WPS3673
The Overhang Hangover
Jean Imbs
HEC Lausanne, FAME and CEPR
Romain Ranciere
CREI and IMF
JEL Classification Numbers: E62, F34, F43, H63
Keywords: Debt Overhang, Kernel Estimation, Debt Contracts, Investment, Debt Relief.
World Bank Policy Research Working Paper 3673, August 2005
The Policy Research Working Paper Series disseminates the findings of work in progress to
encourage the exchange of ideas about development issues. An objective of the series is to get
the findings out quickly, even if the presentations are less than fully polished. The papers carry
the names of the authors and should be cited accordingly. The findings, interpretations, and
conclusions expressed in this paper are entirely those of the authors. They do not necessarily
represent the view of the World Bank, its Executive Directors, or the countries they represent.
Policy Research Working Papers are available online at http://econ.worldbank.org.
For helpful comments, we are grateful to Ilian Mihov, participants at the CEPR Conference on Institutions,
Policy and Growth, INSEAD, May 2005, and at the World Bank's Macroeconomics seminar. Our special
thanks go to Aart Kraay whose input provided the initial impetus to this work. The paper was written while
Imbs was visiting Chicago GSB, whose hospitality is gratefully acknowledged. Financial support from the
Research Department and the Debt and Growth Department at the World Bank, and from the National
Center of Competence in Research "Financial Valuation and Risk Management" is gratefully acknowledged,
as well. The National Centers of Competence in Research (NCCR) are a research instrument of the Swiss
National Science Foundation. Corresponding author. Imbs  HEC Lausanne  Dept of Economics  Lausanne
Switzerland 1015. jimbs@unil.ch.
1 Introduction
Most developing economies can not grow without borrowing to finance the technology gains
and capital deepening that come with economic progress. On the other hand, high levels
of debt are (again) increasingly accused of deleterious effects on economic development, in
what appears to be a revival of a 20yearold debate on the virtues of debt relief. The
notion of debt overhang is coming back to the forefront, and with it a renewed interest in
empirical work identifying what countries have reached the downward segment of a debt
Laffer curve, and why. This paper is part of that effort.
Theoretical arguments supporting the existence of a debt Laffer curve fall into two
broadly defined categories. First are theories based on multiple equilibria, where investment
endogenously collapses beyond a certain level of indebtedness, in preparation for default and
in order to minimize penalty payments, exogenously assumed to equal a fixed proportion of
output. Second are theories where the nature and terms of the optimal debt contract are
affected by the level of existing indebtedness. As debt levels rise, it becomes increasingly
difficult, and eventually impossible, for a creditor with imperfect monitoring technology to
elicit effort on the part of the debtor. The borrowing economy then loses all incentives to
implement policies that are painful in the shortrun but beneficial in the longrun.
In this paper, we evaluate the plausibility of these (relatively) old theories in (relatively)
recent data, using modern empirical techniques. We do this in two steps. First, we imple
ment standard reduced form growth regressions, with a view to selecting a panel of countries
that have effectively gone through a debt overhang episode in a sample of 87 developing
economies. Our selection device is simple. We use a variety of kernel estimators to charac
terize the relation between debt and growth at different levels of indebtedness. This verifies
whether economic growth depends nonmonotonically on debt levels without imposing any
restriction on the shape of the nonlinearity, quadratic or otherwise. In other words, we
investigate the existence and shape of a hypothetical debt Laffer curve in the developing
world. This also identifies precisely the level of debt at which the sign reversal occurs,
i.e. the maximum point on the Laffer curve, again without any parametric restrictions.
Overhang countries are defined as those going through this reversal in sample.1
We pay special attention to the possibility that high indebtedness and low growth arise
simultaneously from omitted variables. This is illustrated in the substantial differences that
exist between the debt Laffer curve implied by growth regressions with or without country
specific intercepts. The estimated number of countries on the downward segment of the
debt Laffer curve is much larger without any fixed effects, which suggests that in many a
case, low growth and high debt occur for timeinvariant, country specific reasons. This is
conceptually very different from debt overhang, which rests fundamentally on a dynamic
argument, i.e. on the withincountry variation in the panel.2 We use our kernel approach
to identify the variables that affect both indebtedness and growth. In particular, we isolate
1Timing issues are treated with care. We estimate the (conditional) effect of initial debt on subsequent
economic growth, and illustrate the importance of simultaneity bias in the context of these growth regressions.
