Financial Frictions, Foreign Currency Borrowing, and Systemic Risk

Transcription

1 USC FBE FINANCE SEMINAR prsntd by Robrt Marquz FRIDAY, Oct. 7, :30 am 12:00 pm, Room: JFF-416 Financial Frictions, Forign Currncy Borrowing, and Systmic Risk Giovanni Dll Ariccia IMF and CEPR Luc Lavn ECB and CEPR Fbruary 2016 Robrt Marquz Univrsity of California, Davis Abstract Firms in mrging markts oftn borrow in forign currncy. Forign currncy borrowing can amliorat financial frictions (by improving firms incntivs and rducing agncy problms) but may incras systmic risk. In addition, through limitd liability, forign-currncy dnominatd dbt acts as a stat-contingnt claim: Borrowrs maximizing profits in local currncy ar shildd from dvaluations (whn rpaying FX dbt is xpnsiv) but pay highr rats in no-dvaluation stats (whn rpaymnt is rlativly chapr). Th rsulting trad-off btwn avrag prformanc and systmic stability, which bcoms strongr whn widsprad bankruptcis incras th risk of failur, provids a rational for rgulatory masurs that limit currncy mismatchs. Kywords: liability dollarization, systmic risk, limitd liability, banking criss JEL Classification Numbrs: E44, E58, G21 Th viws xprssd in this papr ar thos of th authors and do not ncssarily rprsnt thos of th IMF or th ECB. W thank Stijn Classns, Itay Goldstin, and Pirr-Olivir Gourinchas for usful commnts and discussions, as wll as participants at th 2015 Wharton Confrnc on Liquidity and Financial Criss and a sminar at th Swdish Riksbank. Addrss for corrspondnc: Robrt Marquz, Graduat School of Managmnt, On Shilds Av., Univrsity of California, Davis, CA rsmarquz@ucdavis.du.

2 Financial Frictions, Forign Currncy Borrowing, and Systmic Risk Abstract Firms in mrging markts oftn borrow in forign currncy. Forign currncy borrowing can amliorat financial frictions (by improving firms incntivs and rducing agncy problms) but may incras systmic risk. In addition, through limitd liability, forign-currncy dnominatd dbt acts as a stat-contingnt claim: Borrowrs maximizing profits in local currncy ar shildd from dvaluations (whn rpaying FX dbt is xpnsiv) but pay highr rats in no-dvaluation stats (whn rpaymnt is rlativly chapr). Th rsulting trad-off btwn avrag prformanc and systmic stability, which bcoms strongr whn widsprad bankruptcis incras th risk of failur, provids a rational for rgulatory masurs that limit currncy mismatchs. Kywords: liability dollarization, systmic risk, limitd liability, banking criss. JEL Classification Numbrs: E44, E58, G21

3 1 Introduction In this papr w xplor how forign currncy borrowing can amliorat financial frictions in mrging conomis whil at th sam tim incrasing systmic risk. Morovr, w focus on firm-lvl dcisions to maximiz xpctd profits that ar indpndnt of concrns rlatd to th xploitation of govrnmnt actions, or of coordinatd actions across firms (s discussion blow). Undr conditions typical of fixd xchang rat rgims, forign-currncy-dnominatd liabilitis improv firms incntivs and rduc th agncy problm associatd with limitd liability and th unobsrvability of a firm s actions. In doing so, it rducs idiosyncratic risk for firms. Howvr, forign currncy borrowing also xposs th systm to th risk of corrlatd dfaults through xchang rat dvaluation. Forign currncy borrowing (or liability dollarization) has bn a common fatur in svral mrging markt conomis. Figur 1 illustrats this pattrn, showing that th total amount of forign currncy dnominatd dbt, xprssd in U.S. Dollars, by corporations in mrging markts ovr th last dcad has bn larg and incrasing, radily surpassing th $1 Trillion mark. 1 Typically, this liability dollarization rducs th intrst borrowrs pay on thir loans (ths countris gnrally pay a currncy prmium) and has bn associatd with fastr crdit and conomic growth. For instanc, in th run-up to th rcnt global financial criss, among a sampl of Eastrn Europan countris, crdit growth was th fastst in countris that had a largr shar of crdit dnominatd in forign currncy. Liability dollarization, howvr, also incrass systmic risk. Should th country xprinc a sharp currncy dprciation, firms with unhdgd forign-currncy dnominatd dbt would find it difficult to honor thir liabilitis, rsulting in widsprad bankruptcis. Indd, thr is a clar link btwn liability dollarization and th frquncy of criss, in particular in th banking sctor. 2 Liability dollarization also appars to b associatd with mor rigid xchang rat rgims. For xampl, again in Eastrn Europ, countris with currncy boards or rigid pgs (such as Bulgaria, 1 Bruno and Shin (2015) documnt that roughly on third of annual gross bond issuanc by mrging markt non-financial firms is forign currncy (i.., US Dollar) dnominatd - s Figur 1. Apart from Asia, whr th fraction of dbt issuanc that is forign currncy dnominatd is lowr, issuanc in forign currncy dominats in othr mrging markts, such as Latin Amrica (s IMF, 2015, figur 3.11). This bcoms vn mor pronouncd whn focusing on xtrnal borrowing, whr th fraction of such borrowing that is dnominatd in forign currncy is gratr than 80%, as documntd in Du and Schrgr (2015). 2 S Schnidr and Tornll (2004). 1

4 Forign Bank Lnding and FX Dnominatd Bonds in Emrging Markts (Billions of US dollars. Sourc: IMF, Global Financial Stability Rport, 2015) Forign bank lnding + FX bonds FX Bonds Figur 1: Forign currncy dnominatd dbt in mrging markts, , xprssd in U.S. Dollars. Th solid rd lin rprsnts th valu of bonds outstanding from corporations in mrging markt which wr dnominatd in a forign currncy. Th dashd blu lin adds lnding by forign banks, th majority of which is forign-currncy dnominatd. Th countris in th sampl ar Argntina, Brazil, Bulgaria, Chil, China, Colombia, Hungary, India, Indonsia, Kora, Malaysia, Mxico, Pru, Poland, Philippins, Romania, Russia, South Africa, Thailand, and Turky. Estonia, or Latvia) had a much largr shar of crdit to th privat sctor dnominatd in forign currncy than xchang rat floatrs (such as th Czch Rpublic, Poland, and Slovakia). 3 Thr is also som vidnc that th shar of forign currncy lnding in domstic crdit gradually dclind in countris that abandond a fixd xchang rat rgim. 4 In our modl, ntrprnurs borrow in ordr to invst in productiv projcts. A projct s probability of succss dpnds on th ntrprnur s costly ffort. W introduc two basic financial frictions. First, ntrprnurs/firms ar protctd by limitd liability. Scond, an ntrprnur s ffort is unobsrvabl to lndrs and cannot b contractd upon. Ths two frictions gnrat an infficincy in th conomy as thy ntail a backward bnding crdit supply curv (à la Stiglitz and Wiss, 1981). Highr intrst rats rduc th ntrprnur s payoff in cas of succss and thus also rduc hr ffort. Thn, whn th cost of ffort is sufficintly high, thr dos not xist an intrst rat at which th lndr can brak vn givn th xpctd probability of rpaymnt. Put diffrntly, projcts that could b fundd undr prfct information ar rationd out of crdit markts whn th ntrprnur cannot commit to a particular lvl of ffort. W assum that th domstic currncy is xpctd to dprciat rlativ to th forign currncy, 3 S Rosnbrg and Tirpak (2008). 4 S Martinz and Wrnr (2002) for a study on Mxico in th aftrmath of th Tquila crisis. 2

