Large Capital Inflows, Sectoral Allocation, and Economic Performance

K.7 Large Capital Inflows, Sectoral Allocation, and Economic Performance Benigno, Gianluca, Nathan Converse, and Luca Fornaro Please cite paper as: Benigno, Gianluca, Nathan Converse, and Luca Fornaro (2015). Large Capital Inflows, Sectoral Allocation, and Economic Performance. International Finance Discussion Papers 1132. http://dx.doi.org/10.17016/IFDP.2015.1132 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1132 March 2015

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1132 March 2015 Large Capital Inflows, Sectoral Allocation, and Economic Performance Gianluca Benigno, Nathan Converse, and Luca Fornaro NOTE:InternationalFinanceDiscussionPapersarepreliminarymaterialscirculatedtostimulate 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 References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.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.

Large Capital Inflows, Sectoral Allocation, and Economic Performance∗ Gianluca Benigno† Nathan Converse‡ Luca Fornaro§ Abstract This paper describes the stylized facts characterizing periods of exceptionally large capital inflows in a sample of 70 middle- and high-income countries over the last 35 years. We identify 155 episodes of large capital inflows and find that these events are typically accompanied by an economic boom and followed by a slump. Moreover, duringepisodesof large capital inflows capital andlabor shiftoutof themanufacturing sector, especially if the inflows begin during a period of low international interest rates. However, accumulating reserves during the period in which capital inflows are unusually large appears to limit the extent of labor reallocation. Larger credit booms and capital inflows duringthe episodes we identify increase the probability of a sudden stop occurring duringor immediately after the episode. In addition, the severity of the post-inflows recession is significantly related to the extent of labor reallocation during theboom,withastrongershiftoflaboroutofmanufacturingduringtheinflowsepisode associated with a sharper contraction in the aftermath of the episode. Keywords: Capital Flows, Surges, Sectoral Allocation, Sudden Stops JEL Classification: F31,F32,F41,O41 ∗This research has been supported by ESRC grant ES/I024174/1. We thank Carlos V´egh, Alberto Ortiz and Mark Spiegel for their helpful discussions as well as participants in the IDB-JIMF Conference on Macroeconomic Challenges Facing Latin America and the Federal Reserve System Committee on InternationalEconomicAnalysis2014Conference,andseminarparticipantsattheBankofLithuania. Theviewsin this paper are solely the responsibility ofthe author(s) andshouldnot be interpretedas reflecting the views of the Boardof Governorsof the FederalReserve System or of any other personassociatedwith the Federal Reserve System. †London School of Economics, CEPR, and Centre for Macroeconomics; G.Benigno@lse.ac.uk. ‡International Finance Division, Federal Reserve Board; Nathan.L.Converse@frb.gov. §CREI, Universitat Pompeu Fabra and Barcelona GSE; LFornaro@crei.cat.

1 Introduction The last 30 years have seen a sustained process of financial globalization, with countries around the world opening their capital accounts and joining international financial markets. With the passing of time, both in academic and policy circles an initially benign view toward openness to international capital flows has given way to a more skeptical approach. The IMF’s inclusion of capital controls in its recommended policy toolbox epitomizes the shift in thinking (Ostry et al., 2010; WEO, 2011). Not only are episodes of large capital inflows thought to set thestage forsubsequent financial crises, but theimpact of inflows oneconomic performanceduringtranquiltimeshasalsobeencalledintoquestion(Giavazzi and Spaventa, 2010; Powell and Tavella, 2012). Figure1summarizes theexperienceofSpain, whichwasinmanyways typicalofthecountries in the Eurozone periphery. Following the launch of the Euro, Spain received large capital inflows (panel a), coinciding with a consumption boom (panel b). Moreover, Spain experienced a shift of resources out of sectors producing tradable goods such as manufacturing and into the production of nontradable goods, such as construction (panel c). During the same period, Spain saw a slow down in productivity growth (panel d). These developments have led some authors to draw a connection between episodes of large capital inflows and slowdowns in productivity growth, since capital inflows can trigger a movement of resources towardnontradablesectorscharacterizedbyslowproductivitygrowth(Benigno and Fornaro, 2014; Reis, 2013). While the narrative evidence from the Eurozone periphery appears compelling, it remains unclear to what extent these countries’ experience is typical of recipients of large capital inflows. In the second half of the 1990s, Brazil received capital inflows of a magnitude similar to those flowing to the Eurozone periphery (Figure 2, panel a). While Brazil did experience a consumption boom (panel b), the share of employment dedicated to manufacturing was steady or rising, reversing its earlier downward trend (panel c). Similarly, the inflows episode in Brazil saw a net improvement in TFP (panel d). Precisely how periods of large capital inflows affect recipient economies thus remains an open question. Moreover, the issue has acquired new urgency as capital flows to emerging market economies have surged in the five years since the 2008 financial crisis. This paper provides a systematic analysis of how large capital inflows affect macroeconomic performance and the sectoral allocation of productive resources. We examine 155 episodes of largecapitalinflows over thelast35yearsinagroupof70middle- andhigh-incomecountries. We find that these episodes coincide with an economic boom, in which output, consumption, 1

Figure 1: Spain: Capital Inflows and Macroeconomic Performance, 1998-2012 )PDG %( tnuoccA tnerruC 2 0 2− 4− 6− 8− 01− (a): Current Account 1998 2000 2002 2004 2006 2008 2010 2012 dnerT PH morf noitaiveD % 4 2 0 2− 4− (b): Consumption 1998 2000 2002 2004 2006 2008 2010 2012 )PDG%( tnuoccA tnerruC 2− 4− 6− 8− 01− 81. 71. 61. 51. 41. 31. tnemyolpmE latoT fo erahS (c): Employment in Manufacturing 1998 2000 2002 2004 2006 2008 2010 2012 Current Account (%GDP) Trend 1980−2000 Actual Employment Share 95% Confidence Band )PDG%( tnuoccA tnerruC 2− 4− 6− 8− 01− 50.7 7 59.6 9.6 58.6 8.6 PFT goL (d): Total Factor Productivity 1998 2000 2002 2004 2006 2008 2010 2012 Current Account (%GDP) Trend 1980−2000 Actual TFP 95% Confidence Band Sources: IMF BoPS, WDI, PWT, UNIDO, ILO investment, employment, and domestic credit all rise initially. However, once capital inflows subside and credit contracts, the boom leaves place to a recession. Alongside these aggregate macroeconomicdynamics, atthesectorallevelwefindthatlargecapitalinflowsareassociated with an expansion of nontradable sectors, such as services and construction, at the expenses of the sectors producing tradable goods, including agricultural products and manufactured goods. Studying the manufacturing sector in detail, we find that the share of both employment and investment allocated to manufacturing drops during episodes of large capital inflows. In particular, while the reallocation of investment is a general phenomenon in our sample, the reallocation of labor occurs specifically during episodes in which governments do not offset capitalinflowsthroughsubstantialpurchasesofforeignassets, andduringepisodesthatbegin when international liquidity is abundant. Hence, our empirical results are consistent with the predictions of a standard two-sectors small open economy model, according to which capital inflows driven by an increase in access to foreign capital should generate a shift of productive resources out of sectors producing tradable goods, and into sectors producing non-tradable goods (Rebelo and Vegh, 1995; Reis, 2013; Benigno and Fornaro, 2014) We next consider how the behavior of macroeconomic indicators during an inflows episode 2

Figure 2: Brazil: Capital Inflows and Macroeconomic Performance, 1990-2004 )PDG %( tnuoccA tnerruC 2 0 2− 4− 6− 8− 01− (a): Current Account 1990 1992 1994 1996 1998 2000 2002 2004 dnerT PH morf noitaiveD % 5 0 5− (b): Consumption 1990 1992 1994 1996 1998 2000 2002 2004 )PDG%( tnuoccA tnerruC 2 0 2− 4− 80. 60. 40. 20. 0 tnemyolpmE latoT fo erahS (c): Employment in Manufacturing 1990 1992 1994 1996 1998 2000 2002 2004 Current Account (%GDP) Trend 1975−1995 Actual Employment Share 95% Confidence Band )PDG%( tnuoccA tnerruC 2 0 2− 4− 9.5 8.5 7.5 6.5 5.5 PFT goL (d): Total Factor Productivity 1990 1992 1994 1996 1998 2000 2002 2004 Current Account (%GDP) Trend 1975−1995 Actual TFP 95% Confidence Band Sources: IMF BoPS, WDI, PWT, UNIDO, ILO relates to the probability that the episode coincides with a capital flows reversal or a sudden stop. Evidence from probit regressions suggests that, while economic conditions before and during the episodes of large capital inflows are not systematically related to whether or not capital flows reverse sharply, both a larger credit boom and larger capital inflows are associated with a higher probability of a sudden stop, in which a capital flows reversal is accompanied by an output contraction. We also investigate the existence of a relationship between the behavior of the economy during the inflows and the post-inflows slump. Regressing post-episode macroeconomic performancemoregenerally onconditionsbeforeandduring theboom, we findthat largercredit and inflows are associated with a deeper fall in GDP, consumption, investment, employment and TFP at the end of the episode. Moreover, the reallocation of labor out of manufacturing is robustly and significantly related to economic performance after large capital inflows come to an end, with a stronger shift of labor out of manufacturing during the inflows episode associated with a sharper contraction in the aftermath of the episode. Reserve accumulation during the episode appears to dampen the negative after effects of large capital inflows, even when we control for the sectoral reallocation with which it is correlated. It thus appears that foreign reserve accumulation acts through multiple channels to insulate the economy from the dislocation associated with episodes of large capital inflows. 3

