MIT LIBRARIES INIIMi Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/hedgingsuddenstoOOcaba 9 ql^ .1 B31 M415 c).c3 Massachusetts Institute of Technology Department of Economics Working Paper Series HEDGING SUDDEN STOPS & PRECAUTIONARY RECESSIONS: A QUANTITATIVE FRAMEWORK Ricardo J. Caballero Stavros Panageas Working Paper 03-1 May 26, 2003 Room E52-251 50 Memorial Drive Cambridge, This MA 02142 paper can be downloaded without charge from the Social Science Research Network Paper Collection at http://ssrn.com/abstract^XXXXXXX JUN 5 2003 jJBRARJES Hedging Sudden Stops and Precautionary Recessions: A Quantitative Framework nicardo MIT J. and Sta\TOS Panageas Caballero NBER MIT May -' Even which is Abstract managed emerging market economies well financial. are exposed to significant external risk, the bulk of At a moment's notice, these economies that have supported the preceding boom. Even often leads to costly precautionary measures cycle of an 26, 2003* economy that on average needs to if and may be required to reverse the capital inflows such a reversal does not take place, recessions. In this paper, borrow but we its anticipation characterize the business faces stochastic financial constraints. We focus on the optimal financial policy of such an economy under different imperfections and degrees of crowding hedging opportunities. out in its and realistic is simple enough to be analytically tractable but flexible enough to provide quantitative guidance. JEL Codes: Keywords: E2, E3, F3, F4, GO, CI. Capital flows, sudden stops, financial constraints, recessions, hedging, insurance, signals, contingent credit *We The model lines, Eisymmetric information. are grateful to Victor Chcniozliukov, Arvind Krishnamurthy and the participants at the macroeconomics and econo- metric lunches at for financial MIT support. for their comments, and Andrei Levchenko for excellent research assistance. Caballero thanks the NSF Introduction 1 Most emerging economies need tunately, even well moment's borrow from abroad to as they catch managed emerging economies are subject up with the developed world. Unfor- to the "sudden stop" of capital inflows. At a and with only limited consideration of initial external debt and conditions, these economies notice, rnay be required to reverse the capital inflows that supported the preceding boom. The deep contractions triggered by this sudden tightening of external financial constraints have great costs for these economies. Not surprisingly, local policymakers struggle to prevent such crises. During the cycle, the anti-crisis mechanism entails deep "precautionary recessions:" tight monetary and the first sign of symptoms of a potential external crunch. build up of large stabihzation funds and international Over the medium reserves, fiscal contractions at run, the tools of choice are the and taxation of capital inflows. All of tliese precautionary mechanisms are extremely costly. The case of Chile illustrates the point well. of copper, its investors, and Chile's business cycle main export good: so much so that this price is highly correlated with the price has become a signal to foreign and domestic pohcymakers ahke, of aggregate Chilean conditions. As a to external reactions to this signal, the decline in Chilean economic activity sharply is many result of the when the and This contrasts a developed economy not exposed to the possibility of a sudden stop, where in Australia, similar terrns-of-trade shocks are fully absorbed by the current account, with almost no impact on domestic and consumption.' In this paper we do not attempt so differently, or even Instead, internal price of copper falls times larger than the annuity v'alue of the income effect of the price dechne. with the scenario activity many we why to explain why international financial markets treat Chile and Australia the signal for the sudden stops should be so correlated with the price of copper. take the "sudden stops" feature as a description of the environment and characterize the optimal hedging strategies under different assumptions about imperfections the precautionary business cycle that arises The main technical contribution of the paper analytical characterization, but model has two when hedging is also flexible central features: First is and is in hedging markets. We also characterize opportunities are very limited.^ a model that reahstic enough the sudden stop, which we is stjdized enough to allow extensive to generate quantitative guidance. The characterize as a probabilistic event that, once triggered, requires the countrj' to reduce the pace of external borrowing significantly. Second is a signal. 'See , -We focus on the aggregate financial problem vis-a-vis the rest of the world, in an environment where domestic policy e.g., Caballero (2001). managed optimally and compound decentralization is we deal with in this paper by exacerbating the country's exposure to sudden stops. See, Such failures e.g., Caballero and Krishnamurthy (2001, 2003) and Tirole (2002) for articles dealing with decentralization problems. the problems literature on government's excesses is is not a source of problems. Needless to say, these assumptions seldom hold in practice. very extensive. See, e.g., Burnside et al (2003) for a recent incarnation. The Sudden stops have some element of predictability there is a perfect signal is We start by studying a simple environment where the high yield spread, or the price of an important commodity (e.g., that triggers a sudden stop once where the threshold to them. it crosses a well defined threshold. We for the country) then study the more reahstic case may blurred and a sudden stop, while increasingly likely as the signal deteriorates, occur at any time. Within that is, in this model, we develop two substantive themes. First we characterize precautionary recessions: the absence of perfect hedging, the business cycle of the economy follows the signal even sudden stop actually takes place. This is because as the likelihood of a sudden stop Consumption into a precautionary recession. is if no the country goes rises, cut to reduce the extent of the adjustment required if a sudden stop does take place. The second and main theme reduces the extent of the assets and for incurring is crisis in aggregate hedging strategies. An adequate hedging strategy not only the case of a sudden stop, but also reduces the need for hoarding scarce sharp precautionary recessions as the signal deteriorates. We study two polar- opposite types of generic hedging instruments or strategies, as well as their intermediate cases. At one end, the hedging contract relaxes the sudden stop constraint one-for-one. That to relax the constraint. The other extreme is each dollar of hedge can be used if the resources obtained from the motivated by crowding out: hedge facihtate the withdrawal of other lenders, then the main effect of debt in bad states of the world but not to provide fresh resources. excellent substitute for precautionary recessions, but The paper concludes with an is, it illustration of our By the hedge its to reduce the country's is nature, this type of hedging cannot remove the sudden stop main is an entirely. We results for the case of Chile. estimate the probability of a sudden stop as a function of the price of copper and calibrate the parameters needed to obtain sharp consumption drops such as those experienced by Chile in the recent contraction of 1998/99. We then describe different hedging strategies and their impact on the volatility and We discuss credit hnes problems. For example, and its copper. and their indexation to the signal as we argue that le\'els of consumption. one way of reducing asymmetric information Chile could virtually eliminate sudden stops, precautionary recessions, large accumulation of precautionary assets, with a credit line that rises nonlinearly with the price of The cost of this line, if fairly priced, should be around 1-2 percent of GDP. This compared with the costs of the precautionary measures currently undertaken, including is very little when large accumulation of reserves, limited short-term borrowing, and precautionary recessions.^ Currentl)', although futures markets exist ations, the size of the financial problem is for much much of the income-flow effects of larger than that. The markets commodity price fluctu- required for these strategies 'in his Nobel lecture, Robert Merton (1998) highlights the enormous savings than can be obtained by designing adequate derivatives and other contracting technologies to deal with stand to gain the most. Our findings in this paper fully risk support management. He his views. also argues that emerging-market economies do not exist, at least in the magnitude required. In titatively the usefulness of these this sense, our framework also serves to highlight quan- markets and allows us to begin gauging the potential size of the markets to * be developed. we In section 2 has been triggered describe the environment and characterize the optimization problem once a sudden stop — this provides the boundary conditions for the "precautioning phase." In section 3 we The main study the phase that precedes a sudden stop when the country self-insures. is goal of this section to characterize precautionary recessions. Section 4 describes aggregate hedging strategies under different imperfections and the degrees of crowding-out in these markets. an application to the case of Chile. likelihood of a sudden stop and Section 5 illustrates our results through Along the way, we outline an econometric approach to gauging the its correlation with an imderlying signal. Section 6 concludes and is followed by an extensive technical appendix. The Environment and the Sudden Stop Value Function 2 Intertemporal smoothing implies that, during the catch up process, emerging economies typically experience substantial needs for borrowing from abroad. For a variety of reasons that dependence on borrowing The sudden a source of fragility. is we do not model here, this tightening of financial constraints, or the mere anticipation of such an event, generates large drops in consumption. In this section, we formahze such an environment and characterize the optimization problem once a sudden stop has occurred. The next sections complete the description by characterizing the phase that precedes the The Environment 2.1 Endowment and 2.1.1 We assume that {y{i)A > 0), is Preferences the endowment grows at some constant rate, g, during dvU) yO y[t) aspects of this process are to be highlighted. to characterize, •"See, e.g., < i < oo. Thus, the income process described by: -^^gdt, Two crisis. sudden stops are significantly 5 =yo. First, it is > 0. deterministic. In the economies we wish more important than endowment shocks as Krugnian (1988), Froot, Scharfstein. Stein (1989), Haldane (1999), Caballero (2001), commodity indexation of emerging markets debt. The contribution of this for triggers of articles paper relative to that literature quantitative framework and to link the hedging need not to commodities per se, but simply as a signal of financial constraints. is deep advocating to offer a much costlier . contractions.