The Dominant Currency Financing Channel of External Adjustment

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1343 May 2022 The Dominant Currency Financing Channel of External Adjustment Camila Casas, Sergii Meleshchuk, Yannick Timmer Please cite this paper as: Casas, Camila, Sergii Meleshchuk and Yannick Timmer (2022). “The Dominant Currency Financing Channel of External Adjustment,” International Finance Discussion Papers 1343. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2022.1343. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

The Dominant Currency Financing Channel of External Adjustment * Camila Casas † Sergii Meleshchuk ‡ Yannick Timmer§ Abstract Weprovideevidenceofanewchannelofhowexchangeratesaffecttrade.Usinganovel identificationstrategythatexploitsfirms’foreigncurrencydebtmaturitystructureinColombiaaroundalargedepreciation,weshowthatfirmsexperiencingastrongerdebtrevaluation of dominant currency debt due to a home currency depreciation compress imports relativelymorewhileexportsareunaffected. Dominantcurrencyfinancingdoesnotlead toanimportcompressionforfirmsthatexport,holdforeigncurrencyassets,orareactive intheforeignexchangederivativesmarkets,astheyareallhedgedagainstarevaluationof theirdebt.Thesefindingscanberationalizedthroughtheprismofamodelwithcostlystate verificationandforeigncurrencyborrowing. Dominantcurrencypricingmutestheeffects ofdominantcurrencyfinancingonimportsrelativetoproducercurrencypricing. JELCodes: F31,F32,F41,G15,G21,G32 Keywords: Imports, Exports, Foreign Currency Exposure, Capital Structure, Exchange Rates, DebtRevaluation,Hedging *Thisversion: May11,2022. Mostrecentversion: here. WearethankfultoGustavoAdler,LauraAlfaro,MarkAguiar,AdolfoBarajas, ValentinaBruno,CatherineCasanova,LauraCastillo-Martinéz,AlainChaboud,LuisFelipeCéspedes,MariassuntaGiannettiGitaGopinath, BryanHardy,KilianHuber,OlegItskhoki,SebnemKalemli-Ozcan,NobuKiyotaki,SoleMartinezPeria,RobertMcCauleyToddMesser,Carolina Osorio-Buitrón,RicardoReis,AlexanderRodnyansky,AndrésRodríguez-Clare,TimSchmidt-Eisenlohr,LeonardoVillar,LilianaVarela,aswell asseminarandconferenceparticipantsattheAEA2021,CEPRInternationalMacroeconomicsandFinance(IMF)ProgrammeMeeting2020, TheBarcelonaGSESummerForumWorkshoponFirmsintheGlobalEconomy,7thInternationalMacroeconomicsWorkshop,INFER,13thFIW conferenceonInternationalEconomics,1stInternationalConference:FrontiersinInternationalFinanceandBanking,11thifoConferenceon MacroeconomicsandSurveyData(Poster),2ndDCJuniorFinanceConference,LACEALAMES2021AnnualMeeting,BIS,BankofEngland, IMF,BancodelaRepública,FederalReserveBoardforcomments.PartofthisworkwasconductedwhileCasaswasatBancodelaRepública. TheviewsexpressedinthepaperarethoseoftheauthorsanddonotnecessarilyrepresenttheviewsoftheIMF,itsExecutiveBoardorIMF management,theBancodelaRepúblicanortheFederalReserveBoardandtheFederalReserveSystem. †InternationalMonetaryFund.Email:ccasas@imf.org ‡InternationalMonetaryFund.Email:smeleshchuk@imf.org §FederalReserveBoard.Email:yannick.timmer@frb.gov

1 Introduction Theeffectofexchangeratemovementsontradehasbeenoneofthemostdiscussedtopicsin international economics. In traditional models, a depreciation of the home currency causes firms’exportstoexpandandimportstoshrinkduetochangesinrelativeprices(Mundell,1963; Fleming,1962;ObstfeldandRogoff,1995). Standardmodels,however,donottakeintoaccount thatadepreciationofthedomesticcurrencyreducesthenetworthoffirmswithdebtdenominatedinforeigncurrency,thusaffectingtradethroughadifferentchannel. Iffirmsborrowinforeigncurrency,adepreciationofthedomesticcurrencyincreasesfirms’ debt burden and cost of financing, which potentially induces a compression of both exports and imports. Depending on the relative elasticity of exports and imports, the effect of foreign currencyborrowingonthetradebalanceisambiguous. Inthispaper,weshowthatcorporateforeigncurrencyfinancingstronglyamplifiesthenegative effect of a domestic exchange rate depreciation on imports. Firms with a larger balance sheet exposure to the depreciation reduce their imports significantly and persistently more compared to less exposed firms. In contrast, more exposed firms do not reduce their exports during or after the depreciation, as exporters are hedged through their foreign currency revenues. There is substantial heterogeneity across firms in the effect of foreign currency borrowing onimports. First,firmsthatbothimportandexportdonotreducetheirimportsduetoforeign currencyborrowingduringorafterthedepreciation. Thisresultsuggeststhatexportscanprovideahedgeduringadepreciationperiod,especiallyiftheyarealsopricedinforeigncurrency (GopinathandItskhoki,2021).1 Second,importsoffirmswithforeigncurrencyassetholdings are unaffected by the debt revaluation. Third, firms that use foreign exchange derivative contracts are hedged and do not contract their imports as a result of foreign currency borrowing duringthedepreciationperiod. To shed light on the channel through which corporate foreign currency borrowing affects trade we construct a comprehensive dataset with firm-level data from Colombia, a country where trade is almost exclusively priced in US dollars. We merge highly disaggregated trade, loan, balance sheet and financial hedging data, and we use our unique dataset to study the trade response to a sharp and unexpected depreciation of the Colombian peso and its differ- 1Adepreciationincreasesthemarginalcostofthefirmforimporters,buthedgingoccurswhenfirmsexportin USDollars. 1

ential effect on exports and imports depending on firms’ financial heterogeneity in terms of foreigncurrencyborrowing.2 TheColombianpesodepreciatedbetween2014and2015mainly duetoalargedropinoilprices. Asthisshockwasaneventexogenoustoanyindividualfirm, wasnotaccompaniedbythecreditcrunchthatisoftenassociatedwithcurrencycrises,andincreasedfinancingcostsdifferentlyacrossfirms,itprovidesanideallaboratorytostudyhowdifferencesinforeigncurrencydebtaffectnon-commodityfirms’importsandexportsinresponse toadepreciationwhentradeanddebtaredenominatedinthesame(dominant)currency. Usinganovelidentificationstrategythatexploitsfirms’maturitycompositionofdominant currencydebtwestudytheeffectsofUSdollarborrowingonexternaladjustment. Thedetailed loan-level data allows us to compute the increase in debt repayments due to the interaction between exchange rate movements and the maturity structure. Firms that happened to have foreigncurrencydebtmaturingattheheightofthedepreciationfacedalargeincreaseintheir debt repayments. In contrast, for firms whose foreign currency debt matured predominantly justbeforethedepreciation,theirdebtrepaymentsremainedalmostunaffected. Sincethematurity structure can be seen as exogenous to exchange rate movements, the effect of foreign currencyfinancingontradecanlikelybeinterpretedascausal. We decompose firms’ financial exposure to the exchange rate depreciation into a liquidityandawealthshock. Foreigncurrencyleverageexposesfirmstothedepreciationthrougha liquidity shock, and through an additional wealth shock by increasing the debt burden for liabilities that do not immediately mature. Qualitatively, the effects of the increase in the debt burdenandtheliquidityshocksarethesame. Quantitatively,weestimatethataonestandard deviationincreaseinafirm’sforeignleverageratioisassociatedwithanadditionaldecreaseof 10.6%inimportsafterthedepreciation. Whileaonestandarddeviationlargerliquidityshock leadstoa4.7%largerdeclineinimports,aonestandarddeviationlargerwealthshockisassociatedwitha5.9%largerdeclineinimports. Asthewealthshockiscomputedasthesumofthe liquidityshockandthedebtrevaluationnotaffectingdebtrepaymentsduringthedepreciation, thedifferenceinthecoefficientbetweenthewealthandtheliquidityshockcanbeinterpreted astheeffectofthedebtrevaluation.Hence,theliquidityshockexplains80%ofthewealthshock whilethedebtrevaluationonlyaccountsfor20%. We further analyze the underlying financial frictions under the microscope by utilizing a 2Asforeigncurrency debtisalsoalmostexclusively denominatedinUSdollars, weusethewordsdominant currency,foreigncurrencyandUSdollarinterchangeably. 2

broadrangeofinformationfromourloan-leveldata.Non-exportingfirmsthataremorestrongly affectedbyadebtrevaluationthroughthedepreciationfacehigherinterestratesandaremore likelytobecomedelinquentontheirloansafterthedepreciation,evenaftercontrollingforobservedandunobservedtime-varyingcharacteristicsofthelendingbank.Thisisnotthecasefor exportingfirms. Tofurtherensureourresultsareindeeddrivenbytheincreaseinthedebtduetothedepreciationandnotmoregenerallyduetothemacroeconomicenvironment,weperformabatteryof robustnesstests. Weconductaplacebotestwherewereplacetheforeigncurrencywithdomesticcurrencydebtmeasuresandwedonotfindthatfinancialheterogeneityintermsofdomestic debtaffectedtheimportperformanceinresponsetothedepreciation.Wealsocontrolforother firmcharacteristicsthatcanaffectfirm-leveltradeflowssuchasage,sizeandprofitabilityand ourestimatesofthenegativeeffectofdebtdollarizationonimportsarenotaffected.Ourresults arealsonotspecifictothelargedepreciationthatoccurredinColombiain2014. Wecorroborateourmainresultsbyestimatingpanelregressionofimportsandexportsontheinteraction betweenforeigncurrencyleverageandchangeintheexchangerate. These empirical findings can be rationalized in a theoretical framework with Dominant CurrencyFinancing(DCF)underDominantCurrencyPricing(DCP).Webuildaparsimonious modelwithforeigncurrencyfinancingandfinancialfrictionsintheformofcostlystateverification. Borrowing in foreign currency increases the probability of default when the currency depreciates and creates a wedge between the risk-free and the firm-level interest rate. Higher borrowingcostsinduceacompressionofimports. Exportersfaceapositiveprofitabilityshock whichlowerstheirprobabilityofdefaultafteradepreciation. Thisnaturalhedgingmechanism shields exporters from higher interest rates and reduces import compression. Since DCP exportersfaceahigherprofitabilityshockthanproducercurrencypricing(PCP)exporters,exportingandfinancinginthesamedominantcurrencyislesscontractionaryfortradethanforeign currencyborrowingunderPCP. To quantify the magnitude of the effects, we conduct a simple back-of-the-envelope calculation and estimate the share of the total decline in imports attributed to foreign currency borrowing. Colombia’s exports and imports dropped sharply around the depreciation of the Colombianpeso. Importsdroppedby6billiondollarsandexportsdroppedby4billiondollars. Wemultiplythecoefficientofourpreferredspecificationwitheachfirm’sactualexposureand aggregate up. Our estimates imply that the dominant currency financing channel of external 3

adjustment explains around 17%, or 1 billion of the drop in imports but does not explain the dropinexports.3 Whilethispaperfocusesononespecificcountry,itcanbeseenasatypicalemergingmarket withtradepricedpredominantlyinadominantcurrencyandseveralgeneralimplicationscan be drawn from our analysis. First, under producer currency pricing (PCP) where exports are pricedindomesticcurrencyandimportsarepricedandfinancedinforeigncurrency,exporters would not be hedged against a liquidity shock when the domestic currency depreciates. Second,thecompositionofnon-exportingrelativetoexportingimportersiscrucialtounderstand themagnitudeoftheeffectofdominantcurrencyfinancingonexternaladjustment. Incountrieswherethevastmajorityoffirmsareimportersthatdonotexport(e.g. countrieswithlarge wholesaleorservicesectors),theeffectofdominantcurrencyfinancingislikelytobelargerthan forcountrieswithlargemanufacturingsectorsthatbothimportandexport. Third,incountries wherefirmshavesubstantialforeigncurrencyassetholdingsorwhereforeigncurrencyderivative markets are well developed, the effect of foreign currency borrowing on imports during a depreciationislikelytobemoremuted. Overall,theseresultsimplythatdominantcurrencyfinancinghasacontractionaryimpact on trade, but as the effect is only apparent for imports and not for exports, the effect on the tradebalanceisexpansionary. RelatedLiterature Ourpaperrelatestotheliteratureontheeffectofexchangeratemovementsontheexternaladjustment. RecentworkquestionstheconventionalwisdomofbothProducerCurrencyPricing andLocalCurrencyPricingandshowsthattradeismostlyinvoicedinafewvehiclecurrencies, and that the dollar plays a dominant role among them (Goldberg and Tille, 2008; Gopinath, 2015; Amiti et al., 2022). See Gopinath and Itskhoki (2021) for a review. As a result, after a domestic currency depreciation imports drop but exports remain relatively stable and the expenditureswitchingeffectismilderthaninthetraditionalMundell-Flemingsetting.4 3Thisnumbershouldbetakenwithcautionasitignoresgeneralequilibriumeffects. However,webelievethat thiscounterfactualexerciseservesasausefullowerboundbenchmarktounderstandthemacroeconomicrelevanceofourchannelasgeneralequilibriumeffectswouldlikelystrengthentheimpact. 4Theadjustmentprocessoftradedquantitiesdependsonthecurrencyinwhichpricesareset.Underproducer currencypricing(PCP),thelawofonepriceholdsandanominaldepreciation(allelseequal)increasesthedomesticpriceofimportsandreducesthepriceofexportsinthedestinationmarkets.Hence,itleadstoanimprovement ofthetradebalance.However,thereisevidencethatthelawofonepricefailstohold(ObstfeldandRogoff,2000). 4

Gopinathetal.(2020)showthattradeinColombiaisalmostexclusivelyinvoicedisUSdollars and therefore can be seen as a perfect laboratory to study the effects Dominant Currency Pricing(DCP).FollowingGopinathetal.(2020)weuseColombiaasalaboratorytostudyDominant Currency Financing (DCF) under Dominant Currency Pricing (DCP). Under DCF, firms donotonlypricetheirtradeinthedominantcurrencybutalsofinancetheiroperationsinthat samecurrency(Adleretal.,2020).5 Weempiricallyshowthatwhenfirmsnotonlypriceimports andexportsinUSdollarsbutalsoborrowinforeigncurrency,thecontractionaryimpactofexchangeratemovementsonimportsisevenstrongerwhileexportsarenotaffected.AddingDCF toDCPthereforeamplifiestheeffectofthedepreciationonthetradebalance. We also contribute to the large empirical literature on the effects of foreign currency borrowing. Several studies have found negative effects of foreign currency borrowing of firms on investmentwhenthedomesticcurrencydepreciates(Aguiar,2005;Kalemli-Ozcanetal.,2016; Hardy, 2018) or sales Alfaro et al. (2019). Desai et al. (2008) find that the effects of depreciationsonsalesandinvestmentareheterogeneousacrossfirms,andthattheabilitytoovercome financial constraints (as affiliates of multinationals can) play a decisive role in their differential response. Other studies have not been able to confirm these results and find no effect on investment(BleakleyandCowan,2008).6 NiepmannandSchmidt-Eisenlohr(2017a)showthat firmswithmoreforeigncurrencyloansaremorelikelytodefault. VernerandGyongyosi(2018) find negative consequences in response to household foreign currency borrowing after a depreciation.7 Christiano et al. (2021) show that dollarization can help risk-sharing and is not associated with larger negative effects during banking crisis. While all of these papers study the real effects of foreign currency borrowing on several outcomes, they ignore their external adjustmenteffects. Our paper is also closely related to the literature on the trade effects of financial shocks. Amiti and Weinstein (2011) and Niepmann and Schmidt-Eisenlohr (2017c) provide evidence in favor of contractionary export effects of trade finance shocks.8 Paravisini et al. (2014) and AsanalternativeBettsandDevereux(2000)andDevereuxandEngel(2003)proposedthatpricesareinsteadsticky in the currency of the buyer (local currency pricing, LCP). Under LCP a nominal depreciation has no effect on thepriceoftradedgoodsinthedestinationmarkets. Therefore,althoughitworsensthecompetitiveness,ithasa mutedeffectonthetradebalance.SeeLane(2001)forasurvey. 5GopinathandStein(2021)andBahajandReis(2020)alsostudythecomplementaritiesbetweenfinancingand invoicinginthesamecurrency. 6SeeGalindoetal.(2003)forasurvey.SeeBarajasetal.(2017),Restrepoetal.(2014)andEcheverryetal.(2003) forthecaseofColombia. 7ForthetheoreticaldimensionseeforexampleCéspedesetal.(2004),Krugman(1999)orDevereuxetal.(2006). 8SeeNiepmannandSchmidt-Eisenlohr(2017b);Loveetal.(2007);Schmidt-Eisenlohr(2013)formoredetails 5

