Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium Member Libraries http://www.archive.org/details/globalequilibriuOOcaba DEWEY Massachusetts Institute of Technology Department of Economics Working Paper Series A Global Equilibrium Model of Sudden Stops and External Liquidity Ricardo J. Management Caballero Stavros Panageas Working Paper 08-05 September 7, 2007 . RoomE52-251 50 Memorial Drive 02142 Cambridge, MA This paper can be downloaded without charge from the Network Paper Collection at Social Science Research httsp://ssrn.com/abstract=n 08710 HB31 .M415 A Global Equilibrium Model Liquidity Ricardo of Sudden Stops and External Management Stavros Panageas Caballero J. MITandNBER The Wharton School This draft: September 07, 2007* Abstract Emerging market economies, which have much should run persistent current account The counterpart We of these deficits make two contributions model of sadden is deficits in their we use order to smooth consumption intertemporally. dependence on capital in this paper: stops. Second, of their growth ahead of them, either run or First, inflows, wliich can suddenly stop. we develop a quantitative global-equilibrium this structure to discuss practical mechanisms to insure emerging markets against sudden stops, ranging from conventional non-contingent reserves cumulation to more sophisticated contingent instrument strategies. ac- Depending on the source of sudden stops, their correlation with world events, and the quality of the hedging instrument available, the gains from these strategies can represent a substantial improvement over existing practices. JEL Codes: Keywords: E2, E3, F3, F4, .GO, CI. Capital flows, sudden stops, reserves, international liquidity management, world capital markets, swaps, insurance, hedging, options, hidden states, Baj'esian methods. *We at are grateful to Lars Hansen, conference participants at IMF, UPF-CREI for their AEA 2007, and seminar participants comments, and to Herman Bennett, Fernando Duarte and Jose Tessada research assistance. Caballero thanks the circulated as CIRJE, Tokyo, "A Quantitative Model of NSF for financial support. A for excellent partial equilibrium version of this paper Sudden Stops and Liquidity Management," NBER WP 11293, May was 2005. Introduction 1 Emerging market economies, which have much of run persistent current account of these deficits domestic deficiencies, there itself, growth ahead of them, either run or should smooth consumption intertemporally. The funding a perennial source of fragility since is many can suddenly stop. While in the country deficits in order to their is it requires ongoing capital inflows which circumstances the breakdown in capital inflows just amplifies extensive evidence that in many other cases the main culprit is not but the international financial markets' response to shocks only vaguely related to the country's actions. The real costs of this volatility for countries that experience open noticed, but at least as important in terms of their pervasiveness large costs paid by prudent economies. These economies do not to incur in a variety of costly precautionary measures crises are extremely large. Less and cumulative impact, are the fall into open crises but are forced and build large war-chests of international reserves, a trend that has only increased in the aftermath of the capital flow crises of the late 1990s. By now, emerging markets' reserves often exceed 20 percent of their GDP, which contrasts with How the 4 to 5 percent held by developed open economies such as Australia or Canada. are reserves in smoothing the impact of sudden stops unrelated to a country's actions? of them should be accumulated? How to deal with capital flow volatility? How are these fast? Who mechanisms limited by Are there potentially would be the natural counterpart How much mechanisms mechanisms? for these financial constraints? These are among the most pressing questions market economies and the international financial significant conceptual progress over the last for policy-makers and researchers in emerging institutions. Unfortunately, while there has two decades in been understanding some of the hmitations of financial contracting with emerging markets, there has been integral less costly financial effective framework to analyze these questions quantitatively. In much this less progress in providing an paper we take one step in this direction. Our framework considers two representative agents, one from emerging markets from the developed world (W). EM's problem has two main ingredients: significantly higher account deficits. natural lender is than its Second, it W, its but current income so it would like to (EM) and one First, its future income is borrow and run persistent current has difficulty in pledging future income to finance these deficits. willingness to lend varies over time in response to shocks to its The pref- erences, monitoring technology, dechne in capital flows to tions against these EM, sudden a so called sudden stop} stops. EM. At markets available to and output. These shocks, when negative, lead to an equilibrium We common EM sudden environment, The practice by adopting a hedging strategy. costs is and framework and depend on the particular source we estimate a regime switching model of emerging markets, as implied by our theoretical framework. of sudden stops in a group As part of this exercise, estimate the joint incidence of sudden stops with jumps in the price of assets traded in of risk it stops. In our quantitative analysis markets. EM precau- can accumulate noncontingent reserves, but benefits of these strategies are endogenous in our of this discuss a range of options as a function of the risk-sharing a minimum, often possible to improve on this Faced with A calibrated version of the model premia and the size of reserves matches the extent of capital flow accumulation. US we also financial reversals, the behavior We then use the model to quantify the potential gains from adopting hedging strategies and conclude that, depending on the source of sudden stops, their correlation with world events, and the quality of the hedging instrument from these strategies can be substantial. When compared available, the gains to the welfare gains of existing practices, they represent improvements that range between about 10 and 115 percent. Our paper relates to several strands of literature. sudden stop of capital and Russian crises. inflows. The work The main shock that concerns The hterature on sudden stops gained of Calvo (1998) describes the basic momentum us here is a since the Asian mechanics and implications of sudden stops and Calvo and Reinhart (1999) document the pervasiveness of the phenomenon. The modelling of these sudden stops as the tightening of a financial constraint of Caballero among is also present in the and Krishnamurthy (2001), Arellano and Mendoza (2002) and Broner et al. work (2003), others.^ As an intermediate step in developing our substantive argument and quantifying the effects we describe, we model The view that reserves can be used to cushion the impact of external shocks exists at least since Heller (1966), reserves accumulation as a buffer stock model against was enhanced by the work on precautionary savings 'See an earlier version of this paper, Caballero and Panageas (2005), three agents: insure with EM, W, and W. specialists. a macroeconomics during the more reduced form model, but with There, sudden stops are the result of shocks to specialists which Shocks to the monitoring technology play a similar ^See also Kiyotaki and Moore (1997) for in capital flow reversals. for role in the current EM tries to model. a comprehensive model of how financial constraints are linked with cycles. 1980s and has recently returned to the fore with the large accumulation of reserves exhibited by emerging markets since the crises at the end of the 1990s Lee (2004)). (see, e.g., Importantly, the main reason for seeking insurance and hedging in our context is not income fluctuation per-se but the potential tightening of a financial constraint. This motive parallels that highlighted by Froot et al (1993) at the level of corporations, and by Caballero and Krishnamurthy (2001) for emerging markets. While the substantive themes developed in those articles differ from ours, the basic model in our paper is in many respects a dynamic global-equilibrium version of theirs. Closely related to our recommendations are those in the sovereign debt literature. mality of contingent debt and the limitations to of that literature. In particular, the Wright (2000), characterize it imposed by financial The frictions are also a feature work of Kletzer, Newbery and Wright (1992) and Kletzer and feasible financial structures consistent with different degrees of ment by a sovereign borrower and opti- its lenders. In our model we capture financial frictions committhrough monitoring costs, which capture features similar to those of their richer limited enforcement framework. Our paper reinforces much of the message in that literature and provides a tractable model that can be estimated and quantified. The closed interaction between precautionary savings and financial constraints economy framework of Aiyagari (1994). He calibrated such a is also present in the model to estimates of US microeconomic income processes and other parameters and concluded that eventually agents would save enough to relax all a steeper growth rate than W. time varying financial frictions increasing endowment. and recurrent episodes There has been a (which is the index happen because EM faces Hence, EM's incentive to save in order to mitigate the effects of financial constraints. In our The of is model this does not counterbalanced by the reluctance to save in light of a steeply interaction of these two effects leads to a stationary level of reserves sudden stops and consumption drops. significant rise in volatility trades in financial markets, including the we use for our main example of hedging). The finance literature has studied the impact of such trades on the performance of hedge funds and other markets participants e.g., Bondarenko (2004)). VIX From a risk management (see, perspective, the issues faced by these economic agents are similar to those faced by emerging market policymakers. We setup the model in Section 2 and describe global equilibrium under different assumptions on the risk-sharing options available. In Section 3 we estimate the sudden stop process and cahbrate the model. Section 4 presents our main quantitative results and Section 5 concludes. The paper also contains several appendices. A 2 Model of Sudden Stops In this section we in Global Equilibrium describe a model of optimal global risk-sharing in the presence of financial constraints in emerging markets. In the model, systemic crises are caused by events in world financial markets that lead to abrupt changes in capital flows to We emerging markets. start with a characterization of a perfect (sudden stop) risk sharing benchmark, and conclude with a discussion of a more realistic proxy-hedging context where available insurance instruments are only imperfectly correlated with sudden stops. Agents and Endowments 2.1 There are two "representative" agents one from the developed world (W), Et t The parameter p (it is world economy, one from emerging markets (EM) and who have common j, preferences for j 1-7 > represents the discount rate, 7 consumption of agent The world in the 1 e {W, EM} (1) the coefficient of relative risk aversion, Ci the and A^ captures a taste shock, whose properties an endowment economy. EM's current income is will be specified low relative to its future income has yet to catch up), which we capture through a pre-development phase (which of the paper) and a post-development phase. During the former, EM's endowment AfYt, where At is a process that can take either the value the exact dynamics of these transitions shortly) , and —^^fi,dt + where fXf^a > and dBt is g. is or the value A"^ < 1 is independent of the focus described by (we shall specify adBt. (2) EM (permanently) transits from the an exponentially distributed random time This catching-up transition is is given by a standard Brownian motion. pre- to the post-development phase at hazard Yt 1 later. all r*^, with constant other sources of uncertainty in the model. EM's post-development income equal to ^Aflt, where k is > 1. W's endowment is given by PAtYt, where > Ac> /3 > 1 (3) for all t 2.2 Growth-Contingent Debt and Capital FIoavs 0. Since in the pre-development phase EM for to EM's expected growth rate is borrow from W. Moreover, given the stochastic nature of catching up optimal debt contracts are contingent on the transition to development. contingent) debt contracts establish that > t, development arrives if contingent debt contracts is is natural in our model, W provides a flow of resources ftPtYt to EM over the next in the interval dt pt while ft it is Concretely, (growth- infinitesimal time interval dt, in exchange for receiving a promise to a stream of for all s W, higher than that of number the and The otherwise. payments ft^Ys By of contracts (expressed as a fraction of Yt). EM can exchange a stream of post-development for pre-development using these contracts, growth price of these income. In the absence of frictions, growth-contingent debt contracts are claims that complete the market with respect to the uncertainty introduced by the random arrival of development, and ensures that the marginal utilities of W and EM are always proportional to each other, both pre-and post development.'