IMF Bailouts and Moral Hazard *

Transcription

1 IMF Balouts and Moral Hazard * Jong-Wha Lee Korea Unversty and The Australan Natonal Unversty and Kwanho Shn Korea Unversty and Claremont McKenna College March 2005 Abstract Ths paper emprcally nvestgates the extent of nvestor moral hazard assocated wth IMF balouts by analyzng the responses of soveregn bond spreads to the changes n the perceved probablty of IMF balouts of countres undergong fnancal crss. We do not fnd strong evdence that the extent of nvestor moral hazard changed after the non-balout of Russa n August 1998 that sgnaled a modfcaton to IMF nterventon polcy. In contrast, we fnd evdence that nvestor moral hazard s ntensfed for those countres that have stronger poltcal connectons to the IMF and that are thereby more lkely to be baled out by the IMF. Ths pattern prevaled even after the Russan crss. Keywords: IMF, moral hazard, soveregn bond spreads, nternatonal fnancal archtecture JEL No.: F33, F34 * The authors are grateful to Gordon de Brouwer, Shumpe Takemor, Thomas Wllett and semnar partcpants at Claremont McKenna College, Korea Unversty, The 2004 East Asan Economc Assocaton Meetng, and Tokyo Unversty for helpful comments. We would lke to thank Jeromn Zettelmeyer for sharng hs data.

2 I. Introducton Over the last decade, the sze and frequency of nternatonal fnancal dsturbances has contnued to ncrease and the role of the IMF as a crss manager has expanded to meet these challenges. The IMF has moblzed ever-larger rescue packages and become more deeply nvolved n domestc macroeconomc polces and structural reform n developng countres. Whle the IMF s lendng durng the crses helps to avod crashes, t s often clamed by crtcs that the Fund s role of a quas-lender of last resort through ts repeated nterventon wth large-scale fnancal support packages n the nternatonal fnancal markets creates moral hazard. The prospect of future IMF balouts allows nvestors to lend excessvely to member countres at nterest rates that do not adequately reflect underlyng rsks and encourages borrowers to behave n mprudent ways. The purpose of ths paper s to assess the extent of moral hazard assocated wth IMF balouts. To date, there have only been few formal emprcal studes on the mpact of IMF balouts on moral hazard and ther conclusons have been at best mxed (see the survey n secton 2). Ths paper attempts to nvestgate the presence of nvestor moral hazard by emprcally examnng whether IMF balouts encourage excessve rsk takng by nvestors. Our study bulds on prevous emprcal lterature, and n partcular extends the framework of Dell Arcca et al. (2002). Dell Arcca et al. focus on events such as the Russan default that supposedly sgnaled major changes n the polcy stances of the nternatonal fnancal nsttutons towards provdng loans to crss countres. If these events occur exogenously, then any change n the behavor of nvestors, reflected n 1

3 nterests and bond spreads due to changes n perceved probablty of balout of crss countres, wll ndcate whether moral hazard becomes aggravated or mtgated followng the events. Dell Arcca et al. fnd some mxed evdence; namely that moral hazard prevaled before the Russan crss but not n the relaton wth other events such as the Mexcan and Asan crses. An mplct assumpton made by Dell Arcca et al. and n most other studes s that nvestor moral hazard s a phenomenon takng place across all emergng countres. We beleve, however, that the degree of prevalng moral hazard mght dffer across countres dependng on country specfcs. Especally, f a selected group of countres are more lkely to be baled out by the IMF than other countres, then the moral hazard and any adverse effects from t would be concentrated wthn the group. In ths paper we propose another test procedure that delves nto ths possblty. We fnd that the degree of nvestor moral hazard actually dffers across countres. Especally, we fnd evdence that nvestor moral hazard s ntensfed on the countres that are more lkely to be baled out by the IMF. Whle the extent of nvestor moral hazard could change across all countres followng the events that sgnal sgnfcant overall IMF polcy modfcatons, the expected sze of IMF balouts determned by a member country s poltcal connectons to the IMF and by varous country characterstcs plays an mportant role as well. The remander of the paper s organzed as follows. In secton II, we defne the concept of moral hazard trggered by the IMF and survey the prevous emprcal studes on t. Secton III ntroduces the test procedure and apples t to detect prevalng nvestor moral hazard across countres. Concludng remarks follow n Secton IV. 2

4 II. IMF Lendng and Moral hazard 1. The concept of moral hazard Moral hazard s a term that orgnates from analyzng the effects of nsurance. A typcal stuaton s that when a person nsures an asset, the nsured may have less ncentve to mantan the asset properly. Ths s because nsurance reduces the ncentves for the nsured party to take preventve actons. Typcally, the behavor of the nsured s unobservable by the nsurance company or too dffcult to contract on drectly. In recent dscussons on the role of the IMF and reforms toward a new nternatonal fnancal archtecture, moral hazard has been a crtcal ssue. The queston s whether the avalablty of fnancal rescue from the nternatonal fnancal nsttutons lke the IMF n the stuaton of crses encourages lenders and borrowers to behave rresponsbly n ways that may make a crss more lkely. To the extent that the prospect of future IMF balouts s lke the provson of nsurance to both borrowers and lenders, t reduces ther motvatons to take preventve actons. Antcpatng that the perceved default rsk assocated wth nternatonal lendng s dmnshed, nvestors are lkely to lend excessvely to member countres at nterest rates that do not adequately reflect the underlyng rsks. On the other hand, the expectaton of IMF s offcal rescue encourages debtors to behave n mprudent ways. Excessve nternatonal captal flows and mprudent domestc polces wll lkely ncrease the 3

