I M F S T A F F D I S C U S S I O N N O T E

Transcription

1 I M F S T A F F D I S C U S S I O N N O T E September 7, 2012 SDN/12/10 Multlateral Aspects of Managng the Captal Account Jonathan D. Ostry, Atsh R. Ghosh, and Anton Kornek I N T E R N A T I O N A L M O N E T A R Y F U N D

2 INTERNATIONAL MONETARY FUND Research Department Multlateral Aspects of Managng the Captal Account 1 Prepared by Jonathan D. Ostry, Atsh R. Ghosh, and Anton Kornek Authorzed for dstrbuton by Olver Blanchard September 7, 2012 DISCLAIMER: Ths Staff Dscusson Note represents the vews of the authors and does not necessarly represent IMF vews or IMF polcy. The vews expressed heren should be attrbuted to the authors and not to the IMF, ts Executve Board, or ts management. Staff Dscusson Notes are publshed to elct comments and to further debate. JEL Classfcaton Numbers: Keywords: F21, F23 Captal controls, multlateral aspects, rules of the road Author s E-mal Address: jostry@mf.org, aghosh@mf.org, akornek@umd.edu 1 We thank Olver Blanchard, Marcos Chamon, Raphael Espnoza, and Davd Lpton for helpful comments on earler versons of ths paper, and Jane Hazel for assstance.

3 2 Table of Contents Page Executve Summary...3 I. Introducton...4 II. The Pure Theory of Captal Controls...6 III. Survey of Emprcs...15 IV. Implcatons...18 V. Concluson...21 References...23 Boxes 1. Learnng-By-Dong Externaltes n Exportable Producton Equvalence of Captal Controls and Reserve Accumulaton Global Effcency of Correctng Domestc Externaltes Role for Coordnaton under Costly Captal Controls Gans from Coordnaton under Costly Captal Controls...13

4 3 EXECUTIVE SUMMARY The fnancal crss has agan brought home the profound fnancal lnkages across countres, often manfest n hghly volatle captal flows. Ths volatlty has prompted nterest n multlateral prncples to gude polces n both source and recpent countres. Ths note dscusses the analytcal underpnnngs of such prncples, and attempts to draw mplcatons from the analytcs for the desrablty of multlateral coordnaton of country polces. Whle a seres of IMF polcy notes and Board papers have dscussed how an ndvdual country, actng n solaton, should respond to a surge of captal nflows, less attenton has been pad to the multlateral consequences, and the desrablty of nternatonal cooperaton to acheve globally effcent outcomes. Three ssues are of potental concern. Frst s the possblty that captal controls may have the effect of vtatng external adjustment, for example when nflow controls are used to sustan an undervalued currency. Second, as each captal-recevng country that faces excessve captal flows seeks to reduce ts own nflows n support of domestc fnancal stablty, t deflects some captal towards other recpent countres, exacerbatng ther nflow problem. Thrd, polces n source countres (ncludng monetary polcy and prudental regulaton), to the degree they ncrease the volume or rskness of captal flows, may exacerbate problems faced by captal-recevng countres. Some of these consderatons have been taken up by us n earler papers; ths note takes the analytcs further, addng to the exstng foundaton for polcy advce. Ths note s not meant to prejudge the outcome of the IMF s ongong work to artculate a potental nsttutonal vew on captal flows, nor to gude the conduct of blateral or multlateral IMF survellance. We argue that the global welfare mplcatons of captal account regulatons, or polces that mmc ther effects, are threefold. Frst, spllovers from such polces do not necessarly have normatve mplcatons: f polces are justfed from a natonal standpont (n terms of reducng domestc dstortons), under a range of condtons they should be pursued even f they gve rse to cross-border spllovers. Second, however, f polces n one country exacerbate exstng dstortons n other countres, and t s costly for other countres to respond, then multlateral coordnaton of unlateral polces s lkely to be benefcal. Thrd, coordnaton may requre borrowers to reduce nflow controls or, much thorner, agreement by lenders to partally nternalze the rsks from excessvely large or rsky outflows. It s very dffcult to fully spell out the mplcatons of these consderatons n the form of multlateral rules, and ths paper refrans from dong so. Polcy advce mght nevertheless carefully consder stuatons where captal account regulatons seem unjustfed from a macroeconomc or fnancal-stablty standpont, but geared nstead to ganng unfar compettve advantage. Advce mght also rase red flags n stuatons where polces deflect flows excessvely across countres or transmt rsk from source to recpent countres. These consderatons pont n the drecton of usng nflow controls less ntensvely once spllovers are nternalzed, and toward source-country supervsory polces that nternalze the rsks wrought on recpent countres as a result of actons by domestc fnancal nsttutons.