^ and hence > this perspective, the their potential to trigger any collateral since g Prom effects there 0, Reflecting is main importance We a sudden stop. an incentive and riskless interest rate c^ represent date t dimension and group and i r~g it = ^o < 0. times: all ^ > X(0) > 0, exceeds the rate of growth of the endowment, r > g. consmnption and "excess" consumption, respectively, with: c\ The this net debtor position, the country starts with financial wealth, Xt Ct model along their "collateral" effect, country to borrow early on. for the However, the country's total wealth must be positive at Let simplify the is contained in endowment shocks with the signal process (to be described below). Second, its initial where r denotes the endowment shocks of = ~ Ct < K.yt, K < 1 representative consumer maximizes: POO c:)e-*('-*' ds (1) with C*l-7 u(cn The parameters assume r = 6 and 7 is T— 1-7 are the discount rate 6 throughout. with a habit level that = The and 6>0,7>0. , risk aversion coefficient, respectively. For simplicity, we functional form of the utihty function captures an external habit formation, increasing at the rate of the country's growth rate. This economies that exhibit strong growth, since future generations are likely to is a natural assumption in have a higher standard of living than the ciirrent ones.® A frictionless It is benchmark instructive to pause and study the solution of problem (1) subject to a standard intertemporal budget constraint (and absent a sudden stop constraint): = dXt {rXt-cl+y;)dt Xo "Of course, sudden stops reduce growth as well, given but we capture these effects directly through the decline in consumption. In this sense, y can be thought of as potential, rather than actual, output. Accordingly, this type of utility function allows us to impose indebtedness of the country at any point in time a simple reduced form) a barrier on the of: r and on saving, (in - g since: yt- Ct <(l - K)yt t > 0. amount of • = with y* By — (I K)y. standard methods can be shown that the solution to this deterministic problem it c* and the = "total resources" of the country r (Xo + aU for ^:^) f > is given by: 0, remain constant tliroughout: — Xo+ -y r-g ^° (2) -y r-g I ' so that lira —= 1 and, accordingly: lim '-°° Moreover, for any level of the — l-K = r-g' yt debt-to-income ratio below its limit value, the ratio ^ decreases monotonVt ically —. to This case serves as a 2.1.2 There frictionless benchmark in what follows. Signal is a pubhcly observable signal, S(, correlated with the sudden stop and, for simplicity, uncorrelated with world endowments. This signal follows a diffusion process: dst In our basic model, threshold, s, We In our second model, the signal only its likelihood that + adBi. crisis is triggered is is = mi{t e (0,00) S( : time the signal reaches a <s}. (smoothly) influenced by the signal A( We first imperfect. In this case, a sudden stop can happen at any point in time. stochastic time, t, depends on the reahzation of a stochastic 2.1.3 the associate with this event the stochastic time, t: T It is fJ-dt a perfect signal and a St is from above. = jump 5j. In this case, the specification of the process, the intensity of which is given by =exp(Qo -a\St)- Sudden Stops place no hinits on the country's borrowing ability faces a "sudden stop." We up to the stochastic time t. do not model the informational or contractual factors At this time, the country behind this constraint, or the complex bargaining and restructuring process that follows once the sudden stop to produce a quantitative assessment of the hedging aspects of the problem, goal is and fairly robust (across models) constraint. the net present value of is its assume that within It is T a realistic as a temporary we place the constraint on: ~g > e^'^ periods. Formally: obvious that this constraint will be always binding because as ( Xt + -^ remains constant at ) that this constraint can be expressed as a constraint on the + for at time t, financial markets require the country to increase its total resources by ^ the unconstrained benchmark T we look particular, since the "natural" aggregator of a total resources, r We triggered. Since oiu- we simply model the sudden stop For this, and severe constraint on the rate of external borrowing. In country's total wealth is r, as a function of the debt/gdp ratio at time ^ = e^^ r. + To 0(1 - we showed times. It all maximum is in a previous subsection, at then straightforward to show allowable amount of debt/gdp at time see this, redefine e-<'-»'^) (4) and observe that the constraint becomes: X.,r > X. k^ + 0(1 - e-(^-)-)l L ^Ml-^-'"-^^") r - g + J (5) or X.+r ^ X^ " Vt [e^^ + levels of imply higher that for the special case = - e'C-g)^) 0(1 ^ e9T{r-g) esT y;+r Higher - e-(-^)^)] 0(1 levels of (^ and thus make the constraint this constraint literally reduces to the grow any further between r and t tighter. It is also trivial to verify requirement that debt/gdp cannot -\-T-J Xt+T ^ Xt Vr+T " Vr Finally, it is interesting to note that one can think of surpluses the country has to accumulate over the next (5) as a constraint T periods. To on the minimum balance-of- trade see this, start with the intertemporal budget constraint: Xr+ I e-^^-'-^^y; dt= e-'-f'-'^c* dt I + e""^ X. or: r^+^ Xr - e-'-^X^+T = j 4> = the debt/gdp ratio is Even for t.lie case the constraint we consider , ^ !;*fl-e-('-s'^^ e-"<*-">< dt r is -g binding at the time of the sudden stop. This always growing for an unconstrained country as we showed in section 2.1.1. is due to the fact that . and replace (5) into to obtain: it, rT + T j-r+T or, finally: fT + T rT + T - e-''"-^> (y; / cl) dt > X^lle-f'-s)^ + This gives us the constraint in 9!.e-'-^(l - e-<'-5>^)) - 1) + <t>e-'''^ / e-'"*'-^'!/; dt terms of the balance-of-trade surpluses required from a country that faced with a sudden stop. These required surpluses rise with the level of debt (recall that ,Yt endowment the < 0), (p, is and of the country. The Optimization Problem 2.2 Let us assume for indexed to s or t. now that the only financial instrument f Jo riskless debt, so there are maximizes the expected In this case, the country V(X,,y;)=maxE is utility no hedging instruments of a representative consumer: < 7 1 s.t. = dXt [rXi — cl + yl)dt X^+T > Xr+T = Xr -^ = [e^^ + - e-f-'"^)] + ^ylO^_£ll:^^ 0(1 < < e'-^ gdt Vt dst t = hm — iidt + cfdBt X(;)e-''' =0, a.s. 'OO and either Perfect signal: r = inf{i 6 (0,oo) : Imperfect Signal: T = inf{? e (0,c») : St / < or s] \[st)dt<z}, At = e"°~°'*', z ~exp(l) Jo independent of the standard Brownian Ft —Filtration. where z 2.3 The Sudden Stop Value Function and Amplification We is solve the optimization problem in three steps. period, then for the sudden-stop period, we present the first the sudden stop, two, and V^^[Xt, problems before the crisis J/t). and trivial, steps. Starting backwards, we first solve for the post-crisis finally for the period preceding the sudden stop. In this .section The goal to find the value function at the time of of these is which then can be used to find the solution of the optimization and hedging takes place. 8 Post Sudden Stop: 2.3.1 Since we made the problem for i > r t>T + T simplifying assumption that the country suffers only one financial crisis, the maximization +T is simply: y(X.+r,J/;+T)=max/ < Jt+t ^^ 1 -7 e-(*-(-+^)) di (6) s.t. Jt+t Jt+t This problem has the trivial solution: c* = rW^ T + T <t<oo (8) ^^. (9) with VV'|-X.+r + r -g This constant can be interpreted as the "excess" wealth at date t which is 2.3.2 needed to cover the reservation-consumption Sudden Stop: t for the (that is, the wealth in excess of that level kj/j). <t<T + T With the continuation value problem +T function, V{Xr+Tiy*+T)^ from above, we can now write the maximization sudden stop phase: y^^(X,, y;) = max T^ ^e--(t-) dt + e-'^V{X^+T,y;+T) S.t. rT+T j-T Cj*e-^(*-^) / dt= +T Xr+ y;e-''('-^> dt - e"''^ Xr+T Xr+T > Xr+TSince the country allows, the is (10) growing and wishes to expand sudden stop constraint, (10), is its consumption at a faster rate than the constraint always binding (see the appendix for a formal proof): Xr+T = Xt+tGiven this result, the optimization subject to a final wealth condition. A q* problem is straightforward. (11) It is a deterministic consumption problem few steps of algebra show that: = ^_[_rT ^Vl, T<t<T + T (12) with Wi = Ar It is r are now ready crisis in Ar+T- e - g Xt+t easy to see from these expressions that as consumption during the We H The country has c^ falls. rises, to cut back order to satisfy a tighter sudden-stop constraint. to determine the sudden-stop value function: 1-7 Jr 1— 7 1-7 Jr+T r \^ 1~7 J (14) It is easy to verify that our careful choice of the constraint pays off at this stage. The value function simphfies to: = where V{XT,y*) denotes the vahie fimction KV(Xr.y;) in the absence of (15) a sudden-stop constraint and ii" is a constant given by: K= That {1- is, e-'"^)'>(l - e-'"^,^(l - e-('-5)^) - e-''"-*)^)'-^ the dimension of the state space effectively Moreover, up to the constant K , we have arrived is + e"*"^ (<^(1 - + e-'^-^)^) e^^)^"^ reduced from two, (Xr.yr), to one, (Xr at a value function that is (16) . + -^f-z )• identical to the one in the problem without sudden stops. This problem corresponds to the trivial case of an infinite horizon consumption-savings problem under certainty. Note that for 7 > 1, which we assume throughout, While the particular simplicity of our formulae more general. The model also can is K> 1 and increases with 0.^ due to styhzed assumptions, the basic message is be vmderstood as an approximation to a potentially more complicated specification of constraints that result in higher marginal disutihty of debt in the event of a crisis. 'Observe that is for 7 > 1 the function -^-^ — negative for all Z > 0. Thus lower than the continuation value function for the unconstrained problem. of the constraint implies that ha\'ing a higher its is Xt K > 1 The reason we need 7 > does not relax the constraint one demands by the same amount during the sudden stop. When 7 for the crisis. 10 is reflects that the constrained value function below one, for 1 is that the flow aspect one, because the lender does not reduce this effect is strong enough to discoXirage saving — Precautioriciry Recessions 3 Let us now address our main concern and begin to characterize the country's optimization problem prior to the sudden-stop phase. s or T, the country's only borrowing before it we have eissumed Since mechanism We takes place. that there are no hedging instruments contingent on reducing the cost of a sudden stop for show that when some elements the signal of a sudden stop deteriorates, the country a "precautionary recession:" sudden stop. Such behavior that is is, a sharp reduction in of predictability in them. falls into what can be First, we study signal hits a stops stiU are 3.1 minimum is complex for a case where there threshold, more hkely as the Perfect Signal: In labelled as consumption to hmit the cost of the potential many risk. reasons, one of the most important being the uncertainty that surrounds the factors that trigger such crises. In order to isolate the main issues, steps. amount widely observed in emerging market economies, where private decisions and macroeconomic policy tighten on the face of external In practice, such a problem to cut consumption and in addition to a precautionary savings result, the of self-insurance varies over time, because sudden stops have particular, is s, is a perfect (stochastic) signal: for the first time. A we proceed in two sudden stop occurs when this Second, we add (local) uncertainty: while sudden signal deteriorates, they can occur at any time. The Threshold Model In the threshold model, the dynamic programming problem *l rT V{Xt,st,yt) = max£: is: t (17) 1-7 Jt where = T and the evolution of {Xt,St,yt) The boundaries is inf{(i : s < s) Ago} (18) given by: dXt = {rXt-c*+y;)dt (19) — = 9df (20) dst = /.idt of the value function, V{Xt, + adBf Sj, ?