5 probablty of a crss. An IMF loan dffers from an nsurance contract n several aspects. Insurance contracts render permanent transfer of funds from the nsurer to the nsured on realzaton of the rsk. In contrast, IMF fnancal support s not a permanent transfer and comes as a loan to be repad wth nterest. Nevertheless, even f the transfer s temporary, the nterest rate of IMF subsdzed loans s much lower than the rate at whch a country could borrow from prvate captal markets durng a crss. Hence, the nsured beneft from the IMF lendng mght be suffcently substantal to ental debtor moral hazard. From the perspectve of nvestors, t s clear that IMF s lmted nterventon does not provde a complete guarantee of the debt servce. Nevertheless, the ncreased frequency and sze of IMF fnancal support n recent years seem to ndcate there exsts a sgnfcant dstorton of ncentves to nvestors, whch can cause credtor moral hazard. Unlke nsurance contracts, IMF lendng helps to mtgate the real hazard of a crss. The presence of short-term lqudty support from the IMF n the event of a lqudty crss can help to reduce the probablty of runs on a country s nternatonal reserve or currency. The IMF-supported programs mght reduce the severty, such as output losses, of a fnancal crss, and thus lead to mproved prospects for honorng debt contract. However, by mtgatng the real hazard of a crss, IMF lendng aggravates nvestor moral hazard. A decrease n perceved rsk can lead nvestors to undertake rsker and largerscale lendng. When IMF lendng decreases real hazard and smultaneously ncreases moral hazard, separaton between the two effects becomes a crtcal ssue. For example, the 4

6 presence of IMF fnancal support can ncrease nternatonal lendng to borrowers through a decrease n real hazard of a crss or an ncrease n moral hazard, or both. Ths complcaton makes t dffcult to assess the net beneft of IMF nterventon (Rogoff, 2002). The role of IMF lendng on decreasng real hazard s clearly welfare-mprovng. If real hazard benefts outwegh moral hazard costs, then moral hazard n IMF loans would not be a bg concern (Jeanne and Zettelmeyer, 2001). 2. Prevous emprcal lterature The central ssue s whether the moral hazard supposedly caused by the IMF s nterventon s actually present and, f so, s quanttatvely mportant. Kenneth Rogoff, the former drector of the IMF s Research Department, asks where s the emprcal evdence that moral hazard n IMF lendng s mportant, at least n ths crude form? (Rogoff, 2002) The nvestgatons nto whether IMF fnancng leads to moral hazard encounter a number of dffcultes. Frst, t s hard to measure moral hazard precsely. In general, we cannot drectly observe the extent of excessve rsk-takng behavor of credtors and debtors that are nduced by the IMF nterventon. Second, t s dffcult to separate the effects of IMF nterventon from those of other factors. IMF nterventon tself s an endogenous choce that depends on economc and poltcal crcumstances surroundng the global communty as well as member countres. For example, the ncreased frequences of IMF nterventon, rather than beng ntroduced exogenously, can be a response to an ncreased ncdence of crss. In order to dsentangle the effects of IMF programs from those of other factors, we 5

7 have to compare the outcome wth the presence of IMF support to the counterfactual event that would otherwse have occurred. It s dffcult conceptually and practcally to construct ths counterfactual result. Consderng these dffcultes n the emprcal practces, t s not surprsng that there have been few systematc emprcal studes. Recently, there have been new developments of strateges to detect moral hazard. 1 A number of recent emprcal studes n ths area, ncludng Zhang (1999), Lane and Phlps (2000), Kamn (2002), and Dell Arcca et al. (2002), nvestgate whether expectatons of IMF nterventon nfluence nterest rates and bond spreads - whch are measures of the change n perceved nvestor rsk. The hypothess s that an IMF loan decreases the downsde rsks of default and thus encourages nvestors reckless lendng. Therefore, the ncreased IMF nterventon should lower the equlbrum cost of borrowng between debtors and credtors. In addton, most recent studes focus on partcular IMF (non)nterventon events to get around the endogenety of the IMF program. These studes focus on whether an exogenous change n IMF s lendng practce nduces changes n moral hazard behavor. The moral hazard behavor s easest to detect when there are exogenous changes n the ncentve schemes. Zhang (1999) examnes f the Mexcan IMF packages n 1995 aggravated nvestor moral hazard. Snce the spreads on emergng market bonds were actually lowered after the Mexcan IMF packages, t was argued that nvestor behavor became less prudent because of rased expectaton of balouts for crss countres n the future. Zhang formally tested ths hypothess by settng up an equaton for bond spreads based on fundamentals 1 See Dreher (2004) for a broad survey. 6

8 and nternatonal captal market condtons and ncludng a post-mexcan dummy as an addtonal explanatory varable. He fnds that the coeffcent on the dummy varable s postve and nsgnfcant and nterprets that the observed declne n spreads s a reacton to the ncreased lqudty n nternatonal captal markets and mproved fundamentals rather than beng a consequence of moral hazard. Smlarly, by comparng bond spreads over recent years after 1998 wth those pror to the Mexcan crss, Kamn (2002) also fnds that moral hazard has not been present n recent years. Lane and Phlps (2000) broaden the cases and consder the behavor of spreads around the tme of 22 IMF nterventons n the 1990s. They fnd that few of these cases led to a sgnfcant decrease n spreads, whch s consdered as evdence aganst the presence of moral hazard. On the other hand, there are a few studes that may support the presence of nvestor moral hazard. Echengreen and Mody (2001), based on a huge number of prmary-market bond spreads between 1991 and 1999, fnd that the presence of an IMF program sgnfcantly ncreases the probablty of bond ssuance and lowers the spreads of the country wth the program. They nterpret ths as evdence for the "catalyzng" effect of IMF programs because nvestors perceve conclusons of IMF programs as a commtment for reforms, and also suggest t can be reconcled wth the nvestor moral hazard vew. 2 Dell Arcca et al. (2002) also examne f the IMF s nonbalout of Russa n August 1998 decreased nvestor moral hazard. They fnd evdence that soveregn bond spreads n 2 Mody and Sarava (2003) show further results that the effect of IMF programs on bond ssuance and spreads becomes nsgnfcant for the countres wth weak fundamentals such as export growth volatlty, reserve-tomport rato and debt-to-gdp rato, or whch have been n IMF programs for a number of years. Ths pattern of IMF program effects does not seem to match well wth the moral hazard vew. The endogenety problem of IMF programs seems to preval more sgnfcantly n these studes whch focus on all IMF program epsodes nstead of a more-exogenously-held event. See the dscussons below. 7

