A Gravity Model of Sovereign Lending: Trade, Default, and Credit

Author(s): Andrew K. Rose and Mark M. Spiegel

Source: IMF Staff Papers, Vol. 51, IMF Fourth Annual Research Conference (2004), pp. 50-63
Published by: Palgrave Macmillan Journals on behalf of the International Monetary Fund
Stable URL: http://www.jstor.org/stable/30035885
Accessed: 22/11/2010 14:19

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=pal.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

International Monetary Fund and Palgrave Macmillan Journals are collaborating with JSTOR to digitize,
preserve and extend access to IMF Staff Papers.

http://www.jstor.org
IMFStaffPapers
Vol. 51, SpecialIssue
a 2004 International MonetaryFund

A GravityModel of Sovereign Lending:

Trade,Default,and Credit
ANDREWK.ROSEand MARKM. SPIEGEL*
One reason why countriesservice their externaldebts is thefear thatdefaultmight
lead to shrinkageof internationaltrade.If so, then creditorsshould systematically
lend more to countries with which they share closer trade links. Wedevelop a sim-
ple theoreticalmodel to capturethis intuition,then test and corroboratethis idea.
[JELF15,F33]

While
the age of gunboatdiplomacyas a mechanismof creditenforcement
W has long passed, sovereign default is still an exceptional event. This styl-
ized fact indicates that while the source of a sovereign defaultpenalty is still con-
troversial, sovereigns behave as if they consider default costly. Many models of
sovereign debt in the literature(e.g., Bulow and Rogoff, 1989a, 1989b) introduce
explicit defaultpenaltiesto rationalizethis fact. These sanctionsare primarilycon-
sidered to be methods of inhibiting trade.Bulow and Rogoff (1989a) discuss the
difficulties countrieswould experiencein theirtradesubsequentto default,includ-
ing complications associated with avoiding seizure and the interruptionof short-
term tradecredit.

*Roseis B.T.RoccaJr.Professorof International TradeandEconomicAnalysisandPolicyin the Haas

Schoolof Businessat theUniversityof California,Berkeley,NBERresearchassociate,andCEPRResearch
Fellow. Spiegel is SeniorResearchAdvisor,EconomicResearchDepartment,FederalReserveBank of
SanFrancisco.WethankRobFeenstrafor a commentthathelpedinspirethispaper;RosethanksINSEAD
for hospitalitywhile partof this paperwas written.We also thankGerdHaeusler,PhillipLane, Nancy
Marion,PauloMauro,MichaelMussa,andparticipantsat the IMF'sFourthAnnualResearchConference,
andespeciallyMarkWright,for comments.A current(PDF)versionof thispaperandthe STATAdataset
used in the paperare availablevia the Internetat: http://faculty.haas.berkeley.edu/arose.

50
 A GRAVITY OFSOVEREIGN
MODEL LENDING

Nevertheless, there are a numberof reasons why one might doubt the exis-
tence of default penalties. Bulow and Rogoff (1989b) themselves admit that it is
unclearwhetherprivate creditorsenjoy the ability to induce their governmentsto
enforce claims on sovereignborrowers.KletzerandWright(2000) arguethatmost
penalties in models of sovereign lending are not "renegotiation-proof."That is,
Kletzer and Wrightargue that both parties could do better subsequentto a full or
partialsovereign default, if the creditorresists levying a destructivepenalty from
which (s)he would receive no immediate benefit. In brief, there is considerable
uncertainty concerning the viability of penalties for sovereign default. Thus,
empiricalevidence regardingsuch penalties warrantsattention.
Unfortunately,there are only a limited numberof empirical studies concern-
ing such penalties. Ozler (1993) provides evidence of positive, albeit small, pre-
mia chargedto countrieswith defaulthistories.Cline (1987) notes thatBolivia and
Peru experienced interruptionsin their flows of short-termtrade credits subse-
quentto debt renegotiation.In a recentpaper,Rose (2002) providesempiricalsup-
port for the role of tradeas a sovereign enforcementmechanism.His papershows
that sovereign Paris Club reschedulingsare followed by economically and statis-
tically significantreductionsin internationaltrade.
The evidence of Cline and Rose centers on the interruptionof international
trade as a mechanism for sovereign debt repayment.If one believes that the pri-
mary penalties for enforcing sovereign debt obligations are trade related, then
creditors originating from nations with strong bilateral trade ties with a debtor
nation should have a comparativeadvantagein lending to that nation.
In this short paper,we explore this idea. We first present a theoreticalmodel
of internationallending where a debtor optimally chooses its borrowing from
different creditors. These creditors are identical except that they are located in
countries that differ by the strength of their bilateral trade ties with the debtor.
We show that in equilibrium, the patternof borrowing favors the creditor with
higher bilateral trade volume with the debtor.We then test and corroboratethis
idea using an annual panel data set including bilateral trade and international
banking claims from 20 creditor and 149 debtor countries from 1986 through
1999. Using instrumentalvariable (and other) techniques, we find a significantly
positive effect of bilateral trade on bilateral lending patterns. That is, debtors
tend to borrow more from creditors with whom they share more international
trade ties.
While our empirical results supportthe trade sanction sovereign debt model
derived in the paper, the evidence does not necessarily refute pure "reputation-
based"models of sovereign debt in which the penaltyfor defaultis exclusion from
access to futureborrowing(e.g., Eaton and Gersovitz, 1981; Kletzer and Wright,
2000; or Wright,2002), or with mixed models where some combinationof direct
sanctionsandreputationalpenaltiesare applied,such as Kehoe andLevine (1993).1
However, it appearsthat reconciling reputation-basedmodels with the data with-
out introducingnew inconsistencies requires the introductionof some friction,