^ Since there assuming that markets after only one source of uncertainty in the model in the post-development phase, is EM t'^.^ and To W are able to trade simplify the analysis, in W's we will equities implies that assume this trade is W and EM face complete feasible (after ''Note that these contingent contracts can be implemented through equity trades. exist be two equities that give the owner an entitlement Pf^ opment. and P}^ respectively and their dividends be To produce the financed by xtPt''^ / xtYt [1 to the P^ - (Pf^/P^) ( [1 - units of W's equities. P] pre-development {Pfc^i and (5Yt IP%- ) (k/^)] ) ^ The Duffie and and W respectively. Let their prices pre-development and nVt and pVt post devel- and receives a capital gain equal to By P^a arbitrage . Since Xt Pfci is = (f^/P) arbitrary, ^ pBM Xi. 'iv Huang (1985). £,. P^+ one can units of growth contingent contract. In summary, a growth contingent contract e.g. EM assume that there EM's equity holder of such a claim delivers a stream of payments equal to Hence, by purchasing xt units of EM's shares financed by "See of this, payoffs of a growth contingent contract, one needs to purchase xi units of the transition to development takes place. pressed as £t Yt endowment For t^) and use is ( [Pfc^ - (Pfc^ /P^- ) P^+]) once and hence the capital gain can be ex- let xt W's = 1/ shares, ( [l - {Pf^- IP%-) (/t//3)] ) one can create one unit of a equivalent to a carry trade in equities. P^ to denote the price of and W respectively, with 5f^ + S^ ^ Normal Times, 2.3 W's shares and Crises, Sf'^,S^ to denote the holdings of such shares by 1. and Sudden Stops During the pre-development phase, there are two states of nature (Sf = respectively. all t The 1) times. > We EM st: "normal" transition hazards from the former to the latter (st = 0) and "crisis" and vice versa are A and For simplicity, there are no crises after the time of development r'^. (i.e., st — A, for r'^). assume that for each growth contingent contract ft, W pays a monitoring cost equal to itYt, where 'z ^t if Sf shall assume throughout that growth contingent debt i > and g (4) if i We = ={;,..:::: +q St = 1 > These monitoring costs capture the idea that 0. effectively represents uncollateralized lending to subject to a series of incentive problems that are costly to overcome. EM, which in practice is For our purposes, these monitoring costs inhibit intertemporal consumption smoothing between pre-and post-development times. Moreover, we assume that during normal times Af Hence, crises are also times when world output To produce sudden Assumption stops within our framework, 1 Either q Assumption 1 may > or A'^ < fall 1, while during crises Af below the (stochastic) trend in monitoring its trend, so that crisis: it becomes costlier to activities. two extra layers of f^o we assume that the 1. Either the cost generality. First, the growth rate /Xj + St (Mi - ^^o) if t < r"^ preference shock process At in equation (1) 7 is it is also could take different values during crises, normal times, and development: Second, < Yt- For the purposes of calibrating the model and performing robustness checks later on, useful to allow for A'' (or both) I requires that at least one of two events happens during a consumption = we assume that of monitoring increases, or world output declines below sacrifice = given by where ^ > happened expected and 1 until utility A''; time number transitions from normal times to crises that have have is the t. In the special case where of all A — During normal times Benchmark EM precautions against sudden stops. We assume that there are two potential The contracts that can be used for that purpose. which are zero net supply claims that pay an kind are non-contingent bonds (reserves), first interest rate rt per unit of time dt. Second, agents also have access to (sudden stop) hedging contracts. For each of these contacts, of TTtYt to W per unit stays at over the interval dt, then However, if St W until from jumps Sj of time dt while to transits EM happens. delivers the We = Sj 0, where nt over the next time interval 1 back to delivers a payment to EM EM and the contract is If st terminated. receives a flow of Yj (per contract) 0. W payments to W has assume that dt, EM the "price" of such a contract. is W makes no transfers Hedging contracts are hence similar case becomes a standard (1) maximization problem. Perfect (Sudden Stop) Risk Sharing 2.4 the objective 1, in and first, full nature to contingent debt contracts, although in this W has to repay when a transition to a sudden stop commitment to such contracts and hence there is no risk of W defaulting on EM. In summary, sudden stop contracts represent forms of insurance that can help EM hedge against both the occurrence and the duration of sudden stops. 2.4.1 We Net Asset processes can now describe the dynamic budget constraints (reserves), ft the amount of contingent debt as a share of Yt, contracts as a share of Yt, 1. (Pre-development, st we = rt Letting X^ denote j's assets and nt the number of sudden stop have: 0) If i dXf ^ = with EM and W. for < r'^ and [rtXf ^ sj = [nXJ!^ - EM's - Cf ^ + being the prevailing interest rate, and dX^ = 0, Cf + 8 (1 W's {0-ft + net asset process ftpt cisset {Pt - ntnt) process + i)+ Y] is dt (5) is ntnt) Y] dt (6) 2. (Pre-development, = Sj 1) If < i EM time at which last and = s* EM's 1, rEM riEM + rtXr-Cr dX. with r being the r*^ {^' jumped from St T = = SUp{s„ net asset process + ftPt to + nAY, is dt (7) 1 0}. u<t The corresponding W's asset process w dXl 3. (Post-development) If t is ru > dt t*^ then dXf ^ = [nX^"^ _ cf ^ + ciXf ^ [nXf" - Cf' + Let us now summarize our setup. Equations to development "borrows" [t < r*^) whenever St noncontingent bonds. The processes Equations St — I (7) and and hence the 0. development, EM has pt,7rt (8) are similar to (5) n,- the promised payoffs to - + (/3 (5) /,g) KYt] dt «/,g) and and EM. (6), we assume that if EM rit (10) the evolution of assets prior has access to three markets: it hedging contracts, and also invests in except that the economy EM Finally, equations (9) to deliver payoffs to M^uW + ^.'^dP, yt] dt EM (9) are the respective prices in these markets. rt and + 5f ^dFf (6) describe In this regime hedging contracts that the time of transition into development Finally, = (1 units of growth-contingent debt, invests in ft (8) entered and when St is jumped (10) state that after in a regime to 1 EM's where are delivering transition into W as specified by the outstanding growth contracts at (/t-g). enters short positions in non-contingent bonds, then itYtm.in\Xf^ ,0] monitoring costs per unit of time dt. W pays This assumption implies that growth con- tingent contracts and short positions in non-contingent bonds are subject to the same monitoring costs, since they both present forms of borrowing. 2.4.2 Global Equilibrium We now are able to define an equilibrium in the global economy: An Definition 1 nt, TTt, equilibrium a collection of (progressively measurable) processes C^^ is pu ru Xf^\ Xi^, 5f^, maximize (1) (for j — EM) and subject to (5), (7) (1) (for j Markets 3. = W) subject to (6), (8) _ rEM.r^ SfM^gW ^ 1 ft, and Cf^,Xf^,ft, nt and Sf^ , X^ , ft, nt and St^ maximize (10). clear, i.e.: Xf^ + X^ = Remark , (9). Given the price processes rt,ni,pt,P^ the quantity processes Cj^ 2. Cp^ 5j^, Pt^ such that Given the price processes rt,7Tt,pt,Pt^ the quantity processes 1. , Note that all for all t (11) + (yt[{l+l3){l St{A^-l))-itft] tft<T(^ iforallt>T° (13) remaining financial markets clear by construction, since equations (10) recognize the zero net supply nature of the markets for hedging contracts (5)- and growth contingent contracts. The prices definition of equilibrium and that all Before proceeding, markets it clear. is It requires that and Proposition 1 prices. = ^, Yt where i = {W, properties. EM]. some properties of pre-development (Post-development allocations and prices are given in the appendix). For appropriate constants c^'^ /°, f^ there , properties: Xp^ = X^ = its = ^, wherei-{W^,£;M}, following proposition constructs an equilibrium and outlines allocations actions should be optimal given wih be useful to define cj 1. all Next we construct an equilibrium and characterize x\ The standard. for all t. 10 exists an equilibrium with the following 2. W The consumption process for is given by r.Q,W ifst^O cY^ C^'^ if St (14) = I where (15) + 1 -'iP and 3. the consumption process for w Let the constants and u EM is given by (12) after plugging in for be defined as /9 + A-(l-7)(Mi-7f p- (1-7) (mg-ty) and assume <1. When that v st — = ,+ (,.-!-),'A'^(/3 XwA TV K fp while in the state J st 1 (2£) -A -7 0\ -7 - ,o,w 1 (18) -7 — (19) 1 the prices rt and pt are given by the constants ,0,W\ -7 — a+ Kp1\ .l.W Ac P V There are two there is (/? EM's income post development, any form all 1 (20) (21) no accumulation of noncontingent bonds resources for - {c^y/)-^ results in Proposition 1 that are against a sudden stop -7 + K/^)"^ worth highlighting: {Xf^ = contracts to precaution against the arrival of the state in (17) rO,W + l)-(i + g)/i' (3 + 1 -if . fO \ + Kf = (16) • G the prices rt^pt^'n't «re given by the constants a,vi/\ .. C^ is costly. sj = X^ 1. — First, in general equilibrium 0). EM only uses contingent Given the anticipated steep increase of pre-development consumption sacrifice to Since reserves are noncontingent, they imply that states of the world - even states of the world 11 EM is where no sudden stop takes hedge saving place. This makes reserves too costly as a hedging instrument compared to claims that only deliver payoffs exclusively in states of the world where such payoffs are needed Second, Proposition 1 (i.e. states that unconstrained trading in hedging contracts completes markets with respect to the arrival of state st — 1. To see this, note that (15) together with (12) implies that "' (CX\ — St- 0, refers to and s^+ _ A^(/3 /^ ~[ [C}^J where r~ + l)-(i + /3 + l-^/° an instant before the arrival of = 1). The simple limit case (/? — > st — g)/^ \ ' ) and 1 r"^ (C^\ -' _ ^''> "l^Cf.^J to the instant thereafter. (Formally, equality in (22) follows from (15), while the second equality follows first from the goods market clearing condition A during sudden stops). W becomes very large relative to EM, we can sharpen the When oo) (12). results further: Lemma 1 As 13 ^f oo the solution to the system (S0)-(82) satisfies c p^=g-{A'^] i, f^ is — (i + q) ' / (3 -^ 1 and andn = \wA {-') (23) given by -1/7 p A'^- l+P^TT^ = f K + p° 1 ? (24) -1/7 pO I+ttA'' -]- p^Tl ( ^ "^ ^ and (25) Furthermore /° > /\ and since both f^ in equation (22) converges to 1 and f^ have finite limits, the term and hence qEM A^ (jEM Lemma 1 implies that when W is large compared to EM, the only shocks that can make EM's consumption drop when entering a sudden stop are world-wide drop in available resources to EM that is cyclical shocks. caused by capital flow reversals 12 is Hence, the extra smoothed out. To see most this clearly, by jumps monitoring costs in the Then Lemma 1 instructive to focus on the special case it is implies that and there are no it, and (7) = cyclical variations in output, so that A'^ EM's consumption experiences no change upon In order to understand this result, note that since (5) where sudden stops are caused purely = A'' ^^ 1, = entering a sudden stop. and 1 1. Xf^ = equations 0, imply that (/V-/V) EM Equation (26) states that - decline in capital flows (/°p° equation (19) implies that p° > p^ Since /° > it f^ p''' buys an amount of hedging contracts that f^p^) -^ g^ upon entry - i and n > follows that sudden into a The 0. n > fact that P'' ^ 5p - when consumption suspect that insurance must be actuarially A> 1. from fair. This is In this case the value (in state s to 1 is of = 0) of obtaining ^ ^_ / A'^(/3 V fi +^) , * oo, so that non-zero trading.