9 emergng markets have rsen n 1999 and 2000 compared to those n the pre-russan crss perod, and conclude that nvestor moral hazard has been mtgated after the Russan debt crss. 3 In general, however, tests based on market spreads are subject to dentfcaton problems. As ponted out above, there s an ntrnsc dffculty of dsentanglng the effects of the IMF nterventons n queston from those of other crcumstances. For example, the IMF s balouts of East Asa n 1997 probably came together wth the general reassessments of rsks n emergng market economes, as nvestors realzed that even countres wth strong macroeconomc fundamentals were vulnerable to large-scale captal account crses due to nvestor panc or crss contagon. Hence, the perceved rsk must have ncreased, whch tended to offset the reducton of market spread due to moral hazard. In addton, IMF nterventons mght help to mtgate the real hazard of a crss. If ths were the case, IMF supports encourage nvestors to lend more and debtors to borrow more, whch could make t hard to solate the effect of credtor moral hazard on market spreads. Moreover, the change n market spreads mght reflect both the changes n nvestor moral hazard and debtor moral hazard. Changes n borrower moral hazard are n prncple controlled by the change n country fundamentals. The expectaton of fnancal rescue by the IMF dscourages debtors from behavng prudently, and s thereby lkely to deterorate country fundamentals. It s not clear that the emprcal tests control all mportant country fundamentals n the specfcaton. 3 Dell Arcca et al. s approach s dstngushed from most other studes n that they focus on the senstvty of spreads wth respect to fundamentals rather than the level of spreads. See Secton III for a more detaled explanaton of ther approach. 8

10 A few studes such as Kamn and von Klest (1999) and Dell Arcca et al. (2002) attempt to solate the nvestor moral hazard from other factors n nfluencng market spreads by examnng changes n the senstvty of spreads wth respect to fundamentals, rather than changes n the level of spreads, before and after an IMF nterventon. An ncrease n nvestor moral hazard mples that nvestors pay less attenton to dfferences n country characterstcs, thereby leadng to smaller slope coeffcents on country fundamentals n the regressons for spreads. For example, Dell Arcca et al. (2002) fnd that the Russan default has ncreased the senstvty wth whch spreads react to fundamentals and conclude that nvestor moral hazard decreased after the Russan crss. Ths type of test seems to solate the effects of nvestor moral hazard from those of debtor moral hazard, by controllng the change n fundamentals and assessng the changes n slopes. In so far as mtgaton of real hazard s solely reflected n a general declne n spreads and mprovement n fundamentals, ths approach can also dsentangle the moral hazard effect from the real hazard effect of IMF nterventons. There are alternatve approaches focusng on varables other than nterest rates or market spreads. Haldane and Schebe (2003) consder the effect of IMF loans on market captalzaton of UK banks wth sgnfcant exposures to emergng markets. A change n market valuaton of the credtor banks s consdered to capture the (unexpected) change n prce ncentves for credtors to engage n future rsky lendng to emergng market economes that are expected to be subject to IMF nterventon. Regresson analyss shows that the market valuaton of U.K. banks responds postvely to major IMF loan packages. They nterpret ths fndng as evdence for the exstence of credtor moral hazard n the 9

11 sense that the ncrease n net worth of the banks can suggest an ncrease n potental rsktakng behavor, n response to large-scale IMF nterventons. Ga and Taylor (2003) nvestgate whether an ncrease n the avalablty of fnancng under an IMF-supported program nduced debtor behavor of rsk-takng. They test f IMF program partcpaton of member countres has vared wth new polcy measures such as Supplemental Reserve Faclty (SRF) and New Arrangements to Borrow (NAB) that were desgned to facltate fnancal rescues. They fnd that the ntroducton of NAB and SRF has sgnfcantly ncreased the use of IMF resources, partcularly by the debtor countres that are more systematcally mportant, that s, more susceptble to the rsk of contagon. Ths framework has an advantage of avodng the complextes posed by the use of asset prces by measurng debtor s behavor drectly. However, t s not clear that the ntroducton of NAB and SRF s purely an exogenous event. These fnancal facltes were ntroduced to provde larger-scale, short-term fnancng to mtgate the shock of a captal account crss. Then, debtors mght have realzed the ncrease n perceved rsk. That s, an ncrease n IMF program partcpaton followng the ntroducton of new polcy measures can be a result of ncreased real hazard, rather than debtor moral hazard. III. Emprcal Test for Moral Hazard In ths secton we ntroduce a new test procedure that bulds on the work of Dell Arcca et al. (2002). Unlke ther test statstcs that focus on overall moral hazard rrespectve of countres, our test procedure emphaszes that the degree of moral hazard prevalng mght 10

12 dffer across countres. 1. Setup of the model We follow a setup smlar to that used n Dell Arcca et al. that nvestgates how the exstence of nternatonal fnancal nsttutons aggravates nvestor moral hazard. There are a number of rsk-neutral lenders who make loans to debtor countres. There are only two states for the debtor countres: they suffer from a crss or they do not. The probablty of a crss for country s denoted as θ ( x ) whch s expressed as a functon of a vector of observable country-specfc fundamentals, x. If a crss occurs, a country can default wth probablty, (1 ). In other words, the lenders can recover the debt wth the λ recovery rate, λ, even after the occurrence of the crss. Ths s because the nternatonal fnancal nsttuton rescues the crss country and pays back the debt on behalf of the crss country. 4 Note that we assume that the recovery rate vares across countres. The ntroducton of nsurance provded by the nternatonal fnancal nsttuton s denoted as z. The provson of more nsurance can be reflected n the setup through three dfferent channels. Frst, t mght affect observable fundamentals,.e., x x (z), and = hence ndrectly affect the crss probablty,.e., θ ( ( z)). Ths channel represents x debtor-country moral hazard n that as more nsurance s provded, the debtor country may take more rsky actons that deterorate fundamentals and hence ndrectly rase the 4 Ths s a strong assumpton because nternatonal rescue packages do not nvarably nvolve the balout of the nternatonal nvestors. It was the tradton that the IMF dd not lend to countres that were n default or arrears to ther prvate nvestors. However the IMF changed ts practces n the md-1980s to formally adopt a polcy that explctly allowed t to lend to countres n arrears. 11

