‘dollar’ debt in colombian firms: are sinners punished during devaluations?

Emerging Markets Review 4(2003) 417–449

1566-0141/03/$ - see front matter� 2003 Elsevier B.V. All rights reserved.PII: S1566-0141Ž03.00063-3

‘Dollar’ debt in Colombian firms: are sinnerspunished during devaluations?

Juan Carlos Echeverry , Leopoldo Fergusson *,a a,

Roberto Steiner , Camila Aguilarb a

Department of Economics, Universidad de los Andes, Cra 1, 18A-70 Bogota, Colombiaa

Board of Directors, International Monetary Fund, 700 19th St., Washington DC 20431, USAb

Abstract

During the 1990s, the performance of several emerging economies was linked to theiraccess to foreign capital. Colombia was no exception, experiencing a boom and bust cycleassociated with an initial period of real exchange rate appreciation followed by a sharpdepreciation. Although several studies have discussed the recent underperformance of theColombian economy, few attempts have been made at analyzing firm-level data. We rely oninformation for a large sample of firms during 1995–2001 and examine the determinants offoreign indebtedness and the effects on firm performance of holding dollar debt amid realdepreciations(i.e. the so-called ‘balance sheet effect’). In our data set, matching does seemto take place to the extent that firms in more open sectors and exporting firms have highershares of dollar debt. Our estimations also reveal a negative balance sheet effect on firms’profitability, while the effect on investment is generally not significant.� 2003 Elsevier B.V. All rights reserved.

JEL classifications: E22; F31

Keywords: Colombia; Investment; Devaluation; Balance sheet effects

1. Introduction

The traditional expansionary effect of a devaluation predicted by the Mundell–Fleming model has been recurrently subjected to criticism(i.e. Krugman and Taylor,1978), and challenged on new grounds(Calvo, 1999, 2000; Calvo and Reinhart,

*Corresponding author. Tel.:q57-1-339-4949; fax:q57-1-332-4492.E-mail address: [email protected](L. Fergusson).

418 J.C. Echeverry et al. / Emerging Markets Review 4 (2003) 417–449

2000; Dornbusch, 2001). The basic argument in this new strand of literature is thatfirms and governments that borrow in foreign currency and produce an output thatis not entirely tradable face a currency mismatch which, following a devaluation,can produce a balance sheet effect that offsets any enhancement in competitiveness.Largely motivated by the failure of traditional models of balance of payments crisisto explain the financial turmoil in emerging markets during the 1990s, a number ofauthors have appealed to the so-called ‘open economy Bernanke–Gertler’ argument(Krugman, 1999a). According to this view, in a context of financial marketimperfections where net worth affects investment levels(Bernanke and Gertler,1989), substantial levels of foreign currency denominated liabilities imply thepossibility of self-fulfilling crises: a loss of confidence by foreign investors and thecapital flight that results leads to a currency depreciation and a balance sheet effectthat depresses investment.The actual implications and policy prescriptions in this setting have not been

settled on theoretical grounds. On the one hand, some argue that a loose monetarypolicy after a crisis is not a remedy, as it reinforces the currency depreciation andits balance sheet effect(Krugman, 1999b). Aghion et al.(2000, 2001) argue thatthe balance sheet effect of a devaluation might entail a decrease in economic activitywhich reduces money demand and weakens the currency even further. Thus, acurrency crisis is a ‘bad’ equilibrium, with low output and a weak exchange rate.They suggest that if credit supply does not react too strongly to changes in theinterest rate, a tight monetary policy is the correct prescription to avoid a crisis.However, Cespedes et al.(2000, 2002) point out that the offsetting effect of´

increased home output and returns to investment generally imply that the standardMundell–Fleming expansionary effect of devaluations still obtains. Their modelsuggests that fluctuations in output and investment are larger and more persistentunder fixed exchange rates. Nonetheless, balance sheet effects do matter in that theymagnify the effects of foreign disturbances and might lead to a situation offinancialfragility—where devaluations increase the country risk premium.If balance sheet effects matter, a puzzling question is why domestic agents choose

to hold foreign denominated liabilities in the first place. Explanations of the so-called ‘original sin’—the fact that in developing countries agents cannot borrow intheir own currencies at long maturities—range from models that point out moralhazard problems of fixed exchange rates(Burnside et al., 1999; Schneider andTornell, 2001) to those which consider the role of financial underdevelopment(Caballero and Krishnamurthy, 2000).In contrast to the theoretical discussion, empirical work on the determinants of

liability dollarization and its ‘balance sheet effects’ at the firm-level has been scarceand severely hindered by data availability. Exceptions include Bleakley and Cowan(2002) (BC henceforth) study for a sample of publicly traded Latin American firms.Surprisingly, they report that the effect on performance of holding dollar debt duringdevaluation is positive as the negative net worth effect is more than compensatedby the effects of devaluation on ex post earnings. They suggest that this resultsfrom firms’ matching the currency composition of their liabilities and earnings.

419J.C. Echeverry et al. / Emerging Markets Review 4 (2003) 417–449

These conclusions are not supported by Aguiar(2002), who studies investmentin post-crisis Mexican firms. Even though exporters outperform non-exporters interms of profits and sales after devaluation, their investment is constrained as aresult of holding foreign currency denominated debt. Floating the currency alsoimplies an increase in sales volatility, which further reduces investment. Aguiar doesfind that exporting and large firms tend to borrow more in dollars. These results areconsistent with those presented in Gelos(2003), who uses a larger database ofMexican firms and finds that the share of dollar debt is positively correlated withimports, exports and size.Regarding the Asian case, Claessens et al.(2000) describe the balance sheet

composition of a large sample of non-financial firms in East Asia, and argue thatseveral of the vulnerabilities in corporate financial structures that became apparentafter the 1997 crisis were already present during the early 1990s. Both currency andmaturity mismatches were significant, as short-term external financing was neededto finance high investment in the context of low profitability. Nonetheless, thisstudy does not investigate how exchange rate movements translated into firm’sperformance in the presence of such risk factors. Harvey and Roper(1999) findthat Asian corporate managers ‘bet’ their firms by issuing foreign currencydenominated debt and increasing leverage in spite of poor corporate performance.Betting that the exchange rates would remain fixed, Asian corporate managersgreatly exacerbated the crisis.Finally, an important issue for policy discussion that has received some attention

is the role of the exchange rate regime on exchange risk hedging. Arteta(2002)uses a database on deposit and credit dollarization in developing and transitioneconomies to examine whether flexible exchange rate regimes encouragebanks tomatch dollar-denominated liabilities with assets. His results indicate that, if anything,floating regimes tend to exacerbate currency mismatches. According to Martinezand Werner(2002), such results are not supported in the case of publicly listedMexican firms. These authors point out that ‘original sin’ actually implies a sort ofnatural tendency for liability dollarization that goes beyond the choice of exchangerate regime. Their results indicate that Mexican firms have taken exchange rate riskmore seriously after flotation in 1994. Indeed, during the fixed exchange rate regimethe share of dollar debt was mainly determined by firm size and unaffected by thecurrency composition of revenues, whereas during floating exports became the soledeterminant of dollar indebtedness.

2. The Colombian experience in 1995–2001

All in all, the available empirical evidence on the balance sheet effects ofdevaluations is not conclusive. The final verdict has to come from the data and theparticular conditions of firms in specific countries. In this paper, we study the firm-level effects of monetary and exchange rate developments in Colombia during1995–2001. Like many other emerging economies, Colombia experienced increasinglevels of capital inflows during the first half of the 1990s that allowed for arespectable performance in terms of GDP growth. Nonetheless, a curtailment of


Fig. 1. Evolution of main macroeconomic variables, Colombia, 1993–2001.

foreign financing after 1997 coincided with the worst growth performance on recordthereafter(Fig. 1, Panel A). Ample foreign financing brought about an appreciationof the real exchange rate(RER) between 1990 and 1997, and a significant realdepreciation was observed afterwards(Fig. 1, Panel C).These exchange rate developments occurred in the context of different regimes.

Until late 1991 a crawling peg was in place, supported by an array of capital andexchange controls. In 1992 and 1993, and in order to consolidate into one operationthe purchase of foreign currency and the subsequent sterilization, the central bankissued so-called Certificates of Exchange(CE) in order to acquire foreign exchange.Holders of CE’s could keep them to maturity(originally 3 months, eventually 1year) and convert them into pesos at the official exchange rate or they could sellthem upon issuance at a 12.5% discount to the central bank. An informal exchangerate band was thus introduced, since CE’s could be sold in the secondary market.In 1994, the central bank abandoned this sterilization policy, and the issuance ofCE’s. Instead, a formal band was put in place. While capital inflows remainedstrong, the band had to be shifted to allow for nominal appreciation twice. After