This paper contributes to two areas of literature. First, in focusing on episodes of unusually large capital inflows, our work is related to the study of what have been called capital flow bonanzas or surges. Our methodology, taken fromthe literature on credit booms (Gourinchas et al.,2001;Tornell and Westermann,2002;Mendoza and Terrones,2008),identifiesperiodsinwhichthelevel ofcapitalinflowsisunusuallylarge. Bycontrast,theliterature on surges has generally examined the causes (Forbes and Warnock, 2012) and consequences (Reinhart and Reinhart, 2009; Kalantzis, 2014) of unusually large changes in capital inflows. Ghosh et al. (2014) study unusually high levels of capital inflows, but they examine the causes of such episodes specifically in emerging markets, while we focus on the consequences of large capital inflows in both emerging and advanced economies. The experience of Eurozone countries highlights the value of our approach. Capital inflows to Spain grew steadily, eventually exceeding 10 percent of GDP, but never jumped as in a surge. The work closest to our paper is research by Cardarelli et al. (2010) and Caballero (2014), both of whom also examine episodes in which the level of capital inflows is unusually high. Whereas Cardarelli et al. (2010) analyze policy responses to large capital inflows, we study the effects of such episodes on the real economy. Caballero (2014) focuses on how large inflows affect the likelihood of banking crises, whereas our work highlights the way large capital inflows affect the sectoral allocation of resources. Using a slightly different measure of capital inflows than these papers allows us to examine episodes over a longer timespan. Importantly, with respect to existing works, our data include the large capital flows to the Eurozone periphery in the mid-2000s as well as recent capital flows to emerging markets. Second, our work contributes to research on how external factors interact with the sectoral allocation of production to affect economic performance. Rodrik (2008) documents that an undervalued exchange rate is associated with faster economic growth, and presents evidence that the reallocation of resources into the production of tradable goods generates this relationship. Analyzing the impact of sectoral allocation on aggregate productivity in more detail, McMillan and Rodrik (2011) show that a shift of productive resources into relatively less productive sectors has in many countries severely dampened aggregate productivity growth, even as resource allocation within sectors has improved. Our empirical findings are consistent with Rodrik’s in that we show that large capital inflows are associated with both real exchange rate appreciation and a reallocation of resources out of the manufacturing sec- 1 tor, as well as a subsequent slowdown in both output and productivity. Finally, Converse (2014) presents evidence that the financial uncertainty generated by volatile international 1In this sense, ourresults helprationalizethe findings of Alfaro et al. (2014)andGourinchas and Jeanne (2013), which highlight how capital tends to flow toward those developing countries in which productivity growth is slower. 4

capital flows can shift the sectoral allocation of investment in emerging markets, depressing aggregate TFP and growth. The key novelty of our paper with respect to these two literatures lies in the systematic description of how the share of productive resources allocated to manufacturing behaves in a large sample of episodes of large capital inflows. In fact, it has been documented in the context ofexchange ratebasedstabilizationprograms(Rebelo and Vegh, 1995), andofcredit booms(Gourinchas et al.,2001;Tornell and Westermann,2002;Mendoza and Terrones,2008) that the share of tradable sectors in GDP drops with inflows of capital. However, to the best of our knowledge, we are the first ones providing direct evidence on the allocation of labor and investment across sectors in a large sample of inflows episodes, and connecting the sectoral reallocation of resources during the inflows to the post-inflows slump. The rest of the paper begins by describing the data and methodology we use to identify episodesoflargecapitalinflowsinSection2. InSection3weexaminehowkeymacroeconomic variables and the sectoral allocation of productive resources behave during and after inflows episodes. In section 4 we consider how the behavior of the economy during the inflows relates to the probability that an episode ends up in a reversal or a sudden stop and how it affects the post-episode economic performance more generally. Section 5 concludes. 2 Data and Methodology 2.1 Identifying Capital Inflows Episodes As a measure of capital inflows into the economy we use the current account deficit plus the 2 increase in holdings of official reserves. All data on international capital flows are taken from the IMF’s Balance of Payments Statistics (BoPS) data base. Such a broad measure of capitalinflowsmayseematoddswithrecentworkonthesubject, whichhasshownimportant differences in the behavior of private and public flows (Alfaro et al., 2014) as well as gross and net flows (Rothenberg and Warnock, 2011; Forbes and Warnock, 2012; Broner et al., 2013). However, our focus here is the impact of capital flows in recipient countries, meaning that the origins and drivers of those flows is of less importance for our analysis. We do add reserve accumulation to our measure of capital inflows, however, in order to be 2A current account deficit carries a positive sign in what follows, since this corresponds to net capital inflows. 5

3 able to differentiate between large capital inflows and the policy response to them. When the government purchases foreign reservers, it offsets the liabilities incurred when foreigners acquireclaimsondomesticresidents. Byaddingreserve accumulationtothecurrent account, we undo this netting out. In Sections 3 and 4 we explore in detail whether the strategy of reducing the current account deficit through the acquisition of official reserves affects how the economy responds to large capital inflows. Having selected our preferred inflows measure, we normalize by GDP in order to capture 4 the size of the flows relative the economy. We then detrend the normalized series using an HP filter because we observe in the data that numerous economies exhibit medium- or longrun trends in the size of capital inflows, presumably for varying structural reasons. Most obviously, the neoclassical growth model predicts that capital-scarce economies will receive capital inflows that diminish in size as the economy converges to its steady-state level of 5 capital. A downward trend in capital inflows is also consistent with models of convergence to a technological frontier (as in Krugman 1979 or Grossman and Helpman 1991). We are not interested in large capital flows that emerge in the course of a long-run trend, but rather on short- and medium-term jumps in capital inflows that occur along this transition path in response to shocks. Detrending the series allows us to identify precisely such events. In order to identify periods of exceptionally large capital inflows, we follow a procedure commonlyusedinresearchoncreditbooms(Gourinchas et al.,2001;Mendoza and Terrones, 2008) which has more recently been applied to international capital flows (Cardarelli et al., 2010; Caballero, 2014). We calculate the long-run standard deviation of our detrended capital inflows measure for each country, and flag years in which inflows rise more than one 6 standard deviation above their trend. These flagged country-years mark the existence of an episode of large capital inflows. An episode begins when inflows first rise more than half 3Reinhart and Reinhart (2009) describe reserve accumulation less the current account balance as “the best indicator of capital flows,” but ultimately use the current account balance in order to lengthen the period covered by their analysis. In excluding changes in reserves from our net capital inflows measure, our approach is similar to that of Ghosh et al. (2014), who also subtract government borrowing from official sources. 4Specifically, the capital inflows are measured in current US dollars and then normalized by the trend component of GDP in current US dollars. 5Chapter 2 of Obstfeld and Rogoff (1996) provides a textbook treatment on the role of capital flows in the neoclassical growth model. 6 Unlike Gourinchas et al. (2001) and Cardarelli et al. (2010) we take the trend over the entire sample period for each country, rather than a country-year-specific expanding window tend. This is because our rationalefordetrendingdifferssubstantially. Cardarelli et al.(2010)studypolicyresponsestocapitalinflows andthereforedetrendinordertodeterminewhethercontemporarypolicy-makerswouldhaveseentheinflows asunusuallylarge. Wedetrendtodeterminewhethertheinflowsarelargerelativetothelong-runtrajectory oftheeconomy. Thisdifferenceinmotivationmakesthelong-runtrendmoreappropriatethananexpanding window. 6

a standard deviation above their trend level and ends when they again come within half a 7 standard deviation of their trend. The case of Brazil, depicted in Figure 3, demonstrates the merits of our approach to identifying large capital inflows. First, at several points the Brazilian authorities have offset large capital inflows by purchasing substantial foreign exchange reserves. This can be seen in the divergence of the current account (the dotted black line) and our measure of capital inflows (the solid blue line). To highlight the clearest and most recent example, note that during the four years after the 2008 financial crisis, Brazil received approximately US$50 billion in capital inflows, an average of 4.5 percent of GDP per year. At the same time, the country’s foreign exchange reserves nearly doubled. The accumulation of assets by the monetary authorities meant that Brazil’s current account deficit averaged only 2.5 percent of GDP during a time of large capital inflows, much discussed by policy-makers and the media as well as evident in the data. Figure 3: Identifying Episodes of Capital Inflows: Brazil PDG fo tnecreP 6 4 2 0 2− 4− 1980 1985 1990 1995 2000 2005 2010 2015 Current Account Deficit Capital Inflows Detrended Inflows Inflow Episode Trend Thresholds In Figure 3 we also plot the HP trend—the solid black line. This shows how the typical size of capital inflows has varied over time, and supports our use of HP filtered inflows to decide 7In the terminology used by Mendoza & Terrones (2008), we set the entry and exit thresholds for the detrended current account equal to 0.5. 7

when capital inflows are unusually large. What would have been considered an unusually large capital in the late 1980s would not stand out as particularly large in the late 2000s. Although for some countries balance of payments data extend as far back as the 1940s, the IMF Balance of Payments data cover substantially fewer countries prior to the early 1970s. We therefore restrict our attention to capital inflows episodes occurring between 1975 and 2010. We exclude from the analysis countries with a population that never exceeds one million, as well as those with annual GDP that remains below one billion dollars throughout the period we study. This has the virtue of excluding several offshore financial centers where the relationship between capital flows and the real economy might differ substantially from the typical economy. We also remove from our dataset major oil exporters and countries 8 eligible to receive World Bank International Development Association (IDA) assistance. Where oil price movements and donors’ willingness to provide foreign aid determine the external balance, the relationship between capital inflows and the real economy presumably differs substantially from most other economies. We experimented with alternate methodologies for identifying episodes in order to verify the robustness of our results to the use of different capital inflows measures, detrending methods, and thresholds. Thus, we identified episodes using two alternate measures of capital inflows: the raw current account as a share of GDP and the current account in constant US dollars normalized by population. We also detrended the current account using a linear trend rather than an HP filter. Finally, we raised the threshold for identifying episodes from one to 1.5 standard deviations, and (separately) lowered the exit and entry threshold from 0.5 to zero. Using alternate inflows measures does change the set of events that are identified as episodes of large capital inflows, while alternate detrending methods and thresholds alter the average length of the episodes. Since a linear trend is less flexible than an HP trend, the variable can diverge from the trend for longer. Likewise, a lower threshold prolongs the duration of those episodes which do not start and stop abruptly. 2.2 Other Variables Having identified episodes of large capital inflows, we are particularly interested in how these episodes end. Do inflows gradually taper off or do they stop abruptly? Does the economy experience a hard landing once inflows subside? Following the large literature on crises and 8The main criterion for IDA eligibility is a PPP-adjusted per-capita GDP of less than US $1,195. The IDA provides grants as well as concessional lending to eligible countries. 8

sudden stops, we identify capital flow reversals and sudden stops using the methodology 9 developed by Calvo et al. (2004). In this classification scheme, a reversal occurs when the year-on-year change in capital inflows is at least two standard deviations below the mean. A sudden stop occurs when a reversal coincides with an output contraction. We deem a capital inflows episode to coincide with a reversal or sudden stop if one of these events occurs at any point during the episode or in the year immediately after the episode ends. Several authors have suggested a link between aggregate productivity and capital inflows (Aoki et al., 2010) as well as closely related variables such as the real exchange rate (Rodrik, 2008). In order to further explore these links we calculate total factor productivity (TFP) for a broad sample of countries over an extended time period using data on output and investment obtained from the Penn World Tables (Heston et al., 2013). We estimate initial capital stock using the method described in Klenow and Rodr´ıguez-Clare (1997) and calculate capital stock for subsequent years using the annual values of investment obtained from the Penn World Tables. In calculating TFP, we use employment data from the International Labor Organization’s LABORSTA data set rather than the labor force data provided by the Penn World Tables. This ensures that fluctuations in TFP around episodes of large capital inflows are not the result of changes in the unemployment rate. We calculate aggregate total factor productivity using standard growth accounting (e.g. as in Benhabib and Spiegel, 2005). This methodology allows us to measure TFP in nearly all of the 69 countries in which we observe episodes of large capital inflows. Macroeconomic data are from the standard sources, including the IMF International Financial Statistics (IFS) and the World Bank World Development Indicators (WDI). We also analyze international liquidity conditions at the time of capital inflows episodes, taking movements of the effective Federal Funds rate, obtained from the Federal Reserve Economic Database (FRED) as a proxy for changes in the rates attached to international lending. We calculate real rates by subtracting from the nominal rate inflation during the previous year, which we use as a proxy for expected inflation. To measure the risk aversion of major international investors we use the VIX index. The VIX measures the implied volatility of S&P index options and thus reflects the price of risk in U.S. equity markets. When the price 10 of risk and thus the VIX is low, it can be inferred that risk aversion is low. 9Rothenberg and Warnock (2011) and Forbes and Warnock (2012) use this approach to identify both surges and sudden stops in gross capital flows. 10More specifically, we use the “original” VIX index or VXO, which measures the implied volatility of optionsontheS&P100andwhichisavailablesincethelate1980s. Toobtainameasureofriskaversionfrom 1970 to 1986 we regressthe realizedvolatility of the S&P 100 on the VXO for the post-1986 period and use the estimated coefficients to back-cast the VXO. 9

