2015 April

Download as pdf or txt
Download as pdf or txt
You are on page 1of 37

Optimal Trade Costs after Sovereign Defaults


Chenyue Hu and Fudong Zhang

University of Michigan

Abstract
This paper offers new theoretical and empirical insights into the effect of
sovereign defaults on trade. Empirical evidence from the changes in trade
shares after debt renegotiations as well as Aid-for-trade statistics indicates
that sovereign debt renegotiation is not associated with trade sanctions but
with trade incentives offered by creditor countries to debtor countries. Using
a two-country DSGE model with incomplete financial markets, we are able to
explain why trade sanctions are not observed. Our model departs from the
existing literature on sovereign defaults by building on the strategic interaction
between debtors and creditors. We reason that creditors lower trade costs with
debtors in hopes of collecting the remaining debt during debt renegotiations.
The adjustment in turn affects debtors’ default decisions. The model departs
from the existing literature on sovereign defaults by building on the strategic
interaction between debtors and creditors. We solve the model numerically
to determine the optimal trade costs given different combinations of debt and
income levels.


Email:cyhu@umich.edu; fudongzh@umich.edu. We would like to thank Kathryn Dominguez,
Linda Tesar, Javier Cravino, Alan Deardorff and participants at the Umich international/macro
lunch seminar for comments and supports. All errors are ours.

1
1 Introduction

The danger of default exists with every financial loan, and sovereign debt is no
exception. Holders of sovereign debt face additional uncertainty stemming from the
lack of supernational legal entities. The recent debt crises in Europe and Latin
America have demonstrated the need to study both creditors’ and debtors’ incen-
tives and decisions in the initiation, negotiation and settlement process of sovereign
debt contracts. This paper aims to contribute to the discussion by focusing on a
novel mechanism that has been overlooked in previous work.
Globalization since the second half the twentieth century has featured both trade
liberalization and financial mobility across borders. The two channels should not be
studied in isolation, as both are important sources of individual countries’ economic
development as well as world risk sharing. As Tomz and Wright (2013) point out,
theoretical models are missing while empirical evidence is ambiguous over how trade
and sovereign default interact. Our paper addresses this gap in the literature by
providing new empirical and theoretical results that bring together the trade and
borrowing channels to explain sovereign default settlement.
Trade, in previous literature on sovereign default, has played a trivial if any role.
For instance, Bulow and Rogoff (1989) argue that default may lead to a decline in in-
ternational trade, which is interpreted as a constant output loss in their model. Their
approach is followed in the majority of sovereign default papers including Aguiar and
Gopinath (2006), Yue (2010) , Bai and Zhang (2010), to name just a few. Tomz and
Wright (2013) summarize three reasons why trade could suffer after default happens:
(1) creditors’ trade restrictions as a means of punishment (a.k.a. trade sanction), (2)

2
the collapse of trade credit, and (3) creditors’ asset seizures. None of these reasons
can be captured by direct output loss, let alone the strategic behaviors that arise
from these features. Instead, our paper will focus on how trade costs may change
before and after sovereign defaults.
Rose (2005) explains empirically the cost of trade after sovereign default. Using
government-to-government debt default information from the Paris Club, he finds
that debt renegotiations have significantly negative effects on contemporaneous and
lagged trade volume in a gravity regression. We find his results inspiring and intrigu-
ing but not fully explored. Trade volume will naturally fall with the deterioration of
economic terms, which may not be fully picked up by the gravity variables. It is the
relative share instead of the absolute value of trade that measures the existence and
severity of punishment in the bilateral borrowing relationship. We replicate Rose’s
analysis on an expanded dataset that includes fifteen additional years. Similar to
Rose (2005) we find that trade volume falls, but we also find that trade share increas-
es significantly (by around 5%) after debt renegotiation happens. This is a surprising
result that runs contrary to the traditional trade sanction arguments.
Novy (2013) argues that trade share can be used to infer time-varying bilateral
trade costs directly from the model’s gravity equation without imposing arbitrary
trade cost functions. Based on this argument, we hypothesize trade costs change as
a creditor’s reaction to debt renegotiation. As there lacks comprehensive and consis-
tent data on direct measurement of trade costs, we resort to OECD’s data on aid for
trade and find there is noticeable increase in trade-related assistance from creditors
when debt renegotiation happens. This is complementary evidence for lower trade

3
costs after defaults.
Our findings lead us to rethink creditors’ incentives: why would creditors be will-
ing to lower their trade costs with defaulters? In practice, before a default reaches
its final resolution, there is a renegotiation stage where the creditor and the debtor
could agree on debt settlement based on the current income of the debtor and the
size of the debt. Our hypothesis is that in the renegotiation stage, it is sometimes
optimal for the creditor to lower trade costs so that the debtor is more likely to
service the debt.
We build a dynamic stochastic general equilibrium model to develop our hypoth-
esis. Our model differs from a standard sovereign default model in the following
ways. First, it is a two-country model instead of a small open economy. Addition-
ally, because we are interested in whether the model’s prediction of trade shares
can match our empirical findings, we will study a creditor-debtor two-country model
integrated with a world market. Second, our model includes a trade component.
The consumption bundle in a country consists of domestic goods, financial partners’
(whether it be creditor or debtor) goods, and the goods from the world market, with
an elasticity of substitution among them. Third, creditors are risk averse. Credi-
tors in most sovereign default models are risk-neutral and perfectly competitive for
tractability reasons, so that bond prices are directly linked to the world interest rate
once default probability is computed. This assumption will be relaxed in our model
as we assume a concave utility function. We compute a market-clearing bond price
under the assumption of constant relative risk aversion (CRRA).
In our story, the amount creditors hope to collect from debtors induces the ad-