13 probablty of the crss. The second channel s a drect nfluence of the nsurance on the probablty of a fnancal crss,.e., θ ( x ( z), z). Ths drect channel s nterpreted as real crss reducton due to the benefcal role of nternatonal crss lendng by preventng or mtgatng a crss. For example, as more nsurance s provded, ths ncreased safety net mght reduce the probablty of self-fulfllng runs on a country s debt or currency. Alternatvely, the nternatonal lendng may provde the vulnerable country wth the hard currency necessary to mplement the domestc fnancal safety nets or prevent bank runs. Then the probablty of a crss alters ndependently of the changes n fundamentals. 5 The thrd channel s through ts nfluence on the recovery rate. As the nternatonal fnancal nsttuton tends to bal out the crss country more frequently, the probablty of perceved recovery rate rses,.e., λ (z) ncreases. Whle nvestor moral hazard refers to partcular nvestor actons such as an ncrease n rsky lendng or a reducton n montorng, data on nvestor actons are typcally not avalable. However, snce the ncreased recovery rate would have precsely ths effect n the sense that t allows the nvestors to make more rsky nvestments, ths channel s often used to detect nvestor moral hazard. Gven ths setup, we can easly relate the crucal parameters to the spreads of the rsky lendng rate of the debtor countres over the rsk-free rate. Then, by analyzng how an ntroducton of more or less nsurance provded by the nternatonal nsttuton affects these spreads, we can ndrectly nfer the sense of moral hazard occurrng to ether nvestors or debtor countres. 5 Alternatvely we could assume that the lessenng of real hazard occurs through mprovement of fundamentals, n whch case the dstncton between debtor-country moral hazard and real hazard can be made by notng f fundamentals deterorate or mprove. Snce we focus on nvestor moral hazard, we can safely gnore ths alternatve channel that may unnecessarly complcate the analyses. 12

14 For the rsk-neutral nvestors, the ex ante gross lendng rate to country, R, s determned such that expected repayment equals the rsk free rate: R f R = (1) 1 θ (1 λ ) where f R s the gross rsk-free nterest rate. Then, the respectve spread over the rskfree rate for country s: s ln R ln R f θ (1 λ ) (2) where we make approprate approxmatons, ln (1+ε) = ε. Equaton (2) shows that the spread s equal to the crss probablty, θ, multpled by the perceved default rate, 1- λ. Thus, spreads of the lendng rate over the rsk-free rate depend on two factors: the crss probablty and the recovery rate. These two factors are also related to the degree of moral hazard occurrng to debtor countres and nvestors. Snce changes n real hazard are also related to the crss probablty, under our assumptons, we focus on the behavor of the recovery rate so as to assess the degree of nvestor moral hazard. Concentratng on the drecton of changes n spreads, however, cannot unambguously determne the presence of nvestor moral hazard. For example, f the degree of nvestor moral hazard ncreases (that s, λ ncreases), spreads decrease, reflectng that the rsk premum requred to nvestors decreases as the perceved lkelhood of the nternatonal rescue ncreases. At the same tme, f the degree of debtor-country moral hazard ncreases ( θ ncreases), spreads ncrease, reflectng that the rsk premum ncreases as the crss probablty ncreases. In an extreme case, the level of spreads does 13

15 not change f moral hazards that are greatly renforced on both sdes are exactly cancelled out. However, f there s no drect effect of nternatonal crss lendng on the real hazard of default, the drecton of changes n the level of spreads can provde a clue to the qualtatve answer. For example, f the level of spreads ncreases, we can assure that the debtor country moral hazard domnates nvestor moral hazard and vce versa. However, even n ths case, we cannot estmate the quanttatve ncrease n the degree of moral hazard on ether sde because the alteraton n spreads reflects the dfferent degrees of moral hazards that are smultaneously workng n the opposte drectons. In order to dsentangle the factor of nvestor moral hazard, we note that n equaton (2) the senstvty of spreads to the crss probablty depends on the degree of nvestor moral hazard. As the degree of nvestor moral hazard ncreases (that s, λ ncreases), the coeffcent of the default probablty decreases. Ths s the dea pursued by Dell Arcca et al. (2002) n focusng on the slope coeffcent nstead of the level of spreads n order to determne whether nvestor moral hazard ncreases or not. Snce the slope coeffcent s drectly related to the recovery rate parameter (λ ), ts estmaton can provde the quanttatve estmate for the ncrease n the degree of nvestor moral hazard. Ths can be shown by expressng equaton (2) as follows: s = x β + u (3) where the error term, u, s explctly added and θ s assumed to take a lnear form and be mngled nto β. 6 The vector of observable fundamentals, x, appears drectly and 6 A constant term s suppressed for the expostonal convenence. 14

