International Reserves, Credit Constraints, and Systemic Sudden Stops

K.7 International Reserves, Credit Constraints, and Systemic Sudden Stops Shousha, Samer F. Please cite paper as: Shousha, Samer F. (2017). International Reserves, Credit Constraints, and Systemic Sudden Stops. International Finance Discussion Papers 1205. https://doi.org/10.17016/IFDP.2017.1205 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1205 May 2017

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1205 May 2017 International Reserves, Credit Constraints, and Systemic Sudden Stops Samer F. Shousha NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.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 https://www.federalreserve.gov/econres/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at http://www.sssrn.com.

International Reserves, Credit Constraints, and Systemic Sudden Stops∗ Samer F. Shousha† May 4, 2017 Abstract Why do emerging market economies simultaneously hold very high levels of international reserves and foreign liabilities? Moreover, why, even with such huge amounts of international reserves, did countries barely use them during the Global Financial Crisis? I argue that including international reserves as an implicit collateral for external borrowing in a small open economy model subject to exogenous financial shocks can explain both of these puzzling facts. I find that the model can obtain ratios of international reserves and net foreign liabilities to GDP similar to those of Latin American countries. Additionally, the optimal policy implies that the government accumulates international reserves before a sudden stop and that there is a small depletion during it. Finally, an alternative policyofkeepinginternationalreservesconstantattheaveragelevelyieldsresults very similar to those of the optimal policy during sudden stops, highlighting the stabilizing role of international reserves even if central banks do not use them. JEL classification: F32, F34, F41 Key words: international reserves, emerging market economies, sudden stops, international crises ∗This paper is based on the second chapter of my dissertation at Columbia University. I am grateful to Martin Uribe, Stephanie Schmitt-Grohe, and Jose Alexandre Scheinkman for constant guidanceandsupport.Iwouldalsoliketothank,forveryusefulcommentsandsuggestions,SakiBigio, Patrick Bolton, Mariana Garcia, Tommaso Monacelli, Jaromir Nosal, Pablo Ottonello, Ricardo Reis, Ilton Soares, Jon Steinsson, Savitar Sundaresan, and seminar participants at Columbia University. The views in this paper are solely the responsibility of the author and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of any other person associated with the Federal Reserve System. All remaining errors are mine. †DivisionofInternationalFinance,BoardofGovernorsoftheFederalReserveSystem,Washington, D.C. 20551 U.S.A. Email: samer.f.shousha@frb.gov 1

”An economy which maintains an adequate level of reserves gives the rest of the world the assurance that it will honor its commitments in exceptional situations.” Banco Central de Chile (2011) 1 Introduction Although we have seen a remarkable increase in the hoarding of international reserves in emerging market economies, this practice has been the subject of an intense debate. Some authors argue that countries have over-invested in international reserves (Rodrik (2006)), while others see this strong buildup as the optimal response for the possible feedback effects of balance of payments crises (Mendoza (2006)). Moreover, most countries had a small and short-lived international reserves depletion during the Global Financial Crisis (GFC), which is at odds with the conventional wisdom that countries accumulate reserves to provide self-insurance against sudden stops. This paper proposes a new motive for international reserves accumulation, namely its role as implicit collateral for external borrowing in a small open economy subject to external financial shocks. Although policy-makers and financial market participants have often thought that international reserves can serve as collateral for external borrowing,thisrolehasnotyetbeenformallyevaluated.Todothat,Iincludeinternational reserves as collateral for external borrowing in a small open economy model with credit constraints, similar to those in Mendoza (2002) and Bianchi (2011). In this context, I want to understand whether the role of international reserves as collateral for foreign borrowing can explain their high levels in emerging economies and analyze their behavior and that of macroeconomic variables in crises in such an environment. Myframeworkshedssomelightonthepuzzlingfactthatemergingmarketeconomies hold very high levels of international reserves and foreign liabilities simultaneously and these holdings are positively correlated, as we can see in Figure 1.1 In my model, when the economy is hit by an external shock, there is a drastic reduction on the amount of output that can be pledged as collateral for external borrowing. However, as international reserves are very liquid assets, their collateral value is always the same independent of the state of global financial markets. Consequently, the government may choose to pay the cost of holding elevated levels of international reserves during 1. Thecorrelationbetweenthetwovariablesis0.4inthesampleof33countriesshowninFigure1. 2

normal times to relax the credit constraint when the economy is hit by an external financial shock.2 This policy action allows consumers to hold much more debt than would be possible otherwise and softens the drastic effect of negative financial exogenous shocks on consumption. Implicitly, international reserves serve as collateral for a credit line provided by foreign investors in periods when the country’s ability to borrow is heavily constrained by an external financial shock. Figure 1: Net Foreign Liabilities ex-IR and International Reserves (% of GDP) Note:Thedataarethesimpleaveragesampledannuallyfrom1991to2011.Allvariablesareexpressed in percentage points of GDP. Source:Authors’computationsbasedontheupdatedandextendedversionofthedatasetconstructed by Lane and Milesi-Ferretti (2007). Quantitative analysis show that the model does well in several dimensions. I find that we can obtain international reserves holdings close to the average internationalreserves-to-GDP ratio in Latin American countries and these results are robust to different parametrizations. Thus, when we consider the decision by a country to jointly hold foreign debt and international reserves, the government chooses to hold a significant amount of reserves even if we just allow for one-period debt. This result contrasts 2. Rodrik (2006) estimates the income loss due to reserve accumulation in developing countries to be close to 1% of GDP. 3

with those of Alfaro and Kanczuk (2009b) and Bianchi, Hatchondo, and Martinez (2016), who find that the optimal policy when you have only one-period debt is not to hold international reserves at all.3 Moreover, the optimal behavior during crises implies an increase in reserve holdings before a sudden stop and a small reduction during it, which coheres with what was observed in the GFC. Finally, an alternative policy of keeping reserves at a constant level equal to its average value yields results very similar to the optimal policy during sudden stops, highlighting the stabilizing role of reserves even if central banks don’t use them at all, as noted by De Gregorio (2011). I also provide a formal explanation for the behavior regarding international reserves during the GFC. Contrary to the results of almost all models that try to determine the adequate level of international reserves, most countries had a small and short-lived international reserves depletion during the GFC, rebuilding their stocks very quickly after that.4 Due to this fact, Aizenman and Sun (2012) stated that during the GFC the ”fearoflosingreservesseemstoplayakeyroleinshapingtheactualuseofinternational reserves by emerging markets”.5 Although this behavior could lead to a reevaluation of the role of international reserves as insurance against sudden stops, especially when we have a floating exchange rate regime, the financial resilience of emerging market economies during the GFC strongly suggests that having a high level of international reserves can help countries deal with sharp changes in global financial conditions in a much better way even if they are not heavily used.6 This paper is robust to the critique about the assumption of the use of international reserves as collateral because they are legally protected against attachment by creditors under different law systems.7 In my setup, the idea behind the assumption that international reserves serve as an implicit collateral is not based on any contractual arrangement but relies on the fact that there would be strong reputational effects if the government or the private sector defaulted in the presence of international reserves that could be used to comply with these obligations. In fact, Aizenman (2009) and 3. Bianchi,Hatchondo,andMartinez(2016)findthathavinglong-durationbondsiskeytoobtaining significant levels of foreign liabilities and international reserves simultaneously. 4. Dominguez(2012)alsoexamineshowcountriesmanagedinternationalreservesduringtheGFC, showing that they were reluctant to use them. 5. Bussiere et al. (2015) also state that ”international reserves should be viewed as being akin to ‘nuclearweapon’havingadeterrenteffect,ratherthantotruegunpowder,tobeusedinintervention.”. They also conclude that the level of short-term debt is the main determinant of the level of reserves, which supports my conclusions. 6. De Gregorio (2013) also points out the important role of international reserves in the resilience displayed by emerging market economies during the GFC. 7. See Panizza, Sturzenegger, and Zettelmeyer (2009). 4