5 so that th risk fr domstic intrst rat is highr than th forign rat, and that this sprad is du to th xpctation of a larg dvaluation to which markts attach a rlativly low probability. W intrprt ths pso problm conditions as typical of xchang rat pgs and currncy boards in mrging markts (whr th forign currncy can b takn to b th US dollar or th uro). W also assum a lss-than-complt xchang-rat pass-through, so that xchang rat movmnts hav an impact on th solvncy of firms with forign-dnominatd liabilitis. Undr ths conditions, firms driv two bnfits from forign currncy borrowing. First, thr is a pur stat-contingnt pricing ffct drivn by limitd liability for th borrowing firm: dvaluations incras th cost to th borrowing firm (which cars about its rturn in domstic currncy) of rpaying its forign currncy dbt; and larg dvaluations lad to dfault whn th firm is subjct to limitd liability. But dfault is fficint hr sinc it allows th borrowing firm to rallocat rpaymnt from stats of th world in which it is xpnsiv (i.., dvaluation stats) to thos in which it is rlativly chap (whn th pg holds). This occurs through a forign currncy lnding rat that prics dfault risk fairly. 5 Scond, w show that forign currncy borrowing rducs th moral hazard associatd with limitd liability. Th rason is that borrowing in forign currncy acts as a bonding mchanism for th firm: sinc th risk fr rat abroad is lowr, th firm is abl to obtain a lowr intrst rat loan. This crats a highr rturn for th firm whnvr its projct is succssful, and provids th firm with a gratr incntiv to nsur that its rturn actually matrializs, i.., to put in mor ffort. 6 Th trad-off, howvr, is that borrowing in forign currncy xposs th firm to dvaluation risk, which can lad to dfault and hnc bankruptcy costs. Th probability of bankruptcy dpnds on th distribution of possibl xchang rat movmnts: whn th probability of an xchang rat movmnt is high, forign currncy lnding worsns, rathr than amliorats, th agncy problm. It is only whn th dvaluation risk is rlativly low, but th potntial movmnt in th xchang rat larg, that borrowing in forign currncy incrass firm valu through th two channls discussd abov. This is bcaus th ngativ ffct on incntivs of th additional dfault risk is proportional to th probability of dvaluation, whil th positiv ffct stmming from rducd intrst paymnts 5 S Dll Ariccia and Marquz (2015) for a gnral tratmnt of this issu. 6 This is akin to what happns in standard banking modls with risk shifting manating from limitd liability. In thos modls, highr dposit rats tnd to incras risk taking by rducing bank profits (or franchis valus). S for instanc, Matuts and Vivs (2000), Hllmann, Murdock, and Stiglitz (2000), Rpullo (2004), Boyd and D Nicolo (2005), Alln, Carltti, and Marquz (2011), Dll Ariccia, Lavn, and Marquz (2014). 3

6 dpnds in th xpctd xchang rat movmnt. From a policy prspctiv, th papr supports th viw that govrnmnt intrvntion, in th form of (macro) prudntial rgulation and/or capital controls, to curb forign currncy borrowing and th systmic risks associatd with it may b socially optimal. Th papr points to a trad-off btwn suprior productivity and gratr systmic risk (dfind as th risk of widsprad failurs, i.., a crisis). This trad-off bcoms important in th contxt of social costs of failur that may b non-linar and thrfor larg whn risk is systmic. Whil th modl assums risk nutrality throughout, ths concrns ar likly important in practic. W xplor this issu furthr by xtnding th modl to considr th possibility that th failur of a countrparty ngativly affcts firms with succssful projcts and causs thm to fail as wll. This issu bcoms mor important th mor firms fail at th sam tim, and w show that this adds an additional important wrinkl to th problm. In particular, th risk of dvaluation acts as an xtrnality if widsprad failurs may affct a firm s countrpartis, thus having a dtrimntal ffct on that firm s ability to rpay vn if it would b othrwis sound. If a sufficintly larg fraction of firms borrows in forign currncy, othrs (who would hav othrwis borrowd in local currncy) may find it optimal to do th sam as thy ar xposd to th risk of dvaluation through its ffcts via th ral conomy. Th possibility of countrparty failur thus affcts firms choic of whthr to borrow in th domstic or th forign currncy as wll as ntrprnurial ffort - thr is a complmntarity in th choic of borrowing dnomination - and may furthr xacrbat th liklihood and th svrity of a systmic crisis. Undr ths conditions, masurs aimd at limiting forign currncy borrowing may b wlfar improving. Th papr rlats to a broad litratur on how financial imprfctions contribut to shaping intrnational capital flows. In Bris and Koskinn (2002), forign currncy borrowing ariss bcaus govrnmnts find it optimal x post to bail out xporting firms by dvaluating th currncy, thrby rducing dbt ovrhang problms for highly lvragd firms with profitabl xport opportunitis. Our analysis is closst to Rancir, Tornll, and Wstrmann (2008) and Schnidr and Tornll (2004). As in thos paprs, forign currncy borrowing can hlp addrss an agncy problm and incrass output in tranquil tims at th cost of gratr risk of systmic criss. In thos paprs, howvr, crdit rationing hlps to rsolv th asymmtric information problm btwn borrowrs 4