2Even in the case of an economy particularly prone to overhang problems for institutional reasons (for
instance one where creditors can only monitor loans imperfectly), the variation of interest remains over time
and withincountry. There, debt overhang can be particularly severe once the threshold level of indebtedness
is reached, but it will be absent at lower levels, in the same exact way it would be in a country with good
creditors' rights.
2
the sample of overhang countries as implied by (kernel) growth regressions without country
fixed effects, and investigate which controls affect the significance of the effect of debt in
a sample where it is initially significantly negative. We focus in particular on institutional
characteristics of the economy.
The paper's second step involves departing from reduced form growth regressions, and
performing direct tests of the mechanisms that underpin the debt Laffer curve. Overhang
countries are regrouped in a balanced panel which we subject to an event study, where the
onset of debt overhang is the shock whose chronology we seek to characterize. We use our
nonparametric estimates to date the threshold level of indebtedness, and ask three questions
of the event study. First, is investment falling precipitously about the overhang date?
Theories of optimal default suggest investment should respond at or after the overhang date,
but not before. A collapse in investment prior to the overhang date would be suggestive that
an explosion of the debt to GDP ratio is a symptom, rather than the cause of an investment
slump. Second, is economic policy observably deteriorating at or after the overhang date?
Theories of optimal debt contracts suggest incentives alter as a result of reaching a threshold
level of indebtedness; again, the reverse timing is suggestive of reverse causality. Third, do
the terms under which borrowing is contracted worsen noticeably at the threshold debt
level? There, debt overhang occurs because creditors become unable to write incentive
compatible contracts with highly indebted debtors, and choose instead to exact punitive
premia. This effect should be tempered somewhat in environments where creditors are
protected, or have access to monitoring technologies that limit debtors' moral hazard, as
for instance in economies with developed financial markets.
Our results are as follows. Most of our estimates are supportive of a debt Laffer curve  or
at least a negatively sloped relation between debt and growth at high levels of indebtedness.
On average, debt overhang occurs when the face value of debt reaches 55 to 60 percent
of GDP or 200 percent of exports, or when the present value of debt reaches 35 to 40
percent of GDP or 140 percent of exports. Then, initial debt tends to be associated with
subsequently low growth. These thresholds apply withincountries, accounting for country
specific institutional arrangements. They are valid for the average developing economy in
our sample. Still, institutions do matter for debt and growth. In particular, we find that
government effectiveness, the rule of law and bureaucratic quality all correlate positively
with economic growth, and tend to limit debt buildup. Debt overhang may still happen in
economies endowed with good institutions, but for higher values of debt.3
The event study provides clear support for a fall in investment after the onset of over
hang. We uncover some evidence that economic policy in particular price stability deteri
orates. Indices capturing the overall quality of economic policy markedly worsen once the
overhang threshold is reached. Interest rates on new borrowing, on the other hand, tend
to fall. This runs contrary to the theory, but can actually stem from extensive rationing.
Indeed, we find that at the overhang date, quantities lent by the private sector collapse
precipitously, and the bulk of lending originates then from multilateral official agencies, at
concessional rates.
We ascertain our results indeed arise because of overhang mechanisms, and rule out the
following two prominent alternatives. First, we consider the possibility that world interest
3Our event study ignores this heterogeneity, in that it uses the average estimated overhang threshold for
all countries.
3
rates soared during our sample, with the resulting crowding out of investment particularly
prevalent amongst highly indebted economies. Actually, debt service tends to fall over
the event chronology  a result that is consistent with falling interest rates, but not with
crowding out effects on investment. In addition, we find only muted decreases in investment
during overhang episodes in countries where property rights are strongly enforced. This is
consistent with the notion that creditors are best able to monitor debtors when the required
institutions are present, and thus continue to be able to sign optimal debt contracts even
at high levels of indebtedness. But again, this is inconsistent with global interest rate
shocks affecting investment indiscriminately amongst highly indebted economies. By the
same token, we find important differences in the sample formed by low income overhang
countries, relative to the rest of our events. Low income economies seem to suffer the
brunt of the overhang effects we identify. Since both subsamples have by definition similar
indebtedness levels, this suggests once more ours is not a story of high world interest rates
hampering investment in highdebt countries.