Bruno and Shin (2019) show the that firms’ exposed to bank credit shocks reduced their exports.9 We differ from these papers in two dimensions. First, all papers focus solely on the export response of a tightening in financial constraints while we study imports and exports jointly. As pointed out by Blaum (2019), studying the response of imports and exports jointly is crucial to understand the external adjustment process of a country in response to currency movements. Second, none of these papers study shocks affecting firms directly through their borrowinginforeigncurrency.Ourresultsonthemutedeffectonexportssuggestthatcurrencyrelatedshocksthatincreasefirms’debtthroughadomesticcurrencydepreciationareverydifferentfromshocksthatpropagatethroughthebankingsystem. OurmodelisbasedontheTownsendcostlystateverificationmechanism(Townsend,1979) with several extensions. We allow firm’s net worth to be dependent on exchange rate through foreign currency denominated liabilities. As Bernanke et al. (1999) we study general equilibrium implications of net worth shocks but in an open economy context. Akinci and Queralto (2018) and Kohn et al. (2020) also study the implications of foreign currency borrowing. Akinci and Queralto (2018) show that foreign currency borrowing induces imports to drop more but exports to drop less in response to a tightening of US monetary policy. Kohn et al. (2020) moreexplicitlymodelalargedevaluationandconfirmthatfinancialfrictionsdonotcontribute largely to export dynamics. In contrast, our model highlights that exporting in dominant currency can partially offset the negative effects of the debt revaluation as it increases revenues when the currency depreciates. Moreover, we study the implications of foreign currency borrowingundertwodifferentpricingregimes: PCPandDCP. The restofthe paper is structured as follows. In section 2 we present a simple model with financialfrictionsanddominantcurrencyfinancing. Insection3wedescribethedataandthe construction of the variables. In section 4 we discuss the empirical specification. In section 5 wepresenttheempiricalresults. Insection6weconclude. 2 A Simple Model of Dominant Currency Financing In this section we outline a parsimonious model with financial frictions and foreign currency borrowing. We will show how financial frictions affect firm-level cost of borrowing and hence on trade finance shocks. Muûls (2015) shows that firms with credit ratings are more likely to be exporters and importers. 9BermanandBerthou(2009)andBermanandBerthou(2009)providecross-countryevidence. 6

optimal size and imports. A depreciation decreases firm’s net worth and increases the wedge betweentheriskfreeandfirm-levelinterestrate. Entryintoexportingrelaxessomeofthefrictions,reducingtheinterestrateandincreasesfirmsizeandimporting. Attheendofthesection wederiveseveralpredictionsthatwillguideourempiricalanalysis. 2.1 Setup Consider a simple one-period model.10 A firm starts with a stock of net worth denoted by A, reflectingforexamplesomeassets,givenbycash,minussomestockofdebt.Afirmmayborrow B tofinanceitsproductionfromabankthatonlyobservesfirm-levelproductivity,butdoesnot observe the firm’s product appeal, or demand shock denoted by δ. It is costly for a bank to verifyfirm-leveloutput,andthecostofverificationisequaltoγR whereR isfirm’srevenue.The optimalcontractwillhaveaformofthedebtcontract: thebankwilllendb unitsandrequirea repayment of B¯ whenever the firm does not default, see Townsend (1979) for the proof. If the firmdefaults,thebankreceives (cid:161) 1−γ(cid:162) R andthefirmwillget0. Assumethatrevenuestaketheform: R(δ,M)=ρ(M/pM,ϕ)δ whereM isfirm’simportexpenditures.11 Therevenuefunctionisdenotedbyρandisassumed tohavethefollowingproperties: Assumption1. Revenuefunctionisincreasingininput: ρ >0 M Assumption2. Revenueasafunctionofinputisincreasingatadecreasingrate: ρ >0,ρ <0, M M ∂lnρ ∂lnM Assumption3. Revenuefunctionelasticitywithrespecttotheinputislessthan1: ∂lnρ <1 ∂lnM Theassumptionslistedabovearefairlystandardandholdforavastclassofrevenuefunctions in settings with different production functions and market structures. Let the CDF and PDFofdemandshocksbegivenbyF(δ)and f(δ)with(cid:69)[δ]=1. Definethehazardrateas f(x) h(x)= 1−F(x) 10SeesubsubsectionC.1.1foramoredetailedexposition 11Tomakemodelparsimonious,weonlyassumeoneinputintoproduction,althoughthemodelcanbegeneralizedtoincludelabor,capital,andotherinputs. 7

Assumption4. TheCDFandPDFofthedemandshocksaresuchthatxh(x)isincreasingwitha non-decreasingelasticity. ThePDF f(x)anditsderivativeisboundedforallδ¯>0, f(0)=0and lim δ¯f (cid:48) (δ¯)<∞ δ¯→0 Thepreviousassumptionholds,forexample,forthelognormaldistribution. Atthebeginningoftheperiodafirmhasnetworth A=ad+afe, where af and ad arenet domesticandforeignassetsrespectivelyande isthenominalexchangerate(Colombianpesos perforeigncurrency). Forfirmsthatarenetborrowersinforeigncurrencyaf <0and ∂A <0. It ∂e canborrowB andspendsitsresourcesoncompositeinputpMM =A+B wherepM istheprice ofinput.12 Thenfirmpayoffisgivenby: πf =(cid:69)(cid:163)ρ(M)δ−B¯|δ≥δ¯(cid:164)(cid:161) 1−F(δ¯) (cid:162) whereδ¯isthecutoffforthedemandshockthatdeterminestheprobabilityofdefaultgivenby F (cid:161)δ¯(cid:162) .Iftherealizedvalueoftheshockisbelowthecutoff,thefirmdefaultsandthebankextracts alltherevenuesatsomecost. Thiscutoffisimplicitlydefinedbyaconditionthatequatesfirm’s revenueswithfixedpaymenttothebank: B¯=ρ(M)δ¯ Wecanthusexpressfirmpayoffas: πf =ρ(M)(cid:69)(cid:163)δ|δ≥δ¯(cid:164)(cid:161) 1−F(δ¯) (cid:162)−ρ(M)δ¯(cid:161) 1−F(δ¯) (cid:162) Bankpayoffisgivenby: πb=(cid:161) 1−γ(cid:162)ρ(M)(cid:69)(cid:163)δ|δ<δ¯(cid:164) F(δ¯)+ρ(M)δ¯(cid:161) 1−F(δ¯) (cid:162) 12Eventhoughourmainfocusisontheimportedgoods,atthispoint,forparsimoniousreasons,wewillassume thatthepriceofinputsisnotdependentonexchangerate.Relaxingthisassumptionwillnotqualitativelychange ourmainresultsthatfocusontheheterogeneouseffectsacrossfirmswithvariouslevelsofforeigncurrencyborrowing. 8

wherethebankextracts(1−γ)shareofaveragerevenuesbelowthecutoffwithprobabilityF(δ¯), andthebankgetsB¯ withprobability (cid:161) 1−F(δ¯) (cid:162) . Definethefollowingobjects: Ψ(cid:161)δ¯(cid:162)=(cid:69)(cid:163)δ|δ<δ¯(cid:164) ]F(δ)+δ¯(1−F(δ¯)) ζ(δ¯)=γ(cid:69)(cid:163)δ|δ<δ¯(cid:164) F(δ) These objects have an intuitive interpretation. Ψ(δ¯) is the share of profits that goes to the bank inclusive of monitoring costs and ζ(δ¯) is the share of revenues that goes to monitoring costs. Firmandbankprofitscannowberewrittenas: πf =(cid:161) 1−Ψ(δ¯)) (cid:162)ρ(M) πb=(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)ρ(M) Wemakethefollowingassumptiononthemarketstructureofthebankingsector: Assumption5. BanksborrowatarateR tolendtothefirms,areriskneutral,andoperateunder perfectcompetition. 2.2 Solvingthemodel Assumption5impliesthatallbankswillearnthesamereturnR ontheirlendingandearnzero profits,inotherwordsπb=R(pMM−A).Sincebanksarecompetingwitheachother,eachbank willpicksuchlendingB andrepaymentB¯ thatmaximizesfirm’sprofits. Notethatconditional on the lending amount, picking the amount of repayment is analogous to picking the cutoff productivityδ¯. Wecansettheoptimizationproblemofabankas: max ρ(M) (cid:161) 1−Ψ(δ¯) (cid:162) B,δ¯ s.t. ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)=RB pMM =B+A Whensolvingthismodel,weareinterestedinhowtheamountofimportedinputs,borrowing,andrepaymentdependonchangesinexchangerate. Considertwofirmsthatareidentical 9

inallrespectsbutforeigncurrencyleverage.Exchangerateaffectsthosefirmsthroughtheprice of imported inputs pM and net worth A. The former effect will be common across two firms, whilethelatterwilldependontheextentofindebtednessinforeigncurrency. Wewillnowderive several elasticities that will be useful in conveying the intuition of the main results in the modelLetε denotetheelasticityof x withrespectto y. Denotethenominalinterestrateby x,y 1+i = B¯ . Finally,letr ≡ p 1 M ρ M(M,ϕ) betheratioofmarginalrevenuesfromapesospentonim- B R portedinputsandarateofreturnonapesotothebank. Wecanshowthatintheoptimum,the followingcomparativestaticshold: 13 ε =1−ε >0, (1) 1+i,δ¯ Ψ−ζ,δ¯ ε ε = r,δ <0, (2) M,δ¯ ε ρ M,M ε A ε = ρ,M pMM <0. (3) δ¯,A ε (cid:179) pMM −ε (cid:180) −ε M,δ pMM−A ρ,M Ψ−ζ,δ Equation1istheelasticityofnominalinterestratewithrespecttothecutoffdemandshock. The higher this cutoff is, the more likely the firm is to default, and the higher the interest rate willthebankchargetocompensateforthehigherdefaultprobability.Theelasticityofimported inputswithrespecttothedemandshockcutoffisgivenbyEquation2. Higher cutoff implies higher interest rates and thus costlier borrowing for the firm, lower productionandlowerdemandforimportedinputs.FinallyEquation3establishesthattheelasticityofthedemandshockwithrespecttonetworthisnegative.Firmswithmoreownresources are less likely to default since they need to borrow less to produce the same amount as firms withsmallernetworth. Thelastequationiskeytounderstandingthedifferentialeffectsofan exchange rate depreciation for firms with different foreign currency leverage: the firms with higherforeigncurrencyborrowingwillexperienceasharperreductioninnetworth. Definetheratioofforeigncurrencyborrowingtoequityas f =−eaf . Thevariable f isposi- A 13TheproofsareprovidedinsubsubsectionC.1.3 10

tiveforfirmswhoarenetborrowersinforeigncurrency(af <0). Itcanbeshownthat: ε 1+i,e =−fε 1+i,δ¯ ε δ¯,A >0, if f >0, (4) ε =−fε ε <0, if f >0 (5) M,e M,δ¯ δ¯,A Nominalinterestrateincreaseswiththedepreciationofexchangerate,andtheeffectishigher forfirmswithgreaterindebtednessinforeigncurrency.Asaresultthosefirmsimportscontracts relativelymore. Wethusgetthefollowingimplicationsfromthemodelthatcanbetestedempirically:14 Hypothesis1. Adepreciationdecreasesimportsmoreforfirmsthatborrowmoreinforeigncurrency Hypothesis2. Adepreciationincreasesinterestratesmoreforfirmsthatborrowmoreinforeign currency Sofar,weonlyfocusedonthefirmsthatselldomestically.InsubsubsectionC.2.4weextend themodeltoallowforexporting. Considertwofirms,thathavethesameprobabilityofdefault before a depreciation, but one of them is an exporter. The exporting firm will face a positive revenueshockduetothedepreciation,makingitlesslikelytodefault.Inthelimitingcasewhen theprobabilityofdefaultapproacheszero,thebalancesheetchannelwillhavenoeffectoffirm’s behavior and thus the reaction of interest rates and imports will be muted.15 In this case, the exporterfacesalowerinterestratethananon-exportingfirm. Consequently, wewillhavethe followingtestableimplicationsfortheexportingfirms:16 Hypothesis 3. A depreciation does not decrease imports more for exporting firms that borrow moreinforeigncurrency Hypothesis4. Adepreciationdoesnotincreaseinterestratesmoreforexportingfirmsthatborrowmoreinforeigncurrency Colombianfirmsinvoicemostoftheirexportsindollars.17 Thisbehaviorcanberationalized byahighimportshareinproduction, strategiccomplementaritiesAmitietal.(2014,2019), or 14TheproofsareprovidedinthesubsubsectionC.1.3 15Generalizingtheresultforthecasesinwhichexportingfirmsarelesslikelytodefaultthannon-exportersbut thelikelihoodisstrictlylargerthanzeroisworkinprogress. 16TheproofsareprovidedinthesubsubsectionC.2.4 17SeeGopinath(2015). 11

capacityconstraints(Gopinathetal.,2010;Amitietal.,2022)increasingfirmsprofitsrelativeto PCP. In our model, this would translate to a relatively lower probability of default. DCP firms willthusbeevenmorelikelytoapproachthelimitofnodefaultatδ¯→0. Dominantcurrency pricinghencemutestheeffectsofthedominantcurrencyfinancing. 3 Data 3.1 Sources Wecompileddatafromseveralsources. Forourmainspecificationwemergedetailedinformationonloans,tradeflowsandoperationalvariablesfrombalancesheets.Ourcombineddataset coversfirmsaccountingforover90%oftradevalue.18 Commercial loans granted by Colombian banks and other financial institutions to firms comefromthecreditregistryfromSuperintendeciaFinanciera,theagencyinchargeofsupervising the financial sector.19 Every quarter Colombian banks report the amount outstanding andmaturitydateforeachloantheyissuetofirmsandtheeffectiveinterestrates. Importantly, theamountsareallocatedbetweenloansinforeignanddomesticcurrencies. UsingfirmtaxID, weconstructtheamountofoutstandingloansinforeignanddomesticcurrenciesforeachfirm andquarter. Thesedataareavailablebetween2005and2017. ThetradedatacomefromDANE,theColombianstatisticalagency. Thisdatasetcoversthe universe of foreign trade transactions and contains information on imports and exports at a very disaggregated level (by firm, country of origin/destination, product, and month of the year). Public data is available for 2008-2018. We construct total imports and exports by each firm and quarter. Balance sheet data comes from the Orbis global database.20 Originally, the data come from Colombian chambers of commerce and has standard variables from the balancesheetsandincomestatementsofColombianfirms(e.g. assets,liabilities,sales). Wehave data for 2004-2018. We complement our main dataset data with information on firms’ holdings of foreign assets and foreign exchange derivative contracts from Banco de la República, theColombiancentralbank. 18WehadaccesstoconfidentialdatawhileCasasworkedatBancodelaRepública. 19ByColombianbankswemeanbanksoperatinginColombia,regardlessofnationalityofownership. 20TheOrbisglobaldatabaseisprovidedbyBureauvanDijk,aMoody’sAnalyticscompany. 12

3.2 Measuresofexposureandshocks In this section we describe the measures of foreign currency exposure that we construct by merging the balance sheet and loans data. As foreign currency liabilities are the main driver offoreigncurrencymismatchesonColombianfirms’balancesheets, wefocusonforeigncurrencyloanstoconstructourmeasuresofexposure. Insubsection4.1weoutlinetheframeworkfortheregressionanalysis. We construct several measures of firm-level exposure to the exchange rate shock through their indebtedness in US dollars. First, we construct a measure of foreign currency leverage, whichisgivenbytotalloansinforeigncurrencydividedbyfirm-levelassets: (cid:80) L i∈ΛF i FCL = ft (6) ft A ft where ΛF is the set of firm’s f loans with Colombian banks denominated in foreign currency ft in period t, L is the outstanding amount of the loans, and A is the book value of assets of i ft firm f attimet. Thismeasurereflectstheimportanceofforeigncurrencydebtrelativetototal assetsofthefirm. However,itdoesnotdistinguishbetweentheloansofdifferentmaturity. We observefirmshavingforeigncurrencyloanswithquiteheterogeneousfuturematuritydatesin thedata. Wecanleveragethisinformationtostudywhetherfirmswithdebtthatisdisproportionatelydueovershorterhorizonswererelativelymoreaffectedduetotheliquidityshockthat theyfaceduetomuchhigherthananticipateddebtservicepaymentsinthenearrunincomparisontofirmswithlongerdebtmaturity,whofaceawealthshock.21 Inaddition,firmshaving different maturity dates face different shocks due to the difference in the Peso exchange rate between those dates. To draw the distinction between the measures of liquidity vis-a-vis the wealth shock and to take into account depreciation of Colombian currency over time, we use informationaboutthematuritydateofeachloanreportedbyColombianbanksandconstruct 21This identification strategy bears similarities with Duval et al. (2020) and Almeida et al. (2012) who exploit firms’debtmaturitystructurebeforethefinancialcrisis. AsimilarapproachhasalsobeenutilizedbyGiannetti andSaidi(2019). 13

thefollowingfollowingvariables: (cid:80) i∈ΛF 1 T(i)≤t(cid:48)L i ∆e t,T(i) LS ft,t(cid:48) = f,t (7) A ft (cid:80) i∈ΛF 1 T(i)>t(cid:48)L i ∆e t,t(cid:48) +(cid:80) i∈ΛF 1 T(i)≤t(cid:48)L i ∆e t,T(i) WS ft,t(cid:48) = f,t f,t (8) A ft whereT(i)isthematuritydayofloani,∆e isthedepreciationofPesobetweent andT(i), t,T(i) (cid:48) 1 T(i)≤t(cid:48) isanindicatorthatisequalto1iftheloanismaturingbeforet andzerootherwise,and 1 T(i)>t(cid:48) is an indicator that is equal to 1 if the loan matures after t (cid:48) . 22 The first measure, LS (‘liquidity shock’), measures by how much the value of the debt repayments as of t increases (cid:48) due to depreciation between times t and t given the actual depreciation as of their maturity (cid:48) date that falls between t and t . The second measure, WS (‘wealth shock’), adds the increase (cid:48) (cid:48) indebtthatmaturespastt usingtheexchangerateint . Inourbaselineempiricalstrategywe sett to2014Q1,asthedepreciationstartedin2014Q3(seeFigure1). Thisallowsustocapture (cid:48) firms’debtthatmaturesrightbeforeandrightafterthedepreciation. t issetto2015Q3, asit capturedtheheightofthesuddendepreciation. 23 3.3 Descriptivestatistics Figure 1 plots imports, exports and the Colombian peso between 2008 and 2018. The Colombian peso officially switched to a floating status in 1999. Since then, it can be considered a commodity currency, as fluctuations in the peso are strongly correlated with fluctuations in commodity prices. Moreover, commodity price fluctuations are exogenous to the economy; while mining products (mainly oil, coal and nickel alloys) represent a large share of total exports,outputissmallrelativetoworldmarkets. Forexample,in2014Colombia’soilproduction was 1.1% of world oil production but oil and oil products accounted for 52.8% of Colombian exports. Hence, the exchange rate depends heavily on oil prices but Colombia acts as a price takerintheoilworldmarket. In 2014, the price of oil plummeted. The spot price of WTI oil halved from over 100 dollarsperbarrelinJuly2014,to53.45dollarsbytheendoftheyearandreacheditslowestvalue, 22Paymentsneedtobeconvertedtopesosusingtheexchangerateofthepaymentdate(BancodelaRepublica, 2000). 23Resultsarenotsensitivetothechoiceoft andt (cid:48) .FigureA2showsthematuritydistributionofloans. 14