5 4 I. INTRODUCTION The fnancal crss has agan brought home the profound fnancal lnkages across countres, often manfest n hghly volatle captal flows. Durng the 2008 global fnancal crss, captal flows to emergng market economes (EMEs), whch had peaked at $665 bllon n 2007, plummeted to less than $170 bllon n 2008, only to surge agan n 2010 as the global recovery got underway. Followng the US soveregn downgrade, captal flows to EMEs agan dred up, then resumed, and have been bouncng around qute a bt ever snce. Ths volatlty, as well as the polcy responses of EMEs attemptng to cope wth the macroeconomc and fnancal-stablty challenges wrought by swngs n captal flows, has prompted nterest n a set of prncples that could gude polcy n both captal-source and captal-recpent countres. Ths note dscusses the analytcal underpnnngs, and possble contours, of such prncples. A frst queston s how an ndvdual country, actng n solaton, should respond to a surge of captal nflows. Ths queston has been the subject of several recent papers by the Fund, ncludng on the use of monetary, exchange rate, and nterventon polces; captal controls; and prudental regulaton of domestc banks (Ostry et al., 2010, 2011, and 2012; and IMF, 2011a and b). The underlyng premse s that there exsts some dstorton n the domestc economy (such as excessve foregn borrowng (see Kornek, 2010, 2011), or a learnng-bydong externalty n the producton of exportables) that s exacerbated by the captal nflow. Whle the ultmate goal may be to mprove the economy s reslence to fnancal-stablty rsks, the short-term response may call for restrctng the volume of nflows or mtgatng the rsks assocated wth them, the precse nstrument dependng on specfc crcumstances. Ths note provdes further analyss of the possble multlateral consequences of country polces, and of the desrablty of multlateral coordnaton to secure globally effcent outcomes. Three ssues are of concern. Frst s the possblty that captal controls may have the effect of vtatng or preventng external adjustment, for example when nflow controls are used to sustan an undervalued currency. Second, as each captal-recevng country that faces excessve captal flows seeks to reduce ts own nflows n support of domestc fnancal stablty, t may deflect some captal flows towards other recpent countres, exacerbatng ther nflow problem. Thrd, polces n source countres, to the degree they ncrease the volume of captal flows, may aggravate problems faced by captal-recevng countres. Emprcal evdence on deflecton between captal-recevng countres s scant, wth most studes fndng ether no, quanttatvely small, or even ambguous effects. More compellng s the evdence on source country polces, wth studes fndng that a key determnant of captal flows to EMEs and especally of nflow surges s advanced-economy nterest rates. There s also evdence that pervasve captal account regulatons may help to sustan undervalued exchange rates over long perods of tme, whch rases a range of multlateral ssues. Do the possble spllovers from captal account polces gve rse to a need for multlateral coordnaton on the use of such polces? When captal controls are used to frustrate the

6 5 external adjustment process, the answer s planly yes (n the same sprt that trade polces that gve unfar compettve advantage are proscrbed). What ths means n practce s not always clear-cut to be sure nflow controls that support an excessve external surplus are lkely to rase red flags n most cases, but when there s a genune learnng-by-dong externalty n the export sector that, as a practcal matter, can only be nternalzed by an undervalued currency, the bass for controls may be sounder and of less multlateral concern. 2 Beyond ths, there are three mplcatons. Frst, although captal controls may deflect nflows from one country to another, ths does not mean that the frst country should refran from actng: rather, all countres that are concerned about excessve (or excessvely rsky) nflows should take approprate measures to safeguard stablty. Indeed, spllover-effects are natural parts of how the market system adjusts to captal controls, and do not necessarly ndcate neffcency. Second, however, f recpent-country polcy responses do not take account of possble deflecton and the response of other countres, the outcome of uncoordnated polces may be neffcent n the sense that countres may end up mposng controls that are excessvely restrctve from ther own pont of vew. Thrd, globally effcent outcomes may also requre coordnaton between recpent and source countres, although such coordnaton s less obvously n the nterest of source countres, and thus more dffcult to brng about. So, s multlateral coordnaton of country-level polces desrable? The foregong dscusson suggests that such coordnaton may be warranted to prevent controls beng used to underpn beggar-thy neghbor external sector strateges or terms of trade manpulaton, to avod potentally costly wars across recpent countres, and to nduce source countres to take nto account the mpact of ther polces on the level and rskness of outflows. It s mportant to recognze that coordnaton mght not nvolve captal account regulatons as such, but nstead monetary and prudental polces that affect the level and rskness of flows. But desgnng coordnaton n practce s much more dffcult than recognzng ts desrablty n theory. The approprate choce of polcy nterventon and ts calbraton depends crucally on specfc crcumstances, and t s mpossble to fully spell out the nature of desrable coordnaton or multlateral rules ex ante. As such, any rules are lkely to be soft rather than hard and to nvolve the applcaton of judgment. Nevertheless, when deployng tools avalable to manage the level and composton of flows, countres should be guded by multlateral prncples to ensure that ther polces are n lne wth the global nterest. What mght ths mean n practce? Monetary and prudental polces n source countres would need to gve an approprate weght to spllovers on global flows and the 2 Countres have an oblgaton under the IMF s Artcles of Agreement to avod manpulatng exchange rates or the nternatonal monetary system n order to prevent effectve balance of payments adjustment or to gan unfar compettve advantage. The learnng-by-dong externaltes dscussed n ths note are not meant to cover exchange rate polces that would volate ths oblgaton.

7 6 macroeconomc and fnancal-stablty rsks facng recpent countres. Captal account regulatons that deflect flows and amplfy rsks n other countres would smlarly need to nternalze ther external effects. Multlateral coordnaton would seek to encourage the adopton of polces that partally nternalze spllovers. Inflow controls mght thus be used less ntensvely when there are generalzed surges to multple countres, and polces n surplus countres that restrct nflows mght lkewse be adjusted at the margn. Supervsory polces n source countres would take account of rsks mposed on foregn countres by fnancal nsttutons n the source countres. Polces undertaken to gan unfar compettve advantage, wth no prudental beneft or genune learnng-by-dong externalty, would rase red flags, even f dentfcaton of such polces s dffcult n practce. Ths note s organzed as follows. Secton II delves nto the pure theory of captal controls (further analytc results are avalable n Kornek, 2012). Secton III surveys the relevant emprcal lterature. Secton IV dscusses mplcatons of the theory. Secton V concludes. II. THE PURE THEORY OF CAPITAL CONTROLS Before ntroducng the (many) real world complcatons, t s useful to consder frst a hghly stylzed setup n whch departures from the neoclasscal paradgm are ntroduced one at a tme n order to dentfy spllovers, externaltes, and welfare mplcatons of polcy nterventons. Accordngly, we consder a world free of any dstortons. Countres may be large (able to nfluence world nterest rates) or prce-takers n world captal markets. The nterventons contemplated n ths note may nfluence the volume of captal flows or the nterest rate at whch partes transact, regardless of the purpose of the nterventon. These nclude captal controls and also prudental regulatons (the former dscrmnate on the bass of resdency, whle the latter do not). The measures may be ether temporary (e.g., cyclcallyvaryng) or of a more structural nature (for example, the closure of the captal account to certan types of flow). Less obvously, monetary and fscal polces may also have sgnfcant effects on cross-border flows and, to the extent they do, are ncluded n the dscusson here. Terms of trade manpulaton wth market power In a world wthout dstortons, would there be any reason to nterfere wth the free movement of captal across borders? The answer s yes, for the same reason that countres large enough to exert market power may want to mpose export taxes or mport tarffs. Specfcally, credtors may beneft from restrctng the supply of captal, drvng up the world nterest rate and shftng the terms of trade n ther favor. 3 For any ndvdual credtor, the best outcome s one n whch other credtors restrct ther supply so that he can beneft from the better terms 3 In contrast to the optmal tarff argument for goods trade, here credtors suffer a captal loss (dependng on the duraton of assets) but gan n flow terms when nterest rates rse. In what follows, we gnore the frst effect.