of trade
'Whilethe KehoeandLevinemodelallowsfor assetseizure,it doesnot considerinterruptions
in spot markets,whichmightbe consideredanalogousto the directtradesanctionsin ourmodelbelow.

51
 Andrew K.Rose and MarkM.Spiegel

such as superior informationsets to creditors from those countries engaged in

greaterbilateraltrade.2
Ourtheoreticalmodel is presentedin the next section. We then presentthe data
set and methodology and test the model. The paperends with a brief summary.

I. A Model of Sovereign Borrowing

withTrade-RelatedDefaultPenalties
In this section we develop a simple borrowingmodel in which a sovereign debtor
allocates its borrowingacross differentcreditornations,when defaultpenalties are
based on proportionallosses in bilateralgains from trade.
We assume that there are three countries: one borrowercountry, i, and two
creditor countries, a and b. Let r representone plus the world risk-free interest
rate.All countriesare assumed to be small and thereforetake r as given. Lending
banks in the creditor countries are risk-neutraland therefore willing to extend
unlimitedfunds at levels consistent with an expected returnequal to r.
The model has two periods. In the first period, the representativeagent in
lendercountry] (j (= a, b) extends a loan of magnitudeLijin returnfor the promise
of a fixed payment D1jin the second period. In the second period, the agent in
debtor country i makes its default decisions. If the debtor chooses to service its
countryj debt it pays Dij. If the debtordefaults, it suffers a penaltyequal to a frac-
tion 0 of its gains from bilateraltradewith countryj, where 0 < 0 < 1.
Bilateral gains from tradeare exogenous and equal to yTij,where y is a posi-
tive constant and Tijis a randomvariablereflecting total tradebetween country i
and countryj in the second period. Expectationsof Tijare unbiasedand satisfy

= EI(Tij)+ , (1)
where EI(T1j)representsthe period one expected value of T1jand Eiis an i.i.d. dis-
turbancetermwith expectedvalue 0 anda symmetricand single-peakeddistribution
on the interval EiE [E,q]. Let F(s) represent the distribution of s, that is, the
probability that Ei < ,, and f(s) represent its density. The creditor nations are
assumed to only differ in their expected trade volume with the debtor country,
with Ei(Tia) > El
(Tib).
The expected utility function of the representativeagent in countryi satisfies

=
EI(Ui) U(Cl) + PEI(Ci2), (2)

where U' > 0, U" < 0, Citrepresentsconsumptionin countryi in period t (t = 1,2),

and p representsthe debtor's discount rate. The specificationthat debtorutility is
linear in expected second-periodconsumptionis made for analytic simplicity,but
drives none of our results.

2See Wright's(2003) commentson our paperbelow. Wrightarguesthatthe assumptionof superior

informationsets heldby primarytradingpartnersmaygive thosepartnerscomparativeadvantagesin lend-
ing in purereputationmodels with the additionalassumptionof continuoustradein goods to avoid the
"excessivegross flows"problem.