Second, we ensure our results are not driven by the debt crisis of the 1980's and the
ensuing wave of debt rescheduling agreements. In theory, rescheduling may alleviate over
hang issues, in that it could bring the debtor back in the region where incentivecompatible
contracts are possible, and investing in the future optimal. In practice however, measures
of external debt do respond to restructuring episodes. In other words, we may exclude some
countries from our sample of overhang events simply because, in sample, debt ratios fall
back below our estimated thresholds once rescheduling occurs. This tends to exclude cases
when incentives might actually have altered as theory predicts, and thus acts against us
finding any evidence of overhang mechanisms. If anything, excluding rescheduling episodes
should reinforce our results. We investigate this in two ways. First, we simply eliminate
all years between 1979 and 1984 from our sample, and continue finding the same dynamics.
Second, we eliminate from our sample all substantial rescheduling episodes.4 The same con
clusions obtain: debt relief may improve investment and economic policy, but is unlikely to
ease borrowing conditions, as the private sector comes back to take over from concessional
loans.
The rest of the paper is structured as follows. We set the stage in section 2 with a
helicopter tour of the theoretical mechanisms whereby unsustainable debt hampers economic
growth. We also review some of the recent (and less recent) empirical evidence. We next
present, in section 3, our measurement strategy and detail our considerable dataset. Section
4 contains the body of our results, separated into parametric and nonparametric estimates
of a debt Laffer curve. We also discuss the role of country fixed effects in affecting economic
growth and debt accumulation simultaneously. Section 5 then describes the event of interest
 the onset of debt overhang  and tracks its impact on investment, policy choices and the
terms of borrowing. Section 6 reviews some robustness tests, and Section 7 concludes.
2 Overhang Overview
We review the literature on debt sustainability and the theoretical mechanisms whereby
(high level) debt can have deleterious effects on economic growth. We stress in particular
4We choose to eliminate all episodes that reschedule more than 5 percent of the face value of debt.
4
three mechanisms, based on the increasing difficulty in writing incentivecompatible debt
contracts as the level of debt rises, and in particular the possibility that default become
optimal at high debt levels. We also review the relevant empirical work.
Krugman (1988) defines debt overhang as a situation where "the expected present value
of future country transfers is less than the current face value of its debt". In an overhang
situation it may still be profitable for debt lenders to roll over the debt in order to recoup
part of their claims and extract some future country resources. However, if these in turn
depend on the debtor's effort, creditors will have to take into account the incentives effects
of demanding further payments. If all future debtor's resources are to be used to repay its
creditors, there will be little incentive to follow policies that may be painful in the short run
but growthenhancing in the longrun. An optimal debt contract strikes a balance between
two constraints: on the one hand the necessity to set repayments high enough so that lucky
outcomes will effectively generate transfers back to creditors, but exorbitant demands would
compromise any willingness on the debtor's part to increase or even maintain its ability to
repay.5 The higher the level of debt, the harder it becomes to preserve incentives When the
optimal incentivecompatible contract implements a positive level of effort, a suboptimal
contract  like the one that forces maximum repayment  will reduce effort, expected growth
and therefore the present value of repayments as well. This is the basis for a debt Laffer
curve: the present value of debt repayments first increases in debt's face value, up to a
point beyond which the correlation becomes negative. Then, a higher face value of debt is
associated with lower effort, and lower present value of repayments. As long as the ability
to repay depends on growth performance, the negative portion of the debt Laffer curve also
correspond to a negative correlation between debt and growth, where increasing debt tends
to be associated with worsening policy choices.
An important question is why some countries lay on the right side of the debt Laffer
curve, even though debt forgiveness would be Paretoimproving. A classical explanation
builds on a free rider problem: while all lenders collectively would be betteroff financing a
portion of the debt and forgiving the rest, each lender taken individually would prefer to
opt out of the rollover and demand full repayment.
Piketty (1997) shows there might be situations where even a debt contract that elicits
high effort on the borrower's part can itself be suboptimal. In highly indebted economies, or
ones with poor institutions, signing debt contracts that preserve the borrower's incentives
to repay becomes increasingly difficult. Creditors prefer to give up incentives altogether,
expect high repayments only if a lucky state of nature realizes, and thus exact prohibitive
conditions from the borrower. The level of debt where creditors eschew incentivecompatible
contracts is the maximum in a debt Laffer curve, as beyond this level borrowers do not try
anymore to maintain or improve their ability to pay. We should observe a deterioration in
the terms of borrowing along with a lower mean growth.