8 7 of trade on the same or an even hgher volume of flows. By the same token, debtors beneft from restrctng the demand for captal, thereby mprovng ther terms of trade. Whle a theoretcal possblty, t may seem far-fetched to vew large credtors or debtors as delberately seekng to reduce supply or demand for captal n order to manpulate the world nterest rate. But f one consders a broader array of polcy nterventons (rather than, say, just captal controls) that could nfluence the volume or prce of captal flows (for example, monetary polcy), the argument may have an aura of plausblty. To the degree that t does, the case parallels that of the goods market monopolst/monopsonst. Moreover, f both credtors and debtors try to manpulate the terms of trade, then nether succeeds, the world nterest rate remans the same, and both lose by the lower volume of cross-border asset trade. Captal controls to address domestc dstortons Leavng asde ncentves for non-compettve behavor, the modern theory of captal controls stresses dstortons n the domestc economy that result n excessve foregn borrowng as a ratonale for mposng captal controls that can reduce the volume of rsky nflows. The externaltes may be varous. The most obvous are fnancal-stablty rsks (Kornek, 2011). In partcular, ndvdual domestc borrowers may fal to take account of the externalty of ther borrowng on the economy s collateral constrant, whch bnds n the event of a crss. As a result, the compettve equlbrum entals excessve and excessvely rsky (short-term, FX-denomnated) foregn borrowng. The natonal welfare perspectve may therefore call for some form of tax on foregn borrowng, partcularly on the more rsky nstruments. Whle dstortons emanatng from the fnancal sector are an obvous example where foregn nflows mght exacerbate exstng externaltes, t s not the only example. Another possblty s learnng-by-dong externaltes n the producton of exportables. A country wth such an externalty mght engage n sterlzed FX nterventon to acheve an undervalued currency, and nflow controls mght be essental to mantan the exchange rate at the rght level gven the producton externalty (however, ths would not be the frst-best response, whch would be a producton subsdy, see Box 1). The captal nflow controls would ncrease the current account surplus and reduce captal nflows; t s easy to see, moreover, that there exsts a reserve-accumulaton-equvalent to the captal control, as shown n Box 2. Multlateral mplcatons of nflow controls What would be the ramfcatons of such a polcy nterventon? Begn wth the case where there are no dstortons n any other country. The captal control (or other polcy nterventon) by the borrower wth the domestc dstorton reduces nflows to that country (or ncreases ts current account surplus), thereby reducng world demand for captal and depressng world nterest rates (by how much depends on the sze of the country n world captal markets). Borrowers gan and credtors lose. But ths s just a pecunary externalty, nherent to free markets and, from a global perspectve, economcally effcent (Box 3).

9 8 Box 1. Learnng-By-Dong Externaltes n Exportable Producton We consder a two-perod, traded/non-traded good economy that can borrow or lend nternatonally at a rate (1+r). An externalty n the producton of traded goods means that a hgher level of producton, perhaps because of learnng-by-dong, leads to greater productvty n the tradable goods sector n the future. Wth labor the only T T T factor of producton, output n the traded goods sector s gven by: y1 f( l1) ; and y2 A( y1 ) f( l2) where f '( ) 0, f "( ) 0 s a standard neoclasscal producton functon and the productvty parameter A obeys A 1; A'( ) 0. Output n the non-traded goods sector s not subject to the producton externalty: y g(1 l ); y g(1 l ); g' 0, g" 0. N N Decentralzed Economy, Planner s Optmum, and Optmal Government Interventon Snce the learnng-by-dong s external to (.e., cannot be captured by) the frm, the compettve equlbrum entals too lttle producton of the traded good. Specfcally, n the frst perod, the compettve equlbrum sets the margnal rate of transformaton between traded and non-traded goods equal to the real exchange rate (the relatve prce of non-traded goods) takng as gven the externalty: f '( l1) / g'(1 l1) p (1) 1 whereas the natonal socal planner, who takes account of the postve externalty, would allocate a larger share of the labor force to the producton of traded goods: T* * * * [1 A'( y1 ) f( l2) / (1 r)] f '( l1) / g'(1 l1) p (2) 1 The government can reproduce the natonal socal planner s optmum by offerng a subsdy that encourages the producton of traded-goods, where the optmal subsdy would be: * * * s A'( y T 1 ) f( l2) / (1 r) 0 (3) Snce ths subsdy reproduces the natonal socal planner s equlbrum, t s the optmal nterventon; n partcular, producers face the more deprecated real exchange rate (gven by p * / s * ) whle consumers face p. 1 1 Captal Controls Although a producton subsdy to the traded goods sector s the optmal nterventon, t requres budgetary sources to mplement, and a tax on nontraded producton could be dffcult to mplement f that sector s largely nformal. Ghosh and Km (2008) consder several alternatve strateges (export subsdy/mport tarff; consumpton tax; undervaluaton), none of whch s frst best because they dstort the producer and consumer s decson, resultng n too large a current account surplus. Of relevance here s the strategy to mpose a control on captal nflows and support an undervalued exchange rate through sterlzed nterventon. The nterventon absorbs an amount R of nflows. If s the domestc nterest rate at whch households and frms transact, and r s the world nterest rate, then the cost of sterlzed nterventon n an amount R entals a fscal cost of ( r) R. Ths s why the strategy needs some lmt on the amount of nflows (such as captal controls), otherwse the sterlzaton cost would be nfnte. To see that controls-cum-nterventon can be welfare enhancng, t suffces to note that the representatve ctzen s ndrect utlty s ncreasng n R when startng at R=0: ' ' ' T T U'( R) [ (1 r) u2 u1] u2a'( y1 ) f( l2) y1 / R U'( R) R 0 0 (4) At R=0, =r, so the consumer s usual Euler equaton ensures that the frst term of (4) s zero at R=0, whle the second term s necessarly postve because the nterventon serves to deprecate the real exchange rate, encouragng a re-allocaton of labor to the traded goods sector. Although ths strategy can acheve the approprate level of tradables producton, t s not the optmal nterventon because t also dstorts the consumpton decson, mplyng excessve savng and therefore an excessve current account surplus. More generally, whether the polcy nvolves exchange rate undervaluaton or a producton subsdy, there are the usual rsks of polcy nterventon namely that the externalty does not really exst but exporters successfully lobby for the nterventon, whch s socally welfare-reducng, or that the nfant ndustry never grows up. From the perspectve of other countres, currency undervaluaton s worse than a producton subsdy because t tends to vtate ther own attempts to promote tradables producton through undervaluaton. For credtor countres (.e., the source of captal flows), currency undervaluaton may also be worse because t leads to a larger reducton n the recpent country s current account defct, thus shftng the ntertemporal terms of trade aganst credtors.