52
 A GRAVITY
MODELOFSOVEREIGN
LENDING

Debtor income, Yit,is exogenous in both periods. Debtor first-period con-

sumptionsatisfies

Ci = YiI + Lia + Lib. (3)

Since no new funds are obtainedin period 2, the debtor's default decision on
debts from each creditornation is based on maximizing expected second-period
consumption. Conditionalon service on its debt obligations to countryj, debtor
second-periodconsumptionsatisfies

Ci2 = Y2 + - -
jTij Dij g(Dik,Ei), (4)

wherej a k and g (Dik, Ei) representsthe gains from tradenet of the cost of debt
service on the debtor's countryk debt obligations given that the debtormakes its
utility-maximizingdefault decision on its obligations to that country.
Similarly, conditional on default on obligations to countryj, debtor second-
period consumptionsatisfies

Ci2 = + (1 - -
O)y/1j g(Dik, i). (5)

It follows that the debtorchooses to default on countryj when Dij > OyTij.
Define as the realization of Ei that leaves the debtor indifferentbetween
default andei
repayment. satisfies
ij
Dj E
s*. _ 07 (Tij).(6)

Equilibriumin the model is defined as the pair of debt obligations (Dia, Dib)
that maximize expected debtorutility subject to both creditors'zero-profitcondi-
tions. The creditors'zero profit conditions satisfy

(7)
Dij = 1 -
F(*)

wherej=a,b.
Utility maximizationfor the debtorcan be characterizedin terms of two deci-
sions, the overall borrowing level, Li, and the allocation of debt across the two
creditors,Dia and Dib. Considerfirst the allocation decision. Given total borrow-
ing Li, maximizing expected utility subject to the creditors'zero-profitconditions
yields the first-ordercondition

Lia
Li ib,
1
I1 - F(E
Eia
FE ef,)fe
f( ib)
(8)
.

53
 Andrew K,Rose and MarkM.Spiegel

Equation (8) demonstratesthat the debtor skews its borrowing allocation

toward the nation from which the impact of a marginalincrease in borrowingon
its probabilityof defaultis smaller.Since the creditors'risk premiaare symmetric
functions of default risk, equalizing the marginalcost of the last dollar borrowed
in each countryimplies borrowingmore from the countrythatwould have a lower
probabilityof default if borrowinglevels were equal. Of course, doing so pushes
up the probabilityof default in this countryrelative to the other, and narrowsthe
difference between these probabilities.
This result implies that the optimal allocation of borrowingacross countries
mitigates the disparitiesin default risk across countries.This result fits well with
historicalexperience, as countriestypically default on all of their creditorssimul-
taneously, or not at all. It should be stressed, however, that the result is an equi-
libriumoutcome of the model, ratherthan simply assumed.3
Totally differentiating(8) with respect to Lia and E(Tia)yields

Lia +
) r[E(Ti) a> 0,
-) 8E]{[1- )]faE:a)+
aE(Tia a aLia i - F(Ea)]- f(e*)[E(Tia)+ ia2 (9)

where the denominatorcan be signed as negative by the debtor's second-order

condition.
Equation (9) yields our first result. Holding total lending constant, the share
of lending originatingin countrya is increasing in the expected volume of trade
with countrya.
We can now confront the debtor's overall borrowing decision. Maximizing
expected utility in equation(2) over the choice of Li subjectto the creditors'zero-
profit conditions and the debtor'soptimaldebt allocationrule yields the first-order
condition

U' - r+ +E = (10)
Oy{[E(Ti)
f, )

where by (6) and (7), the partialterm satisfies

de r
*i
S= > 0. (11)
dLia -
0y[l F(EFi)] - f(EiaE(Tia)+ Eia]I

Totally differentiatingwith respect to Li and Ei(Tia)yields

+ e -
aE [f'(,I F(E)] +
+Pr[E(Tia)
0, (12)
>

=-
f(Ea)2]
U" ) F()] - f(i)[E(ia)
gE(Ta) - E2

3Inthe limitingcase wherethe cij's are distributeduniformly,the equilibriumborrowingallocation

resultsin the debtordefaultingon bothcreditorsor none.

54
 A GRAVITY
MODELOF SOVEREIGN
LENDING

where the denominatorcan be signed as negative by the debtor's second-order

condition.
Our results demonstratethat an increase in the expected volume of bilateral
tradewith an individualcountryis associatedwith both an increasein overall bor-
rowing and an increasein the shareof overallborrowingoriginatingin thatcountry.
Consequently,the model predictsa positive correlationbetween expected bilateral
tradevolumes and bilaterallending.In the next section, we test this prediction.