1997, the critical situation in international capital markets called for two deprecia-tions of the band, which was eventually abandoned in 1999, when a floating regimewas introduced. In general, the latter half of the 1990s in Colombia was characterizedby unprecedented macroeconomic turmoil and volatility(a brief chronology of themain economic and political events is presented in Table 1).During the second half of 1997 and 1998, the first time the band came under

attack, many observers argued that the lackluster performance of the economy wasassociated with an ill-conceived monetary and exchange rate policy that kept interestrates too high and the currency too strong. Under this interpretation, floating thecurrency should have reverted the trend of the key components of aggregate demand.The stylized facts indicate that during the period of floating the recovery ofconsumption and investment(Fig. 1, Panel D) has been far from satisfactory, whilethe so-called non-traditional exports(those different from coffee, oil and coal) haveperformed reasonably well. Whether this relatively poor performance is associatedwith a protracted effect of having instrumented a tight monetary policy to defendthe currency when it came under attack andyor with the balance sheet effectassociated with the depreciation following flotation is an empirical matter, betteraddressed at the level of the firm. In this paper, we rely on information for a largesample of firms during 1995–2001 and examine the impact of the exchange andinterest rates on the performance of firms with varying degrees of foreign indebt-edness, output tradability, and imported inputs.Since the limited time dimension of our database makes the causal interpretation

of macroeconomic effects on firm performance problematic, we focus on the effectof exchange and interest rate movements on the performance of firms with differentcharacteristics. Our study does not allow us to pinpoint the overall expansionary orcontractionary effect of devaluations, but rather the existence(or lack thereof) ofparticular channels whereby the exchange rate affects firm performance. Moreover,much of the effect of devaluations in Colombia at the macroeconomic level is likelyto occur in the public sector, as the public deficit is largely financed with externaldebt.The paper proceeds as follows. Section 3 describes the data set and Section 4

discusses the basic analytical framework. Section 5 presents our results and Section6 concludes. Our results suggest that matching does seem to take place to the extentthat firms in more open sectors and exporting firms are engaged more often inforeign indebtedness and have higher shares of dollar debt. Also, firm size is(notsurprisingly) the most robust and significant determinant of dollar indebtedness.Although these results and the limited amount of dollar denominated indebtednessin Colombian firms tilt the balanceagainst finding any balance sheet effect ofdevaluations, we find evidence of a negative balance sheet effect on firms’profitability. Results for investment, however, are mixed.

3. Data description and summary statistics

The most ambitious attempt so far to empirically tackle this problem is BC,whose sample consists of 2644 publicly traded firms in 5 countries, including

422J.C

.E

cheverryet

al./

Em

ergingM

arketsR

eview4

(2003)417–449

Table 1Chronology of important political, financial and economic events in Colombia

Date Event

Main political eventsAugust 1994 Ernesto Samper’s presidency begins in the midst of a scandal associated with illegal

campaign contributions from drug traffickers.(A long-lasting legal investigation, theproceso8000, begins. Congress exonerates the president in late 1995, and the United Statesannounces negative assesment of Colombia’s fight against drugs early in 1996. This processof ‘certification’ for drug-figthing was only removed in February 1998).

August 1998 Andres Pastrana, who had formerly lost to Samper and had provided the clues leading to the´proceso 8000, is elected president.

January 1999 Peace process with Colombia’s largest guerrilla group(FARC) begins; government grants ademilitarized zone of 42 000 sq km for conversations.(The peace process and the FARCzone would come to an end in february 2002 after a prolonged crisis of the negotiationsstarting in 2001).

Main economic eventsJanuary 1994 Sterilization policy abandoned and adoption of formal exchange rate band.March 1994–August 1994 Tighter restrictions on domestic credit and foreign loans imposed; new deposit requirements

and restrictions on short-term foreign borrowing(reserve requirements had been relaxed in1991 and re-established in 1993).

December 1994 Band shifted to allow for nominal appreciation.September 1995–November 1995 Introduction of limit on net foreign exchange position of financial intermediaries and

elimination of minimun required exchange rate position of banks and other financialInstitutions.

February 1996 50% non-remunerated deposit requirement imposed on foreign loans with maturities under 3Days.

Fourth quarter 1996 First signs of economic recession.1997 Monetary and fiscal policy shock aimed at increasing money supply and economic activity.1998 Attack on exchange rate band.January 1998 Deposit requirements on foreign loans reduced.August 1998–December 1998 Central bank reduces reserve requirements on domestic deposits.September 1998 Band shifted to allow for depreciation.1999 4.5% yearly contraction of GDP(first yearly negative figure in 70 years).

423J.C

.E

cheverryet

al./

Em

ergingM

arketsR

eview4

(2003)417–449

February-1999–December 1999 Central Bank interest rate cuts in virtually every month.June 1999 Band widened and shifted to allow for depreciation.September 1999 Formal band abandoned and exchange rate allowed to float.December 1999 Agreement on program with the IMF.November 2000 Limit on net foreign exchange position of financial intermediaries is increased from 20 to 50%

of net worthMarch 2001–December 2001 Central Bank interest rate cuts in virtually every month.

Sources: Banco de la Republica and Bekaert and Harvey(2002).


Table 2Number of firms per year and per sector of economic activity

1995 1996 1997 1998 1999 2000 2001*

Agriculture 520 570 611 630 613 573 349Mining 123 138 159 168 152 137 109Manufacturing 1701 1805 1921 1963 1845 1756 1212Electricity, gas and water 14 16 24 18 17 16 13Construction 656 737 853 887 806 704 406Commerce 1438 1541 1738 1846 1716 1622 1058Transport and communications 257 292 350 361 349 333 220Services 1359 1496 1699 1783 1650 1536 873

TOTAL 6068 6595 7355 7656 7148 6677 4240

Source: Authors calculations based on Superintendencia de Sociedades and Superintendencia deValores. Revised data set, see Appendix A.(*) The reason for the decline in the sample is explained inAppendix A.

Colombia. Their sample is biased to Brazil and Mexico and to publicly traded firms,generally the largest and most financially sophisticated ones. We use a morerepresentative database, which covers an average of 8246 firms from 1995 through2001. These firms belong to 66 sectors(4-digit ISIC classification), and most ofthem are under the supervision of the ‘Superintendencia de Sociedades’(Superin-tendency of Corporations).Firms entering after 1995 or leaving before 2001, because they ceased to operate,

lead us to work with an unbalanced panel. We modified the data set in several waysas explained in Appendix A. Table 2 includes the number of firms per year andsector that survived our filtering criteria(see Appendix A).Table 3 presents descriptive statistics for some important firm-level variables(all

variables are defined in Appendix A). The breakdown of liabilities by currencydenomination, maturity and financial vs. trade-related debt is presented in Panel A,whereas Panel B presents descriptive statistics on foreign debt and exports by sector.A more detailed description of our database can be found in www.iadb.orgyRES.Total liabilities (excluding net worth) are close to 45% of total assets both on

average and for the median firm. Noteworthy is the fact that firms hold a largeproportion of short-term debt(close to 75% of total liabilities on average and 90%for the median firm). This is consistent with the available evidence on firms’financial opportunities in Colombia, where internal resources are often the sourceof funding for investment, while short-term debt is a source of working capital.Trade debt(with suppliers) accounts for 19% of total liabilities on average and 9%for the median firm. Financial debt, in turn, represents 29% of total liabilities onaverage.It should be noted that all foreign currency denominated debt is with overseas

creditors, as financial institutions are not allowed to lend in foreign currency. Totaldollar debt includes debt with foreign suppliers and debt with banks and financialcorporations overseas, whereas non-trade or financial dollar debt refers exclusivelyto the latter. The distinction is important, as trade related dollar debt is not usually


Table 3Firm-level variables, descriptive statistics(all ratios in percent, average 1995–2001)

Panel A: All firms

Variable Mean Median

Total debtyTotal assets 44 45Short-term debtyTotal debt 76 91Trade debtyTotal debt 19 9Financial debtyTotal debt 29 22Total ‘Dollar’ debtyTotal debt 5 0Financial Dollar debtyTotal debt(whole sample) 2 0Financial Dollar debtyTotal debt(firms holding dollar debt)* 23 15

ExportsyTotal revenue 5 0

Panel B: Descriptive statistics by industry

FinancialDollar

ExportsyTotalrevenue

DebtyTotaldebt

Mean Median Mean Median

Agriculture 16 0 2 0Mining 4 0 1 0Manufacturing 8 0 2 0Electricity, gas and water 2 0 7 0Construction 0 0 1 0Commerce 3 0 1 0Transport and communications 3 0 3 0Services 1 0 2 0

Notes: (*) On average, 511 firms per year corresponding to nearly 8% of sample.Source: Authors calculations based on Superintendencia de Sociedades and Superintendencia de’

Valores.

considered in the balance sheet story, yet it is an important component of foreigndebt in our sample. The share of total ‘dollar’ debt is low on average(close to 5%)and most firms hold no foreign currency denominated liabilities. Financial dollardebt in the entire sample is significantly lower, close to 2% of total debt on average.Also, even though about 27% of the firms in our sample hold a positive amount ofdollar debt of any kind(not shown in Table), the proportion of firms holding dollardebt that is not trade-related is significantly lower, 8% of the sample. Looking onlyat firms holding financial dollar debt, we find that those firms hold on averageapproximately 23% of their liabilities in the form of(non-trade) dollar debt.Unfortunately, we are not able to identify whether or not foreign debt is acquiredwith a parent firm abroad.With regard to the revenue side, most firms do not export at all, although a few

export their entire output. The average share of income generated abroad, while stilllow, has increased through time(from 4.5% in 1995 to 6.5% in 2001). Regardingthe currency composition of inputs, we rely on sectoral data on imported inputs for


estimation purposes, using the most disaggregated information available from theeconomy’s input–output matrix. For most sectors, imported input shares do notchange much through time, though there is significant heterogeneity in terms ofimport orientation by sector. When examined by sector, exports are important forfirms in agriculture, manufacturing and mining, although most firms do not exportat all, regardless of which sector they are in. Financial foreign debt is also relativelyhigher on average for some sectors, in particular infrastructure: the electricity, gasand water sector and transport and communications(Table 3, Panel B).We now turn to a description of the main correlates of firms’ characteristics. In