10 9 Box 2. Equvalence of Captal Controls and Reserve Accumulaton To what extent s use of captal controls equvalent to sterlzed foregn exchange nterventon? Captal controls reduce the country s net demand for foregn funds and the world nterest rate. Lkewse, nterventon that prevents currency apprecaton leads to a larger current account surplus and downward pressure on world nterest rates. Take the case of a regon that consttutes a fracton m of the world economy and that margnally ncreases ts captal control. Applyng the mplct functon theorem to the equaton (1) n Box 2, hgher captal controls rase the regon s net foregn asset holdngs by: uc( c1) b c /2 0 (1 ) u ( c ) / R Ru ( c ) (1) cc 1 cc 2 The approxmaton assumes that captal controls, the net nterest rate and the dscount rate are suffcently close to zero and that perod 1 and 2 consumpton are close to each other. Calbratng the coeffcent of relatve rsk averson to the standard value of 2 mples that a one percentage pont ncrease n captal controls leads to a reducton n captal nflows/gdp (or ncrease n outflows) of 0.25 percentage ponts. Or, flppng ths around, reserve accumulaton of 1 percent of GDP s equvalent (n terms of the mpact on the global captal markets) to mposng a 4 percent captal control. The world excess demand for bond holdngs s the weghted sum of all bond postons over all regons, B mb. Usng the same approxmatons as above, ts response to a change n the world nterest rate s mb (1 ) u ( c) b / R u ( c ) B m y/ 0 (1 ) ( ) / ( ) 2 cc 1 c 2 R R uc c1 b R Rucc c2 where s global GDP. Combnng (1) and (2), the effect of captal controls n one regon on the world nterest rate s gven by the weghted share of regon 's consumpton n world GDP: R mb R mc / y B R For example, f a country lke Brazl, whch produces about 3.3 percent of world GDP, mposes an nflow tax of 1 percent, t reduces the world nterest rate by percentage ponts. If a country such as Chna that produces about 10 percent of world GDP accumulates 1 percent of GDP n foregn reserves or, equvalently, mposes a 4 percentage pont captal control, then t pushes down the world nterest rate by approxmately 0.4 percentage ponts. In practce, of course, the use of captal controls and reserve accumulaton are not necessarly substtutes: captal controls are often used to support a polcy of sterlzed nterventon. Now suppose, more realstcally, that other borrowers lkewse face dstortons n the form of excessve foregn borrowng. Imposton of captal controls by one country lowers world nterest rates and deflects captal, exacerbatng other borrowers nflow problem. But ths s not a reason for the country to refran from actng, because spllovers here are pecunary externaltes. The effcent outcome s rather for all (excessve) borrowers to act. Ths wll mply lower world nterest rates, beneftng borrowers and hurtng credtors wthout harmng global effcency (Box 3). The same logc apples to a learnng-by-dong producton externalty as to the fnancal-stablty externalty. Whle acton by one country makes t more dffcult for others to nternalze ther externaltes (not everyone can run larger surpluses 4 ), from an effcency standpont, the resultng pecunary externaltes are globally effcent. (3) (2) 4 If exportables are dfferentated products, then all countres can export more. If the learnng-by-dong stems from exportables producton, then all countres can reap the benefts of a producton subsdy. But snce not all countres can run larger external surpluses, the controls-cum-undervaluaton strategy (whch, unlaterally, leads to excessve savng and current account surplus) becomes both unlaterally and multlaterally neffcent.