^ The easiest way then sudden stop to see this is to allow a unit of consumption once s switches given by ^ When is EM becomes perfectly smooth, not true however. (i ^ — implies that adjustments of prices are not enough by themselves to attain consumption smoothing, and there One might proportional to the Additionally, as stop. equation (21) implies that is —^ oo, and A'^ not actuarially fair. = 1, + l)-(I + g)/^ \ "^ H + l-'iP I the above term becomes Nevertheless, consumption is XA > A. Hence sudden stop insurance is not affected by a sudden stop (although marginal utility is). More importantly reserves, EM for our purposes, note that no matter how large to purchase insurance in premium and those that do world where it all not. needs insurance in these states of the world is is A (or V^ EM for that matter). states of the world By prefers to use hedging rather than hold - The reason is any that reserves "force" both the states that command a high risk EM to isolate the states of the Since EM only needs insurance contrast, hedging always allows (i.e. when a sudden and wants to avoid stop occurs). sacrificing pre-development consumption, hedging always a preferred precautionary measure. ^This is in contrast to adjust in such a way many examples of endowment economies in macroeconomics and asset that agents are happy to just hold their endowments without trading. trading in both growth-contingent debt and sudden stop hedging contracts. 13 pricing, where prices Here there is active More generally, when /3 is finite, even though sudden stop contracts cannot eliminate the pres- ence of sudden stops, they can substantially mitigate their effects on consumption. Imperfect (Sudden Stop) Risk Shciring 2.5 In practice there and is a variety of impediments to perfect risk sharing, including agency, behavioral, liquidity problems. risk sharing is In this section we capture some incomplete along dimensions that seem hold their reserves entirely in US and Euro of these realistic. by assuming that sudden stop For the most part, central banks treasuries rather than investing in hedging instruments. Thus, later on we calibrate our model using extreme noncontingent scenario. this our analysis in this section we also introduce assets whose payoffs are correlated perfectly — with sudden stop arrivals. However, in — although not This extension allows us to discuss practical improvements to current central banks' practices in emerging market economies. Concretely, we preserve the environment developed up sharing instruments no longer exist but instead there correlated with the occurrence of sudden stops. now the country will factors, the particular to now, with one exception: Perfect risk an asset with a payoff that is imperfectly As one would expect, the optimal portfolio of is include both, shares of this asset and noncontingent reserves. Among other composition of the portfolio depends on the degree of correlation of the asset with sudden stops, on the level of reserves, and on the sources of sudden stops and their general equilibrium implications. 2.5.1 Danger zones and net asset processes In order to introduce the hedging asset First, there is a Poisson process we decompose the with intensity x that puts danger zones take place at random times rf, instead of r^. 1) i = arrival of EM 1,2,3...., at — 1) preserve the earlier environment and avoids we which we generically denote by r^ is with probability P{s^d (i.e. the state s^d becomes = = 0) 1 - P{s^d = 1). To require throughout that A Let us now assume that there it steps. danger of a sudden stop. These At these times r^, the country enters a sudden stop with probability P{s^d sudden stops into two = xP{s,o = 1). (27) a financial asset with payoff Ft, that also has the potential to 14 jump exhibit a Using J to denote a jump, we have that Ft exhibits jumps at danger times r^. according to a Poisson process with intensity: = xP{J = \j The assumption critical in this section between sudden stops and jumps 1) that there is (28) P{J = in the payoff Ft: some is = 1,s^d Letting $j denote a counting process that increases by correlation, although imperfect, times 1 at 6 1) (0, 1). r^ and , A^ denote a required rate of return (in excess of the interest rate) for holding Ft, the price of asset Ft follows the process — dFf ^ = ndt + - Kdt) (Jd^t (29) i't Without (J — we can condition on those times r^ where we observe loss of generality, and/or a transition into SSfs^o 1) X which = is 1, Hence from now on -'-/^ = X{1- = Pisro the hazard rate for observing either a is corollary (s^D ~ jump in = J 0, J let either a jump us define: (30) 0)) = or a transition into s^d 1. An obvious that there are only tluree possible outcomes that can take place at those times, namely: J= 1 j , or (s^D = 0, J = 1 j , or (s^d ^ J 1, = Oj . Let pj=i,s=i, Pj=i,s=o and pj=o,s=i denote the respective (conditional) probabilities of these three events. We will not consider the possibility of insuring the duration of the sudden stop, and simplify the analysis by assuming that, conditional on entering a sudden stop, the transition out of independent of any jumps in J and happens with intensity The presence dXf^ = ^^ is Xf^ {nXf''' - \;^,Ft - Cf ^ + = 0. [l + (A'^ - l)st + ftPt] Yt} dt the dollar amount invested in the risky asset Ft. Since the asset Ft = [nXf + ^The reason J A. becomes: the equivalent dynamic evolution equation for dXf" is is of a risky asset with the above properties modifies the analysis in only two respects. First, the evolution of where it XlitFt - Cf" +[p+ {A' - X"^ is l)st - that the value function of neither agent Hence the times when s^o = and J = (W ftPt] Yt or EM) - ftYt (i + + itPtJd^t is in zero net supply, siq) } dt - ^tFtJd^t (32) = and experiences any change are irrelevant for equilibrium allocations, prices 15 (31) when s^d and welfare. As with that and we require that <^f > so that 0, EM has to pay upfront for the Ft contracts purchases.^ Finahy, also note that while equations (31) and (32) hold in both states it St reserves, — I, in equilibrium it turns out that = Ct when — St I, since the asset Ft has st = no hedging value in that state. Global equilibrium with imperfect risk sharing 2.5.2 The equilibrium definition in the presence of asset Ft modification that brevity. now is identical to definition 1 with the obvious the consumers optimally choose ^t instead of nt- The next proposition shows how We do not repeat it here for to construct an equilibrium in the presence of imperfect hedging contracts Proposition 2 Define a^t = = ^7- (33) K^ (xt) K^ (xt) given in the appendix, there exists an -^r-'Cf Then, for a system of differential equations , equilibrium with the following properties. 1. Pre-development, the consumption process for EM K° {xt) ifst = \ K'ixt) tfst = Cf^ _ ^* and the respective f is given by (34) W consumption process for is l given by (35) ^i I ^^{Xt) z/st = l where ^An ?P(xt) - {\-r^-~ift-K\xt)) (36) VL\xt) = Ul+l3)k'-{l+q)ft-K\xt)) (37) alternative assumption that will leave our results unaffected the same monitoring costs as growth contingent contracts. 16 is that short positions in asset Ft are subject to The equilibrium values of p{xt) and f{xt) are given 2. P as .fO + Kf P 1/ + Kp = 9- P = if St i V (38) '^ \ [xt) (i n^{xt) where the quantities f°{xt), f^{xt) in the state = 5f + ^ q) if St 0, 1 respectively I (39) are given as the solution to the equations + K '0 V [ Kf° - UXt ni (xt) and K /? 9- .{l-f')+i^xt I K° - + k/1 -VXt' ^' (xt) K 9- (i+q) K (1 (40) {xt) - /^) + UXt (41) K' J The equilibrium value of A* 3. — (xt) given by is -7 'n^xt+l) K=x and + Pj=i,s=o Pj=i,s=iA (42) fi°(Xt) the equilibrium value of n^ixt) solves the equation S,t QHxt+iii Pj=i,s=iA (43) Pj=i,s=q QO{x,.) \ \ ) K°(xt) Proposition 2 asserts that in the absence of perfect hedging instruments construct an equilibrium where to Proposition some 1, reserves in all it is the quantities depend exclusively on xt (and still possible to St). In contrast the absence of a perfect hedging instrument implies that in general its portfolio to insure against the possibility that it EM enters a sudden stop holds and the imperfect hedge does not deliver any payoffs. For our purposes, the most illuminating equation in Proposition 2 tion implies that EM The to see this easiest way = 1 is ^To equation (43). This equa- always wants to hold some amount of imperfect hedges as long as pj=i,5=i is to note that ^j > Otherwise, some amount of ^j St is smaller than EM's see this, note that the , i.e. is in is a root of equation (43) only when pj^ig^i 0. = 0. optimal as long as W's consumption drop upon entrance into as long as denominator on the the numerator needs to be negative = > order to n'(x.+C,) left n°{xt) hand make the K'i^fH,.) K"(xt) side of (43) left 17 ^ < hand side is 8 negative as long as i^-bZ < ) nffjf)- Henee and the right hand side equal. However, this Estimation 3 Before we can proceed with a quantitative analysis of the model and its imphcations, we need to obtain estimates for some of the parameters of the model. In particular we are interested in the and departure arrival sudden rates of the benefits of hedging, we stops, and sudden stops and jumps also estimate the joint incidence of appropriate financial instrument. The next section explains how we obtain The core of our analysis of such process: A and E{9°jel), where Ot is the sudden stop process. In this section is A. We also estimate the these estimates. we drop the simplify notation magnitude of the average = i + fipi, = j or less as described by the on the selection The during first superscript from Oj when there NSS and reversal, no place We for confusion. we had complete data and behave (see the Appendix for details Colombia, Indonesia, Malaysia, Mexico, and Thailand. to find empirical counterparts for the processes describing available resources is SS. For this, we note that in the model these resources can be regular income, Yt, and financial flows, {9t plexities in doing such a decomposition. relative prices is model with no contingent instrument criterion): Chile, step sudden stop {0,l}. use post 1983 data for six emerging market economies for which more we estimate the main parameters defined as el — l)^^ decomposed into In practice, there are several additional com- These stem from the existence of multiple goods whose change during the transitions between NSS and SS and vice versa, the presence of temporary terms of trade shocks, The Data Appendix and endogenous domestic output declines during sudden describes our methodology to deal with those issues. In a nutshell, imate Yt with the permanent component of domestic national income, and of capital flows in terms of imported goods of trade effects. Our main can only happen when is some in The Sudden Stop Process 3.1 To Furthermore, in order to assess their severity. ^j > left hand since Q,^ is {9t — stops. we approx- l)Yt with the sum and the transitory component of exports and terms side variable is decreasing in xt (since increasing in Xt. 18 the ratio of these two, which can be loosely W's holdings of bonds is equal to —Xt), whereas K° interpreted as external financing over "normal" pre-development income: where i the country index. is Since the measurement procedure for On can potentially produce measurement error and the data are time aggregated as opposed to continuous, we assume that Vit is observed with (state dependent) noise, with e^t{s^t)-^N{Q,al{s^t)), Since stops {st t/ij^ = 1) will exhibit economy as the 6 {0,1}. from tranquil times transits we use a standard regime-switching model a the average value of ipj^i jumps Sit ipji during sudden stops during tranquil times (that we (t/)j . ) As part of this procedure, probability of transition from tranquil times ^ ^ Pi{NSS Pi{SS 55) = iV55) [st = Pr(s,,t+A = = Pr(s,,t+A l|s,,t - - Ois,,t sudden ipf^^) and the equivalent value of we can also obtain estimates for the sudden stops 0) to 0) to Hamilton (1989, 1990) to estimate la call = (sf 0) = = 1) 1 - = 1 [st = 1):'' e-^'^ (44) - (45) e-^''^ For the calibration exercises we convert annual transition probabilities into annual frequencies by setting: Given the limited number A = - log [1 - P(Ar55 A = - log [1 - P(55 of -» SS observations we have ^ 55)] /A (46) N55)] /A (47) for each country and the highly nonlinear nature of the hidden states model we are estimating, we use a Bayesian approach (with based on a Gibbs Sampler Appendix. Pi{SS —> (see To obtain more NSS) are the Kim and Nelson precise estimates, same across all [1999]). in describe the precise procedure