Fig. 2, we identify each firm as belonging to one of the three zones in thefinancialforeign debt-exports space:hell, heaven and hedge. Firms are hedged when facingexchange rate devaluation if the share of their output that is denominated in foreigncurrency is ‘similar’ to their share of foreign denominated liabilities. Arbitrarily, weset the upper and lower bounds of the hedge area in the linesshare of foreigndebts(3y2)=share of exports andshare of foreign debts(2y3)=share of exports,respectively. Firms are in hell when their share of foreign debt is significantly1

larger than their share of exports. In the opposite extreme, firms in heaven have alarger proportion of their output in dollars as compared to their share of dollar debt.Obviously, firms in heaven would actually be in hell if a real exchange rateappreciation occurred. Also, the definition of hedging considered here is quitelimited in scope, as it has to do with the extent of mismatch between the currencycomposition of output and liabilities. In our data set, we are unable to observewhether firms use financial instruments such as derivatives or forwards to hedgetheir foreign indebtedness.The distribution of firms and the average value of assets for firms in each zone

are presented for 2000 only, as it varies little through time. Most of the firms in oursample belong to the hedge zone(79%), largely because many firms neither haveforeign debt nor export. Firms in heaven follow in importance(17%), whereas asmaller proportion of firms(4%) are in hell. In terms of size, firms in hell are thelargest on average, whereas those in hedge are the smallest.Another feature of our sample is illustrated in Fig. 3, which depicts the evolution

of fixed capital investment by firms in each of the three zones described.Interestingly, fixed capital investment falls precipitously for all firms in 1997 butthe effect is stronger to those in the hedge zone.A natural alternative to determine whether our sample of firms is sufficiently

representative within the Colombian economy would be to compare the level ofinvestment undertaken by our sample of firms with national accounts data forinvestment at the corporate level. However, since our investment figures come fromcash flow data(see Appendix A), they are not directly comparable to nationalaccounts data. A suggestive figure to gauge whether our sample is representative

It must be pointed out that these bounds are chosen arbitrarily because of their geometrical appeal:1

they imply that the hell, heaven and hedge areas are of the same size. We experimented with alternativedefinitions of these areas and results are very similar. Furthermore, we considered an alternative zoneclassification in thenet exports-foreign debt space; a firm might actually find itself in ‘hell squared’during devaluation if it has negative net exports besides from holding a large share of foreign debt.


Fig.2.

Sha

reof

Non

-Trade

Foreign

Deb

tvs.Exports,20

00.


Fig. 3. Average Fixed Capital Investment, by zone.

refers to non-traditional exports. Once we exclude firms in the coffee, coal and oilsectors, exports of firms in our sample amount to 82% of total non-traditionalexports in the national accounts. Likewise, total dollar debt of firms in our sampleaccounts for approximately three fourths of foreign debt acquired by the non-financial private sector, as reported by the central bank.

4. Framework for the firm-level analysis

Since we closely follow BC’s empirical strategy, it is useful to consider the basicintuition behind their model of the impact of exchange rate movements for firmlevel investment and its variation across firms with different levels of liabilitydollarization (the model with two simple extensions is summarized in AppendixB). In addition to the usual expansionary or ‘competitiveness’ effect of devaluations,they consider the fact that for dollar-indebted firms devaluations might lead to adecrease in ‘net worth’ due to a currency ‘mismatch’ between liabilities and income.This deterioration in balance sheets makes firms appear as riskier investments. Asa result, they face higher interest rates, which bring about a decline in investment.The model can be easily extended in two directions, discussed by the authors but

not incorporated in their model. First, firms might use imported inputs, challengingthe fact that the ‘competitiveness’ effect of devaluation is necessarily positive.Second, firms pay an interest rate that depends not only on their own net worth,but also on the macroeconomic environment. For example, quitting the ‘dogged’defense of a currency allows the domestic interest rate to decrease, fosteringinvestment of firms indebted in pesos. Thus, in attempting to evaluate empirically


the effect of devaluation on investment, it is important to take into account both theextent to which firms tend to import their inputs as well as domestic creditconditions. In our empirical analysis, we explicitly control for the tradability offirms output and inputs and their interaction with foreign exchange movements, aswell as for the interaction of domestic debt with domestic credit conditions.In Section 5, we estimate alternative versions of the following investment

equation:

* *I sm qg D =DRER qdD qbDRERqfD q´ (1)Ž .it i i,ty1 t i,ty1 t ty1 i,t

whereI is the rate of fixed capital investment at timet for firm i, with investmentit

in property, plant and equipment normalized by total assets. The main effect thatwe want to capture is the interaction between the real exchange rate devaluation(DRER), and dollar debt at the beginning oft, . Two alternative definitionsUDt i,ty1

of are considered: the ratio of lagged dollar debt to total assets and to totalUDi,ty1

debt for firm i. Since foreign debt is presumably denominated in dollars, to capturethe ‘balance sheet effect’, we define the real exchange rate(RER) as the nominalbilateral exchange rate with the US adjusted by domestic inflation. By interactingthe (log) percentage change in this real exchange rate index with the share offoreign debt, we capture the differential effect that real exchange rate devaluationhas on investment for firms with varying degrees of foreign debt exposure. Thespecifications consider firm fixed effects,m . An additional set of regressors includesi

firms’ lagged leverageD and the direct effects of foreign indebtedness and realty1

exchange rate devaluation.´ is the error term.t

Although BC’s estimations are a good starting point, they have some limitations.First, putting too much emphasis on any coefficient attached to macro variables isproblematic, because these variables only change through time, and are likely to becorrelated with omitted macro variables that could be captured in a year-specificcomponent of the error term. Thus, the coefficient attached to a macro variable(e.g.RER devaluation) may be inconsistent. On more technical grounds, estimating Eq.(1) implies dealing with a number of econometric problems. For instance, theinconsistency arising from the fact that firm-specific effects might be correlatedwith the set of independent variables may be tackled through the use of the withinestimator, but this estimator does not account for the fact that most right-hand sidevariables might be endogenous. Also, one might be interested in allowing theregressions to have a dynamic structure.A generalized method of moments(GMM) estimator based on the use of lagged

observations of the dependent and explanatory variables allows us to deal with theseproblems (Arellano and Bover, 1995; Bond, 2002). To address the problem ofpossible bias induced by correlation of firm specific effects with independentvariables, the regression equation is differenced. Also, to address the problem ofjoint endogeneity, suitably lagged values of the original(i.e. in levels) independentvariables, including the lagged value of the dependent variable are used asinstruments for the right-hand side variables(i.e. the differenced values of theoriginal regressors) in the transformed equation. The validity of the moment


conditions implicit in this ‘GMM difference estimator’ is tested statistically. First,we present results for a Sargan test of over-identifying restrictions that checks theoverall validity of these moment conditions. Under the maintained assumption thatthe error term of the original dynamic levels equation is serially uncorrelated, thetransformed error term for the difference equation is expected to have serialcorrelation of first order, but not of second order. Thus, we report AR(1) and AR(2)tests on the lack of serial correlation for the transformed error term. These teststatistics are asymptotically normal under the null of no serial correlation.A drawback of the difference GMM estimator is that the instruments available

for the transformed regression equation are weak when the individual series havenear unit root properties. Indeed, if the series are highly persistent, their differencesare nearly innovations and there are no good instruments for near white noise series.Thus, the GMM difference estimator can be subject to finite sample biases. Thispotential bias can be reduced using the ‘GMM system estimator’ proposed byArellano and Bover. This estimator combines the regression expressed in firstdifferences with the original equation expressed in levels. As before, suitably laggedvalues of the dependent variables in levels are used as instruments for the differencedequation, whereas the equation in levels is instrumented with lagged differences ofthe explanatory variables. Both the Sargan and serial correlation tests are examinedin this case. A difference Sargan test is useful in this context, since the set ofmoment conditions specified under the simple difference estimator is a subset ofthe one considered in the system estimator. The difference between the Sarganstatistic obtained under the system estimator and the one obtained under thedifference estimator is asymptotically distributed Chi-square(x ) with degrees of2

freedom given by the difference between the number of degrees of freedom of thesystem estimator and that of the difference estimator. Failure to reject the nullhypothesis of the validity of additional restrictions gives support to the systemestimator.Taking the former considerations into account, in Section 5 we estimate(using

GMM) alternative specifications in which we drop the non-interacted macrovariables and include year specific effects to capture the overall macroeconomicenvironment affecting our sample. We concentrate our attention on the role of firmspecific variables and, perhaps more importantly, their interaction with macroeco-nomic variables.Another important element in our empirical analysis has to do with the determi-

nants of foreign indebtedness, as this might determine the actual sign attached tothe key interaction coefficientg in Eq. (1). BC argue that, in their sample, firmsholding dollar debt invest more than firms holding peso debt in the period followinga devaluation(i.e. g is positive) because they match the currency composition ofdebt with the elasticity of their income to the real exchange rate. They reportregressions where the ratio of dollar debt to total liabilities is a(positive) functionof several proxies for the sensitivity of profits to the real exchange rate. Thus, westart our empirical investigation examining whether this result holds in our data.