11 10 Box 3. Global Effcency of Correctng Domestc Externaltes Can captal controls such as taxes on nflows help address domestc externaltes? And would ths be globally effcent? Box 1 consdered the specfc example of a learnng-by-dong externalty. Ths box generalzes the result and dscusses whether such polcy nterventons are globally effcent. Agan we consder a two-perod, open economy model, where the utlty functon of the representatve agent n country s: U u( c ) u( c ) (1) 1 2 The agent s budget constrants mply c 1 y 1 (1 ) b / R T and c2 y2 b where c s consumpton, y s output, b s foregn bonds (negatve f the country s a debtor), s the captal control (a tax on foregn borrowng, a subsdy on foregn lendng), R s the world nterest rate, and T the lump-sum rebate of the captal control revenues. The frst-order condton characterzng the representatve agent s optmal consumpton and foregn borrowng (or lendng) decson equates the margnal rate of substtuton to the after-tax rate of return: (1 ) u ( c ) Ru ( c ) (2) c 1 c 2 Now suppose that there s a negatve externalty assocated wth foregn borrowng. Kornek (2012) consders externaltes n the fnancal sector such that atomstc domestc agents, who fal to take account of the externalty, engage n excessve (or excessvely rsky) foregn borrowng; Ghosh and Km (2008), Rodrk (2008), and Box 1 above consder a learnng-by-dong producton externalty n the tradable goods sector. Regardless of the specfc nature of the externalty, t can be wrtten (n utlty-equvalent terms) as the representatve agent s true welfare functon s 1 2 externalty when choosng hs optmal consumpton and borrowng). The natonal socal planner does take account of the externalty, yeldng the frst-order condton: 1 c 2 b so that W u( c ) u( c ) b (though the agent gnores the u ( c ) Ru ( c ) R (3) c Comparng (2) and (3), the government can reproduce the natonal socal planner s optmum by mposng a tax on captal flows equal to R / u ( c ) (4) c 1 thus forcng the representatve agent to take account of the negatve externalty (e.g., on fnancal stablty) assocated wth hs foregn borrowng. In a mult-country world, where there are N (not necessarly dentcal) N j countres, the world nterest rate, R, wll adjust to ensure that the captal market clears: 1b 0. The tax on foregn borrowng s optmal from the ndvdual country s perspectve. But s t globally effcent? It turns out that t s. Ths can be shown by consderng the optmal allocaton that would be chosen by a global socal planner who optmzes a weghted sum of natonal welfares, assgnng to country a weght, subject N j to each country s budget constrant and the global market clearng condton j 1b 0, whch has a shadow prce (Lagrange multpler),. The frst-order condton characterzng the global socal planner s optmum s: u ( c ) Ru ( c ) R R/ 1,... N (5) c 1 c 2 At the global socal planner s optmum, 0, because the planner does not need any addtonal bonds to acheve the optmal allocaton (so the shadow value on easng the constrant on the global supply of bonds s zero). Settng v 0 n (5) and comparng to (3) shows that the natonal socal planners (whose optmal decsons are gven n (3)) acheve the same allocaton as the global socal planner (whose optmal allocaton s gven n (5)). The ntuton for why the Nash equlbrum among natonal planners s globally effcent s smple: each planner has one target (.e., to offset the externalty) and one nstrument (the tax on captal flows) a general result n the theory of polcy coordnaton s that when countres have as many nstruments as targets, there s no need for coordnaton as the Nash equlbrum wll be effcent (see Ghosh and Masson, 1994). j

12 11 Costly captal controls The noton that all borrowers that are borrowng excessvely should mpose captal controls mplctly assumes that such controls are costless. In practce, there are lkely to be several types of cost assocated wth such polces: admnstratve cost (.e., payng offcals to admnster the controls); complance costs for fnancal nsttutons and others affected by the controls; and effcency costs due to mperfect targetng of excessvely-rsky forms of nflow. There are also costs from mplementng an export-led growth strategy supported by nflow controls (as an answer to a producton externalty), most notably quas-fscal sterlzaton costs and dstortve effects on consumpton (controls lead to excess savng). Fnally, snce mposton of controls may create vested nterests, these costs may be long lastng. Whle admnstratve and complance costs should not be downplayed, t s the economc dstortons collateral damage caused by mperfect targetng of rsky flows that s lkely to be especally mportant. It stands to reason, moreover, that the more ntensve the control, the greater the collateral damage. For nstance, snce a hgher tax rate wll gve greater ncentve for crcumventon, ts breadth of applcaton wll rse, and caught n the net wll be flows that otherwse would not mert beng restrcted. As such, the cost functon s lkely to be convex: not only ncreasng n the ntensty of the control, but ncreasng at an ncreasng rate. Costlness of controls mples that actons by other borrowers or credtors that exacerbate domestc dstortons (e.g., by lowerng world nterest rates and nducng more nflows where there s already over-borrowng) have substantal multlateral welfare mplcatons. Costly spllovers exsted above, but they were mmateral because (by assumpton) they could be costlessly offset by countres. Even when offsettng polces are not costless, there s no presumpton that actons to offset domestc externaltes should be proscrbed. But t s lkely that n such cases, unfettered unlateral actons wll no longer be globally effcent (Box 4). 5 Ths s perhaps very stark n the case of multple countres attemptng to pursue an export-led growth strategy. Snce t s not possble for all countres to run larger surpluses, the futlty of unlateral polces s readly apparent. Moreover, the beneft-cost calculus for a captal controls war that stfles goods trade s lkely to be much less favorable than when the war s based on deflectng captal flows. The reason s that, n the extreme, when nternatonal asset trade falls to zero because of prohbtve captal controls, the resultng stflng of asset trade also means that fnancal-stablty rsks from cross border flows (the externalty) fall to zero. But when goods trade falls to zero, the resultng stflng of goods trade means not only that gans from trade are extngushed, but also that none of the producton externalty s captured. 5 Specfcally, regardless of whether costs are lnear or convex, there wll be a case for coordnaton among captal-recevng countres. If costs of controls are convex, global effcency wll also necesstate coordnaton between borrowers and lenders, so that the margnal cost of nflow controls n the recpent country equals the margnal cost of outflow controls n the source country. If costs are nstead lnear, then the global effcency crteron does not, tself, determne the optmal splt between nflow and outflow controls.