II. Empirics
GravityMethodology
We are interestedin estimating the effect of internationaltrade on international
debt. However, internationalborrowingmay itself encourage trade;alternatively,
both borrowingand trademay be jointly driven by common factors. That is, it is
importantfor us to consider the possibility that internationalborrowingand trade
are simultaneouslydetermined.
We solve this problem using instrumentalvariables. The popular "gravity"
model of bilateralinternationaltrade provides a wealth of potential instrumental
variables.Many variablesthatare known to be importantdeterminantsof interna-
tional tradeare unlikely to be importantdeterminantsof internationallending pat-
terns. For instance, a pair of landlocked countries engages in less international
trade,while a pair of physically large countriesor those that sharea common land
bordertrademore. But internationallending patternsare unlikely to be affectedby
such features.4We use such variablesas instrumentalvariablesfor tradein a model
of bilaterallending.
Since conditions that lead two countries to be more integratedare likely to
lead to more financial activity between them, our specification for bilateralinter-
nationalborrowinglevels follows the gravity model of internationaltradeclosely:

=
ln(Ci,,) [ ln(YiY) +a21ln(YYj/PopPop)oj +"31n Dij + f4Langij
+ PsContij+ 6FTAij,+ 37Landlij+ 38Islandij
+ + +
91ln(AreaiAreaj) P,1oComColij f1iCurColijt (13)
+ 12Colonyij+ +
13ComNati P14CUijt
+ Yr . Tt, + (pln(Xij) + Eijt

where i andj denotes countries, t denotes time, and the variablesare defined as:
* Cijtdenotesthe value of real lendingfrom i to j at time t,
*
Xijtdenotesthe averagevalue of real bilateraltradebetween i andj at time t,
* Yis real GDP,
* Pop is population,
* D is the distancebetween i andj,

4If bank lendingreflects tradecredits,coefficientestimatesfrom our IV estimationmay be biased

upwards.As our estimatedeffect is large,however,it is unlikelythatcorrectionfor this bias wouldelim-
inateourresults.

55
 Andrew K.Rose and MarkM.Spiegel

* Lang is a binaryvariablethatis unityif i andj have a commonlanguage,

* Contis a binaryvariablethatis unity if i andj sharea land border,
* FTAis a binaryvariablethatis unity if i andj belong to the same regionaltrade
agreement,
* Landlis the numberof landlockedcountriesin the countrypair(0, 1, or 2).
* Island is the numberof islandnationsin the pair(0, 1, or 2),
* Area is the land mass of the country,
* ComColis a binaryvariablethatis unityif i andj were ever colonies after 1945
with the same colonizer,
* CurColis a binaryvariablethatis unity if i andj arecolonies at time t,
* Colonyis a binaryvariablethatis unity if i ever colonizedj or vice versa,
* ComNatis a binaryvariablethat is unity if i andj remainedpartof the same
nationduringthe sample(e.g., the United Kingdomand Bermuda),
* CU is a binaryvariablethatis unity if i andj use the same currencyat time t,
* T,, is a comprehensiveset of year-specificintercepts,
* p andy are vectorsof nuisancecoefficients,and
SEij representsthe myriadotherinfluenceson bilateralcredit,assumedto be well
behaved.
The coefficient of interestto us is (p,the effect of bilateraltradebetween coun-
tries i andj on commercialbank claims by creditorcountryj on debtornation i.
We estimate the model with a numberof techniquesbelow. We begin by using
ordinary least squares (OLS) with standarderrors that are robust to clustering
(since pairs of countries are likely to be highly dependentacross years). We then
use instrumentalvariables, dropping some of the regressorsfrom the right-hand
side of the equationand using them as instrumentalvariables.Finally, we employ
fixed- and random-effectspanel data estimators, with and without instrumental
variables.We use both fixed- and random-effectsestimatorsextensively below.

The Data Set

We use a subset of the panel data set of Glick and Rose (2002); the interested
readeris referredto Glick and Rose for more details.
For the regressandwe use consolidatedforeign claims of reportingbanks on
individualcountries.5These bank loans are providedby the Bank for International
Settlements (BIS) in millions of U.S. dollars for 20 creditorcountriesand almost
150 borrowing countries.6Not all of the areas covered are countries in the con-
ventional sense of the word; we use the term "country"simply for convenience.

5Ourmeasurementof cross-borderobligationsmaycontainerrorsfroma numberof sources.First,the

use of consolidateddatamay not correctlyassign the risk of banks'foreignbranches.Second,"outward
risktransfers"are sometimesused to transferrisksto residentsof othercountries,andourdataset would
notpicktheseup. Still, as theseerrorsfall in theregressandof ourmodeltheyonly maketheeffectof trade
harderto find anddo not appearto introduceany bias issues.
andarepart
6Thesedataareavailablevia theInternetat:http://www.bis.org/publ/qcsvO206/hanx9b.csv
BankingStatisticspublishedregularlyin the BIS QuarterlyReview.Fortechnicalrea-
of the International
sons we usuallyignorea few observationsfromIrelandandSpain;addingthesemakeslittledifferencein
generalto ourresults.