Table 4Determinants of Currency Composition of debt, 1995–2001

Dependent variable: non-trade Foreign debt to Total debt

Tobit Probit Tobit Probit

Independent variablesOpenness dummy 8.94*** 0.45***

(1.25) (0.06)Exports 0.15*** 0.008***

(0.02) (0.001)Imports 0.64 0.035

(0.43) (0.02)Log (Assets) 13.58*** 0.63*** 12.60*** 0.59***

(0.48) (0.02) (0.43) (0.02)Foreign participation 0.03*** 0.0012** 0.04*** 0.002**

(0.01) (0.001) (0.01) (0.0006)D1996 2.38** 0.16** 2.48** 0.17***

(1.01) (0.05) (1.04) (0.05)D1997 y0.78 0.02 y1.20 0.009

(1.02) (0.05) (1.11) (0.05)D1998 y2.21** y0.06 y2.98*** y0.09**

(1.03) (0.05) (1.06) (0.05)D1999 y4.83*** y0.20*** y5.55*** y0.22***

(1.10) (0.05) (1.14) (0.06)D2000 y8.23*** y0.33*** y9.50*** y0.39***

(1.13) (0.06) (1.17) (0.06)D2001 y9.15*** y0.36*** y9.96*** y0.40***

(1.20) (0.06) (1.25) (0.06)Constant y274.28*** y12.70*** y252.44*** y11.79***

(7.85) (0.38) (7.20) (0.40)No of observations 45739 45739 38042 38042No of firms 7738 7738 6446 6446Likelihood Ratio test 1326.1(0.00) 1362.67(0.00) 1135.03(0.00) 1115.36(0.00)PseudoR-square 0.03 0.08 0.03 0.07

Notes: Asymptotic standard errors below coefficients in parenthesis,P-values for regressions statisticsin lower panel appear in( ); *, significant at the 90% level, ** at the 95%, *** at the 99%. See AppendixA for variable definition.

5. Results

5.1. Currency composition of debt

In Table 4, we consider a set of alternative specifications for the determinants ofnon-trade related dollar debt. We are interested in knowing whether larger firms andwith more tradable output use more external credit. Additionally, a number ofauthors have found that firms with international operations are more likely to holdforeign debt. In our data set, the information on whether a firm is a parent or


subsidiary is unreliable, so we include the share of foreign ownership in eachfirm.2

Regressions for the extent of dollar indebtedness are problematic since conven-tional estimators are biased and inconsistent in the context of limited dependentvariables. Thus, we report estimates of a Tobit model for the extent of dollarindebtedness as well as Probit estimations for the likelihood of holding dollar debt.For all estimations, we report(and fail to accept) Likelihood Ratio tests for thejoint lack of significance of the regressors. The results presented in Table 4 showthat foreign debt is positively correlated with firms’ size(i.e. log of assets). In thefirst column, the Tobit estimation reveals that firms with foreign ownership have asignificantly higher share of dollar debt to total debt, and the time dummies indicatea negative trend in the share of financial dollar debt. In this equation, belonging toa relatively open sector(agriculture, mining or manufacturing) is also a significantdeterminant of indebtedness in dollars. Likewise, foreign participation in ownershipis a significant and positive determinant of the extent of foreign indebtedness.Results for size, openness, foreign participation and the time dummies are qualita-tively unchanged in our Probit estimation shown in the second column. These3

results provide evidence of matching of liabilities and income streams. The negativetrend implies that firms were able to reduce their share of foreign debt holdings ata rate that exceeded the devaluation rate. Notice that the time effects are negativefrom 1999 onwards, coinciding with the floating of the currency, probably suggestingthat firms took exchange rate risk more seriously once the exchange rate band wasabandoned.In the final columns, we drop the openness dummy and consider instead firms’

exports and imports as dependent variables. Our regressions show that exports, sizeand foreign participation are significant and positive determinants of foreignindebtedness, whereas the imported input share is not significant. The time dummiesagain show a downward trend.In sum, our results suggest that matching does seem to take place to the extent

that firms in more open sectors and exporting firms are engaged more often inforeign indebtedness and have higher shares of dollar debt. In this regard, it is alsointeresting that financial dollar debt has a downward trend during the devaluationperiod and that imports do not exert a significant effect on financial dollar debt.Finally, as expected, size is the most robust determinant of dollar indebtedness, andthere is also evidence that the degree of foreign ownership increases dollar debt.

Perhaps the most important determinant of the extent of dollar indebtedness is the interest rate2

differential that each firm faces when considering different financing options. Since we are workingwith low frequency data, it is difficult to find a reliable measure of such differential.

To check the sensitivity of the results, we ran regressions for the share of total dollar debt(i.e.3

trade and non-trade related). There are only two significant changes. First, and not surprisingly, thereis a positive correlation between imports and the total share of dollar debt. Second, the somewhatnegative time trend for financial dollar debt does not hold for total dollar debt. A drawback from theprevious estimations is that in our sample there is a substantial number of firms holding no dollar debt.Thus, in regressions, inference is drawn from the difference between a very small share of firms ascompared to the majority of firms holding no dollar debt. There is no simple way to deal with thisissue.


Table 5Zone-performance regressions, 1995–2001

Dependent variable

Fixed capital investment Profits

whole sample 1995–1998 1999–2001 whole sample 1995–1998 1999–2001

Independent variablesZone dummy y0.13 y0.39 y0.61*** y1.56 y1.85 y3.23***

(0.77) (1.16) (0.25) (1.02) (1.38) (0.88)Sectoral GDP 0.06** 0.08 0.01 0.11** 0.06 0.13**

(0.03) (0.09) (0.01) (0.04) (0.11) (0.04)Leverage(ty1) 0.02 0.045** y0.031*** 0.036 y0.024 0.06***

(0.01) (0.02) (0.005) (0.019) (0.02) (0.02)Constant 0.45 y0.45 1.97*** y1.02 2.47** 2.05**

(0.77) (1.070) (0.27) (1.05) (1.27) (0.79)Observations 37842 19801 18041 37842 19801 18041Firms 7637 7292 7146 7637 7292 7146R-square 0.0002 0.0004 0.002 0.0003 0.0007 0.0025Wald test 4.21(0.24) 5.50 (0.14) 42.09(0.00) 10.92(0.012) 3.35 (0.34) 39.37(0.00)

Notes: Random effects GLS regressions; standard errors below coefficients in parenthesis;P-valuesfor regressions statistics in lower panel appear in( ) *, significant at the 90% level, ** at the 95%, ***at the 99%. See Appendix A for variable definition.

5.2. Firm performance

We now turn to the analysis of firm performance. Before considering BC’s Eq.(1) above, we present some regressions for firm investment(net purchases of fixedassets as percent of total assets at the beginning of the period) as a function of the‘zone dummies’(Fig. 2). The ‘zone dummy’ is assigned a value ofy1 for firmsin heaven, 0 for firms in hedge, and 1 for firms in hell. A negative coefficient for4

this dummy would indicate that firms in heaven tend to perform better than hedgedfirms, which in turn perform better than those in hell. Indeed, this is the result weobserve in the first column of Table 5, yet the coefficient is not significant. SectoralGDP growth has a positive effect on performance, whereas lagged leverage is notsignificant. The sign of this coefficient is undetermined a priori, as a high level ofleverage could increase financial constraints and debt payments could reduce theamount of internal funds available for investment, yet a positive coefficient mightarise from the fact that only firms that have access to credit are able to invest.Finally, the Wald test for overall significance of the regressors shows that thisregression has low explanatory power. Thus, in the second and third columns we

It must be pointed out that our dummy specification imposes a special structure as it implicitly4

assumes that the effect for performance of being in hell is the opposite to that of being in heaven. Wechecked the sensitivity of our results by including separate dummies for firms in hell and in heaven. Inthe case of profits, we found similar results in the sense that firms in hell performed worse than thosein heaven in the period following the devaluation. In the case of investment, a puzzle remains sincewhen dummies are included separately the negative effect for firms in hell that is presented below doesnot show up.


repeat the exercise for the periods before and after 1999(with 1999 onwards beingthe period of floating and sharpest real depreciation). Interestingly, we obtain asignificant and negative coefficient for the latter period. Furthermore, the regressionsignificantly improves its overall fit after 1999, indicating that the zone classificationwas a better predictor of performance during the strong depreciation period. SectoralGDP growth is no longer significant in each sub-period and, quite interestingly,lagged leverage changes sign across sub-periods. Apparently, in the latter period thefinancial constraint effect dominated the overall sign attached to this variable.In the last three columns of Table 5, we present additional Zone-Performance

regressions with profits(relative to total assets) as the performance variable. Results5

indicate that there is no effect from the zone-dummy prior to the strong devaluationperiod and a negative effect from 1999 onwards, and the difference in the coefficientsbetween the two periods is significant. Regarding controls, sectoral GDP growthhas a positive effect on performance(except in the first period) while laggedleverage has a positive effect during the late period. Once again, these regressionssignificantly improve their overall fit after 1999.We interpret these results as suggesting the presence of a negative balance sheet

effect for firms that produce non-tradables and are indebted abroad. Since theseresults tend to contradict BC’s, we now turn to the estimation of alternative versionsof Eq. (1) above to confirm that these opposing results stem from differences inthe data set employed rather than from the specification considered. Results from6

a set of regressions including firm-fixed effects are presented in Table 6. The firstpoint that should be highlighted is that the direct effect of the real exchange ratedepreciation is consistently negative and significant, whereas the interaction of thebilateral devaluation with dollar debt is also negative but never significant. In other7

words, the effect of the variation in the exchange rate is negative for all firmsirrespective of the denomination of their debt, and dollar indebted firms do not fareany better(as they do in BC’s sample).As noted in Section 4, if firms attempt to match their revenue and expense

streams, the currency composition of debt might be correlated with a number ofadditional firm characteristics, such as the currency denomination of their revenueand inputs. Thus, in columns 2 and 4, we directly control for the degree of tradabilityof the firms’ output and inputs. When these tradability terms are included, oursample is reduced, as we cannot impute reasonable shares of imports to all of the

Other left-hand side variables considered in unreported regressions are sales and cash flow to assets.5

We consistently find that hedged firms and firms in heaven tend to perform better than firms in hell,particularly after 1999. We also performed regressions in which we interacted the zone dummy variablewith year specific dummies for the period after 1999, obtaining negative coefficients and thus confirmingthat firms in hell tended to perform worse during the devaluation period.