lessthan30dollars,inJanuaryandFebruary2016. Thepeso/USdollarexchangerateincreased fromroughly1850attheendofJuly2014to3435bymid-February2016. Importantly,thepeso depreciationwasnottheresultofamacroeconomiccrisis. AccordingtotheWorldBank’sWDI, annualGDPgrowthin2014-16was4.7%,3%and2.1%,lowerthanduringtheboomofcommoditypricesbutwellabovetheregion’saverage,unemploymentreachedits10-yearminimum,and indicatorsofthefinancialsystemhealthsuchastherateofcapitaltoassetsandtherateofliquid reservestoassetsremainedrelativelyunchanged.24 Thedepreciationoftheexchangeratecoincidedwithadropinbothexportsandimportsof Colombianfirms. Between2012and2014importsandexportswereremarkablystable. Colombianquarterlyexportsandimportsequaledroughly14billiondollars,around39%ofGDP.With the depreciation, trade dropped. While imports dropped from 16 billion dollars to around 10 billiondollarsbetween2014and2018,exportsdroppedfrom14billiondollarsto6billiondollarsin2016butthenrecoveringbackto10billionin2018.25 Table 1 shows summary statistics of firms in terms of their assets, exports and imports.26 Wesplitbetweennon-exporters, exportersandallfirms. Thereare14,618firmsinoursample whichdonotexportbutimport,and7,232exportandimport. Theaveragesizeofafirminour datasetis12 million USD. Average firm-level exports before the depreciationare 127,000 USD butdropto108,000USDafterthedepreciation. Firm-levelimportsdropfrom132,000USDto 118,000USDbetweenbeforeandafterthedepreciation. Themeasuresofexposuretodepreciationduetoindebtednessindollarsarecomputedas oftheendofthefirstquarterof2014,asdescribedinsubsection3.2. Table2showsthatforeign currencyleverageis4.4%fortheaveragefirmrangingbetween0.2%to11.6%betweenthe10th and 90th percentile. The average liquidity shock is close to zero as there are many firms that haveliabilitiesthatmaturebeforethedepreciationstarted. Afirmatthe10thpercentileofthe liquidity shock distribution even faces a negative liquidity shock. A negative shock reflects a decrease in the debt repayments due to the small appreciation before the large depreciation. A firm at the 90th percentile of the liquidity shock sees an increase of their debt repayments purelyduetothedepreciationthatis0.6%oftotalassets. Theaveragewealthshock, whichequalstheliquidityshockplusthedebtrevaluationafter 24SeeFigureA3fortheevolutionoftheoilpriceandthePeso.SeeFigureA4fortheevolutionoftheoilpriceand quarterlyGDPgrowth. 25Notably,exportsalsodecreasedifmeasuredinColombianPesosorinvolumes(netkilograms). 26Sinceimportregressionsarethemainfocusofourempiricalexercise,weonlyfocusonfirmsthateverimported inoursample. 15

2015Q3,equals0.8%oftotalassets. Evenforthewealthshockweseesomefirmswithnegative values, whichbenefit fromthe exchange rate movements before the depreciationstarted. For the comparison purposes, all measures of foreign currency exposure entering the regressions werestandardizedtobemean-zeroandhaveaunitstandarddeviation. 4 Empirical Strategy 4.1 Baselinespecification We use the unanticipated depreciation of the Colombian peso in 2014Q3 as a natural experimenttostudythedominantcurrencyfinancingchannelofaforeignexchangeratedepreciation ontrade.27 Forourbaselinespecificationweestimatetheequationoftheform: ln (cid:161) 1+Y (cid:162)=β×FCE ×1(t ≥t )+controls +(cid:178) , (9) ft f 0 ft ft where Y is either the firm-level exports or imports of firm f at time t.28 Our measure of foreign currency exposure, FCE f , is either FCL ft in t = 2014Q1, LS ft,t(cid:48) with t = 2014Q1 and t (cid:48) = 2015Q3, or WS ft,t(cid:48) with t = 2014Q1 and t (cid:48) = 2015Q3. 1(t ≥ t 0 ) is a dummy that equals oneinanyquarterduringorafterthedepreciation(t ≥2014Q3). Standard errors are double clustered at the firm and quarter level. The set of controls is given by firm and time fixed effects, and hence we interpret the β coefficient as a differential effectofthefinancialexposuretoexchangerateshocksonimportsorexports,withtimefixed effectsabsorbingtheleveleffectonimports(exports)ofallfirms. Thefirmfixedeffectscontrol forthetime-invariantcharacteristicsofthefirm,forexample,theaveragerelianceonimported inputs,size,etc. Itisworthmentioningthatdominantcurrencyfinancingisachoiceforfirms– thatis,firms’foreigncurrencydebtanditsmaturitystructurecanrespondtotheirexpectations 27Rodnyansky(2019)usesthedepreciationoftheRussianrubleduetotheexogenouscollapseinoilpricesin 2014toshowthatmoreproductiveexportersshrankrelativetononexportingfirmsintermsofdomesticrevenue, employmentandprofitabilityfollowingthedepreciation. ? usetheoilshockwithMexicandatatostudythedynamicsofcreditsupply. 28Notethatgiventheabovelogarithmictransformationofthedependentvariablecanintroducebiasesthrough theconstant.Were-estimateallregressionswithaninversehyperbolicsinetransformationofY,andfindqualitativelyandquantitativelyalmostidenticalresults. ExceptforsmallvaluesofY,thistransformationapproximately equalslog(2Y),orlog(2)+log(Y),whichcanbeinterpretedinthesamewayasastandardlogarithmicdependent variable.SeeBurbidgeetal.(1988),orChen(2013)foranapplication. 16

regardingexchangeratemovements.Althoughtheseshocksarenotoriouslyhardtopredict,this potentialendogeneitycouldbiasourresults. Weaddressthisconcernbystudyingthespecific case of an unanticipated, sudden depreciation that was unrelated to the economic situation in Colombia. Given the nature of this shock, we argue it is unlikely that firms’ debt maturity profile that led to variations in the above-defined liquidity shock is systematically correlated withobservedorunobservedcharacteristicsthatcouldbiasourresults. 4.2 ExogeneityofLiquidityShock Ourmainwayofidentifyingacausalimpactofhowforeigncurrencyborrowingaffectstradeis bytheconstructionoftheliquidityshock. Wemakeuseofthefactthatfirms’borrowingneeds tobepaidbackatspecifictimesduringthedepreciationperiod,andthereforearerelatedtoa moreorlessdepreciatedexchangerate. FigureA2showsthedistributionofmaturity,withthe weightedaverageoriginalmaturityataround5years. Asdescribedinsubsection3.3theColombianpesodidnotonlydepreciatestronglybutalso exhibitedlargevolatilitywhichexposesfirmstohighfrequencymovementoftheexchangerate duetotheirdebtmaturitystructure.Whileotherstudiese.g.Almeidaetal.(2012)orDuvaletal. (2020),havemadeuseofasimilaridentificationstrategy,ourapproachisstrengthenduetotwo reasons. First, Almeidaetal.(2012)andDuvaletal.(2020)usetheGlobalFinancialCrisisasa shock for tightening credit conditions and argue that for firms that had debt maturing during the crisis tightened even more. However, the Global Financial Crisis has also been associated with other factors that may discourage firms from investing, e.g. an increase in uncertainty. In our case, as the exchange rate depreciation was driven by a drop in the oil price that was neither caused by the Colombian economy nor affected the Colombian firms through other factorsotherthantheexchangerate, oureffectscanbelikelyattributedtothecausaleffectof the exchange rate. Second, we make use of a high frequency identification by using the date of the maturity instead of just the year or quarter, constructing an arguably more exogenous exposuretotheexchangeratemovements. Ifourargumentholdstrue,wewouldexpecttheliquidityshocktobeuncorrelatedwithobservedandunobservedcharacteristicsthatcoulddrivethedropinstrongerdeclineofimports for non-exporting firms that are faced with a larger liquidity shock. Table 3 shows the results fromacross-sectionalregressionoftheliquidityshockonvariousfirm-levelvariables. Column (1)showsthatwedonotfindastatisticalsignificantrelationshipbetweentheaveragematurity 17

offirms’loansbeforethedepreciationandtheliquidityshockinalinearregression. Whileby constructionthereisanon-linearrelationshipbetweenthematurityofthefirms’loansandthe liquidityshock,wecanruleoutthatfirmsthatonaverageborrowshorterarealsofacedwitha largerliquidityshock.Theabsenceofacorrelationcanbeexplainedbythehighvolatilityinthe exchangerateduringthedepreciation,andillustratedbythefollowingexample. Imagine two firms that borrow at the maturity of 5 years. One firm borrows at the 16th of December2009whiletheotherfirmborrowsatthe21thofDecember2009. Bothfirmsneedto repay1millionUSDollarsonthedaytheloanmatures. Firm1,whichborrowedonthe16thof Decemberneedstopayback2.413billionpesoswhiletheotherfirmthatborrowedonthe21th ofDecemberneedstopayback2.294billionpesos(5%less). Thisexampleshowsthateventhedayoftheweekwhenaloanisscheduledtobepaidback can explain a large fraction of the repayment in local currency. It also shows that a shorter maturitystructure(Firm1)isnotnecessarilycorrelatedwithalargerliquidityshockduetothe largevolatilityintheexchangerateduringthedepreciationperiod,asweshowinTable3. Therefore,asthedayoftheweekwhentheloanhasbeenissuedcanbeseenasexogenous andwecancontrolforthematurityoftheloans,ourregressioncoefficientscanlikelybeinterpretedascausalpartialequilibriumeffects. 4.3 Dynamicresponse Equation 9 estimates the average effect of an additional unit of foreign currency exposure on firm-level imports after the depreciation. However, one might expect the effect to build up and/ordeclineovertime. Hence,inoursecondsetofempiricalexercises,insteadofinteracting themeasureofdominantcurrencyfinancingwithapost-depreciationdummy,weinteractthe measurewithadummyforeachquarter: ln (cid:161) 1+Y ft (cid:162)= (cid:88) β t(cid:48) ×FCE f ×1(t =t (cid:48) )+controls ft +(cid:178) ft , (10) t(cid:48) where Y is either the firm-level exports or imports of firm f at time t. This helps us to pin downthetimingoftheresponseofexportsandimportsinresponsetotheinteractionbetween thedepreciationandthefirms’debtdollarization. Inaddition,weexaminethetradeactivities of firms as a function of their debt dollarization just before the depreciation. This pre-trend 18

analysis helps us to shed light on the question whether firms with more financial exposure to the exchange rate depreciation are fundamentally different from other firms. For example, if firms with more foreign currency debt before the depreciation already started reducing their imports,theinteractionterminEquation9couldjustreflectadownwardtrendintheimports ofthesefirms. 5 Results 5.1 BaselineRegression Table 4 presents the results for Equation 9 for imports. Columns (1)-(3) use foreign currency leverage as the financial exposure to the exchange rate depreciation. When considering all 21,850firms,weseeastrongnegativeimpactoftheinteractionbetweenforeigncurrencyleveragein2014Q1,justbeforethedepreciationstarted,andthedepreciationdummy. Hence,firms thathadalargershareofforeigncurrencydebtbeforethedepreciationimportedlessafterthe depreciationthanbefore. Quantitatively, a one standarddeviationlarger share offoreigncurrency debt29 is associated with a 10% stronger decline in imports. However, this effect masks large heterogeneity across different firms. Around two thirds of all firms do solely import and do not export. We show the regression for only these firms in column (2). The regression coefficient roughly doubles. For non-exporters a one standard deviation larger share of foreign currencydebtisassociatedwitha20%strongerdeclineinimports. Forfirmsthatimportand export,theeffectisonly1.8%andnotstatisticallysignificant.30 Columns (4)-(9) repeat the same exercise but decomposes the increase in leverage due to thedepreciationintoaliquidityandawealthshock. Columns(4)-(6)reportthecoefficientsfor the liquidity shock which captures the increase in debt repayments due to the movements in theexchangerate. Theliquidityshockisdefinedsuchthattheliquidityshockispositivewhen the firm sees an increase in their debt repayments due to the depreciation and negative if the debtrepaymentsfall.Thedebtrepaymentscanfallforthefirmifithasveryshort-termmaturity 29Onestandarddeviationofforeigncurrencyleverageis3.2% 30Theeffectisextremelyrobustacrosssectors. FigureA5displaysthenumberoffirmsintheregressionsample bysector. FigureA6estimatestheregressionincolumn(1)separatelyforeachbroadsectorandplotsthecoefficient,withallbeingnegative. FigureA7estimatesasector-specific(2-digitNAICS)codecoefficientandplotsthe coefficientagainsttheshareofexportersinthesector. Consistentwiththedifferenceinthecoefficientbetween column(2)and(3),sectorsthathavealargershareofexportersaremoreshieldedfromthenegativeimporteffect offoreigncurrencyborrowing. 19

debtthatmaturesbeforethedepreciationhappened.31 Theeffectoftheliquidityshockonimportsisaroundonehalfoftheforeigncurrencyleverage. Inparticular,aonestandarddeviationlargerliquidityshockleadstoadeclineinimports relativetothepre-depreciationperiodbyaround5%. Asforforeignleverage,theeffectissolely driven by non-exporters for which the response is around 7% and statistically signifanct. Exportersreducetheirimportsby2.5%moreiftheyfaceaonestandarddeviationlargerliquidity shockbutthecoefficientisnotstatisticallysignificant. Lastly,columns(7)-(9)showtheaverageresponseofimportsinthedepreciationperiodasa functionofawealthshock,whichcombinestheliquidityshockduringthesharpdepreciation withtheshocktowealththatoccursduetodebtrevaluationforlaterperiods. Asabove, firms thatarefinanciallymoreexposedtothedepreciationoftheexchangerateduetoawealthshock compressimportsrelativelymorecomparedtootherfirms.Thispatternissignificantlystronger forfirmswhichimportanddonotexport,ratherthanfirmsthatparticipateinbothactivities.32 Intermsofeconomicsignificance,aonestandarddeviationlargerexposuretotheexchange ratedepreciationthroughtheirdebtdollarizationisassociatedwitha12%largercompression ofimportsifmeasuredbythewealthshockfornon-exporters. Thenumbershrinkstoaround 1%forexportersandisnotstatisticallysignificant. Table 5 repeats the same exercise as Table 4 but for exports. The number of observations significantly drops as many more firms import than export. In our sample only around 7,200 firmsexport. TheresultsforexportsdifferstarklyfromtheresultsshowninTable4. Whilethe coefficients are mostly negative, the estimates are much smaller and not statistically significant. Neitherforeignleveragenorthewealthorliquidityshockindicatesthatfirmswithlarger 31InTable3wetestwhethertheliquidityshockiscorrelatedwithotherpre-existingobservablecharacteristics (in2013)atthefirm-level,suchastheiraveragematurity,theirprofitability,theirage,andfindthatfirmsthathave largerliquidityshocksareextremelybalancedintermsofothercharacteristics,whichsuggestthatunobservable characteristicsarealsounlikelytobecorrelatedwiththeliquidityshock. Whencontrollingfortheseobservable characteristicsinteractedwiththepostdummy,aswellassector-timefixedeffects,wefindthatourbaselinecoefficient remains extremely stable, suggesting no substantial bias in the estimated coefficient, see Table A1. In unreportedresultswealsouseameasureofrevenueproductivityasacontrolandtheresultsremainunchanged. However,duetodataconstraintsweareonlyabletoestimateproductivityforarelativelysmallsampleoffirms. Wealsofindthattheresultsarerobusttocontrollingforfirmsize,whichisagoodproxyforstrategiccomplementaritiesAmitietal.(2014,2019). 32Weconfirmthatthisresultisrobusttofocusingexclusivelyonquantities.TableA2showsthatfirmsthathave beenexposedmorecontractedthenumberofunitsimportedaswellthekilogramsimportedmore.TableA3shows thatexposedfirmsalsoreducetheuniquenumberofproductsaswellasthediversityofcountriesfromwhichthey import.Wetestformallyforthestatisticaldifferencebetweennon-exportersandexportersinTableA4.Wecannot rejectthehypothesisthattheeffectofbeinganexporteroffsetstheeffectfornon-exporters. Thisresultisalso robustwhenforotherpotentialconfoundingfactorsthatarecorrelatedwithexporterstatusarecontrolledfor. 20

dominantcurrencyfinancingfacealargerdropinexports.33 ThisresultisdifferentfromevidenceinParavisinietal.(2014)orAmitiandWeinstein(2011) who show that firms that face financial shocks reduce their exports. Why is the evidence different? Paravisini et al. (2014) and Amiti and Weinstein (2011) study shocks to banks that are unrelatedtoexchangeratemovementstowhichfirmsarelikelynothedgedagainst.Incontrast, in our case firms that have dollarized debt face a negative shock when the domestic currency depreciates,butthisnegativeshockisoffsetbyanincreaseintheirexportrevenuesastheyare almostexclusivelypricedindollars(Gopinathetal.,2020). 5.2 DynamicsResponse In this subsection we shed light on the dynamics of the response in imports and exports over time. As in Table 4, we split the firms into importers that do not export and firms that import andexport. Figure2showstheevolutionofimportsasafunctionofaonestandarddeviation larger financial exposure to the depreciation and the Peso/US Dollar exchange rate. The behaviorofimportsisnotstatisticallydifferentbetween2012Q3and2014Q2. Atthesametime the peso/US dollar exchange rate was relatively stable at around 2000 pesos per dollar. The factthatimportsdidnotbehavedifferentlyasafunctionofdominantcurrencyfinancinginthe period before the depreciation serves as a useful placebo test. Once the peso starts to depreciate in 2014Q3, firms with a larger financial exposure to the exchange rate depreciation contract their imports significantly more. The effect of dominant currency financing on imports fornon-exportersaccumulatesovertime,whichstressestheimportanceofstudyingtheeffect inadynamicsettingthatallowsfirmstorespondwithalag. Threeyearsafterthedepreciation the effect reaches a coefficient of -0.4. This coefficient implies that firms with a one standard deviationlargerfinancialexposuretotheexchangeratedepreciationcontractimportsby40% morerelativetobeforethedepreciationperiod. Figure3showstheevolutionofimportsforexportersinresponsetoaonestandarddeviation largerdominantcurrencyfinancing. Theestimatedcoefficientisclosetozerobothbeforeand afterthedepreciationperiodandisnotstatisticallysignificant. Thisresultsuggeststhatfirms that export and import were insulated from the dominant currency financing channel due to higherrevenuesinlocalcurrencysinceexportsarepricedinUSDollars. 33Thisresultisrobustwhencontrollingforotherfactors,seeTableA5. 21