4
justment of bilateral trade costs. At the same time, the change in trade costs affects
debtors’ probability to service the debt. At the end of the day, default probabilities,
bond prices and optimal trade costs are all endogenously derived as the solution to
the general equilibrium model. Trade and debt channels are more correlated and
interactive in our model than in any previous work.
Our contribution is three-fold. First, we identify an interesting but overlooked
phenomenon through our empirical analysis, which calls the widely-accepted trade
sanction argument into question. Second, we propose a new mechanism which links
bilateral trade and bilateral borrowing. Third, we develop computation techniques
that allow us to numerically solve a sovereign default model with more realistic fea-
tures, such as risk-averse creditors.
In our new approach, we have maintained several important features from pre-
vious work. Eaton and Gersovitz (1981) propose financial autarky as a means to
support debtors’ incentive to repay the debt. In our model, defaulters are also de-
nied access to new loans. In terms of empirical analysis, our paper is in line with
Martinez and Sandleris (2011) who find that debtors’ bilateral trade with creditor
countries does not fall more than trade with other countries. On the computa-
tion side, we follow Hatchondo et al. (2010)’s recommendation and use cubic spline
interpolation rather than discrete state space technique to approximate the value
functions to reduce computational burdens. Our paper is also related to the recent
work of Gu (2015) but with a different focus. She introduces vertical integration in
production between a creditor and a debtor to examine the dynamics of terms of
trade and trade volume, while our work aims to provide an answer to the optimal

5
trade costs a creditor imposes on a debtor after debt renegotiations take place.
The remainder of the paper proceeds as follows: Section 2 presents the empirical
findings. Section 3 describes the model as well as the properties of the recursive
equilibrium. Section 4 elaborates on the algorithm, parameterizations and numerical
results. Section 5 concludes.

2 Empirical Analysis

In this section, we present our findings about the effect of sovereign defaults on
trade. We are interested in the dynamics of trade shares after debt renegotiations.
Trade share is a more accurate measure of trade sanctions or benefits than trade
volume: if there were trade sanctions, creditors would disproportionably depress
their trade with debtors. Hence, trade sanctions indicate lower creditor-debtor trade
shares after debt renegotiations.
Following Rose (2005), we track sovereign default episodes since 1956 from the
Paris Club. It is an informal group of financial officials from 19 of the world’s biggest
economies, which provides financial services such as war funding, debt restructur-
ing, debt relief and debt cancellation to indebted countries and their creditors. We
recognize that there are diverse forms of international lending besides the debt ex-
changes between governments1 , yet the Paris Club has remained a central player in
the resolution of developing and emerging countries’ debt problems. We can track
1
Besides government to government bilateral debt under the Paris Club umbrella, debtor coun-
tries also issue commercial bank debt under the London Club, or issue bond debt. For detailed
elaboration and comparison of different forms of sovereign debts, see Das et al. (2012).

6
Figure 1: The share of debtors’ goods in creditors’ and non-creditors’ imports

the date, list of creditors, amount of debt and terms of treatment. Another reason
that we only consider government-to-government bilateral agreements is that private
lending does not have as direct impact as public lending on trade flows. After all,
governments are the major players to design trade policies and sign trade treaties.
Before we move to regression analysis, it is intuitive to show graphically changes
in trade shares around sovereign default periods. In Figure 1, we plot the share of
debtors’ goods in creditors’ and non-creditors’ imports, averaged across all the de-
fault episodes. Trade share reaches its trough in the default year (denoted as zero
on the x-axis) when debtors’ economies experience the hardest hit. However, it is
noticeable that debtors’ trade with creditors is able to recover sooner and better than
that with non-creditors: while the trade share in non-creditors’ imports is lower than
the level before defaults, the trade share in creditors’ imports bounces back and even
higher than the level before defaults.

7
I herein use a panel regression to quantify the effect of sovereign defaults on trade.
The first step we take is to replicate Rose’s (2005) results with fifteen more years of
data. The original gravity model in Rose (2005) is

ln(Xijt ) = β0 + β1 ln(Yi Yj )t + β2 ln(Yi Yj /P opi P opj )t + β3 lnDij + β4 Langij

+β5 Contij + β6 F T Aijt + β7 Landlij + β8 Islandij + β9 ln(Areai Areaj )

+β10 ComColij + β11 CurColijt + β12 Colonyij + β13 ComN atij + β14 CUijt
X X
+β15,0 IM Fijt + β15,k IM Fijt−k + φREN EGijt + φm REN EGijt−m + ijt
k m

Xijt is the trade flow between country i and j at time t. Y denotes real GDP
and P op denotes population, so that Y /P op is income per capita. Dij represents
the distance between i and j and Area represents a country’s land mass. Binary
variables include Lang (common language), Cont (common border), F T A (regional
trade agreement), ComCol (common colonizer after 1945), CurCol (colonies at time
t), ComN at (part of the same nation at time t) and CU (same currency). Landl
and Island are the numbers of landlocked and island countries in the country pair,
which take the value of 0,1, or 2. REN EGijt is a dummy variable which is unity if i
and j renegotiate debt at time t and zero otherwise. IM Fijt is one(two) if one (both)
of i or(and) j begin an IMF program at t and zero otherwise. Lagged REN EG and
lagged IM F are also listed as explanatory variables, considering the change in trade
flow is a gradual and persistent process.
Our first goal is to extend Rose’s data by 15 years to reflect the recent trends in

8
sovereign defaults. In collecting the data, we do our best to choose similar, if not the
same data sources as Rose, in order to make the results consistent and comparable.
We get the trade data from the ‘Direction of Trade Statistics (DOTS)’ dataset by
the International Monetary Fund (IMF). The values are in current US dollars. We
deflate them by the US CPI (82-84=100) from BLS to get the real value. GDP
and population data are taken from World Bank’s ‘World Development Indicator’.
In the case of missing values, we turn to Penn World Table. Values of other com-
mon gravity variables including distance, contiguity, language and colonization are
available in the CEPII dataset2 . The information about regional trade agreements is
updated with the records from the World Trade Organization. Lastly, we get the list
for the IMF programs from Axel Dreher. See Table 5 in the Appendix for detailed
categories.
Our results about trade volumes are similar to Rose’s, in both the sign and
magnitude of the estimated coefficients. Table 1 lists the estimates in fixed-effect
and random-effect models with contemporaneous and fifteen lags of REN EG, the
dummy variable of debt renegotiation. In all the cases (i.e. bilateral trade, trade
from debtor to creditor (denoted as country1to2), and trade from creditor to debtor
(country2to1)), the linearly-combined coefficients of contemporaneous and lagged
debt renegotiation — 15
P
t=1 REN EG — are all negative, whether we employ a fixed-

effect or random-effect model. This result indicates that bilateral trade volumes
between a creditor and a debtor decrease after a sovereign default.