We obtain data on manufacturing sector employment, value-added, and investment from the UNIDO INDSTAT2 database. As the UNIDO data are in nominal terms, we deflate them using the aggregate GDP deflator (taken from the WDI), as is standard in the literature (e.g. Kroszner et al., 2007; Ciccone and Papaioannou, 2009; Gupta and Yuan, 2009; 11 Levchenko et al., 2009; Rajan and Subramanian, 2011). Appendix 5 provides detailed descriptions of which data were drawn from which source. 2.3 Descriptive Statistics Our baseline methodology identifies 164 episodes of large capital inflows occurring in 70 countries between 1975 and 2010. Of these, 54 took place in advanced economies and 67 in 12 emerging markets. AfulllistoftheseepisodesisprovidedinAppendix 5. Ourmethodology captures nearly all well known examples of large capital inflows. These include events in emerging markets such as the lead-up to crises in Latin America in the early 1980s, the exchange-rate-based based stabilization programs in the region later in the decade, which were accompanied by large inflows (V´egh, 1992; Rebelo and Vegh, 1995), and the run-up to the East Asian crises during the mid-1990s. In addition, our sample includes advanced country cases such as Scandinavia and the United Kingdom in the early 1990s, and the Eurozone periphery in the mid-2000s. We also pick up less well-known episodes that did not end in a crisis, such as inflows to Canada in the late 1980s. Importantly, the episodes we identify include the large capital inflows to emerging markets such as Brazil, Indonesia, and Turkey following the 2008 crisis. The number of episodes we identify is consistent with the findings of Reinhart and Reinhart (2009), who identify 207 capital flow “bonanzas” in middle- and high-income countries between 1980 and 2007, of which 112 last more than one year. Figure 4 plots the number of countries undergoing episodes of large capital inflows in each year. The number of episodes varies substantially over time, with increases in the number of episodes in the early 1980s and 1990s, and again in the late 2000s. Notably, the number of countries receiving exceptionally large inflows was significantly larger during the most recent surge in episodes than in the past. Presumably this pattern reflects the fact that 11Since industry-leveldeflators arenotavailablefor a broadset ofcountries,the alternateapproachtaken by Koren and Tenreyro (2007) is to use US industry-level deflators. We use the method most widely used in the literature. 12 Wedefineemergingmarketsbroadly,includinginthiscategorycountriesineithertheJPMorganEmergingMarketBondIndex(EMBI)ortheS&P/InternationalFinanceCorporationEmergingMarketsDatabase Investable Index (S&P IFCI Index). Advanced economies are the high-income members of the OECD. 10

governments have consistently liberalized controls on capital inflows since the 1970s, as documented, for example, by Chinn and Ito (2006). The type of countries receiving large inflows has also fluctuated. During the late 1980s advanced economies were nearly the only countries receiving large inflows. More recently, the majority of large inflows episodes have taken place in emerging markets, although other economies, which comprise smaller and relatively poorer countries sometimes called frontier markets, have also seen their share in the number of episodes increase. Figure 4: Frequency of Large Inflows Episodes Over Time edosipE swolfnI gniogrednU seirtnuoC fo rebmuN 04 03 02 01 0 5 0 5 0 5 0 5 0 7 8 8 9 9 0 0 1 9 9 9 9 9 0 0 0 1 1 1 1 1 2 2 2 Advanced Economies Other Countries Emerging Markets Table 1 provides descriptive statistics about the episodes of large capital inflows that we identify, broken down by region. Overall roughly one third of the episodes occur in advanced economies, while Latin America, Asia, and Eastern Europe have experienced similar shares of the episodes. The average episode of large inflows lasts approximately three and a half years, with little variation across regions in the typical length. With the exception of Asia, the size of the current account relative to the economy during these episodes is substantially larger in emerging markets than in advanced economies. The measure of capital inflows that we use to identify episodes of unusually large flows is deliberately general, capturing net inflows of all types apart from those initiated by the domestic government in each country. However, in Table 2, we look more closely at the 11

Table 1: Capital Inflows Episodes: Summary Statistics Number of Ave. Duration Ave. CA Episodes (% of total) (years) Deficit (%GDP) Total 155 3.5 4.7 Advanced Economies 54 (35.3) 3.4 3.2 Latin America 28 (18.3) 3.4 5.5 Asia 24 (15.7) 3.7 1.9 Eastern Europe 25 (16.3) 3.6 7 Middle East 5 (3.3) 3.8 9.6 Sub-Saharan Africa 17 (11.1) 2.9 5 Sources: IMF BoPS, Authors’ Calculations behavior of component flows in each episode. Overall half of these episodes coincide with unusually largegrossinflows, withthissharesignificantly higherforemerging economiesthan 13 for the rest. Portfolio flows—so-called hot money—are large in 37 percent of episodes, a share that is constant across country groups. In 45 percent of episodes, FDI flows were unusually large, and again this share does not vary substantially between country groups. Finally, large flows in the residual other flows category, which is primarily comprised of cross-border lending by banks but also includes trade credit, were present in just over 60 percent of the cases we study. This finding is consistent with, for example, recent work by Bruno and Shin (2014) documenting the important role played by banks in cross-border capital flows. Moreover, in this paper we study not only recent episodes but also episodes that took place in the late 1970s and early 1980s when bank lending played a relatively more important role in cross border capital flows. Bank flows also were a substantial part of the capital flows in the Eurozone in the 2000s. Table 3 examines the relationship between the capital inflows episodes that we identify, capital flow reversals, and sudden stops. Of the episodes of unusually large capital inflows that we study, 123 (77 percent) end in a reversal as defined by Calvo et al. (2004). Just 14 over 40 percent inflows episodes coincided with a sudden stop. Table 3 suggests that the probability than an episode ends up in a capital flows reversal is similar for advanced and emerging economics, while sudden stops occur somewhat more frequently in advanced economies. 13Here we use the same criteria to identify unusually large component flows that we used in identifying large net inflows. 14Recall that according to Calvo et al. (2004) and others (Rothenberg and Warnock, 2011; Forbes and Warnock, 2012), a reversal occurs when the year on year change in capital inflows is at least two standard deviations below the mean. A sudden stop occurs when a reversal coincides with an output contraction Calvo et al. (2004). 12

Table 2: Capital Inflows Episodes and Types of Capital Flows Advanced Emerging Other Economies Economies Economies Total Total Episodes: 54 67 34 155 Of which, coincide with: Large Gross Inflows 23 42 14 79 (% of Group Total) (42.6) (62.7) (41.2) (51) Large Portfolio Inflows 22 24 12 58 (% of Group Total) (40.7) (35.8) (35.3) (37.4) Large FDI Inflows 25 31 14 70 (% of Group Total) (46.3) (46.3) (41.2) (45.2) Large Other Inflows 28 45 24 97 (% of Group Total) (51.9) (67.2) (70.6) (62.6) Large Reserve Accumulation 26 37 23 86 (% of Group Total) (48.1) (55.2) (67.6) (55.5) Sources: IFS, WDI, Authors’ Calculations Table 3: Capital Inflows Episodes, Reversals, & Sudden Stops Advanced Emerging Other Economies Economies Economies Total Total Episodes: 54 67 34 155 Of which: Ending in Reversal 45 53 24 122 (% of Group Total) (83.3) (79.1) (70.6) (78.7) Of which: Ending in Sudden Stop 19 19 4 42 (% of Group Total) (35.2) (28.4) (11.8) (27.1) Sources: IFS, WDI, Authors’ Calculations 13

3 Event Study 3.1 Aggregate Economic Variables In this section we characterize the behavior of several macroeconomic variables during a typical episode of large capital inflows. To this end, we compute the mean and median path of a set of macroeconomic indicators across all our episodes. In order to capture both the buildup and end phase of each episode, we consider nine-year windows that begin two years before the start of each inflows episode. In general, this window captures the point at which the variables first diverge from their trend level as well as the trough of the post-boom drop. As we saw in the previous section, many of the episodes in our sample occur in the late 2000s, and thus a full six years of data are not available after the end of these episodes. To ensure that the patterns we uncover in this section do not reflect mere changes in the composition of the sample, we include here only episodes for which a full nine years of data are available. Asisstandardinmuch oftheliterature(e.g. Gourinchas et al., 2001;Mendoza and Terrones, 2008; Cardarelli et al., 2010), we focus on the cyclical component of each variable by looking at the deviations from an HP trend. In each of the graphs in this section, time zero marks the start of the episodes. Vertical lines mark the start and the average length of an inflows 15 episode, which is just over three years. Figure 5 paints a stark picture of how domestic variables behave during a typical episode of large capital inflows. First, large inflows are associated with an economic boom. In fact, at the peak of the typical episode GDP is around 2 percentage points above trend. The boom is driven by a significant rise in consumption, and by an even more marked increase in investment. The boom is also accompanied by a significant rise in private credit, suggesting the existence of a link between capital inflows and access to credit by the private sector. Both a rise in employment and in measured TFP contribute to the increase in production. However, since we measure TFP using a Solow residual, we cannot distinguish whether the rise in TFP is due to an improvement in productivity, or to increased capacity utilization 16 during the economic boom that accompanies episodes of large capital inflows. In contrast with the boom taking place during the inflows, the aftermath of the typical episodeoflargecapitalinflowsischaracterizedbyaneconomiccontraction. Infact,beginning 15This is slightly shorter than the average length in Table 1 because here we include only episodes with complete data. 16See Basu and Fernald (2001) for the evidence on the procyclicality of capacity utilization and the challenges it poses for measuring TFP over the business cycle. 14