56
 A GRAVITY
MODELOFSOVEREIGN
LENDING

Table 1. OLSEstimatesof Effectof Tradeon Claims

Default 0.54 (0.04)

Withoutcontrols 0.75 (0.02)
Levels 0.0001 (0.00003)
Levels withoutcontrols 0.0001 (0.00003)
1990 0.51 (0.05)
1995 0.53 (0.07)
Only industrialdebtors 0.74 (0.04)

Equationestimatedis Claimsi,,= (pTradei,,,

+ SXijt + Eijt.
Robuststandarderrors(clusteredby country-pairs)recordedin parentheses.
Interceptsandyeareffects not recorded.

(The creditorcountries and debtor countries are listed in the appendix.)The data
are provided semi-annuallyfrom 1986; we average the data to annual series by
simple averaging.We convertnominalbank claims to a real series by deflatingby
the U.S. CPI (1982-1984 = 1). Almost half the claims are reportedto be zero. This
makes the log transformationpotentially importantand questionable;we investi-
gate it furtherbelow.
The most importantregressoris the level of internationaltrade.We use bilat-
eral tradeflows takenfrom the IMF'sDirectionof TradeStatisticsdataset, deflated
by the U.S. CPI.7To this we add populationand real GDP data (in constantdol-
lars).8We exploit the CIA's WorldFactbookfor a numberof country-specificvari-
ables. These include:latitudeandlongitude,land area,landlockedandislandstatus,
physically contiguousneighbors,language,colonizers, and dates of independence.
We use these to create great-circledistanceand our other controls.We obtaindata
from the WorldTradeOrganizationto create an indicatorof regional tradeagree-
ments, andwe include:EEC/EC/EU,US-IsraelFTA,NAFTA,CACM,CARICOM,
PATCRA,ANZCERTA,ASEAN, SPARTECA,and Mercosur.Finally,we add the
Glick and Rose (2002) currencyunion dummyvariable.
Descriptive statistics for the data set are tabulatedin the appendix.

Results
We begin our investigation by estimating equation (12) with OLS. Our results
appearin Table 1.

7Bilateraltradeon FOB exportsandCIFimportsis recordedin U.S. dollars;we deflatetradeby the

U.S. CPI.We createan averagevalueof bilateraltradebetweena pairof countriesby averagingall of the
fourpossiblemeasurespotentiallyavailable.
8Whereverpossible,we use WorldDevelopmentIndicators(takenfromthe WorldBank'sWDI2000
CD-ROM)data.Whenthe dataareunavailablefromthe WorldBank,we fill in missingobservationswith
comparablesfromthe PennWorldTableMark5.6, and(whenall else fails) fromthe IMF'sInternational
FinancialStatistics.The serieshave been checkedandcorrectedfor errors.

57
 Andrew K.Rose and MarkM.Spiegel

Our default estimates include the entire set of regressors (i.e., all 14 coeffi-
cients are estimated, as well as the set of time-specific intercepts).In this specifi-
cation, the estimate of the all-important(p coefficient is 0.54, with a robust
standarderrorof 0.04. This elasticity is not only consistent with our theory,but is
highly significant. With a t-statistic of over 15, the coefficient is different from
zero at any reasonablelevel of statisticalsignificance. The effect is also econom-
ically significant; an increase in tradeof 1 percent is associated with an increase
in bilateral lending of over 0.5 percent, all other things being equal. Of course,
since there are capital flows above and beyond the bank lending that we consider
(through,for example, stock and bond markets, as well as foreign direct invest-
ment), even this considerable elasticity should probably be considered a lower
bound.
The rest of the table provides a series of robustnesschecks. For instance, the
second row reports(p if the other controls are droppedfrom the equation(i.e., we
set 3 = y = 0); in this case, the effect is even more significant. Since many of the
creditorcountrieshave not extended loans to some of the debtorcountries,many
observations of the dependent variable are zero and are thus dropped from the
equation estimated in naturallogarithms.Therefore,the thirdand fourth rows of
the table reportcomparableestimates of (p when both trade and bank claims are
included in untransformedlevels. Yet (p remains statisticallysignificant when the
key relationshipis estimatedin levels.9
The fifth and sixth rows of the table move away from panel data analysis to
cover only cross sections for two years in the middle of the sample, 1990 and
1995. However, the results are essentially unchangedfrom the default specifica-
tion. The seventh and final row includes only observations between industrial
countries(i.e., those with IFS countrycodes less than 200). If anything,the results
become mysteriously larger;they certainlyremainpositive and highly significant
in both the economic and statisticalsenses.'0
To summarize,the effect of internationaltradeon bankclaims seems positive,
significant, and robust in simple OLS estimation. The question is whether this
result stands up to greatereconometric scrutiny.