Figure 4 plots the estimated coefficient for the interaction between financial exposure to theexchangeratedepreciationandquarterdummiesfortheexportresponse. Asinthecaseof importsbyexportersthelineforexportsisentirelyflat. Thelogicisthesameasforimporters thatexport. Whileitmaybethecasethatexportsdeclineinresponsetoafinancialshockasexportersfacedifficultiestofinancetheirworkingcapital,thischannelisnotatworkhereasfirms thatexportarehedgedduetotheirhigherdomesticcurrencyrevenuesgiventheappreciation oftheUSdollarinwhichtheirexportsarepriced. FigureA1jointlyplotsthecoefficientsofbothexportersandnon-exporters,highlightingthe starkdifferenceintheirimportresponsesduringthedepreciationepisode. 5.3 Hedging Firmsthatborrowinforeigncurrencycanbehedgednaturallythroughtheirexportrevenuesin dollars, or through other sources, such as foreign exchange derivatives and through the accumulationofdollarassets.34 Whenfirmsexperienceadebtrevaluationindomesticcurrencydue toadepreciationofthedomesticexchangerateduetoliabilitiesinforeigncurrency,thevalue of foreign currency assets appreciates at the same time in domestic currency terms. Hence, the assets held in foreign currency can serve as a hedge to dampen the negative effects of the increaseinthedebtburden. Firmsthataccumulatedforeigncurrencyassetsbeforethesharp depreciation of the peso could have drawn down their foreign currency assets to either repay their revalued debt or pay for dollar-priced imports, dampening the negative effect of foreign currencyborrowingonimportsduringandafterthedepreciation. Similarly, firms can tap the derivatives markets to hedge their exposure to exchange rate fluctuationbyengaginginaforwardcontractthatlocksintheexchangerateforthepurchaseor saleofacurrencyonafuturedate. Forexample,ifafirmisexpectingtorepaydebtindollarsat acertaindate,itcanlockinthecurrentforwardexchangeratefortherepaymentdatetohedge againstadollarappreciation. Ontheagreeddate,thecounterpartsexchangetheamountsthey hadcommittedto. Incaseofanunexpectedpesodepreciation,thefirmpaysasmalleramount of pesos to obtain the pre-committed amount of dollars than it would in a spot transaction. Thefirmcanthenusethegainsfromtheforwardcontracttorepaytheincreaseddebtburden 34Frootetal.(1993)showthatifexternalsourcesoffundingaremorecostlytocorporationsthaninternallygeneratedfunds,therewilltypicallybeabenefittohedging.Alfaroetal.(2020)developamodelthatincorporatesthe jointdecisionoffinancingandhedging.Lyonnetetal.(2021)studytheinteractionbetweenhedgingandcurrency choice. 22

orpaythehigherpesopricefordollar-invoicedimports. Formoredetailsonthefunctioningof over-the-counterforeignexchangederivativemarkets,seeHauetal.(2021).35 Table6showsthepercentageoffirmsthathadaccumulatedforeignassetsandtheshareof firms that have outstanding foreign exchange derivative positions in 2013. Around 13% of all firms are hedged either through dollar assets or are active in the derivative markets. Derivative markets are relatively infrequently used among Colombian non-financial corporations, withonly2.9%ofallfirmshavingaforeignexchangederivativecontractoutstandingin2013.36 11.6% of all firms have outstanding assets in foreign currency. These statistics differ largely between exporters and non-exporters. While almost 30% of exporting firms hedge, only 5% ofnon-exportersdoso. Bothforeignexchangederivativemarketsandforeignassetsaremore likelytobeusedashedgingdevicesforexportersthanfornon-exporters. We complement our baseline regression equation to test whether the usage of foreign exchangederivativesmarketsortheaccumulationofassetsindollarscushionsthenegativeimpactofforeigncurrencyborrowing. Weestimatethefollowingequation: ln (cid:161) 1+Y (cid:162)=β×FCE ×1(t ≥t )+γ×FCE ×Hedge ×1(t ≥t )+controls +(cid:178) , (11) ft f 0 f f 0 ft ft whereY arefirm-levelimportsoffirm f attimet. FCE istheforeigncurrencyexposure. 1(t ≥ f t ) is a dummy that equals one in any quarter during or after the depreciation, (t ≥2014Q3). 0 Hedge is defined as a dummy that equals one if the firm (i) had FX derivative markets posif tionsoutstandingin2013(ii)heldforeigncurrencyassetsin2013or(iii)either(i)or(ii). Theset ofcontrolsisgivenbyfirmandtimefixedeffectsaswellasHedge ×1(t ≥t ). f 0 Weinterprettheβcoefficientastheeffectofforeigncurrencyborrowingonimportsduring and after the depreciation for firms that do not hedge (Hedge = 0). The coefficient on the f tripleinteraction(FCE ×Hedge ×1(t ≥t )),γ,reflectsthedifferentialeffectoffirmsthatdo f f 0 hedge(Hedge =1). Anegativecoefficientforβshowsthatfirmsthatdonothedgeandhave f moreforeigncurrencyliabilitiesreducetheirimportsmorerelativetofirmswithfewerforeign 35Foreign-ownedfirmsmayalsobehedgedtosomeextentKalemli-Ozcanetal.(2016).Inunreportedresults,we controlforforeignownership,butourresultsremainunchanged. 36ThevastmajorityoftheforeignexchangeratederivativepositionsarenetforwardpurchasesofUSdollars, implyingfirmshedgingagainstadepreciationoftheColombianpeso. Duetofrequentmisreportingissuesofthe sign of the forward position, we only define a dummy that takes the value one if the firm reports outstanding positioninanydirection. 23

currency liabilities. This confirms our baseline result shown in Table 4. The coefficient β is largerinabsolutevaluesthaninthebaselineresult,whichalreadysuggestsacushioningeffect ofhedgingonimportcompressionduetodebtrevaluation. Apositivecoefficientγshowstheeffectisweaker(smallerimportcontraction)forfirmsthat hedge. The sum of the two coefficients reflects the estimated effect of the post-depreciation importcontractionduetoforeigncurrencyborrowingforhedgingfirms. Table7showstheresultsforeithertypeofhedging. Thecoefficientonthetripleinteraction betweenFCE ×Hedge ×1(t ≥t )isconsistentlypositiveacrossspecificationsandarecloseto f f 0 thedoubleinteractioninabsolutevalues. Thisresultshowsthatonceafirmhedgesitsforeign currency liabilities through either foreign exchange derivatives or foreign currency assets, the effectofdominantcurrencyfinancingonimportsduringthedepreciationperiodisabsent.This result can be confirmed (in unreported tables) by re-estimating the regression for only firms that hedge. In these results, the double interaction between foreign currency borrowing and thepostdummyisstatisticallyinsignificant. In Table 7 the triple interaction is positive across all specifications, but the magnitude is largestforfirmsthatdonotexport. Asnon-exportingfirmsaredrivingthenegativeimpactof foreign currency borrowing on imports during the depreciation, the result suggests that nonexportingfirmsbenefitmostfromhedgingcomparedtofirmsthatarealreadyhedgedthrough theirexportrevenuesindollars.Indeed,columns(3),(6),and(9)showthatthetripleinteraction isnotstatisticallysignificant,andneitheristhedoubleinteraction. Exportingfirms,whichdo nothedge,eitherthroughforeigncurrencyassetsorforeignexchangederivatives,donotseea contractionoftheirimportsduetoforeigncurrencyborrowingduringthedepreciation,asthey arealreadynaturallyhedgedthroughtheirdollarexportrevenues. Theadditionalhedginghas anon-significantpositiveeffectonimports. Next,weanalyzewhetherbothtypesofhedging,eitherthroughderivativesorthroughforeign currency assets can protect firms from the adverse effect of foreign currency borrowing on imports during a depreciation. Table 8 shows the results for firms that do borrow through foreigncurrencyassetsbutnotnecessarilythroughthederivativemarkets.ThedummyHedge isredefinedasoneifthefirmhasforeigncurrencyassetsoutstandingbeforethedepreciation (in 2013). The results closely resemble the results presented in Table 7. Firms that have assetsinforeigncurrencybutdonotexportdonotcontracttheirimportsduetoforeigncurrency borrowingduringthedepreciationperiod. 24

Thisisalsotrueforfirmsthathedgetheirforeigncurrencyexposurethroughtheforeignexchangederivativemarkets(Table9). WeredefinetheHedge dummytooneifthefirmhasoutstanding foreign currency derivative positions outstanding before the depreciation (in 2013). For firms that have derivative positions outstanding the triple interaction offsets the effect of foreigncurrencyborrowing,whichisconsistentwiththeideathatfirmshedgetheforeigncurrencyborrowingwiththeiroutstandingderivativepositions.Indeed,whilethedummyvariable doesnotindicatewhetherthefirmspeculatesorhedges,orwhetherthefirmboughtorsoldUS dollarsforward,inunreportedresultsweshowthatthepatternisonlyexistentforfirmswhich arenetpurchasersofdollarforwardpositions. 5.4 Quantification This section aims to quantify the effect of dominant currency financing on the trade adjustment. Ourestimatesallowustocalculatethecounterfactualtradeflowsintheabsenceofforeigncurrencyborrowing,ignoringgeneralequilibriumeffects.37 Using our estimates from Table 4 we calculate the response in imports and the trade balance due to foreign currency borrowing. The estimated effects can then be used to infer the counterfactualresponseoftradeflowstoexchangeratemovementsforotherlevelsofforeigncurrencyfinancing. WeshowtheresultsofourcounterfactualexerciseinFigure5andFigure6. InthecaseofColombia,foreign-currencyfinancingwassubstantialformany importingfirms, eventhoughtheaveragelevelofdependenceonforeigncurrencyloansismoderate. Ifallfirms hadnoforeign-currencyborrowing,thepeak-to-troughimportcontractionisestimatedtohave been17%percentsmaller. 5.5 Lending InthissectionwereestimateEquation9, replacingexportsandimportswiththeamountborrowedindomesticandforeigncurrency. Thisexerciseallowsustotestwhetherfirmsthatfaced larger financial constraints after the peso depreciation due to a higher dominant currency financing indeed had to delever. Table 10 shows that firms with larger corporate dollarization strongly contract their foreign currency borrowing. This result holds when we use either for- 37See Nakamura and Steinsson (2018) for a discussion of extrapolating aggregate effects from cross-sectional regressions. 25

eigncurrencyleverage, thewealthshockortheliquidityshockasanindependentvariable. In contrast,localcurrencyborrowingincreasesforfirmswithlargerdominantcurrencyfinancing astheyseemtosubstituteforeigncurrencyloanswithdomesticcurrencyloans,consistentwith evidenceinHardy(2018). Overall, anaveragefirmwithnon-zeroforeigncurrencyborrowing, hadroughly20%ofitsdebtdenominatedinforeigncurrency. Asaresult,thisfirmwouldseea 3%declineintotalborrowingaftertheshock,takingintoaccountforeignanddomesticloans, andabstractingfromthevaluationeffectsofexchangeratemovements. Thisisconsistentwith theresultsofKalemli-Ozcanetal.(2020)whoshowthatthefirmsthatweremoreexposedtothe negativeeffectofmismatchesontheirbalancesheets(thoseoperatingincountrieswithhigher foreign currency debt) delever more after currency depreciations than firms in less exposed countries.38 5.6 FinancialFrictionsundertheMicroscope Inthissectionweexploretherichnessoftheloan-leveldatasettoshedlightontheexactmechanism through which the revaluation of debt feeds into financial frictions and leads to a contractioninimports. First we focus on the interest rates of the loans l that have been issued to firm f at day d by bank b. We regress the interest rate at which the loan is issued on the different measures of foreign currency exposure, as discussed above. The loan-level data allows us to control for bank-supply, by including bank-time(month) fixed effects. If firms that are financially more exposedtothedepreciationborrowmorefrombanksthataremorenegativelyaffectedbythe depreciation,thenegativesigninourbaselineresultscouldbedrivenbyaspuriouscorrelation betweenbanks’creditsupplyandfirms’foreigncurrencyexposure. Thebank-timefixedeffect allows us to compare firms with differential foreign currency exposure with each other that borrowfromthesamebankinthesamemonth. Inparticular,weestimatethefollowingregression: r =α+β FCE ∗Post +α +α +(cid:178) (12) lfd(t)b 1 f t bt f fd(t)b wherer istheinterestratethatischargedforloanl tofirm f atissuancedated(t)bybankb. 38TableA7estimatesaloan-levelregressionwhereweconfirmthatexposedfirmssubstitutefromforeigncurrencyborrowingtolocalcurrencyborrowingwithoverallborrowingremainingstatisticallynotdifferentfromzero. Thiseffectisnotsignificantlydifferentforexportersvs.non-exporters. 26

FCE are the different measures of foreign currency exposure. Post is a dummy that equals 1 after the depreciation and 0 otherwise. α are bank-month fixed effect and α are firm fixed bt f effects. Asbefore,standarderrorsareclusteredatthefirmlevel. Table11showstheresults. Onaverage,firmsthataremoreexposedtothedepreciationfinanciallyfacehigherinterestratesafterthedepreciationstartedcomparedtobefore,reflected inthepositivecoefficient. However, consistentlywithourresultsonimports, thisresultisentirely driven by firms that do not export. For exporters, the coefficient is statistically insignificantforalltypesofexposuremeasuresandevennegativeforthesimpleforeigncurrencyleveragemeasureaswellasthewealthshock. Theseresultsindicatethatnon-exportingfirmsthatexperiencealargerappreciationoftheir debtvaluesarechargedhigherinterestrates,sheddinglightonthefinancialfrictionsthatthese firmsfaceandprovidingevidenceonapotentialchannelforwhytheycontractimports. Thehigherinterestratesthatarechargedbybanksarepotentiallyduetohigherprobabilitiesofdefaultsofthenon-exportingfirmsthatfaceadebtrevaluation. Thisisnotthecasefor exportingfirms,astheyarehedgedagainsttherevaluationoftheirdebtduetotheirappreciationofexportrevenues. Next, we test whether the higher default risk of non-exporting firms with a larger appreciation of their debt due to the depreciation also feeds actually into delays in their payment of loans. Wecreateabalancedpanelforeachbank-firm-datecombinationanddefineadummythat equals1iffirm f isbehindpaymentintheirloanstobankbattimet andestimatethefollowing linearprobabilitymodel: Delinquent =α+β FCE ∗Post +α +α +(cid:178) (13) fbt 1 f t bt f fbt whereDeliquent isequaltooneiffirm f isdelinquentonaloantobankbattimet and fbt zero otherwise. FCE is either the foreign currency leverage just before the depreciation, the f liquidity shock, or the wealth shock, as described above. α are bank-time fixed effects and bt α arefirmfixedeffects. Thebank-timefixedeffectscanagaincontrolfortime-varyingbankf specificfactorsthatcouldleadtoaspuriouscorrelationbetweenourforeigncurrencyexposure measuresandwhetherthefirmisdelinquentontheirloansafterthedepreciation.Forinstance, certainbanksmaybebetteratscreeningandmonitoringborrowers,whichleadstolowernon- 27

performing loans during the depreciation.39 If the better screening and monitoring ability is correlatedwiththeaverageforeigncurrencyexposureoftheirborrowers,thiscouldintroduce abiasintheestimates. Afterabsorbingforbank-timefixedeffects,wecontrolforthispotential correlationbetweentheerrortermandthetreatmentvariable. Table12showstheresults. Onaverage,firmswithlargerfinancialexposuretotheexchange rate depreciation are more likely to fall behind their loan payments than their counterparts. ThiscanbeseeninColumns(1),(4),and(7). However,thiseffectisexclusivelydrivenbynonexporting firms. For exporting firms the effect of larger debt appreciation due to foreign currencydebtisnotsignificantlyaffectingtheprobabilityoffallingbehindindebtpayments. Theseresultsconfirmourbaselineresultandsuggestthatsinceexportersarehedgedthrough higherUSdollarrevenues,theyarestillabletorepaytheirloans. 5.7 Placebo In this section we conduct a placebo exercise where we replace the foreign currency leverage withdomestic currencyleverage. Thisexercisehelpsustoruleoutthatourmeasuresoffinancialexposuretothedepreciationarenotsimplymeasuresoffinancialvulnerabilitiesingeneral. Ifourmeasurescapturedfinancialconstraintswhicharenotrelatedtoforeigncurrencyborrowing,ourresultswouldstillbeindicativeofafinancialchannelofexternaladjustmentbutnota debtdollarizationchannel. Table13indeedshowsthatifwereplaceourmeasuresofdebtdollarizationwithdomesticleverage,wedonotfindsignificantnegativeeffectsonneitherimports nor exports. This evidence is strongly suggesting that foreign currency borrowing rather than domesticfinancialheterogeneityacrossfirmsisdrivingtheimportresponse. 5.8 PanelRegression Inthissectionweaimtogeneralizeourresultstoalargertimewindow. Whileoneadvantageof theevent-studyapproachisthatthecausalidentificationismorecredible,ourresultsmaynot beexternallyvalidoutsideoflargedevaluationepisodes.Weusedatabetween2008and2018to constructanannualpaneldatasetwithlogimports,logexports,foreigncurrencyleverageand the Colombian peso/US dollar exchange rate. Although quarterly data is available, we move 39SeeforexamplePierriandTimmer(2022),whoshowthatbanksthatinvestmoreininformationtechnology aremoreresilienttofinancialshocks. 28

toanannualpanel,aswehaveseeninFigure2thatthedominantcurrencyfinancingchannel works with a lag to the exchange rate movement. We regress log imports and exports on the laggedforeigncurrencyleverageandthemovementinthepeso/dollarexchangeratebetween t−2andt−1controllingforfirmandyearfixedeffects. Table14showstheresults. Thefirstrowshowsthatfirmsthathadhigherforeigncurrency leverageinthepreviousyearexportandimportmoreinthecurrentyearintheabsenceofexchange rate movements. However, once we see a depreciation of the peso relative to the US dollar,onlynon-exporters(column(2))shrinktheirimportssignificantlyinthefollowingyear. ThisisaconfirmationofourbaselineresultsshowninTable4. Asintheevent-studyapproach, imports of firms that both export and import do not respond to the interaction between foreigncurrencyleverageandtheexchangerate.Incontrasttotheevent-studyapproach,oncewe bundlenon-exportersandexporterstogetherwedonotfindasignificantimportresponse. 6 Conclusions Inthispaperweanalyzethedominantcurrencyfinancingchannelofexternaladjustment. We study whetherfirmswithmore foreigncurrency borrowingareaffecteddifferently interms of their trade response compared to firms that borrow mainly in domestic currency. As foreign currencyborrowingisanendogenouschoice,weimplementanovelidentificationstrategythat helpsusshedlightonthecausalimpactofdebtdollarizationonexternaladjustment. Wecomputetheincreaseinforeigncurrencydebtpaymentsthatissolelyduetothedepreciationofthe currencyasfirmshavetheirdebtmaturingatdifferentpointsintime. Firmsthathavethemajorityoftheirdebtmaturingbeforethestartofthedepreciationare not affected by the depreciation through their balance sheet. In contrast, firms which have debtrepaymentsscheduledattheheightofthedepreciationfacealargeincreaseintheirdebt repayment. This foreign currency liquidity shock strongly predicts a contraction in imports. However,thiseffectissolelypresentforfirmsthatdonotexport.Exportersarenaturallyhedged against these negative balance sheet shocks through their revenues in foreign currency and therefore they not only do not shrink their imports in response to the shock, they also do not seetheirexportsrespond. We conclude that foreign currency mismatches increases the potency of the external adjustment. This channel can be seen as an unintended side effect from borrowing in foreign 29

currency from the perspective of external adjustment. However, the sharp decline in imports canhaverealeffectsonfirmswhoarereliantonimportedcapitalgoodsorintermediateinputs. 30