Figure 5: Capital Inflows Episodes and the Domestic Economy 2 1 0 1− 2− GDP −2 −1 0 1 2 3 4 5 6 2 1 0 1− 2− Consumption −2 −1 0 1 2 3 4 5 6 01 5 0 5− Investment −2 −1 0 1 2 3 4 5 6 01 5 0 5− Private Sector Credit −2 −1 0 1 2 3 4 5 6 1 5. 0 5.− 1− 5.1− Employment −2 −1 0 1 2 3 4 5 6 1 5. 0 5.− 1− 5.1− Total Factor Productivity −2 −1 0 1 2 3 4 5 6 dnerT PH morf noitaiveD tnecreP Mean Median One SE Band Note: t=0 at start of capital inflows episode. Vertical lines mark start and average duration of episodes. Sources: IMF BoPS, WDI, UNIDO, ILO 15

with the fourth, or fifth in the case of private credit, year after the start of the episode all the variables, apart fromTFP, fall significantly below trend. Employment exhibits a particularly large fall, since the magnitude of its drop below trend after the end of the episode is larger than the pickup occurring at the start of the episode. This pattern suggests that the return of capital inflows to their long run trend might cause economic disruption, a point on which we will return in section 4.2. Figure 6: Capital Inflows Episodes and the External Sector dnerT PH morf noitaiveD tnecreP 4 2 0 2− 4− Real Exchange Rate −2 −1 0 1 2 3 4 5 6 PDG fo tnecreP 5 4 3 2 1 0 Current Account Deficit −2 −1 0 1 2 3 4 5 6 PDG fo tnecreP 4 3 2 1 0 Change in Reserves −2 −1 0 1 2 3 4 5 6 Mean Median One SE Band Note: t=0 at start of capital inflows episode. Vertical lines mark start and average duration of episodes. Sources: IMF BoPS, WDI, UNIDO, ILO Figure 6 examines the path of external variables during episodes of large capital inflows. Large capital inflows coincide with an appreciation of the real exchange rate, represented by a rise in the index plotted in Figure 6, peaking at just over two percent above its trend level late in the episode. This finding is consistent with the real exchange rate appreciations associated with credit booms (e.g. Gourinchas et al., 2001; Mendoza and Terrones, 2008), and with exchange-rate-based stabilization programs (V´egh, 1992; Rebelo and Vegh, 1995), which constitute a subset of the episodes we study here. The real exchange rate remains above its trend value for approximately five years, or two years longer than the length of an average episode. The current account deficit goes from an average of just under two percent of GDP prior to start of the episode to between five and six percent in the first two years after the start of the episode, before returning to its original level after five years. At the same time, foreign reserves increase in the period before the start of the episode to fall back 17 to 1% above trend during the average length of the episode. Hence, on average, the impact of the capital inflows on the current account is only partially offset by the accumulation of reserves by the central bank. 17We measurereserveaccumulationusingthe netchangeinofficialreservesfromthebalanceofpayments, which gives the increase in reserves net of valuation changes. 16

Figure 7: Capital Inflows Episodes and International Conditions dnerT PH morf noitaiveD tnecreP 2. 0 2.− 4.− 6.− Real Effective Federal Funds Rate −2 −1 0 1 2 3 4 5 6 xednI XIV 32 22 12 02 91 VIX Index −2 −1 0 1 2 3 4 5 6 Mean Median One SE Band Note: t=0 at start of capital inflows episode. Vertical lines mark start and average duration of episodes. Sources: IMF BoPS, WDI, UNIDO, ILO In Figure 7, we look at the international liquidity conditions during episodes of large capital inflows, as captured by two measures of financial conditions in the US. First, we take the US real interest rate as a proxy for the international interest rate. The typical episode is preceded by a period of low interest rates, with the real Fed Funds rate significantly below its HP trend. The US interest rate then rises to or slightly above its trend level, although the standard error bands indicate that the level of interest rates during these episodes varies substantially. We do not investigate here whether low international interest rates have a causal role in generating episodes of large capital inflows. However, the pattern of low rates preceding such episodes is consistent with panel data evidence from Fratzscher (2012) that U.S. interest rates are an important driver of portfolio flows, as well as with the VAR analyses by Bruno and Shin (2013) and Rey (2013) showing that lower U.S. interest rates drive increases in cross-border lending by banks. Second, we test whether prevalent attitudes towards risk in major financial markets vary around the episodes that we identify, using the VIX index as a measure of risk aversion (Figure 7, right panel). As the episode begins, the VIX is on average below its long run average (the horizontal line in the graph), indicating that risk aversion is lower than usual when episodes begin. Risk aversion rises during the first two years before returning to its 18 long run average around the time the typical episode ends. As with global interest rates, we do not examine in detail whether risk appetite is a cause of inflows episodes, but we do note that the pattern we uncover is consistent with the findings of Forbes and Warnock (2012) and Fratzscher (2012) as well as Rey (2013), who present evidence of a causal role 18We observe the same pattern if we employ an alternate measure of risk such as the spread between the yield on medium-grade corporate bonds (rated Baa by Moody’s) and that on highly rated (Aaa) corporate bonds. 17

for changes in risk appetite in driving cross-border capital flows. 3.2 Sectoral Allocation of Production Having characterized the aggregate behavior of the economy during our episodes, we now turn our attention to the sectoral allocation of production. In particular, we are interested in documenting how the composition of GDP and the allocation of productive resources across different sectorsbehaveduring largecapitalinflows. Aswasthecasewiththemacroeconomic variables we examined, we detrend the sectoral shares using an HP filter, because these exhibit clear time trends in nearly all countries in the sample. In advanced economies, the sectoral shares of tradables in general, and manufacturing in particular, fall steadily over time, reflecting a structural shift towards services. By contrast, the importance of tradables and manufacturing rises steadily over time in most emerging and developing economies. Figure 8 plots changes in the shares of gross value added produced by four sectors: agriculture, mining, services, and manufacturing. In the top left panel, we see that the share of agriculture in value added drops significantly during the typical episode and returns to its trend level when the episode ends. To the extent that agricultural products are tradable, this is consistent with two-sector small open economy models in which a consumption boom 19 is accompanied by a shift in production towards nontradable goods. However, the top right panel of Figure 8 provides some evidence that the share of mining rises above trend during episodes of large capital inflows. Since metals and hydrocarbons are tradable goods, this appears at odds with the idea that capital inflows episodes are associated with a shift out of tradables production. At the same time, the data show substantial heterogeneity, with particularly wide confidence intervals. We therefore suspect that some of the episodes in our sample correspond to periods in which funds from abroad are used to finance the development of mineral resources. Again consistent with the theoretical literature, the share of value added in services is on average slightly below its trend level before the episode begins, then rises to its trend level or slightly above for the duration of the typical episode. Finally, manufacturing value added is at or above its trend level at the start of these episodes, but drops steadily for four years before beginning to return to trend. The fall in manufacturing value added is consistent with, among others, Rebelo and Vegh (1995), Rodrik (2008), and Kalantzis (2014) who find that manufacturing value added typically falls during episodes of real exchange rate appreciation. However, precisely during 19See Rebelo and Vegh (1995) and Benigno and Fornaro (2014). 18

Figure 8: Capital Inflows Episodes and Sectoral Allocation of Value Added 4. 2. 0 2.− 4.− Agriculture Share of Value−Added −2 −1 0 1 2 3 4 5 6 3. 2. 1. 0 1.− 2.− Mining Share of Value−Added −2 −1 0 1 2 3 4 5 6 4. 2. 0 2.− 4.− Services Share of Value−Added −2 −1 0 1 2 3 4 5 6 2. 0 2.− 4.− Manufacturing Share of Value−Added −2 −1 0 1 2 3 4 5 6 dnerT PH morf noitaiveD stnioP egatnecreP Mean Median One SE Band Note: t=0 at start of capital inflows episode. Vertical lines mark start and average duration of episodes. Sources: IMF BoPS, WDI, UNIDO, ILO periods of real exchange rate overvaluation, the sectoral share of real value added may not give reliable information on the sectoral allocation of productive resources. Consider an episode of real appreciation. The domestic price level rises faster than the foreign price level, but the price of tradable goods will move together with international prices more closely than will the price of nontradable goods. As a result, episodes of real appreciation will tend to be periods in which the price of nontradables like services rises faster than the price of tradables like manufacturing. However, as discussed in Section 2, standard practice when using sectoral data for a wide sample of countries (including the WDI data we use here) is to deflate all sectors using the GDP deflator, due to the limited availability of data on 20 sectoral price changes. As a result, real value added in tradables, including agriculture and manufacturing, will mechanically tend to grow more slowly than real value added in services during periods of real appreciation. To have a better sense of how capital inflows affect sectoral production, we therefore look at the sectoral allocation of productive resources during the episodes we study. This allows us to determine the extent to which production is truly shifting, irrespective of movements in output prices. In particular, we examine employment in the manufacturing sector as a share 20An exception is Kalantzis (2014), who uses sectoral price deflators for a narrower sample of countries. 19

Figure 9: Capital Inflows Episodes and Sectoral Allocation of Resources 4. 2. 0 2.− 4.− Employment in Manufacturing −2 −1 0 1 2 3 4 5 6 1 5. 0 5.− 1− 5.1− Investment in Manufacturing −2 −1 0 1 2 3 4 5 6 dnerT PH morf noitaiveD stnioP egatnecreP Mean Median One SE Band Note: t=0 at start of capital inflows episode. Vertical lines mark start and average duration of episodes. Sources: IMF BoPS, WDI, UNIDO, ILO of total employment and investment in manufacturing as a share of total investment. Here we limit our analysis to manufacturing in order to maximize the number of capital inflows episodes included in the analysis, since data on the shares of employment and investment allocated to agriculture, mining, and services are not widely available. Figure 9 makes clear that the share of productive resources allocated to manufacturing drops during episodes of large capital inflows. In fact, while the share of manufacturing in both employment and investment is above trend when the episode begins, by the end of the episode both shares are significantly below trend. Hence, Figure 9 provides direct evidence of a reallocation of productive resources out of manufacturing, and presumably into nontradable sectors, during episodes of large capital inflows. In this sense, of the two country cases highlighted in the introduction to this paper, the case of Spain rather than that of Brazil is typical of countries experiencing unusually large capital inflows. We now consider whether the reallocation of resources across sectors is connected with two other dimensions: the extent to which the government accumulates foreign reserves during the episode, and the international liquidity conditions when the episode begins. Let us start with the accumulation of foreign exchange reserves by the central bank. Standard two-sector small open economy models predict that the allocation of productive resources between tradable and nontradable sectors should respond to changes in the current account, 21 rather than to capital inflows per se. Hence, theory suggests that, to the extent that 21See Rebelo and Vegh (1995) and Benigno and Fornaro (2014). See also Benigno and Fornaro (2012), which present a theoretical framework in which the accumulation of reserves by the central bank induces a 20

Figure 10: Capital Inflows Episodes and Sectoral Allocation, High and Low Reserve Accumulation 4. 2. 0 2.− 4.− Below Average Reserve Accumulation −2 −1 0 1 2 3 4 5 6 4. 2. 0 2.− 4.− Above Average Reserve Accumulation −2 −1 0 1 2 3 4 5 6 dnerT PH morf noitaiveD stnioP egatnecreP Employment in Manufacturing and Official Reserves Mean Median SE Band Note: t=0 at start of inflows episode; Sources: IMF, World Bank, UNIDO reserve accumulation by the central bank offsets the impact of capital inflows on the current account, we should expect the reallocation of resources out of manufacturing to be larger, when the accumulation of reserves by the central bank during an episode is smaller. Motivated by this insight, we compare the behavior of the share of employment in manufacturing in episodes with below-average reserve accumulation to those with above average 22 reserve accumulation. The results are illustrated by Figure 10. Where reserve accumulation is below average, the share of employment in manufacturing is on average at or slightly above trend when the episode begins, but drops significantly below trend during the second and third year of the episode before moving back towards its trent level four years after the episode begins. Moreover, the magnitude of the drop is much larger than was the case for the entire sample. By contrast, episodes in which reserve accumulation is above average show a rise in the share of employment in manufacturing as the episode begins. The share 23 then moves gradually back towards its trend level. Hence, the behavior of the share of employment in manufacturing suggests that the accumulation of reserves by the central bank might mitigate the contraction in manufacturing during episodes of large capital inflows, in line with the predictions of standard two-sectors small open economy models. The behavior shift of resources toward the tradable sector. 22Once again we measure reserve accumulation using the net change in official reserves from the balance of payments, which gives the increase in reserves net of valuation changes. 23These patterns hold when the median rather than the mean is used to divide episodes, and regardless of whether reserve accumulation is normalized by GDP or by the level of capital inflows. 21