7 and lndrs so that, in th absnc of bailout guarants, risk is corrctly pricd at th margin. Hr, whil risk is corrctly pricd in quilibrium, lndrs cannot condition thir pricing on an ntrprnur s ffort. As a rsult, systmic risk associatd with forign currncy borrowing can mrg vn in th absnc of bailout guarants. From this point of viw, our papr idntifis an additional mchanism linking systmic risk and conomic prformanc. Svral othr paprs focus on th intraction btwn liability dollarization and govrnmnt bhavior. In Jann (2009), a sovrign s inability to protct forign crditors rights rsults in a systm dominatd by short-trm loans. This short maturity structur provids govrnmnts with incntivs to nforc forign contracts. Howvr, it coms at th cost of risking liquidation (i.., a run ) triggrd by ngativ productivity shocks (a similar thm is in Tirol, 2003). In Vlasco and Chang (2004), forign currncy borrowing mrgs as a raction to th xpctation that th cntral bank will choos a fixd xchang-rat rgim. Thn, th financial instability that a dvaluation would caus through balanc-sht ffcts inducs th cntral bank to fight xchang rat flxibility, validating xpctations. Undr ths conditions, committing to xchang-rat flxibility, if fasibl, is wlfar improving. A similar analysis is in Chamon and Hausmann (2005). In Jann (2005), forign currncy borrowing is an outcom of domstic montary policy. If montary policy mitigats dfault risk in th privat sctor, firms will tnd to borrow in domstic currncy. If, on th othr hand, th montary nvironmnt dos not protct firms against low ralizations of thir domstic currncy incom, firms will borrow in forign currncy bcaus borrowing in domstic currncy can rsult in unbarably high ral dbt burdns if th xpctd domstic montary policy dos not matrializ x post. In Korink (2011), forign currncy dbt mrgs from an optimal portfolio choic problm with a risk prmium on local currncy dbt. Th advantag of local currncy dbt is that it mitigats conomic volatility. Local currncy dbt mrgs at low lvls of volatility of consumption and th xchang rat, as wll as whn risk prmia on local currncy dbt ar low. Most of th xisting work on th xplanations for forign currncy borrowing, such as that discussd abov, rly on firms trying to xploit som kind of govrnmnt policy that will thn bnfit thm. Ths govrnmnt distortions rang from xpctations of govrnmnt bailouts as in Schnidr and Tornll (2004) and Bris and Koskinn (2002) to dvaluations as in Rancir, Tornll, and Wstrmann (2008), Vlasco and Chang (2004), and Chamon and Hausmann (2005) 5

8 to montary policy as in Jann (2005). Our modl shows that, vn with risk nutrality for all partis, forign-currncy borrowing can b optimal in th absnc of any govrnmnt distortions whn othr financial imprfctions such as limitd liability and dviations from th law of on pric ar prsnt. All that is rquird is a lss-than-complt pass-through of xchang rat movmnts on local currncy prics. This maks xchang rat movmnts and th currncy composition of liabilitis rlvant for borrowrs solvncy. And, critically, it allows for larg currncy dprciations to triggr limitd liability protction. At th sam tim, th implid dviations from intrnational prics provid a justification for th assumption that local borrowrs maximiz thir profits in local currncy. In contrast, with a complt pass-through (whn local currncy prics adjust in tandm with th xchang rat) th currncy composition of liabilitis is irrlvant and a dprciation cannot, pr s, triggr a borrowr s insolvncy. Put diffrntly, all borrowrs would b naturally hdgd if xchang rat movmnts wr to pass through prfctly onto prics in th local currncy. Thr is ampl vidnc that xchang rat pass-throughs ar lss than complt in practic and that thr can b long-livd dviations from th law of on pric (s, for instanc, Gopinath t al., 2010, and th survy by Rogoff, 1996). This, togthr, with th vidnc of th larg balanc sht ffcts associatd with currncy dprciations in countris with prvasiv dollarization of liabilitis (s, for instanc, Calvo t al., 2004) suggst that th incomplt pass-through assumption undrlying our modl is ralistic. Whil w cast th analysis in th contxt of domstic- vrsus forign-currncy borrowing, w bliv that svral insights from our framwork apply mor broadly. In particular, th cntral finding that a rduction in idiosyncratic risk, and th rlatd fficincy gains, may com at th cost of gratr systmic risk applis to othr contxts. For instanc, considr th trad-off btwn fixdand variabl-rat dbt contracts. Undr normal conditions, short-trm rats will b lowr than longtrm ons, allowing for bttr borrowr incntivs, much th way that forign currncy borrowing dos in our modl. Howvr, such short trm contracts will lav firms xposd to potntially sharp incrass in thir dbt burdn, in a similar fashion to how dvaluation affcts firms in our modl. Whil intrst rat changs will typically b small and gradual, unlik dvaluation in our modl, thr ar cass in which vn marginal changs will imply paymnt difficultis for crtain borrowrs. For xampl, this kind of ffct was obsrvd for a larg fraction of subprim borrowrs whn thir 6

9 contracts rst, suggsting that th basic idas hr may b applid to a broadr contxt such as th maturity composition of dbt rathr than its currncy dnomination. A rlatd argumnt has also bn mad in th contxt of th us of short trm, or vn dmandabl, dbt by banks, whr a risky financing choic can hlp allviat a moral hazard problm that othrwis would lad to xcssiv risk taking or, quivalntly, too littl ffort in monitoring (s Calomiris and Kahn, 1991, and Diamond and Rajan, 2001). Th papr procds as follows: Sction 2 xamins a sris of stylizd facts and th mpirical litratur on forign currncy borrowing and financial criss. Sction 3 prsnts th main modl. Sction 4 xamins th cas with no moral hazard, whil Sction 5 studis th cas with moral hazard. Sction 6 xtnds th modl to th cas of countrparty risk. Sction 7 discusss th tradoff btwn total xpctd output and aggrgat risk. Sction 8 concluds and brifly discusss th policy implications of th modl. 2 Stylizd facts and mpirical vidnc In this sction, w rviw th mpirical litratur on forign currncy borrowing and financial criss, and prsnt som stylizd facts that ar consistnt with th prdictions of our modl. Much of th micro-lvl mpirical litratur on th dtrminants of forign currncy borrowing and th balanc shts ffcts that aris as a rsult of currncy dprciations whn firms borrow in forign currncy has focusd, du to data limitations, on larg and publicly tradd firms. Allayanis, Brown, and Klappr (2003) invstigat th capital structur of 327 larg, publicly tradd firms in East Asia around th tim of th East Asian financial crisis and collct data on thir local, unhdgd forign, and hdgd forign currncy dbt. Thy find that intrst rat diffrntials ar a ky dtrminant of th us of forign currncy dbt, and that th markt valu of firms that usd financial hdgs to synthtically convrt forign currncy dbt into local dbt wr hit particularly hard during th East Asian financial crisis. Blakly and Cowan (2008) study th currncy composition of th dbt of 500 publicly tradd firms in Latin Amrica during th priod 1990 to 1999, a priod of substantial xchang rat volatility in this rgion of th world, and find that th snsitivity of firms invstmnts dos not dpnd on th currncy composition of thir dbt bcaus firms tnd to match th currncy composition of thir dbt with th lasticity of thir incom to th 7

