42) (0.28) (0.25) (0.29) (0.31) (0.21) (0.42) (0.18) (0.22) (0.18) (0.20) (0.28) Observations 15 15 15 15 15 14 14 14 14 14 14 14 R-squared 0.56 0.42 0.56 0.44 0.43 0.64 0.62 0.58 0.66 0.62 0.58 0.62 Note: Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. 24

Table 5: Examination of the April-August 2013 event: the depreciation pressure index (a) Cumulative stress Dependent variable: (1) (2) (3) (4) (5) (6) (7) (8) Depreciation pressure index Apr-May Apr-Jun Apr-Jul Apr-Aug Apr-Sep Apr-Oct Apr-Nov Apr-Dec Vulnerability index 2012 0.098* 0.12*** 0.12*** 0.14*** 0.13*** 0.11*** 0.10** 0.10** (0.047) (0.028) (0.026) (0.028) (0.033) (0.031) (0.037) (0.038) Observations 18 18 18 18 18 18 18 18 R-squared 0.21 0.53 0.59 0.60 0.50 0.45 0.33 0.32 (b) Incremental stress Dependent variable: (1) (2) (3) (4) (5) (6) (7) (8) Depreciation pressure index Apr-May May-Jun Jun-Jul Jul-Aug Aug-Sep Sep-Oct Oct-Nov Nov-Dec Vulnerability index 2012 0.098* 0.084** 0.055 0.12*** -0.021 -0.082* 0.027 0.058 (0.047) (0.032) (0.046) (0.030) (0.046) (0.040) (0.045) (0.038) Observations 18 18 18 18 18 18 18 18 R-squared 0.21 0.30 0.08 0.52 0.01 0.21 0.02 0.12 Note: The depreciation pressure index is based on currency depreciation and losses in foreign exchange reserves, with larger values indicating stronger depreciation pressure. Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 25

Table 6: Examination of the April-August 2013 event: the increase in bond yields (a) Cumulative stress Dependent variable: (1) (2) (3) (4) (5) (6) (7) (8) Increase in bond yields (ppt) Apr-May Apr-Jun Apr-Jul Apr-Aug Apr-Sep Apr-Oct Apr-Nov Apr-Dec Vulnerability index 2012 -0.013 0.022 0.047* 0.072** 0.077** 0.062** 0.076*** 0.094*** (0.011) (0.023) (0.025) (0.024) (0.028) (0.021) (0.023) (0.027) Observations 14 14 14 14 13 13 14 14 R-squared 0.11 0.07 0.22 0.44 0.40 0.44 0.47 0.50 (b) Incremental stress Dependent variable: (1) (2) (3) (4) (5) (6) (7) (8) Increase in bond yields (ppt) Apr-May May-Jun Jun-Jul Jul-Aug Aug-Sep Sep-Oct Oct-Nov Nov-Dec Vulnerability index 2012 -0.013 0.036* 0.025* 0.026*** 0.0059 -0.015 0.015 0.018** (0.011) (0.017) (0.013) (0.0073) (0.0070) (0.0093) (0.011) (0.0072) Observations 14 15 15 15 14 14 14 15 R-squared 0.11 0.25 0.22 0.50 0.06 0.18 0.13 0.32 Note: Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 26

Table 7: Differentiation across EMEs during past events: the depreciation pressure index (a) Dependent variable: depreciation pressure index Dependent variable: (1) (2) (3) (4) (5) (6) (7) Depreciation pressure index Aug 1994 - Mar 1995 Jun 1997 - Jan 1998 Jul-Nov 1998 Mar-Oct 2002 Aug 2008 - Feb 2009 Jun-Dec 2011 Apr-Aug 2013 Vulnerability index (y-1) -0.0079 0.042 0.078 0.055 0.071** 0.097*** 0.14*** (0.015) (0.043) (0.051) (0.043) (0.033) (0.026) (0.027) Observations 18 15 19 18 19 20 20 R-squared 0.02 0.07 0.12 0.09 0.21 0.43 0.61 (b) Dependent variable: stock market change (%) Dependent variable: (1) (2) (3) (4) (5) (6) (7) Stock market change (%) Aug 1994 - Mar 1995 Jun 1997 - Jan 1998 Jul-Nov 1998 Mar-Oct 2002 Aug 2008 - Feb 2009 Jun-Dec 2011 Apr-Aug 2013 Vulnerability index (y-1) -2.13* -0.27 -0.67 -0.12 -0.25 -0.041 -0.074 (1.07) (0.52) (1.11) (0.48) (0.36) (0.30) (0.27) Observations 12 10 15 16 16 17 17 R-squared 0.28 0.03 0.03 0.00 0.03 0.00 0.01 Note: The depreciation pressure index is based on currency depreciation and changes in foreign exchange reserves, with larger values indicating stronger depreciation pressure. Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 27