III.Results
We now proceed to instrumentalvariables estimation.We use five instrumental
variables for (the log of) trade: (the log of) distance between the countries;the
land border dummy; the numberof landlocked countries;the number of island
nations; and the log of the productof the countries' area. We accordingly set the
appropriate0 coefficients to zero (i.e., drop them from the equation, leaving the
remainingvariablesas controls). The estimates are tabulatedin Table 2a.
Despite the use of instrumentalvariables that are both plausibly exogenous
and correlatedwith trade, the key results do not change with IV estimation.The
9Box-Coxtestsimplythatthe naturallogarithmictransformation
is quitereasonable,andthatthelevel
transformationis rejectedin favorof the log transform.
10Though if we includeonly developingcountryborrowers,our estimateremainssignificantat 0.53
(standarderrorof 0.04).

58
 A GRAVITY
MODELOFSOVEREIGN
LENDING

Table 2a. IVEstimatesof Effectof Tradeon Claims,

Geographic Instruments

Default 0.41 (0.07)

Without controls 0.50 (0.04)
Levels 0.00006 (0.00001)
Levels without controls 0.00007 (0.00002)
1990 0.52 (0.10)
1995 0.40 (0.10)
Only industrial debtors 1.03 (0.07)

Equation estimated is Claimnsi,jr= tpTradei.ijt

+ PWi,jt +
Eij.t.
Robust standarderrors (clustered by country-pairs) recorded in parentheses.
Intercepts and year effects not recorded.
Instrumentalvariables for trade are: distance: land border; number landlocked; number of island
nations; log of area.

Table 2b. IVEstimatesof Effectof Tradeon Claims,

Excludable Instruments
(P
Default 0.80 (0.40)
Without controls 0.83 (0.07)
Levels 0.00004 (0.00001)
Levels without controls 0.00005 (0.00001)
1990 0.59 (0.37)
1995 1.13 (0.49)
Only industrial debtors 0.79 (0.29)

Equation estimated is Clairnsij, = p(Trade;ij,r

+ PZi,j,t + Eij.
Robust standarderrors (clustered by country-pairs) recorded in parentheses.
Intercepts and year effects not recorded.
Instrumental variables for trade are: common language; regional trade agreement; same nation.

default estimate is somewhat smaller, averagingperhaps0.4. But it remainseco-

nomically and statisticallysignificant;it is also robust to a numberof economet-
ric perturbations."
Table 2b reports sensitivity analysis with respect to the set of instrumental
variables.Insteadof the five geographicvariables,we use threewhose coefficients
are usually insignificantin OLS estimatesof equation(14): the common language
dummy; the regional trade agreement dummy; and the same nation dummy.
Again, the estimates of (pseem economically and statisticallysignificant.12
Again, if we includeonly developingcountryborrowers,our estimateremainssignificantat 0.38
(standarderrorof 0.08).
variables,ourkey resultof a positiveeffect
12Ifwe use lags (e.g., of the GDPterms)as instrumental
of tradeon borrowingis not changed.

59
 Andrew K.Rose and MarkM.Spiegel

Table 3. IVEstimatesof Effectof Tradeon Claims,

Controllingfor TotalClaims/Debt
Control TotalClaims TotalDebt
Default 0.40 (0.07) 0.42 (0.07)
Withoutcontrols 0.42 (0.04) 0.27 (0.04)
Levels 0.00005 (0.000004) 0.00006 (0.00002)
Levels withoutcontrols 0.00005 (0.000006) 0.00006 (0.00002)
1990 0.47 (0.10) 0.56 (0.09)
1995 0.37 (0.10) 0.42 (0.10)
Only industrialdebtors 0.48 (0.23) 1.10(0.20)
OLS 0.29 (0.03) 0.39 (0.02)
Equation estimated is Clainmsij.= (pTradeij,t+ pWij,t+
Robuststandarderrors(clusteredby country-pairs) i.j,t. in
recorded parentheses.
and
Intercepts year effects not recorded.
Instrumental variablesfor tradeare:distance;landborder;numberlandlocked;numberof island
nations;log of area.