13 12 Box 4. Role for Coordnaton under Costly Captal Controls As dscussed n Box 2, when there are domestc externaltes assocated wth captal flows, governments can nduce domestc agents to take account of them by mposng taxes on the flows. Moreover, these (uncoordnated) taxes are globally effcent. But are there crcumstances under whch they would not be effcent and coordnaton would be requred? The answer s yes, provded the captal controls themselves ental costs. Agan we consder a two-perod model where the representatve agent dsregards the externalty assocated wth captal flows. The natonal planner must choose the optmal tax on captal flows to maxmze where the cost functon welfare takng account of the welfare cost of mposng captal controls, gven by ( ), s assumed to be ncreasng and convex (ncreasng at an ncreasng rate) n the tax rate: ' 0, " 0: W u( c ) u( c ) b ( ) (1) 1 2 The resultng frst-order condton characterzng the natonal socal planner s optmum s: u ( c ) Ru ( c ) R '( )(1 )( R ) (2) c 1 c t cc t c t where u ( c )/ u ( c ), t 1,2. The terms n (2) trade off the benefts of foregn flows for consumptonsmoothng aganst the domestc externalty but also the cost of mposng captal controls. Solvng for the natonal planner s optmal tax rate shows that t wll be postve as long as there s an externalty, but t wll fall short of fully offsettng the externalty because of the costs assocated wth the mposton of the control: 0 R / u ( c ) (3) c 1 By contrast, when controls are not costly, the planner fully offsets the externalty (see eqn. (4) n Box 2). What tax rates would a global planner choose for each country, =1,..N? It can be shown that they must satsfy: N j j j 1 (1 ) '( ) 0 (4) that s, the global socal planner wll try to mnmze the aggregate global cost of the captal controls by spreadng the use of controls across countres (gven the convexty of the captal control cost functon, t s better to have two countres mpose relatvely low controls than to have one country mpose hgh controls). More precsely, the weghted average margnal dstorton mposed by captal controls must be zero as n (4). If there are no externaltes assocated wth captal flows ( 0 ), then the soluton to (4) s smply 0 the global planner sets all tax rates equal to zero. Ths s ntutve, and s the only case where the natonally chosen tax rates (3) concde wth those chosen by the global planner (4). More generally, (3) and (4) wll dffer, and the uncoordnated equlbrum among natonal planners wll be globally neffcent (ntutvely, each planner now has two targets offsettng the domestc dstorton and mnmzng captal control costs but stll only one nstrument, so the uncoordnated equlbrum s neffcent). Consder the followng examples and some for the calbraton exercses n Box 5: Example 1 (Symmetrc countres). Consder a world economy of k 1 dentcal countres that mpose captal controls to offset domestc externaltes, ncurrng a cost ( ) 0. Snce countres are dentcal, however, the captal controls smply deflect captal to each other, wth no net effect on flows or the externalty. Therefore, t s Pareto effcent to set all controls to zero, thus avodng the cost of controls. Indeed, that s the global socal planner s soluton (the only soluton to equaton (4) for symmetrc countres s ( ) 0 0 ). Example 2 (Two countres wth asymmetrc externalty). Consder a world wth countres (borrower) and j (lender). Country experences an externalty from excessve nflows 0, so ts planner mposes an nflow control, 0 whereas country j wll not mpose any controls. Ths equlbrum s neffcent and the optmalty condton (4) s volated. A global planner would lower the tax rate n country and mpose a tax on outflows n country j. Gven the convexty of costs, the sum of the deadweght loss from the (lower) nflow control plus the outflow control wll be less than the cost of a hgh nflow control n the borrowng country.

14 13 Box 5. Gans from Coordnaton under Costly Captal Controls Ths Box provdes a smple calbraton of the gans from coordnaton under costly captal controls. We focus on captal flows between two regons a borrowng regon that exhbts domestc externaltes and a lendng regon that s not subject to externaltes. The relatve sze of the two regons s assumed to be three-to-one, roughly correspondng, for example, to the relatve szes of the US and Latn Amerca. In our example, reported n the frst panel of Table 1, the borrowng regon mports captal n the amount of 3.1 percent of GDP at a world nterest rate of 4.1 percent n the lassez fare equlbrum. We assume that the externalty n the borrowng regon s 6 percent, and that both regons experence a 2 resource cost of mposng captal controls that takes the functonal form ( ). The parameter s set to 0.05 so that planners n the borrowng regon fnd t optmal to correct only two-thrds of the externalty n the absence of coordnaton, mposng a 4 percent nflow tax. Snce there are no externaltes among lenders, domestc planners n that regon mpose no controls on lendng. In the resultng equlbrum, reported n the second panel of Table 1, captal flows n the borrowng regon declne to 2.3 percent of GDP, correspondng for example to the stuaton n Brazl n The world nterest rate falls to 3.1 percent per year due to the lower worldwde demand for captal. In the Nash equlbrum among planners, captal controls ncrease the welfare of borrowers by the equvalent of a percent ncrease n consumpton and reduce the welfare of lenders by.009 percent due to the lower nterest rate. (If borrowers compensated lenders for ther loss, they would stll enjoy a welfare gan of percent.) Table 1. General Equlbrum Effects of Captal Controls wth and wthout Coordnaton Lassez Fare Nash Equlbrum Full Coordnaton Coordnaton wth Restrcted Lender Borrower Lender Borrower Lender Borrower Lender Borrower Lender captal control 0% 0% 4.04% 0% 3.13% -0.90% 3.72% set to 0 CA/GDP -3.09% 1.03% -2.32% 0.77% -2.33% 0.78% -2.38% 0.79% nterest rate 4.12% 3.09% 4.03% 3.17% welfare LF % % % % % % welfare Nash % % 0 If all countres n the two regons coordnate ther use of captal controls, a more effcent global equlbrum can be acheved. As reported n the thrd panel of Table 1, the borrowng regon reduces ts nflow tax from 4 percent to 3.1 percent and the lendng regon mposes a control of 0.9 percent on outflows. The sum of the two controls s almost unchanged from the Nash equlbrum, but snce the cost of mposng captal controls s convex, the two regons reduce the aggregate cost () by sharng the burden of regulaton. The net global welfare gan from coordnaton s equvalent to a.0017 percent ncrease n consumpton of the borrowng country. Observe that ths gan s an order of magntude lower than the gans of gong from lassez fare to the uncoordnated Nash equlbrum among natonal planners. (For ease of nterpretaton, we assume that n coordnated equlbra, a transfer from lenders to borrowers keeps the welfare of lenders unchanged from the Nash equlbrum.) Fnally, we consder a global planner who nternalzes the general equlbrum effects of captal controls on the world nterest rate but cannot mpose captal controls n the lendng regon. Ths stuaton may reflect for example that ths regon s an mportant fnancal center where the mposton of captal controls would carry prohbtve costs. The resultng equlbrum s reported n the last two panels of Table 1. All countres n the borrowng regon now mpose captal controls that are smaller than the unlaterally optmal controls n the Nash equlbrum, but larger than n the case of full coordnaton. As a result, the world nterest rate and welfare margnally ncrease compared to the Nash equlbrum, but the gans are smaller than n the case of coordnaton wth lenders. The global planner reduces the captal control for borrowers even f lenders cannot adjust ther controls because he recognzes that lowerng borrowers controls rases the world nterest rate, whch also dscourages borrowng but wthout mposng the deadweght loss (). (For ths form of coordnaton to lead to a Pareto mprovement, the global planner agan needs to compensate borrowers for the hgher nterest burden by makng a compensatory transfer from lenders.)