16 reflects the rsk factors that nfluence the crss probablty, θ. An ncrease n x s supposed to ncrease the rsk factors. The slope parameter, β, s assumed to reflect the degree of nvestor moral hazard that nfluences the perceved default rate, 1- λ. Dell Arcca et al. s strategy s to estmate equaton (3) before and after an event that changes the way nternatonal crss lendng s provded. If ths change leads to a dfferent degree of nvestor moral hazard, bascally the slope parameter, β, wll reflect ths change. By estmatng the slope parameter, they argue that any change n the degree of nvestor moral hazard can be detected. They take the Russan default as an event that sgnals that the perceved lkelhood of future nternatonal crss lendng s lessened, whle the Mexcan and Asan crses are taken as events that sgnal the opposte. They consder an ncrease n β after the Russan crss, that s, spreads becomng more senstve to rsk factors after less nsurance s provded by the nternatonal fnancal nsttuton, as evdence of nvestor moral hazard prevalng before the crss. They also consder a decrease n β after the Mexcan and Asan crses as evdence of nvestor moral hazard prevalng more afterwards. Ther actual emprcal results show evdence of nvestor moral hazard n the event of the Russan crss but not n that of the Mexcan and Asan crses. We extend the basc setup of equaton (3) by explctly allowng the recovery rate to vary across countres. As wll be explaned n detal n the next subsecton, the lkelhood of recevng an IMF loan hnges on the member country s poltcal connectons to the U.S. and other major shareholdng countres of the IMF, as well as on varous country characterstcs. We wll show that ntroducng the varyng recovery rate greatly enhances 15

17 the ft of the equaton. 7 More mportantly, relaxng the assumpton of the constant recovery rate allows a new method of testng nvestor moral hazard. As the approval rate ncreases across dfferent countres, snce balout s more lkely, nvestors should be less senstve to the rsk factors f nvestor moral hazard prevals. Then nvestor moral hazard s detected by examnng whether the slope coeffcent of spreads also changes n lne wth the varyng recovery rate. In other words, we can be assured that nvestor moral hazard prevals f the slope coeffcent for countres more lkely to be baled out s lower than that for countres less lkely to be baled out. In fact, the dfferent recovery rate across countres s analogous to ntroducng events alterng the overall lkelhood of lendng provded by the nternatonal fnancal nsttuton. Whle exstng studes ncludng Dell Arcca et al. focus on an event that would alter the slope coeffcent of all countres by changng the overall lkelhood of lendng, our test nvestgates f the slope coeffcent vares across countres as the lkelhood of lendng dffers across countres. Our slope test procedure s especally useful, as we can analyze the dataset for a relatvely short tme perod that does not correspond to the event that alters the general atmosphere of the nternatonal lendng. Furthermore, we do not need to worry about the possblty that the event may affect overall real hazard as well. Our test procedure can also mmc exactly that used by Dell Arcca et al. For example, after an event alters the overall nsurance provded by the nternatonal fnancal 7 Dell Arcca et al. actually show that even f they allow the recovery rate to depend on the fundamentals, under some crcumstances, ther proposton stll follows. However, n our setup we allow the recovery rate to depend not only on the fundamentals but also on poltcal connectons to the IMF. 16

18 nsttuton, the slope coeffcent of every country changes. By comparng the slope coeffcent for every country across before and after the event, we can also assess f the event affects the degree of nvestor moral hazard or not. Our test procedure starts wth the followng modfcaton of equaton (3): 8 s β + u (4) = x ( 0 + β1z ) + u = x β 0 + x β1z where the recovery rate s explctly modeled to vary as z changes across countres. The parameter z measures the lkelhood of IMF balouts for each country gven the country fundamentals. We derve the proxy for based on the recent work by Barro and Lee (2004) that nvestgates how a country s poltcal connectons to the IMF as well as other characterstcs affect the probablty and sze of IMF lendng. The parameter,, wll be defned so that a country wth hgher s more lkely to obtan IMF loans. Then the evdence of nvestor moral hazard s detected by focusng on the slope coeffcent of x. z z z For example, f nvestor moral hazard s aggravated, then the slope coeffcent of x wll be lowered n absolute value or equvalently β 1 wll take a dfferent sgn from β 0 so that the slope coeffcent, β 0 + β 1 z, decreases n absolute value (.e., s less responsve to the country fundamentals) as z rses. 9 The test of Dell Arcca et al. also mples that the 8 The model can be extended by allowng that the lkelhood of IMF balouts can have a drect effect on the level of spread, ndependently from the effect on the slope coeffcent. The specfcaton can be expressed as, s = x β 0 + x β1z + δz + u (4) The man emprcal results reported below hold true n ths alternatve specfcaton. 9 Whle t s desrable to defne all the rsk factor varables, x, consstently n such a way that a hgher value mples a hgher rsk, some rsk factor varables we use n the actual estmaton are defned as they change nversely wth the rsk. In ether case, however, a lower coeffcent n absolute value mples less senstvty to the rsk factor. 17

19 estmate of β 0 + β 1 z, becomes smaller followng an event that ncreases overall nsurance provded by the nternatonal fnancal nsttuton. 2. IMF Governance and Lendng Decson To ntroduce the varyng recovery rate, t s essental to analyze the IMF decson process to provde loans to crss countres. In fact, the decson on partcpaton n an IMF program can be determned endogenously by varous factors. 10 The partcpaton n an IMF program s a jont decson between a partcular country and the IMF whch requres the agreement of both. On the demand sde, a country seeks IMF fnancal assstance when t faces certan economc dffcultes. For example, a currency crss nduces countres to seek IMF fnancal support. On the supply sde, the decson of the IMF to provde loans depends on the current economc stuaton of the country and ts past economc performance. The IMF evaluates the country economc crcumstances n order to determne f a lendng program s warranted. The IMF also evaluates whether the country s commtted to undertake any condtons that the IMF wll mpose. Barro and Lee (2004) ntroduce another poltcal-economy dmenson of IMF decson-makng, and show that a country s poltcal connectons to the IMF affect the probablty and sze of IMF lendng. They proxy the poltcal connectons by several nsttutonal and geopoltcal varables: the sze of the country s quota at the IMF, the sze 10 See Brd and Rowlands (2001), Brd et al. (2004) and Barro and Lee (2004) for determnants of IMF loans. 18