/<), can be found readily. (21) On one end, we showed in the previous section that: ViX.,,,y;) where K > 1 measures the intensity of the = crisis K{l}l'i^^0— and is 11 given in (16). (22) ' On we show the other end, the appendix that in goes to St eis infinity, the value function converges to the value function of the deterministic problem with no sudden stops: £n^y(X.,.,y;) = That ,_;^ (23) the value function becomes independent of the signal as the crisis event becomes less and less is, The ^ (;j V value function satisfies the following Hamilton-Jacobi-Bellman (henceforth f.*i-T = max <! -^ - Vxcl I - rV + Vx {rX + \ yl) + Vy-yy', + Km + HJB) likely. equation: 1 ^Vsst' (24) subject to the boundary conditions (22) and (23). In the appendix, we show that the solution has the form: 1-7 V{XuSt,y:) = ^) ^^ + (-^- a{str^ for some twice The difTerentiable function a{st). Mm latter satisfies the = a{s) s—*oo and solves mization an ordinary in (24) differential with respect to = (24). Cj exists function a{s) is and is unique. Ma, We This setup allows for Given the value function, (27) = {Vx)-"^ (28) . -^^ — ^_'' . to (26) and (27). We show in (29) the appendix that the solution to s. As the signal deteriorates toward s, a[st) 12 The in likely. feedback form. from (28) that the optimal consumption policy takes the form: .-Ml. 1. this function rises rapidly, wealth (and marginal wealth) as the sudden stop becomes more follows yield: l^K)^ ^2 ( ,.. =0 + -CTM(7-l)^-^+a,J an exphcit characterization of the optimal consumption policy it These steps plot this function (multiplied by r) in the left panel of Figure decreasing with respect to reflecting the increasing \'alue of (-\ K^'-^ Third, di\'ide the resulting expression by which can be solved numerically subject ODE (26) / to get: l-ra + this r equation that can be obtained in three steps. First, carry out the maxi- c* Second, substitute this into boundary conditions: (-] V a{s) (25) -7 1 The effect of limit. Conversely, when the mechanism: as the and hence exposure to reduce external debt Ito's consumption signal worsens, precisely the precautionary recession Applying As the signal can be seen clearly in this expression. Lemma to the right to the hand falls for crisis s rises, a{s) falls any given level of toward its frictionless income and debt. This becomes imminent, consumption falls in is order sudden stop. we side of (30), obtain the country's (excess) consumption process: ^ = 0^ + () dt-^adBt. 1 2 Cl \ a (31) a J This process reinforces the above conclusion and shows that consumption has a nontrivial diffusion term. Without hedging, consumption "picks up" the Up to the functions ^, which can be (— ^cr) savings, whereas the second term signal news. Positive is news about the and fulfills the function of a hedging strategy. evaluated numerically, the rest of the terms in this expression The have a straightforward interpretation. volatility first is term is positive also positive signal increases and captures the and captures the effect of precautionary sensitivity of consumption consumption whereas negative news decreases it. to This the outcome of the country's attempt to accumulate resources as the sudden stop becomes imminent. As can be seen in the right panel of Figure deteriorates. 3.2 That is, 1, which plots ^, this effect becomes more intense as the the sensitivity of consumption to news about the signal increases during downturns. Imperfect Signal: The jump model Countries do not have a perfect signal for their sudden stops. Variables such as terms-of-trade and the conditions of international financial markets raise the probability of a sudden stop, but as in the threshold model. In this section we capture sudden stop a probabihstic function of the underlying The setup and optimization problem T, in (18). Now the sudden stop is this Thus, we replace (18) = it is never as stark dimension of reality by making the trigger of a signal. are exactly as in (17)-(21), with the exception of the stopping time, a Poisson Jump>-Event with intensity: A(5t)=e°''-"»^S T signal a, >0. for: inf {i e (0, go) : / \{st)dt < z}, A(s,) = e""""'^' , 2 ~ exp(l). Jo Once the sudden stop takes value function still place, events unfold exactly as in the threshold model. takes the form: rj 13 1-7 In particular, the Function a Function ra(s) /a 1.25 -0.15 -0,2 -0.25 105 - 2 2 4 Level o( the state Variable Figure 1: Functions ra{s) s=-0.6,so =0,r (left = 0.09,T= of the state variable is 1, panel) and = 4 Level of the state Variable ^ 0.35e''^,a (right panel). = 0.23,7 set to 0. 14 = 7,a The parameters used = 0.03, k = 0.8, yo = example are: -O.S.The drift for this I.Xq = The only difference In this case the that is HJB V^^ now can be reached from any equation for the value function which A(s) s for Vxc;} - rV With still rather than from just 0, s. is: r*l-T - inax{^^ > + Vx{rX + j/*) + Vy-y'g + essentially identical steps as in the previous subsection, V^^i + 1 ^-a^V^s + [V^^ X{st) we can show that our stylized V framework yields a simple solution of the form: 1-7 V= (^' bistY + ^) 1-7 Precautionary Savings 3.2.1 Let us pause and focus on the case where sudden stops are totally unpredictable, aj 6(s) is no longer a function of the algebraic equation for signal s, replacing the function V in the HJB = 0. Since in this case equation yields a simple b: l-rb+-\ 0. 7 where RIt is now straightforward to obtain the (excess) consumption function from the envelope theorem: b Applying simple differentiation to this expression, we obtain the country's (excess) consumption process: dc; -A dt 1 7 Relative to the threshold model, there reflects sudden stop correlated with the signal self-insurance new drift term in the excess consumption process. This term is s, tsiking place in the next instant. But, because this probability is not there are no precautionary recessions. In this case, the pattern of saving for not a soruce of business cycles. Precautionary Recessions 3.2.2 Let us in a the additional precautionary savings attributable to the local uncertainty introduced by the strictly positive probabihty of a V is now the > return to the general case, where ai HJB equation yields the following 1 — r6 + bsi-i + 0. After a few simplifications, substituting the function ODE: (7-1)^+6. 15 0, 2 7 (32) which differs from (29) only The boundary probabiUty of a in the last term. conditions are also different. First, crisis. when Second, as conditions worsen, there goes to infinity there s is is still a strictly positive no equivalent to the tlu-eshold * where a crisis happens with probability one. Let b* bt b* and 6, are = = lim b{s). lim b{s) given by: = b* r = b, Again, it is K'/-'-. straightforward to obtain the (excess) consumption function from the envelope theorem: (^"^) ,. Finally, applying Ito's lemma ,33, hand side of to the right we obtain (33), the country's (excess) consumption process:^ 7 dc* + 1 fabsY 1 1 b 7 b' . (It b J This case integrates the insights of the pre\'ious models. tributable to local uncertainty, but the section, ^ is amount - jadBt There is ongoing precautionary savings at- same reasons varies with the signal. For the more to a one-standard-deviation increase in the signal new precautionary term (-A(S() (|) precautioning against the Poisson paper we quantify these 4 of the previous strictly negative, so that the conclusions of the pre^aous section also carry tlirough here. deteriorating signal increases the need to accumulate resources and accordingly a (34) ' jump — 1 ), by a factor of —^f^- which we e\'ent that can occur also at found A makes consumption respond Finally, the drift in the case a\ = 0, term includes and captures any time. In the apphcation section of the effects in the context of Chile. Aggregate Hedging Precautionary savings and recessions are very costly and imperfect self-insurance mechanisms for smoothing the impact of a sudden stop. using deri\'atives 'We In this section, and insurance contracts. are obviously focusing on a ca.se where the Of we enlarge the options of the country and allow it to hedge course, the effectiveness of the hedging strategy depends on jump lias not yet taken place so that dq 16 = 0, where q is the poisson event. how the contracts and instruments that are available to the country, stop constraint, and The issue of how how accurate the crisis-signal is. the fresh funds relax the country's constraint optimal assistance mechanisms and is We not well understood. first case, is at the core of the current debate do not characterize optimal hedging strategies under different scenarios. In the these contracts enter into the sudden We about try to solve this debate but rather begin by exploring two polar cases: the hedge cannot relax the constraint directly but only improves the initial conditions of the country once hits the constraint. it This would be the case, for example, when the resources from the hedge are used entirely to pay other lenders.'" In the second That relaxes the sudden stop constraint. case, the country can buy hedging that directly each dollar received from the hedge can be used to is, fulfill the sudden stop constraint. In this section, we isolate the for analytical tractabihty, closed-form solutions. signal case, 4.1 for the constraint. this do this model that allow us to obtain use these results as an approximation benchmark for the more reahstic imperfect the focus of the empirical section. moment that the country litis no mechanism of injecting net resomces into the sudden stop However, the country faces complete hedging markets before the sudden stop we can only with An is because we can use complete-markets tools in We Hedging Precautionary Recessions Assume sense, which We above distinction and focus on the simpler threshold model. its interpret the sudden stop as a time when resources at the outset of the alternative interpretation is financial markets close all arises. In this and the country is left Xr- crisis: that the hedge is used (crowded out) by existing lenders and the sudden- stop constraint remains unchanged, except for the positive effect of a reduced initial debt (hence, there is a reduction in the required balance of trade surplus). Let us re-write the dynamic budget constraint for dXt where dFt denotes the = {rXt - c; t > + profit/loss in the futures position Since under the assumption of zero risk as: y;)dt and + tTj ntdFt is the number of future contracts. premium and any constant convenience yield for copper, d, have: Ft we can apply Ito's lemma = to obtain: dFt Recall that our 5tef'--'^«^-*) model has no straight = oFtdBf default. Implicitly, however, 17 it does allow for limited rescheduling. we Defining the portfolio process as pt = dXt tt; = Ft, we obtain - {rXt + cl y;)dt Thus we modify the dynamic programming problem spondingly, we refer to this hedging portfoUo as + ptodBf (35) in (17) to allow for a hedging portfolio pt (corre- p— hedging): ^»l-7 rT V{Xt.St,yl) the new dynamic budget constraint: = msxE (36) 1-7 Jt where r and the evolution of {Xt,St,yt) is = inf{(i now given dXt = < s) c* + y^) dt A oo} by: - {rXt -^ = s : + ptcrdBt (37) gdt (38) y't dst It = + adBt. j-idt turns out that the solution to this problem (39) simpler than the no-hedging problem because is we can use weU-knowTQ techniques from the complete markets case." Following a derivation similar to the no-hedging case, one verifies that the value function of the problem in the presence of complete hedging is given by (see the appendix): V{Xuy:,St) = "' (x a^{s^)^ ^ I ' -'>V""' 1-7 with a''(st) = - (l + e-^'(^-^)(K^ - 1)) Correspondingly, the consumption policy in feedback form Wliile it case, this is may appear from not so. To this expression that not see this, apply Ito's lemma (40) is: much to the right has changed with respect to the no-hedging hand side and simphfy (see the appendix), to obtain: dc* That is, excess consumption is = 0. constant throughout the pre-sudden stop phase. precautionary recessions, and the signal does not affect consumption. "This is possible because markets are complete at all dates but r. 18 There are no more > How can this be reconciled with the consumption expression Xt was simply made Xt- While in the no-hedging case component, which once replaced imphes that all + e-^^i^-s.)