Figure 11: Capital Inflows Episodes and Sectoral Allocation, High and Low International Interest Rates 4. 2. 0 2.− 4.− Below Average Fed Funds Rate −2 −1 0 1 2 3 4 5 6 4. 2. 0 2.− 4.− Above Average Fed Funds Rate −2 −1 0 1 2 3 4 5 6 dnerT PH morf noitaiveD stnioP egatnecreP Employment in Manufacturing and International Interest Rates Mean Median SE Band Note: t=0 at start of inflows episode; Sources: IMF, World Bank, UNIDO of manufacturing investment, on the other hand, shows no divergence between episodes with low versus high reserve accumulation (to conserve space, we do not include these graphs). We now turn to the role of the international liquidity conditions at the onset of the episodes that we identify. In general, easy access to credit from abroad generates a boom in consumption. While the increase in tradable consumption results in a current account deficit, nontradable consumption requires a shift of resources out of the tradable sector and into the production of nontradables (see Rebelo and Vegh, 1995; Benigno and Fornaro, 2014, for a detailed theoretical exploration of this mechanism). So the sectoral allocation of productive resources can be driven also by international financial conditions. Figure 11 compares the behavior of the share of employment in manufacturing in episodes characterized by below-average Federal Funds rate at the start of the episode, to those with above average Federal Funds rate. The left panel shows that, for those episodes that were preceded by below average Federal Funds rates, manufacturing employment drops throughout the duration of the typical episode before beginning to recover. Instead, where the Feds Funds rate is high when the episode begins, the share of manufacturing employment rises significantly before returning to its trend level around the time the typical episode ends. In addition, a nearly identical pattern emerges if we divide the episodes according to the level of the VIX just before the start of each episode (see Appendix C). Where risk aversion is low at the outset manufacturing employment drops, while where risk aversion is above average, manufacturing employment rises. Hence, the reallocation of employment out of manufac- 22

turing seems to be a feature of those episodes that take off when international liquidity is abundant. In the first part of this section, we showed that on average productive resources shift out of manufacturing during episodes of large capital inflows, and indeed the reallocation of investment out of manufacturing appears to be a general feature of period in which capital inflows are unusually large. However, we also find that the reallocation of employment out of manufacturing is not a universal feature of the episodes in our sample. Rather, employment shifts out of manufacturing during episodes in which reserve accumulation has been relatively low but moves very little in cases where governments actively purchase foreign assets. Employment also shifts out of manufacturing during episodes which begin at times of abundant international liquidity. In the next two sections, we show that this distinction–episodes in which reallocation occurs versus those where it does not—is particularly important, because the allocation of employment is significantly related to how the economy fares in the aftermath of large capital inflows. 4 The Aftermath of Large Capital Inflows 4.1 Capital Flow Reversals and Sudden Stops Policymakers often cite the risk that an episode of large capital inflows might create the conditions for a financial crisis and a recession as one of the key reasons why it is necessary to monitor and intervene in capital flows. In fact, the event study in Section 3 showed that on average episodes of large capital inflows set up the stage for a slump. In this section, we ask whether the behavior of several macroeconomic indicators, and in particular of the sectoral allocation of production, before or during the episode can provide any information about whether the episode is likely to end in a hard or a soft landing. We begin by testing how various economic variables are related to two broad measures of the outcome of each 24 episode: Whether or not the episode coincides with a reversal or a sudden stop. We model the probability that a sudden stop will occur during or immediately after episode i using a probit specification in which y is equal to one if a reversal (regression 1 in Table 4) i or a sudden stop (regression 2) occurs during episode ior in theyear immediately afterwards. Theresults arenearlyidentical ifwe employ alinear probabilitymodel ora logisticregression 24As in Section 2 we identify sudden stops using the methodology of Calvo et al. (2004). 23

model (these results are provided in Appendix C). Pr(y i = 1|X i ) = Φ (γ 1 INFLOWS i +γ 2 CREDIT i (1) ′ ′ ′ +γ ALLOCATION +γ FED FUNDS +γ POLICY ) 3 i 4 i 5 i Since episodes of large inflows are associated with credit booms, we first examine whether the size of the credit boom affects the probability of a sudden stop or reversal. In particular, we include in the regression the variable CREDIT , the average value of HP-detrended real i credit to the private sector during the episode i. The variable INFLOWS is the average i value of our HP-detrended capital capital inflows measure (the current account deficit plus reserve accumulation) during the episode. The vector ALLOCATION contains two variables: the average share of manufacturing in i total employment during the episode and the average share of manufacturing investment, measured as share of total investment. Once again, we measure the allocation variables as the deviation from their HP trends. In the previous section, we found that episodes of large capital inflows coincided with larger than normal shifts of resources out of the manufacturing sector. Here we examine whether these shifts render the economy vulnerable to a sudden stop. The event study also indicated that the real federal funds rate was on average lower than its trend level just before episodes of large inflows. Therefore FED FUNDS is a vector of i two variables: the average US effective Federal Funds rate, in real terms, in the three years prior to the start of each episode and the average value of the real Fed Funds rate during each episode. Finally, we include a vector of four variables (POLICY ) capturing the policies in place i beforeandduringtheseepisodes. Totest whetherpolicy-makers caneffectively guardagainst sudden stops by accumulating foreign reserves once capital inflows grow unusually large, we include the average purchase of new reserves during the episode, measured as a share of 25 GDP. We also examine whether holding a pre-existing stockpile of foreign reserves can benefit the economy by including in the regression the level of foreign reserves before the episodestarts. Finally, weinclude adummy variableequal tooneifthecountryhasafloating exchange rate regime at the start of the episode (constructed using data from Ilzetzki et al., 2008, and updated through 2012) as well as the Chinn-Ito measure of financial openness when the episode begins (Chinn and Ito, 2006). 25We use the change in reserves from the balance of payments, which captures the change in reserves net of valuation changes. 24

Table 4: Probit Regression Results Episode Characteristics, Reversals, and Sudden Stops Dependent Variable: Reversal Sudden Stop (1) (2) Capital Inflows1 0.086 0.141* (0.072) (0.076) 2 Private Credit -0.007 0.052*** (0.014) (0.019) Manuf. Employment3 -0.181 -0.496 (0.424) (0.532) Manuf. Investment3 -0.062 0.101 (0.142) (0.163) Fed Funds Rate4 0.08 0.135 (0.091) (0.108) Fed Funds Rate4 0.123 -0.214** Before Episode (0.09) (0.103) 5 Reserve Accumulation -0.143* 0.026 (0.086) (0.096) Initial Reserves5 0.035** -0.105*** (0.016) (0.031) Floating ER6 -0.382 -0.832** (0.349) (0.423) Financial Openness7 -0.009 0.187 (0.106) (0.14) Observations 91 91 Pseudo R-Squared 0.097 0.396 Robuststandarderrorsinparentheses;**p<0.01, ** p<0.05, * p<0.1. 1 Percentage points deviation from HP trend. 2Real, per capita terms; log deviation from HP trend. 3Share of total, Percentage points deviation from HP trend. 4Percentage Points. 5PercentofGDP.6Dummyforfloatingexchangerateregime,basedonIlzetzkietal. (2008). 7Chinn-Ito index of financial openness. Pre- and post-peak values are averages for 3 years before and after the year capital inflows peak. See Appendix 5 for data sources. The regression results reported in Table 4 show that the variables we consider do not provide 25

much information on whether the episode will coincide with a reversal of capital inflows. Although reserve accumulation and the level of reserves appear statistically significant, the model fit as captured by the pseudo R-squared is poor. Thus, it appears that neither the domestic or foreign macroeconomic conditions we consider, nor the policy variables we analyze, are systematically related to whether large inflows will end abruptly or smoothly. By contrast, the model appears much more informative about whether the episodes we examine will coincide with a sudden stop. A larger expansion of domestic credit is significantly 26 associated with an increased probability of a sudden stop. Even controlling for domestic credit, the capital inflows variable is also significantly related to the probability of sudden stops. This suggests that the presence of unusually large capital inflows puts the economy more at risk of a sudden stop than does a purely domestic credit expansion. This is consistent with the work of Caballero (2014) who finds that surges in capital inflows increase the risk of banking crisis even in the absence of a lending boom. We also find that episodes of capital inflows that start when the Fed Funds rate is low are more likely to end up in a sudden stop. Moreover, although reserve accumulation during the episode does not enter significantly, a higher pre-episode level of reserves is significantly associated with lower vulnerability to a sudden stop. Finally, a floating exchange rate significantly reduces the probability that a sudden stop will occur. Instead, we do not find evidence of an impact on the likelihood that a sudden stop occurs from the extent to which productive resources are reallocated across sectors during the inflows. 4.2 Economic Performance when Capital Flows Fall We now investigate the relationship between, on the one hand, what happens before and during large capital inflows and, on the other, macroeconomic performance in the aftermath of the episode. We saw in the previous section that macroeconomic and policy variables provided some information about the risk that a sudden stop would occur, and the literature has indeed shown that sudden stops have significant negative consequences for economic performance (Calvo and Reinhart, 2000; Gourinchas and Obstfeld, 2012). However, more than 70 percent of the episodes in our sample do not end in a sudden stop. We therefore 26Gourinchas and Obstfeld (2012) find that credit expansion increases the probability of both banking and currency crises in emerging markets, but to a lesser extent in EMEs. Thus our work confirms that this credit-crisis relationship holds once we restrict our sample to periods of large capital inflows. citeCR2000 document the close relationship between sudden stops and banking crisis. 26

estimate the following model y i = α+β 1 INFLOWS i +β 2 CREDIT i + (2) ′ ′ ′ β ALLOCATION +β FED FUNDS +β POLICY +ε 3 i 4 i 5 i i Where the dependent variable y is the average of a measure of economic performance after i,t the end of episode i. The dependent variables we consider are the average values of GDP, consumption, investment, employment, and TFP (all HP detrended) during the three years after the end of the episode. We use the same set of explanatory variables as in the previous section. We first examine whether the extent of the credit boom (CREDIT ) or the size of capital inflows i (INFLOWS ) affect economic outcomes after the episode. In the Section 3, we found that i episodes of large capital inflows coincided with larger than normal shifts of employment and investment out of the manufacturing sector. Here we examine whether these shifts (again measured by the vector ALLOCATION ) adversely affect economic performance after capi ital inflows come to an end. As in our analysis of sudden stops, FED FUNDS includes i the average US effective Federal Funds rate in both the three years prior to the start of each episode and during each episode. 27 And we again include a vector of variables POLICY i capturing the policies in place before and during these episodes, including reserve accumulation during the episode, the level of reserves before the start of the episode, and the exchange rate regime and degree of de jure capital openness in place when the episode begins. The coefficients on capital inflows in table 5 are always negative and generally significant. This indicates that the size of the capital inflows the economy receives during the episode is systematically related to how the economy fares once inflows come to an end. A larger credit boom during the episode also has a negative relationship with post-episode macroeconomic outcomes. This confirms that the episodes we examine are typical of credit booms more generally. However the capital inflows variable is significant even after we control for the size of the domestic credit boom. Therefore the negative impact of the booms in our sample on post-episode output is significantly larger than would be the case during a purely domestic credit boom. The positive and significant coefficient on the share of manufacturing employment in regression (1) indicates that less reallocation of employment away from manufacturing during the 28 episode is significantly associated with a less severe recession afterwards. Likewise, less 27Results are nearly identical if we use the VIX index as a measure of international liquidity conditions. See Appendix C 28Inalargemajorityofthe episodesinoursample,theshareoflaborthemanufacturingsectorfallsbelow 27