The middle column of Table 3 adds a control for the (log of the) total credit
extended by the creditor country, as suggested by our theoretical analysis; the
right-handcolumn controlsfor the (log of) total debt incurredby the debtorcoun-
try.Again, the results remaineconomically and statisticallysignificant.
Finally,Table4 reportsresultswhen panel estimatorsare used insteadof more
traditionalregressions.The middle columns reportOLS fixed- and random-effects
estimates of (p for a variety of different specifications. The former takes into
account all country-pairfactors that influence tradewhethermeasuredor not, and
is thus an exceptionally good robustness check. The right-handcolumn reports
instrumentalvariablesestimatesusing a randomeffects estimator(the fixed-effect
estimator is infeasible since the geographic variables are time-invariant).Yet
despite all the econometricfirepower,the estimate of (premains significant;it has
a t-statisticof almost 9 and an economically large effect.13
We conclude that our hypothesis that bank credit is extended across interna-
tional bordersalong the lines of internationaltradeis corroborated.

IV.Summary
It is plausible to believe that countriesservice their foreign debts at least in partto
avoid the reducedtradethattypically follows internationaldefault.If so, sovereign
borrowerswill enjoy superiorcredit terms from creditorcountriesfor which this

13Lending may be motivatedby servicingFDI, ratherthanthe sovereignriskissues consideredin the

theoryabove.To test this, we adda controlin the formof the naturallogarithmof FDI sourcedfromthe
creditorcountry.We obtainedthe bilateralFDI datafrom the OECD'sInternationalDirect Investments
Yearbook1980-2000. This dataset is annualandunavailablefor manycountriesin our sample,contain-
ing only some 2,600 observations.Whenwe add this controlto our defaultIV regression(in logs, with
controls)its coefficientis indeedpositive and significant.Still, the log of traderetainsan economically
andstatisticallysignificantcoefficientof 0.62 (with a robuststandarderrorof 0.11).

60
 A GRAVITY LENDING
MODELOF SOVEREIGN

Table 4. IVEstimatesof Effectof Trade Levelon Claims, Panel Estimators

Estimator OLS,RE OLS,FE IV,RE
Default 0.31 (0.01) 0.19 (0.02) 0.52 (0.06)
Withoutcontrols 0.38 (0.01) 0.19 (0.01) 0.52 (0.03)
Levels 0.00003 (0.000001) 0.00002 (0.000001) 0.00006 (0.00001)
Levels withoutcontrols 0.00003 (0.000001) 0.00002 (0.000001) 0.00007 (0.000003)
Only industrialdebtors 0.46 (0.06) 0.28 (0.07) 0.96 (0.19)
Equation estimated is Claiinsij,, = (pTradei,, + 3Wi,.t+ Eij.
Robuststandarderrors(clusteredby country-pairs) recordedin parentheses.
Interceptsandyeareffects not recorded.
Instrumental variablesfor tradeare:distance;landborder;numberlandlocked;numberof island
nations;log of area.

penalty is disproportionatelyhigh. In this paperwe have provideda simple theo-

retical model that formalizes this intuition.We have also empiricallyinvestigated
and confirmed the hypothesis that internationaltrade patternsdetermine lending
patterns.
In futurework it would be interestingto extend this analysis to otherforms of
internationallending, above and beyond bank loans. We think this is a good place
to pass the torch to others.

APPENDIX
DescriptiveStatistics

Sample Mean Std. Dev. Min Max

Claims 31,787 561 3529 0 146061
Log realclaims 19,769 3.69 2.53 -1.20 11.5
Log realtrade 28,809 11.6 2.81 -0.55 20.3
Controls:Log distance 28,809 8.32 0.59 5.37 9.41
Log realGDP 25,126 49.6 2.50 42.3 58.0
Log realGDPpercapita 25,102 17.3 1.07 14.1 21.1
Commonlandborder 28,809 0.003 0.053 0 1
Commonlanguage 28,809 0.173 0.379 0 1
Log areas 28,809 23.8 3.25 12.20 32.3
Numberlandlocked 28,809 0.286 0.496 0 2
Numberof islands 28,809 0.301 0.489 0 2
Regionaltradeagreement 31,787 0.009 0.094 0 1
Samenation 28,809 0.003 0.054 0 1
Colonialhistory 28,809 0.051 0.221 0 1
Currentcolony 28,809 0.003 0.057 0 1
Currencyunion 28,809 0.003 0.055 0 1