(K~ = {rXt the variation in precautioning portfolio, p, is always negative. when Its the signal goes to of this investment position is now the behavior of includes a hedging portfolio it is -l)J r-g^ \ is c; + y;)dt + ptadBt absorbed by Xt rather than by consumption.^^ The hedging is largest when the This means that the coimtry and the amount signal is is at the trigger point s and always shorting the asset that of shorting rises as the signal worsens. The is counterpart a reduction in the countries' external debt as the signal deteriorates. However, this form of insurance the coimtry can do is - absolute value infinity. perfectly correlated with the signal, surpluses it is in in dXt AU of riskless debt, The answer pti l goes to in (41)? will to arrive at the not entirely remove the consumption drop during the sudden stop. sudden stop with required to have during the crisis. less debt, Because of and borrowing throughout the pre-sudden stop phase for and hence to reduce the this drop, the country precautionary reasons. It is still size of the trade cuts consumption simply no longer state (signal)-dependent. Optimal consumption before the sudden stop is constant at (see the appendix): V _l_ ^0 E[j^e-rtdt + Since K > 1, it clearly is lower than the 4.2 Hedging Sudden Stops Let us now assume that there level of e-rrl<l!l consumption in a framework without sudden stops: a second asset, H. with the property that is it can be excluded from the sudden stop constraint and hence can be used directly to overcome the forced savings problem. This can be thought of as a credit line that does not crowd out alternatiT.'e funding options (we refer to this case as /i— hedging). Formally, the problem becomes: .1— V{Xt,St,yt) = E max it '^The appendix provides a formal proof of and return to a fuller characterization in this 1 (43) -7 statement, which follows steps similar to those in Karatzas and the implementation section. 19 Wang (2000), where = T and the evolution of {Xt, Ht,st,yt) inf{(t now given is : s < Aoo} s) (44) by: dXt = {rXt dHt = vHt+ptadBt+dNt vHt+ptodBt+dNt (46) gdt (47) - c* + dst = ndt + adSt Ht > Q, i>l ij;) dt + ptodBt - idNt (45) (48) . (49) Observe that we now have a second portfoho process pi and an increasing process dNt representing Moreover, every addition into Ht from Xt additions into the second type of asset. markup ^ fee, C to imply that Ht Since 1- To be more Xt+t ^ X + , the presence of a that at time t the holdings of the asset we explain we can This is = X;~, < </>< e"-^. follows that the constraint will be non-binding Ht > h* (50) because the country can use its write: Xt^=Xt-^Ht. It will as: rather than Xt- to relax the constraint directly, so that Ht below. e9^) +<?i^^(l-e-('"-»>^) now reads Xr~ markup would seem not the case precisely because Ht has the ability to is sudden stop constraint precise, let us re-write the ('?^(l-e-''"-s>^) the form of we assume ^t- In other words Notice that the constraint Ht this in order to relax the constraints as ^r+T>^r- holdings of arri\'e in dominated by Xt- However, is relax the constraint: be added to X^.- income flows tlie accompanied by a is (51) if (Xt- + ^^] (52) where h' To see why this Xt+ is so, Xt- + Ht = Xt- ('0(1 - (<A(1 xr; e-f'-s'^) h* [Xt- + e-(^-')^) + e^^) ) V = - Ht = note that for = = y'r+T T — g _ — -^ ) at time + <t>^^{l — T Vt r e^^ - l) + g 20 Q (53) . t+ total resources e-<'-s)^) + (e^^ ^ - l) ' become: ^ yr_ T — 9 However, as we established in equation the unconstrained solution automatically satisfies the relation: (2), A^+ H r which implies that A Xt+t ^ Xr , i.e. - — A.T-+T H r g the constraint is 1 g Ht > satisfied automatically for modification of the martingale approach of Karatzas and Wang > and characterizing the solution < Cf,pt,pt,Xt, Ht,dNt finding - (2000) any feasible consumption-portfolio plan we Applying in this context. = ^^ + '^(1 [y*t-cl)e-^'dt-i E{e-^^Hr) = Ho + Eie-^^ {Xr We will assume that at time + e( f e-'-'dNt^ ^Hr)) it can structure as a "markup," ^, on it the country starts with desires. I.e., this credit line. Then -^f- ) • Ito's is finite lemma to almost surely, e-''dNt] j . =Xo + iHo + Eff interesting interpretation of this equation line that + get: E{e-'^{Xr+(Hr)) = Xo + An X^.- get: Eie-'-^X^) Combining the above two equations we I particularly well suited to is e~^*Xt and e~^* Ht, and because we have a bounded portfolio and a stopping time that for h* is it Hq = E(j as follows: can choose 0, {yt {yl - c*) e"''' dt\ so that - we end up cl) e-^' dt\ Suppose the country Hr "state by state." the price of the credit line with: (54) . offered a contingent credit is But assume too that it pays is: Pcci=^E{e-^^H^). With this definition we can rewrite (54) as: E(e— XO = which is the standard budget constraint (Xo - Pcc) (e.g. + E (j^ [y; of Section 4.1.) - c,*) e'^' dt\ after subtracting the initial cost of the contingent credit line from Xq. Prom now on, it will prove useful to express the amount of the credit line as a fraction of \Xt- + ;rf^ ) so that: hr For It is T.'alues of h < h* the constraint will be binding and consumption will drop during the sudden stop. not difficult to show that in this case consumption after the sudden stop 21 (i > r + T) is equal to: — which follows from Given Xras long as h {t < t < T , < this (8). the same quantity as in the no-hedging case because the sudden stop constraint binds is However, the h*. + T) and (4) since Ht <. = entirely used during this period is for = h . Consumption during the crisis is now equal to: -^^^ (X^- + -^) (l+h- e--^<P{l - e-(-^)^) - e-t-^)^) — e~^^ r — q 1 which consumption during the sudden stop phase effect of the credit line is to raise \ h* rises to + h')[X^-+-^^]=c%r- r[l Accordingly, the value function at time r is: vss ^ c= (55) 1-7 + h'y-n^^^,.y + [(1 (^^ -W c^ i^ii„<,,.j }J where: ''^ Ki = {l- Now e-'-^)^(l + h- + e""-^!! + e"'-^ (1 + /i*))^-^ /i*)^"^ < K. the optimization problem has the same form as before except that the intertemporal constraint takes into account the fact that the consumer essentially faces two types of Adopting the Cox-Huang (1989) methodology and Wang ping time in Karatzas and min k max E is X and H. application to problems involving a its random stop- are able to reduce the problem to the following static problem: e-'-'dt + e-'-'V^^(Xr,Hr)-k( f Cte'^Ut + f V 1-7 V^o Ct,Xr,Hr e'"'^ {Xr + e.Hr Jo s.t. where V^^ we (2000), sissets, e( f c'e-'-' dt + e-'-" {Xr + ^Hr)) =Xo + e[ f yle-"' dt\ given by (55). The presence of dynamically complete markets allows us to reduce the problem to an essentially static problem. Parties can contract on the payoffs to be transferred "state by state." In other words, the objective inside the brackets It is is maximized state by state and the optimal ''It might seem puzzling why time r+ are now given by in the ab.sence of Xr rephcated dynamically. easy to show that in this framework one can derive optimal consumption to be: Cj* in the payoflF is {l a markup, C= (^)'' (1 -I- /i')'"'' + h") [X^- + -^^ it will be the case that ) . = yt-i,iG (O.r). when h = h' and not Actually one can (l ^ + h')/ (x^T \^ complete absence of sudden stops. 22 C= ea.sily show^ in = + -^^1 T-g J (-)"' • The rea.son is that total resources at the framework of the threshold model that X:^"" ^ + -^^. where Xi""" denotes the ^ 1 g level of where fc a constant that is is determined a way that the intertemporal budget constraint in such is satisfied. The crucial step In the appendix is the maximization of the problem involving the continuation value function at time r. we show that the solution to this problem . Xr- = Ifpt ^ T{l h* is: —J- + - Vt r-v h*) [\ - + e-'"^){\ h*) with K-i) o<r — Notice that as ^ > 1, /i ^> /i*. As might be expected <i. in this case, the constraint, the country will optimally choose h = consumption adjustment during the sudden stop be required. will h* > li ^ . 1 it will Because of the homotheticity of the problem, the optimal credit initial wealth. To complete the solution to the overall problem, and combine everything to solve + costs nothing to avoid be true that h us return to the time let < h* and does not depend on the line ratio + /i''p*or" il + h* £[e-''^] (i Jo - it some level of budget constraint for c*: dt 1 where 1 e-^(^-^) (1 e-^(^-^> + T -1 /i°P*Or" (l + Vo Xo Xo + Vo ft*) where as usual: A The credit line reduces the consumption. However, since = (56) magnitude of the sudden /i°p' < h* , there is still the remaining need for a precautionary recession. significantly less 4.3 stop, scope which translates into higher pre-sudden stop for The economy hedging; pt takes the slack and eliminates still suffers through the sudden stop, but than without hedging. Imperfect Signal In general, crises will be correlated with a signal but not but the nature of the solution can occur for any to contrast it s, is perfectly. This complicates the p-insurance case similar to that of the threshold model. On the other hand, the fact that r considerably complicates the /i-hedging case. In this section with the perfect-signal scenario. We return 23 we develop the former case to the ft— hedging case in the empirical section. The steps of the derivation are similar to those in the threshold model, appendix. The optimal (excess) consumption and portfolio 6f(..)(A', policies and we relegate them to the become: + ;^) Unfortunately the function b^ {s) can no longer be characterized in closed fonn, but it can be computed numerically. Applying Ito's lemma to the right hand dc* side of (57), we obtain the process: / A(s, dt. 1 (.59) c; Comparing unhedged case this expression to that of the eUminates the diffusion term. in (34), However, unlike the hedged case shows that the in the threshold possibility of hedging model, there is still pre- sudden-stop precautionary savings and these fluctuate over time as the signal changes. This happens because the sudden-stop An 5 In tliis jump is not directly contractible. Illustration: we section illustrate case study since Chile flows' volatility. is The Case of Chile our main results througli an application to the case of Chile. an open econom>\ with most of Moreover, the price of copper (its its This is a good recent business cycle attributable to capital main export) is an exceUent indicator of investor attitude toward Chile. 5.1 Calibration While our purpose here is only illustrative, and hence our search for parameters some time describing our estimation/calibration Estimating 5.1.1 we still we use only the we spend of the key function A(s). rare events. This highlight our procedure because Similarly, a key issue rather informal, X{s) Sudden stops are very severe but ever, is is it makes it hard to estimate X{s:) with any precision. How- provides a good starting point for an actual implementation. determining the components of the signal, logaritluri of the price of copper. In s. Given the hmited goal of a true implementation 24 it also this section, would be worth including some global risk-financial indicator, such as the U.S. high yield spread or the moving trends from the EMBI+, and removing slow price of copper.