10 xchang rat. For a sampl of larg US firms, Kdia and Mozumdar (2003) find that firms issu forign currncy dbt mainly to hdg thir xposur to forign currncis. Similarly, Kloharju and Niskann (2001) find for a sampl of larg Finnish firms that hdging faturs prominntly in th dcision to borrow in forign currncy, with firms for which xports rprsnt a largr fraction of sals mor likly to rais forign currncy dbt. At th sam tim, thy also find that firms tnd to borrow in forign currncy whn th forign intrst rat is rlativly low, consistnt with carry trad xplanations. Many of ths rsults may b skwd by th focus on larg and stock-xchang listd firms that ar oftn in a bttr position to us financial hdgs compard to small firms (ithr bcaus of know-how or conomis of scal). In fact, Glos (2003), using data on 500 Mxican firms, shows that firm siz is a ky dtrminant of forign currncy borrowing in addition to imports and xports. Morovr, a larg fraction of ths firms has natural hdgs against xchang rat risks bcaus thy oprat in th tradabl sctor and hav significant forign currncy rvnus. Not surprisingly, much of this litratur finds rlativly small balanc sht ffcts associatd with forign currncy borrowing during currncy criss (for rviws, s Galindo, Panizza, and Schiantarlli, 2003, and Kamil, 2008). Brown, Ongna and Ysin (2009) ar th first to study th dtrminants of forign currncy borrowing for a rprsntativ sampl of firms that includs small firms using survy data on firms in Eastrn Europ. Thy find that firms that naturally gnrat a largr fraction of incom in forign currncy, such as xporting firms, ar mor likly to borrow in forign currncy, whil intrst rat diffrntials and xchang rat volatility do not xplain th us of forign currncy borrowing. Brown, Kirschnmann, and Ongna (2009) study a rprsntativ sampl of Bulgarian firms and find that forign currncy borrowing is not only drivn by dmand factors but is partly supply-drivn by banks that prfr to lnd in forign currncy to minimiz currncy mismatchs in thir balanc shts, vn whn borrowrs rqust loans in domstic currncy. Obviously, this still xposs th banks to crdit risks arising from balanc sht ffcts of thir borrowrs in cas of currncy dprciation. Rancir, Tornll, and Vamvakadis (2010) also study a rprsntativ sampl of firms in Eastrn Europ and focus on forign currncy borrowing by firms with no forign currncy incom. Thy 8

11 find that currncy mismatchs rduc intrst rats and nhanc growth of small firms in nontradabl sctors, thrby contributing to growth in tranquil tims, whil at th sam tim incrasing th probability of criss. Thy argu that th xpctation of govrnmnt bailouts in th vnt of a currncy crisis is on of th mchanisms that fostrs th us of forign currncy borrowing by firms that fac borrowing constraints. Th mpirical link btwn forign currncy borrowing and boom-bust cycls has ld countris to implmnt rgulatory policis to slow forign currncy borrowing during crdit booms, although ths policis hav typically mt with only limitd succss, mainly bcaus ths policis ar gnrally asy to circumvnt through, for instanc, dirct borrowing from abroad (Rosnbrg and Tirpak, 2008). At th macro-lvl, a larg mpirical litratur links banking and currncy criss to crdit booms accompanid by an ovrvalud currncy, although most of this litratur dos not distinguish btwn local and forign currncy borrowing (s, for xampl, Kaminsky and Rinhart, 1999). In an xcption, Iz and Lvy-Yyati (2003) show that th us of forign-currncy dbt can b linkd to macroconomic uncrtainty, including th rlativ volatility of domstic inflation and th ral xchang rat. Using data on forign currncy borrowing from th IMF s Vulnrability Exrcis Databas (not publicly availabl), Figur 2 shows a clar link btwn th dgr of forign currncy borrowing in th country and th occurrnc of banking criss, as dfind in Lavn and Valncia (2008), in a sampl of 114 countris. Forign currncy borrowing from banks in countris that xprincd banking criss ovr th priod 1970 to 2010 stood at 24.8 prcnt on avrag compard to only 15.0 prcnt in countris that did not xprinc a banking crisis ovr this priod. Ths mpirical findings ar consistnt with th prdiction from our modl that forign currncy borrowing xposs borrowrs to xchang rat risk and xposs lndrs to dfault risk from dvaluation-drivn balanc sht ffcts. On th dposit-taking sid of banks, D Nicolo, Honohan, and Iz (2003) show that dollarization is associatd with dpr financial dvlopmnt, spcially in high inflation nvironmnts. This is consistnt with th modl in Caballro and Krishnamurthy (2003) who argu that limitd financial dvlopmnt rducs th incntivs for forign lndrs to ntr mrging markts. Using data from Lvy-Yyati (2006) on th dgr of dollarization of dposits in th country, 9

12 Figur 2: Forign currncy lnding to GDP and occurrnc of banking criss, Figur 3 shows a clar link btwn dollarization and th occurrnc of banking criss, in lin with th findings on th link btwn forign currncy lnding and banking criss shown in Figur 2. Th ratio of forign currncy dposits in total dposits is about 29 prcnt on avrag for countris that xprincd a banking crisis ovr th priod 1970 to 2004 compard to only 18 prcnt for countris that did not xprinc a banking crisis ovr this priod. A rlatd litratur studis th link btwn xchang rat rgims and banking crisis. Burnsid, Eichnbaum, and Rblo (2001) argu that banks in countris with a fixd xchang rat rgim do not compltly hdg th xchang risk that ariss from th currncy mismatch btwn thir assts and liabilitis in anticipation of govrnmnt bailouts, and that such opn forign xchang positions maks banks pron to banking criss associatd with currncy criss. Empirical studis gnrally find that fixd xchang rat rgims, and spcially thos with hard pgs, ar mor pron to banking criss than flxibl xchang rat rgims or thos with adjustabl pgs, and that banking criss in fixd xchang rat rgims ar mor costly in trms of svrity of crisis and output losss (Eichngrn, 2002, Dmac and Martinz Pria, 2003, and Husain t al., 2005). Figur 4 shows that forign currncy borrowing is mor prvasiv in countris with fixd xchang rat rgims, which togthr with Figur 2 suggsts that th currncy composition of borrowing may b a ky drivr linking fixd xchang rat rgims to criss. In th contxt of our modl this association can b intrprtd as fixd xchang rats bing associatd with highr xpctd 10