References Adler,Gustavo,CamilaCasas,LuisCubeddu,GitaGopinath,NanLi,SergiiMeleshchuk,CarolinaOsorio-Buitron,DamienPuy,andYannickTimmer(2020)“Dominantcurrenciesand externaladjustment”,IMFStaffDiscussionNote. Aguiar,Mark(2005)“Investment,devaluation,andforeigncurrencyexposure:Thecaseofmexico”,JournalofDevelopmentEconomics,78(1),pp.95–113. Akinci, Ozge and Albert Queralto (2018) “Exchange rate dynamics and monetary spillovers withimperfectfinancialmarkets”,FRBofNewYorkStaffReport(849). Alfaro,Laura,GonzaloAsis,AnushaChari,andUgoPanizza(2019)“Corporatedebt,firmsize and financial fragility in emerging markets”, Journal of International Economics, 118, pp. 1– 19. Alfaro,Laura,MauricioCalani,andLilianaVarela(2020)“Currencyhedginginemergingmarkets: Managingcashflowexposure”. Almeida,Heitor,MurilloCampello,BrunoLaranjeira,ScottWeisbenneretal.(2012)“Corporatedebtmaturityandtherealeffectsofthe2007creditcrisis”,CriticalFinanceReview,1(1), pp.3–58. Amiti,Mary,OlegItskhoki,andJozefKonings(2014)“Importers,exporters,andexchangerate disconnect”,AmericanEconomicReview,104(7),pp.1942–78. Amiti,Mary,OlegItskhoki,andJozefKonings(2019)“Internationalshocks,variablemarkups, anddomesticprices”,TheReviewofEconomicStudies,86(6),pp.2356–2402. Amiti,Mary,OlegItskhoki,andJozefKonings(2022)“Dominantcurrencies:Howfirmschoose currencyinvoicingandwhyitmatters”,QuarterlyJournalofEconomics. Amiti,MaryandDavidEWeinstein(2011)“Exportsandfinancialshocks”,TheQuarterlyJournalofEconomics,126(4),pp.1841–1877. Amiti,MaryandDavidEWeinstein(2018)“Howmuchdoidiosyncraticbankshocksaffectinvestment? evidence from matched bank-firm loan data”, Journal of Political Economy, 126 (2),pp.525–587. 31

Bahaj,SaleemandRicardoReis(2020)“Jumpstartinganinternationalcurrency”. BancodelaRepublica(2000)“Paragrafo1delarticulo79delaresolucion8de2000”,Technical report,Superintendencia. Barajas,Adolfo,SergioRestrepo,MrRobertoSteiner,JuanCamiloMedellin,andCésarPabón (2017)CurrencyMismatchesandVulnerabilitytoExchangeRateShocks: NonfinancialFirms inColombia,InternationalMonetaryFund. Berman,NicolasandAntoineBerthou(2009)“Financialmarketimperfectionsandtheimpact ofexchangeratemovementsonexports”,ReviewofInternationalEconomics,17(1),pp.103– 120. Bernanke, Ben S, Mark Gertler, and Simon Gilchrist (1999) “The financial accelerator in a quantitativebusinesscycleframework”,Handbookofmacroeconomics,1,pp.1341–1393. Betts,CarolineandMichaelBDevereux(2000)“Exchangeratedynamicsinamodelofpricingto-market”,JournalofinternationalEconomics,50(1),pp.215–244. Blaum,Joaquin(2019)“Globalfirmsinlargedevaluations”,WorkingPaper. Bleakley, HoytandKevinCowan (2008) “Corporate dollar debt and depreciations: much ado aboutnothing?”,TheReviewofEconomicsandStatistics,90(4),pp.612–626. Bruno,ValentinaandHyunSongShin(2019)“Dollarexchangerateasacreditsupplyfactor– evidencefromfirm-levelexports”. Burbidge,JohnB,LonnieMagee,andALeslieRobb(1988)“Alternativetransformationstohandleextremevaluesofthedependentvariable”,JournaloftheAmericanStatisticalAssociation, 83(401),pp.123–127. Céspedes, Luis Felipe, Roberto Chang, and Andres Velasco (2004) “Balance sheets and exchangeratepolicy”,AmericanEconomicReview,94(4),pp.1183–1193. Chen, M Keith (2013) “The effect of language on economic behavior: Evidence from savings rates,healthbehaviors,andretirementassets”,AmericanEconomicReview,103(2),pp.690– 731. 32

Christiano, Lawrence, Nurbekyan Armen, and Dalgic Husnu (2021) “Financial dollarization inemergingmarkets: Efficientrisksharingorprescriptionfordisaster?”,mimeo. Desai,MihirA.,C.FritzFoley,andKristinJ.Forbes(2008)“Financialconstraintsandgrowth: Multinational and local firm responses to currency depreciations”, The Review of Financial Studies,21(6),pp.2857–2888. Devereux, MichaelBandCharlesEngel (2003) “Monetary policy in the open economy revisited: Pricesettingandexchange-rateflexibility”, TheReviewofEconomicStudies, 70(4), pp. 765–783. Devereux,MichaelB,PhilipRLane,andJuanyiXu(2006)“Exchangeratesandmonetarypolicy inemergingmarketeconomies”,TheEconomicJournal,116(511),pp.478–506. Duval,Romain,GeeHeeHong,andYannickTimmer(2020)“Financialfrictionsandthegreat productivityslowdown”,TheReviewofFinancialStudies,33(2),pp.475–503. Echeverry, Juan Carlos, Leopoldo Fergusson, Roberto Steiner, and Camila Aguilar (2003) “Dollardebtincolombianfirms: aresinnerspunishedduringdevaluations?”,EmergingMarketsReview,4(4),pp.417–449. Fleming,JMarcus(1962)“Domesticfinancialpoliciesunderfixedandunderfloatingexchange rates”,StaffPapers,9(3),pp.369–380. Froot, KennethA,DavidSScharfstein, andJeremyCStein (1993) “Risk management: Coordinating corporate investment and financing policies”, Journal of Finance, 48 (5), pp. 1629– 1658. Galindo,Arturo,UgoPanizza,andFabioSchiantarelli(2003)“Debtcompositionandbalance sheeteffectsofcurrencydepreciation: asummaryofthemicroevidence”,EmergingMarkets Review,4(4),pp.330–339. Giannetti, Mariassunta and Farzad Saidi (2019) “Shock propagation and banking structure”, TheReviewofFinancialStudies,32(7),pp.2499–2540. Goldberg,LindaandCédricTille(2008)“Vehiclecurrencyuseininternationaltrade”, Journal ofInternationalEconomics,76(2),pp.177–192. 33

Gopinath, Gita (2015) “The international price system”,Technical report, National Bureau of EconomicResearch. Gopinath, Gita, Emine Boz, Camila Casas, Federico J Diez, Pierre-Olivier Gourinchas, and M Plagborgmøller (2020) “Dominant currency paradigm”, American Economic Review, 110 (3),pp.677–719. Gopinath,GitaandOlegItskhoki(2021)“Dominantcurrencyparadigm: areview”. Gopinath, Gita, Oleg Itskhoki, and Roberto Rigobon (2010) “Currency choice and exchange ratepass-through”,AmericanEconomicReview,100(1),pp.304–36. Gopinath,GitaandJeremyCStein(2021)“Banking,trade,andthemakingofadominantcurrency”,QuarterlyJournalofEconomics. Hardy, Bryan (2018) “Foreign currency borrowing, balance sheet shocks and real outcomes”, BalanceSheetShocksandRealOutcomes(November22,2018).BISWorkingPaper(758). Hau, Harald, Peter Hoffmann, Sam Langfield, and Yannick Timmer (2021) “Discriminatory pricingofover-the-counterderivatives”,ManagementScience. Kalemli-Ozcan, Sebnem, Herman Kamil, and Carolina Villegas-Sanchez (2016) “What hinders investment in the aftermath of financial crises: Insolvent firms or illiquid banks?”, ReviewofEconomicsandStatistics,98(4),pp.756–769. Kalemli-Ozcan,Sebnem,XiaoxiLiu,andIlhyockShim(2020)“Exchangeratefluctuationsand firmleverage”,WorkingPaper. Kohn, David, Fernando Leibovici, and Michal Szkup (2020) “Financial frictions and export dynamicsinlargedevaluations”,JournalofInternationalEconomics,122,p.103257. Krugman, Paul (1999) “Balance sheets, the transfer problem, and financial crises”, Internationalfinanceandfinancialcrises,Springer,pp.31–55. Lane,PhilipR(2001)“Thenewopeneconomymacroeconomics: asurvey”,Journalofinternationaleconomics,54(2),pp.235–266. 34

Love, Inessa, Lorenzo A Preve, and Virginia Sarria-Allende (2007) “Trade credit and bank credit:Evidencefromrecentfinancialcrises”,JournalofFinancialEconomics,83(2),pp.453– 469. Lyonnet,Victor,JulienMartin,andIsabelleMejean(2021)“Invoicingcurrency,firmsize,and hedging”,JournalofMoney,CreditandBanking. Mundell, Robert A (1963) “Capital mobility and stabilization policy under fixed and flexible exchange rates”, Canadian Journal of Economics and Political Science/Revue canadienne de economiquesetsciencepolitique,29(4),pp.475–485. Muûls,Mirabelle(2015)“Exporters,importersandcreditconstraints”,JournalofInternational Economics,95(2),pp.333–343. Nakamura,EmiandJónSteinsson(2018)“Identificationinmacroeconomics”,JournalofEconomicPerspectives,32(3),pp.59–86. Niepmann,FriederikeandTimSchmidt-Eisenlohr(2017a)“Foreigncurrencyloansandcredit risk: Evidencefromusbanks”,Technicalreport,CESifoGroupMunich. Niepmann, FriederikeandTimSchmidt-Eisenlohr (2017b) “International trade, risk and the roleofbanks”,JournalofInternationalEconomics,107,pp.111–126. Niepmann, Friederike and Tim Schmidt-Eisenlohr (2017c) “No guarantees, no trade: How banksaffectexportpatterns”,JournalofInternationalEconomics,108,pp.338–350. Obstfeld,MauriceandKennethRogoff(1995)“Exchangeratedynamicsredux”,JournalofPoliticaleconomy,103(3),pp.624–660. Obstfeld, Maurice and Kenneth Rogoff (2000) “New directions for stochastic open economy models”,JournalofInternationalEconomics,50(1),pp.117–153. Paravisini,Daniel,VeronicaRappoport,PhilippSchnabl,andDanielWolfenzon(2014)“Dissectingtheeffectofcreditsupplyontrade: Evidencefrommatchedcredit-exportdata”,The ReviewofEconomicStudies,82(1),pp.333–359. Pierri,NicolaandYannickTimmer(2022)“Theimportanceoftechnologyinbankingduringa crisis”,JournalofMonetaryEconomics. 35

Restrepo,Sergio,JorgeNiño,andEnriqueMontes(2014)“Delcalcescambiariosdelasfirmas nofinacierasencolombia”,BorradoresdeEconomíaNo.805. Rodnyansky, Alexander (2019) “(un) competitive devaluations and firm dynamics”, Available atSSRN3095698. Schmidt-Eisenlohr, Tim (2013) “Towards a theory of trade finance”, Journal of International Economics,91(1),pp.96–112. Townsend,RobertM(1979)“Optimalcontractsandcompetitivemarketswithcostlystateverification”,JournalofEconomictheory,21(2),pp.265–293. Verner,EmilandGyozoGyongyosi(2018)“Householddebtrevaluationandtherealeconomy: Evidencefromaforeigncurrencydebtcrisis”,AmericanEconomicReview. 36

Table1: Descriptivestatistics,firmcharacteristics Non-exporters Exporters Allfirms mean sd mean sd mean sd Assets 8,810 154,309 19,655 208,475 12,400 174,183 Exportsbefore2014Q2 0 0 385 6,959 127 4,007 Exportsafter2014Q2 0 0 326 4,584 108 2,646 Importsbefore2014Q2 55 229 288 642 132 429 Importsafter2014Q2 48 219 260 615 118 410 Numberoffirms 14,618 7,232 21,850 AllvariablesreportedinthousandsofUSdollars.AssetsrefertototalassetsoffirmsasreportedinOrbisdatabase attheendof2013.TradedataistakenfromDANEandiscalculatedatthequarterlylevel.Thesampleislimitedto firmsthateverreportedinourdata.Exportersaredefinedasfirmsthateverexportedinourdata. Table2: Descriptivestatistics,independentvariables Percentiles mean 10 25 50 75 90 Foreignleverage 4.4% 0.2% 0.4% 2.0% 6.3% 11.6% Liquidityshock 0.1% -0.2% 0.0% 0.0% 0.0% 0.6% Wealthshock 0.8% -0.1% 0.0% 0.2% 0.9% 2.6% Seesubsection3.2forthedefinitionofthevariables. 37

Table3: BalanceTable (1) (2) (3) (4) (5) Size BankCreditSupply Age Profits Maturity ∗ LS 0.033 -0.001 0.006 -0.002 0.007 (0.017) (0.001) (0.009) (0.007) (0.010) N 14,680 14,680 14,502 14,566 12,994 Thetablereportstheestimatedβcoefficientfromacross-sectionalfirm-levelregressionofpre-depreciation(2013) firmlevelcharacteristicsontheliquidityshock(LS).Sizereferstologassets,BankCreditSupplyistheexposureof firmstoabankcreditsupplyshockconstructedinthespiritofAmitiandWeinstein(2018).Ageisthelogageofthe firm,profitsisEBITDAovertotalassets,andMaturityisthelogoftheaveragematurityoffirms’loans.***,**,and* indicatesignificanceat10%,5%,and1%levelsrespectively. 38

Table4: Imports ln(imports) (1) (2) (3) (4) (5) (6) (7) (8) (9) Post×FCL -0.106 ∗∗∗ -0.215 ∗∗∗ -0.018 (0.025) (0.047) (0.023) Post×LS -0.046 ∗∗ -0.069 ∗ -0.024 (0.020) (0.034) (0.021) Post×WS -0.059 ∗∗∗ -0.120 ∗∗∗ -0.009 (0.020) (0.032) (0.024) Sample All Non-X X All Non-X X All Non-X X FirmFE Y Y Y Y Y Y Y Y Y TimeFE Y Y Y Y Y Y Y Y Y N 524,943 350,934 174,009 524,943 350,934 174,009 524,943 350,934 174,009 ThetablereportstheestimatedβcoefficientfromEquation9withln (cid:161) 1+imports (cid:162) ontheleft-handsideatthequarterlylevel.Columns (1)-(3)reporttheestimatedcoefficientoftheinteractionofaPost withforeigncurrencyleverage(FCL).Incolumns(4)-(6)and(7)-(9) Post isinteractedwithliquidity(LS)andwealthshocks(WS)respectively. FCL,WS,andLSwerenormalizedtohavezeromeananda standarddeviationof1. Post isdefinedasabinaryvariableequalto1fortheperiodafter2014Q2. Thesampleislimitedtothefirms thateverimported. Columns(1),(4),(7)reporttheresultsforallimporters,columns(2),(5),(8)reporttheresultsfornon-exporters, andcolumns(3),(6),and(9)reporttheresultsestimatedonthesampleofexporters.Exportersarefirmsthateverexportedinourdata. Seesubsection3.2forthedefinitionoftheindependentvariables.Standarderrorsareclusteredatthefirmandquarterlevel.***,**,and *indicatesignificanceat10%,5%,and1%levelsrespectively. 39