Obstfeld and Rogoff (1996) show that Krugman (1988) debt overhang problem can
be reformulated as the outcome of a simple twoperiod consumptioninvestment decision.
Suppose that a debtor country has to make a risky investment decision while an inherited
stock of debt is due to mature the following period. For a given investment, a higher
debt level increases the number of states of nature where default occurs. Assuming that the
5Aghion and Bolton (1997) and Piketty (1997) provide a general characterization of this problem and its
dynamic implications
5
default penalty is proportional to output, the inherited debt plays the role of an effective tax
on investment: as default becomes more likely or optimal, the borrower's purpose becomes to
minimize penalty payment, that is minimize future output, for instance reducing investment.
Inherited liabilities have a debt overhang effect on investment. Once again, debt forgiveness
will increase investment as well as the present value of debt repayments.
Krugman (1988) suggests that a way to escape the tradeoff between debt forgiveness (to
preserve incentives) and debt financing (to obtain maximum repayment in good states of the
world) is to convert debt into statecontingent claims. Cohen and Sachs (1986) and Cohen
(1995) develop this view in an infinite horizon model of debt and growth with a risk of debt
repudiation. At first, high growth is financed with increasing debt to GDP ratios until an
endogenous debt ceiling is reached. When the credit constraint binds, growth performance
depends on the repayment strategy followed by creditors, and its implication on debtors
incentives. The optimal repayment strategy is to let the performing debt assets grow with
the expected growth of the economy. If this is implemented, growth is faster than under
autarky and a crowding in effect ensues, with debt service negatively correlated with the
borrower's investment decisions. But such a "smooth payments" policy requires that the
creditor be able to monitor the borrower's investment strategy. If the nature of institutions
or contractual arrangements are such that monitoring cannot be ensured, the creditors
optimal strategy is to claim a constant share of output. This amounts to a distortionary
debt tax on output, leads to inefficiently depressed levels of investment and low growth.
The terms of borrowing for highly indebted economies should once again worsen observably
once the overhang zone is reached, and the severity of this response should depend on the
creditors ability to monitor borrowers' investment policy.
Finally, the political economy can shed some light on the reasons why countries end up
highly indebted, but this literature does not address directly the mechanisms that link debt
and growth. For instance Velasco (1997) shows that fragmentation in fiscal authorities can
create a tragedy of commons, which results in overspending and excessive debt accumula
tion.6 Alesina and Tabellini (1989), in turn, explain why successions of government with
different distributional goals creates fiscal uncertainty that generates capital flight, low in
vestment and overaccumulation of external debt. There, high debt and low growth prevail
simultaneously because of institutions that are prone to overborrowing, and that tend to
divert investment from efficient uses, rather than as cause and consequence. High levels
of debt do not inherently alter borrower's behavior or incentives, and most importantly,
debt relief would not prevent renewed debt accumulation, low investment and low growth.
Distinguishing between these two possibilities is obviously of paramount importance, if only
for policy reasons.
The empirical literature on debt and growth has followed two strands. A first set of pa
pers have attempted to test directly the potential crowdingout effect of debt on investment.
The second approach fits in the empirical growth literature, and investigates the reduced
form (conditional) effects of debt on growth in crosscountry regressions, with particular
focus on the presence of nonlinear relations. Cohen (1993) finds that the level of debt
had no significant impact on investment during the debt crisis of the early eighties. Over
the same period however, the surprise increase in debt payments correlated negatively with
6 The benchmark model in this literature is Barro's (1979) model of optimal level of public debt where
debt is used to smooth the effect of distortionary taxation.
6
investment, thus suggesting a crowding out effect. In contrast, Warner (1992) shows that
some significant determinants of investment which are unrelated to debt can explain well
the decline observed in highly indebted countries in the eighties. In particular, the combi
nation of an increase in world interest rates and a fall in commodity prices can account for
most of the observed decline in investment.7
Patillo, Poirson and Ricci (2002) [henceforth PPR] follow the alternative route. They
estimate the conditional correlation between debt and growth in the context of standard
panel growth regressions, and investigate whether the sign reverts at high enough debt
levels. They find clear evidence that debt becomes detrimental for growth in highly indebted
economies, and quantify the threshold levels in the thus confirmed debt Laffer curve using
a variety of debt measures.