2
It is a square gravity dataset for all pairs of countries, downloadable at http://econ.sciences-
po.fr/thierry-mayer

9
Table 1: Linearly Combined Contemporaneous and Lagged Effects of Debt
Renegotiation on Trade Volumes

Coefficient Std. Err. t [95 percent Conf. Interval]


bilateral FE -1.098 0.118 -9.300 -1.329 -0.867
bilateral RE -1.608 0.119 -13.490 -1.842 -1.375
trade 1to2 FE -1.416 0.150 -9.460 -1.710 -1.123
trade 1to2 RE -2.177 0.151 -14.410 -2.473 -1.881
trade 2to1 FE -1.426 0.144 -9.930 -1.708 -1.145
trade 2to1 RE -1.891 0.144 -13.090 -2.174 -1.608

After replicating Rose’s original results, we go a step further to analyze trade


shares. We believe it is the relative but not the absolute change in trade that reflects
the existence and severity of trade sanctions after defaults take place. To this end, we
create two variables: Impw1to2 is the share of creditors’ goods in debtors’ imports,
and Impw2to1 is the share of debtors’ goods in creditors’ imports. Then we replace
trade volumes with these two measures as the dependent variable in the regression.
In addition, we add exchange rates as an independent variable as currency depre-
ciation may bias the results. For instance, the collapse of South America during the
1970’s debt crisis affected the currency values of nearly all the countries in the whole
region. The covariance between exchange rates across Latino countries was different
from that between Latino countries and developed countries, which mattered for the
change in trade shares. To correct this bias, we collect data on exchange rates from
the International Financial Statistics (IFS). The original data are in the units of
currency per US dollars, which can be converted to obtain bilateral exchange rates
between two arbitrary currencies.

10
Table 2: Linearly Combined Effects of Debt Renegotiation on Trade Share

Coefficient Std. Err. t p > |t| [95 percent Conf. Interval]


share 2in1 0.0590 0.0048 12.2000 0.0000 0.0495 0.0684
share 1in2 0.0628 0.0047 13.4600 0.0000 0.0537 0.0720

Table 2 lists the regression results in the fixed-effect model. We present the coeffi-
cient and the standard error of 15
P
t=1 REN EG, the linear combination of coefficients

on paris,paris1-15. From the table, we find that debt renegotiations have significant-
ly positive effect on trade shares; a sovereign default episode is associated with a 5%
increase in the share of debtors’ goods in creditors’ imports. This number is impres-
sive, given the number of trade partners available nowadays in the integrated world
market. We believe this increase in trade shares indicates that sovereign defaults do
not lead to trade sanctions, but are instead associated with trade benefits.
Trade shares have been used by trade economists to uncover trade costs. This
approach is developed by Head and Ries (2001) and extended by Novy (2013), who
derives a micro-founded measure of bilateral trade costs that indirectly infers trade
frictions from observable trade data. The measure turns out to be consistent with
a broad range of leading trade theories including Ricardian and heterogeneous-firm
models. The bilateral comprehensive trade costs are calculated as

Xii Xjj 2(σ−1)


1
τij = ( ) −1
Xij Xji

where Xij and Xji denote bilateral trade, while Xii and Xjj denote domestic expen-
diture. τij represents the geometric average of trade costs between countries i and j
relative to domestic trade costs within each country. Its value reflects the additional

11
costs that trading goods between i and j involves, as compared to when the two
countries trade these goods within their borders. It covers tariffs, transportation
costs, and other unobservable trade barriers. It is straightforward to see an increase
in trade shares is equivalent to a decrease in trade costs. Thus, based on the in-
crease in trade shares, we hypothesize that bilateral trade costs between creditors
and debtors decrease after defaults happen.
While our argument will be stronger if we can support our hypothesis with a
consistent and continuous data set of visible and invisible trade costs, such data set
is rare.3 Alternatively, we turn to OECD’s Aid-for-trade dataset to see whether the
efforts to boost bilateral trade are strengthened when debt renegotiations happen.
We restrict our attention to the categories of aid that are directly related to trade
policy adjustment (See Table 7 for details). Figure 5 plots the change in creditors’
trade-related aid to debtors around the following three default episodes: Honduras
(2004), Congo (2008) and Burundi (2009). In the years of sovereign defaults, credi-
tors double or triple their trade-related aid to help defaulters out. Instead of trade
sanctions, they offer generous trade benefits. These case studies serve as indirect
evidence for our hypothesis that creditors lower trade costs with defaulters.
To sum up the empirical section, sovereign renegotiation is associated with in-
creased bilateral trade shares between debtors and creditors. This empirical result,
in line with Martinez and Sandleris (2011), contradicts the prediction of the trade
sanction theory. Based on Novy (2013)’s trade costs theory and Aid-for-trade data,

3
Bilateral tariff and non-tariff data from the World Bank’s WITS are discontinuous and available
only for the past decade. Trade costs in our paper are broader in definition, so it is hard to find
direct comprehensive evidence.

12
we believe bilateral trade costs decrease after debt renegotiations.

3 Model

In the empirical section, we challenge the conventional trade sanction theory of


sovereign default. Our explanation for this interesting observation is that a creditor
is willing to compromise in the trade channel in order to minimize its loss from the
financial side. In other words, when the creditor finds the debtor on the brink of
defaulting, it is willing to lower trade costs to boost the debtor’s exports such that
the debtor is more likely to service the debt. The reduced trade costs will in turn
determine a debtor’s willingness to repay. Our model features this strategic interac-
tion between the two parties in the Markov perfect equilibrium.

3.1 Model Environment

In the model, there is a creditor, a debtor and the rest of the world (ROW).
Although commonly used for sovereign default problems, a model with a small open
economy is not able to capture the strategic interaction between countries. Mean-
while, a standard two-country model is not helpful in studying the trade shares after
sovereign defaults. To this end, we will build a creditor-debtor two-country model
integrated with a world market (or ROW).
The creditor and the debtor are endowment economies with goods specific to

13
country i = c, d (c denotes the creditor and d denotes the debtor). For simplicity,
we assume the income of the creditor Ā is constant over time and large enough for
the country to always be the lender. Meanwhile, the income of the debtor follows an
AR(1) process:
yt = ρy yt−1 + (1 − ρy )ȳ + t

with its long-run mean ȳ and innovation t ∼ N (0, σ 2 ).