De Gregorio (2013) point out that the credibility of Brazil’s, Mexico’s and Korea’s anti-crisis measures unveiled in the second half of 2008 was reinforced by their massive stock of reserves. Related Literature. This paper is related to the literature that tries to explain international reserves accumulation in emerging market economies. Some authors argue that international reserves accumulation has a mercantilist motive and is related to competitiveness in international trade. Dooley, Folkerts-Landau, and Garber (2004a) attribute this motivation specifically to China, where a strategy of export promotion and consequently the desire for a depreciated currency leads to sizable reserve accumulation. Moreover, international reserves could serve as collateral for foreign direct investment and all the learning externalities that might come with investment in the tradable sector (Dooley, Folkerts-Landau, and Garber (2004b)). To address this issue, Korinek and Serv´en (2016) build a stylized model that incorporates learning-byinvestment externalities and a capital-intensive tradable goods sector. Their calibrated model suggests that the welfare benefits of reserve accumulation are outweighed by its costs for standard parameter values. This work contributes to this literature by introducing another way by which international reserves can serve as an implicit collateral; namely, for foreign borrowing by private agents. Anotherstrandoftheliteratureseesinternationalreservesaccumulationasaformof precautionarysavingsagainstsuddenstopsandrolloverrisk(AizenmanandLee(2007); Durdu, Mendoza, and Terrones (2009); Alfaro and Kanczuk (2009b); Jeanne and Ranciere (2011); Bianchi, Hatchondo, and Martinez (2016); Hur and Kondo (2016)). This paper departs from this literature by introducing international reserves as an implicit collateral for foreign borrowing in a small open economy model subject to exogenous financial shocks. I show that this feature leads to optimal ratios of international reserves and foreign liabilities to GDP that are similar to what we observe in Latin America. Moreover, this result is obtained in an environment where there is only short term debt, contrary to the findings of Alfaro and Kanczuk (2009b) and Bianchi, Hatchondo, and Martinez (2016), who argue that in the presence of only nonstate-contingent short-term debt the optimal holdings of international reserves are very close zero. In fact, as shown by Broner, Lorenzoni, and Schmukler (2013), emerging economies borrow mostly short term because investors charge a higher risk premium on long-term bonds.8 Additionally, another implication of the models previously stud- 8. Broner, Lorenzoni, and Schmukler (2013) analyze a database on sovereign bond prices, returns and issuance at different maturities for eleven emerging economies - including Argentina, Brazil, 5

ied in the literature is that, in the event of a crisis, countries should heavily reduce their international reserves holdings, which is at odds with the behavior we see in the data for countries with a floating exchange rate, especially during the GFC. In my framework, the international reserves depletion is small and short lived because there is a trade-off between using the reserves today to increase tradable consumption and keeping international reserves untouched to be able to borrow more tomorrow. Finally, extending the precautionary approach, some authors argue that international reserves accumulation is a tool for managing domestic financial instability and smoothing exchange rate fluctuations in the presence of underdeveloped domestic financial markets. Obstfeld, Shambaugh, and Taylor (2010), for example, build a model based on the idea that, in a double drain scenario, domestic capital flight is financed through withdrawals of domestic bank deposits. In their setup, the growth of banking systems and financial markets in emerging market economies explains almost all of the recent buildup of reserve holdings. Dominguez (2010) also focuses on the implications of underdeveloped capital markets for emerging market economies. Following Caballero and Krishnamurthy (2001), she shows that underdeveloped capital markets lead to an undervaluation of international resources by the private sector, increasing the exposure of these economies to capital shortfalls. In this environment, international reserves accumulation can mitigate the costs of this excessive exposure, working as insurance against sudden stops. I contribute to this literature by developing a framework where the quantitative implications of the role of international reserves as implicit insurance for private-sector foreign borrowing can be evaluated. I also show that this feature explains international reserves holdings of Latin American countries over the last 25 years. Layout. The rest of the paper is organized as follows. Section 2 illustrates the mechanism behind my results in a simple environment. Section 3 builds a quantitative business cycle model that includes international reserves as collateral for foreign borrowing. Section 4 details the calibration and simulation of the model, presents its unconditional moments and the behavior of different variables during crises, and analyzes the implications of an alternative policy where we keep international reserves at constant levels for all periods. Section 5 evaluates the robustness of the results to some specific parameters. Section 6 concludes. Colombia, and Mexico - during the period from 1990 to 2009. 6

2 Three-Period Economy I present a simple model to provide some intuition on how the mechanism works. The economylastsforthreeperiods,receivesadeterministicendowmentonlyinthelastone, and might face an exogenous shock in the intermediate period that limits the amount of borrowing to a multiple of the international reserves held by the government, which gives a motive for reserve accumulation. I present the full model in the next section. 2.1 Environment The economy lasts for three periods t = 0,1,2. There is only one good and a representative agent receives a deterministic sequence of endowments given by y = y = 0 and 0 1 y > 0. The household only values consumption in periods 1 and 2 and maximizes the 2 discounted expected future flow of utility using a subjective discount factor β ∈ (0,1). Households can borrow from abroad subject to an exogenously determined interest rate r. I assume for simplicity that β(1 + r) = 1 and the utility function is given by u(c) = ln(c). The economy is subject to a ”sudden stop shock” in period 1. If the sudden stop shockmaterializes,borrowinginperiod1islimitedtoamultipleκir oftheinternational reserves held by the government. A sudden stop occurs with probability π(cid:15)[0,1]. At t = 0, the government can accumulate reserves through lump-sum taxation on households. The only reason to accumulate international reserves is to use them as collateral for external borrowing if the economy is faced with a sudden stop shock in period 1. Let b denote the bond purchased by agents in period t. A negative value means t+1 an issuance of bonds by households. The budget constraints for each period for the whole economy are given by IR ≤ −b 1 1 c ≤ (1+r)b −b +IR 1 1 2 1 b (s ) ≥ −κirIR 2 0 1 c −y 2 2 b (s ) ≥ − 2 1 1+r c ≤ y +b (1+r) 2 2 2 where s denotes a sudden stop state and s denotes a normal state. Figure 2 shows 0 1 7