^'' With these caveats behind us, let us describe the reduced-form Markov regime switching model we use to estimate A(s). Unlike the standard regime-switching model, ours has time-varying transition probabihties. Our left-hand side variable is = Yt The growth the regime o-Q- tJ.o + the regime Otherwise, the regime "normal" and growth is is = = Q\zt 0,Xs,seit,t+i)) Pr(24+i =0\zt where poi = 1 — Poo, Pio = = following: = exp"/.'^' ^(-")<^- = poo Pr(2t+i 1 =Pio l,Xs,s€{t,t+i)) — g, and Xt stands and 1. When the value of normal variable with mean Hq and variance just a assumed to be the is by the process:^^ 0. that takes values Zt "abnormal" and growth has a lower mean, transition matrix between the two states Pr{zt+i is to be generated £-7^(0,1),^! < l[^^=i}^ii+a^_^s, rate of Yi depends on an unobserved regime is 0, we assume aggregate demand, Yj, which + ^Zg = Pr(2t+i Mi> l\zt = = l|-2t ^nd a variance 0,X,^se(t,t+i)) = = Poi l,Xs^se{t,t+i)) for the regressors that enter the The a\. = 9 hazard function (the logarithm of the price of copper in this application).'^ For the purpose of estimating A(s), we use annual GDP data starting from 1976 to 1999 and monthly copper data for the same period (source, International Financial Statistics).''' For estimation, we use a Bayesian approach that is Bayesian inference seems natural in this context, since suitable for the very few datapoints that it we have. allows exact statements that do not require as- ymptotic justification. To estimate the parameters of interest (namely (ao,ai)) we use a multimove Gibbs Sampler as described in (1989). Kim and Nelson (1998,1999) and is For details we refer the reader to these references. '''Removing the slow moving trend not only seems to improve the based on the The fit filtering basic idea but also is is to algorithm of Hamilton augment the parameter a key ingredient in designing long run insurance and hedging contracts. Few investors/insurers are likely to be willing to hold long-term risk on a variable that may turn out to be non-stationary. '*A11 aggregate quantity data are highly correlated during large crises; hence the particular series used is immaterial for our purposes. '^Strictly speaking the quantity exp~ J« .^(^v)<'u jg unobservable, since we cannot observe the continuous path of copper, only discrete points. However, w"e can obtain copper data at reasonable high frequencies so that w"e can safely ignore this issue and calculate the integral by a Rieman sum as exp-/.'^'^<-")''"=^exp-£^(-'')^" '^There due is a caveat here. to political turmoil Our extended-sample combined w-ith starts abnormal years these do not influence the estimation states). ]t is in this sense that the from 1972, but all the years up to 1976 are part of a deep contraction extremely weak external conditions. However, since the extended-sample sample relevant of X{s) (w-hich for is estimated from the transitions from norma! to abnormal the estimation of the latter starts 25 starts during in 1976. For details see below-. by treating the unobserved states as parameters. Then we set (/ig, ^ij , ctq. f1 , <?, oo, cti we sample Kim and Nelson the conditional means a'o>/^i leads to a normal / trucated normal posterior.'* (1999). Given the same Conditional on the states and the draw from and the draw from Similarly, fixing the states class. beta prior leading to conjugate beta posterior. The samphng gamma (i-Iq, where we sample from a bivariate normal priors (/^q, /.ij) se which lead to posteriors in we draw iii,C(j.Oi) new draw we can using a conjugate use a Metropolis Hastings- Random walk- centered at the previous iterations' estimate Robert and Casella (1999). This provides us wdth a new draw form as described in q of (qq.oi) presents the difficulty that there we does not seem to be a natural conjugate prior to use and thus accept-reject step draw from the posterior distribution using a conjugate normal / trucated normal prior which sample from the posterior chstribution of {oQ,ai) using inverse the a draw from the posterior distribution of and conditional on these parameters we draw from the posterior chstribution of the ) states in a sigle step as described in of the states fix (/Iq! Mi-f^o. '^i- (/, «i) <^o. the states again, b;ised on the and conditional on this new draw then a standard result in Bayesian computation that the stationary distribution of the etc. It is iterate the algorithm by filtering parameters sampled with these procedure (treating the unobserved states as parameters too) coincides with the posterior distribution of the parameters.'^ A first output of this procedure Figure 2 which plots the probabihty of being in an abinormal state.The is model recognizes roughly three years out of the 24 as abnormal The following the Asian/Russian crises. was significantly more years: the early 1980s and the recent episode early 1980s episode corresponds to a devastating debt-crisis, severe than the recent one. In fact, the recent episode appears to be a and mix of a milder sudden stop and a precautionary recession. An interesting observation about these probabihties is that they allow us to identify the abnormal regimes with great confidence. To improve the tightness of the posterior confidence intervals concerning the parameters of interest (ao,Q:i) we observe X] where x, takes the \'alue 1 if that conditional [Xtlog(poo(«o-a:i)) there is on the + (1 -Xi)log{poi(Qo,ai))] no transition to state to notice that the hkelihood for (qq, oi ) states the (log-) likelihood function becomes: 1, and takes the value depends on the data only through the otherwise. It is interesting filtered states. In other words '*For details see Albert and Chib (1993). To reduce computational time and used proper priors was Normal / for (/jqiMi) ^"d satisfy the technical conditions required for the applicabibity of the {o'0,o>l) and improper priors Truncated normal with means (0,0) for the rest of the we used independent normal Gibbs Sampler, no The proper prior for (aiqi/'i) and a diagonal covariance matrix. The standard deviations where to be roughly 5 times the range of the sample, so that the priors for (ao,o;i) parameters. priors centered at had effectively no influence on the estimation. with a standartl <leviation of 10. cho.sen Similarly Even when we experimented with more diffuse priors the algorithm jjroduced virtually identical results but computational time was significantly increased. because convergence was significantly slower. Most importantly though, u.se in the calibration exercise we used completely flat jjriors. 26 for the results that we report in Table 1 and that we Probability of 0.6 abnormal regime - 0.3 1975 1980 Figure all 1985 2: 1995 1990 Probability of abnormal regime other parameters of the model including the GDP-data are relevant for our purpose of estimating (qq. ch) only to the extent that they influence the identification of the states. In other words directly on the states we would be able to get Given the few data points and the quite rid of the noise introduced by clear identification of the if we were to condition filtering. abnormal states we report directly the posterior distribution of (ao.«i) conditional on the early 80's and 1999 being the abnormal states, we report the posterior distributions of the parameters (ao.cn) conditional on the states that are identified as abnormal.^" Table reports these results. 1 Correspondingly, in what follows we use qq = 2.5, qi = 5.2 as our benchmark values for the function X{s). 5.1.2 We Other parameters calibrate 4> and the sample, which T is to generate the average cumulative consumption drop caused by the sudden stops in about 8-10% of GDP. One such configuration The parameters k and 7 ^"^I.e., states that h.i\'c are calibrated to generate reasonable is T= 1 amounts and <p = 0.35e''-^. of steady-state debt (and a postfrior probability of being abnormal above 0.75. Tlie conditioning the precision of the e.stimates and seems to be warranted just by a visual inspection of Figure 27 2. is done in hence order to increase Posterior Distributions mean "1 Table Deviation 10 25 50 75 90 -2.522 1.067 -3.826 -3.027 -2.316 -1.789 -1.420 -5.170 3.387 -9.259 -6.802 -4.708 -3.040 -1.515 std. Posterior Distributions of the Parameters (ao,ai) Conditional on the States that are Identified as 1: Abnormal. 10, 25 etc. refer to the respective quantiles. reasonable amounts of insurance needs) together with significant precautionary fluctuations (see below, in particular, to explain 1998-9 episode). For The to a much of the precautionary recession experienced by Chile at the early stages of the we tliis, set = k interest rate r is set to 0.09 0.8 and 7 = and the growth rate path that grows significantly faster than that Xo, to Finally, to 0.03. level for a The latter is (yo) to 1, and is catching up, and set the initial debt-to-GDP 0.5. we approximate the process for copper by a tlriftless Brownian motion with constant volatihty which we estimate using monthly data from 1972 to 1999 (source, IPS). and normalize the 5.2 a constant-rate approximation few years, while the country We normahze initial GDP then decelerates below that level forever. ratio, 7. initial value of s to We find a value of roughly a = a, 0.23 0. Aggregate Hedging Recall that our concern in tliis economy is to stabilize both precautionary recessions and sudden stops. Let us start with the former. 5.2.1 Hedging Precautionary Recessions Evaluating at in ^ as 1 , and s = in the calculation of ^ , we can approximate the volatility of log consumption our model before the sudden stop takes place by: -(l-K)--^rr==0.01. That is. despite the absence of income fluctuations in our model, fluctuations in precautionary behavior can account for about a fourth of consumption volatility in the regimes that are characterized as normal by our algorithm. More importantly, the contribution of precautionary fluctuations sudden stops, as Panel shown by Figure (a) in this figure a sudden stop takes place. is particularly relevant near 3. shows a random realization of the path The main s that runs for nearly eight years before features of this path are not too different from the realization of the 28 0.24 — ^ 023 No Hedge Hedge y ^\\ g "•^' I 0.22 C « <D ;j:/!L ,../> -- 'V:yr •\i^ 0.21 ..