Figure 1: Financial performance April-August 2013 Note: Exchange rates are expressed as units of local currency to the dollar, so that positive changes in exchange rates indicate depreciation. The depreciation pressure index is based on currency depreciation and losses in foreign exchange reserves, with larger values indicating stronger depreciation pressure. As an alternative, the depreciation pressure index 2 is based on currency depreciation, losses in foreign exchange reserves, and policy rate increases, with larger values indicating stronger depreciation pressure. ycneuqerF 01 5 0 (1) -10 0 10 20 Exchange rate depreciation (%) ycneuqerF 8 6 4 2 0 (2) -30 -20 -10 0 10 Change in FX reserves (%) ycneuqerF 01 8 6 4 2 0 (3) -2 -1 0 1 2 Change in policy rate (ppt) ycneuqerF 6 4 2 0 (4) -1 0 1 2 3 4 Depreciation pressure index ycneuqerF 6 4 2 0 (5) 0 2 4 6 Depreciation pressure index 2 ycneuqerF 3 2 1 0 (6) 0 .5 1 1.5 2 2.5 Increase in LCY bond yields (ppt) ycneuqerF 3 2 1 0 (7) -20 -10 0 10 Stock market price change (%) ycneuqerF 4 3 2 1 0 (8) -.5 0 .5 1 Increase in EMBI spreads (ppt) ycneuqerF 5 4 3 2 1 0 (9) 0 .2 .4 .6 .8 1 Increase in CDS spreads (ppt) 28

Figure 2: Exchange rate depreciation and macroeconomic fundamentals during April-August 2013 in ug bz tksf id cl m p tx g har ph ma pceopk ru ko ta ch Note: Exchange rates are expressed as units of local currency to the dollar, so that positive changes in exchange rates indicate depreciation. )%( noitaicerped etar egnahcxE 52 02 51 01 5 0 (1) in ug bz stfk id armcx plg ph mtha pkco ru pe ko ta ch -5 0 5 10 CA/GDP (2012) R-squared= 0.47 )%( noitaicerped etar egnahcxE 52 02 51 01 5 0 (2) in ug bz sf tk id th phmaar m p x g cl ru ppkeco ta ko ch 0 20 40 60 80 Reserves/GDP (2012) R-squared= 0.20 )%( noitaicerped etar egnahcxE 52 02 51 01 5 0 (3) 20 40 60 80 100 ST debt/FX reserves (2012) R-squared= 0.25 in ug bz tk sf id cplg pmh thxar ma ru pe co pk kota ch )%( noitaicerped etar egnahcxE 52 02 51 01 5 0 (4) in ug bz sf tk id mcal th pmhx pg ar pceo ru pk ta ko ch 0 20 40 60 80 Gross gov. debt/GDP (2012) R-squared= 0.27 )%( noitaicerped etar egnahcxE 52 02 51 01 5 0 (5) in ug bz sf tk id ma cxp r lh ma t p h g pk pe rcuo ko ta ch 2 4 6 8 10 12 Inflation, 3yma (2012) R-squared= 0.23 )%( noitaicerped etar egnahcxE 52 02 51 01 5 0 (6) -10 0 10 20 30 Credit/GDP runup (2012) R-squared= 0.01 29

Figure 3: Increase in local currency government bond yields and macroeconomic fundamentals during April-August 2013 tk id co bz mx sf ru ko in th ma ch ta ph )tpp( sdleiy dnob YCL ni esaercnI 5.2 2 5.1 1 5. 0 (1) itdk co bz mx sf ru ko in mtha ph ch ta -5 0 5 10 CA/GDP (2012) R-squared= 0.45 )tpp( sdleiy dnob YCL ni esaercnI 3 2 1 0 1- (2) id tk co bz mx sf ru ko th ma in ch ta ph 0 20 40 60 80 Reserves/GDP (2012) R-squared= 0.44 )tpp( sdleiy dnob YCL ni esaercnI 5.2 2 5.1 1 5. 0 (3) 20 40 60 80 100 ST debt/FX reserves (2012) R-squared= 0.47 id tk co bz mx sf ru ko th ma in ch ta ph )tpp( sdleiy dnob YCL ni esaercnI 5.2 2 5.1 1 5. 0 (4) id tk co bz mx sf ru ko ma th in ch ta ph 0 20 40 60 80 Gross gov. debt/GDP (2012) R-squared= 0.03 )tpp( sdleiy dnob YCL ni esaercnI 5.2 2 5.1 1 5. 0 (5) id tk co bz mx sf ru ko in ma th ch ta ph 2 4 6 8 10 12 Inflation, 3yma (2012) R-squared= 0.11 )tpp( sdleiy dnob YCL ni esaercnI 5.2 2 5.1 1 5. 0 (6) -10 0 10 20 30 Credit/GDP runup (2012) R-squared= 0.00 30