the timing of decisions and correspondent utilities at each period for the simple model when all budget constraints are satisfied with equality. Figure 2: Timing of Decisions and Utilities - Simple Model 2.2 Analytical Results A social planner maximizes the expected utility by choosing the optimal level of international reserves and consumption at t = {1,2}. If the economy is not subject to a sudden stop shock, the solution is trivially c∗ = c∗ = y /(2+r). However, a sudden stop 1 2 2 may prevent agents from borrowing in period 1 if there are no international reserves in place. Substituting the budget constraints for different states into the utility function, the problem for the social planner is given by max π{ln[(κir−r)IR ]+βln[y −(1+r)κirIR ]}+(1−π){ln[−b (s )−rIR ]+βln[y +b (s )(1+r)]} 1 2 1 2 1 1 2 2 1 IR1,b2(s1) Thecostsandbenefitsofholdingreservesareclearfromthesocialplanner’sproblem:on the one hand, there is the cost of carrying reserves from period 0 to 1, which is given by rIR as the agents must issue bonds to finance the acquisition of international reserves; 1 on the other hand, if the economy is hit by a sudden stop shock, it allows an increase 8

in consumption by κirIR in period 1. 1 Using the first-order conditions, the optimal level of reserves holdings is given implicitly by the following expression: y −(2+r)kirIR (cid:18) 1−π (cid:19) r(2+r)IR 2 1 1 − = 0 y −κirIR (1+r) π y −r(1+r)IR 2 1 2 1 which has as solution a linear function in y 2 IR = K(r,π,κir)y 1 2 where the constants are given by9 K −K K(r,π,κir) = 2 3 K 1 K = 2r(2+r)(1+r)κir 1 K = (2+r)κirπ +r(2+r−π) 2 (cid:113) K = K2 −2πK 3 2 1 and has the following properties (i) IR is strictly increasing in π 1 (ii) IR is strictly decreasing in r 1 (iii) IR is strictly decreasing in κir 1 (iv) IR is strictly increasing in y as K(cid:15)(0,1/2] 1 2 Consequently, if the probability of a sudden stop is high, the optimal level of international reserves is also higher as insurance against it. Moreover, if the opportunity cost of holding international reserves is high, the optimal international reserves holdings are lower. Additionally, if the collateral value of international reserves is high, the optimal holdings are also lower as we need less international reserves to get the same foreign borrowing level. Finally, if output will be higher in the future, it pays to carry more international reserves as insurance against the bad state. 9. K(r,π,κir) = (K +K )/K violates the feasibility conditions in the economy as it is greater 2 3 1 than 1, so we have as a unique solution K(r,π,κir)=(K −K )/K . 2 3 1 9

In the next section, I present a detailed model that will allow me to evaluate the quantitative implications of the role of international reserves as collateral to account for the level of international reserves holdings in emerging market economies and their behavior during crises. 3 Model This section presents a small open endowment economy where foreign creditors constraint the amount that they lend to a share of tradable output and a multiple of the international reserves held by the government. In this setting, the main purpose of international reserves is to facilitate external borrowing when the economy is hit by an exogenous shock that drastically reduces the amount of output that can be pledged as collateral. This way, the government faces a trade-off between the benefits of keeping international reserves that serve as collateral for foreign borrowing in bad times and the cost of carrying this stock of reserves, as we saw in the simple model in section 2. After detailing the model, I present both the competitive equilibrium and the socially optimal one. 3.1 Theoretical Framework I model a small open endowment economy where the preferences of the representative consumer are represented by a time-separable utility function, (cid:40) (cid:41) ∞ (cid:88) E βtU(c ) (1) 0 t t=0 where β ∈ (0,1) is a subjective discount factor. The consumption basket is a CES aggregator with elasticity of substitution η between tradable cT and nontradable goods cN, t t c t ≡ A(cT t ,cN t ) = [ω(cT t )−η +(1−ω)(cN t )−η]− η 1 Every period, consumers receive an endowment of traded yT and nontraded yN t t goods. Markets of contingent claims are incomplete so consumers can only trade oneperiod bonds on international capital markets.10 The face value of these bonds specifies 10. I limit my analysis to short-term debt instead of long-term debt as in Bianchi, Hatchondo, and 10

the amount that will be paid in the next period, b . I normalize the price of tradt+1 ables to 1 and define the relative price of nontradables as pN. I also assume that t the government accumulates international reserves through lump-sum taxation τ . The t household’s budget constraint is consequently cT +pNcN +b +τ = yT +pNyN +(1+r )b (2) t t t t+1 t+1 t t t t t and as the government runs a balanced budget and only taxes agents to accumulate international reserves, its budget constraint is given by τ = ∆IR (3) t+1 t+1 The central assumption of the model is that creditors constrain the amount that they lend to a fraction κT of tradable income plus κir times the total stock of internat tional reserves b ≥ −[κTyT +κirIR ] (4) t+1 t t t where κT is an exogenous parameter that represents the state of international finant cial markets. I assume that both κT and r can take two different values, κT,N and rN, t t which are related to normal times, and κT,C and rC, which are related to disruptions in financial markets, capturing the feature that extreme capital flows episodes are significantly related to global risk, as we can see for example in Calvo (2005) and Forbes and Warnock (2012).11 The level of international reserves is taken as given from the perspective of the households. We can think of this borrowing limit as being the result of an incentive constraint coming from information asymmetries between borrowers and lenders and the presence of underdeveloped financial markets, which leads to limited enforcement. For simplicity, I assume that the borrowing limit is exogenously given. The possibility of using international reserves as collateral has been challenged by different authors such as Alfaro and Kanczuk (2009b). However, although central bank Martinez(2016)basedontheresultsofBroner,Lorenzoni,andSchmukler(2013).Theyarguethatthe predominance of short-term debt in developing countries happens because investors charge a higher risk premium on long-term bonds, and this relative cost increases even more in a crisis, making it much cheaper for emerging markets to borrow short-term. Alfaro and Kanczuk (2009a) also show that the optimal structure for emerging market economies is usually to have only short-term debt, although in their model this arises from the fact that the costs of defaulting increase more than the benefits when maturity increases. 11. Eggertsson and Krugman (2012) also have a model where views about safe levels of leverage change abruptly over time, an event they call a Wile E. Coyote moment based on the famous Road Runner cartoon. 11