in the we assume that the parameters pi{NSS -^ SS), countries, whereas i'i'^^, ipf^ are allowed to differ. Tables 'These equations approximate the continuous time model by can be at most one transition We flat priors) its discrete analog, by imphcitly assuming that there a time interval of A. This approximation becomes exact as 19 A— > 0. Average Std Dev 5% 25% 50% 75% 95% A 0.19 0.050141 0.11337 0.15434 0.1868 0.22153 0.27772 A 0.14769 0.046718 0.079083 0.11403 0.14322 0.17685 0.23111 Table 1: Posterior distribution of A, A. Average Std Dev 5% 25% 50% 75% 95% Chile -0.091 0.030 -0.132 -0.109 -0.094 -0.076 -0.041 Colombia -0.051 0.009 -0.064 -0.057 -0.052 -0.047 -0.036 Mexico -0.084 0.015 -0.106 -0.093 -0.085 -0.075 -0.059 Indonesia -0.066 0.009 -0.081 -0.072 -0.067 -0.061 -0.049 Malaysia -0.164 0.023 -0.199 -0.178 -0.164 -0.149 -0.126 Thailand -0.146 0.023 -0.179 -0.161 -0.147 -0.133 -0.107 Table 1 2: Posterior distribution of capital flow reversal during sudden stops (77) and 2 present these estimates. Based on the posterior medians, we conclude that sudden stops are large, leading to last for about dechnes in available resources sometimes beyond 10 percent of 6.5 years previous sudden stop) in interest rates . and they occur about every 12 years (about Our measure and turmoil but of the capital flow reversal. of sudden stop also the effects of is meant GDP, 5.5 years after exiting the to capture not only the initial spike sudden stops that remain well Our estimates suggest that their effects after the onset these effects can be quite large: Countries take a long time in resuming significant borrowing after experiencing severe capital flow reversals. Finally, Figure 2 It illustrates summarizes the output of the corresponding Gibbs sampler It is apparent from this figure that our procedure does capture the transitions into sudden stops that occur around major financial The VIX and the To evaluate our economies. the path of the share of economies with posterior probabilities of being in sudden stop above 0.5 during a given period. 3.2 for joint incidence of the usefulness of hedging, when a country transits into a sudden we need sudden stops crises. zind VIX jumps to find an instrument that exhibits stop. If such 20 an asset exists, then it is jumps at times straightforward to use Chile 0.1 h- 0.05 Y ^v'^«^ 5- l\ -0.05 -0.1 Colombia W^ ' 1 ' \ 1990 1995 Mexico 1985 1990 1995 2000 Indonesia 0.05 V 2 CO 0.5 w. \^ ^ ) -0.05 1985 1990 1995 2000 1990 1985 Malaysia 0.2 0.1 :'" \ - to 1985 Figtixe a combination of detail how Our to 1990 1: tp'^ sliort dollar 2000 1995 1985 and posterior probability dated put and when the jump call is of being in a call and otherwise. (We explain in options shortly.) Rather, we seek to show that there exist assets with such properties and Moreover, by finding an asset that delivers payoffs stops, 2000 for 6 different countries. in the underlying asset occurs obtain an order of magnitude of the benefits of expanding sudden SS 1995 not to conduct a thorough search for the optimal risky instrument for specific countries' portfohos. of 1990 options on the asset in order to synthetically approximate approximate such a claim with goal here ^; Va ^u. . 0,1 1 2000 ^. 0.5 a payoff of 1995 Thailand we can use risk EM's strategies to include such an asset. in states of the premia on that asset to world with increased incidence infer the price of obtaining payoffs in such states. With this purpose in mind, we chose the formed from quoted put and call CBOE options on the Volatility S&P 21 Index (VIX). This is a traded index 500, available since the mid-1980s. The VIX Sudden Stops 0.9 0.8 LTCIVI/Russia-* I Crises in Asia Asian Crisis S« 0.4 Tequila Crisis 4- 0.1 1983 1989 1987 1985 Figure 1991 2: 1993 Year jumps in the calls on the VIX, making it S&P 1997 1999 2001 2003 Fraction of Economies in SS captures traders' anticipations of volatility in the by various puts and 1995 500.^" US stock market over the next month as implied As we show below, sudden stops often coincide with ^^ a good candidate for our purposes. '"For more details on the construction of the VIX, sec http://www.cboe.com/micro/vix/vixwhite.pdf ^'See also the recent work CDS' prices of, Pan and Singleton (2007) documenting the tight correlation between and the VIX. 22 EM's sovereign The VIX Process 3.2.1 Let us postulate a continuous time process of the VIX, described by an Ornstein-Uhlenbeck (i.e. a continuous time analog of an AR(1) process) with jumps dlog{VIX) a (log) is = -a (\og{VIX) - log(y/X)) a parameter that controls the speed of VIX, uvix to a standard mean second and third terms and then to is Xj^^dt] (48) . the average level of remove \og{VIX) and and standard deviation acj). is Finally, (f> assumed to be drawn Given these assumptions, the discrete time counterpart. Since (48) its discrete time observations follow an its this predictable isolating the residuals. in intensity Xj respectively. martingale differences. in (48) are an Ornstein Uhlenbeck process, is fi^ jump we approximate the above process by In estimation, - [(f>d^t log(y/X) reversion, Brownian motion and a Poisson process with from a normal distribution with mean step + avixdZt + the instantaneous volatility of the process, dZt and d^t capture the increments is captures the (potentially random) magnitude of the first dt AR(1) is process. Hence, the component by estimating an AR(1) process for log(V^/X) For small time intervals, these residuals are characterized (ap- proximately) by a mixture of normals:^^ £t where pvix = (1 =1- -py/x)iV(My/A'A>^WA'A) +pvixN{iiyjxA e"-^-''^, fiyj^ = + i^^,a^vix^ + (49) ^l) -Xjfi^A. In principle, estimation of this process can proceed from this point on along standard estimation techniques (see, e.g., Caballero and Panageas [2004]). Such estimation delivers estimates of the underlying parameters of the process for \og{VIX). Given these parameters, we can also estimate the posterior probabilities that the VIX also estimate the joint incidence of jumps has exhibited a jump in a given month, and we can then in the VIX and the arrival of sudden stops. Unfortunately, these estimates are not directly useful in practice, because in real markets one can only find contracts whose payoffs are contingent on the the month). However, there are no contracts on whether the '^The source of the approximation to the continuous time possibility of more than one jump and infrequent jumps and in hrait is level of the VIX A. 23 (say, at has exhibited a the end of jump over the that the discrete approximation excludes the the interval A, which seems reasonable relatively small time intervals VIX if we want to focus on relatively large course of a month, because such jumps are unobservable (and impossible to define) with discrete time trades. To measure the strategy, correlation between the arrival of sudden stops we adopt an alternative procedure. ^'^ In particular, We monthly data. of (49) using we let first > = e|J 0) < and consider a contract that interval of lif log(F/Xt+A)>log(V^/A7^J is above small A, the payoff a payoff that it is more is 1 if 1 if +e the unexpected change in the \og{VIX) over an also important to note that in the limit e. It is small, this contract pays off and only Qt+A approaches if there is a jump in the where VIX A and becomes from arbitrarily otherwise. Hence, for the contract described in section 2.5.1.^^ Given that (50) measurable with respect to the market participants' information useful be defined as otherwise this contract delivers a payoff of A A = {1- aA) \og{VIXt) - aA\og[VIX) \0 is, such that delivers the following payoffs: Q^^^^i That e, 0.01. the expected level of ]og{VIX) after a time interval of log(y/Xi%A) feasible perform a Bayesian estimation then use the obtained estimates and set a cutoff value Pr{et In a next step we and the payoffs of a Qt+A = this point to think directly of the event set at time t is + A, as a "jump". 1 Furthermore, figure 3 illustrates how to approximate the payoff Qt+A by combining a standard call option with strike price ]og{VIX^_^_^) \og{VIX^_^_^) + e + 6. The + e f e + 5 \ogiVIXt+A) < log(y/Xf+^) + e 6if\og{VIXt+A)>^og{VIX^^^) + (51) if As was shown by Breeden and Litzenberger Given '^If anything, this this option with strike price < [ a-e. call resulting payoff is^^ Qt+A = Qt+A together with a short approximation result, (1978), when ^ 5 and the existence we obtain that /im^^o ^"^ = of a rich set of options of various procedure biases the results toward finding a lower "correlation", because it introduces the potential of "over-identifying" jumps, and hence understating the benefits of hedging. ^''When A = course of that ^^When 1/12 month (i.e. is monthly data) the probability that Qt+A 1%. This log{VIXt+A) e is so by the construction of [\og{VIX^+^) will be 24 when no jump has occured over the e. + e,logiVIXt+A) + e + 5] (log(V/X.%^)+e). 1 then Qt+A = \og{VIXt+A) - Payoffs Payoff of Long Call Option Sum of the Payoffs I \ogiVIX „^) Payoff of Stio rt Call Option Figure 3: Payoffs of a long call option with strike price B= with strike price strike prices, \og{VIX^_^^) we assume +e+ S. The A= \og{VIXf_^^) solid line presents the directly that the payoff given by is (50) +e and a short sum of the call option two payoffs. a feasible payoff given existing securities. Importantly Qt+A for for small 6 is our purposes, equation (51) implies that the arbitrage free price of the payoff given (approximately) as the price difference of two standard strike prices \og{VIX^_^_^) + e and \og{VIX^_^^^) +e+ 5 respectively, divided by options with call 5. Denoting that price as P'^t-^^^ a feasible strategy to replicate a payoff such as (29) in section 2.5.1 follows: Every time interval payoff such as (50). there is If A (say every one or two months) an the (unexpected) change in the is less than can pay P<5«+a and obtain a e , then Qt+A a discontinuity that produces an (unexpected) change larger than the risk compensation A* in equation (29) as'^ A* details VIX EM on how we used the prices '^In the limit where 5 — > and A— » of existing this VIX ~ call e, — 0. If then Qt+A P^'+^/A. In the appendix we = however, 1- Hence, give further options to obtain an estimate for the average approximation becomes exact. 25 given as is Average Std Dev 5% 25% 50% 75% 95% Pr{SS\J) 0.236 0.096 0.095 0.165 0.226 0.296 0.413 Pr{J) 0.159 0.032 0.109 0.137 0.158 0.180 0.215 Table 3: Posterior distribution of Pr {SS\J) and Pr( J) based on quarterly data, value of P'^t+A and hence A*. Based on that analysis our baseline value for A* 3.2.2 It is is 0.94. Joint jumps, risk premia and implied preference shocks now straightforward to measure the joint incidence of sudden stops and positive payoffs to a claim such as the one defined in equation (50). Figure 4 gives a incidence. It shows the residuals of an those instances when AR(1) model log(y/X). The shaded areas represents these residuals are above e and hence the times constructed in equation (50) would pay off 1 dollar. onset of the gulf war) in 1997 (around the Asian after 9/11/2001, for visual depiction of this joint first and around the beginning The shaded crisis), in when a claim like the one areas occur in the early 90's (at the 1998 (around the Russian/LTCM of the U.S. corporate scandals crisis), and the Argentinean default. These large movements in the VIX at times of crisis, together with the increased incidence of sudden stops during such times, suggest that there can be used as for hedging purposes. "jump" times and we will We is some will refer to the times denote such events with J Conditioning on the times where J = 1, it is joint incidence of the = 1 when two events, which the payoff of equation (50) is 1 in analogy to section 2.5.1. straightforward to use a Bayesian procedure similar to section 3.1 in order to estimate the parameters P(SS\J - 1) = Pl^h£^ and PiJ) = 1 - e"^^^ PJ = 1,5 = +PJ=1,S=1 where pj^i,;^i,pj^i^s^Q etc. were defined in section 2.5.1. The exact procedure appendix. The resulting posterior distributions are given in table With the estimates of P{J) from table 3 , namely Xj the estimates for A (from section 3.1) and also Aj, and 26 of the arrival intensity — - log [1 - P{SS\J = described in the 3. we can obtain estimates a manner analogous to equations (46) and (47) is 1) Xj in P{J)] /A. In turn, given from table 3, we can identify VIX Residuals and Jumps Jan92 Figure 4: VIX residuals. Jan94 The grey JanOO Jan98 Jan96 Jan02 areas correspond to instances where the VIX residual is above the jump-cutofF. all the parameters of section 2.5.1 by solving the following linear system of equations Aj X(PJ = 1,5 = 1 +PJ = l,S=0) (52) A X(PJ=l,s=l +PJ=0,s=l) (53) PJ = 1,S=1 Fv{SS\J) (54) PJ=l,s=0 +PJ=l,s=l I for the four for A, Xj, unknowns pj^j and P{SS\J = 1), of equations (52)-(55) are x PJ=l,s=\ + PJ=l,s=0 + Ps=l.J=0 ^^j, Ps=i,j=o, Pj=i,s=o and x- Based on the median posterior values the values of pj=i,5=i, Ps=i,J=o, Pj=i,s=o and = 0-74, pj^i^s=i = 0.23, 27 (55) ps=i,j=o = 0.06, x that ps^ij=o = solve the system 0.71. Calibration and Results 4 In this section we input the estimates obtained in the previous section into a cahbration exercise, and then use the resulting quantitative model to evaluate the gains from different precautionary Since the standard practice of central banks in emerging markets strategies. is to hold reserves almost exclusively in the form of noncontingent assets, we calibrate the model of section 2.5.1 assuming that a country does not use any hedging instruments. This standard practice seems to be more the result of inertia how much would to ask and domestic pohtical economy welfare increase if and hence factors, is it reasonable countries adopted optimal hedging strategies. All the counterfactual exercises are done in general equilibrium, and hence take into account the endogenous reaction of prices. Parcuneters 4.1 We calibrated the model following two different approaches. In the CRRA = preferences with no global preference shock (A this kind, this approach is able to match quantity data in actual markets, especially during times of turmoil. 