-'^'^M 0.2 ;'.;/" / 0.19 4 time Figure 3: p-Hedging for the Chilean case price of copper during the 1990s. In particular, the large rise in s near the middle of the path followed by a sharp dechne toward the end of the period is reminiscent of the path of the price of copper from 1996 to 2000. The dashed hne in the bottom left panel (b) illustrates the corresponding path of (excess) consumption generated by our model without hedging. The More hne interestingly, drift in the process is due to average precautionary savings. one can clearly see the precautionary recessions caused by the decline in the top right panel (c) shows the impact of the latter in s. The dashed on the current account. In particular, the sharp decline in the deficit in the current account from 6 to about 3.5 percent of GDP toward the end of the sample. This suggests that about half of the current account adjustment observed in Cliile during the 1998-9 crisis can be accounted by optimal precautionary behavior in the absence of hedging. The rest of the adjustment could be accounted for by excess adjustment (some have argued that the central bank contracted monetarj' policy excessively during this episode) and by the partial sudden stop itself (the "sovereign" spread tripled during this episode). Now, let the country hedge by shorting copper futures, which do not relax the sudden stop directly but 29 Xr only through their effect on to the (p-hedging). The dashed paths of the unhedged economy. Panel (b) clearly shows that hedging, even of this very hmited kind, virtually eliminates precautionary fluctuations. consumption from precautionary cycles improvement in the current latter are reflected in a As a practical matter, are very large. The is Interestingly, panel (c) account as size of s deteriorates The difference comes from the comes from hedge-transfers sharp dechne in external debt as s worsens (panel it is shows that the insulation of not done through a smoothing of the current account, which looks hedged and unhedged economies. virtually identical in the The show the paths corresponding solid lines in panels (b) to (d) fact that the in favor of the coimtry. (d)). important to point out that the hedge-ratios required to achieve this success the imphed portfolios can be calculated as (evaluating at s = 0): ^(Xo + (1-k)-^)«-0.56. r b That the notional is, very large amount g would have to exceed 50 percent of GDP. This in the futures position when compared with the existing futures markets for copper, pointing to the or related markets for sudden-stop-insurance purposes. However, it is is need to develop these not a disproportionate number when compared with the very large and much costher practice of accumiilating international reserves (around 30 GDP). percent of and Finally, in as indicated in section 4, this strategy smoothing the sudden stop hedging of this sort — if Our estimates show itself. subject to full crowchng out — is not very effective that in case of a sudden stop, consumption with exceeds consumption in the unhedged case only by about 2% for consumption drops of roughly 8-10%. Hedging Sudden Stops 5.2.2 now analyze the opposite extreme and assume Let us (/i-hedging) . Full insurance in (Xo + case requires a transfer in the case of a sudden stop ~^] (1 h' tliis -K)yo\ _ = / ,., (0(1 Importantly, note that 15 percent of tion to be in case of in the case of „-(r-s)^^ e-(^-^)^) + , „oT e^^ - of: ,(l-«)yo' Afv l) (Xo + r — 0.15. g GDP exceeds the 8-10 percent we calibrated the dechne in consumpis because 15 percent covers not only the drop in consumption a sudden stop but also the precautionary sa\dngs and recession that precedes the sudden stop. depends on the initial NPV of such a claim, we must multiply the number above by E[e~'"'^] for different Using the benchmark case, we see that GDP E[e~'"^]. Since the value of the signal s and on the estimates of a's, which are very imprecise, report in Table 2 the value of percent of - a sudden stop. This In order to calculate the latter that hedging directly relaixes the sudden stop constraint (0.15 times 0.536) if we combinations of these parameters. "fairly" priced full insurance would cost the country about 8 contracted when the price of copper 30 is at "normal" levels and about 5 — — Table percent (0.15 times 0.323) cost of 2: s = s = (2.5,5.2) (5,5.2) (2.3,0) 0.536 0.247 0.517 0.323 0.139 0.523 1.5(7 Values of E[e~'"^] for Different Combinations of (qq, oi). when While these amounts are small when compared with the at very high levels. sudden stops and precautionary recessions, they for these countries to undertake, Fortunately, insurance. In much what even still amounts which are probably too large involve such markets existed. if amounts of of the advantage of insurance can be obtained with significantly smaller follows, we take h and as given n maxE solve: r.*i f.*i ^ 1 r -7 s.t. dXt = {rXt-cl + y;)dt Xq+ = Xq- — Po Po In other words costing Po which is we assume e-'-^hiXr- + ^^ r-g that the country starts with an initial debt priced fairly common knowledge is =E and denominated to both the borrower as a fraction h and the < Xq- , and purchases insurance h* of the quantity X^-- tools Po-^^ type of contingent claim results we developed The The first in a problem that we can in the previous sections, plus a fixed point results are summarized row shows Fqi the in Table second cj, problem stop. The next three rows while the last three columns do the same for sq There are three main lessons to be learned from insurance is easily solve numerically maximum show consumption = The first with the The right before the final row describes three columns present values for 1-5(t. this table. First, clearly much of the cost of a lack of paid for with the large precautionary behavior that the economy needs to undertake without ^'Eveii though the contingent claim signed by the borrower and the lender seems complicated, not varying Using value attained by excess after the crisis, respectively. the standard deviation of c*+ for a sample of 1000 simulations. 0, -^f-. for the initial price of the insurance, while the third row shows the sudden stop takes place, during the sudden stop, and = X^- + 3. consumption along the path before the sudden So setup lender, so the lender prices this claim understanding the subsequent optimization problem of the borrower and accordingly the resulting path of this particular + ^^. The ver>' much and can be approximated reasonably well by 31 Xq + -^, it isn't since especially for large h. -V^- H ^^ is . 1.5rT So h/h': 0.4 0.8 0.000 0.019 0.039 0.252 0.213 0.247 0.252 0.248 0.253 0.243 0.252 0.254 0.212 0.244 0.252 0.224 0.247 0.253 0.010 0.002 0.002 0.011 0.002 0.001 4+ 0.125 0.175 0.229 0.131 0.177 0.230 <+T 0.300 0.269 0.2.59 0.313 0.271 0.260 Standard Deviation of c*^ 0.013 0.002 0.001 0.014 0.002 0.001 Initial Cost c5 C'r- Standard Deviation of c*^^^. -c;_ Table insurance. be seen in The level dimension of — c*^ . less itself is significantly 0.000 0.032 0.063 0.198 0.243 0.226 /i- Hedging 3: cj, while the cyclical component can Second, much of the precautionary costs can be removed with very limited amounts of insurance; a value of h stop 0.8 can be seen in the value of this effect the difference between c*max 0.4 = harder to smooth, but 0.4 does nearly the same as one of 0.8. Third, the possible to make a difference with h significantly it is still sudden than h" 5.2.3 Asymmetric Beliefs and Contingent Credit Lines Before concluding, we discuss an important practical country's actions and there is significant issue. Sudden stops are not entirely exogenous to the asymmetric information about these actions between borrowers (the country) and lenders. Suppose tlien that the financial markets overstate (from the \'iewpoint of the country) the constant part of the hazard, qq. In particular, The borrower on the other hand takes this borrower and the lender be E"" we assume number to be Oq that the lender takes this constant to be Qq << Qq . . Let the expectation operators of the and E^, respectively, so that the problem becomes: "-""' ^-ds 1-7 + e-^^V^^ {Xr^) s.t. dXt -= [rXi -c*+yl)dt Xq+ = Xq- — Pq Po = E' e-'-^h{X°1' + In this case the country obviously will find the price of the insurance "unfairly" high. 32 One way to reduce u;=3 uncontingent 0.4 hlh": Cost 0.8 0.032 0.063 0.009 0.020 c5 0.219 0.247 0.250 0.238 0.246 <- 0.219 0.246 0.251 0.245 0.255 <+ 0.128 0.175 0.228 0.155 0.187 Standard Deviation of c*+ 0.009 0.002 0.001 0.016 0.026 Table the extent of this problem this CEise the country fiscal deficits is for the ft— Hedging with 4; asymmetric Beliefs country to add another contingency to the credit hne. would benefit from making the hne contingent on the value of crises not related to s as this feature 0.4 0.000 Initial run up 0.8 — much it is less likely optimal for than the markets - for example, it clear that in Since the coxmtry sees s. may be not to pay for insurance in states when s it It is certain that is it will not high. Let us capture by assuming that the borrower and the lender enter an agreement whereby the lender agrees to pay: Vr /(sO/iLy!? + to the borrower and the fraction of h paid out depends on St fiSr) exp{—w{s-r = 1 Obviously < /(s,-) and only when Table 4, s,- is < 1. Taking Qq ~ 2.5, + 2a)) + exp(— u;(st- + 2a)) a^ = 5 and w= very low (2 standard deviations below 3, i.e. 0), a claim that pays quite steeply the results are reported in the when first half of and contrasted with the case without the additional contingency. By adding the additional state contingency, a country that is confident of its "good behavior" is able to lower the price of the claim without incurring a very significant rise in risk exposure. 6 Final In this paper Remarks we characterized many aggregate hedging strategies. For the key issues but realistic aspects of sudden stops, precautionary recessions, and the corresponding this, we built a model simple enough to shed analytical hght on some of enough to provide some quantitative guidance. We showed that even after removing aU other sources of uncertainty, stops and the mere anticipation of them has the the combination of infrequent sudden potential to explain a large share of the volatile business 33 cycle experienced by emerging market economies. This important source of volatility could be overcome with suitable aggregate hedging strategies, but these would require developing large new financial markets. In our application, we hinted at one aspect of the insurance arrangements that could faciUtate the development of such markets by reducing the inherent asymmetric information problems. credit hues for and financial instruments being contingent example, Chile could remove many its line to the price of oil of the signaling problems and US GDP, and so There are several aspects of an aggregate we did not risk this discuss accumulation of international reserves, as these are these issues properly, signal worsens. Of until the crisis is we also need to course, a country imminent, but may for it fears from the IMF's Contingent Credit Mexico could do the same by on. management model the maturity structure of debt and how reserves financed with long-term debt argued on indicators that are exogenous to the country. Thus, Lines by adding a clause linking the size of the line to the price of copper. indexing We strategy that we left unexplored. In particular, changes with the signal, dominated by oiu credit s. Similarly, lines. we did not However, building partially substitute for overpriced credit lines. In order to address ha\'e a better understanding of the behavior of the supply side as the would want to postpone accumulating reserves and borrowing long term it is highly urdikely that the lenders will go along with this strategy. intend to explore these dimensions in future work. 34 We References [1] JH Albert and Chib S. Bayes inference via gibbs sampling of autoregressive time markov mean and variance [2] C. Burnside, of Political [3] Ricardo J. shifts. J. Bus. Econ. Eichenbaum M., and Rebelo Ricardo J. Ricardo J. Prospective deficits and the south east asian crisis. Journal Caballero. Macroeconomic Volatility in Reformed Latin America., volume 25. Inter- American first edition, 2001. Caballero and Arvind Krishnamurthy. International and domestic collateral constraints in a model of emerging market [5] Stat., 11(1):1-15, 1993. Economy., 2003. Development Bank, Washington, D.C., [4] S. series subject to crises. Journal of Monetary Economics., 48(3):513-548, 2001. Caballero and Arvind Krishnamurthy. Excessive dollar debt: Financial development and underinsurance. Journal of Finance., 2003. [6] John C. Cox and Chi-fu Huang. Optimal consumption and portfolio pohcies when asset prices foUow a diffusion process. J. [7] Econom. Theory, 49(l):33-83, 1989. Francis X. Diebold, Joon-Haeng Lee, and Gretchen Weinbach. Regime Switching with Time-Varying Transition Probabilities in Nonstationary Time Series Analysis and Cointegration, chapter 10, pages 283-302. Oxford University Press, 1994. [8] Darrell DufEe, Pedersen Lasse, and Ken Singleton. Modehng sovereign yield spreads: A case study of russian debt. Graduate School of Business, Stanford University, forthcoming. Journal of Finance, 2000. [9] Wendell H. Fleming and H. Mete Soner. Controlled Markov processes and viscosity solutions, volume 25 of Applications of Mathematics. Springer- Verlag, [10] Kenneth A. Froot, David S. Scharfstein, New York, and Jeremy first edition, Stein. Ldc debt: 1993. Forgiveness, indexation, and investment incentives. Journal of Finance, 44(5):1335-50, 1989. |11] Andy Haldane. Private sector involvement in financial Financial Stability Review, |12] James D. Hamilton. 