assets are legally protected against attachment by creditors under the U.S. Foreign Sovereign Immunities Act of 1976 and comparable laws, the argument for the inclusion of international reserves as collateral relies on the reputational costs of a default in external borrowing by the government or the private sector in the presence of international reserves that could be used to fulfill these obligations. In practice, we usually see a positive correlation between the stocks of international reserves and short-term foreign debt. Dominguez (2012), for example, find that countries that accumulated larger stocks of reserves prior to the GFC also had higher short-term-debt-to-GDP ratios. I will show later that this is also the case for the crises episodes I study in this paper. 3.2 Competitive Equilibrium Households choose {cT,cN,b } to maximize expected utility (1) subject to the t t t+1 t≥0 budget constraint (2) and the borrowing limit (4), taking b , pN, τ , IR , κT and r 0 t t+1 t t t as given. Defining G(cT,cN) ≡ U(cid:48)(c )A (cT,cN), the first-order conditions are t t t 1 t t G(cT,cN) = λ (5) t t t (cid:18) 1−ω (cid:19)(cid:18) cT (cid:19)η+1 pN = t (6) t ω cN t λ = β(1+r )E λ +µ (7) t t t t+1 t µ ≥ 0, µ [b +κTyT +κirIR ] = 0 (8) t t t+1 t t t Market clearing conditions are given by cN = yN (9) t t τ = ∆IR (10) t+1 t+1 Definition 1 (Decentralized Competitive Equilibrium): A competitive equilibrium is a set of processess {cT,b ,µ } satisfying t t+1 t t≥0 G(cT,yN) = β(1+r )E G(cT ,yN )+µ (11) t t t t t+1 t+1 t b = yT −cT +(1+r )b −∆IR (12) t+1 t t t t t+1 (cid:2) (cid:3) b ≥ − κTyT +κirIR (13) t+1 t t t 12

(cid:2) (cid:3) µ ≥ 0, µ b +κTyT +κirIR = 0 (14) t t t+1 t t t given processes {yT,yN,IR } and the initial condition b . t t t t≥0 −1 3.3 Socially Optimal Equilibrium So far I have stated that households solve their optimization problem taking the stock of international reserves as exogenously given. To determine the optimal amount of international reserves at each period t, I write the optimization problem in recursive form and solve the social planner’s problem. The social planner’s problem consists in choosing {IR , b , cT} given {IR , b , yT, yN,κT,r } to maximize expected utility t+1 t+1 t t t t t t t subject to the budget constraint and the collateral constraint V(IR,b,y,κT,r) = max u(c(cT,yN))+βE{V(IR(cid:48),b(cid:48),y’,κT(cid:48),r(cid:48))} IR(cid:48),b(cid:48),cT subject to cT +b(cid:48) +IR(cid:48) = yT +b(1+r)+IR b(cid:48) ≥ −(κTyT +κirIR) t The first order conditions associated with this problem are now G(cT,yN) = β(1+r )E G(cT ,yN )+µ (15) t t t t t+1 t+1 t G(cT,yN) = βE {G(cT ,yN )+µ κ } (16) t t t t+1 t+1 t+1 ir (cid:2) (cid:3) µ ≥ 0, µ b +κTyT +κirIR = 0 (17) t t t+1 t t t Note that the competitive and the socially optimal equilibria have the same Euler equation and differ only because now the planner also chooses the optimal level of international reserves through equation (16). Thus, to implement the social planner’s equilibrium as a competitive equilibrium, the planner chooses the optimal IR given t+1 current conditions and then finances it through lump-sum taxation of the households making τ = ∆IR . t+1 t+1 13

4 Quantitative Analysis Thissectioncalibratesandsimulatesthemodel,showingthatitcanyieldinternationalreserves-to-GDP ratios close to what we see in practice. I also find that the cyclical behaviors of the current account and net foreign liabilities excluding international reserves are very close to what we observe in practice, while that of international reserves is somewhat different. Moreover, the optimal policy leads to international reserves accumulation before a sudden stop and a small depletion during it, which is close to what we see in the data. Finally, I evaluate the behavior of the model under a simpler passive rule for international reserves accumulation where the central bank keeps international reserves levels constant and find that the behavior of consumption in crises under the passive rule is very similar to what is obtained under the optimal policy. 4.1 Long-Run Business Cycle Moments in the Data I begin the analysis by looking at the main statistics regarding international reserves, net foreign liabilities excluding international reserves, and current account balance for the larger Latin American countries, shown in Table 1. As we can see, the average ratio of international reserves to GDP is close to 10% while that of net foreign liabilities exinternational reserves is 36%. Moreover, international reserves are acyclical while the other variables are countercyclical. Table I Summary statistics - Latin America (% of GDP) Average Std Autocorr. Correl(y) International Reserves 9.9% 2.1% 0.55 0.07 Net Foreign Liabilities ex-IR 36.0% 11.9% 0.66 -0.31 Current Account -1.6% 2.5% 0.68 -0.40 Note: The data are the simple average of the indicators for the five main Latin American countries (Argentina, Brazil, Chile, Colombia and Mexico). To calculate the standard deviations and correlations, I detrend the ratios of the log of Real GDP, International Reserves to GDP and Net Foreign Liabilities excluding International Reserves to GDP taking out a linear and a quadratic trend. The Current-Account-to-GDP ratio is not detrended, as it is stationary. The data are from the World DevelopmentIndicatorsdatabasefromtheWorldBank,andtheupdatedandextendedversionofthe dataset constructed by Lane and Milesi-Ferretti (2007) complemented by the updated international capital flows database constructed by Alfaro, Kalemli-Ozcan, and Volosovych (2014). The data are sampled annually from 1991 to 2015. 14

4.2 Sudden Stops Episodes Following Calvo, Izquierdo, and Mejia (2008) and Alberola, Erce, and Serena (2016), I focus on systemic sudden stops, i.e., episodes triggered by an exogenous financial shock.12 I use the JP Morgan EMBI Global Index to identify periods of global financial stress in emerging market economies. These periods are defined as quarters where there is a spike in the EMBI Global spread with respect to its two-year moving average. This way, I have four global financial stress events iny 1995, 1999, 2002 and 2009, which are the well-known Tequila, Russian-Asian, Argentine, and Global Financial Crises. Using these global crisis dates we can then identify the sudden stop episodes, which are those dates where the country experiences a one standard deviation reversal in the current account conditional on being in a global crisis year. The methodology yields eight sudden stop episodes for the five Latin American countries studied in this paper. The list of episodes is in Table 2. Table II Sudden Stops Episodes Country Years of Sudden Stops Argentina 1995, 2002 Brazil 2002 Chile 1999, 2009 Colombia 1999 Mexico 1995, 2009 Figure 3 shows the average behavior of the ratios of the current account, international reserves, and net foreign liabilities excluding international reserves to GDP in crises. The behavior of these variables is close to what was obtained in previous works by Eichengreen, Gupta, and Mody (2008) and Jeanne (2007). The ratios of both international reserves and net foreign liabilities excluding international reserves to GDP increase in the onset of a sudden stop episode and decrease afterward. The real exchange rate appreciates before the sudden stop, suffers a strong depreciation during it, and stays at this more depreciated value afterward. 12. Calvo, Izquierdo, and Mejia (2008) argue that focusing on systemic sudden stops is desirable because they exclude idiosyncratic crises that can result from factors such as political turmoil and disasters.TheseidiosyncraticcriseshaveseveraldifferentfeaturescomparedtotheonesIisolatehere. 15