1 first for all approach we use standard As with t). all well but not the large risk Within calibrations of premia observed approach, we allow for two this first subcases -one where cyclical components in world output are small (case la) and hence sudden stops are caused almost exclusively by changes in monitoring costs, and one where sudden stops coincide with significant global downturns (case lb). In the second approach (case 2) jumps in the preference shock price of risk. helps us fit We As in order to calibrate interest rates also allow in this case for a EM common across all cases. for the three cases The parameter g difference between these cases is we income of a developed economy. The parameter in The consider. set to 0.03 to is that W's We first twelve parameters approximately match the speed of set k = 3, an emerging market economy g\og{K) of (logarithmic) growth of we in the global growth rates during sudden stops, which convergence estimated by Barro and Sala-i-Martin (2003). Since for varies across them. Table 4 reports the parameters are in The main the volatility of interest rates. willingness to insure drop and fluctuations we allow controls the relative size of so that the expected rate — EM 3.2% higher than that and W in the model. are interested in estimating the benefits of hedging for a large part of the world, such 28 Case la Case lb Case 2 g 0.03 0.03 0.03 K 3 3 3 13 7 7 7 A 0.14 0.14 X 0.74 0.74 0.74 Ps=i,j=i 0.23 0.23 0.23 Ps=o,J=i 0.71 0.71 0.71 Ps=i,j=o 0.06 0.06 0.06 i 0.01 0.01 0.01 Q 0.35 0.35 0.35 7 7 7 7 P 0.01 0.01 0.01 a 0.027 0.02 0.02 Mo 0.018 0.018 0.03 Ml 0.018 0.018 Mg 0.018 0.018 0.018 1 1.8 0.96 0.96 A 1 A'i . 0.99 Table 4: Parameters used in the calibration exercises. 29 . 0.14 GDP we added the (PPP adjusted) as Latin America, GDP compared that to the American of Latin US, of EU countries and Japan The parameters countries.-'^ and then p^^u^Q, pj=i,s=i, Pj=i,s=o and X are those that solve the system of equations (52)-(55) as determined in the previous section. 9% GDP of flows A was estimated and reported in tranquil times. become The in table value of q We 1. set high is We sudden stops. practically zero during chose so as to i chose 7 to generate reserves accumulation in the 20 — 30% growth rates of (see Y fiQ = fii — fXQ = but assume some variation in the riskless interest rates. ^^ In case 2 we 2 Y A calibrate premia (assuming that originally only in set the developed economies match the observed volatility in real match the observed VIX's to approximately in this market). Finally, in all cases case la produces a standard deviation of the cyclical component corresponding to what The drop output. ^^ in cases risk we set The drop At to capture the standard deviation of the cyclical components in output in the US. US we For cases la and lb growth rates in order to W participates from a Beveridge-Nelson decomposition of we preferences. Finally, preserve (approximately) the average growth rate growth rate of we three cases capital range. 0.018, so as to capture Campbell and Cochrane 1999). In case CRRA uses parameters varies across the different cases. to all p to a low number in order to reduce set when one set of capital flows of about enough, so that in the high risk-free rates that are typically obtained, The next match is in obtained lb and 2 produces a model-implied standard deviation of cyclical component that corresponds to the one estimated from a standard H-P volatility of post-war We US filter decomposition for the US.^'' output produced by the model We set ay so that in all three cases the annual about 0.028, corresponding approximately to annual is data.^^ also estimated the model by matching the relative size of US, and East Asian economies (Indonesia, Korea, Singapore, Thailand and '^The presence EU and Japan to Latin American countries Malaysia.) obtaining similar results. of preference shocks necessarily increases the volatility of the interest rate. By allowing for some small time variation in output growth that volatility can be reduced within reasonable levels. Since in case 2 we are interested in matching asset prices, we allowed such time variation in the stochastic trend. However, we also simulated the model allowing for constant growth rates and preference shocks. This did not affect the results substantively. ^'See Nelson (2006), p. 12. ^°See Nelson (2006), p. 12. ^'Note that the total variance of output is given as E [Vart (Yt+i — is the variance of the predictable component Var[Et (Vt + i total volatility stays the somewhat the same compared total volatility of to case lb. Yt)] — Yt)] + Var smaller, we assign a In case 2, the variation in output compared to case 16, 30 but this effect [Et (V/.+i — Yi)] . Since in case la) larger value to growth rates fJ,Q,fJ.i ay, so that tends to reduce has a negligible effect on total volatility Model Data Case la Case lb Case 2 Reserves 0.31 0.32 0.2 0.17 Mean Resource Drop 0.091 0.091 0.091 0.1 1.01 1.04 1.33 1.39 0.14 0.14 0.01 0.020 0.008 0.04 0.04 [0-0.05] Mean Jump Risk Intensity ratio (A*/Aj) Mean Interest Rate Volatility of the interest rate Table Table 5: Comparison between model and data, assuming no hedging. 5 reports several statistics accumulated and the corresponding values column are based on the countries that The jump as ^ we consider as reported in Table 9 in the appendix. risk intensity ratio is GDP) (In the Mean W participates resource drop refers to the in this market). The data on the average real interest volatility of the real interest rate is from data one can only obtain estimates about the volatility of the realized anticipated real interest rate, as an interval.) data obtained by our estimates of \j and A* in the previous section on bonds. These estimates provide an upper bound on the real return in the during the sudden stop as reported in table Campbell and Cochrane (1999), while the (Jermann 1998). reserves are following: Reserves refers to the historical average levels of reserves for the (assuming that only rate are from The numbers our sample of countries. ^^ in average reversal in capital flows (as a fraction of 2. when only noncontingent generated by the model and hence we report the volatility of the unobserved volatility of the anticipated real interest rate The models produce reasonable numbers accumulation and the for reserve the sudden stops. However, as explained above, only case 2 is size of designed to match asset prices by introducing an additional source of shocks to the marginal utility of consumption. In particular, the jump premium" in As is needed to generate jumps implicit in the they generate a positive (about 20 beisis VIX during drift in the in crises. marginal marginal Jumps utility consistent in ^4^ also help utility of with the rise in the "risk reduce the interest rate, since consumption. points reduction in annual volatility). "*To obtain the stationary quantities we simulated 20000 development. 31 artificial yearly data, conditioning on no transition to Hedging 4.2 Case la Case lb Portfolio (imperfect correlation) 0.87 0.86 0.78 Portfolio (perfect correlation) 1.00 1.00 1.00 Proportional reduction in stationary reserves (imperfect correlation) 0.16 0.16 0.23 Proportional reduction in stationary reserves (perfect correlation) 0.42 0.47 0.62 Table Case 2 Portfolio allocated to hedging and proportional reduction in stationary non-contingent 6: reserves (reserves prior to hedging divided by reserves after hedging). In this section we study focuses on the former. spent in hedging, that The the effect of hedging on portfolio decisions and welfare. first available (imperfect correlation) occurrence of a sudden stop is and is + for OfYt — for the case Cf), The where only options on VIX are an upper bound where an asset perfectly correlated with the available (perfect correlation). invested in hedging instruments. two rows two rows report the share of the flow of precautionary savings X*^^l{rtXt is Table 6 We refer to — portfolios are evaluated at xt clear in all cases: the bulk of the precautionary resources, this ratio as the "portfolio" The message 0.^'^ of these even when the correlation is imperfect, should be allocated to hedging instruments rather than reserves accumulation. We already knew from Proposition sudden stop insurance that EM is is 1 that this "portfolio" notion is always 1 when there exists both perfectly correlated with the sudden stop and also insures against the duration of th& sudden stop ("complete risk sharing"). Once the price of risk correctly assessed, a country always prefers a hedging instrument to non-contingent reserves, irrespective of the magnitude of preference shocks that this mechanism remains intact even stops (second row). This hedging The is limited to mechanism VIX is when A there and the associated is weakened but risk premia. Table 6 shows no insurance against the duration of sudden still remains dominant (first row) even when instruments. third and fourth rows of 6 describe a stock rather than a flow concept. ^''Evaluating the portfolio at xt = is conservative. When we solve the model the portfolio at levels of xt that are higher than zero, the denominator rtXt numerator X'^,, making the portfolio rise toward 1. 32 + They for all three cases BtYt — are based on and we compute Ct decreases faster than the a simulated run of the model with and without hedging. ^^ decline in average non-contingent reserves that different cases that reserves when What is it we consider. The is The two rows report the proportional achieved with various forms of hedging in the table shows that EM holds between 16 and 62 percent less can hedge. We the impact of these changes on welfare? by computing the equivalent variations particular form of insurance. For this, from hedging strategies income required to compensate in let assess the gains for the absence of a us define the no-insurance case as one in which EM only trades in growth contingent contracts without any access to either noncontingent bonds or any form of hedging. In such a world EM's utility is: (A. fsPsy-'Y, 1-7 Jq Jo where oo 1-7 JtO Correspondingly, country (in let V'^*(0, Yq) and V"*"''(0, Yq) denote the EM's utility in a world in which the allowed to use reserves only to insure against sudden stops and hedging instrument ins is addition to reserves), respectively, both evaluated at zero initial reserves. variation is defined as the additional pre-development income that needs to be given to an with no access to insurance or reserves, in order to achieve Since utility Then, the income is homogeneous of degree 1 —7 = (l + fc^^*)^-^Eo / Jo yi"- = (1 + fc^"^)'^^ where the fc's Yo) Eo and F*"*(0, Yq) respectively. with respect to the country's income, we have that: y-=^(0,yo) (0, y^'^*(0, Yo) EM A / '^'^^ ^ 1-7 r Jo A^ i^^ + fsPs)^~'"^s 1-7 e'P'ds \ .ps ^^ + V^^" ^ ydev represent the proportional income variation. Since our goal in this section is to evaluate the improvement over the standard practice of accumulating noncontingent reserves, we ^""See footnote 22 for a description of the simulation design. Importantly, when we compute average reserves, we also take into account states of the world stop (as in Proposition 1), it when Si = 1. When EM has to hold non-contingent bonds has no way to insure the duration of the sudden in this regime. Because of this, even when there is perfect correlation between the arrival of the sudden stop and the payoffs of the contingent instrument, the stationary reserves will always be non-zero, unlike in Proposition 1. 