20 of the natonal staff at the IMF, and each member country s economc and poltcal proxmty to the major shareholdng countres of the IMF, ncludng the Unted States, France, Germany and the Unted Kngdom. The proxmty varables are constructed based on patterns of votng n the U.N. General Assembly and blateral trade flows. Table 1 reports the estmaton result, whch s reproduced from the fndng of Barro and Lee (2004, Table 3). The dependent varable s the rato of approved IMF loans to GDP for country, averaged over the fve-year perod , , , , or By takng account of the censorng of the dependent varable at zero, the Tobt model s appled to panel data for 130 countres over the fve fve-year perods from 1975 to The explanatory varables nclude country-specfc economc factors that nfluence the exstence and sze of IMF loan programs such as the rato of foregn reserves to mports, per capta GDP, total GDP, and per capta GDP growth rate. They are the values at the begnnng of the perod or an average over the prevous fve-year perod. In addton, the regresson ncludes the nsttutonal and geopoltcal factors that measure each country s poltcal-economy connectons to the IMF. The regresson also ncludes perod dummes and a dummy for OECD membershp. The regresson result n Table 1 shows the sgnfcant nfluence of economc factors, as well as poltcal and nsttutonal factors, over the sze of IMF lendng. All of the poltcal-economy varables are jontly statstcally sgnfcant. More detaled dscusson of the results s avalable n Barro and Lee (2004). Based on the estmaton result, we can construct the predcted value of the loansze for the fve-year perod from 1995 to Table 2 reports the estmate for the predcted loan sze based on the regresson result n Table 1. Ths varable s constructed for 19

21 the 16 EMBI Global countres (EMBIG) ncluded n our data set. 11 Venezuela (1.23), Mexco (1.18), South Afrca (1.15), Phlppnes (1.08) and Panama (1.07) comprse the group of the hghest fve countres, whle Korea (0.48), Chna (0.66), Thaland (0.69), Argentna (0.73) and Brazl (0.78) are the lowest fve countres. 3. Emprcal test for nvestor moral hazard based on spreads In the lterature, two dfferent sources for bond spreads are wdely used: the launch spreads contaned n Captal Data s Bondware dataset and the secondary-market spreads ncluded n J.P. Morgan s Emergng Markets Bond Index (EMBI). Whle Captal Data s Bondware dataset covers broadly 54 emergng countres, t contans prmary spreads that are observed only at the tme of ssue. Ths creates a complcated problem because t consttutes a hghly unbalanced panel. Snce the decson of ssuance s lkely to depend on the factors that nfluence the level of spreads as well, a smple OLS estmaton of spreads s vulnerable to a selecton bas. 12 On the other hand, the EMBI or the EMBI+ datasets cover only a small number of countres. A more recent dataset, EMBI Global, covers a much broader set of 18 emergng countres, but only for a much shorter perod startng from January Ths dataset has an mportant advantage, however, because t conssts of a balanced panel based on secondary market spreads at daly frequency. The nstruments are manly Brady bonds 11 Among the 18 countres consdered n Dell Arcca et al., Bulgara and Croata are not ncluded because some varables necessary to calculate the predcted values are not avalable. 12 Echengreen and Mody (2004) get around ths problem by employng a two-step equaton of a sample selecton model. 20

22 and Eurobonds, but a small number of trade loans and local market nstruments are contaned n the calculaton of the weghted average ndex. To avod the complcated econometrc ssue, we have decded to choose the EMBI Global dataset for our emprcal analyses. The major event durng the coverage of the dataset s the Russan crss. Whle Russa was wdely beleved to be too bg to fal, somewhat surprsngly, the nternatonal communty dd not prevent the default. Ths can be nterpreted to sgnal that the polcy of nternatonal fnancal nsttutons changed to provde less support to crss countres. A number of prevous studes ncludng Dell Arcca et al. (2002) focus on ths changng behavor of nvestors to compare moral hazard before and after the crss. Followng the Russan crss, snce a perod of prolonged turbulence contnued for a whle, the perod mmedate after the crss should be excluded from the post-crss perod for a far comparson wth the pre-crss perod. Otherwse, the estmated coeffcents that are supposedly derved from a statc model of stable relatonshp between spreads and fundamentals tend to be contamnated by the hghly unstable turbulence perod around the tme of the crss. One dffculty wth ths approach, however, s that there s no objectve way to determne the excluson perod and an error n ths determnaton may generate a based result. To compare our emprcal analyses wth those of Dell Arcca et al., we follow ther choce of the pre-crss perod from January to June 1998 and the post-crss perod from Aprl 1999 to December The data frequency used s monthly. Before we present the results based on the test procedure devsed n ths paper, Table 3 shows pre- and post-russan crss regresson results based on equaton (3) that were 21

23 utlzed by Dell Arcca et al. We choose the control varables that are used n the most preferred specfcaton n Dell Arcca et al. 13 The results under Sample 1 are based on the whole set of 18 emergng countres and are dentcal to those reported n Dell Arcca et al. Snce our new test procedure requres the Barro-Lee ndex that s used as a proxy for, our sample dscards two countres, Bulgara and Croata, due to a lack of data nvolved wth constructon of the Barro-Lee ndex. For the comparson later, the results under Sample 2 z are based on the set of 16 emergng countres only. However the two results are very smlar and hence our explanatons below can be appled to both results. When the slope coeffcents of the rsk factors are compared between pre- and postcrss regresson results, although the magntude of the change s not always statstcally sgnfcant, the absolute value of most of the coeffcents ncreases. Dell Arcca et al. (2002) take ths fndng as supportng evdence for the prevalng of nvestor moral hazard before the Russan crss. However, the coeffcents of some varables have the wrong sgn, whch makes the nterpretaton of ther test results dffcult. For example, the coeffcent of current account surplus s postve, whch mples that a country s spread wll ncrease f ts current account surplus ncreases. However, f the crss probablty s nversely related to the volume of current account surplus, the coeffcent should be negatve. Another example s the coeffcent of real credt growth, whch s estmated to be negatve n contradcton to the expectaton. Now we swtch to our preferred specfcaton of equaton (4) n whch dfferences n the approval rate across countres are explctly consdered. Table 4 reports the 13 See table 5 n Dell Arcca et al. (2002). They selected ths specfcaton through a general-to-specfc procedure usng a rch rght-hand sde dataset. 22