Other than the two countries, there is a world market that both countries can
interact with. Specifically, this market consists of two parts — financial and goods
markets. In the world financial market, there is a risk-free asset called world bond
with rate r. Meanwhile, the world goods market supplies one kind of consumption
good. For simplicity, we assume a country can trade one-for-one domestic goods for
goods in the world market: ciw = cwi .
Country i’s objective is to maximize its expected lifetime utility:

∞ 1−γ
X
t
Ci,t
E0 β
t=0
1−γ

where utility takes the form of constant relative risk aversion (CRRA) with con-
sumption
1
Ci,t = [θii cρii,t + θij cρij,t + (1 − θii − θij )cρiw,t ] ρ

The consumption composite of country i consists of domestic goods (cii ), foreign


1
goods (cij ) and world goods (ciw ) with elasticity of substitution 1−ρ
. We further
assume preference is symmetric across countries θii = θii = θh and θij = θji = θf .

14
The market clearing condition of goods i states that

cii + cji + cwi = yi

Let pii represent the price of goods i in its source country. There is a trade cost
τij > 1 imposed by country i on goods coming from country j, reflecting trade re-
strictions like tariffs. Thus, the effective price of imports from country j to country i
is pij = τij pjj . As we are mainly interested in the impact of creditor’s trade policies
on debtor’s default decisions, we assume τdc ≡ 1. An implication of this assumption
is that there is no trade retaliation on the debtor’s side. On the other hand, the
creditor has some flexibility in adjusting trade costs τcd , ∈ [τ , τ ]. Tariffs also become
part of the creditor’s income for the model to yield a non-corner solution to τcd . As
we will show later, τcd is a crucial policy instrument that affects not only bilateral
trade but also bilateral debt. Lastly, the trade costs between a country and the rest
of the world are set equal to zero for simplicity: τiw = τwi = 1, i = {c, d}.
The debtor issues one-period risky bonds to the creditor. The bond market fea-
tures limited enforcement since the debtor can default on its debt. There are two
default states (S0 , S1 ) :
State 0 (S0 ): The debtor repays the bilateral debt previously and retains its financial
ties with the creditor.
State 1 (S1 ): The debtor defaults previously and is stuck in financial autarky.
In S0 , the debtor chooses from two default options (D ∈ {D0 , D1 }). It either
services the debt (D0 ) and stays in S0 , or defaults (D1 ) and downgrades to financial
autarky in the next period. In S1 , it no longer issues debt and consumes its endow-

15
ment.
The timeline of the model is summarized in Figure 2. At the beginning of period
t, the debtor can issue risky bond b to the creditor if it is in S0 . The creditor lends
money, chooses risk-free asset bc from the world financial market and sets trade cost
τ . When the one-period bond matures at t + 1, the debtor observes the realization of
its current endowment and chooses either to repay the debt so as to stay in S0 , or to
default and move to S1 . Meanwhile the creditor sets τ 0 based on state variables b, bc
and y. If the debtor defaults previously, it is in financial autarky (S1 ). Following
Aguiar and Gopinath (2006), there is an exogenous probability λ for the debtor in
S1 to regain access to borrowing.

Figure 2: Timeline

3.2 Recursive Equilibrium

The state space of the model consists of default states s ∈ S = {S0 , S1 } and a
set of fundamental macroeconomic variables including the debtor’s income, bilateral
bond holdings and the creditor’s wealth w. Denote the set as x = (y, b, w) ∈ X.

16
Agents’ value functions and decision rules will depend on S × X. In this section, we
solve for the creditor’s and the debtor’s problems and define the equilibrium of the
model.

3.2.1 Debtor’s Problem

In S0 , the debtor enters a period with b and observes the endowment realization y.
If it chooses not to default, it issues a new bond b0 at the price q(y, b0 , w) (denominated
in the debtor’s goods price). If it chooses to default, its debt b is written off but it
moves to financial autarky at the beginning of next period. Denote the value function
of a debtor who has not previously defaulted by Vd (S0 , y, b, w).

Vd (S0 , y, b, w) = max{W0 (y, b, w), W1 (y, b, w)}

where W0 (y, b, w) is the welfare by choosing D0 and W1 (y, b, w) is the welfare by


choosing D1 . A debtor makes its default decisions upon the comparison of the two
welfare levels

Ds = arg max Ws (y, b, w), s = {0, 1}


s

More specifically, W0 (y, b, w) can be expressed as

W0 (y, b, w) = max 0 U (Cd ) + βE[Vd (S0 , y 0 , b0 , w0 )|y]


Cd ,≥0,b

17
subject to
Cd + q(y, b0 , w)b0 ≤ y + b

Since everything is denominated in debtor country’s domestic good, the debtor’s


total expenditure on consumption is

cdd pdd + cdc pcc + cdw pdw


Cd =
pdd
pcc
From now on, we normalize pdd to be one and define pdd
≡ p. Thus, Cd = cdd + cdc p +
cdw . We also discipline the level of bonds with the financial constraint following
Aiyagari (1994)

b0 ≥ −
r

where r is calibrated to the world interest rate. As long as the debtor does not
borrow b0 > 0, it saves the money in the world financial market at rate qf = 1
1+r
.
Similarly, W1 (y, b, bc ) the welfare of choosing D1 follows

W1 (y, b, w) = max U (Cd ) + βE[Vd (S1 , y 0 , 0, w0 )|y]


Cd ≥0

subject to
Cd ≤ y

A country in S1 is in financial autarky, but there is an exogenous probability λ


for it to return to S0 in the next period. Hence, its value function becomes

Vd (S1 , y, 0, w) = max U (Cd ) + β(λE[Vd (S0 , y 0 , 0, w0 )|y] + (1 − λ)E[Vd (S1 , y 0 , 0, w)|y]


Cd ≥0

18
subject to
Cd ≤ y

3.2.2 Creditor’s Problem

The creditor’s problem is contingent on the debtor’s state. When the creditor
deals with the debtor who hasn’t defaulted in the last period, its value function is