15 14 Coordnaton among borrowers Coordnaton between borrowng countres mposng controls on captal nflows s necessary because, f they fal to coordnate, each country may end up mposng controls that are too hgh from ts own perspectve. Ths result actually follows drectly from the country s costbeneft calculus n choosng the level of captal controls gven ts dstortve cost. When the country gnores the repercussons on other borrowers of the deflected captal flows and ther lkely reacton to them, t overestmates the beneft of the controls because t fals to recognze that the equlbrum reducton n nflows wll be smaller once other countres react by mposng ther own controls. Put dfferently, at the end of the captal control war, borrowng countres would fnd that, from ther own perspectves, captal controls are too hgh (and the volume of flows too low): each borrower that mposed controls would be better off f they could jontly agree to lower controls from ther uncoordnated equlbrum levels. Source-country polces The second neffcency concerns the role of credtor countres n managng outflows. If the cost of captal controls s convex, then t would be more globally effcent to splt that cost between borrowng countres (who would mpose controls on nflows) and lendng countres (who would mpose controls on outflows). 6 One obstacle to such coordnaton s that captalexportng countres do not drectly beneft from movng from the uncoordnated equlbrum (where they do not mpose controls or adjust other polces e.g., monetary polcy to take account of the mpact on borrowng countres, and therefore do not ncur any assocated costs) to the coordnated equlbrum (where they ncur such costs). Nevertheless, t may be n the nterests of captal-sendng countres to mpose some restrant on outflows, especally when these are exacerbatng global fnancal-stablty rsks. Frst, they may smply want to contrbute to global stablty, recognzng that t s a publc good. Second, nasmuch as crses n recpent countres mply losses for fnancal nsttutons n source countres, t may be n the latter s nterest to reduce the rsk of such crses. 7 In that case, measures that reduce the volume or rskness of outflows consttute a form of prudental regulaton for banks n credtor countres. Thrd, credtor countres enjoy a terms of trade 6 Coase s (1960) theorem on the rrelevance of the assgnment of externalty-generatng rghts does not apply here because convex costs mply that technologcal effcency requres splttng the burden of captal controls between source and recpent countres. 7 A volumnous lterature documents the mpact on source countres of fnancal crses n borrowng countres: see, e.g., Sachs and Huznga (1987) on the Latn Amerca debt crss (whch posed a systemc threat to U.S. banks); and Boughton (2012) for dscusson of the Asa and Russan crses. If source countres have ther own ncentve to restrct outflows (perhaps because of negatve repercussons of a crss n the borrowng country), then global effcency requres the equalzaton of the margnal cost of nflow controls n the recpent country to the margnal cost of outflow controls n the source country; when these costs are lnear (.e., not convex), the global effcency crteron does not pn down the optmal splt between nflow and outflow controls.

16 15 gan on ther flow of lendng by reducng the supply of captal, so measures that reduce outflows would be n ther nterest even f undertaken n order to reduce fnancal-stablty rsks n recpent countres. In ths case, there s a happy concdence whereby the (ncdental) exercse of monopoly power by source countres helps debtor countres whch would otherwse be over-borrowng. In sum, once we depart from the frst-best world of perfect markets there may be vald economc reasons for nterventons to address domestc externaltes such as those that result n excessve or excessvely rsky foregn borrowng. And, as a matter of logc, n a second best world, polcy actons by borrowers or lenders can exacerbate exstng externaltes such that cross-border spllovers have global welfare mplcatons. In general, ths does not mean that countres should refran from polcy nterventons but t does mean that effcent outcomes wll lkely requre coordnated polces. Ths s a fortror the case when captal controls are beng pursued to support an export-led growth strategy. When naked mercantlsm s the bass of polcy at the country level, t s clear that the polcy s nether unlaterally nor multlaterally effcent. But even when such polcy s unlaterally desrable, say because of genune learnng-by-dong externaltes n exportables producton, the need for multlateral coordnaton to avod self-defeatng polces s very clear. 8 III. SURVEY OF EMPIRICS Any call for polcy coordnaton both among captal-recpent countres, and between captal-exportng and captal-mportng countres s predcated on the relevant spllovers beng quanttatvely mportant. Indeed, f captal controls (and other polces that have smlar effects) do not have a sgnfcant mpact even on the flows to the country mposng them, then the whole dscusson s moot. In ths secton, therefore, we revew exstng evdence on three questons: do captal controls alter the volume or composton of captal nflows? Does the mposton of controls by one borrower deflect flows to others? And do polces n source countres apprecably affect the volume of cross border flows to captal-recpent countres? Volume and composton of captal nflows Most of the evdence on the effectveness of captal controls suggests that they have lttle effect on overall flows. Ths fndng relates to countres that have relatvely open captal accounts (many of today s emergng market countres, for example); there s lttle doubt that countres that are less well ntegrated n global captal markets can nfluence the magntude of flows through a range of admnstratve measures and quanttatve lmts. In general, the effectveness of controls n regulatng the magntude of flows depends on how extensve they 8 Moreover, learnng-by-dong externaltes seem more relevant for countres behnd the technologcal fronter, not for those at the fronter. So the bass of an undervaluaton strategy s lkely to be fundamentally asymmetrc.