33 report the welfare gains as the proportional gain over fc'"^*.^^ Table 7 shows that the extra benefits Case la Case lb Case 2 Imperfect Correlation 20 15 8 Perfect Correlation 91 60 44 Complete Markets 114 81 60 Table 7: Percent increase in the relative effectiveness of hedging (fc^"'' — for various /c^*^^) /fc'"**, cases. from enriching the precautionary strategy with hedging instruments ranges between when only VIX of k^"^^ to 114% results we move as is used, from 44 to 91% to the complete markets case. Importantly, = 1 utilitj' others, 5 all "hurts" W more than making the in the price of risk. The W's income for risk falls sharing is and in case 2 W's marginal larger in case la than in the benefits of hedging larger. Remarks Emerging market economies hold by developed economies latter, the and from 60 that in cases lb and 2 the transition into state (In case lb This means that the potential increases). Fined in case la. is and 20% these are general equilibrium and therefore do take into consideration the endogenous changes reason for the difference between the three cases St in the perfect correlation scenario, 8 levels of international reserves that greatly (relative to their size). This would seem paradoxical given that, unlike the former face significant financial constraints with The paradox disappears once exceed the levels held much of their these greater financial constraints also of volatility, which countries seek to smooth. This is growth ahead of them. become an important source the context we have modelled, analyzed, and assessed quantitatively. ^"The absolute value of k^"" itself instance in cases la), lb) and 2) magnitude and any k''"^ is rather small in line with 0.26, 0.32 all welfare calculations in the Lucas tradition. and 0.23 percent of GDP as Lucas's estimates of the costs of business cycles (0.1 percent of links welfare gains, problems. is respectively. This GDP). is in For the same order of Since the absence of heterogeneity, between temporary and permanent consumption components may be understating the magnitude of we use a concept An added advantage of proportional increases in welfare benefits that should is that it makes our results less sensitive to 34 be more immune to these assumptions about risk aversion etc. The main contributions of this paper are twofold: equilibrium model of sudden stops. Second, to insure emerging markets against sudden we use First, we develop a quantitative global- this structure to discuss practical stops, ranging mechanisms from conventional non-contingent reserves accumulation to more sophisticated contingent strategies. Depending on the source of sudden stops, their correlation with world events, and the quality of the hedging instrument available, the gains from these strategies can be substantial. On the first contribution, the model is matching the extent of capital flow successful in the behavior of risk premia and the size of reserves accumulation. On the second one, reversals, we estimate that the potential gains from adopting hedging strategies are, in a worst case scenario, about 10 percent more than conventional reserves management, but the gains can also reach up to efficient 115 percent as the quality of hedging instruments improves and the correlation between sudden stop arrivals and marginal investors' risk attenuates. All these strategies imply a substantial reduction accumulation and in the rate of reserves productive investments. in many For example, in our case practical instances up reserves free for (which calibrates quantities and prices), we 2 compute that our representative country would need may to hold between 23-62% less reserves than it does without hedging (depending on the quality of the hedging instrument). There are several natural extensions frictionless trading in (imperfect) to our work. Our model assumes hedging instruments to explain observed risk premia. This mech- anism allows us to reproduce variations in the marginal utility of consumption of the representative agent in a spirit similar to Campbell and Cochrane (1999). As a ful since it preference shocks and first pass, this approach safeguards that welfare computations are consistent with asset prices. An is use- interesting extension of our model would be to assume some form of segmentation between bond markets and hedging instruments and derive the observed in these markets. It is risk premia from the heterogenous participants reasonable to conjecture that in such a world, increasing participation in the markets for (imperfect) hedging instruments by agents in the developed world would outcomes resemble the cases la) and lb) in make the our model. Furthermore, we have assumed a representative agent within each of the regions of the world. It is quite apparent that in reality there are plenty of financial frictions within each of these regions that clearly interact with the aggregate risk By the same token, it management and sharing problem we have discussed. seems important to endogenize the production side of the economy, as 35 in practice real investment and employment are severely 36 affected by sudden stops. Appendix 6 A Proofs Proof of Proposition Shreve 1998), Ciiapter { ^w W's endowment post development = pEt The absence of arbitrage is f J- so that results in (Karatzas and dynamically complete and hence there exists a for all for any > r2 (56) i > r*-" is then (57) Ht implies that the value of the payouts that W's intertemporal budget t H-.G ^nds. I t _ dW /t-gP^, is (jEM p,'^ « on development. By the first Ht such that p-p{t~r1) value of a claim to by focusing start market post development 4, tlie stochastic discount factor The We 1. W is collecting constraint post-development from EM post development is dt ' + ' (58) + Using (56) inside (58) and simplifying, leads to i+^fro)prc+x!:^a = c: The market IT J^-^o + w -<w Ht Cf — / / J^a H^G —rrrdt w C C% E^a e-K*-?) I 4 1-7 dt (59) clearing condition (12) together with (56) implies that ^t = e-p(t-T?) H-G ^t ( for all t > T° (60) + Using (60) inside (59) and (57) allows us to compute the expectation of the integral on the right hand side of (59) exphcitly. at time r^ Using the fact that Yt is lognormally distributed, allows us to compute W's consumption as Cw i {l3 w + Kf,c)+yx'^G (61) y:> For future reference, of W at time t% as it will also be convenient to combine (61) with (56) and (1) to obtain the value function A'^roy^ {X^G,Y^a) where 1 —7 ' ;y 37 ' (62) The rest of the proof We equilibrium. save space. A is V^{Xt, the processes for straightforward and is C^ ^Cf^ ,it, ft,nt and hence W's value function can be expressed ' when Ft) if 1 We are optimal. constitute an we omit to it by showing start Since the preference shocks enter multiplicatively in the objective, the value function A in markets clear shall verify this directly. Verifying that W. optimality for ' devoted to showing that the prices and quantities of proposition remains to check It homogenous is = Sj 1. Specifically, the A as ''V°{Xt,Yt) when Hamilton Jacobi Bellman (HJB) equation = St and as W when for is sj = given by = (i2LLl + v°{nXr-Cr + ma^ +^yi>o + \v?WYy- -{p + where V'^ when = Sj = is + 9)V° X (63) the value function in development. + Vk {nXr - Cr + ^7 J +V^-Yfio (64) + and \vi-yo\y'' - (p + A + ff) [PA' -h{pt+-i first is + q)- W order conditions for = V? ,- /^ in n.) Y,) = Vl, fori Y,; h) + = the states j 0, 1 0,1 ~XV' (X^, Y,) (65) can be written as K,(^+«/r'' + n) = 9-——^-=^ . (j The equation in (66) follows by re-arranging the first , C^^ 5:^ - given by (16). + gV^ [X^ (64) (63) imply the following first order conditions for P^ where v for V^ (Cn"^ = Similarly the Hamilton Jacobi Bellman equation Similarly, the given by 1 is inp Equations {P-ft{pt+i)+nt7T,)Y,)+gV^{xr,YtJt)+XAV^Xr,Yun,)] (^ + 3i) for j = 0, first (66) order condition for /t and the second equality follows by combining (65) with (62) and noting that in the postulated allocation Xf^'^ the = Xf^ = same for all t. Note that equation (66) is identical to the equations (19) steps as in equations (63), (64), (65) and (66) for . Pt=9 To determine the optimal choice of Ut for EM i-^fl-Z/V^ ) .^J-, W in state 0, denote the time of entry and t the time of exit from state respect to rir = 0,1 shall first St Repeating (21). gives for J we and = 1, compute V^ differentiating (67) [X^ ,Yt;nt) both sides . Letting r of (64) with and using the envelope theorem gives = -V^Yt+Vl^ (nAT - cr + (/3A' -ft{Pi + + q)i 38 nr) Yt)+VlyY^i,+^Vlyy<rlY^--[p + A + g) V^^^ The Feynman-Kac Theorem above to the differential equation is and Shreve 1991) and (Oksendal 1998)) imphes that a solution given by = -E n^-^t'-^Vi-ytdil = -E Vn. The second f/^'e^"'*-^' {c'''^)-'Yl-'d^ = -A = \^n^r V°Yt (63) and -1— {c°'^r^ ^^ (69) ^ W nor EM, the analog of equation (69) ' holds as well, so that nt Finally since Equations = — AtJ? Since there are no constraints in the choice of Ut for neither for j (68) order conditions with respect to n^ to obtain first EM {Y^c'-'^f-'vo Having computed Vn^ we can now return to explicitly the associated expectation. TTt for =- equality in (68) follows from (65) whereas the third equality follows by using lognormality of Yt and computing take (see (Kaxatzas Xf^ = X^ = ^^ '—-. in the postulated equilibrium, we (70) also know that cO.i^^ = /3_(pO_^-)^o^^„ ^i,w ^ A^/3-(p^+i + g)/i-n (72) ^^.BM ^ i+pOfO_^^ (73) ^i,BM ^ A-^+pV^+n (74) (66), (67), (69), (70), (71)-(74) 0, 1. = Aro— Plugging (73)-(74) into (71) form a system of ten equations (69), in ten unknowns {p> , f^ ,n,n,c^'^ jC^'^'^) combining (69) with (70) and using the market clearing condition (12) leads to ^i.w Ad(^+i)_(^ + g)/ 1 ^i,iv (75) Equation (75) implies that ci,tv_ Ad(/3+l)-(i + g)/l (76) p + i^lfo cO.iv Using (76) inside (69) we arrive at equation which is precisely equation (18). Using (66) and (77) inside equation (71) leads to -7 .0,w,,_..,;),.„„,,_,;(jLt3QZ,.,,„^f l^^')A'-gr'^' 39 | » (78) Solving for n from equation (72), and using (76) and (66) gives + ^f n-^ {P n^A" A'^{P+l)-{i + q)P f rO,W (79) P+l-lf° J Using (79) inside (78) and rearranging, we arrive P + 1 at -gH^^n^iP + ^fV /O.A.^ig^/^ X^AA^(^I^1^,±^ „o,w (80) 1-7 Combining (66) and = (67) for j leads to 0, 1 niP+^^fy K (/? " {{P + - 1) A"^ + (i 1 + - if — IM-^ q) « \. , (81) (cO,W)-T 1/ [P + nf n-7 -p+g) " - P = g- ^ rO,W\ (82) vd>'^^) where + l)A''-(7+g)/i P + l-~if° (/3 By solving the system of equations (80), (81) and (82) for obtain n, into (77) to obtain tt, into (66) to obtain p° f^,P we can c°''^', then substitute into (79) to and p^ and then inside (71)-(74) to obtain the rest of the allocations. To complete the verification that the postulated allocation the consumption processes Cf^ optimality for W it C^^ and X^^ X^ (cry = E, -''"') \ (34)"'^ e(^''"-" ° '^" *" " {C'^y ^- Nt =Et{e (^) (/o'('-.-o)'i") f-T\Nt (a) ' (^w\-i (Cf^) ,l,M/\ r° - p + A M,W is a martingale. Applying -T + p+ g nfo\ r.O,W 40 for all {Ct) Hence, an equivalent way to test whether the Euler equation holds _ and W respectively. To show i T> and i leads to {]J'(r„-p)du) (Cr) [A) EM are optimal for , Multiplying both sides of this equation with e e an equilibrium, we need to show that check that the asserted allocations satisfy the familiar Euler equation; suffices to (A)"' , is \ is Lemma Ito's for all to check in state i and T> t. whether Zj, defined cis gives —y - 1 dt + dMt (83) where dMt represents martingale increments. Using equation A hence the Euler equation holds. when St = and driftless argument can be applied to show that the Euler equation holds to EM, our assumption that short positions in bonds are subject to the same monitoring costs growth contingent contracts implies that EM. is 1. Turning as similar (17) inside (83) implies that Zt In particular, it suffices to ^(;„V„_„,.„) Equation ^^;v, borrowing at the rate rt + EM it is sufficient to check that X^'^ = check the following pair of Euler inequalities for ^^^^y, ^ (84) asserts that it is Hjf <..-.,..) ^^ ^^;v. not optimal either. To show equation (84), development and r^ denote the instant after all t ^c^Myl\ ^^ does not want to invest in bonds at the rate a "corner" solution is < ^jj ^ and t'^ ^^d all Sj T> 6 for {0, 1} t. (84) vt, whereas (85) asserts that let r't denote the instant before development. Then for states j = 0, 1 equations (81) and (82) imply that ^I;em C'}^ J^)=\^\ +-i' g^'^ + ''" C^f \ ' j For times prior to development (r'EM\ where the follows is first line from the {t < t'^) 1 ' and any T> — i I \ C'-j""' I / C M 7^ < I (86) I -G we obtain !r'EM\~''i follows from the fact that TOT lr'W\ ( W's Euler equation — '' holds as an equality and the inequality (if T< r*^) or is larger than 1 (if T> r"^) by equation (86). To show (85), apply Ito' use (17), (20) to obtain _k ^ ' ^ e-.