Vc (S0 , y, b, w) = max U (Cc ) + βE[π00 (y, b0 , w)Vc (S0 , y 0 , b0 , w0 )|y


Cc ,b0c ≥0,τ0

+π01 (y, b0 , w)Vc (S1 , y 0 , b0 , w0 )|y]

subject to
q(a, b0 , w)b0 b
Cc − + qf b0c ≤ yc − + bc
p p

where
ccc p + τcd ccd + ccw p
Cc =
p

πmn (y, b0 , w0 ) represents the debtor’s probability of going to state Sn from state Sm
conditional on y. There is a cutoff income value y ∗ of the debtor below which it will
default. Thus, we have

Z ȳ
0 0 ∗
π00 (y, b , w) = P r(y > y |y) = f (y 0 |y)dy 0 = 1 − π01 (y, b0 , w)
y∗

If the debtor is in the default state, the creditor’s value function Vc (S1 , y, b, w) is

Vc (S1 , y, b, w) = max U (Cc ) + βE[λVc (S0 , y 0 , 0, w0 )|y + (1 − λ)Vc (S1 , y 0 , 0, w0 )|y]


Cc ,b0c ≥0,τ1

19
subject to the budget constraint

Cc + qf b0c ≤ yc + bc

The creditor’s financial wealth is its aggregate holding of the two bonds. Since there
is possibility of default, we need to multiply risky asset by the debtor’s repayment
decision D ∈ {1, 0} where D = 1 represents the repayment case and D = 0 represents
the default case.
w = D(−b) + bc

3.2.3 Bond Price

The creditor can choose between two assets: a risky asset and a risk free asset.
The former is the bilateral bond at price q. The latter is the bond purchased from the
1
world financial market at qf = 1+r
. If the debtor is in the default state, the creditor’s
saving which is the difference between its income and consumption is used solely to
purchase risk-free asset bc . If the debtor has good credit history, the creditor’s saving
is divided between b and bc . In this case, the bilateral bond price can be determined
by the creditor’s Euler equation

∂Vc ∂V 0
q = βE c0
∂Cc ∂Cc

The right hand side is the expected marginal utility from tomorrow’s consumption,
which incorporates the default probability of the debtor. As is pointed out by Lizara-
zo (2013), the bond price is higher in the case where creditors are risk-averse due to

20
the fact that there is covariance between creditors’ consumption and debtors’ default
decisions.

3.2.4 Goods Price

p denotes the creditor’s goods price pcc relative to the debtor’s goods price pdd .
Based on the creditor’s budget constraint,

ccc p + τcd ccd + ccw p q(a, b0 , w)b0 b


− + qf b0c = yc − + bc
p p p

we find p is determined jointly by debt b, wealth w and trade cost τ . In the model,
the creditor chooses optimal wealth and trade costs to maximize its utility. In this
process, it is considering the gains from both the lending channel and the trade
channel. This explains why τ may deviate from its value when the two countries
do not borrow and lend to each other. The debtor anticipates the lower trade cost
and strategically makes its default decisions. This mechanism can be used to explain
why both debt levels and default probabilities are higher than expected.

3.2.5 Markov Perfect Equilibrium

We now proceed to define the equilibrium of the model.


A Markov Perfect Equilibrium consists of the debtor’s value function Vd (S, X), the
creditor’s value function Vc (S, X), bond holdings b0 , b0c , consumption choices Cc , Cd ,
default decisions D, trade costs τ , bond pricing schedules q(y, b0 , w), and relative
goods prices p, such that

21
1. Given the bond prices q, goods prices p, trade costs τcd , the creditor’s wealth
w and consumption Cc , the debtor chooses optimal Cd , D and b0 to maximize its
expected lifetime utility.

2. Given the debtor’s default decisions D, bond holdings b0 and consumption Cd ,


the creditor chooses optimal τcd , bc and Cc to maximize its expected lifetime utility.

3. Bond markets clear at q and goods markets clear at p.

4 Computation

In this two-country model, the creditor and the debtor decide interactively their
policy rules. The numerical solution to the model is found over the space of three
state variables, b the bilateral bond,w the creditor’s wealth and y the debtor’s in-
come.
We first divide all the three state variables into grids and compute the initial
value function at each grid based on different default states. Second, we derive in-
teractively the optimal choice of bond holding of both countries and the creditor’s
optimal trade cost τ . In this process, we approximate the value function by cubic
spline interpolation, which is significantly more efficient and accurate than the dis-
crete state space technique which is commonly used for the computation of sovereign

22
default problems, as is pointed out by Hatchondo et al. (2010). After we find optimal
policy functions, we solve for the debtor’s default decision and update its value func-
tion. We continue the iterating process until the difference between value functions
in consecutive iterations is smaller than the precision criterion. The algorithm is
described in detail below.

4.1 Algorithm

Step 1. Discretize b, w, y and compute the corresponding consumption of the


debtor at all the grid nodes. In different default states S0 , S1 , calculate the u-
tility from consumption V00 , V10 . The initial value guess is the higher of the two
V 0 = max{V00 , V10 }.

Step 2. In default state S1 , solve for the creditor’s optimal choice of tariff τ1 and
bond holding bc . With τ1 , calculate the price level that clears the goods market and
the resulting debtor’s value function V11 .

Step 3. In repayment state S0 , guess an initial value of tariff τ00 and calculate the
corresponding price level.

Step 4. Given the creditor’s choice, solve the debtor’s problem to get the optimal
borrowing in the next period b0 , with which to update the best responding bond
holding b0c and τ01 by maximizing the value function of the creditor.

23
Step 5. Continue the iterating process until τ0 converges, at which time compute
the debtor’s interpolated value function V01 .

Step 6. Compare the debtor’s value function V01 , V11 , and find the maximum
V 1 = max{V01 , V11 }.

Step 7. Repeat Step 2 - Step 6, until value function converges, |V i+1 − V i | ≤ v .