Regressions including total dollar debt were also run, and some of the results are reported below.6

We performed estimations for end of period and average percentage change of RER. To ease reading,we shall present results with the latter measure, noting which results change when end-of-perioddepreciation is considered instead. Estimations were undertaken using DPD for OX developed byManuel Arellano, Stephen Bond and Jurgen A. Doornik.

Whereas dollar debt is interacted with the bilateral real exchange rate devaluation, when devaluation7

enters independently, it is measured by the effective(multilateral) real exchange rate devaluation.


Table 6Fixed capital investment regressions(BC), 1995–2001

Dependent variable: fixed capital investment

dollar debt to total assets dollar debt to total debt

Direct effectsD Log (RER) y0.23*** y0.29*** y0.23*** y0.30***

(0.06) (0.08) (0.05) (0.09)

InteractionsD Log (bilateral US ‘RER’)=D*(ty1) y0.28 y0.02 y0.07 y0.09

(0.51) (0.04) (0.30) (0.50)D Log (RER)=Exports(ty1) 0.047** 0.05**

(0.02) (0.02)D Log (RER)=Imports(ty1) 0.006 0.006

(0.005) (0.005)

ControlsD*(ty1) y0.09 y0.01 y0.03 y0.007

(0.10) (0.01) (0.06) (0.01)D (ty1) y0.01 y0.0001 y0.011 y0.001

(0.02) (0.002) (0.03) (0.002)Exports(ty1) y0.003 y0.003

(0.003) (0.003)Imports(ty1) y0.06*** y0.005***

(0.001) (0.001)Sectoral gdp growth y0.026 y0.0005 y0.027 y0.0005

(0.04) (0.005) (0.05) (0.004)R2 0.012 0.02 0.015 0.0190Observations 37612 23156 27962 13227Firms 7567 4572 5619 3004F-test 4.41(0.00) 31.43(0.00) 4.31 (0.000) ** 31.54 (0.000)**

Notes: Firm-fixed effects regressions, using within estimator. Robust standard errors in parentheses.P-values forF-test of overall lack of significance ofregressors appear in( ); *, significant at the 90%level, ** at the 95%, *** at the 99%. For variable definition see Appendix A and Eq.(1).

firms in the sample. The direct effect of the devaluation is still negative andsignificant, and its interaction with the degree of foreign indebtedness remainsinsignificant. We do find evidence, however, that exports, when interacted with realexchange rate devaluation have a positive impact on investment. Other controls aregenerally not significant. The only exception refers to the share of imports exertinga negative effect on investment. Likewise, sectoral GDP growth does not have asignificant effect on investment in any of the specifications.In short, these results, which replicate in our sample the estimation undertaken

by BC, a set of publicly traded Latin American corporations, indicate thatdevaluations have a negative effect on investment and that foreign indebtedness(ifanything) makes matters worse.As explained in Section 4, although these estimations are suggestive, they have

limitations in that they do not deal with the possible endogeneity of right-hand sidevariables and in that the interpretation of the direct coefficient of the RER


devaluation is unclear in such a short panel. In what follows we drop the non-interacted macro variables and include year specific effects to capture the overallmacroeconomic environment. Since BC argue that their result is driven by the fact8

that dollar indebted firms see their sales and earnings rise after a devaluation, inunreported regressions we also employed the same framework used for investmentto examine the effects of a devaluation on earnings and sales. Results actually lentsome evidence to the fact that dollar indebted firms experienced a decline ratherthan an upsurge in cash flow and sales during devaluation when compared to peso-indebted firms.Moving away from BC’s framework, in Tables 7 and 8, we consider an alternative

set of regressions where we control for potentially relevant omitted factors andexamine, together with the behavior of investment, the response of profits as analternative performance measure. We include the interaction between dollar debt atthe end ofty1 and alternative channels for the real exchange rate, and interactions9

of exchange rate terms with lagged exports and imports. As noted in Section 4, wemight see firms investing more after a devaluation not because individually theybenefit from a ‘competitiveness effect’ but because collectively this allowed for alooser monetary policy. Likewise, prior to a devaluation, we could see theminvesting less because they face higher interest rates under a dogged defense of thecurrency. To account for these possible effects, we include interest rate termsinteracted with different measures of indebtedness. Time-specific effects are alsoincluded.Although we present results for only one measure of dollar debt, short-term debt

and exports, results do not change when alternative definitions are included. We10

estimate static fixed effects specifications as well as dynamic GMM estimations.11

In unreported regressions we considered a specification for investment where besides from the key8

interaction term and time-specific effects, only lagged leverage and the lagged ratio of dollar indebtednessto assets are included as additional regressors. We also considered specifications where the share ofdollar debt was interacted with time specific dummies. Results indicated that it is difficult to find asignificant degree of heterogeneity in firm response to exchange rate movements depending on theirlevel of foreign indebtedness. Though probably not surprising considering the limited degree of foreignindebtedness of Colombian firms, these results reveal that BC’s result is unlikely to hold in our sampleof firms (i.e. dollar indebted firms do NOT fare any better during devaluations).

In the case of fixed capital investment, we interacted firm level variables both with thechanges in9

macro variables and with theirlevels. The latter specification could be motivated by anaccelerationistapproach(see Bond et al., 1997) whereby the desired level of capital depends, for instance, on thelevel of the exchange and interest rates so that the ratio of investment to total assets depends on thedevaluation rate and the rate of change of the interest rate.

The exchange rate devaluation refers to average devaluation of the bilateral real exchange rate,10

short-term debt is the share of short-term domestic financial debt scaled by total financial debt andexports are obtained at the level of the firm. In the case of exports and imports, interactions of amultilateral real exchange rate are considered instead. Regressions were run for end of perioddepreciation, the overall share of short-term indebtedness and for sectoral data on exports. Results weremostly unchanged.

Estimating dynamic specifications with the OLS and within estimators is useful, since the former11

is usually biased upwards and the latter downwards. For all reported estimations, we run OLS and fixedeffects regressions to check that our GMM estimators, presumably consistent, lied between the two.The difference in the estimators were often large, suggesting the presence of significant firm-specificeffects.

437J.C

.E

cheverryet

al./

Em

ergingM

arketsR

eview4

(2003)417–449

Table 7Detailed fixed capital investment regressions, 1995–2001


Independent variables *Changes *Levels

Fixed effects GMM difference GMM system Fixed effects GMM difference GMM system

Dependent variable(ty1) 0.066** 0.031 0.058** 0.031(0.028) (0.024) (0.020) (0.020)

InteractionsD*(ty1)=Bilateral US RER in«* 0.116 0.0910 0.150 y0.002 0.0002 y0.001

(0.166) (0.217) (0.150) (0.003) (0.002) (0.001)Exports(ty1)=RER in«* 0.317* 0.110 0.19** 0.13** 0.16** 0.14*

(0.170) (0.12) (0.070) (0.042) (0.08) (0.080)Imports(ty1)=RER in«* 0.03 y0.130 y0.149 y0.02 y0.507* y0.55**

(0.03) (0.290) (0.227) (0.02) (0.275) (0.240)Short-term financial domestic debt(ty1)=Real interest lending rate in«* 0.38 0.030 0.059 0.70 y0.166 y0.110

(0.51) (0.14) (0.164) (0.73) (0.229) (0.196)

ControlsShort-term financial domestic debt(ty1) 0.01 y0.002 0.023* y0.76 0.192 0.137

(0.012) (0.020) (0.012) (0.818) (0.266) (0.221)D (ty1) y0.016 y0.195 0.102 y0.016 -0.30* 0.076

(0.030) (0.217) (0.096) (0.030) (0.171) (0.090)D*(ty1) y0.046 0.019 0.017 0.10 0.162 0.120

(0.033) (0.053) (0.042) (0.215) (0.151) (0.120)Exports(ty1) y0.003 y0.005 0.005 0.584 y0.753** y0.66*

(0.007) (0.032) (0.013) (0.193) (0.349) (0.370)Imports(ty1) y0.004 y0.150 y0.066 0.097 2.34* 2.6**

(0.003) (0.110) (0.046) (0.088) (1.380) (1.124)Sectoral GDP growth y0.007 y0.032 y0.076 0.001 y0.120 y0.027

(0.070) (0.142) (0.160) (0.078) (0.140) (0.150)

Time effects YES YES YES YES YES YES

438J.C

.E

cheverryet

al./

Em

ergingM

arketsR

eview4

(2003)417–449

Table 7(Continued)


Independent variables *Changes *Levels

Fixed effects GMM difference GMM system Fixed effects GMM difference GMM system

Observations 23164 23164 23164 23164 23164 23164Firms 4570 4570 4570 4570 4570 4570Wald test( joint) 26.81(0.003) 25.41(0.008) 21.95(0.025) 29.60(0.000) 46.69(0.000) 31.55(0.001)Wald test(time) 154.2(0.000) 29.30(0.000) 114.0(0.000) 166.1(0.000) 18.85(0.002) 33.13(0.000)Sargan test 18.55(0.818) 50.18(0.276) 20.60(0.714) 47.30(0.379)Difference Sargan test 31.63(0.047) 26.7 (0.144)**AR(1) y12.29(0.000) y12.25(0.000) y12.11(0.000) y12.25(0.000)AR(2) 0.1316(0.895) y0.676(0.499) y0.321(0.748) y0.676(0.499)

Notes: Time effects included. Robust standard errors in parentheses;P-values for Wald test of overall lack of significance of regressors and for jointsignificance of time dummies appear in( ); * denotes significance at the 90% level, ** at the 95%, *** at the 99%. Lags datedty2 and ty3 for dollardebt, short-term debt, leverage and exports were included as instruments for GMM difference estimations. See Appendix A for variable definition.