*(l)'^'(Cf-)where dMt =^ / fact that either equal to Lemma and \ ' is ^ =-rtdt + g\ ^ ^ ^ U^.-) - ^ A K-) ^ ^ dt + dMt (87) a martingale increment. Equations (81) and (82) imply that -7 (88) 41 Combining Assuming and (88) tliat z/ (87) < 1, if we obtain holds. This concludes the verification that the postulated Xf^ = are optimal for Lemma Proof of Then equation c^,wi^Q,w = = for j = fr > ^ ) /3 and letting Using these facts inside it 1, (89) with (25), (73) follows that /° and (81) yields Proof of Proposition W is -^ oo shows that c°-^^ /P /3 (19), (21) 2. > (l /i. + and — >• 1. (18) gives (23). Since ^^^> ^A-^) and using • = A'^ q}-'^^ jcP^^^^ implies that 1-/1 /p°\^ Rearranging the above equation yields EM and combining Verifying that the proposed allocations and prices constitute an equi- W are the same as in proposition 1 given by equation (62), while the value function of 1/° (25). Finally, but tedious manipulations. (24) after straightforward, librium pre-development follows similar steps to the proof of proposition function of both the asset process (/V-/V)A^ 0, 1 1 ( Cf^, and c''^^/c°'^^', equations (73). (74) imply that Furthermore, applying equation (67) t^z consumption process Dividing both sides of (80) by 1. _ "- Since a submartingale and hence (85) is EM. (76) implies that c^'^' /P —* A"^. A-* {Cf^y follows that e(^o('-"+'"-p)'^«) (a)'^' (X,c,nc) = ^i 1 — 7 EM \k{\ Post development, the value 1. This implies that the value function of agent . is - /,g) + vx,a\^-^ (90) ;y vEM Since in the proposed equilibrium all —= the pre-development equilibrium prices depend exclusively on '-^ — 5^, the value function of both agents can j = {EM, W}. = V of agent Yi, W Si, and Xl where in state Sj = Hcn^\ yO^^^^w_cW^(^0_f^(^p^^j^^^^^\^'^Y,)+gV^{xr,Yuf,) [pj^i.s=iAV' +V°Yf,y^o Computing the be expressed as a function of For instance, the Bellman equation for the value function max +X still first (^X^-lYuYt) +pj=i,s=oV° (xr (91) +PJ=o,s^iAV° {X^ ,Yt)] + ^V°yalY'-{p + X + g)V°^ order condition for V°X:Yt -Ct^t.i^t) is ^^ leads to = X [pj=i,s=iAVk (Xf - lYt, Yt) + pj=i,s=oV^ [x^ - ItYu Yt)] 42 Yt (92) By combining jjj'BM _ -x}^ condition for (92) with the first order condition for and using the definition of Xt EM By combining (42) when pj=i,s=i = arrival of St We = and 0, (Cl^)"^), using (35), noting that equation (33) we arrive at (42). in then zr^ PJ^l,S=lA we (93) jumps there are no joint when = (V^, The analogous first order is K=x and the consumption ^j = (93) also explain in the text, equation (43) implies that Hence, unless there are joint jumps in the marginal 0. in the As we arrive at (43). in the asset Ft, the jumps +PJ=1,5=0 marginal optimal holdings of Ft are utility of = S,t 0. By consumption and the jumps utility of consumption sj = ^j = assumption, in state in Ft and therefore 1 1. next turn to the determination of the optimal /° and the equilibrium price p". Differentiating (91) with respect to /°, using (35) and (34) and repeating the same steps as order conditions (38) and (39). Deriving the analogous with (38) and (39) leads to (40) when = St and first to (41) as functions of K° {xt) ^. Sofar we have derived when EM St = 1. To We 1, leads to the first and combining them simplify notation, will also we shall denote apply the same shorthand the quantities of interest (/°, /'^,p°,p\ A* and and K^{xt). To complete the equilibrium verification, it remains to derive these functions K°,K^. utilizing the Euler equation. Formally, define tq to To Using similar steps as all Proposition order conditions for the solution to (40) as f° instead of the more lengthy f'^(K°{xt)). notation to /"^,p°,p\A* and in in Proposition 1, = be the smallest time after time inf {X, = s>t we obtain that for t We do such that Xto by this = : 0} any time t <T < tq, the following Euler equation must characterize EM's consumption A''^^Ct'^My'r^E^^^(r,-p)(T^t)^'^T^^EMy-r^^ An identical equation needs to hold for W Following identical steps to the proof of Proposition St = imply that Applying Ito's e''"'"''^*^'^' Lemma {C^y (94) and 1, e<'''-'')*34'^' one can show that equations (94) and (95) when (Cf ^)"'^ are both (local) martingales when Xt> 0. to Xt gives dxt = d (^^\ =^+Xtd (y] 43 = a{xi)dt - XtadBt (96) where a[xt) is defined a^ a{xt) Using (96) and applying Ito's = - (r°(xt) Lemma Mo + ^') to e''"'"'''*^ ' Xt - - XUt + K'^ixt) when {C^^')~"^ 1 = s* + f?P° (97) gives d[e(-.-p)*A'^'(Cf^)^^] l"^"' (nEM\~y eCr.-P)*^'"' (C, = <!'-"(a:t)-p-7(Mo- +X Pj=i,s=i^ d.K" dxi I [ where dM^'^ is +^+5 y) + Pj=i,s=o i^° (xt) 9 f f 2\ \ (98) ^ '^ 2 , , Since a (local) martingale. 7 (d,T[p "'"^t , of K°,K'^, ^, ^^ to determine i<"°, and Xf We > \ / I Mo - y > 2 Kf-^G +-^- + 5 1 nUx, + i, dit I dM^ we can (98) r (k°,K\ Tit no is 2 2 I I \ y'^^^t) ~ 2\ y^t'^) + , '^ Xi a local martingale. Since set -^ with the must VX^a - also hold for . Finally W, namely (99) 1 (xt) , , n^{xt)Y'' 0° [xt] " (dxi)' (7+1) no e'^'~''^*yl ' The terms them equal and use equations [97]) ^, ^^,xt) fn°{xt+~Q brackets in equation (99) must be zero. so = fi° [Xt) \ where a (equation + Pj=i,s=o Pj=i,s=iA f dn° -y definition of zero allows us to solve for r°[xt) as a function +P— n° a local martingale, the term inside is denote this function as r°[xt) shall + dMf ^ dt ''' [Cf^^ is 2 s dTt Combining the brackets of (98) curl}' / dA" K^, we observe that an expression analogous to r''(xt)-p-j +X ' 9 Xf T\ '^ e^'''"'''^*A the curly brackets in equation (98) must be zero. observation that the term inside the i^° [xt] {Cf) is and a local martingale, the term inside the curly (42), (93) to '^^[xt)^ no + dM;w inside the curly brackets of (98) (40), (41), gi- +xPj=o,s=iA dt [xt) 44 and (99) are both zero, obtain [K^xtY K° (100) {xt) dK° = dxt 7 d-n° dx. {oi{xt) A'o 0.° Using equation -jxta-) + 2 QP \ we obtain (36) axt i^t and By ^ = ^^^ ti^ ^*' "^ similar line of derivations allows " dn' 7 ni {a{xt) A'l 5(a;t) = - -yxtCJ 'n°{xt ni (xi ) (103) {r'^{xt) - Mi + o"^) Xt = - '— -^ H ' 2 non-contingent bonds when Xj B + q)-+X d^n' d/<' di, dx. where us to obtain a differential equation in state namely I, g{i = (102) using (36), (101) and (102) inside (100) allows us to obtain an ordinary differential equation involving^^ ^°'-^^' "d^' St --/-l^ -(+:«- /J.)) ^-/;. - K^{xt) + A'' - A'l fii I + ' I ' ^' 2 EM flp\. Proving that f^W \ 1 \ A'^ not choose to borrow in will follows similar steps to the proof of Proposition 1 . Data For the construction of 0^j, we used data from the World Bank's World Development Indicators Database, and from the International Monetary Fund's International Financial of the variables While and corresponding in the model there is Statistics (IFS). Table 8 presents a list sources. a single good, in the data the computation of more cumbersome ipt is since there are multiple goods, exchange rate fluctuations, intermediate goods, and so on. All our steps below are aimed at this, we isolating in ?/)( the component of external resources and income which transitory in nature. For let: ^' where is A^ and X in local currency; Y -r r/D ^ correspond to real nontradables and exports; and E and CF x-. V \^'^^> Trend' [Ar,r^"'^+[(^;^-0.5X,)] Px and Pm to export to the nominal exchange rate and capital flows. and import prices Real nontradables are constructed from: A^t = TT- iGDPt J {Px,t - 0.5PM,t)Xt,) N,t 'Note also that a{xt) can be expressed as a function of A'°, K^, is a function of A'°, A'\ ' ' ^, ^^C dxi ' (dxt) and Xf 45 ^-, f^ ^.; and xt since the interest rate Source Series Nominal GDP [GDP) World Development Indicators (quarterly CPI (P) and annual) IFS (quarterly and annual) Nominal Exports {PxX) World Development Indicators and IFS (local currency) (quarterly Nominal Imports {P^M) and annual) World Development Indicators and IFS and annual) (local currency) (quarterly Real Exports {X) World Development Indicators (local currency) (annual) Real Imports (M) World Development Indicators (local currency) (annual) Nominal Capital Flows (CF) IFS (dollars) (quarterly Nominal Exchange Rate {E) World Development Indicators and IFS (quarterly Net Factor Payments (NFP) and annual) and annual) IFS (annual) (dollars) Table 8: Data used in the construction of 46 ip. Mean Median Min Max Chile 20.6 21.5 15.4 22.7 Colombia 11.3 10 8.4 16.6 5.5 5.6 2.7 7.2 Indonesia 11.3 7.4 4.7 19.8 Malaysia 29.4 28.6 19.6 42.3 Thailand 21.3 20.8 14.1 28 Mexico Table GDP where Reserves for various countries as a percent of 9: is the country's GDP, and the term 0.5PM,t removes a proxy ( — p"^ - — 0.5A'tj is Pa'.j filter, for intermediate inputs mary, the denominator We effect. extending the series as We end-of-series bias in this procedure. applied the in export-production. The expression decompose between trends and cycles using a much as we could in order to reduce the effect of the to the log of the corresponding variable. In filter equation (104) measures the average (trend in for the years 1990-2003. the price of nontradables approximated by the local CPI, captures the terms of trade standard Hodrick-Prescott GDP level) of total sum- income and resources, while the numerator attempts to capture the cyclical component of external resources. We work with a sample of the following developing countries/emerging markets; donesia, Malaysia, Mexico, Thailand. We chose them from (2004) plus Malaysia. However, since their sample for the is dimension (wc used data from 1983 to 2003), we dropped with domestic macroeconomic which we did not have good issues, or did deflators. flows. we constructed We all list of countries in Calvo, Izquierdo quarterly series for the countries that either were closed economies Our marginal drop was Korea, GDP il'it for Korea ), solve this quarterly series for the years using a related scries approach with quarterly data on capital be equal to the annual figure we computed directly countries (dates are As a shown reference, table 9 problem we use a linear when we for our results remained essentially unchanged. using equation (104). Unfortunately, we lack quarterly data for some components of To and Mejia 1990s only, and we needed a longer time series not have complete data. restrict the average of the quarterly values to the 1990-2003 period. Colombia, In- However, when we re-estimated our model with Korea included (using only capital flows data divided by nominal Finally, the Chile, lack quarterly data. in parenthesis): Chile (1990), shows reserves for the six (or a quadratic) interpolation We some countries method in to obtain used interpolation methods for the following Colombia (1990-1995), and Malaj'sia (1990-1998). economies we study. 