61
 AndrewK.Roseand MarkM.Spiegel

Creditor Countries with Claims Reported

United States United Kingdom Austria Belgium

Denmark France Germany Italy
Netherlands Switzerland Sweden Canada
Japan Finland Greece Iceland
Ireland Malta Portugal Spain

Debtor Countries with Claims Reported

Afghanistan, Islamic State of Ecuador Madagascar

Albania Egypt Malawi
Algeria El Salvador Malaysia
Angola Equatorial Guinea Maldives
Argentina Ethiopia Mali
Australia Falkland Islands Malta
Bahamas, The Fiji Mauritania
Bahrain French Polynesia Mauritius
Bangladesh Gabon Mexico
Barbados Gambia, The Mongolia
Belize Ghana Morocco
Benin Gibraltar Mozambique
Bermuda Greece Myanmar
Bhutan Grenada Namibia
Bolivia Guatemala Nauru
Botswana Guinea Nepal
Brazil Guinea-Bissau Netherlands Antilles
Brunei Darussalam Guyana New Caledonia
Bulgaria Haiti New Zealand
Burkina Faso Honduras Nicaragua
Burundi Hong Kong SAR Niger
Cambodia Hungary Nigeria
Cameroon Iceland Oman
Cape Verde India Pakistan
Cayman Islands Indonesia Panama
Central African Republic Iran, Islamic Republic of Papua New Guinea
Chad Iraq Paraguay
Chile Israel Peru
China Jamaica Philippines
Colombia Jordan Poland
Comoros Kenya Portugal
Congo, Dem. Rep. of Kiribati Qatar
Congo, Rep. of Korea, Rep. of Romania
Costa Rica Kuwait Rwanda
C6te d'Ivoire Laos Sao Tome and Principe
Cuba Lebanon Saudi Arabia
Cyprus Lesotho Senegal
Djibouti Liberia Seychelles
Dominica Libya Sierra Leone
Dominican Republic Macau SAR Singapore

62
 A GRAVITY
MODEL LENDING
OFSOVEREIGN

Solomon Islands Syrian Arab Republic Uruguay

Somalia Tanzania Vanuatu
South Africa Thailand Venezuela, Reptiblica Bolivariana de
Sri Lanka Togo Vietnam
St. Lucia Tonga Western Samoa
St. Vincent Trinidad and Tobago Yemen, Republic of
St. Helena Tunisia Yugoslavia, Federal Republic of
Sudan Turkey Zambia
Surinam Uganda Zimbabwe
Swaziland United Arab Emirates

REFERENCES
Bulow, Jeremy,and KennethRogoff, 1989a, "A ConstantRecontractingModel of Sovereign
Debt," Journal of Political Economy, Vol. 97, No. 1, pp. 155-78.
, 1989b,"SovereignDebt:Is to Forgiveto Forget?"AmericanEconomicReview,Vol.79,
No. 1, pp. 43-50.
Cline, William R., 1987, Mobilizing Bank Lending to Debtor Countries (Washington: Institute
for International
Economics).
Eaton,Jonathan,andMarkGersovitz,1981, "Debtwith PotentialRepudiation:Theoreticaland
EmpiricalAnalysis,"Reviewof EconomicStudies,Vol. 48 (April),pp. 289-309.
Glick,Reuven,andAndrewK. Rose, 2002, "Doesa CurrencyUnionAffectTrade?"European
EconomicReview,Vol. 46, No. 6, pp. 1125-51.
Asset Markets,"Reviewof
Kehoe,TimothyJ., andDavid K. Levine, 1993, "Debt-Constrained
Economic Studies, Vol. 60, pp. 865-88.
Kletzer,KennethM., and BrianD. Wright,2000, "SovereignDebt as Intertemporal
Barter,"
American Economic Review, Vol. 90, No. 3, pp. 621-39.
Ozler,Sule, 1993, "HaveCommercialBanksIgnoredHistory?"AmericanEconomicReview,
Vol. 83, No. 3, pp. 608-20.
Rose, Andrew K., 2002, "One Reason Countries Pay Their Debts: Renegotiationand
InternationalTrade,"NBER WorkingPaper 8853 (Cambridge,Massachusetts:National
Bureauof EconomicResearch).
Wright, Mark L.J., 2002, "Reputationsand Sovereign Debt" (unpublished;Palo Alto,
California:StanfordUniversity).
"New EmpiricalResults on Default:A Discussion of 'A GravityModel of
, 2004,
SovereignLending:Trade,Default,andCredit',"IMFStaffPapers,Vol. 51 (specialissue),
pp. 64-74.

63