As will become clear, parts of this paper are largely inspired by PPR. One key difference
however pertains to the actual estimates of debt levels beyond which the marginal effects
of debt become negative. PPR's results suggest the marginal effects of debt are negative
for face values larger than 30 to 115 percent of exports, or 5 to 90 percent of GDP, and for
present values larger than 30 to 295 percent of exports, or 5 to 50 percent of GDP. The
imprecision in their results may be due to their using quadratic functional forms, or spline
estimators that select the threshold level on the basis of goodness of fit criteria (Rsquared).8
In contrast, the kernel approach we adopt enables more precision, in that it identifies the
very first sample where debt correlates negatively with growth, when countries and years
are ranking by increasing level of indebtedness. The median debt in that sample indicates
the threshold of interest, which does not depend on the arbitrary choice of a functional
form.
Finally, Clements, Bhattacharya and Nguyen (2003) estimate a quadratic relation be
tween debt and growth, in a sample of low income countries. Their estimates point to an
important role for public investment, which they argue high levels of debt tend to crowd
out.
3 Data
Of crucial importance for our purposes is the strategy we adopt to capture a country's level
of indebtedness. We focus on measures of gross external debt, for a sample of developing
economies that includes low and middle income according to the World Bank classification.
Financial flows are largely unilateral over our sample for a vast majority of developing
economies, which justifies our focus on gross measures.9 As is well known, a large proportion
of debt in developing economies is also external. At the very least, this is the component of
debt that relief programs propose to target, and thus presumably the most relevant from the
standpoint of discussing debt overhang. We use two measures: Total Outstanding External
7More precisely, estimating the determinants of investment to GDP ratios (debt excluded) before the
debt crisis performs well outofsample during the high debt period.
8In Patillo, Poirson and Ricci (2003), the same authors use growth accounting techniques to identify
the sources of the growth effects of debt. They conclude that debt deteriorates growth via lower capital
accumulation and a fall in Total Factor Productivity.
9Information on net debt is also substantially harder to come by for the type of coverage we endeavor.
7
Debt (TOD), taken from the Global Development Finance dataset, which tracks the face
value of the stock of external debt, and a Present Value measure (PV), computed as the
discounted sum of future external debt payments. The former is standard in the empirical
literature on debt, whereas the latter is more recent and relies on specific assumptions
regarding the discount rates and amortization profiles. PPR use both and so do we. We do
however use two PV measures, one constructed by Easterly (2001) (PVE) and the second
by Dikhanov (2004), PVY.
The two measures differ in three ways. PVE builds from aggregate countrylevel data
on the terms of borrowing, whereas PVY is based on loanbyloan data that are aggregated
up to the country level. In addition, PVY allows for currencyspecific and timevarying
discount rates. Unlike PVE, it is however restricted to public and publicly guaranteed
debt. While this is a narrower measure than PVE, it will also provide some robustness
checks on the importance of debt ownership. Both measures assume a linear amortization
schedule.
We follow PPR and construct two ratios for each debt measure, expressed relative to
GDP or to exports. The ratio of debt to exports captures the external resources effectively
available to cover external debt liabilities, and is often used by practitioners, but it is also
more sensitive to term of trade shocks, and thus more volatile, than ratios to GDP
The rest of our data are standard, but offer relatively broad coverage. We construct
a panel of observations for 87 developing economies over the period 19692002, which we
use in two empirical exercises. First in standard growth regressions, based on three or
fiveyear averages; second in an event study where we use all the available time variation.
Our control variables in the growth analysis are classic and inspired from the robust sets
proposed in Levine and Renelt (1992).10 They include initial income, openness to trade,
population growth, secondary schooling and the growth rate of the terms of trade. We also
experimented with the fiscal balance, with no changes in conclusions. Construction of these
variables is standard; the sources used are the World Development Indicators, the World
Economic Outlook and the International Financial Statistics.