4.2 Calibration

Parameters in the model are chosen in our best effort to match either stylized
facts or classical literature on the topic. The coefficient of relative risk aversion σ
is set to 2. Discount factor β is set to be relatively low as in Aguiar and Gopinath
(2006) to speed up convergence of solution and to get a reasonable prediction of
default occurrence. We set the elasticity of substitution between goods ρ to be
2 and the weight of domestic/partner’s goods in consumption is θh = θf = .3 in
the benchmark case. These two parameters are important in reflecting the relative
significance of bilateral trade. We will do a numerical exercise by looking at value
functions and default decisions when varying the values of ρ. Also following Aguiar
and Gopinath (2006), we assume income in the debtor country follows an AR(1)
process with coefficient of autocorrelation ρy = .9 and standard deviation 3.4%. The
advantage of choosing the parameter values in a classic paper is that we can directly
compare our results, and highlight the contribution our model — which is the trade
channel — to the existing literature. To this end, we also temporarily set bc = 0 and

24
Parameter Description Value
β quarterly discount factor 0.80
σ coefficient of relative risk aversion 2
r international risk-free rate 0.01
λ probability autarky ends 0.1
Income process
ρy coefficient of autocorrelation in endowment of debtor 0.9
σy standard deviation of endowment shocks of debtor 0.034
ȳ average endowment level of debtor .00058
Ā constant endowment level of creditor log2
In the benchmark case
θ weight of home/partner’s goods in consumption 0.3
ρ elasticity of substitution between goods 0.75

Table 3: Parametrization

focus on bilateral lending. To start with, we assume the endowment of the creditor
is twice that of the debtor Ā = log2. The relative economy size also comes into play
in affecting the creditor’s willingness to adjust trade costs and forgive debt.
All the parameter values are summarized in Table 3.

4.3 Results

4.3.1 Comparison with Previous Work

We first compare the performance of our model with that of Aguiar and Gopinath
(2006) (AG for short hereafter) in capturing the features of sovereign defaults. We
use 150 simulation samples with 500 periods and report statistics in the Table 4.
Among all the statistics, consumption volatility and average debt ratio are similar
across models. Trade-balance volatility is much greater in my model, as the price

25
Variable Description AG’s result Our result
std(c) consumption volatility 4.37 4.03
std(tb/y) trade-balance volatility .17 2.81
corr(y, c) correlation between income and consumption .99 .79
corr(y, tb) correlation between income and trade-balance -.33 -.10
avg(b/y) average debt ratio .27 .34
d% default probability .02% .48%

Table 4: Comparison across Models

adjusts based on the two countries’ endowment as well as creditor’s trade costs. The
correlation between income and consumption turns out to be smaller in our mod-
el, partly due to the additional uncertainty from changes in trade costs and goods’
prices. Our trade balance is counter cyclical, but the value is greater than that in AG
since creditors adjust trade costs to boost debtors’ exports. Lastly, both the debt
level and default probability are much higher in our model. It implies that trade
benefits encourage debtors to take on more debt than what they can afford to repay.

4.3.2 Trade Costs

In this part, we evaluate the adjustments in trade costs. The following two graphs
present the changes in τ in the two default states S0 and S1 given different combi-
nations of endowment y and debt b.
It is easy to spot the monotonic relationship between τ1 and y. When there is
no outstanding debt in S1 , a debtor’s price of exports negatively comoves with its
endowment. As the elasticity of substitution between goods is below unity in the
baseline case, the price adjusts in the same direction as the tariff revenue. Thus it

26
is in the creditor’s interest to set a high trade cost when the debtor’s endowment is
low. Moreover, the optimal tariff in the default state is independent of initial debt b
as the tariff does not affect repayment probability.
In S0 with outstanding debt (which corresponds to the debt renegotiation stage
in data), the optimal tariff not only covaries with the debtor’s endowment but also
the debtor’s amount of outstanding debt. For a relatively low level of debt, when
we control for b, we find τ0 decreases in the debtor’s endowment y. This fact can be
explained by the same reasoning as in the S1 state: trade policies do not matter for
the debtor’s default decision because it is always in the debtor’s interest to service
the debt. Hence, the creditor chooses trade costs that will maximize its revenue. We
also find in this region that controlling for the level of y, τ0 first decreases and then
increases in initial debt. This is largely due to the curvature of the interior solu-
tion to the goods market clearing condition. We find interesting jumps in optimal
tariffs above a certain debt level. It is within this region that the debtor is on the
brink of defaulting and has non-smooth choices of b0 . The shape of the surface can
be explained by the following reasons. When debt is high, the debtor has higher
probability to default. To avoid the financial loss of sovereign defaults, the creditor
is willing to sacrifice in the trade channel by choosing a lower value of τ . Hence,
the solution to the optimal τ0 plummets in the region. It is worth-noting that the
creditor and the debtor are best responding to each other’s choices. In expectation
of lower τ in S0 , the debtor is also willing to take more debt than in an ordinary
setting.
Next, let us compare side by side τ0 and τ1 by fixing the initial debt level to a

27
Figure 3: Optimal Trade Costs in S0 and S1

high level and a low level.


In the case where initial debt is equal to zero, τ0 and τ1 are very close in value. τ0
is slightly greater since the debtor is going to borrow from the creditor in the current
period, the loss in wealth caused by lending is partially compensated by the increased
tariff revenue. Once the level of debt goes up, τ0 is going to be significantly lower
than τ1 . This is consistent with the main empirically finding of the paper: when
the debtor is on the brink of defaulting, the creditor has the incentive to lower trade
costs in order to increase the debtor’s repayment probability.

5 Conclusion

This paper identifies the increase in bilateral trade shares between a creditor and a
debtor when sovereign default happens. The finding runs contrary to the traditional
trade sanction theory. We build a model which incorporates the trade channel in a
sovereign debt problem to account for the phenomenon. The model builds on the

28
Figure 4: τ under different endowment

strategic interaction between the creditor and the debtor. By solving the model
numerically, we are able to capture counter-cyclical trade balance and high default
probability that are closer to data than other models.
We consider extending our model in the following ways so that it reflects reality
better. First, we can build a production-economy model instead of endowment-
economy model. Many debtors are in need of developed countries’ support for capital
goods and investment. By introducing two sectors (consumption goods and capital
goods) into the model, the two countries will be more dependent on each other.
Second, we consider introducing a partial default state into the model to reflect the
renegotiation stage in sovereign defaults better. The equilibrium will feature financial
haircut, grace period and dynamics in trade simultaneously. But the extension does
come at the cost of a higher level of computation complexity. Lastly, we can relax
the assumption of constant creditors’ income, and study the creditors’ incentives
in different economic conditions. To sum up, there is much interesting interaction

29
between the trade channel and the borrowing channel. We hope future research will
explore the mechanisms in depth so that we can have a better understanding of
sovereign defaults.