Table5: Exports ln(exports) (1) (2) (3) Post×FCL -0.010 (0.025) Post×LS -0.025 (0.023) Post×WS -0.024 (0.026) FirmFE Y Y Y TimeFE Y Y Y N 169,232 169,232 169,232 ThetablereportstheestimatedβcoefficientfromEquation9withln (cid:161) 1+exports (cid:162) ontheleft-handsideatthe quarterly level. Column (1) reports the estimated coefficient of the interaction of a Post with foreign currency leverage(FCL).Incolumns(2)and(3)Post isinteractedwithliquidity(LS)andwealthshocks(WS)respectively. FCL,WS,andLSwerenormalizedtohavezeromeanandastandarddeviationof1. Post isdefinedasabinary variableequalto1fortheperiodafter2014Q2. Seesubsection3.2forthedefinitionoftheindependentvariables. Standarderrorsareclusteredatthefirmandquarterlevel. ***,**,and*indicatesignificanceat10%,5%,and1% levelsrespectively.Thesampleislimitedtothefirmsthateverexportedinourdata. 40

Table6: Descriptivestatistics,hedging Non-exporters Exporters Allfirms ShareofFirms(2013) HedgeForeignAssets 4.0% 26.9% 11.6% HedgeFXDerivatives 1.6% 5.6% 2.9% Hedge(ForeignAssetsorFXDerivatives) 5.1% 28.7% 12.9% Numberoffirms 14,618 7,232 21,850 HedgeForeignAssetsistheshareoffirmsthathaveoutstandingforeignassetsin2013.HedgeFXDerivativesisthe shareoffirmsthathaveoutstandingforeignexchangederivativesin2013.Hedge(ForeignAssetsorFXDerivatives) istheshareoffirmsthathaveeitheroutstandingforeignassetsin2013oroutstandingforeignexchangederivatives in2013. Thesampleislimitedtofirmsthateverreportedinourdata. Exportersaredefinedasfirmsthatever exportedinourdata. 41

Table7: HedgingandImports ln(imports) (1) (2) (3) (4) (5) (6) (7) (8) (9) Post×FCL -0.150 ∗∗∗ -0.270 ∗∗∗ 0.002 (0.045) (0.062) (0.046) Post×FCL×Hedge 0.140 ∗∗ 0.246 ∗∗∗ 0.000 (0.051) (0.079) (0.054) Post×LS -0.071 ∗∗ -0.085 ∗∗ -0.049 (0.030) (0.039) (0.037) Post×LS×Hedge 0.086 ∗∗ 0.105 0.059 (0.040) (0.071) (0.046) Post×WS -0.084 ∗∗∗ -0.131 ∗∗∗ -0.020 (0.029) (0.036) (0.042) Post×WS×Hedge 0.107 ∗∗∗ 0.140 ∗ 0.044 (0.038) (0.071) (0.049) Sample All Non-X X All Non-X X All Non-X X FirmFE Y Y Y Y Y Y Y Y Y TimeFE Y Y Y Y Y Y Y Y Y N 524,943 350,934 174,009 524,943 350,934 174,009 524,943 350,934 174,009 ThetablereportstheestimatedβandγfromEquation11withln (cid:161) 1+imports (cid:162) ontheleft-handsideatthequarterlylevel.Columns(1)- (3)reporttheestimatedcoefficientoftheinteractionofPostwithforeigncurrencyleverage(FCL)andtheinteractionofPost×FCL×Hedge. Incolumns(4)-(6)and(7)-(9)Postisinteractedwithliquidity(LS)andwealthshocks(WS)andthehedgingdummyrespectively. FCL,WS, andLSwerenormalizedtohavezeromeanandastandarddeviationof1. Hedgeisadummythatequalsoneifthefirmhaseitherforeign currencyassetsoutstandingorhasforeigncurrencyderivativepositionsoutstandingin2013andzeroifnot. Post isdefinedasabinary variableequalto1fortheperiodafter2014Q2.Thesampleislimitedtothefirmsthateverimported.Columns(1),(4),(7)reporttheresults forallimporters,columns(2),(5),(8)reporttheresultsfornon-exporters,andcolumns(3),(6),and(9)reporttheresultsestimatedonthe sampleofexporters. Exportersarefirmsthateverexportedinourdata. Seesubsection3.2forthedefinitionoftheindependentvariables. Standarderrorsareclusteredatthefirmandquarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. 42

Table8: ForeignCurrencyAssetsHedgingandImports ln(imports) (1) (2) (3) (4) (5) (6) (7) (8) (9) Post×FCL -0.142 ∗∗∗ -0.236 ∗∗∗ -0.029 (0.039) (0.055) (0.040) Post×FCL×Hedge 0.126 ∗∗ 0.147 ∗ 0.045 (0.047) (0.078) (0.050) Post×LS -0.064 ∗∗ -0.082 ∗∗ -0.039 (0.027) (0.038) (0.033) Post×LS×Hedge 0.077 ∗ 0.118 ∗ 0.042 (0.038) (0.068) (0.044) Post×WS -0.080 ∗∗∗ -0.125 ∗∗∗ -0.026 (0.026) (0.035) (0.036) Post×WS×Hedge 0.104 ∗∗∗ 0.110 0.057 (0.036) (0.072) (0.046) Sample All Non-X X All Non-X X All Non-X X FirmFE Y Y Y Y Y Y Y Y Y TimeFE Y Y Y Y Y Y Y Y Y N 524,943 350,934 174,009 524,943 350,934 174,009 524,943 350,934 174,009 ThetablereportstheestimatedβandγfromEquation11withln (cid:161) 1+imports (cid:162) ontheleft-handsideatthequarterlylevel.Columns(1)- (3)reporttheestimatedcoefficientoftheinteractionofPostwithforeigncurrencyleverage(FCL)andtheinteractionofPost×FCL×Hedge. Incolumns(4)-(6)and(7)-(9)Postisinteractedwithliquidity(LS)andwealthshocks(WS)andthehedgingdummyrespectively. FCL, WS,andLSwerenormalizedtohavezeromeanandastandarddeviationof1.Hedgeisadummythatequalsoneifthefirmhasforeign currencyassetsoutstandingin2013andzeroifnot. Post isdefinedasabinaryvariableequalto1fortheperiodafter2014Q2. The sampleislimitedtothefirmsthateverimported.Columns(1),(4),(7)reporttheresultsforallimporters,columns(2),(5),(8)reportthe resultsfornon-exporters,andcolumns(3),(6),and(9)reporttheresultsestimatedonthesampleofexporters.Exportersarefirmsthat everexportedinourdata. Seesubsection3.2forthedefinitionoftheindependentvariables. Standarderrorsareclusteredatthefirm andquarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. 43

Table9: ForeignExchangeDerivativeMarketsHedgingandImports ln(imports) (1) (2) (3) (4) (5) (6) (7) (8) (9) Post×FCL -0.124 ∗∗∗ -0.251 ∗∗∗ -0.003 (0.033) (0.054) (0.028) Post×FCL×Hedge 0.127 ∗ 0.388 ∗∗∗ -0.060 (0.062) (0.114) (0.058) Post×LS -0.050 ∗∗ -0.074 ∗ -0.023 (0.023) (0.036) (0.024) Post×LS×Hedge 0.039 0.081 -0.005 (0.050) (0.115) (0.051) Post×WS -0.066 ∗∗∗ -0.134 ∗∗∗ -0.003 (0.023) (0.034) (0.028) Post×WS×Hedge 0.088 ∗ 0.226 ∗∗ -0.020 (0.048) (0.103) (0.049) Sample All Non-X X All Non-X X All Non-X X FirmFE Y Y Y Y Y Y Y Y Y TimeFE Y Y Y Y Y Y Y Y Y N 524,943 350,934 174,009 524,943 350,934 174,009 524,943 350,934 174,009 ThetablereportstheestimatedβandγfromEquation11withln (cid:161) 1+imports (cid:162) ontheleft-handsideatthequarterlylevel.Columns(1)- (3)reporttheestimatedcoefficientoftheinteractionofPostwithforeigncurrencyleverage(FCL)andtheinteractionofPost×FCL×Hedge. Incolumns(4)-(6)and(7)-(9)Postisinteractedwithliquidity(LS)andwealthshocks(WS)andthehedgingdummyrespectively. FCL, WS,andLSwerenormalizedtohavezeromeanandastandarddeviationof1.Hedgeisadummythatequalsoneifthefirmhasforeign currencyderivativepositionsoutstandingin2013andzeroifnot. Post isdefinedasabinaryvariableequalto1fortheperiodafter 2014Q2.Thesampleislimitedtothefirmsthateverimported.Columns(1),(4),(7)reporttheresultsforallimporters,columns(2),(5), (8)reporttheresultsfornon-exporters,andcolumns(3),(6),and(9)reporttheresultsestimatedonthesampleofexporters.Exporters arefirmsthateverexportedinourdata.Seesubsection3.2forthedefinitionoftheindependentvariables.Standarderrorsareclustered atthefirmandquarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. 44

Table10: Borrowing ln(FCborrowing) ln(LCborrowing) (1) (2) (3) (4) (5) (6) Post×FCL -0.587 ∗∗∗ 0.104 ∗∗∗ (0.117) (0.026) Post×LS -0.207 ∗∗∗ 0.029 (0.059) (0.017) Post×WS -0.391 ∗∗∗ 0.068 ∗∗∗ (0.097) (0.020) FirmFE Y Y Y Y Y Y TimeFE Y Y Y Y Y Y N 524,943 524,943 524,943 524,943 524,943 524,943 The table reports the estimated β coefficient from Equation 9 with ln (cid:161) 1+borrowing (cid:162) on the left-hand side at the quarterly level. Columns(1)-(3)reporttheresultsforborrowinginforeigncurrencyandcolumns(4)-(6)useborrowinginlocalcurrencyonthelefthandside. Columns(1)and(4)reporttheestimatedcoefficientoftheinteractionofaPost withforeigncurrencyleverage(FCL).In columns(2),(5)and(3),(6)Postisinteractedwithliquidity(LS)andwealthshocks(WS)respectively.FCL,WS,andLSwerenormalized to have zero mean and a standard deviation of 1. Post is defined as a binary variable equal to 1 for the period after 2014Q2. See subsection3.2forthedefinitionoftheindependentvariables. Standarderrorsareclusteredatthefirmandquarterlevel. ***,**,and* indicatesignificanceat10%,5%,and1%levelsrespectively.Thesampleislimitedtothefirmsthateverimportedinourdata.Exporters arefirmsthateverexportedinourdata. 45

Table11: InterestRate InterestRate (1) (2) (3) (4) (5) (6) (7) (8) (9) FCL×Post 0.040 0.134 ∗∗∗ 0.001 (0.028) (0.049) (0.036) LS×Post 0.052 ∗ 0.074 ∗ 0.043 (0.029) (0.041) (0.039) WS×Post 0.031 0.090 ∗ -0.006 (0.032) (0.052) (0.040) Sample All Non-X X All Non-X X All Non-X X FirmFE Y Y Y Y Y Y Y Y Y Bank×TimeFE Y Y Y Y Y Y Y Y Y N 364,928 220,732 143,774 364,928 220,732 143,774 365,286 220,732 143,774 The table reports the estimated β from Equation 12 with r the interest rate charged by bank b at day d to firm f on the left-hand side.Columns(1)-(3)reporttheestimatedcoefficientoftheinteractionofPostwithforeigncurrencyleverage(FCL).Incolumns(4)-(6) and(7)-(9)Post isinteractedwithliquidity(LS)andwealthshocks(WS).FCL,WS,andLSwerenormalizedtohavezeromeananda standarddeviationof1. Post isdefinedasabinaryvariableequalto1fortheperiodafter2014Q2. Thesampleislimitedtothefirms thateverimported. Columns(1),(4),(7)reporttheresultsforallimporters,columns(2),(5),(8)reporttheresultsfornon-exporters, andcolumns(3),(6),and(9)reporttheresultsestimatedonthesampleofexporters.Exportersarefirmsthateverexportedinourdata. Seesubsection3.2forthedefinitionoftheindependentvariables.Standarderrorsareclusteredatthefirmandquarterlevel.***,**,and *indicatesignificanceat10%,5%,and1%levelsrespectively. 46

Table12: Delinquency Delinquent (1) (2) (3) (4) (5) (6) (7) (8) (9) Post×FCL 0.002 ∗∗∗ 0.006 ∗∗∗ 0.001 (0.001) (0.002) (0.001) Post×LS 0.001 ∗ 0.003 ∗∗ 0.000 (0.001) (0.001) (0.001) Post×WS 0.001 ∗∗ 0.004 ∗∗∗ 0.001 (0.001) (0.001) (0.001) Sample All Non-X X All Non-X X All Non-X X FirmFE Y Y Y Y Y Y Y Y Y Bank×TimeFE Y Y Y Y Y Y Y Y Y N 3,635,780 2,165,253 1,470,378 3,635,780 2,165,253 1,470,378 3,635,780 2,165,253 1,470,378 ThetablereportstheestimatedβfromEquation13withDelinquent asadummyvariableontheleft-handsideatthequarterlylevel that is equal to 1 if the firm is delinquent on a loan at quarter t to bank b. Columns (1)-(3) report the estimated coefficient of the interactionofPostwithforeigncurrencyleverage(FCL).Incolumns(4)-(6)and(7)-(9)Postisinteractedwithliquidity(LS)andwealth shocks(WS).FCL,WS,andLSwerenormalizedtohavezeromeanandastandarddeviationof1. Post isdefinedasabinaryvariable equalto1fortheperiodafter2014Q2. Thesampleislimitedtothefirmsthateverimported. Columns(1),(4),(7)reporttheresults forallimporters,columns(2),(5),(8)reporttheresultsfornon-exporters,andcolumns(3),(6),and(9)reporttheresultsestimatedon thesampleofexporters. Exportersarefirmsthateverexportedinourdata. Seesubsection3.2forthedefinitionoftheindependent variables. Standard errors are clustered at the firm and quarter level. ***,**, and * indicate significance at 10%, 5%, and 1% levels respectively. 47

Table13: PlaceboDomesticLeverage ln(imports) ln(exports) (1) (2) (3) (4) Post×DomesticLeverage -0.006 -0.007 -0.001 0.001 (0.004) (0.005) (0.007) (0.003) Sample All Non-X X All FirmFE Y Y Y Y TimeFE Y Y Y Y N 523,719 350,094 173,625 168,848 The table reports the estimated β coefficient from Equation 9 with ln(1+imports) on the left-hand side in columns (1)-(3) report andln(1+exports)incolumn(4)atthequarterlylevel. Column(1)reportstheestimatedcoefficientoftheinteractionofaPost with domesticcurrencyleverage(DCL).DCLwasnormalizedtohavezeromeanandastandarddeviationof1.Column(1)reportstheresults forallfirms,column(2)focusesonnon-exporters,whilethesampleincolumns(3)and(4)islimitedtoonlyexporters.Standarderrors areclusteredatthefirmandquarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively.Thesampleislimited tothefirmsthateverimportedinourdata.Exportersarefirmsthateverexportedinourdata. 48

Table14: Panelregression ln(imports) ln(exports) (1) (2) (3) (4) ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ FCL i,t−1 5.008 6.715 3.586 2.098 (0.386) (0.651) (0.443) (0.329) FCL i,t−1 ×∆ER t−1 -0.737 -5.044 ∗∗ -0.465 -1.840 (1.254) (2.225) (1.458) (1.312) Sample All Non-Exporters Exporters All FirmFE Y Y Y Y TimeFE Y Y Y Y N 230,340 148,590 81,750 230,340 Thetablereportstheestimatedβcoefficientfrompanelregressiondiscussedinsubsection5.8withln(1+imports)ontheleft-hand sideincolumns(1)-(3)reportandln(1+exports)incolumn(4)attheannuallevel.Seesubsection3.2forthedefinitionofFCL,which wasnormalizedtohavezeromeanandastandarddeviationof1. Column(1)reportstheresultsforallfirms,column(2)focuseson non-exporters,whilethesampleincolumns(3)and(4)islimitedtoonlyexporters.Standarderrorsareclusteredatthefirmandquarter level. ***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. Thesampleislimitedtothefirmsthateverimportedin ourdata.Exportersarefirmsthateverexportedinourdata. 49

Figure1: Imports,exportsandtheexchangerateinColombia ThethickredlinerepresentsaggregateexportsandthedashedblacklinerepresentsaggregateimportsinColombiainmilliondollars. Tradevariablesareplottedontheleft vertical axis and indexed before the depreciation. The thin black line represents exchangerate(ColombianPesosperUSdollar,rightaxis). 50

Figure2: EstimatedImpactofImportsbyNon-Exportersasafunctionoflargerdominantcurrencyfinancing Thefigureplotspointestimates(reddots),95%(verticalbars)and99%(greyshade)confidencebandsoftheeffectofforeigncurrencyleverageonimports(leftverticalaxis).See Equation10forspecification. ThethinbluedottedlinerepresentsaverageColombian Peso/USdollarexchangerate(rightaxis). Standarderrorsareclusteredatthefirmand quarterlevels. Thesampleislimitedtofirmsthatimportedatleastonceandneverexportedinourdata. 51

Figure3: EstimatedImpactofImportsbyExportersasafunctionoflargerdominantcurrencyfinancing Thefigureplotspointestimates(reddots),95%(verticalbars)and99%(greyshade)confidencebandsoftheeffectofforeigncurrencyleverageonimports(leftverticalaxis).See Equation10forspecification. ThethinbluedottedlinerepresentsaverageColombian Peso/USdollarexchangerate(rightaxis). Standarderrorsareclusteredatthefirmand quarterlevels. Thesampleislimitedtofirmsthatimportedandexportedatleastonce inourdata. 52

Figure4: EstimatedImpactofExportsasafunctionoflargerdominantcurrencyfinancing Thefigureplotspointestimates(reddots),95%(verticalbars)and99%(greyshade)confidencebandsoftheeffectofforeigncurrencyleverageonexports(leftverticalaxis).See Equation10forspecification. ThethinbluedottedlinerepresentsaverageColombian Peso/USdollarexchangerate(rightaxis). Standarderrorsareclusteredatthefirmand quarterlevels. Thesampleislimitedtofirmsthatimportedandexportedatleastonce inourdata. 53