In the event study, we track the response over time of investment, macroeconomic policy,
and the terms of borrowing. Of these, only investment is readily available and its measure
relatively uncontroversial. We choose to investigate the dynamic response of two policy
related variables: inflation and government expenditures. If the conduct of economic policy
does indeed deteriorate at high levels of debt, we conjecture that at least one of these
measures will show a systematic response. We also include an index computed by the World
Bank, the Country Performance International Assessment, or CPIA, meant to summarize in
one number an assessment of the overall quality of policy stance.11 Because we are interested
in the terms under which debt is contracted, we finally bring in additional information from
Global Development Finance publications, with a view to isolating changes in the rate of
interest for different countries and debt levels. In particular, we collect the value of new
10Doppelhofer, Miller and SalaiMartin (2004) recently used Bayesian techniques to isolate an alternative,
updated set of control variables. The approach is however purely crosssectional, and its results cannot be
used in the present context where withincountry growth determinants are of the essence. For instance, the
most robust correlates of economic growth include the extents of confucianistic or protestant religions, or
geographic binary variables, none of which lend themselves to a panel estimation.
11These data are confidential.
8
loan agreements contracted by either private or official creditors, as well as the average rate
at which they are contracted.12
Finally, several institutional factors can affect the relationship between external debt
and growth, as well as the onset of an overhang episode. We seek to characterize the
institutional arrangements likely to result in both high indebtedness and low growth. We
investigate the role of government efficiency, as institutions prone to tolerate the official
squandering of resources are likely to hamper growth while they also facilitate debt build
up. We use the measure of bureaucratic quality constructed in the International Country
Risk Guide (ICRG), as well as government effectiveness, one of the Kaufmann, Kraay and
Mastruzzi (2004) (KK) indicators.
4 A Debt Laffer Curve
We first revisit the role of debt as implied by standard growth regressions, paying particular
attention to timing issues. We then introduce our kernel estimator to characterize a debt
Laffer curve with as little parametric assumptions as possible. Finally, we discuss the dis
crepancies that arise when comparing between and withincountries estimates. We relate
these differences to the role of institutions, and investigate which tend to result simultane
ously in high debt and low growth
4.1 Debt and Growth Regressions
Debt overhang should prevail only for high enough levels of indebtedness. Below that, the
relation between debt and growth is theoretically ambiguous. For instance Barro and Salai
Martin (1995) show, in an augmented Solow growth model, that access to foreign borrowing
leads to a faster rate of convergence. In that context however the role of external debt
should be entirely reflected in the rate of investment and should not have an independent
effect of growth. Debt finances investment, and thus fosters growth, but no direct effect is
discernible. For this reason, we omit investment from the set of control variables in what
follows. We want to allow for a possible channel that works via investment, since this is
one of the prominent theoretical possibilities.13
For purposes of comparison with the existing empirical literature, we first briefly reassess
the link between debt and growth in the context of a linear panel approach using non
overlapping windows of three and fiveyear.14 We consider the general growth specification
yit+1  yit = Dit + Xit0 + i + t + it (1)
where yit is the log of per capita GDP, Dit a debt ratio, Xit a vector of control variables
and i and t are country and time fixed effects, respectively. There are six measures of
12These data include new money lent as part of restructuring deals, and thus does not strictly focus on
the terms of purely new contracted debt.
13In fact, including investment yields overall similar results, with perhaps slightly higher estimated over
hang thresholds. This indicates that investment is a relevant channel for overhang effects, but not the only
one, which is confirmed later in the paper.
14With threeyear windows, the dataset includes up to eleven time units from 19691972 to 19992002.
With fiveyear windows, the dataset includes up to seven times unit from 19711975 to 19762000.
9
Dit: we normalize each of our debt measure, total outstanding debt and the two alternative
present value measures (PVE or PVY) with either the value of exports or nominal GDP.
The set of control variables, in turn, includes: population growth, secondary schooling,
investment to GDP and the growth rate in terms of trade. We estimate equation (1) using
three techniques: i) Ordinary Least Squares, ii) Fixed Effects, and iii) a GMM system
estimator. The GMM estimator controls for the bias resulting from the correlation between
the lagged dependent variable and the fixed effect. The GMM system estimator corrects for
the imprecision of the difference estimator by jointly estimating equation (1) in difference
and in level. The set of instruments used with GMM are lagged levels for the difference
equations and lagged differences for the level equation.