13 No banking criss (0.18) (n=59) Banking criss (0.29) (n=76) Nots: Bars dnot th avrag dgr of dollarization ovr th priod across countris dpnding on whthr or not thy xprincd a systmic banking crisis during th priod , as dfind in Lavn and Valncia (2008). Dgr of dollarization is th ratio of forign currncy dposits in total dposits. Data on th occurrnc of banking criss ar from Lavn and Valncia (2008), Systmic Banking Criss: A Nw Databas, IMF Working Papr 08/224, and data on th dgr of dollarization ar from Lvy-Yyati (2006), Financial Dollarization: Evaluating th Consquncs, Economic Policy, January, pp Numbr of country obsrvations (n) btwn brackts. Figur 3: Dgr of dollarization and occurrnc of banking criss, dvaluations dspit highr probabilitis (α) of th xchang rat rmaining constant. 3 Modl Considr an conomy populatd by ntrprnurs/firms that invst $1 in risky assts that rturn y whn succssful and 0 othrwis. A firm s ffort dtrmins th probability of succss, q, at a cost c 2 q2. Th cost c rflcts country lvl institutional considrations that mak it difficult for firms to stablish good govrnanc structurs, such as bcaus of th poor nforcmnt of invstor rights. Firms hav no initial funds and nd to borrow in ordr to invst. Th loan contract spcifis th gross intrst rat (i.., on plus th nt intrst rat) r L to b rpaid by th borrowr. This is an opn conomy and firms can borrow in a comptitiv crdit markt, in ithr th domstic or a forign currncy. Th two currncis ar linkd by a standard intrst parity condition: r f = r f, whr r f is th gross (crdit) risk-fr intrst rat in domstic currncy, r f its quivalnt in forign currncy, th xpctd futur xchang rat chang, and th currnt xchang rat. For simplicity, w assum that xchang rat movmnts ar govrnd by a binomial distribution: th xchang rat stays constant with probability α, and dprciats by with probability 1 α. 7 7 W us th simplst xchang rat procss possibl to illustrat th main ffcts in th modl, which stm from limitd liability and agncy problms within th firm. Th main rsults can all b shown to hold for mor gnral distributions of xchang rat movmnts as long as thy includ th possibility of larg dvaluations which occur rlativly infrquntly. 11

14 No fixd xchang rat (10.7%) (n=237) Fixd xchang rat (19.3%) (n=97) Nots: Th rd (blu) bar dnots th avrag prcntag of forign currncy lnding to nominal GDP across country-yar obsrvations ovr th priod during which th country did (not) hav a fixd xchang rat rgim. Data on classification of xchang rat rgims from Rinhart, Carmn and Knnth Rogoff, 2004, Th Modrn History of Exchang Rat Arrangmnts: A Rintrprtation, Quartrly Journal of Economics 119(1): 1-48, and data on prcntag of forign currncy lnding to nominal GDP from th IMF s Vulnrability Exrcis Databas. W dfin fixd xchang rat rgims as xchang rat rgims with prannouncd or d facto pgs (classification cods 2, 3 or 4 in Rinhart and Rogoff). Data on xchang rat rgims ar avragd ovr th priod Numbr of country-yar obsrvations (n) btwn brackts. Figur 4: Exchang Rat Rgims and Forign Currncy Lnding, Thus, w can rwrit th intrst parity condition as r f = r f ( 1 + (1 α) ). (1) Without loss of gnrality, w normaliz th currnt xchang rat to 1, so that masurs not just th absolut chang in th xchang rat, but also th rlativ chang. 4 Optimal currncy dnomination with xognous projct risk W first considr th cas whr th firm must choos th dnomination of its dbt - domstic or forign currncy - but its projct s succss probability is xognous and indpndnt of ffort. Studying this issu allows us to isolat th ffct of currncy dnomination stmming purly from limitd liability without confounding it with ffcts rlatd to th moral hazard problm that may aris from th firm s ffort dcisions. Spcifically, w assum that a projct s probability of succss is fixd by an xognous paramtr q 0, and is not subjct to th firm s control. This is quivalnt to assuming that ffort q lads to projct succss q 0 + q, but that th cost c =, so that no firm would vr put in any additional ffort to incras projct succss abov q 0. Whn borrowing in domstic currncy, a firm s xpctd profits can b writtn as Π = q 0 (y r L ), (2) 12

15 which rflcts th fact that th firm is protctd by limitd liability and that th projct will only pay off with probability q 0. Whn th projct dos pay off, th cash flow from th projct is y, and th firm rpays th lndr th promisd amount r L. Sinc thr is no ffort for this bnchmark cas, th firm s xpctd profits ar thn just q 0 (y r L ). Th intrst rat on th loan has to rflct th risk associatd with th projct, q 0. With no moral hazard, and comptitiv crdit markts, w hav q 0 r L = r f r L = r f q 0. W can substitut this into th xprssion for xpctd profits to gt ( Π = q 0 y r ) f = q 0 y r f. q 0 It is usful to us th intrst parity condition to xprss this in trms of th forign risk fr rat, in which cas w gt Π = q 0 y r f ( 1 + (1 α) ). W can writ th firm s xpctd profits whn borrowing in forign currncy in similar fashion: Π f = αq 0 (y r L), (3) which rflcts th fact that th projct only pays off with probability q 0, but also th firm s rturn nt of loan rpaymnt is only positiv if th domstic currncy dos not dprciat. Th loan intrst rat must again rflct th lndr s xpctations of actually bing rpaid. Assum that thr ar no liquidation costs and that in cas of dprciation all of th projct s rvnu accrus to th lndr. In that cas, th lnding rat on th forign dnominatd loan must satisfy which givs Plugging r L back into th profit function givs αq 0 rl + (1 α) q 0 y + = r f, ( Π = αq 0 y r f (1 α) y + αq 0 α rl = r f (1 α) y αq 0 α +. ) = q 0 y + ( ) α + rf +. (4) W can now stablish th following. 13