7, A new crisis: Analytics and public policy approaches. 1999. approach to the economic analysis of nonstationary time series and the business cycle. Econometrica, 57(2):357-384, 1989. [13] J. Michael Harrison. Mathematical Brownian motion and Statistics: Probability stochastic flow systems. and Mathematical 1985. 35 Statistics. Wiley Series in Probability John Wiley & Sons Inc., New and York, [14] loannis Karatzas, John P. Lehoczky, and Steven E. Shreve. Optimal portfolio and consumption decisions for [15] AM J. a "small investor" on a finite horizon. SI loannis Karatzas and Steven E. Shreve. Brownian motion and stochastic calculus, volume 113 of Graduate Texts in Mathematics. Springer- Verlag, [16] Control Optim., 25(6):1557-1586, 1987. loannis Karatzas and Hui Wang. Utility New York, second edition, 1991. maximization with discretionary stopping. SIAM J. Control Opttm., 39(l):306-329, 2000. [17] Chang-Jin Kim and Charles R. Nelson. State-space models with regime smtching Classical and Gibbssampling approaches with applications. [18] CJ Kim and CR MIT Press, Cambridge Nelson. Business cycle turning points, a new MA and London, 1999. coincident index, dependence based on a dynamic factor model with regime switching. Rev. Econ. and tests of duration Statist., 80(1): 188-201, 1998. [19] Krugman Paul. Financing vs. forgiving a debt overhang. Journal of Development Economics, 29:253-68, 1988. [20] Christian P. Robert and George Casella. Monte Carlo statistical methods. Springer- Verlag, New York, 1999. [21] Jean Tirole. Inefficient foreign borrowing: A dual-and-common agency perspective. Mimeo, Toulouse, 2002. [22] Ferdinand Verhulst. Nonlinear differential equations and dynamical systems. Verlag, Berhn, second edition, 1996. 36 Universitext. Springer- Appendix 7 Propositions and Proofs for section 7.1 Proposition The optimal solution of the r period problem 1 C;°P* X-r+T and Moreover the maximum in (61) the above statements, we have Proof. To establish the = and thus j/i-e*-^ rT / = yx = max{X, — r r -g (60) (61) } g r 1-7 X always given by is That . is, the constraint is always binding. Combining that the optimal consumption-wealth pair is given by (12) first assertion we need to show that the proposed pohcy and (11). (we normalize satisfies without loss of generahty) + e-^^V{X°'" + ^2^)> c* , X^, for e-'-'^vldu 1-7 Jo < u < > By T. f feasibility „* + e'^^ViXr + du ^^) r-g we have + e^'-^Xr e-'-'c^du Jo Xt where we can focus without e-^^^ r-g any other admissible policy pair ^*l-7 i-T t au Xo+ f Jo By T<t<T+T (fcl)-^, /*opt\ 1-7 g-..iS^J Jo for = is: ki is determined as the (unique) solution to: r T 2.3. loss of generality X > on deviations that satisfy the first equation as concavity of the utility function and the terminal Value function 1-7 L it an equality. follows that: r-g ^ 1-7 \Jo ^ / ^-ruc_u l^H_J < f I €-"" e-^- (c*°P'^ (C')"'^ Jo Now we constant fci du 1-7 Jo r-g + e-^^[V{XT + V + ^^^) ^^)-V{X^' 'r-g r-g - X°P' X -^)(Xt « - CP*)du + e-^^V'iX^"' + -^^^^MXt r-g fc*. c*°P')du e~^'^V'{X°J" need to distinguish between two cases: such that the budget constraint (62) is a) if X^^ > X satisfied and the then V'°p* result follows last expression: ^-1 I / (/ J e-^-{c: - cl°P')du + e-^^(Xr - 37 X'^p') 1 = = ki = (C^*)"^ upon substituting for a in the since both policies satisfy: 7o Jo b)When Xj^ = X, is feasible, it it is still the case that Xt > X = X^^ also satisfies These two arguments can be combined to (c^"''') ^ — and because the alternative candidate policy kx, Because of concavity of . V^'this also impMes V'{X + -^) < fci- get: / e-'-"(CP')"^«-CP')du + e-''^V"(X + ^^)(Xr-X°'") r - g = - X""*) f e-'""A:i« - CP')du + e-^'^V'(X + -^^^){Xt '^-9 Jo < Jo / + e-'-"fci(c:-c°P')du e-''^fci(,Yr-X°P') ' = Jo This verifies the optimality of the proposed policy. For the second assertion, that the constraint otherwise. Then is always binding, the argument runs as follows. Suppose k\ is given as: r r - g Under our (counterfactual) assumption we should have: X^ {4>{\ - e"''"-''*^) + + 0^^(1 - e^^) r or - e-(''-9>^) < Xr + r -g g — Xr < - g e-*'"-^'^) - r (63) since (1 But = - es^) < consumption when combined with the transverality condition 0. Propositions and Proofs for section 3.1 In this section the (/.(I (63) contradicts non-negativity of lim,_ooe-'-'.Y, 7.2 - same we proof all the steps of the perfect signal case. steps but has a few steps which are much harder numerical simulations. 38 The proofs to proof, hence of the imperfect signal case follow we only \^lidate them though our ) , The proof consists of the following steps. First value function should satisfy. text subject to the required that we use Then we prove the we boundary conditions that the establish rigorously the ODE existence of a unique solution to the boundary conditions. Then we in order to apply a (classical) verification verify the "tail'' Theorem along the presented in the condition on the value function lines of Fleming and Soner (1993) p. 172. Proposition 2 The value function of the problem with sudden stops for \imV{Xr,s,y;)= Proof. we We 7 > focus on the allow hedging or not (i.e., IV {Xr^^) {in s satisfies: 1-7 (64) two steps and does not depend on whether whether we require that the portfolio p of the optimization problem). First one obtains an upper sudden stops, which in this case > 1-7 The proof proceeds 1 case. s = or whether p determined as part is bound on the value function of the problem with naturally given by the value function of the problem in the absence of is sudden stops: 1-7 V NSS 1 J- To see why this is —7 an upper bound notice that the following inequahty holds for (Xr^ + V \-Hu-t){<LJLdu >• -'Al-I ^ e >" -'^^ au + 1^7-<oo^e 1{t < oo}e-^(^-"K f- I 1-7 any ^ feasible policy: 1-7 r-c, '- \Ft : 1-7 Vr l--> < E ^ 1-7 ^ \Ft 1-7 \rj 1-7 -r(u-t) V'-n) 1-7 since for 7 > when one -du + 1{t < oo}e-^^^-*^V^^^{Xr,y;)\Ft XxH 1 1-7 < 0. This holds true for any feasible consumption /portfolio policies and thus. evaluates this inequality at the optimal plan c*°P',p*: V{XuSt,y:) / op(\l-7 „-r(u—t) V-u'—^J 1-7 /IN du + 1{t < l^*opt \ 1 -7 r -r{ 1-7 7 (xT oo}e-'"(^-*^A' 1-7 / [{t < oo}e 39 +^y~'' 1 N 7 (X°P' + ^^) 1-7 -\Ft 1-7 l^( = - ( < max E * X^"'!' 1-7 + du 1{t < V^^^iXuVl) oo}e-''("-''l/(X^,y;)|Ft 1-7 1-7 rj The last equality follows true as from the principle of dynamic programming. Since upper bound. The lower bound —> oo. This establishes the st consumption /portfolio policy in the Xt > —-^ti for ^U ^ )• It is also witliin it satisfies the the class of pohcies that also it is upon observing that the established absence of sudden stops c^ss pNSS /portfoho plan in the presence of sudden stops (since also is this holds for all Sj a (=^0) is still consumption feasible intertemporal budget constraint and mandate Pu = 0. Accordingly we have the two inequalities: (x;NSS V^''-^=\XuSuVl)>E l-l Kt , r—a \Ft 1-7 1-7 -r(.-t)i5^^ VSs.v=P°-\x^,suy;)>E ^^ 1-7 i: However by standard arguments it is ^ e-'-(^-"l{r — easy to show that as s oo > < oo}K , Pr(T -) \rj ^{xr' + ^^) 1 < Q) — » \Ft - 7 for all finite Q, and thus 1-7 „*NSS\'^-'^ lim s—•oo E / e-(.-<)ifl^ Jt 1 1-7 du + e-'-(^-')A-l{T < oo} ( 1-7 V^ 1-7 du\Ft \Ft V^'^^C^t.yr) 1--I 1 — 7 yi^ss Since the upper and the lower bound coincide to as s ^r J Before following we can invoke a standard verification Theorem we — » oo, the claim is estabhshed. establish the existence of a solution to the ODE: -yC Even though this > -^rC + ^iCs + second order non-linear hard to establish that it ODE has a unique solution that ifsso"^ = (65) does not seem to have any closed form solution satisfies the required in the following proposition: 40 boundary conditions. it is This is not done ODE (65) variables a"^ = C Proposition 3 The change in has a unique solution satisfying the boundary conditions (26), (27), with the . Proof. The steady state of this 2nd order nonhnear ODE can be obtained at once as: 7 C^"" The Theorem now a system of two first - (66) basically follows from the stable manifold Theorem upon reformulating the system as order ODE's, realizing that the system has one positive and one negative eigenvalue and applying the stable manifold Theorem (see thus = we do not provide any details apart Verhulst(2000) e.g. p. 33.) The argument is particularly easy and from the fact that the linearized system has one positive and one negative eigenvalue. Indeed, one can approximate the solution by means of a hnear approximation around the steady state value to get: 7(C^^)^+(7-l)(C^^)"MC^-C^^)-r7C + Ma + ^a.a2 = The two (67) eigenvalues of the characteristic polynomial of this equation are: ^ ,±^,^-,2o-r Obviously one eigenvalue is positive, one is negative and this establishes the claim after a few straight- forward steps and the use of the stable manifold Theorem. Remark 4 A ODE simple reformulation of the (65), yields the function a{s) that we use in the text and is obtained by defining: and rewriting the ODE 7aT-i as: ^ _^^^7 + ^^a^-ifl^ + 1^2 (7(7 - l)a^-2(a,)2 + 70^-ia,,) = or: l-ra + ;ia3 + ^a2U7-l)i^ + a,, =0 (69) ) subject to the boundary constraints: a{s) lim a{s) 5 The last step in verifying the fact the problem at hand is —too = A-^f-l (70) = (71) ( ) V ^ / that the conjectmed Value function to invoke a verification Before we do that we prove an almost obvious is indeed the Value function for Theorem along the lines of Fleming Lemma of independent interest: 41 that is and Soner (1993) p. 172. Lemma 1 Consider a consumption policy that has the feedback form: c* = A{st) (Xt + Then Xt > Xj^^^ where Xj^^^ ^"j the asset process that results is , K-^'^r < A{st) < , r Vst e {-s, +00) when one uses the optimal consumption policy in the absence of sudden stops: c^ss ^ (,^Nss ^ _yL_ ^ r-g \ Proof. This result is trivial. One solves for the asset process that results from the two policies to find that ^NSS The Xo - equivalent calculation for the optimal policy d (e-(^'-/o ^(«.)d-)x,) Now integrating both sides X, = ^^ -5 yp Ct = -(r = ,-frt-L- A(s,Adu) - - 1) conditional on a path of A(s.)du) yl(si))e-('-'--^o and rearranging Xo + ^e('-*-/o f^gt (e^' r ^^ ^ e'i^'-fi A{st) [ r-g - \. r - result now foUows To invoke a Lemma since A{st) verification < Theorem we 2 Assume (i)^ < C{st) < gives: Ms.)du) f r Vst € f' ^-Hr-g)^-fi .4{.„)d.)(^ {5, _ +00). need to show finally K {^Y Vs, that: € {s,+oo). Then - lim Proof. For any t it is E e-'-'l{r > t}C(s] N 1-7 = 1-7 the case that e-'-'l{T > t}C{s) > 1-7 1--) E [xr e-''*l{T >^}A- + ^ 1-7 1-1 KE e-'''l{T>^} ^^'-^''")dXt g ^^; Xo-^2_('eSt_e('*-/,M(^..)<'.)^ r - g \ J The S( gives: \V {xr' + ^) 1-7 42 ^(^.,) _ ^)^.