A similar approached was followed by PPR. An important difference concerns the treat
ment of the potential endogeneity in the debt and growth relation. Specifically, it is crucial
to separate the period used to measure GDP growth from that used to measure indebted
ness as a ratio of this very same GDP. A period of high growth will mechanically reduce
the debt to GDP ratio and induce a negative relation that bears no relation to overhang
mechanisms.15 In what follows we compare the relation between average GDP growth and
debt ratios, where they are both computed over the same period, to that between initial
debt and subsequent growth. We show using average values tend to generate a negative bias
in the estimation of the debt effect. Using initial debt ratio is also more consistent with the
theoretical prediction that relates high debt levels to lower subsequent growth performance.
Further, we present results based on a panel constructed using fiveyear averages, in order
to filter out the effect business cycles fluctuations. Tables 1, 2 and 3 summarize our findings.
Table 1 presents the results of the estimation using Present Value of Debt (PVY) and
threeyear windows. Two features emerge. First, a negative and significant link between
debt and growth appears only in OLS estimations; second the point estimate on debt is
reduced when initial rather than average debt ratios are used. This will be a regularity
across all debt measures.
Regarding the control variables, we find that the convergence terms is negative and
significant in all regressions. The coefficient for the other variables tend to exhibit the
expected sign but are not always significant.16 From the Sargan test and secondorder
Serial Correlation test, we conclude that overall validity of the instruments used in GMM
estimation cannot be rejected.
Table 2 summarizes the effect of debt on growth when the regressions reported in Table
1 are performed with various measures of debt ratios. The reduction of the debt coefficient
when we use initial rather than average ratio is a common pattern across all regressions.
Table 3 presents regressions results obtained with fiveyear nonoverlapping windows and
using initial debt ratios. Only one estimation out of eighteen exhibits a negative and
significant effect of the initial debt ratio on growth. Overall, we can conclude that there
15This problem is not solved when internal instruments are used in the GMM estimation. The validity of
the GMM identification relies on the assumption of weak exogeneity of the explanatory variables, which is
not fulfilled under the scenario we discuss here.
16The relatively low tstatistics in the two steps GMM system estimation can be largerly attributed to
the use of the Windmeijer (2005) small sample correction.
10
is no robust linear evidence of a negative relationship between debt and growth in the full
sample. Of course, this may well reflect the prevalence of nonlinearities.
4.2 Two Kernel Estimators
In this section, we propose a nonparametric empirical strategy meant to assess the existence
of a debt Laffer curve, while imposing as little structure on functional forms as possible. In
addition, we seek to identify a subsample of countryyears in which overhang mechanisms
may be operating.
We use two kernel estimators. The first one derives from sequential estimations of equa
tion (1) on rolling subsamples of observations, ranked by their initial level of indebtedness.
N countryyears observations are ranked by increasing values of (initial) debt ratios. Let
j denote the rank of each countryyear observation according to this ordering. The kernel
bandwidth, denoted by l, is chosen arbitrarily, but robustness along this margin is ensured.17
We then estimate equation (1) on the first l observations, roll the subsample over by one
unit and perform a new estimation on the thus modified sample. We stop when we reach the
subsample of l observations with highest indebtedness. Each subsample is characterized
by its midpoint debt level Dj, the median (initial) debt ratio computed for each window.
Formally, for each j (1,Nf l + 1), we estimate

yit+1  yit = jDit + Zit0j + it, (i,t) lj (2)
where Zit = [Xit,i,t] and lj denotes the subsample j of l countryyears observations.
Each individual coefficient is derived from a parametric linear estimation, but we do
not impose any functional forms on the relationship between growth and debt, nor on that
between growth and the other control variables. Crucially, estimates of j at high levels
of debt do not depend on observations at low debt levels. This is consistent with theory,
where the onset of an overhang episode corresponds to dramatic changes in incentives once,
and only once, the debt threshold is reached. This independence feature will be absent from
any parametric estimation of nonlinearities performed over the full sample.