16 Proposition 1 Whn thr ar no bankruptcy or liquidation costs, and whn uncovrd intrst parity (UIP) holds (as in (1)), firms prfr to borrow in forign currncy rathr than domstic currncy. Proof: Using th xprssions abov, profits for th borrowing firm ar highr whn borrowing in forign currncy than whn borrowing in domstic currncy if Π = q 0 y ( ) ( α + rf + > q 0y rf 1 + (1 α) ) = Π. Rarranging yilds q 0 y ( ) α + + r f ( + ) + > q 0 y r f ( + (1 α) ), which is quivalnt to r f > q 0y ( + ). (5) Th right hand sid is th xpctd valu of th output (q 0 y) xprssd in forign currncy (q 0 y) in th vnt of a dvaluation, so that w divid by +. This by assumption is lss than th forign risk fr rat, rf. Thrfor, (5) is always satisfid and borrowing in forign currncy is always prfrrd whn UIP holds and thr is no liquidation or bankruptcy costs. To undrstand th sourc of th gain associatd with borrowing in forign currncy, it is usful to compar th rsult from Proposition 1 to what would obtain if th borrowing firm wr subjct to unlimitd liability and could commit to fully rpay th lndr in all stats of th world. For as of xposition, normaliz th initial xchang rat so that = 1. Whn liability is unlimitd, th firm s payoff whn borrowing in forign currncy is Π = q 0 [( y r f ) (1 α) (1 + ) r f ] (1 q0 ) [ αr f + (1 α) (1 + ) r f ]. For simplicity, lt q 0 = 1, in which cas th xprssion bcoms Π = y rf (1 + (1 α) ) Not that this is idntical to th xpctd profit from borrowing in local currncy. Indd, absnt limitd liability or othr frictions, th intrst parity condition is constructd by imposing quality 14

17 in th firm s payoffs rgardlss of in which currncy it borrows. Notic, howvr, that ( ) Π Π α + 1 = y rf 1 + y + r f (1 + (1 α) ) ( ) = (1 α) rf y, 1 + which is positiv sinc w assumd that, in th dvaluation stat, th firm s rvnu is not sufficint to covr th forign risk-fr rat: r f > y 1+. Indd, this is th sam condition as in th proof of th proposition givn that with unlimitd liability th firm is indiffrnt btwn borrowing in ithr th domstic or th forign currncy. Thus, Π < Π, th profits that obtain undr limitd liability whn q 0 = 1. Th rason is that whn liability is unlimitd and th firm is rquird to rpay th sam amount, rf, in all stats of th world (i.., whthr dvaluation occurrd or not), th firm bars th ntir cost associatd with th dvaluation, which is (1 α). From th prspctiv of th borrowing firm, howvr, this is suboptimal sinc whn th currncy dprciats, which occurs with probability 1 α, it bars a cost 1 + pr unit it has to rpay, rathr than th cost (normalizd to 1) whn no dprciation has occurrd. In othr words, dprciation of th domstic currncy maks forign currncy dnominatd dbt rpaymnt mor xpnsiv for a firm that cars about its rturn or profits masurd in domstic currncy. Limitd liability, by contrast, allows th firm to dfault whn th dvaluation is larg, and shift paymnts it would othrwis hav to mak into stats of th world whr no dvaluation has takn plac and as a consqunc th firms viws it as chapr to rpay. Th lndr, who is risk nutral, is indiffrnt as long as in xpctation h rcivs th forign intrst rat rf. Indd, th largr is th probability that no dvaluation taks plac (i.., th largr is α), th asir it is for th borrowing firm to incrass its rpaymnt in no-dvaluation stats, and still obtain th rturn y from its invstmnt. Thus, for α larg nough, th savings obtaind from bing abl to rpay in low cost (in trms of th cost of buying forign currncy) rathr than in high cost stats allows th firm to borrow mor chaply in forign currncy. This gain only prsnts itslf as a rsult of limitd liability, which is triggrd whn a sufficintly larg dvaluation taks plac. 8 It is important to not as wll that, unlik much of th xtant litratur, hr th prfrnc for borrowing in forign 8 Dll Ariccia and Marquz (2015) provids a gnral tratmnt of this issu showing that a contract with statcontingnt paymnts ngativly corrlatd with th xchang rat lvl is prfrrd to a non stat-contingnt contract. 15

18 currncy ariss purly bcaus of limitd liability, and dos not dpnd on attmpts to xploit possibl govrnmnt actions, or coordinatd actions across firms. To our knowldg, ours is th only rational firm-lvl xplanation for liability dollarization of firms. Th xtrm rsult in Proposition 1 is prdicatd on th fact that thr ar no frictions associatd with a firm s dfault. In particular, thr ar no liquidation or bankruptcy costs that aris whn th firm fails to rpay. Mor ralistically, on may wll xpct a trad-off btwn th losss that may aris undr bankruptcy and th gains from bing abl to shift paymnts across stats. To introduc a ral cost associatd with possibl bankruptcy, assum going forward that undr dvaluation, th lndr rcivs nothing back from th borrowr. This would b consistnt, for instanc, with a vry larg dvaluation that lavs littl on th firm s balanc sht, and which subsquntly gts lost as part of bankruptcy procdings. Whil xtrm and also prhaps somwhat unralistic, as w show latr this assumption in fact biass against borrowing in forign currncy. Undr this assumption, w can stablish th following variant of Proposition 1. Proposition 2 For q 0 [0, 1], and kping th siz of th xpctd dvaluation, (1 α), constant, thr xists a valu α < 1 such that if th probability of no dvaluation, α, is gratr than α, firms prfr to borrow in forign currncy rathr than domstic currncy. Proof: Undr th assumption that thr is no rcovry whn a dvaluation taks plac and th firm dfaults, th promisd rpaymnt on th forign loan, rl, must satisfy Plugging this back into (3) yilds From 6 abov, Π > Π αq 0 r L = r f r L = r f αq 0. ( ) Π = αq 0 y r f = αq 0 y rf αq. (6) 0 αq 0 y r f > q 0y r f ( 1 + (1 α) ). As α 1, th lft hand sid convrgs to q 0 y rf, which is largr than th right hand sid, which is indpndnt of α givn that th xpctd dvaluation, (1 α), is hld constant. Thrfor, for α larg nough Π > Π. 16