Table 8Profits regressions, 1995–2001

Dependent variable: profits

Fixed effects GMM difference GMM system

Independent variablesDependent variable(ty1) y0.300 0.029

(0.280) (0.045)InteractionsD*(ty1)=Bilateral US RER y0.009** y0.009* y0.007*

(0.004) (0.005) y0.003Exports(ty1)=RER y0.060 0.003 0.005**

(0.158) y0.003 y0.002Imports(ty1)=RER y0.01 y0.007 y0.029**

(0.03) (0.008) (0.01)Short-term financial domestic debt(ty1)= y0.41 0.003 0.00005Real interest lending rate (0.40) (0.003) y0.002

ControlsShort-term financial domestic debt(ty1) 0.51 y0.4 0.003

(0.400) (0.321) (0.25)D(ty1) 0.034 0.387 0.299**

(0.047) (0.400) (0.150)D*(ty1) 0.609** 0.470 0.321

(0.272) (0.434) (0.256)Exports(ty1) 0.358 y0.435 y0.713**

(0.789) (0.3062) (0.312)Imports(ty1) 0.035 0.821 3.93***

(0.140) (0.918) (1.73)Sectoral GDP growth y0.090 0.075 0.591

(0.150) (0.132) (0.352)

Time effects YES YES YESObservations 23164 23164 23164Firms 4570 4570 4570Wald test( joint) 18.20(0.050) 25.34(0.008) 16.92(0.110)Wald test(time) 16.37(0.006) 12.62(0.049) 114.0(0.000)Sargan test 13.81(0.965) 39.13(0.718)Difference Sargan test 25.32(0.189)AR(1) y1.638(0.101) y1.453(0.146)AR(2) 0.405(0.686) y0.578(0.563)

Notes: Robust standard errors in parentheses;P-values for Wald test of overall lack of significance ofregressors and for joint significance of time dummies appear in( ); * denotes significance at the 90%level, ** at the 95%, *** at the 99%. Lags datedty2 andty3 for dollar debt, short-term debt, leverageand exports were included as instruments for GMM difference estimations. See Appendix A for variabledefinition.

Regarding the latter, Sargan and AR tests do not provide clear evidence against thenull hypothesis that the GMM System estimator is well specified. Nonetheless, sincein some cases the null hypothesis is marginally accepted, we present both the GMMDifference and System estimations. Two-step estimates with robust standard errorsare presented. Second-step standard errors were computed using Windmeijer(2000)


finite-sample correction. It should also be noted that we instrument for the level offoreign debt, leverage, exports and the maturity composition of debt. Also, althoughwe run regressions using all available lags as instruments, we only report those inwhich three lags were used.Table 7 presents a set of estimations for fixed capital investment. In the first three

columns, firm level variables are interacted with measures for thechanges in themacro variables; i.e. dollar debt, exports and imports are interacted with the(log)real exchange rate devaluation, whereas the degree of short-term indebtedness isinteracted with the(log) percentage change of the real interest lending rate. We failto find any significant heterogeneity in terms of the response of firms to exchangeand interest rate movements. Actually, the terms included in our fixed effectsregressions besides from the time specific effects are just marginally jointlysignificant. In the case of the dynamic GMM regressions, there is some evidence ofpersistence in the level of investment, but key interactions are once again insignifi-cant, except for the interaction of exports with the RER devaluation in the systemestimation. Also, under the GMM System regression we find a significant effect ofshort-term debt on investment.When macro variables are interacted in levels as in the final three columns of

Table 7, we are able to capture somewhat more heterogeneity. In particular, theinteraction of the real exchange rate index with exports is positive and significantin all fixed effects and GMM system specifications, whereas the interaction withimports is negative and significant in the dynamic regressions. However, theinteraction of dollar debt and the real exchange rate is not significant. Regardingnon-interacted firm level variables, the only important change refers to the level ofimports having a positive effect on investment under the GMM regressions. Asbefore, time effects are significant, and there is evidence of persistence in investmentin the GMM difference regression.Finally, in Table 8, we use profits(as percent of total assets) as the dependent

variable, while the macro variables in the interactions enter in levels. Interestingly,we obtain a significant negative effect of the dollar debt interacted with the exchangerate. This result is somewhat puzzling since if net worth affects investment levels,decreased profits should imply lower investment. There is also some evidence that12

exporting firms tend to have larger increases in their profits in times of devaluations,whereas the impact of having a higher share of imported inputs is not robust, thoughnegative in the preferred System GMM regression. A noteworthy result is that thereappears to be no persistence in firm profitability. Regarding the role of additionalfirm level variables, results are not robust to alternative specifications.In sum, this section suggests that while there is evidence of a negative balance

sheet effect on firms’ performance as measured by profitability, results for investmentare inconclusive. On the one hand, firms that are not highly indebted in dollars andexport part of their output tend to outperform dollar-indebted, non-exporting firmsaccording to our zone-performance regressions. Nonetheless, this result is not robustto alternative dummy specifications(see footnote 4) and the interaction of dollar

When total dollar debt is considered instead of non-trade debt this result still holds.12


indebtedness with the exchange rate terms is generally not significant in theinvestment regressions. Finally, the real interest rate interaction is rarely significantlynegative.13

6. Conclusions

Recently, Colombia has experienced increased macroeconomic volatility. After aperiod of significant currency appreciation associated with large capital inflows andforeign investment in the oil sector, the real exchange rate experienced a sharpdepreciation in response to capital outflows at the end of the decade. While amongColombian policy makers the favorable view of exchange rate devaluation for firminvestment is generally alleged, there is a recent and increasing concern in theliterature and in policy-making circles for the possible detrimental effects ofdevaluations in the presence of foreign indebtedness. Foreign denominated debt, itis argued, leads to a negative balance sheet effect that constraint firms’ investment.This paper contributes to this debate on empirical grounds. We examine the

determinants of debt composition, investment and profits for a representative sampleof Colombian firms in the period 1995–2001. Our results suggest that matchingtakes place to the extent that exporting firms and those in more open sectors aremore likely to have foreign indebtedness and hold higher shares of dollar debt. Firmsize is the most robust and significant determinant of dollar indebtedness, anunsurprising result as certain scale is required to access international capital markets.Although the previous results and the limited amount of dollar denominatedindebtedness in Colombia tilt the balanceagainst finding any negative balance sheeteffect of devaluations, we find evidence in favor of the latter on firms’ performanceas measured by profitability. Results for investment, however, are inconclusive.Although firms that were not highly indebted in dollars and exported part of theiroutput tended to outperform others during the devaluation period, the interaction ofdollar indebtedness with the exchange rate is generally not significant in ourinvestment regressions.In sum, there is evidence of matching in Colombian firms; yet, and contrary to

the results reported by Bleakley and Cowan dollar indebted firms not necessarilyfare better during devaluations. Answering our title’s question, for a relevant groupof firms ‘sins’ are indeed punished. We find evidence that those firms that areheavily indebted in US and whose exports are not significant exhibit negativebalance sheet effects during devaluations, both in investment and profitability.However, in our set of panel regressions these results are relatively less robust.This paper might serve as a good starting point for further research. For instance,

the difficulty in disentangling the effect of dollar indebtedness on firm performanceis probably due to the fact that there is a small degree of dollarization of liabilities.Thus, it might be worthwhile to analyze the subset of firms that engaged in dollarindebtedness vis-a-vis other similar firms that did not. Sample selection models`

This finding is consistent with studies on the determinants of investment at the macro level in13

Colombia(see Ocampo et al., 1988; Fainboim, 1990).


popularized by Heckman and others could help in this regard. Also, further researchmight concentrate on those firms that went bankrupt during this period of study, asa special case of vulnerability in our sample.14

Acknowledgments

The views expressed in the paper do not implicate the IMF or its Board ofDirectors. We are indebted to Maurice Kugler for his valuable input during theinitial stages of this project. Preliminary versions of this paper were presented atIDB seminars in Madrid and Boston College, at the IMF, Universidad de los Andes,at the NBER Inter-American Seminar on Economics and at the Eight AnnualLACEA Meeting. We received useful comments from seminar participants, partic-ularly Mark Aguiar, Adolfo Barajas, Stephen Bond, Arturo Galindo, Eduardo Levy-Yeyati, Marcela Melendez, Ugo Panizza, Roberto Rigobon, Fabio Sanchez, Fabio´ ´Schiantarelli, Hernan Vallejo and Alejandro Werner. We are also indebted to Hoyt´Bleakley for useful discussions on the empirical approach and to Campbell Harveyfor his detailed comments. The Superintendencia de Sociedades in Colombiagenerously provided us with access to its database, which was processed incollaboration with Francisco Mejia. Financial support from the IDB is gratefullyacknowledged.