Figure 3: Macro Dynamics around Sudden Stops Events Note:Thefive-yearwindowiscenteredonasuddenstopoccurringattimet.Thelistofcountriesand sudden stops is given in Table 2. All variables are expressed in percentage points of GDP except for the Real Exchange Rate. Source: Authors’ computations based on the World Bank World Development Indicators database and the updated and extended version of the dataset constructed by Lane and Milesi-Ferretti (2007) complemented by the updated international capital flows database constructed by Alfaro, Kalemli- Ozcan, and Volosovych (2014). 4.3 Functional Forms and Calibration The utility function has a constant relative risk aversion (CRRA), ie c1−σ −1 U(c) = 1−σ The endowment process follows a VAR(1): log(y ) = ρlog(y )+(cid:15) t t−1 t with |ρ| < 1 and (cid:15) ∼ N(0,V). I use an average of the process estimated for t 16

Argentina, Brazil, Chile, Colombia and Mexico13, which yields as ρ and V (cid:34) (cid:35) 0.920 −0.314 ρ = 0.277 0.573 (cid:34) (cid:35) 0.00248 0.00142 V = 0.00142 0.00143 I discretize the process into a first-order Markov process with four grid points for each shock using the methods of Terry and Knotek (2011), which allows for arbitrary error covariance structures.14 κT can take two values, κT,H, which is related to normal times, and κT,L, which t is related to disruptions in international financial markets. The probability of entering a period of disruptions in international financial markets is given by π while the probability of going back to normal times is given by ψ. All of the benchmark parameter values can be seen in Table 3. A period in the model refers to a year. Table III Calibrated Parameter Values Parameter Value Source/target Interest rate in normal times rN =0.03 Sample average Interest rate in financial distress rC =0.08 Sample average Risk aversion σ =2 Standard value Atemporal elasticity of substitution 1/(1+η)=0.8 Conservative value Weight on tradables in CES ω =0.23 Share of tradable output = 23% Discount factor β =0.932 Average NFL ex IR-GDP ratio = 36.0% Probability of entering financial distress π =0.2 CA reversal = 3.9% Probability of going back to normal times ψ =0.4 CA recovery = 1.4% yT credit coefficient in financial distress κT,L =0.2 CA standard deviation = 2.5% IR credit coefficient κir =2.87 Frequency of Sudden Stops = 6.4% The interest rate r is set to 3% in normal times and 8% during disruptions in financial markets. The values are the averages of the real interest rates calculated for allcountriesin thesampleduring eachtypeof event.15 Thecoefficient ofriskaversionis set to 2, which is a standard value in quantitative business cycle analyses for emerging 13. See the Appendix for a description of the construction of each time series 14. I would like to thank Ed Knotek for providing the code to implement this method. 15. The country specific interest rate in international financial markets is measured as the sum of J. P. Morgan’s EMBI+ sovereign spread and the U.S. real interest rate. The U.S. real interest rate is measured by the interest rate on the three-month U.S. Treasury bill minus a measure of the U.S. expected inflation. 17

market economies. The range of estimates for the atemporal elasticity of substitution 1/(1+η) is between 0.40 and 0.83, as we can see in Mendoza (2005), so I use 0.8 as a conservative value. The parameter ω defines the share of tradable goods in the CES aggregator and is defined such that we have a 23% share of tradable production, which is the average for the countries in my sample. The subjective discount factor β is set to match the average ratio of net foreign liabilities excluding international reserves to GDP for Latin American countries, which is 36% for the period from 1991 to 2015. This criterion yields a beta of 0.932, which is reasonable for an annual frequency. I calibrate κT,H such that the collateral constraint is never binding in normal times.16 The parameters concerning the behavior of international financial markets disruptions are κT,L; the credit coefficient for tradable output in financial distress periods; π, the probability of entering a financial distress period; and ψ, the probability of going back to normal times. The parameters are set to obtain a current account standard deviation of around 2.5%, a current account reversal of close to 3.9% of GDP in the year of a sudden stop (compared to the average of the previous three years) and a posterior reduction in the current account result of 1.4% (compared to the average of the three years after the sudden stop), which were obtained from the data analysis shown previously. This procedure yields κT,L equal to 0.2, π equal to 0.2 and ψ equal to 0.4. The parameters values obtained for π and ψ are consistent with what is observed in the sample, as we have four international crises in 25 years and they usually last for two years. Finally, I obtain κir by matching the frequency of sudden stops for my sample of countries.Iobtainκir equalto2.87,whichseemsreasonablebecause,asnotedbySiritto (2016), the collateral solves an asymmetric information problem about the resources available to the borrower at the time of repayment, creating incentives to tell the truth and allowing agents to borrow more funds than by just selling the assets.17 16. As κT,H is very high, the lower bound of bond holdings becomes b min > − r y m m T a in x , which is the largest debt that the country can repay. 17. Garc´ıa-Schmidt (2015) includes asymmetric information in a model of sovereign borrowing with default and finds that it improves the fit for debt and spreads a lot, which indicates that this is an important feature of emerging market economies’ debt markets. 18

4.4 Borrowing and International Reserves Decisions Figure 4 shows the bond decision rules for both κT,H, which I call the normal period, and κT,L, which I call the crisis period. As the average value of tradable output is equal to 1, we can interpret the results as ratios of the average level of tradable output. As we can see, for the same level of current bond holdings, agents decide to have more debt in t+1 when the tradable output is lower during periods when the collateral constraint is not binding. However, when it is binding, agents are restricted to a level of debt around 1.5 times the tradable output. Figure 4: Bond Policy Functions Note: The bond policy functions are calculated for IR =0.40, which is the average stock of internat tional reserves in tradable units. Figure 5 shows the international reserves decision rules for, again, both normal and crisis periods. As we can see, the decision about international reserves holdings depends crucially on whether we are in normal or crisis times. In normal times, the higher the current debt, the more international reserves are accumulated because the country is in a more dangerous zone, closer to a binding collateral constraint if international finance conditions turn out to be bad in the following period. During crises, there is a tradeoff between reducing international reserves to consume more today and 19

keeping the reserves in case the crisis lasts. Thus, international reserves holdings are kept somewhat around the current level when the collateral constraint is binding and debt levels are not too high as an additional insurance if the crisis continues. However, there is some reduction in international reserves holdings when the current debt is very high to compensate for the strong deleveraging necessary in the current period due to the binding constraint. Figure 5: International Reserves Policy Functions Note: The international reserves policy functions are calculated for IR = 0.40, which is the average t stock of international reserves in tradable units. 4.5 Long-Run Business Cycle Moments In this section, I compare the model unconditional moments to the data. To do so, I conduct a one-million-period simulation of the model by drawing a sequence of endowments {yT,yN} and tradable output credit constraint coefficients {κT} from their t t t distributions and feed them to the policy functions to get the time-series for {b , cT, t t cN, IR }. t t Table4showsthat,ingeneral,Iamabletoreproducethemainmomentsinthedata. First, I manage to obtain an average international-reserves-to-GDP ratio very close to 20