Table 5: Regression Results Episode Characteristics and Economic Performance Dependent Variable: GDP1 Consumption1 Investment1 Employment1 TFP1 (1) (2) (3) (4) (5) Capital Inflows -0.159 -0.474*** -0.665* -0.075 -0.293** (0.119) (0.139) (0.383) (0.111) (0.122) Private Credit -0.069*** -0.041 -0.260*** -0.033* -0.059*** (0.024) (0.032) (0.086) (0.019) (0.016) Manuf. Employment 1.499*** 1.812** 3.814** 1.273** -0.201 (0.548) (0.744) (1.896) (0.557) (0.438) Manuf. Investment -0.136 -0.078 -0.513 0.034 0.061 (0.232) (0.221) (0.807) (0.283) (0.206) Fed Funds Rate3 -0.151 -0.06 0.195 0.065 -0.001 (0.134) (0.152) (0.528) (0.124) (0.145) Fed Funds Rate3 0.138 0.212 0.419 0.201 0.179 Before Episode (0.122) (0.128) (0.442) (0.133) (0.111) Reserve Accumulation4 0.188 0.314* 1.203*** 0.093 0.392*** (0.127) (0.164) (0.42) (0.129) (0.133) 4 Initial Reserves 0.008 -0.018 -0.063 0.01 0.012 (0.026) (0.022) (0.082) (0.024) (0.018) Floating ER5 0.644 0.53 3.603** 0.974** 0.257 (0.408) (0.468) (1.498) (0.45) (0.432) Financial Openness6 -0.098 0.058 -0.268 -0.1 0.181 (0.155) (0.188) (0.594) (0.159) (0.179) Observations 91 90 87 83 83 R-Squared 0.356 0.336 0.447 0.258 0.483 Robust standarderrors in parentheses; ** p<0.01,** p<0.05,* p<0.1. Dependent variables are averagevaluesforthe3yearsaftereachepisodeends. 1Real,percapitaterms;logdeviationfrom 2 3 4 HP trend. Percentage points deviation from HP trend. Percentage Points. Percent of GDP. 5Dummy for floating exchange rate regime, based on Ilzetzki et al. (2008). 6Chinn-Ito index of financial openness. See Appendix 5 for data sources. 28

reallocation away from manufacturing is associated with higher consumption, investment, andemployment. Bycontrast, Table5 shows no systematic relationship between the shareof total investment allocated to manufacturing during episodes of large inflows and subsequent economic performance. These findings suggest that, once we control for other relevant factors, the sectoral allocation of labor is significantly related to economic performance in the post-boom period. These findings are related to the analysis of Giavazzi and Spaventa (2010), who discuss the importance of the sectoral allocation of production for current account sustainability. However, our results indicate that the allocation of labor is more informative regarding post-episode 29 performance than the sectoral allocation of investment. International liquidity conditions as measured by the Fed Funds rate do not appear significantly relatedtoeconomicperformance aftertheepisodeends. Thus our results indicatethat abundant international liquidity does not significantly affect macroeconomic variables once we control for the two channels through which it might affect the domestic economy—capital inflows and domestic credit conditions. Moreover, we saw in Section 3 that the extent of reallocation of employment away from manufacturing is greater in episodes that start during periods of low U.S. interest rates. Sectoral reallocation thus appears to be another channel through which abundant international liquidity affects macroeconomic outcomes in these episodes, but once we account for reallocation, U.S. interest rates themselves have no independent impact. Turning to the policy variables in our regression, reserve accumulation during episodes of largeinflowsisalwayspositivelyrelatedtopost-episodemacroeconomicoutcomes, andnearly always significantly so. By contrast, the level of official reserves prior to the start of the episodeisnever significant andfluctuatesinsign. Moreover, theevidence fortheeffectiveness of the other two policy measures we study is not particularly strong. A floating exchange rate is positively associated with post-episode performance, but significantly related only to investment and employment. Financial openness at the start of the episode, on the other hand, does not appear to affect subsequent economic performance. The relatively parsimonious specification we employ explains between one third and one half of the variation in the macroeconomic outcomes we analyze, and the size of the coefficients in itstrend(referbacktoFigure9). Whendiscussingourresultswethereforeinterpretcoefficientsasestimates of the impact of reallocation out of manufacturing. 29This result might be due to frictions to the reallocationof labor across sectors once the inflows subside. For instance, the combination of nominal wage rigidities and a fixed exchange rate prevents the fall in real wagesthatmightbeneededtoreallocatelaborinthetradablesectorsintheaftermathofanepisodeoflarge capitalinflows,andthusgenerateunemployment(seeSchmitt-Groh´e and Uribe(2011)andFornaro(2012)). 29

Table 5 are economically meaningful. For instance, in the mid-2000s Ireland experienced an episodeoflargecapitalinflows, duringwhichemployment inthemanufacturing sectorran0.4 percentage pointsbelow itsHPtrend. According totheresults inTable 5, thisreallocationof labor is typically associated in the aftermath of the inflows with GDP being 0.6 percentage points lower that it would have been without such reallocation, and investment being a 1.5 percent lower. Like the Eurozone periphery, countries in Eastern Europe received large capital inflows during the mid-2000s. In these countries (Poland, Hungary, Bulgaria, and the Baltic Republics) the share of the labor force in manufacturing actually rose to between 0.6 and 0.9 percentage points above its trend. Our results imply that this reallocation would typically coincide with post-episode GDP 0.9 and 1.3 percent higher than without reallocation, investment two or three percent higher, and employment between 0.8 and 1.1 percent higher. Of course, from our simple empirical model it is not possible to draw conclusions about the channels that generate a correlation between the share of employment in manufacturing and the behavior of macroeconomic variables in the aftermath of an episode of large capital inflows, nor about the directions of causality. However, we think that the relationships uncovered by this empirical analysis are suggestive enough to warrant further research, perhaps aiming at empirically testing some of the channels suggested by the theoretical literature. 5 Conclusion This paper has analyzed the experiences of 69 middle- and high-income countries that underwent episodes of large capital inflows between 1975 and 2010. A large majority of these episodes end in a sharp reversal of capital inflows, but less than a third of these reversals are sudden stops in which output contracts. Our event study shows that in the typical episode output rises initially but then drops below trend as capital inflows subside. This is also true of investment, consumption, and employment. A credit boom also accompanies the episodes in our sample, collapsing when the episodes end. Aggregate productivity follows a similar path, remaining below its trend level for morethanthree years after the episode ends. The episodes that we identify typically begininyearswhenUSinterestratesarebelowaverageandwhenriskappetiteinUSfinancial markets is higher than average. Large capital inflows also coincide with a shift of both labor and capital out of the manufacturing sector. While the reallocation of investment is a general feature of episodes of 30

large capital inflows, the reallocation of labor away from manufacturing is a phenomenon particular to episodes in which the accumulation of reserves by the central bank is low, as well as to episodes that begin during periods of abundant international liquidity. Our regression analysis reveals that post-episode economic performance is significantly and negatively relatednot onlyto thesize ofthecredit boomgenerated by capital inflows and the magnitude of those inflows, but also the extent to which labor moves out of manufacturing, with a stronger shift of labor out of manufacturing during the inflows episode associated with a sharper contraction in the aftermath of the episode. By contrast, international liquidity conditions and the allocation of investment are uninformative regarding the severity and length of the post-boom downturn. Our findings therefore indicate that policy-makers should monitor the sectoral allocation of labor during periods of exceptionally large capital inflows. In fact, a shift in employment out of manufacturing may signal increased risk of a hard landing. However, on a positive note, our results also indicate that foreign exchange reserves management might help policymakers in dealing with the labor reallocation out of manufacturing during episodes of large capital inflows. References Alfaro, L., Kalemli-Ozcan, S., and Volosovych, V. (2014). Sovereigns, upstream capital flows and global imbalances. Journal of European Economic Association, forthcoming. Aoki, K., Benigno, G., and Kiyotaki, N. (2010). Adjusting to capital account liberalization. CEP Discussion Papers dp1014, Centre for Economic Performance, LSE. Basu, S. and Fernald, J. (2001). Why is productivity procyclical? why do we care? In New Developments in Productivity Analysis, pages 225–302. University of Chicago Press. Benhabib, J. and Spiegel, M. M. (2005). Human capital and technology diffusion. In Aghion, P. and Durlauf, S. N., editors, Handbook of Economic Growth, volume 1, Part A, pages 935 – 966. Elsevier. Benigno, G. and Fornaro, L. (2012). Reserve accumulation, growth and financial crises. CEPR Working Paper 9224. Benigno, G. and Fornaro, L. (2014). The financial resource curse. The Scandinavian Journal of Economics, 116(1):58–86. 31

Broner, F., Didier, T., Erce, A., and Schmukler, S. L. (2013). Gross capital flows: Dynamics and crises. Journal of Monetary Economics, 60(1):113 – 133. Bruno, V. and Shin, H. S. (2013). Capital flows and the risk-taking channel of monetary policy. Working Paper 18942, National Bureau of Economic Research. Bruno, V. and Shin, H. S. (2014). Capital flows, cross-border banking and global liquidity. Review of Economic Studies. Forthcoming. Caballero, J. A. (2014). Do surges in international capital inflows influence the likelihood of banking crises? The Economic Journal. Forthcoming. Calvo, G. and Reinhart, C. (2000). When capital infows come to a sudden stop: Consequences and policy options. In Kenen, P. and Swoboda, A., editors, Key Issues in Reform of the International Monetary and Financial System, pages 175–201. International Monetary Fund. Calvo, G. A., Izquierdo, A., and Mejia, L.-F. (2004). On the empirics of sudden stops: The relevance of balance-sheet effects. Working Paper 10520, National Bureau of Economic Research. Cardarelli, R., Elekdag, S., and Kose, M. A. (2010). Capital inflows: Macroeconomic implications and policy responses. Economic Systems, 34(4):333 – 356. Chinn, M. D. and Ito, H. (2006). What matters for financial development? capital controls, institutions, and interactions. Journal of Development Economics, 81(1):163 – 192. Ciccone, A. and Papaioannou, E. (2009). Human capital, the structure of production, and growth. The Review of Economics and Statistics, 91:66–82. Converse, N. (2014). Uncertainty, capital flows and maturity mismatch. Mimeo. Forbes, K. J. and Warnock, F. E. (2012). Capital flow waves: Surges, stops, flight, and retrenchment. Journal of International Economics, 88(2):235–251. Fornaro, L. (2012). International debt deleveraging. CREI Working Paper. Fratzscher, M. (2012). Capital flows, push versus pull factors and the global financial crisis. Journal of International Economics, 88(2):341 – 356. {NBER} Global. Ghosh, A. R., Qureshi, M. S., Kim, J. I., and Zalduendo, J. (2014). Surges. Journal of International Economics, 92(2):266 – 285. 32