Figure 4: Identification of past events of financial stress Note: The start and end dates for each of the 13 events in chronological order are as follows: (1) August 3 to November 9, 1990; (2) September 16, 1994 to March 10, 1995; (3) July 4 to August 29, 1997 and October 31, 1997 to January 9, 1998, merged in one single event; (4) August 7 to November 13, 1998; (5) July 14 to December 22, 2000; (6) April 19 to October 11, 2002; (7) July 20 to August 17, 2007; (8) October 26, 2007 to March 21, 2008; (9) September 12, 2008 to February 27, 2009; (10) April 16 to May 28, 2010; (11) July 29 to December 16, 2011; (12) March 2 to June 8, 2012; and (13) May 10 to August 30, 2013. 31

Figure 5: Depreciation pressure and economic fundamentals during past events of financial stress ar bz ma ph co ta pgth in idu p g e ko cl pk tk ch sf Note: For each event, the vertical axis shows the depreciation pressure index (based on currency depreciation and losses in foreign exchange reserves), with larger values indicating stronger depreciation pressure. 5. 0 5.- 1- (1) Aug 1994 - Mar 1995 ru pg pk ta co cl in bz pe ar tk sf ch ug mx 10 15 20 25 30 Vulnerability index (1993) R-sq = 0.02 exludes MX 3 2 1 0 1- (2) Jun 1997 - Jan 1998 bz tk mxco pk pe pg ug t c a h cl ar sf ph in ktho ma id 10 15 20 25 30 Vulnerability index (1996) R-sq = 0.07 ex. ID KO MA PH TH 3 2 1 0 1- 2- (3) Jul - Nov 1998 bz pg co ph mx cl tk ma pe th ta chko i i r n d u sf pk 10 15 20 25 30 Vulnerability index (1997) R-sq = 0.12 excludes RU 4 2 0 2- 4- (4) Mar - Oct 2002 10 15 20 25 30 Vulnerability index (2001) R-sq = 0.09 excludes AR UG ko mbxz tk id pg co ma in sf pe aurg cl ta ph ch pk th 3 2 1 0 (5) Aug 2008 - Feb 2009 tk in sf ru bz ipdg marxpk tath ko co ma ug cl ch phpe 0 10 20 30 40 Vulnerability index (2007) R-sq = 0.21 excludes RU 4 3 2 1 0 1- (6) Jun - Dec 2011 in pk id bz sf tk ar th ma ug phru cl mpxg pe co ta ko ch 10 20 30 40 Vulnerability index (2010) R-sq = 0.43 4 3 2 1 0 (7) Apr - Aug 2013 10 15 20 25 30 35 Vulnerability index (2012) R-sq = 0.61 32

Figure 6: Stock market declines and economic fundamentals during past events of financial stress cl ko ta pe chid ma th in ph bz ar Note: For each event, the vertical axis shows the percentage change in the local stock market index. 01 0 01- 02- 03- 04- (1) Aug 1994 - Mar 1995 mx ch ta in ru sf bz cl ar pe 10 15 20 25 30 Vulnerability index (1993) R-sq = 0.28 exludes MX TK 01 0 01- 02- (2) Jun 1997 - Jan 1998 ktho ma ph ch ta cl in mx sf ar id pebz tk 10 15 20 25 30 Vulnerability index (1996) R-sq = 0.03 ex. ID KO MA PH TH TK 04 02 0 02- 04- (3) Jul - Nov 1998 co ru ch pe th ma cl tksf in mx id ko ph ta bz 10 15 20 25 30 Vulnerability index (1997) R-sq = 0.03 excludes RU 02 0 02- 04- (4) Mar - Oct 2002 10 15 20 25 30 Vulnerability index (2001) R-sq = 0.00 excludes AR UG ch cl co ma ko sf bz phmx tath inid ar tk pe 01- 02- 03- 04- 05- (5) Aug 2008 - Feb 2009 msxf ph th peid ma bz cklo co in tk ch ta ar ru 0 10 20 30 40 Vulnerability index (2007) R-sq = 0.03 excludes RU 01 0 01- 02- 03- (6) Jun - Dec 2011 sf ar ma co ta in ko mx ru ch ph bz th id pe cl tk 10 20 30 40 Vulnerability index (2010) R-sq = 0.00 01 5 0 5- 01- 51- (7) Apr - Aug 2013 10 15 20 25 30 35 Vulnerability index (2012) R-sq = 0.01 33