Appendix A: Data sources and variable definition

In this appendix, we describe the data sources and the variables included in thepaper. As stated in Section 3, our main data source for firm-level variables is theSuperintendencia de Sociedades, which supervises both unlisted and publicly tradedfirms in Colombia. Only firms with assets of at least 20 000 legal minimum monthlywages now have to report to the Superintendencia, but the sample also includes15

smaller firms. Also, due to procedural changes there was a non-negligible decreasein the number of firms in 2001. Until 2000, all firms had to report their financialstatements. Starting in 2001, only larger(‘vigiladas’) have to do so. Though theSuperintendencia de Sociedades database covers both unlisted and publicly tradedfirms, a number of publicly traded firms only transfer their financial statements tothe ‘Superintendencia de Valores’. We relied on information from the Superinten-dencia de Valores in those cases. The ‘Superintendencia de Sociedades’ databasehas not been previously used in a systematic way for economic analysis. Thus, inattempting to stay with only high-quality information, we modified the data set inseveral ways.

Mejıa (2003) is a first approximation to this issue.14 ´The current minimum monthly wage is US$110. Hence, only firms with assets above US$2 million15

are subject to mandatory reporting.


Table A1

Variable Description

Macroeconomic variablesReal GDP growth Annual percentage change of real gross domestic product. Fig. 1,

Panel A. Source: DANE.Inflation Annual percentage change in Consumer price index. Fig. 1, Panel

B. Source: DANE.Real exchange rate Multilateral(effective) real exchange rate index, computed as a

weighted average of bilateral exchange rates with a group of20 trade partners. RER presented in Fig. 1, Panel C,corresponds to this measure. For estimation, theinflation-adjusted bilateral exchange rate with the United Statesis used instead, as foreign debt is denominated in dollars.Both measures are highly correlated(the correlationcoeffcient is 91%). Source: Banco de la Republica.

Investment Total private investment from National Accounts figures. Presentedas percent of total output in Fig. 1, Panel D. Source: DANE.

Real interest lending rate Average lending rate of financial institutions. Source:Superintendencia Bancaria.

Foreign debt acquired by Financial external debt of the private sector. Source: Banco de lathe private sector Republica.

Sectoral variablesImported input share Imputed to each firm by mapping each firms’ sector with the most

disaggregated sector available in National Accounts Data.Sectoral imported input shares were computed as the ratioof imported intermediate purchases by each sector to totalintermediate purchases, both domestic and imported. Suchdata is available from the economy-wide input–outputmatrix, with 60 sectors being the thinner disaggregationavailable. Source: DANE.

Sectoral GDP growth Annual percentage change in real output by each of 60 sectors ofeconomic activity. Source: DANE.

Firm level variablesTotal debt or leverage Total liabilities(excluding net worth) as reported in firms’ balance

sheets. Normalized by assets for estimation and desciption.Sources: Firms financial statements, Superintendencia deSociedades and Superintendencia de Valores.

Short-term debt Part of Total debt as defined above that is due in less than 1 year.Sources: Firms financial statements, Superintendencia deSociedades and Superintendencia de Valores.

Short-term financial Financial debt denominated in local currency. For estimation, it isdomestic debt scaled by total financial debt.

Sources: Firms financial annexes, Superintencia de Sociedades andfirms finacnial statements, Superintendencia de Valores

Trade debt Debt acquired by firms with suppliers. Sources: Firms financialstatements, Superintendencia de Sociedades andSuperintendencia de Valores

Total ‘Dollar’ debt Debt acquired by firms with foreign financial banks or corporations(both short-term and long-term) as well as with foreign suppliers.In the case of information coming from Superintendencia deValores, the amount of foreign debt with foreign suppliers is


Table A1 (Continued)

Variable Description

unavailable(thus, description of Total foreign debt inTable 3 is restricted to firms reporting to theSuperintendencia de Sociedades). Source: Firms financialannex, Superintendencia de Sociedades.

Non-trade or financial Debt acquired by firms with foreign financial banks or corporations.‘Dollar’ debt In the case of information coming from Superintendencia de

Valores, it refers to the value of foreign currency financial debt,both short and Long-term. Source: Firms financial annex,Superintendencia de Sociedades and firm’s financialstatements, Superintendencia de Valores.

Exports Refers to operational income generated abroad. For estimation anddescription it is normalized by total operational income. Source:Firms financial statements, Superintendencia de Sociedadesand Superintendencia de Valores.

Investment in fixed Net purchases of fixed assets from cash flow data. For estimationcapital1 and description purposes, it is normalized by the total value of

pre-existing assets. Sources: Firms financial annexes,Superintencia de Sociedades and Superintendenciade Valores.

Profits Operational profits(operational income net of operational expenses).For estimation purposes it is normalized by the total value ofpreexisting assets. Source: Firms financial statements,Superintendencia de Sociedades and Superintendenciade Valores.

Foreign participation in Share of company that is owned by foreign investors. Source: FirmsOwnership financial annex, Superintendencia de Sociedades, and firms

financial statement, Superintendencia de Valores.Log(Assets) Logarithm of real value of assets. Assets are deflated by sectoral

producer price index to get an estimate at constant prices.Sources: Firms financial statements, Superintendencia deSociedades and Superintendencia de Valores.

An ideal measure should consider the rate of change of the capital stock series at replacement cost.1

Nonetheless, we relied on information from cash flow data for several reasons. First, while our datasetincludes firms’ fixed asset valuation, in practice the series is probably correlated with firms’ tax burdenand access to credit. Correcting balance sheet data on fixed assets in order to get an estimate of thereplacement cost of capital was not feasible either as reasonable assumptions on rates of investment,depreciation and price of capital for different sectors and types of capital are unavailable for Colombia.There being no satisfactory measure of the capital stock at replacement costs from the balance sheetinformation in our data set, nor a convincing set of assumptions to correct the book value of fixed assets,the rate of investment from cash flow information provides a better measure of the evolution of capitalstock.

The following were excluded:

1. Firms that do not appear in the sample for at least 4 consecutive years. Thisresults in dropping 6700 firms, which account for roughly 44% of the number offirms in the original sample, yet many of which only appear for 1 or 2 years.

2. 65 firms that have no change at all in their level of assets or liabilities inconsecutive years.


3. 6 firms reporting unrealistically low levels of assets. In particular, firms whoseassets do not exceed $100 000 Colombian pesos(US$35 at current exchangerates), which is nearly a third of the legal minimum monthly wage.

4. 868 firms displaying inconsistent accounting information, including:● firms having liabilities that exceed the value of their assets(812 firms)● negative operational income(4 firms)● short-term assets larger than total assets(7 firms)● firms reporting negative values for their total liabilities, any of its componentsor on interests on their financial liabilities(24 firms)

● firms in which components of liabilities exceed the total(foreign, domestic,trade and financial, 21 firms)

5. For estimation purposes, we also check the sensitivity of our results to theexclusion of outliers(firms for which our measures of investment lie in the upperor lower 3% of the sample). In a number of estimations, we are also obliged todrop firms for which we do not have(or are unable to impute) denomination ofoutput and inputs in terms of currencies. 1064 firms(964 belonging to the retailsector) are dropped because of these criteria.Table A1

Appendix B: The BC set-up: summary and extensions

B.1. A summary of the BC set-upIn a two-period world, a continuum of firms holding a fractionb of their total

liabilities (normalized to 1) in dollars seek to maximize their profits in periodtq1as given by:

p e K ;b sg e ;b F K yr W K . (2)Ž . Ž . Ž . Ž .tq1 tq1, tq1, tq1 tq1 t tq1

The first term at the right hand side of Eq.(2) are earnings before interestpayments. For each firm, the capital stock at periodt, K is predetermined, as theirt

fraction of dollar debt. Functiong captures the response of profits to changes in thereal exchange rate. Firms borrow capital at an interest rate that is decreasing in networth (W):

W sp y be q 1yb (3)Ž Ž ..t t t

Devaluations reduce net worth because they increase the domestic currency valueof foreign liabilities. Firms chooseK so as to maximize Eq.(2) subject to Eq.tq1

(3) and to an exchange rate level in periodtq1 that exhibits persistence,e stq1

m(e ). The F.O.C. implicitly defines an optimal demand for capital, whose derivationt

with respect to the exchange rate leads to BC’s competitiveness and net-worth


channels on investment.16

dKtq1 w zx |su g9 e ;b m9 e qs g9 e ;b F K yb . (4)Ž . Ž . Ž . Ž .t tq1 t t t tq1y ~det

The first term in Eq.(4) is the ‘competitiveness effect’. BC consider the case inwhich g9(e ;b)G0. As long as the exchange rate exhibits persistence is such thattq1

m9(e )G0, this implies a positive competitiveness effect of devaluations on invest-t

ment. A more general case—for instance, if imported inputs are important inproduction—should consider thatg9(e ;b) might be negative. The second term intq1

Eq. (4), capturing the net worth channel is ambiguous. Ifg9(e q1;b) is negative, at

devaluation reduces earnings and net worth, leading to a decline in investment.However, for a sufficiently strong matching of liabilities and income streams(g9(e ;b)40), earnings increase with a devaluation, leading to higher investmenttq1

and compensating the rise in foreign denominated liabilities(as captured byyb).Following BC, we consider a ‘neutral’ exchange rate for which the peso value of

debt is identical for all firms, so that for all periods and allb and˜ ˜g9 e;b sgŽ .