30
References

Aguiar, M. and Gopinath, G. Defaultable debt, interest rates and the current ac-
count. Journal of International Economics, 69(1):64–83, 2006.

Aiyagari, S. R. Uninsured idiosyncratic risk and aggregate saving. The Quarterly


Journal of Economics, pages 659–684, 1994.

Bai, Y. and Zhang, J. Solving the feldstein–horioka puzzle with financial frictions.
Econometrica, 78(2):603–632, 2010.

Bulow, J. and Rogoff, K. Sovereign debt: Is to forgive to forget? The American


Economic Review, 79(1):pp. 43–50, 1989.

Das, U. S., Papaioannou, M. G., and Trebesch, C. Sovereign Debt Restructurings


1950-2010: Literature Survey, Data, and Stylized Facts. International Monetary
Fund, 2012.

Eaton, J. and Gersovitz, M. Debt with potential repudiation: Theoretical and em-
pirical analysis. Review of Economic Studies, 48(2):289–309, 1981.

Gu, G. W. A tale of two countries: Sovereign default, trade, and terms of trade.
2015.

Hatchondo, J. C., Martinez, L., and Sapriza, H. Quantitative properties of sovereign


default models: solution methods matter. Review of Economic Dynamics, 13(4):
919–933, 2010.

31
Head, K. and Ries, J. Increasing returns versus national product differentiation as
an explanation for the pattern of us-canada trade. American Economic Review,
pages 858–876, 2001.

Lizarazo, S. V. Default risk and risk averse international investors. Journal of


International Economics, 89(2):317 – 330, 2013. ISSN 0022-1996.

Martinez, J. V. and Sandleris, G. Is it punishment? sovereign defaults and the decline


in trade. Journal of International Money and Finance, 30(6):909–930, 2011.

Novy, D. Gravity redux: measuring international trade costs with panel data. Eco-
nomic Inquiry, 51(1):101–121, 2013.

Rose, A. K. One reason countries pay their debts: renegotiation and international
trade. Journal of Development Economics, 77(1):189–206, 2005.

Tomz, M. and Wright, M. L. Empirical research on sovereign debt and default.


Technical report, National Bureau of Economic Research, 2013.

Yue, V. Z. Sovereign default and debt renegotiation. Journal of International Eco-


nomics, 80(2):176–187, 2010.

32
Appendices

A Tables and Charts

Table 5: IMF Programs

IMF SBA IMF Standby Arrangement agreed


IMF EFF IMF Extended Fund Facility Arrangement agreed
IMF SAF IMF Structural Adjustment Facility Arrangement agreed
IMF PRGF IMF Poverty Reduction and Growth Facility Arrangement agreed
IMF SBA 5IMF Standby Arrangement in effect for at least 5 months
in a particular year
IMF EFF 5 IMF Extended Fund Facility Arrangement in effect
for at least 5 months in a particular year
IMF SAF 5 IMF Structural Adjustment Facility Arrangement in effect
for at least 5 months in a particular year
IMF PRGF 5 IMF Poverty Reduction and Growth Facility Arrangement in effect
for at least 5 months in a particular year