17 16 are, whether the country mantans the necessary admnstratve and nsttutonal nfrastructure to enforce controls, and the extent of nvestors ncentves to crcumvent them. The lack of convncng evdence on the mpact of controls on the overall level of captal flows lkely reflects a number of factors. For countres wth relatvely open captal accounts, captal controls have represented margnal changes that have been undertaken alongsde a range of other measures whch also mpact flows, ncreasng the dffculty of solatng the effects of captal control measures themselves. Measurng the ntensty of captal controls has also been a problem plagung emprcal nvestgatons. And perhaps foremost are fundamental econometrc dentfcaton problems f countres that are facng large nflows are the ones that mpose controls, t s not surprsng that econometrc studes fnd no, or even a postve, relatonshp between controls and the magntude of nflows. In lght of these dentfcaton problems, the exstng emprcal studes should not lead one to conclude that controls have no tracton n managng the level of flows. The emprcal evdence that controls have a systematc mpact n tltng the composton of flows, moreover, also casts doubt on the noton that the aggregate volume of flows s mpervous to the level of controls. Ths s because, f there are no aggregate volume effects but sgnfcant compostonal effects, there would need to be full offsets across dfferent types of flow n response to controls. Ths seems hghly mplausble, gven the dfferent forces that lkely drve the ndvdual components of the captal account. If, for example, controls help to curtal nflows of short-term debt, t s much more lkely that total nflows would be reduced than that another component of the captal account e.g., foregn drect nvestment would rse dollar-for-dollar wth the reducton n short-term debt. We conclude that the lack of strong evdence n favor of aggregate effects of captal controls reflects more the dffcultes of gettng the data to speak loudly than a lack of tracton of the polces themselves. Lterature revews and meta-studes (ncludng Magud et al. (2011), Habermeer et al. (2011) and Ostry et al. (2010, 2011)) suggest that captal account polces have tracton along a number of dmensons monetary polcy ndependence, the exchange rate path, and compostonal effects (see also Klen, 2012, forthcomng). Whle evdence of an mpact on the total volume of flows s thus hard to come by, ths may reflect dentfcaton problems more than realty. Dverson and cross-border spllovers If captal controls mpact the aggregate level of flows, there s obvously potental for them to have welfare mplcatons for other countres, e.g., by dvertng flows n stuatons where such dverson would amplfy exstng dstortons. Is there any evdence to confrm ths noton? Drect evdence of cross-border effects from captal account polces s mxed, an unsurprsng fndng n the lght of the lmted evdence that such polces have szable effects on captal flows to the country mposng the measure. Forbes et al. (2011) fnds small average effects on flows to other countres from the mposton of captal controls by Brazl:

18 17 some countres saw ncreases n nflows as a result of the Brazlan measures, whle others saw decreases. IMF (2011) usng a slghtly dfferent methodology/data comes to a smlar concluson, that captal account polces have on occason led to ncreases or decreases n flows to other countres. Both studes suggest dfferent explanatons to ratonalze the results, ncludng search for yeld (whch pushes captal to countres wth no controls) or amplfed rsks of the mposton of controls n other countres (deflectng captal toward countres deemed less lkely to mpose controls), whch may lead total outflows from source countres to dmnsh. The latter result that captal s deflected away from other countres lkely to mpose controls means that a captal control war between captal-recevng countres s unlkely to occur. Inasmuch as countres that are facng excessve nflows are more lkely to want to mpose controls, t may also mean that the deflecton s effcent. Source-country polces What are the effects of source country polces on captal flows to emergng economes? There s certanly plenty of evdence along a broader dmenson of ths queston, for example, the spllover effects from monetary polcy n source countres through cross-border flows (e.g., Calvo et al., 1993; Taylor and Sarno, 1997; Renhart and Renhart, 2008; Ghosh et al., 2012). Indeed, emprcal studes typcally fnd that advanced economy nterest rates (or other proxes for monetary polcy) are among the most mportant determnants of captal flows to EMEs. 9 The exstence of spllovers from a range of polces (monetary, fnancal, fscal, regulatory, and structural) has been documented n the seres of spllover reports undertaken by the Fund over the past year: such spllovers can and do operate n part through crossborder fnancal flows and can have welfare mplcatons when the resultng flows exacerbate pre-exstng dstortons. Spllovers from macroprudental polces are also evdent n cases where, for example, defcences n domestc prudental or regulatory polces amplfy the extent of rsky cross-border captal flows (IMF, 2011). To sum up, emprcal evdence on the mpact of captal account polces s mxed, and generally stronger n terms of fndng an effect on the composton of flows than on the aggregate level. However, the fndng of neglgble effects on aggregate flows may reflect econometrc dentfcaton problems more than realty. Smlarly, the fndng of small crossborder spllovers may reflect more the small measured unlateral effects than a true absence of spllovers (and possbly also the fact that measures mposed to date have been small n the grand scheme of thngs). The potental for spllover effects from captal account polces, moreover, would seem to be a salent rsk were captal controls to become larger n the future. Whatever stand one takes on the strength of the emprcal evdence, the logcal arguments presented n Secton II, and ther mplcatons presented n Secton IV below, wll reman vald. If the spllovers and dstortons are strong, then the case for multlateral 9 Note that the argument, sometmes made, that quanttatve easng produces a wall of lqudty that washes over emergng market countres, should really be recast n terms of the prces of dfferent assets, wth the actons by central banks engaged n QE provdng ncentves to nvest n rsker assets, ncludng rsker foregn assets.