47 V't for Details on the econometric procedure of Sections 3.1, 3.2.1, 3.2.2. C^ Section 3.1: To estimate the process described Gibbs Sampler. The Gibbs Sampler states (See Kim and is in this section by now a standard methodology posterior distribution of all However, the same first SS), p{SS -^ of country > for we run i We allow we assume NSS) at time is differ across countries. that the transition probabilities into and out of a we assume that the common is joint probability of a Fx{SS = l\ip^;'^). As was drawn from the In the next step in the text, we use In the next step we take these whether each economy is in SS [NSS) first "f = {i/'t , ~ V'f We first; To Kim and with a = (3 = we draw an the convention that I's and O's as given. Kim and ipi " gamma Nelson (1999)) ''Pi 1 I i'^ef) sample of I's and O's c) By We i priors: a) a from these posterior Then we use well known NSS. to this information to beta prior [0, 1] b) for p{NSS — > we knew determine this in steps as described in Kim and "P^^^ o-^p, o-f|} , SS), p{SS an (improper) normal prior — NSS) » for ip^ that lead to posteriors that depend only on the data a truncated (improper) normal and an inverse (improper) Kim and shall denote Effectively this allows us to proceed as if Once again we do prior for [cr^f^) ^^ explained in independent of each other. ^ To do that corresponds to a Sudden Stop and or not at a given point in time. updating we use conjugate and an (improper) inverse <^^f^ o'ff a sudden stop at a specific point in time. in (artificial) which coincides with a uniform prior on 1 in the Nelson (1999) to determine a start with determining the posterior distribution of {ip^^^, ipf^ facilitate the , ''Pi or the second (SS) regime. as described in filter the posterior distributions of the parameters in $. Nelson (1999). jump across countries. to fix a set of initial parameters a standard Hamilton (1989,1990) type probabilities. for Nelson (1999) present by repeat this process for each country separately and obtain one sequence per country. this as (see Kim and and then determine the posterior probabilities that a particular realization } t to exploit knowledge is parameters of the model to all sequence of posterior probabilities that a given countrj' was We models involving hidden the others) to construct the joint all modify the basic model that across countries. Moreover, step of the procedure time (fixing at a a simultaneous transition into a sudden stop p{NSS — tpit We in order to obtain precise estimates VIX and The parameters. all the countries into a single sample. sudden stop are in estimating Nelson (1999) for an introductory treatment). The basic idea about the conditional distribution of one parameter pooling we apply a Bayesian methodology, by using a Nelson (1999). results in Bayesian Finally, we assume that gamma all prior priors are statistics the posterior distributions are in the same class as these conjugate priors and there are simple closed form expressions for the parameters of the posterior distributions. observations for all The updating oi p{NSS the countries. So, for each country 48 -^ 55), and p{SS -^ we count the number NSS) is done by pooling the of transitions into and out of sudden stops, and the episodes for number total normal times and of periods in ^ ~ SS) beta + 1 ( ^ap, \ sudden We stop. a normal year we add all number the is We stops. then add all the countries and find the posterior distributions as follows all piNSS where af^ sudden in marked of observations + 1 ^ ^NSS 1 I normal years that are followed by a year marked as as shall refer to this as a transition to a sudden stop. Conversely, followed by another normal 3'ear (NSS). This count is them up same probabilities to be the precise estimates. number which into a single for all countries, counts the times that done country by country, but then used in the updating process. is we is af ^^ By restricting the transition exploit the panel dimension of the data in order to obtain Similarly, the posterior for the other parameter in the transition ^ + matrix given by the is following formula ^ NSS) p(SS where b^^^ number the is of observations ~ beta marked + 1 ( as ^^^^^ ^ E ^f^ ) sudden stops that are followed by a year marked eis "normal". Conversely, bf^ represents the other case, namely when a transition out of a sudden stop does We not occur. specific record the random draws - parameters {ip^^^, ipf^ NSS) p{SS -^ draw of the paths of I's of a) the paths of I's tp^^^, a^f^, o-f|} and Then we repeat the above procedure . and O's for the posterior distribution of these c) and each country, b) the country the "pooled estimates" of several times each country and the parameters. random draws O's for and By p{NSS -* SS),and we record the new at each time properties of the Gibbs sampler, coincides in law with the posterior (joint) distribution of the parameters. all Section 3.2.2: "jump" in the cutoff value e. By VIX, using the cutoff value e for the i.e. months when the VIX we determine the months residuals of the estimated To determine a distribution oip{SS\J) we proceed AR(1) process in which we observed a of the as follows. First, for VIX exceeded the each country we draw paths from the posterior distribution of the states (NSS,SS)^, as provided by the Gibbs Sampler. Given the model, the only relevant observations for the conditional probability are the ones is in NSS or has just transitioned to a SS. model, we discard of each SS. there is a all Then we jump in the VIX all is in also determine all those times where the states switch from either in that quarter or the quarter before. the times when n}v55 as the numbers of observations i SS, except the one that marks the beginning allow for the possibility of delayed data reporting etc. Let this we the country Hence, in accordance with the data generating process of the the quarters in which the country look at when there was a jump when country i in the is 49 NSS We to this ^''^^ss—ss number be cither in A'^55 or just simultaneously allow this short number be given by VIX. Let 55 and moved ny J- window to Similarly, Finally, define to a SS. We repeat procedure this Tiggj, for and n'-j, each country separately. After completing the above procedure for n^f^gg all countries In accordance with the Bayesian methodology that across countries. we sum we have been using throughout, we use a beta distribution with uniform priors to obtain the posterior distributions p{J) - beta (1 + E,:nj,, 1 p {SS\ J) - beta (1 + T-m^Nss-^ss.j, + i:in%ss) 1 + In a nutshell, for each iteration of the Gibbs sampler, we determine the VIX coincided with a transition into a SS for each country. imposes the constraint that all VIX. The benefit into a SS and jumps U Estimation of Risk Premia To estimate the risk in the premium same countries have the we A* is that number Then we pool if P"" {VIX, K, t) is ^ We ^^ '^^ ' . To estimate used quotes^^ from the Letting k = ^7^, we is ^\l''^'^''^^ VIX, estimates^''. Huang K (1985). As explained we on VIX fit first call in the and r days to expiration, when then the (arbitrage free) price of a payoff such as (50) a non-parametric function P'-^ [VIX, K, is simply r) to the data. options since the beginning of this market in early 2006. = 0.0306isr-1.115/c-|-0.34/c2-0.0535/c^ -0.0002r/C - 0.0304TA: ... random draws + 0.0243rfc2 - + 0.0024fc4 0.008tA:^ + (105) O.OOlrfc* + .. additional controls a robustness check we also computed recording the between transitions then estimated the following regression: ^ "As in the followed an approach similar to (Ait-Sahalia and Lo 2000), which itself this quantity CBOE when a jump across countries. This effectively we can obtain more accurate the price of a call option with strike price the underlying value of the index of times joint probability distributions builds on the ideas of (Breeden and Litzenberger 1978) and Duffie and text, S»"!/) p[J ) and p{SS\ J) without imposing any prior, i.e. by just of: ^in'j p^^> at each iteration k of the distribution. (SS\ J) Gibbs Sampler and computing the empirical mean of the corresponding stationary The two approaches delivered very similar results, suggesting that our results are not influenced by the assumption of a uniform prior. ^^In particular, we used the midpoint between bid and ask 50 where additional controls included a l/VIX, a constant, VIX, To avoid micro-structural problems t,t'^. with very short maturities,^^ we focused attention on values of r e [30,90]. Furthermore, since we are interested in isolating the price of risk for upward jumps in the VIX we focused on was estimated on 2476 date-price combinations. The resulting E? of the regression was that the regression specification accounts well for the observed variation in differentiated the estimated P^ {VIX,K,t) in equation (105) on short-dated options but with expiry dates and at the average value of regression (105). To be VIX first term in over an interval r- k is + 0.17^ {\og{VIX) - set so that the probability of .01. The second term accounts 2.88 (based on data since 1986). The reversion in monthly data. Plugging in these numbers, As was shown 0.105. in the text, this {VIX)) having a jump in the the distance between the average (log) underlying price in the mean \og{VIX) = log we used call for Since . is '^^ ^^^'"^''"^ at r days a large fraction of the out of the money of bid This is and the next highest increment why we of ask. 40, we set residuals-assuming no reversion in (VIX) = VIX jump by comparing 2.54 to its long run speed of mean that the value of a claim such as (50) at very short intervals. call = . options log number imphes that A*(r/360) = problem then an interior value to the is 0.105. Solving for A* gives 0.94. ^^The data are quoted at prices with discrete increments. For the out of the money consider, this discreteness presents a We we focus attention coefficient 0.17 accounts for the we obtain regression which suggests K in the text VIX mean 0.83, The option prices. sure that r consistent with the definition of "jumps" that approximately is make 1. call we evaluated — larger than 30 days, over the sample in order to r The with respect to > /c call As maturity nears options that we to less than thirty options start to converge to the lowest possible increment This discreteness leave out the very short maturities. 51 is less of a. problem for longer maturities. References Ait-Sahalia, Y., and A. Lo (2000): "Nonparametric Risk Management and Implied Risk Aversion," Journal of Econom.etrics, 94, 9-51. AiYAGARi, R. (1994): "Uninsured Idiosyncratic Risk and Aggregate Saving," Quarterly Journal of Eco- S. nomics, 109(3), 659-684. Arellano, C, and Economies: An Mendoza G. E. Open "Credit Frictions and 'Sudden Stops' in Small (2002): Equilibrium Business Cycle Framework for Emerging Markets Crises," NBER Working Papers: 8880. BONDARENKO, 0. "Market Price of Variance Risk and Performance of Hedge Funds," mimeo. (2004): University of Illinois-Chicago. Breeden, and R. H. Litzenberger D., (1978): "Prices od state-contingent claims implicit in option prices," Journal of Business, 51, 621-651. Broner, G. Lorenzoni, and F., Term?," mimeo. Caballero, R. in a Caballero, R. work," and (2003): "Why Do Emerging Economies Borrow Short Department of Economics. Emerging Market of J., S. (2001): "International and Domestic Collateral Constraints Monetary Economics, 48(3), 513-548. Crises," Journal of Panaceas NBER Working (2005): Sciimuckler and a. Krisiinamurthy J., Model MIT S. "Contingent Reserves Management: (2004): An Applied Frame- Papers: 10786. "A Quantitative Model of Sudden Stops and External Liquidity Management," NBER Working Paper 11293. Calvo, G. a. (1998): "Capital Flows and Capital-Market Crises: The Simple Economics Sudden Stops," of Journal of Applied Economics, 1(1), 35-54. Calvo, G. A., A. Izquierdo, and L.-F. Mejia of Balance-Sheet Effects," NBER Working Calvo, G. A., and C. M. Reinhart (1999): (2004): "On the Empirics of Sudden Stops: The Relevance Papers: 10520. "Capital Flow Reversals, the Exchange Rate Debate, and Dollarization," Finance and Development, 36(3), 13-15. Campbell, J. Y., and J. of Aggregate Stock DuFFiE, of J. D., and Froot, K. Cochrane (1999): "By Force Market Behavior," Journal of C. FU Few Long-lived H. Huang of Habit: Political (1985): "Implementing A Consumption-Based Explanation Economy, 107(2), 205-251. Arrow-Debreu Equilibria by Continuous Trading Securities," Econometrica, 53(6), 1337-1356. a., D. S. Scharfstein, and J. C. Stein (1993): "Risk Management: Coordinating Corporate Investment and Financing Policies," Journal of Finance, 48(5), 1629-1658. 52 Hamilton, D. (1989): "A J. New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle," Econometnca, 57(2), 357-384. (1990): "Analysis of Time Series Subject to Changes in Regime," Journal of Econometrics, 45(1-2), 39-70. Heller, H. R. "Wealth and International Reserves," Review of Economics and (1970): Statistics, 52(2), 212-214. Jermann, U. "Asset Pricing in Production Economies," Journal of Monetary Economics, pp. J. (1998): 257-275. Karatzas, I., AND S. E. Shreve Browman motion and (1991): Texts in Mathematics. Springer- Verlag, New stochastic calculus, vol. 113 of Graduate York, second edn. Methods of mathematical finance, Applications of Mathematics. Springer- Verlag, (1998): Kim, C.-J., and C. R. Nelson (1999): State-space models with regime switching: sampling approaches with applications. KlYOTAKI, N., AND J. MoORE MIT Press, and Gibbs- Cambridge and London. (1997): "Credit Cycles," Journal of Political Kletzer, K., D. M. Newbery, and B. D. Wright Classical New York. (1992); Economy, 105(2), 211-248. "Smoothing Primary Exporters' Price Risks: Bonds, Futures, Options and Insurance," Oxford Economic Papers, 44(4), 641-671. Kletzer, K. M., and B. D. Wright Economic Review, Lee, "Sovereign Debt as Intertemporal Barter," Am.erican 90(3), 621-639. "Insurance Value of International Reserves," mimeo. IMF. J. (2004): Nelson, C. R. (2000): (2006): "The Beveridge-Nelson Decomposition in Retrospect and Prospect," mimeo. Federal Reserve of Atlanta. Oksendal, B. (1998): Stochastic differential equations: An introduction with applications, Universitext. Springer- Verlag, Berlin. Pan, J., AND K. CDS J. Singleton (2006): Spreads," Stanford and 737/ "Default and Recovery Implicit in the MIT Working paper. 53 Term Structure of Sovereign