33
Table 6: Effect of Debt Renegotiations on Trade Volumes

bilateral bilateral trade 1to2 trade 1to2 trade 2to1 trade 2to1
FE RE FE RE FE RE
paris -0.0115 -0.0114 0.0259 0.0124 -0.0496 -0.0459
(-0.30) (-0.32) -(0.54) (-0.25) (-1.08) (-0.99)
parisl1 -0.0438 -0.0515 -0.0432 -0.0639 -0.102* -0.106*
(-1.14) (-1.32) (-0.89) (-1.30) (-2.19) (-2.25)
parisl2 -0.0358 -0.044 -0.0194 -0.0392 -0.0832 -0.0882
(-0.92) (-1.11) (-0.39) (-0.79) (-1.76) (-1.85)
parisl3 -0.057 -0.0746 -0.0732 -0.103* -0.0705 -0.0861
(-1.45) (-1.87) (-1.48) (-2.05) (-1.48) (-1.79)
parisl4 -0.0376 -0.059 -0.0183 -0.0516 -0.0724 -0.0937
(-0.95) (-1.47) (-0.37) (-1.02) (-1.51) (-1.93)
parisl5 -0.0568 -0.0662 -0.0507 -0.0695 -0.115* -0.125*
(-1.43) (-1.64) (-1.01) (-1.36) (-2.38) (-2.56)
parisl6 0.0226 -0.00173 0.0153 -0.0228 -0.0317 -0.0539
(-0.56) (-0.04) (-0.3) (-0.44) (-0.65) (-1.10)
parisl7 -0.0389 -0.07 -0.0862 -0.134** -0.053 -0.0812
(-0.96) (-1.71) (-1.69) (-2.58) (-1.08) (-1.64)
parisl8 -0.0588 -0.0972* -0.0904 -0.148** -0.0588 -0.094
(-1.44) (-2.35) (-1.76) (-2.85) (-1.19) (-1.88)
parisl9 -0.0794 -0.112** -0.116* -0.167** -0.043 -0.0737
(-1.89) (-2.63) (-2.19) (-3.11) (-0.84) (-1.43)
parisl10 -0.109* -0.144*** -0.156** -0.209*** -0.0897 -0.122*
(-2.55) (-3.32) (-2.89) (-3.82) (-1.73) (-2.33)
parisl11 -0.136** -0.178*** -0.132* -0.192*** -0.168** -0.206***
(-3.10) (-4.00) (-2.39) (-3.43) (-3.17) (-3.84)
parisl12 -0.0615 -0.108* -0.0911 -0.158** -0.0728 -0.115*
(-1.33) (-2.31) (-1.56) (-2.67) (-1.30) (-2.04)
parisl13 -0.0779 -0.130** -0.0963 -0.166** -0.120* -0.169**
(-1.63) (-2.68) (-1.60) (-2.70) (-2.07) (-2.87)
parisl14 -0.125* -0.195*** -0.212*** -0.303*** -0.118* -0.184**
(-2.53) (-3.88) (-3.40) (-4.80) (-1.97) (-3.03)
parisl15 -0.196*** -0.277*** -0.277*** -0.378*** -0.175** -0.251***
(-3.83) (-5.34) (-4.31) (-5.80) (-2.83) (-4.01)
imf -0.128*** -0.155*** -0.201*** -0.233*** -0.144*** -0.170***
(-28.48) (-34.03) (-33.79) (-38.87) (-25.29) (-29.80)
imfl1 -0.0248*** -0.0401*** -0.0366*** -0.0552*** -0.0290*** -0.0432***
(-5.08) (-8.11) (-5.71) (-8.50) (-4.73) (-6.98)
imfl2 -0.00372 -0.0132* -0.000829
34 -0.0122 0.00468 -0.00298
(-0.72) (-2.53) (-0.12) (-1.79) -0.73 (-0.46)
imfl3 -0.000458 -0.0109* 0.0129 0.000624 0.00785 -0.000803
(-0.09) (-2.00) -1.85 -0.09 -1.18 (-0.12)
imfl4 0.0281*** 0.0134* 0.0425*** 0.0247*** 0.0239*** 0.0116
(-5.1) (-2.39) (-5.95) (-3.41) (-3.48) (-1.68)
imfl5 0.104*** 0.0720*** 0.147*** 0.109*** 0.125*** 0.0981***
(-19.68) (-13.48) (-21.66) (-15.77) (-19.09) (-14.9)
custrict 0.421*** 0.398*** 0.389*** 0.387*** 0.296*** 0.325***
(-10.54) (-10.33) (-7.43) (-7.68) (-5.88) (-6.7)
ldist 0.255*** -1.334*** 0.0988 -1.556*** 0.515*** -1.438***
(-4.04) (-80.74) (-1.21) (-70.94) (-6.58) (-66.55)
lrgdp 0.332*** 0.543*** 0.216*** 0.528*** 0.487*** 0.676***
(-54.99) (-119.28) (-25.7) (-84.79) (-61.38) (-113.06)
lrgdppc 0.155*** 0.00498 0.357*** 0.102*** 0.0924*** -0.0336***
(-18.36) (-0.79) (-30.63) (-11.97) (-8.41) (-4.10)
comlang -0.0284 0.243*** 0.0850** 0.419*** -0.271*** 0.107***
(-1.27) (-12.56) (-2.81) (-16.05) (-9.51) (-4.33)
border 0.0191 0.976*** 0.0234 0.983*** -0.158 1.177***
(-0.1) (-11.27) (-0.09) (-8.63) (-0.62) (-10.44)
regional 0.295*** 0.275*** 0.291*** 0.264*** 0.321*** 0.312***
(-10.39) (-9.58) (-8.1) (-7.28) (-9.27) (-8.92)
landl 1.158*** -0.559*** 1.202*** -0.623*** 0.689*** -0.658***
(-26.48) (-27.48) (-20.6) (-22.97) (-12.35) (-24.75)
island 0.422*** 0.197*** 0.436*** 0.219*** 0.456*** 0.254***
(-20.25) (-11.84) (-15.76) (-9.9) (-16.89) (-11.74)
lareap 0.496*** 0.172*** 1.419*** 0.259*** 0.531*** 0.149***
(-5.67) (-36.3) (-12.35) (-40.79) (-4.74) (-23.9)
comcol 0.574*** 0.197*** 0.551*** 0.222*** -0.216* 0.195***
(-6.99) (-5.03) (-4.88) (-4.22) (-2.08) (-3.8)
curcol 0.348*** 0.539*** 0.374*** 0.617*** 0.644*** 0.797***
(-4.05) (-6.39) (-3.43) (-5.76) (-6.13) (-7.75)
colony 0.314** 1.408*** 0.327** 1.534*** 0.261* 1.368***
(-3.23) (-18.2) (-2.66) (-15.43) (-2.21) (-14.21)
comctry -0.701*** -1.077*** -0.670*** -1.132*** -0.776*** -1.146***
(-12.23) (-19.33) (-9.24) (-16.06) (-11.11) (-16.93)
cons -18.87*** -6.088*** -37.82*** -7.988*** -28.41*** -11.06***
(-8.99) (-33.77) (-13.72) (-33.16) (-10.59) (-47.02)
t statistics in parentheses. * significant at 10%, * significant at 1%, * significant at 1%
Country 1 denotes a debtor; country 2 denotes a creditor.

35
Table 7: Trade Policy, Regulations and Trade-Related Adjustment

CRS Code Description Clarifications / Additional notes on coverage


33110 Trade policy and adminis- Trade policy and planning; support to ministries and departments
trative management responsible for trade policy; trade-related legislation and regula-
tory reforms; policy analysis and implementation of multilateral
trade agreements e.g. technical barriers to trade and sanitary and
phytosanitary measures (TBT/SPS) except at regional level (see
33130); mainstreaming trade in national development strategies
(e.g. poverty reduction strategy papers); wholesale/retail trade;
unspecified trade and trade promotion activities.
33120 Trade facilitation Simplification and harmonisation of international import and ex-
port procedures (e.g. customs valuation, licensing procedures,
transport formalities, payments, insurance); support to customs
departments; tariff reforms.
36

33130 Regional trade agreements Support to regional trade arrangements [e.g. Southern African De-
(RTAs) velopment Community (SADC), Association of Southeast Asian
Nations (ASEAN), Free Trade Area of the Americas (FTAA),
African Caribbean Pacific/European Union (ACP/EU)], including
work on technical barriers to trade and sanitary and phytosanitary
measures (TBT/SPS) at regional level; elaboration of rules of origin
and introduction of special and differential treatment in RTAs.
33140 Multilateral trade negotia- Support developing countries effective participation in multilateral
tions trade negotiations, including training of negotiators, assessing im-
pacts of negotiations; accession to the World Trade Organisation
(WTO) and other multilateral trade-related organisations.
33150 Trade-related adjustment Contributions to the government budget to assist the implemen-
tation of recipients own trade reforms and adjustments to trade
policy measures by other countries; assistance to manage shortfalls
in the balance of payments due to changes in the world trading
environment.
Source: OECD Aid for Trade
Figure 5: Aid for Trade during Sovereign Defaults

37

You might also like