19 Proposition 2 shows that whn th risk of dvaluation is sufficintly low, but th actual dvaluation if it occurs is larg, borrowing in forign currncy may b optimal rlativ to having dbt dnominatd in domstic currncy vn whn th dvaluation causs a loss for all partis. In othr words, pso problm conditions push firms toward borrowing in forign rathr than domstic currncy. This prfrnc for a currncy mismatch btwn assts and liabilitis, which occurs for th sam rason as in Proposition 1, ariss dspit th fact that a parity condition holds btwn th domstic and th forign risk fr rat, that th firm is risk nutral and thr is no agncy problm, and that thr is no govrnmnt action (.g., a bailout) bing xploitd. 5 Moral hazard and currncy choic In this sction, w allow firms to dtrmin th probability of succss of thir projcts, q, at a cost c 2 q2. As dscribd blow, bcaus of lvrag and th non-obsrvability of ffort, this introducs a moral hazard problm. 5.1 Domstic currncy borrowing Much as abov, whn a firm borrows in local currncy, it s xpctd profits can b writtn as Π = q(y r L ) c 2 q2, which is th sam as (2) xcpt that w subtract th cost of ffort, c 2 q2. Maximizing this with rspct to th lvl of ffort givs q = y r L. c Th intrst rat chargd on th loan has to rflct th lvl of risk associatd with th projct. Suppos that invstors or lndrs conjctur a lvl of ffort q C. Sinc lndrs ar comptitiv, this thn mans that q C r L = r f r L = r f q C. In quilibrium, lndrs blifs about th amount of ffort that will b supplid must b corrct, which mans that q C = q. W can substitut this into th xprssion for optimal ffort q to obtain q = y r f q c, and thn solv for q as { } y + y 2 4cr f q = min, 1, (7) 2c 17

20 whr (7) rflcts th fact that th positiv root that solvs for th quilibrium valu of ffort is Parto optimal (this can b asily shown). 9 Th constraint that q 1 rflcts th fact that q is th probability of projct succss and hnc cannot xcd 1. Throughout, w focus on th cas whr thr is an intrior solution for th firm s ffort, so that q < 1. It is straightforward to s that paramtr valus xist that guarant q < 1 in quilibrium. W also assum that financing is viabl, which amounts to assuming that q is a ral variabl. A sufficint condition to guarant this is that y 2 4cr f > 0. W com back to this issu latr whn w xplor th conditions undr which invstmnt, which ntails financing, is fasibl. W can now invrt th xprssion for optimal ffort to obtain r L = y qc, which, aftr substituting for q yilds r L = y c y + y 2 4cr f 2c = y 2 y 2 4cr f 2. Using th optimal valu q, w can writ th quilibrium xpctd profits as Π = q(y r L ) 1 2c (y r L) 2 = 1 c (y r L) 2 1 2c (y r L) 2 = 1 2c (y r L) 2. Substitut now for th optimal r L to obtain Π = 1 2c ( y + y 2 4cr f 2 Finally, again as abov, w can us th intrst rat parity condition, (1), to writ th quilibrium profits Π as a function of th forign risk fr rat, rf, and th xpctd xchang rat movmnt, ) 2. (1 α) : Π = 1 2c y + y 2 4crf ( 1 + (1 α) 2 ) 2. (8) Not that lvrag and th fact that risk cannot b pricd at th margin gnrats a moral hazard problm: In th absnc of limitd liability, th firms ffort choic would b q = y c q. Thn, sinc quilibrium ffort is blow its socially optimal lvl (and lndrs ar comptitiv), borrowrs would bnfit from a mchanism that allowd thm to rduc th moral hazard problm. 9 Whil in principl th ngativ root may also b part of a Nash quilibrium, w assum going forward that th Parto dominant solution - th positiv root - will b chosn. 18

21 5.2 Forign currncy borrowing W assum, as abov, that th possibl dprciation is larg nough that, in th vnt of a dprciation, th firm would go bust and dfault on its loan. Similar to abov, w can writ a firm s xpctd profit whn it borrows in forign currncy as Π = αq(y r L) c 2 q2, which is ssntially th sam as (3) xcpt that it subtracts th cost of ffort. W maximiz ths profits Π with rspct to ffort to obtain q = min {( y r L c ) } α, 1. (9) As abov, w will focus on th cas whr an intrior solution xists, so that q < 1. Now, sinc firms only rpay whn th currncy rmains stabl, for banks/invstors to b willing to lnd in forign currncy th intrst rat nds to compnsat thm for both th borrowr idiosyncratic risk, 1 q, and th dvaluation risk, 1 α. As abov, w assum that undr dvaluation, th lndr rcivs nothing back from th borrowr. This would b consistnt, for instanc, with a vry larg dvaluation that lavs littl on th firm s balanc sht, and which subsquntly gts lost as part of bankruptcy procdings, and biass against borrowing in forign currncy. Undr this assumption, w hav that, givn a conjcturd lvl of ffort q C and comptitiv crdit markts, th promisd rpaymnt on th forign loan, rl, must satisfy q C αr L = r f. From this w can solv for th quilibrium forign dnominatd loan rat, r L, as r L = r f q C α. As abov, w can substitut r L into th xprssion for th optimal ffort q givn in (9) and solv for q to obtain q = 1 2c ( ) yα + y 2 α 2 4crf. Noting that r L = y c q α, w can substitut for q and obtain ( ) c 1 r L 2c yα + y 2 α 2 4crf = y = y y 2 α 2 4cr α 2 f, 2α 19

22 which givs us th quilibrium loan rat whn th firm borrows in forign currncy. Givn th quilibrium loan rat r L and ffort lvl q, w can rplac ths in th xprssion for th firm s xpctd profits as Π = q (y r L)α c ( 2 ( q ) 2 y r ) = L α(y r c L)α 1 2c α2 (y r L) 2. Simplifying, Π bcoms Π = 1 2c yα + y 2 α 2 4crf 2 2, (10) which again xprsss th firm s quilibrium profits as a function of th forign risk fr rat. 5.3 Equilibrium dbt currncy dnomination W can now study undr what conditions firms prfr to borrow in forign rathr than domstic currncy whn projct succss is ndognous. W stat th following rsult, which rplicats th rsult in Proposition 2 for th cas whr projct succss is ndognous. Proposition 3 Kping th siz of th xpctd dvaluation, (1 α), constant, whn q is ndognous thr xists a valu α < 1 such that if th probability of no dvaluation, α, is gratr than α, firms prfr to borrow in forign currncy rathr than domstic currncy. Proof: A firm will prfr to borrow in forign currncy if Π > Π. Using (8) and (10), w can writ this inquality as 1 yα + y 2 α 2 4crf 2c 2 2 > 1 2c y + y 2 4crf ( 1 + (1 α) 2 ) 2. ( yα + y 2 α 2 4crf y > y + 2 4crf 1 + (1 α) ). (11) If (11) is satisfid, thn borrowing in forign currncy will b optimal for th firm. From hr, on ss that as α and incras so as to kp (1 α) constant, hnc kping th domstic riskfr rat constant, Π f incrass whil Π rmains constant. As α 1, th lft hand sid convrgs to y + y 2 4crf, which is strictly gratr than y + y 2 4cr f sinc rf < r f whnvr thr is a positiv risk of a dvaluation. 20