24 estmaton results. For each rsk factor varable, the coeffcent of the nteracton term wth the predcted loan-sze s reported below t. In both columns of table 4, reportng pre- and post-crss results, the Barro-Lee ndex s found to be qute relevant n determnng spreads snce, n most cases, the coeffcents of the nteracton terms are statstcally very sgnfcant. Further, the coeffcents of the varables wth the wrong sgn are converted nto the rght one;.e., the coeffcent of current account surplus becomes negatve and that of real credt growth, postve. Thus the overall ft of the specfcaton mproves by addng up the nteractve terms wth the predcted loan-sze varable. Ths s also confrmed by the fact that 2 R s hgher than before. In the new specfcaton, nvestor moral hazard s detected by assessng whether the slope coeffcent of rsk factors decreases n absolute value as z ncreases. In other words, f nvestors respond less senstvely to the rsk factors of a country that s more lkely to be baled out by the IMF, we can be assured that nvestor moral hazard prevals. Snce the slope coeffcent s the sum of the coeffcents of the rsk factors and the correspondng nteracton term, an equvalent way s to check f the coeffcent of the latter takes the opposte sgn to that of the former. As explaned, our method has an advantage n that t can be appled to each subsample perod separately. We frst examne the regresson results for the pre-crss perod. Column 1 depcts a general pattern that the coeffcents of nteracton terms take the opposte sgn from those of the correspondng rsk factor varables, consstent wth the asserton that nvestor moral hazard prevaled. The only excepton s the current account for whch the coeffcent of the nteracton term s statstcally nsgnfcant. 23

25 For the post-crss perod, whle relatvely more varables are not statstcally sgnfcant, the same pattern emerges of the opposte sgn between the coeffcents of the rsk factors and those of the correspondng nteracton terms. For example, real growth, fscal balance and poltcal nstablty are not statstcally sgnfcant. However, for other varables such as current account, real credt growth, and sze varables that are statstcally sgnfcant, we observe the same pattern as before. Hence, the overall evdence seems to support the hypothess that bond spreads of a country whch s more lkely to be baled out by the IMF tend to be less responsve to the country fundamentals. Thus, we can conclude that the moral hazard also prevaled even after the Russan crss. One of the man fndngs of Dell Arcca et al. s that moral hazard decreased n general after the Russan crss. Do our analyses support ths concluson? Ths ssue can be nvestgated by comparng the slope coeffcents before and after the crss. Snce the value of the slope coeffcents depends not only on the coeffcents of the rsk factors and the nteracton terms, but also the actual value of z, t s convenent to represent ths comparson for the range of the entre support of. When we compare the pre- and the post-crss perods for the rsk factor, t s not always clear whether nvestor moral hazard z has ncreased or not. For example, when we use the pont estmates n column (1), the slope coeffcents of real credt growth ndcate that the slope (that s, the response of spread to credt growth) becomes unformly larger n the entre support of z after the crss, demonstratng that nvestor moral hazard became less prevalng after the crss. However, for the other varables, the changes n the slopes are ambguous and depend on z the value of. 24

26 Overall, our evdence ndcates that the Russan default may not necessarly have decreased nvestor moral hazard. Instead, our evdence seems to suggest that, rrespectve of the tme perod, nvestor moral hazard prevals n those countres that are more lkely to be baled out by nternatonal fnancal nsttutons such as the IMF. V. Concludng Remarks Ever snce the large-scale IMF balouts for Mexco n 1994 and East Asa n , countless proposals for the new nternatonal fnancal archtecture have been put forward. 14 In lne wth ths, attenton has focused on how to change the role of the IMF to ncrease ts effectveness n managng fnancal crses and also on how to lmt the moral hazard assocated wth IMF balouts. The ncreasng ncdence of large-scale fnancal crses and the rapd global transmsson of each crss support the case for an nternatonal lender of last resort (ILLR) (Kndleberger 1989, Fscher 1999). The scale of captal outflows when a crss breaks out has contnued to ncrease n recent years, thereby substantally rasng the cost of each crss. The herd behavor exhbted by nvestors often trggers a severe lqudty crss n a country. In ths regard, an nternatonal nsttuton such as the IMF can play a desrable role by enablng an llqud but solvent country to survve and by stemmng the contagon of the crss to neghbor countres. However, expandng the role of the IMF as a stronger ILLR would undoubtedly 14 Refer to De Gregoro et al. (1999), Echengreen (1999) and Park and Wang (2001) among others for the dscusson of new nternatonal fnancal archtecture. 25

27 create a large moral hazard problem, rasng expectatons of more frequent nterventons by the IMF wth a larger scale of fnancal support than at present. Such an ncreased-scale of IMF-led balouts has certanly lead to excessve borrowng and lendng. A large IMF could become more actvely engaged n preventatve montorng and early supervson of fnancal actvty, whch mght help to reduce the problem of moral hazard but cannot completely elmnate t. Our emprcal fndng shows that nvestor moral hazard stll prevals, even after the Russan default that supposedly sgnaled a dfferent stance on IMF s role as an ILLR. Hence, nvestments n countres that are lkely to be excessve benefcares of the ILLR are subject to moral hazard behavor. Even f t were possble to succeed n reformng the general structure, the IMF would not be able to prevent ts major shareholders, lke the Unted States, from seekng to bal-out countres that have stronger poltcal connectons. Ths rases a concern that, as long as ths favortsm contnues, moral hazard may not be sgnfcantly mtgated. Therefore, when we dscuss the role of the IMF as an ILLR, the governance structure of the nsttuton, and n partcular the decson-makng procedure wth respect to the approval of rescue loans, should be crtcally examned. Otherwse, any attempt to promote the role of the IMF as an ILLR along the lnes of current polcy prescrptons s merely lkely to heghten the moral hazard problem nherent n the system. Prvate-sector nvolvement (PSI) has also been a hot ssue n the recent dscussons on the nternatonal fnancal archtecture. In order to reduce credtor moral hazard, there have been suggestons of balng-n the prvate sector, whch mples that nvestors should 26