the data. Second, I find countercyclical fluctuations for the current account and net foreign liabilities excluding international reserves, and I find procyclical fluctuations for the real exchange rate, again with results close to the data.18 However, I also obtain countercyclical international reserves, which is at odds with what we see in practice. Finally, the standard deviation of both international reserves and net foreign liabilities excluding international reserves are higher than what we see in the data. Table IV Long-Run Business Cycle Moments Targeted Moments Model Data Average NFL ex-Reserves-to-GDP ratio 36.0% 36.0% Frequency of Sudden Stops 6.4% 6.4% σ(CA/Y) 2.4% 2.5% Reversal 3.1% 3.9% Recovery 1.8% 1.4% Non-Targeted Moments Model Data Average Reserves-to-GDP ratio 10.8% 9.9% σ(IR/Y) 10.8% 2.1% σ(NFL ex-IR/Y) 30.3% 11.9% ρ(y,IR/Y) -0.64 0.07 ρ(y,-b/Y) -0.66 -0.31 ρ(y,CA/Y) -0.42 -0.40 ρ(y,REER) 0.77 0.30 The high standard deviation of international reserves and net foreign liabilities exinternational reserves might be explained by the absence of any adjustment costs for agents to change their foreign assets positions, as in Schmitt-Groh´e and Uribe (2003). The presence of convex portfolio adjustment costs would curb the volatility of both international reserves and net foreign liabilities ex-international reserves, which might lead to numbers closer to the data. It could also solve the issue of the countercyclicality of international reserves, as the government would accumulate more reserves during good times to avoid paying high adjustment costs when it gets closer to a binding collateral constraint. As this was not the main subject of this paper, I decided not to include any adjustment costs. 18. I use the relative price of nontradables as a measure of the real exchange rate in the model. 21

4.6 Sudden Stops Experiments I now analyze the dynamics of the model during a sudden stop and compare it with the data. To construct the implied sudden stop events using the model, I use the following steps (i) Identify crisis events: I define t as a crisis event where we get a current account reversion of one standard deviation and a binding collateral constraint; (ii) Compute averages of macro quantities of the model centered around these events, were t represents the crisis episode; (iii) Compare the outcomes with the average crisis in the data. As we can see in Figure 6, the model can generally explain the behavior of macroeconomic variables in sudden stops. First, the optimal policy implies that the economy enters the crisis with a higher level of international reserves than what we see in the data. Moreover, international reserves have a small depletion after the onset of a crisis both in the model and in the data. Second, the optimal policy is to keep international reserves somewhat stable after a sudden stop and consequently the model cannot explain the rebuilding of international reserves levels after crises that we see in practice. Finally, the behavior of the ratio of net foreign liabilities excluding international reserves to GDP is close to what we see in the data, although we find a higher and more stable level before and after the crisis in the model. 4.7 A Passive Central Banker Inowcomparetheoptimalpolicywithapassivecentralbankerwhokeepsinternational reserves constant at the average level in tradable goods units in the base scenario for all periods, regardless of the state of the economy. The behavior of different variables during a crisis can be seen in Figures 7 and 8. The economy with a constant level of international reserves enters the sudden stop with a lower level of net foreign liabilities excluding international reserves but also has to deleverage as it enters the crisis with a lower level of international reserves. Moreover, the implied path of tradable consumption is almost the same for both economies, which implies that the welfare benefits of holding international reserves during sudden stops are quite similar if the central bank behaves optimally by accumulating more international reserves before sudden 22

Figure 6: Dynamics around Sudden Stops Events stops and depleting some of them during it; or if it keeps a constant buffer that allows agents to maintain their level of external borrowing. Finally, the level and behavior of international reserves during a crisis are very close to what we see in the data. These results can explain the fear of losing international reserves identified by Aizenman and Sun (2012) during the GFC and is consistent with what De Gregorio (2011) noted: ”Countries hoard reserves because they see them as a safety net for periods of financial stress but, in practice, they seldom use them (...) reserves play a stabilizing role simply because they are there and not necessarily to be used.” However, this passive policy increases the frequency of sudden stops to 8.8% and reduces welfare when the country is far from hitting the collateral constraint, at which point it would be optimal to reduce the level of international reserves to consume more. Moreover, when the country is in states where the collateral constraint might bind in the near future, the optimal policy is to hold a higher level of international reserves than the average to allow for more consumption if the economy is hit by a negative shock in international financial markets. 23

Figure 7: Dynamics around Sudden Stops Events - Alternative Policy vs Data Figure 8: Dynamics around Sudden Stops Events - Alternative vs Optimal Policy 24

5 Can the Model Explain Any Level of International Reserves? I now analyze what would happen to the results if I change the value of κir and adjust the subjective discount factor β accordingly to get the same average level of net foreign liabilities ex-international reserves. As we can see in Figure 9, the optimal ratio of international reserves to GDP is between 9% and 16% of GDP. In fact, changing κ mainly changes the frequency of sudden stops, which goes from 15% to 4%. This ir result shows that, contrary to what some people might expect, the model cannot tautologically generate any level of international reserves just by changing the value of κ . ir Figure 9: International Reserves - Sensitivity to κ ir 6 Conclusion Why do emerging market economies simultaneously hold very high levels of international reserves and short-term foreign liabilities? This work explains this puzzling fact by explicitly introducing international reserves as collateral for external borrowing in a dynamic, stochastic model of a small open economy with credit constraints subject to exogenous financial shocks. Ifindthatthemodelcanexplaintheobservedaverageratioofinternationalreserves 25

to GDP in Latin America without considering any additional motives for international reserves accumulation. Moreover, the optimal policy implies that the government accumulates international reserves before a sudden stop, and there is a small depletion during it. Finally, the welfare implications of the optimal policy are quite similar to those of a policy of constant international reserves, which sheds some light on the fear of losing international reserves observed in the recent GFC. It is important to emphasize that I abstract from some potentially important features of models where foreign liabilities and international reserves are chosen together. First, I do not consider the role of international reserves to reduce output costs in sudden stops. Including this feature would unambiguously lead to an increase in the optimal level of reserves. Second, I do not consider the possibility of sovereign default. On the one hand, Alfaro and Kanczuk (2009b) show that holding international reserves increase the country’s willingness to default and consequently make external debt more costly. On the other hand, Levy Yeyati (2008) argues that international reserves reduce the probability of default during crises and consequently reduce spreads in external borrowing. Therefore, the effect of international reserves accumulation in the cost of external borrowing when we allow for sovereign default is still debatable. Finally, I abstract from any exchange rate management policies, which might be another important motive for international reserves accumulation. The role of international reserves as collateral for foreign borrowing is an important and unexplored aspect of the recent process of international reserves accumulation by emerging market economies, which is still a puzzle in the international macroeconomics literature. The policy implications of this feature and its potential consequences for the policies pursued by multilateral institutions such as the IMF and central banks around the world make it an important area for future research. 26