Giavazzi, F. and Spaventa, L. (2010). Why the current account may matter in a monetary union: Lessons from the financial crisis in the euro area. Discussion Paper 8008, Centre for Economic Policy Research. Gourinchas, P.-O. and Jeanne, O. (2013). Capital flows to developing countries: The allocation puzzle. The Review of Economic Studies. Gourinchas, P.-O. and Obstfeld, M. (2012). Stories of the twentieth century for the twentyfirst. American Economic Journal: Macroeconomics, 4(1):226–65. Gourinchas, P.-O., Valdes, R., and Landerretche, O. (2001). Lending booms: Latin america and the world. Econom´ıa, 1(2):47–99. Grossman, G. and Helpman, E. (1991). Innovation and Growth in the Global Economy. MIT press. Gupta, N. and Yuan, K. (2009). On the growth effect of stock market liberalizations. Review of Financial Studies, 22(11):4715–4752. Heston, A., Summers, R., and Aten, B. (2013). Penn World Table Version 7.1. Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania. Ilzetzki, E. O., Reinhart, C. M., and Rogoff, K. (2008). Exchange rate arrangements into the 21st century: Will the anchor currency hold? Kalantzis, Y. (2014). Financial fragility in small open economies: firm balance sheets and the sectoral structure. Working Paper 505, Banque de France. Klenow, P. and Rodr´ıguez-Clare, A. (1997). The neoclassical revival in growth economics: Has it gone too far? In Bernanke, B. S. and Rotemberg, J. J., editors, NBER Macroeconomics Annual 1997, Volume 12, pages 73–114. MIT Press. Koren, M. and Tenreyro, S. (2007). Volatility and development. The Quarterly Journal of Economics, 122(1):243–287. Kroszner, R. S., Laeven, L., and Klingebiel, D. (2007). Banking crises, financial dependence, and growth. Journal of Financial Economics, 84(1):187 – 228. Krugman, P. (1979). A model of innovation, technology transfer, and the world distribution of income. Journal of Political Economy, 87(2):pp. 253–266. 33

Levchenko, A. A., Ranci`ere, R., and Thoenig, M. (2009). Growth and risk at the industry level: The real effects of financial liberalization. Journal of Development Economics, 89(2):210 – 222. McMillan, M. S. and Rodrik, D. (2011). Globalization, structural change and productivity growth. Working Paper 17143, National Bureau of Economic Research. Mendoza, E. G. and Terrones, M. E. (2008). An anatomy of credit booms: Evidence from macro aggregates and micro data. Working Paper 14049, National Bureau of Economic Research. Obstfeld, M. and Rogoff, K. S. (1996). Foundations of international macroeconomics. MIT Press Books. Ostry, J. D., Ghosh, A. R., Habermeier, K., Chamon, M., Qureshi, M. S., , and Reinhardt, D. B. (2010). Capital inflows: The role of controls. IMF Staff Position Note 10/04, International Monetary Fund. Powell, A. and Tavella, P. (2012). Capital inflow surges in emerging economies: How worried should lac be? IDB Working Paper 326, Inter-American Development Bank. Rajan, R. G. and Subramanian, A. (2011). Aid, dutch disease, and manufacturing growth. Journal of Development Economics, 94(1):106 – 118. Rebelo, S. and Vegh, C. A. (1995). Real effects of exchange-rate-based stabilization: An analysis of competing theories. In Bernanke, B. S. and Rotemberg, J. J., editors, NBER Macroeconomics Annual 1995, Volume 10, pages 125–188. MIT Press. Reinhart, C. and Reinhart, V. (2009). Capital flow bonanzas: An encompassing view of the past and present. In Frankel, J. and Pissarides, C., editors, NBER International Seminar on Macroeconomics 2008, pages 9–62. University of Chicago Press. Reis, R. (2013). The portuguese slump and crash and the euro-crisis. Brookings Papers on Economic Activity. Rey, H. (2013). Dilemma not trilemma: The global financial cycle and monetary policy independence. Paper Presented at the 25th Jackson Hole Symposium. Rodrik, D. (2008). The real exchange rate and economic growth. Brookings Papers on Economic Activity, pages 365–412. 34

Rothenberg, A. D. and Warnock, F. E. (2011). Sudden flight and true sudden stops. Review of International Economics, 19(3):509–524. Schmitt-Groh´e, S. and Uribe, M. (2011). Pegs and pain. NBER Working Paper No. 16847. Tornell, A. and Westermann, F. (2002). Boom-bust cycles in middle income countries: Facts and explanation. IMF Staff Papers, 49:111–155. V´egh, C. A. (1992). Stopping high inflation: An analytical overview. Staff Papers - International Monetary Fund, 39(3):pp. 626–695. WEO (2011). World economic outlook. IMF Occasional Papers, International Monetary Fund. 35

Appendix A: Data Appendix A Data Sources Table A-1: Data Sources Variable Source Current Account IMF Balance of Payments Statistics Reserves IMF Balance of Payments Statistics Effective Fed Funds Rate FRED Baa-Aaa Corporate Bond Spread FRED VXO Index Bloomberg Real Exchange Rate WDI Output WDI Consumption WDI Investment WDI Credit to the Private Sector WDI Tradables Value-Added WDI Nontradables Value-Added WDI Total Employment ILO LABORSTA Manufacturing Employment UNIDO INDSTAT2 Manufacturing Investment UNIDO INDSTAT2 TFP Penn World Tables Exchange Rate Regime Ilzetzki, Reinhart, & Rogoff (2008) Capital Controls Chinn & Ito (2006) B Notes on the Construction of Selected Data Series Total Factor Productivity: We calculate total factor productivity (TFP) using data on output and investment obtained from the Penn World Tables (Heston et al., 2013). We estimate initial capital stock using the method described in Klenow and Rodr´ıguez-Clare (1997) and calculate capital stock for subsequent years using the annual values of investment obtainedfromthePennWorldTables. WeuseemploymentdatafromtheInternationalLabor Organization’s LABORSTA data set rather than the labor force data provided by the Penn World Tables. This ensures that fluctuations in TFP around episodes of large capital inflows are not the result of changes in the unemployment rate. We calculate aggregate total factor productivityusingstandardgrowthaccounting(e.g. asinBenhabib and Spiegel,2005). This methodology allows us to measure TFP in nearly all of the 69 countries in which we observe episodes of large capital inflows. The VIX: To measure the risk aversion of major international investors we use the VIX index. The VIX measures the implied volatility of S&P index options and thus reflects the price of risk in U.S. equity markets. When the price of risk and thus the VIX is low, it can 36

be inferred that risk aversion is low. More specifically, we use the “original” VIX index or VXO, which measures the implied volatility of options on the S&P100 and which is available since the late 1980s. To obtain a measure of risk aversion from 1970 to 1986 we regress the realized volatility of the S&P 100 onthe VXO for the post-1986period anduse the estimated coefficients to back-cast the VXO. 37

C List of Episodes Table A-2: Capital Inflows Episodes in Advanced Economies Ave. Curr. Acct. Country Start Year Peak Year End Year Episode Length Deficit (%GDP) United Kingdom 1987 1990 1990 4.0 3.7 Austria 1975 1979 1981 7.0 3.6 Austria 1995 1997 1998 4.0 2.5 Belgium 2008 2009 2009 2.0 1.0 Denmark 1985 1986 1987 3.0 4.2 Denmark 1997 1998 1999 3.0 -0.4 Denmark 2008 2008 2009 2.0 -3.2 France 1980 1981 1983 4.0 1.1 France 2006 2008 2008 3.0 1.1 Germany 1978 1979 1982 5.0 0.2 Germany 1991 1993 1995 5.0 1.2 Germany 1998 2002 2003 6.0 -0.0 Italy 1980 1981 1981 2.0 2.4 Italy 1988 1988 1991 4.0 1.4 Italy 2006 2009 2011 6.0 2.7 Netherlands 1977 1979 1980 4.0 -0.1 Netherlands 1992 1993 1993 2.0 -3.0 Netherlands 2000 2001 2002 3.0 -2.3 Netherlands 2008 2009 2009 2.0 -4.8 Norway 1976 1976 1977 2.0 11.4 Norway 1987 1988 1988 2.0 4.2 Norway 1996 1997 1999 4.0 -4.7 Sweden 1975 1976 1977 3.0 1.5 Sweden 1990 1991 1993 4.0 2.4 Canada 1975 1977 1978 4.0 3.9 Canada 1987 1989 1989 3.0 3.3 Canada 1998 1999 1999 2.0 0.5 Canada 2009 2010 2010 2.0 3.2 Japan 1980 1981 1981 2.0 0.3 Japan 1995 1995 1996 2.0 -1.7 Japan 2003 2004 2004 2.0 -3.4 Finland 1990 1993 1994 5.0 3.1 Greece 1999 1999 2000 2.0 6.7 Greece 2006 2010 2011 6.0 12.0 Ireland 1980 1982 1982 3.0 10.7 Ireland 1993 1994 1995 3.0 -2.9 Ireland 2006 2009 2009 4.0 4.2 Portugal 1981 1982 1982 2.0 12.8 Portugal 1989 1990 1991 3.0 0.3 Portugal 2008 2009 2010 3.0 11.4 Spain 1987 1988 1991 5.0 2.2 Spain 2005 2005 2008 4.0 9.0 Australia 1981 1983 1983 3.0 4.3 Australia 1986 1987 1990 5.0 5.1 Australia 2003 2004 2006 4.0 6.2 New Zealand 1984 1985 1986 3.0 11.4 Source: IMF, Authors’ Calculations 38

Table A-2: Capital Inflows Episodes in Advanced Economies (continued) Ave. Curr. Acct. Country Start Year Peak Year End Year Episode Length Deficit (%GDP) New Zealand 2005 2005 2007 3.0 7.0 Cyprus 1982 1983 1984 3.0 9.2 Cyprus 1989 1991 1992 4.0 6.2 Cyprus 1999 2000 2001 3.0 3.4 Cyprus 2005 2007 2008 4.0 9.1 Israel 1982 1983 1983 2.0 8.9 Israel 1993 1994 1997 5.0 4.3 Israel 2008 2010 2010 3.0 -2.8 Source: IMF, Authors’ Calculations 39