23 Much lik in Proposition 2, Proposition 3 stablishs that an incras in th siz of a larg dvaluation that occurs with only a small probability - a pso-problm - favors forign currncy borrowing. Th rsult stms from two ffcts associatd with limitd liability. Th first is that discussd in th prvious sction: limitd liability allows th firm to shift forign-dnominatd dbt paymnts from th dvaluation stat in which thy ar xpnsiv to th non-dvaluation stat in which thy ar chap. Th scond ffct oprats through a rduction in moral hazard. Borrowing in forign currncy has two ffcts on a firm s ffort. By lowring th intrst rat th firm has to pay whn succssful, it lads to gratr ffort (lowr risk taking). At th sam tim, howvr, by xposing th firm to dvaluation risk, it has th opposit ffct. Th rduction in intrst rat is proportional to th xpctd dprciation of th local currncy and invrsly proportional to th probability of dprciation; which also dtrmins th incras in dfault risk. It follows that whn th probability of dprciation is low (larg α), but th xchang rat movmnt conditional on dprciation,, larg, th nt ffct from borrowing in forign currncy on firms ffort is positiv. W show this formally in th following corollary: Corollary 1 Whnvr it is optimal to borrow in forign currncy, so that Π > Π, th firm also xrts mor ffort and rducs risk mor whn borrowing in forign currncy than whn borrowing in domstic currncy: q > q. Proof: Whn th firm borrows in domstic currncy, optimal ffort is givn by q = y + ( ) y 2 4cr f y + y 2 4crf 1 + (1 α) =. 2c 2c By contrast, whn it borrows in forign currncy, optimal ffort is yα + y 2 α 2 4cr q f =. 2c From this, q > q yα + y 2 α 2 4crf y > y + 2 4crf which is th xact sam condition that guarants Π > Π. ( 1 + (1 α) ), (12) Th corollary stablishs an quivalnc btwn a firm s ffort and its optimal choic of financing arrangmnt. Spcifically, it stablishs that th prfrrd dnomination of dbt, in th sns 21

24 of maximizing th firm s profit, is also th on that lads to gratr ffort. Not that, as bfor, th rsult holds prcisly bcaus th domstic borrowr valus his profit - and hnc consumption - in trms of his domstic currncy, whras th lndr valus rpaymnt xprssd in trms of th forign currncy. In othr words, it dpnds on th assumption of an incomplt xchang-rat pass-through, an issu that has bn shown to b rlvant mpirically (s Gopinath t al., 2010). It can radily b shown that if pass-throughs wr always complt, th currncy composition of liabilitis would b irrlvant in this modl. On simpl intrprtation of this condition is as a fixd xchang rat rgim whr a chang in th forign risk fr rat, r f, is immdiatly rflctd onto an qual chang in th domstic rat r f bcaus of th fr flow of capital and th fact that with a (rlativly crdibl) fixd xchang rat rgim xpctations of a dvaluation will not b affctd by th chang in th forign intrst rat. By contrast, if α is low but th xpctd dprciation is also small (to maintain consistncy with th intrst rat parity condition), thn th risk of bankruptcy costs and lowr ffort associatd with forign currncy borrowing imply that profits ar highr whn borrowing in domstic currncy. As abov, this can b intrprtd as a flxibl xchang rat rgim, whr adjustmnts in th xchang rat ar mor frqunt (i.., lowr α) but also typically smallr. W not, howvr, that (11) is writtn ntirly in trms of th forign rat rf, maning that throughout w ar assuming that parity is maintaind by adjustmnts in th domstic risk fr rat r f. In othr words, th xrcis conductd hr cannot radily b intrprtd as rprsnting a flxibl (i.., fully floating) xchang rat rgim. W discuss this cas in mor dtail latr. As a final point, w show that th assumption that undr dvaluation th lndr rcivs no rpaymnt actually biass th firm against forign borrowing. Suppos that instad undr dvaluation thr is som rsidual amount lss than what is promisd to th lndr, rl, and which th lndr can rcovr in cas of dfault. Th xprssion for th firm s profit will rmain unchangd sinc undr dvaluation, which occurs with probability 1 α, thr will still b nothing lft for th firm. Howvr, sinc th lndr rcovrs somthing, th quilibrium loan rat rl should b lowr, ( ) ctris paribus. Givn that th optimal ffort dcision for th firm is givn by q y r = L c α, this implis that q will b highr, so that th firm s projct is mor likly to pay off. This has an additional fdback ffct onto rl sinc th loan rat will also b lowr whn th probability of full 22

25 rpaymnt, q, incrass. Both of ths ffcts togthr imply that th firm s quilibrium xpctd profit Π will b highr whn th lndr obtains som rcovry in cas of dvaluation. Thrfor, th assumption w usd abov, that no such rcovry xists, in fact biass our rsults against th optimality of forign dnominatd borrowing. 5.4 Crdit rationing So far w hav assumd that th paramtrs ar such that crdit markts clar. Yt, our modl admits crdit rationing (a la Stiglitz and Wiss, 1981). This occurs whn th cost of ffort c is larg nough that moral hazard prvnts lndrs from braking vn. To s this, considr that th condition for a firm to obtain crdit in domstic currncy is that y 2 4cr f 0. In contrast, if th firm wr abl to commit to a crtain lvl of ffort, it could obtain crdit undr th lss stringnt condition y 2 2cr f 0. This raiss th qustion of whthr forign currncy dnominatd loans rlax th borrowing constraint for firms/countris that might hav bn othrwis rationd out, givn that, undr crtain conditions, it raiss thir quilibrium ffort. W show that this is indd th cas in th following rsult. Dfin c as th maximum ffort cost such that firms can obtain crdit domstically. 10 Likwis, w us c to dnot th maximum ffort cost such that borrowrs can obtain forign currncy dnominatd dbt. Formally, for borrowrs/conomis with highr ffort costs (7) and/or (9) do not admit a ral solution. Proposition 4 Kping th xpctd dvaluation, (1 α), constant, whn th risk of dvaluation is sufficintly low (i.., α is larg) but th siz of th possibl dvaluation is larg ( w hav c > c. is larg), Proof: Th marginal borrowr in domstic currncy is on for whom y 2 4cr = 0, which aftr som rarranging dlivrs th following thrshold valu of c: c = 4r f y 2 ( 1 + (1 α) Th quivalnt thrshold valu for forign currncy borrowing is c = α2 y 2 4rf. 10 Mor prcisly, givn that th cost of ffort q is c 2 q2, c is th thrshold valu of th paramtr for th cost function abov which (i.., for c > c) firms ar unabl to obtain crdit. ). 23