References Aizenman, Joshua. 2009. “Reserves and the crisis: a reassessment.” Central Banking Journal 19 (3): 21–26. Aizenman, Joshua, and Jaewoo Lee. 2007. “International Reserves: Precautionary Versus Mercantilist Views, Theory and Evidence.” Open Economies Review 18 (2): 191–214. Aizenman, Joshua, and Yi Sun. 2012. “The Financial Crisis and Sizeable International Reserves Depletion: From Fear of Floating to the Fear of Losing International Reserves?” International Review of Economics and Finance 24:250–269. Alberola, Enrique, Aitor Erce, and Jos´e Maria Serena. 2016. “International reserves and gross capital flows dynamics.” Journal of International Money and Finance 60:151–171. Alfaro, Laura, Sebnem Kalemli-Ozcan, and Vadym Volosovych. 2014. “Sovereigns, Upstream Capital Flows and Global Imbalances.” Journal of European Economic Association 12 (5): 1240–1284. Alfaro, Laura, and Fabio Kanczuk. 2009a. “Debt Maturity: Is Long-Term Debt Optimal?” Review of International Economics 17 (5): 890–905. . 2009b. “Optimal Reserve Management and Sovereign Debt.” Journal of International Economics 77 (1): 23–36. Bianchi, Javier. 2011. “Overborrowing and Systemic Externalities in the Business Cycle.” American Economic Review 101 (7): 3400–3426. Bianchi, Javier, Juan Carlos Hatchondo, and Leonardo Martinez. 2016. “International Reserves and Rollover Risk.” Unpublished. Broner, Fernando A., Guido Lorenzoni, and Sergio L. Schmukler. 2013. “Why do Emerging Economies Borrow Short Term?” Journal of the European Economic Association 11 (s1): 67–100. Bussiere, Matthieu, Gong Cheng, Menzie Chinn, and Noemie Lisack. 2015. “For a Few Dollars More: Reserves and Growth in Time of Crises.” Journal of International Money and Finance 52:127–145. 27

Caballero, Ricardo, and Arvind Krishnamurthy. 2001. “International and Domestic Collateral Constraints in a Model of Emerging Market Crisis.” Journal of Monetary Economics 48 (3): 513–548. Calvo,Guillermo.2005.“CrisesinEmergingMarketEconomies:AGlobalPerspective.” NBER Working Paper, no. 11305. Calvo,Guillermo,AlejandroIzquierdo,andLuis-FernandoMejia.2008.“SystemicSudden Stops: The Relevance of Balance-Sheet Effects and Financial Integration.” NBER Working Paper, no. 14026. De Gregorio, Jos´e. 2011. “International Reserve Hoarding in Emerging Economies.” Unpublished. . 2013. “Resilience in Latin America: Lessons from Macroeconomic Management and Financial Policies.” IMF Working Paper, no. WP/13/259. Dominguez, Kathryn M.E. 2010. “International Reserves and Underdeveloped Capital Markets.” In NBER International Seminar on Macroeconomics 2009, edited by Lucrezia Reichlin and Kenneth West, 193–221. University of Chicago Press. .2012.“ForeignReserveManagementDuringtheGlobalFinancialCrisis.”Journal of International Money and Finance 31:2017–2037. Dooley,Michael,DavidFolkerts-Landau,andPeterGarber.2004a.“TheRevivedBretton Woods System: The Effects of Periphery Intervention and Reserve Management on Interest Rates and Exchange Rates in Center Countries.” NBER Working Paper, no. 10332. . 2004b. “The US Current Account Deficit and Economic Development.” NBER Working Paper, no. 10727. Durdu, Ceyhun Bora, Enrique G. Mendoza, and Marco E. Terrones. 2009. “Precautionary Demand for Foreign Assets in Sudden Stop Economies: An Assessment of the New Mercantilism.” Journal of Development Economics 89 (2): 194–209. Eggertsson, Gauti, and Paul Krugman. 2012. “Debt, Deleveraging, and the Liquidity Trap: A Fisher-Minsky-Koo Approach.” Quarterly Journal of Economics 127 (3): 1469–1513. 28

Eichengreen, Barry, Poonam Gupta, and Ashoka Mody. 2008. “Sudden Stops and IMF Supported Programs.” In Financial Markets Volatility and Performance in Emerging Markets, edited by Sebastian Edwards and Marcio Garcia, 291–366. University of Chicago Press. Forbes,Kristin,andFrancisWarnock.2012.“CapitalFlowWaves:Surges,Stops,Flight and Retrenchment.” Journal of International Economics 88 (2): 235–251. Garc´ıa-Schmidt, Mariana. 2015. “Volatility of Sovereign Spreads with Asymmetric Information.” Unpublished. Hur, Sewon, and Illenin O. Kondo. 2016. “A Theory of Rollover Risk, Sudden Stops, and Foreign Reserves.” Journal of International Economics 103:44–63. Jeanne, Olivier. 2007. “International Reserves in Emerging Market Countries: Too Much of a Good Thing?” Brooking Papers on Economic Activity 38 (1): 1–79. Jeanne, Olivier, and Roman Ranciere. 2011. “The Optimal Level of International Reserves for Emerging Market Countries: a New Formula and Some Applications.” The Economic Journal 121:905–930. Korinek, Anton, and Luis Serv´en. 2016. “Undervaluation through Foreign Reserve Accumulation: Static Losses, Dynamic Gains.” Journal of International Money and Finance 64:104–136. Lane, Philip, and Gian-Maria Milesi-Ferretti. 2007. “The External Wealth of Nations Mark II: Revised and Extended Estimates of Foreign Assets and Liabilities, 1970– 2004.” Journal of International Economics 73 (2): 223–250. Levy Yeyati, Eduardo. 2008. “The Cost of Reserves.” Economic Letters 100 (1): 39–42. Mendoza, Enrique. 2002. “Credit, Prices, and Crashes: Business Cycles with a Sudden Stop.” In Preventing Currency Crises in Emerging Markets, edited by Sebastian Edwards and Jeffrey Frankel, 335–392. University of Chicago Press. . 2005. “Real Exchange Rate Volatility and the Price of Nontradable Goods in Economies Prone to Sudden Stops.” Economia 6 (1): 103–148. . 2006. “Lessons from the Debt-Deflation Theory of Sudden Stops.” American Economic Review 96 (2): 411–416. 29