Table A-3: Capital Inflows Episodes in Emerging Economies Ave. Curr. Acct. Country Start Year Peak Year End Year Episode Length Deficit (%GDP) Turkey 1993 1994 1997 5.0 1.1 Turkey 2005 2007 2007 3.0 5.4 South Africa 1975 1976 1976 2.0 5.9 South Africa 1981 1983 1984 4.0 3.2 South Africa 1995 1996 1997 3.0 1.4 South Africa 2004 2005 2008 5.0 5.2 Argentina 1997 2000 2000 4.0 3.4 Argentina 2004 2006 2007 4.0 -2.3 Brazil 1995 1996 1996 2.0 2.6 Brazil 2000 2000 2001 2.0 4.0 Brazil 2007 2008 2011 5.0 1.5 Chile 1978 1979 1981 4.0 8.6 Chile 1990 1991 1997 8.0 2.8 Colombia 1981 1981 1982 2.0 6.6 Colombia 1993 1994 1997 5.0 4.7 El Salvador 1981 1982 1982 2.0 5.4 El Salvador 1989 1990 1990 2.0 3.8 El Salvador 2007 2007 2008 2.0 6.6 Mexico 1980 1980 1981 2.0 5.9 Mexico 1990 1990 1993 4.0 4.7 Peru 1994 1997 1997 4.0 7.1 Peru 2007 2009 2012 6.0 1.9 Lebanon 2008 2008 2009 2.0 16.7 Egypt 1979 1980 1982 4.0 6.6 Egypt 2005 2006 2010 6.0 -0.1 India 2006 2006 2007 2.0 0.8 Indonesia 1995 1995 1996 2.0 3.3 Korea 1979 1981 1981 3.0 6.2 Korea 1991 1993 1996 6.0 1.4 Korea 2009 2010 2011 3.0 -2.6 Malaysia 1981 1981 1983 3.0 11.4 Malaysia 1991 1991 1993 3.0 5.5 Malaysia 2003 2003 2004 2.0 -12.1 Pakistan 1992 1995 1996 5.0 5.1 Pakistan 2006 2008 2009 4.0 5.5 Philippines 1978 1981 1982 5.0 6.1 Philippines 1991 1995 1996 6.0 3.6 Philippines 2010 2011 2011 2.0 -3.1 Thailand 1989 1989 1991 3.0 6.6 Thailand 1994 1994 1996 3.0 7.2 Thailand 2005 2009 2010 6.0 -2.6 Vietnam 2007 2008 2008 2.0 9.9 Morocco 1976 1976 1977 2.0 15.4 Morocco 1981 1981 1982 2.0 12.0 Morocco 1990 1991 1991 2.0 1.1 Morocco 1999 2000 2001 3.0 -0.9 Tunisia 1977 1978 1978 2.0 9.4 Tunisia 1982 1983 1984 3.0 8.1 Tunisia 1992 1994 1994 3.0 6.5 Source: IMF, Authors’ Calculations 40

Table A-3: Capital Inflows Episodes in Emerging Economies (continued) Ave. Curr. Acct. Country Start Year Peak Year End Year Episode Length Deficit (%GDP) Tunisia 2006 2007 2008 3.0 2.7 Bulgaria 2006 2006 2008 3.0 22.6 Russia 1995 1996 1998 4.0 -1.1 Russia 2006 2006 2007 2.0 -7.4 China 1993 1995 1996 4.0 0.1 China 2003 2003 2005 3.0 -4.0 Ukraine 2005 2009 2010 6.0 2.2 Slovak Republic 2005 2006 2007 3.0 5.7 Estonia 2006 2008 2008 3.0 13.5 Latvia 2006 2007 2007 2.0 22.5 Hungary 1993 1994 1995 3.0 8.0 Lithuania 1995 1996 1998 4.0 9.4 Lithuania 2006 2008 2008 3.0 12.9 Slovenia 2001 2001 2002 2.0 -0.6 Slovenia 2007 2007 2008 2.0 4.8 Poland 1992 1994 1995 4.0 2.1 Poland 1998 2000 2000 3.0 5.8 Poland 2007 2008 2010 4.0 5.5 Source: IMF, Authors’ Calculations 41

Table A-4: Capital Inflows Episodes in Other Economies Ave. Curr. Acct. Country Start Year Peak Year End Year Episode Length Deficit (%GDP) Costa Rica 1980 1983 1983 4.0 11.7 Costa Rica 2006 2006 2007 2.0 5.4 Dominican Republic 1979 1979 1980 2.0 8.4 Dominican Republic 1991 1991 1993 3.0 4.0 Dominican Republic 2000 2000 2001 2.0 3.6 Dominican Republic 2005 2009 2011 7.0 5.8 Guatemala 1991 1992 1993 3.0 5.0 Guatemala 2000 2000 2001 2.0 6.1 Paraguay 1978 1980 1980 3.0 6.0 Paraguay 1986 1986 1987 2.0 11.8 Jamaica 1981 1983 1985 5.0 11.4 Jamaica 2000 2000 2001 2.0 6.2 Jordan 1991 1992 1992 2.0 12.4 Jordan 2005 2008 2009 5.0 12.2 Sri Lanka 1979 1983 1983 5.0 10.7 Sri Lanka 1991 1992 1995 5.0 5.5 Singapore 1980 1981 1982 3.0 10.4 Singapore 1990 1991 1993 4.0 -9.3 Singapore 2008 2011 2012 5.0 -19.0 Mauritius 1978 1981 1981 4.0 11.6 Mauritius 1988 1988 1990 3.0 3.9 Mauritius 1999 2000 2000 2.0 1.9 Namibia 1999 2000 2001 3.0 -1.5 Namibia 2008 2010 2011 4.0 -0.3 Belarus 2007 2008 2011 5.0 10.1 Albania 1988 1990 1994 7.0 5.3 Albania 2008 2009 2009 2.0 15.5 Croatia 1995 1996 1997 3.0 7.2 Macedonia, FYR 1998 1999 2000 3.0 4.2 Macedonia, FYR 2005 2010 2011 7.0 4.9 Romania 1990 1991 1992 3.0 6.0 Romania 2004 2005 2007 4.0 10.2 Source: IMF, Authors’ Calculations 42

Appendix B: Robustness Checks Figure B-1: Capital Inflows Episodes and Sectoral Allocation, High and Low VIX 4. 2. 0 2.− 4.− Below Average VIX −2 −1 0 1 2 3 4 5 6 4. 2. 0 2.− 4.− Above Average VIX −2 −1 0 1 2 3 4 5 6 dnerT PH morf noitaiveD stnioP egatnecreP Employment in Manufacturing and International Interest Rates Mean Median SE Band Note: t=0 at start of inflows episode; Sources: IMF, World Bank, UNIDO 43

Table B-1: Alternate Specifications: Episode Characteristics, Reversals, and Sudden Stops Specification: Linear Logit Dependent Variable: Reversal Sudden Stop Reversal Sudden Stop (1) (2) (3) (4) Capital Inflows1 0.017 0.014 0.145 0.221 (0.017) (0.02) (0.125) (0.135) Private Credit2 -0.001 0.011*** -0.01 0.091** (0.003) (0.004) (0.028) (0.04) Manuf. Employment3 -0.049 -0.098 -0.357 -0.723 (0.112) (0.093) (0.774) (1.031) Manuf. Investment3 -0.015 0 -0.091 0.127 (0.042) (0.033) (0.278) (0.33) 4 Fed Funds Rate 0.016 0.026 0.137 0.245 (0.026) (0.026) (0.161) (0.198) Fed Funds Rate4 0.028 -0.047* 0.195 -0.378** Before Episode (0.023) (0.026) (0.158) (0.192) Reserve Accumulation5 -0.033 0.01 -0.247 0.075 (0.021) (0.022) (0.154) (0.186) Initial Reserves5 0.007** -0.008** 0.061* -0.182*** (0.003) (0.004) (0.033) (0.065) Floating ER6 -0.111 -0.176* -0.727 -1.457* (0.096) (0.091) (0.636) (0.841) Financial Openness7 0.001 0.064** -0.015 0.322 (0.027) (0.028) (0.185) (0.257) Observations 91 91 91 91 R-Squared 0.087 0.33 0.0944 0.392 Robust standard errors in parentheses; ** p<0.01, ** p<0.05, * p<0.1. 1 2 Percentagepoints deviationfromHPtrend. Real,percapitaterms; logdeviationfromHP trend. 3Shareoftotal,Percentagepoints deviationfromHP trend. 4Percentage Points. 5Percent of GDP. 6Dummy for floating exchange rate regime, based on Ilzetzki et al. (2008). 7Chinn-Ito index of financial openness. Pre-andpost-peakvalues areaveragesfor3 yearsbefore andafter the year capital inflows peak. See Appendix 5 for data sources. 44

Table B-2: Alternate Specification: Episode Characteristics and Economic Performance Dependent Variable: GDP1 Consumption1 Investment1 Employment1 TFP1 (1) (2) (3) (4) (5) Capital Inflows -0.172 -0.479*** -0.647* -0.079 -0.282** (0.108) (0.122) (0.366) (0.108) (0.119) Private Credit -0.061*** -0.033 -0.255*** -0.035* -0.055*** (0.021) (0.029) (0.079) (0.019) (0.014) Manuf. Employment 1.581*** 2.026*** 4.501** 1.469** 0.029 (0.543) (0.719) (1.962) (0.608) (0.441) Manuf. Investment -0.186 -0.151 -0.655 -0.032 -0.003 (0.239) (0.208) (0.793) (0.313) (0.187) 3 VIX Index -0.041 -0.072 -0.123 -0.039 -0.063 (0.057) (0.066) (0.209) (0.05) (0.046) VIX Index3 0.156** 0.189** 0.339 0.05 0.128* Before Episode (0.076) (0.086) (0.266) (0.072) (0.067) Reserve Accumulation4 0.206* 0.315** 1.168*** 0.096 0.380*** (0.107) (0.139) (0.393) (0.117) (0.129) Initial Reserves4 0.006 -0.021 -0.071 0.004 0.005 (0.022) (0.02) (0.079) (0.022) (0.018) 5 Floating ER 0.680* 0.536 3.472** 0.884** 0.176 (0.406) (0.452) (1.579) (0.403) (0.423) Financial Openness6 0.062 0.208 -0.093 -0.114 0.249 (0.146) (0.172) (0.595) (0.168) (0.191) Observations 91 90 87 83 83 R-Squared 0.384 0.379 0.454 0.238 0.503 Robust standarderrors in parentheses; ** p<0.01,** p<0.05,* p<0.1. Dependent variables are average values for the 3 years after each episode ends. 1Real, per capita terms; log deviation fromHPtrend. 2PercentagepointsdeviationfromHPtrend. 3VXOindex,1986-2012,andbackcasted through 1970. 4Percent of GDP. 5Dummy for floating exchange rate regime, based on 6 Ilzetzki et al. (2008). Chinn-Ito index of financial openness. See Appendix 5 for data sources. 45