Figure5: CounterfactualImportswithoutForeignCurrencyDebt Thefigureplotstheactual(thickblueline)andcounterfactual(dashedredline)imports toColombia. Thecounterfactualimportsarecomputedassumingfirmsin2014Q2had zeroforeigncurrencyleverage. Seesubsection5.4forthedetailsofthecounterfactual exercise. 54

Figure6: CounterfactualNetExportswithoutForeignCurrencyDebt Thefigureplotstheactual(thickblueline)andcounterfactual(dashedredline)Colombia’s net exports. The counterfactual net exports are computed assuming firms in 2014Q2 had zero foreign currency leverage. See subsection 5.4 for the details of the counterfactualexercise. 55

Appendix A Tables TableA1: BaselinewithControls ln(imports) (1) (2) (3) (4) (5) (6) (7) Post×LS -0.068 ∗ -0.071 ∗∗ -0.069 ∗∗ -0.069 ∗∗ -0.069 ∗∗ -0.069 ∗∗ -0.070 ∗∗ (0.034) (0.033) (0.033) (0.033) (0.033) (0.033) (0.033) Size×Post -0.044 ∗∗ -0.045 ∗∗ -0.016 -0.011 -0.014 (0.016) (0.016) (0.022) (0.022) (0.022) BankCreditSupply×Post -0.212 -0.269 -0.270 -0.290 (0.242) (0.237) (0.237) (0.237) Age×Post -0.138 -0.141 -0.131 (0.083) (0.083) (0.083) Profits×Post 0.109 ∗∗ 0.109 ∗∗ (0.042) (0.041) Maturity×Post 0.080 ∗∗∗ (0.025) Sample Non-X Non-X Non-X Non-X Non-X Non-X Non-X Sector-TimeFE N Y Y Y Y Y Y FirmFE Y Y Y Y Y Y Y TimeFE Y Y Y Y Y Y Y N 350,934 305,518 305,518 305,518 305,518 305,518 305,518 Thetablereportstheestimatedβcoefficientsfrompanelregressiondiscussedinsubsection4.1complementedwithpre-depreciation (2013)firmlevelcharacteristicscontrolsinteractedwiththePostdummyandwithsector-timefixedeffects. Sizereferstologassets, BankCreditSupplyistheexposureoffirmstoabankcreditsupplyshockconstructedinthespiritofAmitiandWeinstein(2018). Age isthelogageofthefirm,profitsisEBITDAovertotalassets,andMaturityisthelogoftheaveragematurityoffirms’loans. ***,**,and* indicatesignificanceat10%,5%,and1%levelsrespectively. 56

TableA2: Imports: Quantities Units Kilograms (1) (2) (3) (4) (5) (6) Post×FCL -0.338 ∗∗∗ -0.317 ∗∗∗ (0.073) (0.074) Post×LS -0.099 ∗ -0.105 ∗∗ (0.050) (0.049) Post×WS -0.154 ∗∗∗ -0.161 ∗∗∗ (0.044) (0.043) Sample Non-X Non-X Non-X Non-X Non-X Non-X FirmFE Y Y Y Y Y Y TimeFE Y Y Y Y Y Y N 350,934 350,934 350,934 350,934 350,934 350,934 Thetablereportstheestimatedβcoefficientfrompanelregressiondiscussedinsubsection5.8whereincolumns(1)-(3)thelefthand sidetakesthelogofthenumberunitsimportedandcolumns(4)-(6)takesthelogkilogramsimportedbyfirm-quarter.Seesubsection3.2 forthedefinitionofFCL,whichwasnormalizedtohavezeromeanandastandarddeviationof1. Standarderrorsareclusteredatthe firmandquarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. 57

TableA3: Imports: ExtensiveMargin #Countries #Products (1) (2) (3) (4) (5) (6) Post×FCL -0.038 ∗∗∗ -0.075 ∗∗∗ (0.009) (0.017) Post×LS -0.009 -0.026 ∗ (0.006) (0.013) Post×WS -0.018 ∗∗∗ -0.041 ∗∗∗ (0.006) (0.011) Sample Non-X Non-X Non-X Non-X Non-X Non-X FirmFE Y Y Y Y Y Y TimeFE Y Y Y Y Y Y N 350,934 350,934 350,934 350,934 350,934 350,934 Thetablereportstheestimatedβcoefficientfrompanelregressiondiscussedinsubsection5.8whereincolumns(1)-(3)thelefthand sidetakesthelogoftheuniquenumberofimportcountrydestinationsandcolumns(4)-(6)ittakesthelognumberofuniqueproducts byfirm-quarter. Seesubsection3.2forthedefinitionofFCL,whichwasnormalizedtohavezeromeanandastandarddeviationof1. Standarderrorsareclusteredatthefirmandquarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. 58

TableA4: Robustness: Exporter ln(imports) (1) (2) (3) (4) (5) (6) Post×FCL -0.189 ∗∗∗ -0.187 ∗∗∗ -0.186 ∗∗∗ -0.189 ∗∗∗ -0.188 ∗∗∗ -0.193 ∗∗∗ (0.043) (0.044) (0.045) (0.043) (0.043) (0.043) Post×FCL×exporter 0.176 ∗∗∗ 0.151 ∗∗∗ 0.151 ∗∗∗ 0.151 ∗∗∗ 0.150 ∗∗∗ 0.151 ∗∗∗ (0.049) (0.049) (0.050) (0.050) (0.050) (0.050) Sample Nocontrols Size BankCreditSupply Age Profits Maturity p:β +β =0 0.574 0.289 0.289 0.247 0.244 0.202 1 2 N 461,746 461,746 461,746 461,746 461,746 461,746 Thetablereportstheestimatedβcoefficientsfrompanelregressiondiscussedinsubsection4.1complementedwithatripleinteraction withanexporterdummyinsteadofsplittingthesample. Columns(2)-(6)introducesequentiallyadditionaltripleinteractionsofpredepreciation(2013)firmlevelcharacteristicswithPost andFCL ontopoftheonesincludedinthepreviouscolumns. Sizerefersto logassets,BankCreditSupplyistheexposureoffirmstoabankcreditsupplyshockconstructedinthespiritofAmitiandWeinstein (2018). Ageisthelogageofthefirm,profitsisEBITDAovertotalassets,andMaturityisthelogoftheaveragematurityoffirms’loans. p:β +β =0displaysthepvaluefromat-testwhetherthesumofPost×FCLandPost×FCL×exporter areequaltozero.***,**,and 1 2 *indicatesignificanceat10%,5%,and1%levelsrespectively. 59

TableA5: Robustness: Exports ln(exports) (1) (2) (3) (4) (5) (6) (7) Post×FCL -0.010 -0.020 -0.009 -0.009 -0.005 -0.005 -0.003 (0.025) (0.025) (0.025) (0.025) (0.025) (0.025) (0.025) Size×Post -0.055 ∗∗ -0.055 ∗∗ 0.019 0.020 0.019 (0.020) (0.019) (0.020) (0.020) (0.020) BankCreditSupply×Post 0.031 -0.048 -0.046 -0.056 (0.342) (0.338) (0.338) (0.339) Age×Post -0.355 ∗∗∗ -0.355 ∗∗∗ -0.349 ∗∗∗ (0.073) (0.074) (0.072) Profits×Post 0.008 0.007 (0.017) (0.017) Maturity×Post 0.051 (0.036) Sample Non-X Non-X Non-X Non-X Non-X Non-X Non-X Sector-TimeFE N Y Y Y Y Y Y FirmFE Y Y Y Y Y Y Y TimeFE Y Y Y Y Y Y Y N 169,232 151,764 151,764 151,764 151,764 151,764 151,764 Thetablereportstheestimatedβcoefficientsfrompanelregressiondiscussedinsubsection4.1complementedwithpre-depreciation (2013)firmlevelcharacteristicscontrolsinteractedwiththePostdummy. Sizereferstologassets,BankCreditSupplyistheexposure offirmstoabankcreditsupplyshockconstructedinthespiritofAmitiandWeinstein(2018). Ageisthelogageofthefirm,profitsis EBITDAovertotalassets,andMaturityisthelogoftheaveragematurityoffirms’loans. ***,**,and*indicatesignificanceat10%,5%, and1%levelsrespectively. 60

TableA6: Investment ∆Investment (1) (2) (3) (4) (5) ∗∗ ∗ ∗∗ FCL -0.035 -0.101 -0.062 -0.100 -0.101 (0.027) (0.045) (0.041) (0.057) (0.043) Exporter 0.068 (0.050) FCL×Exporter 0.102 ∗ (0.055) Sample All Non-X Non-Hedge Non-X,Non-Hedge All N 16,924 10,621 14,225 9,956 16,924 Thetablereportstheestimatedβcoefficientsfromacross-sectionalregressionofthechangeininvestmentbetweenpre-depreciation and2014ontheforeigncurrencyleverageofthefirmbeforethedepreciation. ***,**,and*indicatesignificanceat10%,5%,and1% levelsrespectively. 61

TableA7: NewBorrowing ForeignCurrency (1) (2) (3) (4) (5) (6) (7) (8) (9) FCL×Post -0.284 ∗∗∗ 0.234 ∗∗∗ 0.008 (0.059) (0.050) (0.014) FCL×Post×exporter 0.034 -0.048 0.021 (0.077) (0.068) (0.019) LS×Post -0.100 ∗∗ 0.105 ∗∗∗ 0.003 (0.049) (0.039) (0.013) LS×Post×exporter -0.013 -0.058 -0.013 (0.069) (0.057) (0.018) WS×Post -0.092 ∗ 0.124 ∗∗∗ 0.012 (0.050) (0.039) (0.017) WS×Post×exporter -0.032 -0.048 -0.001 (0.069) (0.054) (0.021) Sample All All All All All All All All All FirmFE Y Y Y Y Y Y Y Y Y Bank×TimeFE Y Y Y Y Y Y Y Y Y N 364,911 364,911 364,911 364,911 364,911 364,911 364,911 364,911 364,911 ThetablereportstheestimatedβfromLoan =α+β FCE ∗Post +β FCE ∗Exporter ∗Post +α +α +(cid:178) where i,f,d(t),b 1 f t 2 f f t b,t i f,d(t),b Loanisthelogoftheloanvolumeofloan(i)tofirm(f)atissuancedate(d)bybank(b). Columns(1)-(3)reporttheresultsforloansin foreigncurrency,columns(4)-(6)reporttheresultsforloansinlocalcurrency,andcolumns(7)-(9)reporttheresultsforallloans. Post isinteractedwithliquidity(LS)andwealthshocks(WS).FCL,WS,andLSwerenormalizedtohavezeromeanandastandarddeviation of1.Postisdefinedasabinaryvariableequalto1fortheperiodafter2014Q2.TheExporter dummytakesthevalueoneifafirmever exportedinourdata. Seesubsection3.2forthedefinitionoftheindependentvariables. Standarderrorsareclusteredatthefirmand quarterlevel.***,**,and*indicatesignificanceat10%,5%,and1%levelsrespectively. 62

B Figures Figure A1: Estimated Impact of Imports by Non-Exporters and Exporters as a function of larger dominant currencyfinancing Thefigureplotspointestimatesforexporters(greendots)andnon-exporters(pinkdots) andtheir95%(verticalbars)confidencebandsoftheeffectofforeigncurrencyleverage onimportsforexportersandnon-exporters. SeeEquation10Thesampleislimitedto firmsthatimported. 63

FigureA2: MaturityStructureofLoans Thefigureplotstheloanfrequencybytheoriginalmaturityinmonths. Thered(blue) barsreflectloanshaveanoriginalmaturityofnon-whole(whole)years.Theblacksolid (dashed)linerepresentstheaverage(weighted)maturity.Thematurityiswinsorizedfor longerthan120months. 64

FigureA3: OilPriceandthePeso ThefigureplotstheOilPrice(leftaxis)andtheColombianPeso/USdollarexchangerate (rightaxis). 65

FigureA4: ColombianGDPgrowthandthePeso ThefigureplotstherealGDPgrowth(leftaxis)andtheColombianPeso/USdollarexchangerate(rightaxis). 66

FigureA5: NumberofFirmsperSector Thefigureplotsthenumberoffirmsineachsectorconsideredinourregressionsample. 67

FigureA6: SectorHeterogeneity Thefigureplotsthesector-levelcoefficientofourbaselinespecificationEquation9for eachsector. 68

FigureA7: SectorHeterogeneityandExportShare Thefigureplotsthesector-level(2-digitNAICS)coefficientofourbaselinespecification Equation 9 estimated for each sector separately against the share of exporters in this sector.Thesizeofthecirclerepresentsthenumberoffirmsinthesector. 69

C Analytical model This Appendix details the proofs used in the paper. We start with a simple model that follows Bernanke et al. (1999)first,andlateraddexporting. C.1 Simplemodel C.1.1 Set-up Definethefollowingobjects: Ψ(cid:161)δ¯(cid:162)=(cid:69)(cid:163)δ|δ<δ¯(cid:164) ]F(δ¯)+δ¯(1−F(δ¯)) (14) ζ(δ¯)=γ(cid:69)(cid:163)δ|δ<δ¯(cid:164) F(δ¯). (15) WLOGassumethat(cid:69)[δ]=1,then: Ψ(cid:48) (δ¯)=1−F(δ¯) (16) ζ(cid:48) (δ¯)=γδ¯f(δ¯) (17) and ζ(δ¯) δ¯(1−F(δ¯))=Ψ(δ¯)− . (18) γ Wecanthenexpressfirmprofitsas: πf =ρ(M) (cid:161) 1−Ψ(δ¯) (cid:162) Here ρ(M) are expected revenues of the firm. M are the inputs of the firm (imports). 1−Ψ(δ¯) is the share of revenues that goes to the firm. Ψ(δ¯) is the share of profits that goes to the bank inclusive of monitoring costs. Notethatbankprofitscanbewrittenas: πb=(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)ρ(M) 70

andsoζ(δ¯)istheshareofrevenuesthatgoestomonitoringcosts. M isprocuredatprice pM andisfinancedby debt(B)andequity(A). pMM =B+A. Hence,theproblemcanbewrittenas: maxρ(M) (cid:161) 1−Ψ(δ¯) (cid:162) (19) M,δ¯ s.t. ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)=R(pMM−A) (20) wheretheconstraintimpliesthattheprofitsofthebankareequaltotherisk-freerateRtimestheamountlentto thefirm pMM−A,asbanksarecompetitive. ThepropertiesoffunctionsΨandζwilldependonthefunctional formsofthedistributionofδ. C.1.2 Optimization Lagrangianisgivenby: L=ρ(M) (cid:161) 1−Ψ(δ¯) (cid:162)+λ(cid:161)ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)−R(pMM−A) (cid:162) (21) FOC: M : 1 ρ (M) (cid:163) 1−Ψ(δ¯)+λ(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)−λR=0 (22) M pM δ¯: −ρ(M)Ψ(cid:48) (δ¯)+λρ(M) (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162)=0 (23) λ: ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)=R(pMM−A) (24) Assume for now that ER affects only net worth of the firm and that there is an interior solution.40 Then from Equation23wecanexpress: λ(cid:161)δ¯(cid:162)= Ψ(cid:48) (δ¯) . Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) Let 1 ρ (M) r ≡ pM M . (25) R 40SeeBernankeetal.(1999)fortheconditionsunderwhichuniquesolutionexists. 71

This is the ratio between marginal revenues from a dollar spend on imported inputs and a rate of return to the bank. FromEquation22weget: λ(δ¯) r(δ¯)= (26) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) We canshow thatr (cid:48) (δ¯)>0 and thus ε >0,41 inother wordsthe wedge betweencostoffunds to the bankand rδ¯ expected return to capital is increasing with the cutoff. Note that we can express the actual interest rate on the loanas: B¯ ρ(M)δ¯ 1+i = = . (27) B B SinceB¯=ρ(M)δ¯,asrevenuesatcutoffmustbethesameastotalpayments. FromEquation24: ρ(M) R = . (28) B (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162) PlugEquation28intoEquation27toget Rδ¯ 1+i = . (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162) C.1.3 Derivationsofelasticities Wecannowderiveseveralelasticitiesofinterest. Ψ(δ¯)−ζ(δ¯)−δ¯(Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯)) δ¯ ε =R 1+i,δ¯ (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)2 Rδ¯ (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162) Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) =1−δ¯ Ψ(δ¯)−ζ(δ¯) =1−ε Ψ−ζ,δ 41SeeBernankeetal.(1999)fortheproof. 72

Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) ε Ψ−ζ,δ =δ¯ Ψ(δ¯)−ζ(δ¯) (29) (1−F(δ¯))δ¯−γδ¯2f(δ¯) = <1 (30) (1−F(δ¯))δ¯+(1−γ)(cid:69)[δ|δ<δ¯]F(δ¯) Sinceγδ¯2f(δ¯)>0and(1−γ)(cid:69)[δ|δ<δ¯]F(δ¯)>0,thenumeratorishigherthandenominator.Itfollowsthatε Ψ−ζ,δ < 1andsoε >0 1+i,δ¯ ConsiderEquation25: ε =ε ε r,δ¯ ρ M,M M,δ¯ ε ε = r,δ¯ <0, M,δ¯ ε ρ M,M asε r,δ¯ >0andε ρ M,M <0byassumptionthatmarginalrevenuesaredecreasingininputs. ConsiderFOCforλ: ε ε ε +ε ε =ε ρ,M Mδ¯ δ¯,A Ψ−ζ,δ¯ δ¯,A pMM−A,A pMM −A ε ε ε +ε ε =ε +ε ρ,M Mδ¯ δ¯,A Ψ−ζ,δ¯ δ¯,A pMM,A pMM−A A,A pMM−A pMM A ε ε ε +ε ε =ε ε − ρ,M Mδ¯ δ¯,A Ψ−ζ,δ¯ δ¯,A M,δ¯ δ¯,ApMM−A pMM−A − A ε = pMM <0. δ¯,A ε (cid:179) ε − pMM (cid:180) +ε M,δ¯ ρ,M pMM−A Ψ−ζ,δ¯ Notethatbyassumptionε ρ,M <1. Sincefirmsstartwithnon-negativenetworth A>0and pM pM M M −A >1(assuming that pMM−A > 0 or there would be no need to borrow in the first place ), otherwise they would default before the beginning of the period and start with zero net worth. Since we proved that ε <0, the first term in the M,δ¯ denominatorispositiveasaproductoftwonegativenumbers. Bernankeetal.(1999)showthatε ispositive, Ψ−ζ,δ¯ 73

so denominator is positive, and the ratio is negative as the numerator is negative. Hence, the demand shock cutofffordefaultdecreaseswithnetworth(firmswithhighernetworthhavelowerprobabilityofdefault). C.2 Modelwithexporting C.2.1 Set-up Tomakeprogress,assumethatrevenuefunctiontakesanexplicitform ρ (M)=M σ σ −1 (31) NX fornon-exporters,and σ−1 σ−1 ρ (M)=M σ +M σ Fe; M +M =M (32) X D F D F for exporters. These functional forms can arise when demand curves are isoelastic up to some normalization. HereF reflectstherelativesizeoftheforeignmarket. Notethatexportrevenuesnowdependonexchangeratee. C.2.2 Optimization WecanrewriteLagrangianEquation21as: L= (cid:181) M σ σ −1 +M σ σ −1 Fe (cid:182) (cid:161) 1−Ψ(δ¯) (cid:162)+λ (cid:181)(cid:183) M σ σ −1 +M σ σ −1 Fe (cid:184) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)−R(pM(M +M )−A) (cid:182) (33) D F D F D F TakingFOCswithrespecttoM ,M yields D F 1 σ−1 M − σ 1(cid:163) 1−Ψ(δ¯)+λ(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)=λRpM (34) pM σ D 1 σ−1 M − σ 1 Fe (cid:163) 1−Ψ(δ¯)+λ(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)=λRpM (35) pM σ F Fromwhichitfollowsthat: −1 −1 M σ =M σ Fe (36) D F 74

or M =M (Fe) σ (37) F D NotethatsinceM =M +M : D F M (cid:161) 1+(Fe) σ(cid:162)=M (38) D M M = (39) D 1+(Fe) σ σ M(Fe) M =(Fe) σ M = (40) F D 1+(Fe) σ Thepreviousequationtellsusthatthefirmwillusetheshare 1 ofinputstoproducegoodssolddomestically, 1+(Fe)σ withthissharefallinginF –therelativesizeoftheforeignmarket PluggingEquation39andEquation40intotheexpressionfortherevenuesforexportersEquation32weget: (cid:181) (cid:182)σ−1 (cid:181) σ (cid:182)σ−1 M σ M(Fe) σ ρ (M)= +Fe (41) X 1+(Fe) σ 1+(Fe) σ (cid:195) (cid:183) (cid:184)σ−1 (cid:183) σ (cid:184)σ−1(cid:33) =M σ σ −1 1 σ +Fe (Fe) σ (42) 1+(Fe) σ 1+(Fe) σ Let (cid:183) (cid:184)σ−1 (cid:183) σ (cid:184)σ−1 1 σ (Fe) σ 1+κ(e)= +Fe (43) 1+(Fe) σ 1+(Fe) σ 1+(Fe)1+σσ σ −1 = (44) (1+(Fe) σ ) σ σ −1 1+(Fe) σ = (45) (1+(Fe) σ ) σ σ −1 =(cid:161) 1+(Fe) σ(cid:162) σ 1 (46) 75

Notethat1+κ(e)>1itisincreasingine sinceσ>1 ρ (M)=(1+κ(e))ρ (M) (47) X NX Thelastexpressiontellsustherevenuesofexportersare1+κ(e)timeslargerthantherevenuesofnon-exporters conditional on the same amount of inputs used M where ρ (M) ≡ ρ(M) = M σ σ −1 . So now we can solve the NX problem: max(1+κ(e))ρ(M) (cid:161) 1−Ψ(δ¯) (cid:162) (48) B,δ¯ s.t. (1+κ(e))ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)=R(pMM−A) (49) andcomparetheelasticitiesofinterestacrossexportersandnon-exportersbysettingκ(e)=0. Notethatinthis caseexchangeratewillaffectthefirmsalsothroughparameterκ(e) C.2.3 Optimization WestartwithsettingtheLagrangian: L=(1+κ(e))ρ(M) (cid:161) 1−Ψ(δ¯) (cid:162)+λ(cid:161) (1+κ(e))ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)−R(pMM−A) (cid:162) (50) FOC: M : (1+κ(e)) 1 ρ (M) (cid:163) 1−Ψ(δ¯)+λ(cid:161)δ¯(cid:162)(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)−λ(cid:161)δ¯(cid:162) R=0 (51) M pM δ¯: −(1+κ(e))ρ(M)Ψ(cid:48) (δ¯)+λ(1+κ(e))ρ(M) (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162)=0 (52) λ: (1+κ(e))ρ(M) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)=R(pMM−A) (53) FromEquation52wegetthesameexpressionforλasbefore: λ(cid:161)δ¯(cid:162)= Ψ(cid:48) (δ¯) (54) Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) Nowthewedgebetweenmarginalproductandriskfreerateisgivenby: 76

1+κ(e)ρ (M) r ≡ pM M (55) R FromEquation51weget: λ(δ¯) r(δ¯)= (56) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) Notethattheexpressionfortheinterestrateisnowgoingtobe: B¯ (1+κ(e))ρ(M)δ¯ 1+i = = (57) B B FromEquation53 (1+κ(e))ρ(M) R = (58) B (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162) andhence Rδ¯ 1+i = (59) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162) The expressionforthe nominal interestrateisthe same asbefore andwe will have thesame expression for the elasticityofinterestratewithrespecttoδ¯. NotethatfromEquation(53)wecanshowthatexportershavelower probabilityofdefault. Sinceincreasein1+κ(e)increasesLHS,bycontradictionwecanshowthatinresponseδ¯ hastodecreasetomakebothsidesoftheequationequalagain.42 C.2.4 Elasticities Letε 1+κ(e),e ≡ϑ>0. WecannowuseEquation55toget: ε r,e =ϑ+ε ρ M,M ε M,e (60) 42ThisisdrivenbythefactthatRHSelasticitywithrespecttoMisgreaterthan1,whileelasticityofρ(M)withrespecttoMislessthan 1,soifδ¯increasesforexporters,whichleadstoafallinM andRHSfallsmuchmorethanLHS. 77

Notethatwecanalsoexpresselasticityofthewedgeas: ε =ε ε (61) r,e r,δ¯ δ¯,e andhence ε r,δ¯ ε δ¯,e =ϑ+ε ρ M,M ε M,e (62) ϑ+ε ε ε = ρ M,M M,e (63) δ¯,e ε r,δ¯ Let pMM =nandε =−l. Asaresult,n>1 −A =1−n<0andl >0byassumption(weassumethatfirms’ pMM−A A,e pMM−A net worth is negatively affected by a depreciation, which will happen, for example, when they borrow more in foreigncurrencythantheyownforeigncurrencyassets). FromEquation53 ϑ+ε ρ,M ε M,e +ε Ψ−ζ,δ¯ ε δ¯,e =nε M,e +(1−n)(−l) (64) SubstituteEquation63intoEquation64toget: ϑ+ε ε ϑ+ε ρ,M ε M,e +ε Ψ−ζ,δ¯ ρ ε M,M M,e =nε M,e +(1−n)(−l) (65) r,δ¯ (cid:195) ε ε (cid:33) (cid:195) ε (cid:33) ε M,e ε ρ,M + Ψ−ζ ε ,δ¯ ρ M,M −n =−l(1−n)−ϑ 1+ Ψ ε −ζ,δ¯ (66) r,δ¯ r,δ¯ Andhence: (cid:179) ε (cid:180) (n−1) −ϑ 1+ Ψ ε −ζ,δ¯ ε =l + r,δ¯ (67) M,e ε ρ,M + ε Ψ−ζ ε ,δ¯ ε ρ M,M −n ε ρ,M + ε Ψ−ζ ε ,δ¯ ε ρ M,M −n r,δ¯ r,δ¯ Equation67tellsusthatforeignexchangeratemovementshavetwoeffectsonfirmsize.Thefirsttermistheeffect throughforeigncurrencyliabilities. Thesecondtermistheconventionaleffectthroughhigherexportearningin localcurrency. Sincen>1>ε ρ,M (thelastinequalityisbyassumption),ε ρ M,M <0(alsooneoftheassumptions), ε >0, and ε >0 the denominator in both terms is negative. Since n >1 and l >0, the numerator in the Ψ−ζ,δ¯ r,δ¯ first term is positive and hence the whole term is negative – as was the case with non-exporters, exchange rate 78

depreciationreducesnetworthofthefirmwhichmakesitriskierforthebankstolend. NotethatthesecondterminEquation67ispositive. Thedenominatorisnegativeaswediscussedabovethe numeratorisnegativesinceϑ>0.Depreciationleadstohigherrevenuesandstimulatesfirm’sexpansionthrough thischannel. Notethatbecauseofthissecondchannel,exporterswillalsobelargerandhavelowerprobabilityofdefault. In the empirical specification, we will estimate the effect of exchange rate depreciation conditional on foreign currencyborrowing,capturebyl inthismodel. Inotherwords,wewillestimatethetermproportionalto: (n−1) β∝ (68) ε ε ε ρ,M + Ψ−ζ ε ,δ¯ ρ M,M −n r,δ¯ while the second term in Equation 67 that is common within exporters (and within non-exporters) who have different initial levels of foreign currency leverage will be captured by firm fixed effects. Note that under our simplifyingassumptionsonrevenues,ε ρ,M = σ σ −1 andε ρ M,M = − σ 1 andthus (n−1) β∝ (69) ε σ−1− 1 Ψ−ζ,δ¯ −n σ σ ε r,δ¯ Hencetherelativestrengthofthelendingchannelbetweenexportersandnon-exporterswilldependontheratio ε of Ψ−ζ,δ¯ . Notethat ε r,δ¯ Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) ε Ψ−ζ,δ =δ¯ Ψ(δ¯)−ζ(δ¯) (70) Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯)1−F(δ¯) =δ¯ (71) Ψ(δ¯)−ζ(δ¯) 1−F(δ¯) Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯)1−F(δ¯) =δ¯ (72) Ψ(δ¯)−ζ(δ¯) Ψ(cid:48) (δ¯) 1−F(δ¯) 1 =δ¯ (73) Ψ(δ¯)−ζ(δ¯)λ(δ¯) wherethethirdlinecomefromEquation16andthefourthlinecomesfromEquation54. 79

Notethat r (cid:48) (δ¯) ε = δ¯ (74) r,δ¯ r(δ¯) UsingEquation56: λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)−λ(δ¯) (cid:163)−Ψ(cid:48) (δ¯)+λ(cid:48) (δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)+λ(δ¯) (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162)(cid:164) r (cid:48) (δ¯)= (75) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)2 PlugEquation54intoEquation74toget: λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)−λ(δ¯) (cid:104) −Ψ(cid:48) (δ¯)+λ(cid:48) (δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)+ Ψ(cid:48) (δ¯) (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162) (cid:105) r (cid:48) (δ¯)= Ψ(cid:48)(δ¯)−ζ(cid:48)(δ¯) (76) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)2 λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)−λ(δ¯) (cid:163)−Ψ(cid:48) (δ¯)+λ(cid:48) (δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)+Ψ(cid:48) (δ¯) (cid:164) = (77) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)2 λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164)+λ(cid:48) (δ¯)λ(δ¯) (cid:163)(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)−λ(δ¯)λ(cid:48) (δ¯) (cid:163)(cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) = (78) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)2 λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164) = (79) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)2 λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164) λ(δ¯) = (80) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)λ(δ¯) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164) = r(δ¯) (81) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)λ(δ¯) where in Equation 80 we multiplied numerator and denominator by λ(δ¯) and in Equation 81 we substituted Equation56. PlugEquation81intoEquation74toget: λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164) δ¯ ε = r(δ¯) (82) r,δ¯ (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164)λ(δ¯) r(δ¯) λ(cid:48) (δ¯)δ¯ (cid:163) 1−Ψ(δ¯) (cid:164) = (83) λ(δ¯) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) Note that from Equation 69 we know that the reaction of imports as a function of a probability to default will 80

ε dependontheratioof Ψ−ζ,δ¯ whichwecancalculateusingEquation73andEquation83: ε r,δ¯ ε δ¯ 1−F(δ¯) 1 Ψ−ζ,δ¯ = Ψ(δ¯)−ζ(δ¯)λ(δ¯) (84) ε r,δ¯ λ(cid:48)(δ¯)δ¯ (cid:163) 1−Ψ(δ¯) (cid:164) λ(δ¯) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) 1−F(δ¯) = Ψ(δ¯)−ζ(δ¯) (85) λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) Assumethatexportersareveryunlikelytodefault(theprobabilityofdefaultapproacheszero). Weknowthatthe ratioaboveispositive. Considerthelimitoftheratiowhenδ¯→0: ε 1−F(δ¯) lim Ψ−ζ,δ¯ =lim Ψ(δ¯)−ζ(δ¯) (86) δ¯→0 ε r,δ¯ δ¯→0λ(cid:48) (δ¯) (cid:163) 1−Ψ(δ¯) (cid:164) (cid:163) 1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)(cid:164) andnotethat: lim (cid:163) 1−F(δ¯) (cid:164)=1 (87) δ¯→0 bydefinitionofCDF lim (cid:163)Ψ(δ¯)−ζ(δ¯) (cid:164)=0 (88) δ¯→0 fromEquation14andEquation15asbothexpectationsgotozeroanddensityisassumedtobebounded lim (cid:163) 1−Ψ(δ¯) (cid:164)=1 (89) δ¯→0 forthesamereasonasabove limλ(δ¯)=1 (90) δ¯→0 81

as Ψ(cid:48) (δ¯) limλ(δ¯)=lim (91) δ¯→0 δ¯→0 Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) 1−F(δ¯) =lim =1 (92) δ¯→01−F(δ¯)−γδ¯f(δ¯) whereEquation92comesfromEquation16andEquation17. Asaresult: lim1−Ψ(δ¯)+λ(δ¯) (cid:161)Ψ(δ¯)−ζ(δ¯) (cid:162)=1 (93) δ¯→0 Finally: Ψ(cid:48)(cid:48) (δ¯) (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162)−Ψ(cid:48) (δ¯)(Ψ(cid:48)(cid:48) (δ¯)−ζ(cid:48)(cid:48) (δ¯)) limλ(cid:48) (δ¯)=lim (94) δ¯→0 δ¯→0 (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162)2 −Ψ(cid:48)(cid:48) (δ¯)ζ(cid:48) (δ¯)+Ψ(cid:48) (δ¯)ζ(cid:48)(cid:48) (δ¯) =lim (95) δ¯→0 (cid:161)Ψ(cid:48) (δ¯)−ζ(cid:48) (δ¯) (cid:162)2 f(δ¯)γδ¯f(δ¯)+(1−F(δ¯)(γf(δ¯)+γδ¯f (cid:48) (δ¯)) =lim (96) δ¯→0 (cid:161) 1−F(δ¯)−γδ¯f(δ¯)) (cid:162)2 =γδ¯f (cid:48) (δ¯)=0 (97) wherethelastlinecomesfromthe(Assumption4). Asaresult, ε lim Ψ−ζ,δ¯ =∞ (98) ε δ¯→0 r,δ¯ andhence limβ=0 (99) δ¯→0 Hence,exporterswhohaveaprobabilityofdefaultclosetozerowillnotreacttotheeffectsofexchangerate through the balance sheet. In addition, there is theoretical and empirical evidence that firms who price their exports in dominant currency are more profitable. In terms of this model, higher probability implies a lower probability of default and thus DCP exporters are even more likely than exporters not to react to exchange rate 82

movementsthroughtheirbalancesheets. ε Assumingthatdemandshocksaredistributedlognormally,wecanalsoshowthatthe Ψ−ζ,δ¯ termisdecreas- ε r,δ¯ ing for any δ¯ and since exporters have a lower cutoff demand shock, they will be less elastic to exchange rate movementsviathefinancingchannel(seeFigureA8). ε FigureA8: Demandshockcutoffand Ψ−ζ,δ¯ underlognormaldistribution ε r,δ¯ 10 9 8 7 6 5 4 3 -10 -8 -6 -4 -2 0 2 4 C.3 Dominantcurrencypricing Considernowthecaseinwhichallintermediateinputsareimported,thefirmsstartwithanoptimalpricingand productiondecision,butcan’tchangethepriceoftheexportedgoodaftertheexchangerateshock. Therearetwo typesoffirms: somesetpricesinpesos(PCPexporters),andsomesetpricesindollars(DCPexporters). Finally, assumethatforeigndemandisisoelastic. Assumethatdomesticcurrencydepreciates. Inafrictionlessworld,the firmswouldprefertokeeptheirmarkupconstant,andsince100%ofinputsispricedinforeigncurrency,export pricewouldcomovewiththeexchangerate. Inthisworld,theoptimalexportpricewouldbeconstantindollars and DCP firms would hence be more profitable. Note that since the price of their exports in foreign currency doesn’tchange,theDCPexporterswillusethesameamountofinputs MF toproduceexportedgoodsasbefore the exchange rate depreciation. What happens when we introduce financial frictions into the analysis? Using Equation (54), Equation (55), and Equation (56) we can show that optimal production and hence export would 83

weaklydecrease.4344 SinceDCPexporterswouldusethesameamountofinputsfortheproductionofexported goods MF while PCP firms would increase their exports (as their price in foreign currency drops) – they would befurtherawayfromtheoptimalallocation, hencelessprofitable, whichwouldinducebankstochargethema higherinterestrateanddecreasetheiroutputdomesticallybyrelativelymorethantheDCPfirms. 43Bycontradiction,assumethatproductionisconstantorincreases,thenfromEquation(55)wewouldhaveadropinr aspM would increase by more than 1+κ(e) and from Equation (56) this would lead to a drop in δ¯This would violate Equation (56) as M would increasefasterthanρ(M)andpM wouldincreasefasterthan1+κ(e) 44Outputwouldn’tchangeforthosefirmswhoonlyexportasthenκ(e)andpM wouldmoveone-to-one. 84