28 bear part of the burden. One proposal nvolves empowerng the IMF to functon as a sort of nternatonal bankruptcy court whch can mpose payment standstlls (Krueger 2001). A payment standstll could compel credtors to act collectvely n ther best nterests, and thus help to avod a dsruptve asset-grabbng race (Sachs, 1995). If the payment standstll scheme proves vable, the government and the IMF can declare debt standstll as an alternatve to a large-scale loan for a country, partcularly n a lqudty crss. In the recent dscusson of PSI, ts facltaton of orderly debt restructurng n the event of a crss has been much emphaszed, whereas ts role n preventng moral hazard has receved lttle consderaton. PSI would help to mtgate the nvestor moral hazard assocated wth IMF balouts. However, the current proposals for PSI would actually ncrease the moral hazard behavor from the debtor sde. Our emprcal analyss focuses on moral hazard from the nvestor sde. It remans an nterestng and mportant ssue as to whether the expectaton of IMF balouts encourages borrowng countres, especally those closely connected to the IMF, to behave mprudently or not. We plan to nvestgate ths ssue n future research. 27

29 References Barro, R and Lee, J, 2004, 'IMF Programs: Who Is Chosen and What are the Effects?', Workng Paper, Harvard Unversty, forthcomng Journal of Monetary Economcs. Brd, G. and D. Rowlands, 2001, IMF Lendng: How Is It Affected by Economc, Poltcal and Insttutonal Factors, Journal of Polcy Reform, 4, Brd, G., M. Hussan and J. Joyce, 2004, Many Happy Returns? Recdvsm and IMF, Journal of Internatonal Money and Fnance, 23, De Gregoro, Jose, Barry Echengreen, Takatosh Ito, and Charles Wyplosz, 1999, An Independent and Accountable IMF. Geneva, Internatonal Center for Monetary and Bankng Studes and Center for Economc Polcy Research. Dell'Arcca, G, Schnabel, I and Zettelmeyer, J, 2002, 'Moral Hazard and Internatonal Crss Lendng: A Test', IMF Workng Paper No 02/181. Dreher, Axel, Does the IMF Cause Moral Hazard? A Crtcal Revew of the Evdence, Mmeo Echengreen, Barry, 1999, Toward a New Internatonal Fnancal Archtecture: A Practcal Post-Asa Agenda Washngton: Insttute for Internatonal Economcs. Echengreen, B. and Ashoka Mody, 2001, "Bal-Ins, Balouts, and Borrowng Costs, IMF Staff Papers 47: Echengreen, B. and Ashoka Mody, 2004, "Do Collectve Acton Clauses Rase Borrowng Costs?". Economc Journal 114, Fsher, Stanley, 1999, "On the Need for an Internatonal Lender of Last Resort," Journal of Economc Perspectves. Vol.13, Issue 4. 28

30 Ga, Prasanna and Ashley Taylor, 2003, "Internatonal Fnancal Rescues and Debtor Country Moral Hazard," Bank of England, mmeo. Haldane, A G and Schebe, J, 2003, 'IMF Lendng and Credtor Moral Hazard', Bank of England, mmeo. Haldane, A and Taylor, A, 2003, 'Moral Hazard : How Does IMF Lendng Affect Debtor and Credtor Incentves?', Fnancal Stablty Revew, June. Internatonal Monetary Fund "Collectve Acton Clauses: Recent Developments and Issues, prepared by the Internatonal Captal markets, Legal and Polcy Development and Revew Departments. Jeanne, O and Zettelmeyer, J, 2001, 'Internatonal Balouts, Moral Hazard and Condtonalty', Economc Polcy, October, 33, Kamn, S. B, 2002, 'Identfyng the Role of Moral Hazard n Internatonal Fnancal Markets', Federal Reserve Board Internatonal Fnance Dscusson Paper No Kamn, S B, and K. von Klest, 1999, 'The Evoluton and Determnants of Emergng Markets Credt Spreads n the 1990s,' BIS Workng Papers No. 68. Kndleberger, Charles P Manas, Pancs, and Crashes: A Hstory of Fnancal Crses. New York: Basc Books. Krueger, Anne O "Internatonal Fnancal Archtecture for 2002: A New Approach to Soveregn Debt Restructurng". Natonal Economsts' Club Annual Members' Dnner. Amercan Enterprse Insttute. (IMF homepage). Lane, P and Phllps, S, 2000, 'Does IMF Fnancng Result n Moral Hazard?', IMF 29

31 Workng Paper No Mody, Ashoka, and Dego Sarava, 2003, Catalyzng Prvate Captal Flows: Do IMF Programs Work as Commtment Devces, unpublshed manuscrpt. Park, Yung Chul and Yunjong Wang, 2002, "What Knd of Internatonal Fnancal Archtecture for an Integrated World Economy?", Asan Economc Papers 1(1), Rogoff, Kenneth, 2002, Moral Hazard n IMF Loans: How Bg a Concern? Fnance and Development 39(3), September. Sachs, Jeffrey, "Do We Need an Internatonal Lender of Last Resort?" mmeo., Harvard Unversty Zhang, X A, 1999, 'Testng for 'Moral Hazard' n Emergng Markets Lendng', Insttute of Internatonal Fnance Research Paper No