\ ^ — The inequality follows from the assumption first The second from the previous lemma and and the fact that '- -^^ —iz:~ is the monotonicity of the value function and the last limit follows from the fact that one can tri^dally show that in the standard model without sudden stops e The in a rather straightforward — C~ a{s) (If it increasing). To establish that maxima extremum. Then asis") = — ra < it 1) 1 to be a local 0. Then the region where 2) 1 — ra > 0. > 1 s* — Then To or we establish the claim it we decreasing, would have just need to to < show that Now < minima. So suppose indeed that there would have to be 0. exists ODE) it one point s* that that a^^ maximum, which local < < would be is a local > 0, so that s* would have impossible since and a^sis**) > which is it would still would have to be a local maximum. But minimum Lemma Lemma impossible. consumption process: derives the properties of the 3 The optimal consumption process ^= del c^ Proof. Straightforward apphcation of 1 satisfies: , Ito's • / oa,\'^ , a, ,„ + l)(—^) dt^^adB, ^ o(7 2 \ a Lemma ~ / to: «(»,) along with equation (69.) 43 in s** to the right one can apply a standard verification Theorem and verify that indeed the proposed function following be a{s**)). so that 5* 0, is the Value function for the problem. The everywhere cases: (Since -necessarily- a{s*) has to be true that Uss a{s*) is have at least one section where order to satisfy the boundary condition at infinity there would have to be a local of s* satisfying a{s**) a(s) order for a{s) to satisfy the Boundary condition at infinity there would have in that ra -. are going to establish 2 contradictions by studying points has to be true (in order to satisfy the , it it is and a decreasing function that stays between -K^'~' is and accordingly we have 2 minimum. But to exist another s** in , exited the "band" [^, ^K^^''] that could be local ^' by means of two contradictions. Suppose a{s) solves (69) subject to will derive the result the boundary conditions (70), (71). decreasing. ^J manner: Proposition 5 The function We ^ \j:) prove that any solution to a{s) (respectively C{s)) will be bounded. This can be done final step is to Proof. a negative number. a V is Propositions and Proofs for section 4.1. 7.3 We first first show that the excess consumption process, Cj, one is to apply Ito's lemma to Cj is There are two ways constant. and observe that both parts of the Wang for distributions of hitting times of tliis. The resulting Ito process (informally speaking the dt and the dBt terms) are identically zero. Another more direct way methodology developed by Karatzas and to see is to use the martingale (2000) to deal with this prot>lem along with standard formulas Brownian Motion. This approach particularly appealing in the context is of this section because of the presence of complete markets (up to the point where the sudden stop takes place). We start Lemma with the first approach. We have the following result: 4 Under perfect hedging "excess" consumption, Cj , is constant for t < t Proof. To simplify notation somewhat we define aP(st) = (e-^'("-^(Ki -l) + J and observe that The consumption pohcy can then be aP{st) r-g V as: a Lemma (73) f^^^ydi-^) 1 and the portfoho policy l=0 reexpressed by means of (41) and (72) as ' Ito's (72) this function solves the ordinary differential equation: Ja2<, + ^iaP-ra^ + Applying l) to the right hand r-g \ side of this expression gives: dc' = g ) \aP Cidt + C2dBt where aP \ a? \ oP \ (aPfV r - aP \ r - r-g T g \ aP J — aP J \ aP A2 " 44 ' ^ J T ) 9 2 2 —g and T- g aP \ J \ aP J aP r \ - g J \ aP ' In other words perfect hedging leads to complete smoothing in our example, despite the possibility of a fall in consumption when the state variable crosses the critical level. Let us prove the same result with the martingale methodology developed by Cox and Karatzas, Lechoszky and Shreve (1987), as be using the results Lemma 5 in Wang Karatzas and it facilitate the proofs later on. (1989) and For oui exposition we will (2000). Without loss of generality, take = t = ci wiU Huang The optimal 0. policy is given as: (fc-i/^)l{«<r} \r\K) r-( where k solves the intertemporal budget constraint: Xo + -^ = fc-i/^ ( E e-''"du / + 1{t < oo} e"''" J or k Xo + = (Ej^e'-'-^du Proof. We ^1 + e-'--^l{r < provide only a sketch of the proof. The reader is oo}) referred to Karatzas and Wang (2000) for details.^^ The objective is: maxEf "I'P" [ ^^^^e-''''du + \Jo 1-7 e-''''V{Xr,y*r)l{T < c»} s.t. dXt and the We (2000). = rest of the evolution equations {rXt - c* + y*)dt + aptdBt remain unchanged. proceed to show formally that the proposed strategy is optimal borrowing from Karatzas and Wang Let us denote for an alternative admissible strategy {c*,X*) satisfying the intertemporal budget constraint with equality the Value of the objective: Jicp) = e( + [ ^^e-'-^du -7 \yo Karatzas and Even though it Wang (2000) do not strictly cover the case seems easy to extend their e-'-"T/(X,,y:)l(r < cxd} 1 we are considering here, since our problem is on an infinite horizon. approach to cover our case too, we refrain from doing so and just note that the techniques used for the limited hedging case at the beginning of section 4 along with the observation about the optimal policy and the differentiability of the a —function are enough to apply a standard verification Theorem. 45 where Let / denote the Fenchel Legendre Transform of the function - = max f{x) — f{k) /: kx so that this maximization e.g. for - 1 —7 e ~ = max e " — kc„ \^1— 7 X yields: and similarly for X-r- Then it is the case that: J{c,p) = El < £ (If^-^ +k (e < E = E / +k I e-^'^'cldu -^ Xo + r El This verifies optimality of using well / the strategy. the e-''^.Y^l{T - 9/ ^-^^—^ known formulas about + e-'-'^du -^ - first < + e-"K(X^,y;)l{T<oo} Y^e-^"du + e-^^V{Xr,y*)l{T < / j e-'-''dK \ + < oo} j + oo e-^^y(X;P',y;)l{T < oo} + Xo+^^ r - e-'-''du g + e-'-^V{X°P\y;)l{T<oo}\ We will now use the optimal strategy to obtain the Value function hitting times of Browmian Motions. V{Xo + ^2-^ r — g 46 ;,-(l-7)/7 - [-. -7 E -e-^"du + 1-7 jt-(i—r)/7 / K^/-i\^ fK'^f e- 1{t 1-7 < oo} 1-7 (^» ^) + {e 1-7 e-^'dif + l{T<oo}j e-'-" = 1-7 (^0 ^) + 1-7 1-7 -\ —7 1 (l_e-^'(^-^)+e-^><=-^)/^^)'' yr -Ai(s-s) 1-7 The only important Vry (A-- - + 1) 1 V step in this straightforward derivation This involves a well known formula which can be found is going from the second to the third hne. Harrison (1985) e.g. in p. 47 Proofs and Propositions for section 4.2 7.4 Proposition 6 For 1 < ^ < ^* ,where: c* the solution to the A problem (74) is given by: c* = /c-7,t e (0,r) -i Xr- l^opt r is r{\ ^ /j* + h*) Vl - r _(l_e-rT)(j_^;j defined by 1-^ e-'-T'il+h'] and h' is k~^ given by: IS given in the + e text. E\e~''^] (l f e-'^'du Jo and g + fe°P'C)r" '*' r (l since: 47 + /i*) Vo Xo + T -g where A is given in (56), we get finally that + (1 + h*) Proof. Adopting the Cox-Huang (1989) methodology and stopping time Karatzas-Wang (2000) max k c-^.XrMr where V^^ is its -9 apphcation to problems involving a random are able to reduce the problem to the following static problem: ^^du e"-'V^^{Xr,R-r) -k( [ 1-7 + E min we T e-^V-, + e""" {Xr + ^Hr \Jo IJo given by (55). By maximizing the objective inside the integral one can derive optimal con- sumption to be: = c; The crucial step is k-^,t e (0,t) the maximization of the problem involving the continuation value fimction at time max [V^^iXr. Hr) - k{Xr + where V^^ is given above. Assuming for the moment Hr<C[ we can ^Hr)] r: (74) that Xr- + r- g solve the problem (74) using (55), to get: 1 - e-'-^rB {B(Xr-+ -^^ ]+Hr] r - + e-'-^ (1 + h')''^ Xr- + T g [l - e-^^r { B { Xr- + r where: 5= l-e-'"''(l-h/z*) Let us define fCi so that we — k arrive at: 1-7 Vr k.^B + e-^' {l-\-h*y-^ (Xr- + T Xr- + y'r -g k,il-iB) Y^ ,\l—l T-gJ e-"-^ (1 -h /i*) 48 = k, - - g + Hr g k^ or fci(l-^-B) Vr X,- + T -g e-'-T (1 /i*)i-^ + fci(l-^(l-e-^(l + e-'-^(l (1 It is an interesting observation {that we + will + 1-e + + e-^-^Cl < F < Now 1, F ' > + - + e-'-^(l hi Hr (1 - as: + H^ kxi + Hr hi -rr\-y (l-e-^) /i*)'-'^ fci(l-^g) B hji-JB) B e-rT) *^i-^ e-'-^(l+/i*) get the ratio Hr Xr-+^ we combine the preceding equations to get: h We first = {l-e-'-'){l verify that indeed h (1 < h*. .*.f( + h*) This is (l-e-'-^d + (l-e-^^(l + - (1 + r - e-'-^(l + hn) + h* + h*) h* Kh* smce: > F e--'''(l (since ^ > 1) and thus (1 - e-''^)(l + /(*) /i*)) true since: .J^ - e-^^){l + h*)[^\ (f)' 49 + ft-) <0. 1 we get: (75) C 1. kijl-^B) B e-'-^)'^ > l,^B < Zi*) the holdings of the second (hedging) asset are determined (1 To /i.) use later) to note that since ^ 1-^ and we have that +C e-rT(l /1*) /i*))) + /i*)-^ /i*)(l + 1 1 > . Finally we want to provide conditions in order to exclude that h becomes negative and this tvill be the case whenever: / £ 1 h^ h' <e-'-^(l + /r) {l + + h'){l-e-^] 1 or i>e 1 = (76) Accordingly we only focus on cases where ^ As might be expected due on the < c^* to the homotheticity of the problem, the optimal reserve ratio does not To complete level of initial wealth. the solution to the overall problem budget constraint and combine everything together to Xo + Vo r k-^E -g + du Jo A- k-^< can be determined ^ (1 > The 1. + (1 + -— + /J-) /1-) as: r -' -; • r e-'^'du and we have that F < l,r k J r which shows that us return to the time let get: e / depend (1 final step of + A'o h') the proof is + r - g to use the formula for E [e '"^] from section 4.2. Remark 7 The optimal portfolios can much smaller 7.5 We be derived in a manner similar to sectioii ^.2. Their magnitude is in this case. Propositions and Proofs for section 4.3 give a sketch of the claims. section 3.1. We = Formal proofs would proceed along the lines of the respective proofs in focus only on pt insurance for simplicity. In this case the Bellman Equation becomes: (c*)^-^ max{— c" 1 — 1 Vxc'} + max{-a p -y p'V'xx 2 + ^Xs^p} 1— -rV + VxirX + y') + Vyv' g Once again the optimal consumption and + K/z + Ja^^/,, + A(,., portfolio policies are given c* = V^ 50 ^ C 1-7 by (assuming a V C^— Value function): and - p= Notice that the portfoho strategy to risk is Vx? a "pure" hedging strategy, in the sense that there are no demands due premia (we assumed them away by positing that the sources of consumption growth). In variance motives. where C'^{st) this way we risk are imcorrelated are able to distill out the pure hedging component, excluding As before we conjecture a value function mean- of the form: the same boundary conditions as in the no-hedging case (section satisfies with aggregate 3.2.). Under this conjecture the optimal portfolio process becomes: Notice that the sign and magnitude of hedging One can show i.e. that this they involve short non-hnear ODE influenced by the derivative of C(sj) with respect to Plugging back into the value function and simphfying we get a second order for C{st). _ r^C" + Cf ,< + Cif ia^ + ia^illl^^ + A(.,) [C - C"] ^ defining CH we St- a decreasing function and thus not surprisingly the hedging demands are negative, is sales. 7C"^ Now is .H\l [b") get r^{b"v + ^^i{b"y-'h^ + -:^i{b"y-' 1 ,(l-7)7^(6f)M^^]''"' + X{st) [A (7 + H i)(S b" b: - {b"r] -r{b")-y or after simplifying: C b" 2 Notice that once again this ODE is [{b"V 7 very similar to the ODE with no hedging, namely: \2 Q = l-rb + bsii + up to the absence of the term ^0-^(7 — portfolio pohcies a (7-l)%^0. 2 l)-^-|^. In 7 1 b~< terms of the 6— function the optimal consumption and become: 51 Vt Lemma 6 Under perfect hedging "excess'' consumption, ct - Kyt = i.e. c*., t < r follows the process: dcT ^ c / A(5t) dt C{st Proof. The proof proceeds in a virtually identical way 52 as in section 4.2. 4^00 O ^f .J ij HOV 1 1 Z003 Date Due MIT LIBRARIES 3 9080 02613 2461