. The differential effect on investment of a devaluation across firmsdu dst ts s0db dbwith varying levels of ‘dollar’ indebtedness is:

w zw z dg9 e ;b dg9 e ;bŽ . Ž .tq1 td dKtq1 w zx | x |s u m9 e qs F K y1 (5)x | Ž . Ž .t t t Tq1y ~db de db dby ~ y ~t

B.2. The effect of imported inputsFrom Eq.(5), it is clear that the effect on investment of a devaluation can be

either increasing or decreasing inb. BC consider the case of ‘weak’ matching of

liabilities, where . This assumption hinges on the fact thatg9(e ;b)dg9 e ;bŽ .tq1

17G0 tq1dbhas been assumed positive. What if imported inputs are so important that

g9(e ;b)-0? Then it might be that ; firms with more dollar debtdg9 e ;bŽ .tq1

F0tq1 db

Where16

w zF9 KŽ .tq1x | w xu s y G0 and s s g9 e ;b F K yb G0.Ž . Ž .t t t tq1y ~g e ;b F0 KŽ . Ž .tq1 tq1

Presumably, risk averse firms will choose a composition of debt that will match the exchange rate17

sensibilities of their balance sheet and income stream. It could also be the case that creditors chargemore to firm’s without a proper currency matching. In equilibrium, there would be a correlation betweencurrency composition of liabilities and the ‘tradability’ of output.


face a sharperdecrease in profits as the exchange rate depreciates. In this case, adevaluation unambiguously decreases investment, and investment falls more in firmswith higher dollar debt.

B.3. The effect of lower domestic interest ratesQuitting the ‘dogged’ defense of the exchange rate allows interest rates to

decrease. Thus, in addition to the idiosyncratic decrease in net worth, a macroeco-nomic channel affects the rate at which firms rent capital. Firms maximize profitsin period tq1, now given by:

Uw zx |p e ,K ;b sg e ;b F K y r W qr e ye ;b K (6)Ž . Ž . Ž . Ž . Ž .tq1 tq1 tq1 tq1 tq1 t tq1 t tq1y ~

In Eq. (6), the second term for the interest rate shows that higher devaluationexpectations imply higher domestic interest rates. When the policymaker quits the‘dogged’ defense of the exchange rate, and a devaluation does occur, expectationsfor a devaluation disappear or decrease, and interest rates fall:r9 (e ye ;b)G0.*

tq1 t

Firms with a higher fraction of peso debt will be favored more by a decrease indomestic interest rates. F.O.C for firm’s optimization implicitly defines an optimal18

demand for capital, which, in addition to BC’s competitiveness and net-worthchannels depends on a ‘macroeconomic channel’:

dKtq1 w zx |su g9 e ;b m9 e qs g9 e ;b F K yb qd m9 e y1 . (7)Ž . Ž . Ž . Ž . Ž Ž . .t tq1 t t t tq1 t ty ~det

The macro effect(third term on the RHS of Eq.(7)) has a simple interpretation.As long as 0-m9(e )-1—i.e. allowing the exchange rate to weaken today will nott

lead to a more than proportional weakening tomorrow—a devaluation has a positiveimpact on investment. The total effect on investment is still ambiguous; ultimatelyan empirical matter. The differential effect on investment of a devaluation acrossfirms can be found by implicit differentiation of Eq.(7). This total effect is thesum of the competitiveness and net worth channels as well as an additional termcapturing the ‘macroeconomic channel’:

w zw z dg9 e ;b dg9 e ;bŽ . Ž .tq1 td dKtq1 w zx | x |s u m9 e qs F K y1x | Ž . Ž .t t t tq1y ~db de db dby ~ y ~t

Udr 9 m e yeŽ Ž . .t t1q m9 e y1 . (8)Ž Ž . .tg e ;b F0 K dbŽ . Ž .tq1 tq1

18

w z1 Ux |dsy r 9 m e ye G0.Ž Ž . .t ty ~g e ;b F0 KŽ . Ž .tq1 tq1


The way the macroeconomic effect on investment varies for firms with differentproportions of dollar debt—the third term at the RHS of Eq.(8)—may be interpretedas follows. First, notice that under the assumption of moderate persistence of theexchange rate—i.e.(m9(e )y1-0—concavity of the production function impliest

that . Thus, the sign of the expression is deter-1

= m9 e y1 )0Ž Ž . .tg e ;b F0 KŽ . Ž .tq1 tq1

mined by . Recall thatr 9(m(e )ye ))0 and note that for highlydr9 m e yeŽ Ž . .t t

*t tdb

‘dollar indebted’ firms interest rates might increase less if devaluation expectationsare high, since such firms presumably depend less on domestic credit conditions.

Thus, is negative and so is the ‘macroeconomic’ channel. In otherdr9 m e yeŽ Ž . .t t

dbwords, for firms with a high proportion of dollar debt a devaluation, by decreasingdomestic interest rates, will result in a lower increase in investment. The combinedeffect of changes in earnings, in liabilities and in domestic interest rates on thedemand for capital is theoretically ambiguous.

References

Aghion, P., Bacchetta, P., Banerjee, A., 2000. Currency crises and monetary policy in an economy withcredit constraints. Eur. Econ. Rev. 45, 1121–1150.

Aghion, P., Bacchetta, P., Banerjee A., 2001. A corporate balance-sheet approach to currency crises,Study Center Gerzensee, Working Paper No. 01.05.

Aguiar, M., 2002. Devaluation, Foreign Currency Exposure and Investment: The Case of Mexico.University of Chicago, Mimeo.

Arellano, M., Bover, O., 1995. Another look at the instrumental-variable estimation of error componentsmodels. J. Econometrics 68, 29–52.

Arteta, C., 2002. Exchange rate regimes and financial dollarization: does flexibility reduce currencymismatches? Center for International and Development Economics Research. Working Paper No.C02–123.

Bekaert, G., Harvey, C., 2002. A Chronology of Important Financial, Economic and Political Events inEmerging Markets: Colombia. In http:yywww.duke.eduy;charveyyCountry_riskychronologyycolombia.htm.

Bernanke, B., Gertler, M., 1989. Agency costs, net worth, and business fluctuations. Am. Econ. Rev.79 (1), 14–31.

Bleakley, H., Cowan, K., 2002. Corporate Dollar Debt and Devaluations: Much Ado About Nothing?,Mimeographed document, MIT.

Burnside, C., Eichenbaum M., Rebelo S., 1999. Hedging and financial fragility in fixed exchange rateregimes. NBER WP No. 7143.

Bond, S., 2002. Dynamic panel data models: a guide to micro data models and practice. CemmapWorking Paper, CWP09y02.

Bond, S., Elston, J., Mairesse, J., Mulkay B., 1997. Financial factors and investment in Belgium,France, Germany and the UK: a comparison using company panel data. NBER WP No. 5900.

Caballero, R., Krishnamurthy A., 2000. Dollarization of liabilities: underinsurance and domestic financialunderdevelopment, NBER WP 7792.

Calvo, G., 1999. Fixed vs. Flexible Exchange Rates. University of Maryland, mimeo.Calvo, G., 2000. Capital Markets and the Exchange Rate. University of Maryland, mimeo.Calvo, G., Reinhart, C., 2000. Fear of floating, NBER WP 7993.


Cespedes, L., Chang, R., Velasco, A., 2000. Balance sheets and exchange rate policy, NBER WP 7840.Ćespedes, L., Chang, R., Velasco, A., 2002. IS-LM-BP in the Pampas, NBER WP 9337.Ćlassens, S., Djankov, S., Lang, L., 2000. East asian corporates: growth, financing and risks over thelast decade. Emerging Market Quarterly. Springer.

Dornbusch, R., 2001. A primer on emerging market crises, NBER WP8326.Fainboim, I., 1990. Inversion, tributacion y costo de uso del capital en Colombia: 1950–1987. Ensayos´ ´Sobre Polıtica Economica 18, 7–50.´ ´

Gelos, R.G., 2003. Foreign currency debt in emerging markets: firm-level evidence form Mexico. Econ.Lett. 78 (3), 323–327.

Harvey, C., Roper, A., 1999. The asian bet. In: Harwood, A., Litan, R.E., Pomerleano, M.(Eds.), TheCrisis in Emerging Financial Markets. Brookings Institution Press, pp. 29–115.

Krugman, P., Taylor, L., 1978. Contractionary effects of devaluation, J. Int. Econ. 8(3).Krugman, P., 1999a. Analytical Afterthoughts on the Asian Crisis. MIT, mimeo.Krugman, P., 1999b. Balance sheets, the transfer problem, and financial crises. In: Isard, P., Razin, A.,Rose, A.(Eds.), International Finance and Financial Crises. Kluwer, Dordrecht Essays in Honor ofRobert P. Flood.

Martinez, L., Werner A., 2002. The exchange rate regime and the currency composition of corporatedebt: the Mexican experience. Prepared for the NBER-Interamerican Seminar on Economics, July,2001.

Mejıa, 2003. El efecto de hojas de balance de la devaluacion en Colombia: 1998–2001. Bogota,´ ´ Ćolombia, Universidad de los Andes, Mimeographed document.

Ocampo, A., Londono, J., Villar, L., 1988. Comportamiento del Ahorro y la inversion: evolucion˜ ´ ´historica y determinantes. In: Lora, E.(Ed.), Lecturas de Macroeconomıa Colombiana. Tercer Mundo-´ ´Fedesarrollo, Bogota, pp. 13–90.´

Schneider, M., Tornell, A., 2001. Boom-Bust Cycles and the Balance Sheet Effect. UCLA, mimeo.Windmeijer, F., 2000. A finite sample correction for the variance of linear two-step GMM Estimators,Institute for Fiscal Studies Working Paper Series No. W00y19.

‘dollar’ debt in colombian firms: are sinners punished during devaluations?

Documents