Obstfeld, Maurice, Jay Shambaugh, and Alan Taylor. 2010. “Financial Stability, the Trilemma, and International Reserves.” American Economic Journal: Macroeconomics 2 (2): 57–94. Panizza, Ugo, Federico Sturzenegger, and Jeromin Zettelmeyer. 2009. “The Economics and Law of Sovereign Debt and Default.” Journal of Economic Literature 47 (3): 651–698. Rodrik, Dani. 2006. “The Social Cost of Foreign Exchange Reserves.” International Economic Journal 20 (3): 253–266. Schmitt-Groh´e, Stephanie, and Mart´ın Uribe. 2003. “Closing Small Open Economy Models.” Journal of International Economics 61 (1): 163–185. Siritto, Cecilia. 2016. “Collateralizing Liquidity.” Unpublished. Terry, Stephen, and Edward Knotek. 2011. “Markov-Chain Approximations of Vector Autoregressions: Application of General Multivariate-Normal Integration Techniques.” Economic Letters 110 (1): 4–6. 30

A Data The dataset includes annual data for Argentina, Brazil, Chile, Colombia, and Mexico. The sample period is from 1991 to 2015. I chose the main Latin American Countries as my sample because they belong to the set of countries included in J. P. Morgan’s EMBI+ data set for emerging-country spreads and have consequently access to borrowing in international markets. Annual series for nominal GDP (total and sectoral), total international reserves, and current account balance in U.S. dollars; total and sectoal real GDP; and real effective exchange rates are from the World Bank’s World Development Indicators databank. The net foreign liabilities positions are from the updated and extended version of the External Wealth of Nations dataset constructed by Lane and Milesi- Ferretti (2007), complemented by the updated international capital flows database constructed by Alfaro, Kalemli-Ozcan, and Volosovych (2014). The JP Morgan EMBI Global Index spread used to identify the Sudden Stop episodes is measured using data on spreads from JP Morgan. The endowment process for each country is estimated using the HP-filtered cyclical component of tradables (agricultural and manufacturing industry) and nontradables (industry ex-manufacturing and services) GDP. B The High Frequency Behavior of International Reserves Although we do not see much international reserves depletion during crisis on an annual basis, there might be larger international reserves losses if we look at higher frequencies.19 Aizenman and Sun (2012), for example, show that most emerging market economies began exhibiting large international reserves losses during the second half of 2008 and regained most of their losses by the first quarter of 2009. If we look at quarterly data, there is indeed a larger international reserves depletion also in my sample, as we can see in Figure B.1, where the average international reserves level falls around 15% in US dollars. However, this fall is compensated by the fall in GDP in US dollars due to currency depreciation and recessions experienced by some countries and, consequently, we still get on average an increase in the ratio of international reserves to GDP, coherent with their annual counterpart. Moreover, although it is true that 19. I thank Pablo Ottonello for pointing out this fact. 31

we see some cases of stronger international reserves depletion, as in Argentina in the 2002 crisis, these episodes are related to fixed exchange rate regimes, which are not the subject of this paper, where I abstract from studying the effects of different exchange rate policies on international reserves accumulation. Finally, the fact that countries international reserves holdings start to recover rapidly is another evidence that although they might be used for another purpose in the short term, their role as collateral leads to an urgency of having them back quickly to the previous levels. Figure B.1: International Reserves around Sudden Stops - Quarterly Data C Do International Reserves Serve as Collateral? I have assumed that international reserves are used as collateral, is this really the case? Unfortunately, as the main idea is that reserves serve as an implicit collateral, we cannot infer directly from any database if this is true. However, an implication of my setup is that during periods of international financial stress countries with a higher level of international reserves before these crisis are able to hold more net foreign liabilities excluding international reserves during the crises. I check this implication for several emerging market economies, the results are presented in Figure C.1. As we can see, there is a strong positive correlation between b and IR and thus it indicates t+1 t that foreign lenders do lend more during crises to countries that have a higher level of international reserves. 32

Figure C.1: Net Foreign Liabilities ex-International Reserves and International Reserves in International Crises Note: Each point represents data for a specific country during an international crisis episode. The years of international financial stress are 1995, 1999, 2002 and 2009. The value of international reservesismeasuredinthebeginningoftheyearofaninternationalfinancialstressperiodwhilethatof net foreign liabilities ex-international reserves is measured in the beginning of the following year. The countries included are Argentina, Bolivia, Brazil, Bulgaria, Chile, Colombia, Costa Rica, Czech Republic, Dominican Republic, Ecuador, Egypt, El Salvador, Guatemala, Honduras, Hungary, Jamaica, Jordan, Korea, Malaysia, Mexico, Paraguay, Peru, Philippines, Poland, Romania, South Africa, Sri Lanka, Thailand, Tunisia, Turkey and Uruguay. Source:Authors’computationsbasedontheupdatedandextendedversionofthedatasetconstructed by Lane and Milesi-Ferretti (2007). D Additional Tables and Figures Table D.1 International Reserves (% of GDP) 1991-1995 1996-2000 2001-2005 2006-2010 2011-2015 Argentina 4.0% 7.4% 10.1% 11.8% 6.9% Brazil 3.7% 6.2% 6.5% 8.8% 15.1% Chile 17.5% 19.8% 17.9% 11.3% 14.9% Colombia 11.3% 9.2% 9.8% 8.7% 10.7% Mexico 3.7% 4.8% 6.6% 8.6% 13.1% 33

Table D.2 Net Foreign Liabilities ex-International Reserves (% of GDP) 1991-1995 1996-2000 2001-2005 2006-2010 2011-2015 Argentina 14.1% 30.3% 67.0% 11.0% 3.1% Brazil 24.1% 33.2% 45.7% 32.9% 48.6% Chile 51.4% 56.7% 56.8% 22.9% 30.5% Colombia 25.7% 37.5% 34.2% 30.9% 38.6% Mexico 34.6% 38.8% 34.9% 41.8% 54.2% Table D.3 Current Account (% of GDP) 1991-1995 1996-2000 2001-2005 2006-2010 2011-2015 Argentina -2.5% -3.8% 3.7% 2.0% -1.5% Brazil -0.2% -3.6% -0.3% -1.0% -3.3% Chile -2.5% -2.9% 0.0% 2.2% -2.4% Colombia -1.1% -2.7% -1.1% -2.5% -4.1% Mexico -4.4% -2.1% -1.5% -1.1% -2.0% Figure D.1: International Reserves - Full Sample - US$ billions Note:ThedataarethesumofthevalueforthefivemainLatinAmericancountries(Argentina,Brazil, Chile, Colombia and Mexico). The shaded areas are the systemic sudden stops events identified in section 4.2. Source: Authors’ computations based on World Bank World Development Indicators database. 34

Figure D.2: International Reserves - Full Sample - % of GDP Note: The data is the simple average of the indicator for the five main Latin American countries (Argentina, Brazil, Chile, Colombia and Mexico). The shaded areas are the systemic sudden stops events identified in section 4.2. Source: Authors’ computations based on World Bank World Development Indicators database. 35