While this approach has the simplicity afforded by a "rolling window" interpretation, it
does not go without problems. First and foremost, it allows for all covariates to depend non
parametrically on debt levels. This complicates interpretation, as estimates of the impact
of debt on growth taken from successive samples are not directly comparable. A difference
could arise because other covariates also change in significance or in importance. Second,
even if the linear coefficients were stable across samples, the presence of a nonlinearity
creates a bias, which in turn can affect all estimates. We address these concerns using the
partiallinear kernel estimator introduced in Robinson (1988). The approach involves a se
quence of parametric and nonparametric regressions, with straightforward intuition. The
nonlinearity is first eliminated from both dependent and independent variables, through bi
variate kernel estimations. Ordinary leastsquares using the two resulting residuals provides
unbiased estimates of the linear coefficients. With these in hands, the dependent variable is
17We experimented with bandwidths going from 150 to 300 observations. Unsurprisingly, the shape of
the Laffer curve smoothens as l increases, but our main result of a a significantly negative effect of debt on
growth at high debt levels always prevails.
11
purged of its linear determinants, and the residual used to estimate the nonlinear relation
of interest.
and E{Zit / Dit} for all (i,t) lj, . We then construct the corresponding residuals
More formally, we first use simple kernel techniques to estimate E{(yit+1  yit) / Dit}
unbiased estimates of j. Finally, we implement a kernel estimator of (yit+1yit)Zit0j on
(yit+1  yit)  E{(yit+1  yit) / Dit} and Zit  E{Zit / Dit}, and use least squares to obtain
debt. The approach addresses the issues created by the hybrid nature of the relation we seek
to identify, but at some efficiency costs. To ascertain significance, we present bootstrapped
standard errors.18
Using GMM estimators in either kernel estimation is in practice unwieldy. Indeed,
the ordering of our observations by countryyear does not guarantee the presence of the
relevant lags necessary to build up the internal instrument matrix.19 We are therefore
restricted to kernel estimations based on OLS and Fixed Effects. When using the latter, each
individual coefficient estimate is based on a withincountry estimation, but as the sample of
countryyears changes across windows, variation in estimates also reflects betweencountries
differences.20
In contrast, PPR look for nonlinearities imposing one of two functional forms. First
they investigate the significance of a linearquadratic component, and second they impose
a piecewise affine function or linear spline. But the linearquadratic specification can be
misleading in that it can confuse monotonic concavity with nonmonotonicity. Furthermore,
linear quadratic functional forms tend to deliver estimation results that can depend heavily
on extrema.21 The piecewise linear estimation solves some of these issues, but it can deliver
vastly different results depending on the number of assumed discontinuity points. In general,
a drawback of nonlinear parametric specifications is they often lead to conclusions based
on outof sample thresholds. This is especially problematic when the aim is to select sub
samples of observations with certain properties, as is the case for the purposes of our event
study.
Our kernel estimates are derived from a panel of fiveyear averages, using the initial
values of total outstanding debt or the updated measure of debt present value constructed
by Dikhanov (2004), PVY. Figures 1 and 2 present the kernel estimations for ratios of debt
to GDP and to exports, respectively. The bandwidth is set to l = 200. On each scatterplot,
partial effect of debt on growthf(j)(1,Nl+1) on the vertical axis. For each figure, the left
midpoint level of debt ratios (Dj)j(1,Nl+1) are on the horizontal axis and estimates of the
and right panels report estimates as implied by OLS and panel Fixed Effects, respectively.
18We follow Yatchew (2003) , and bootstrap the residuals corresponding to the last kernel in the procedure.
We simulate 10,000 repetitions, with replacement, for each step of the kernel.
19In addition, the poor small sample properties of GMM make it undesirable to run this estimator on
truncated slices of our initial panel, with each much fewer observations.
20The Monte Carlo simulations in Hauk and Wacziarg (2004) suggest that our inability to use a GMM
estimator may not be problematic after all. Their results suggest the archetypical growth regression is
actually best estimated by Ordinary Least Squares, even in the presence of countryspecific intercepts and
a lagged dependent variable. GMM estimations suffer from small sample biases, whereas the withingroup
estimator exacerbates measurement error. In our case, the additional covariate, debt, demands that we
investigate the importance of country fixed effects, in order to isolate the dynamic impact of increasing debt
on growth.
21Especially when applied on logarithms.
12
Points in bold correspond to estimates of j significantly different from zero, at the 10
percent confidence level.
OLSbased kernel estimates suggest debt starts having significant deleterious effects on
growth when it reaches just above half of GDP. With fixed effects on Figure 1, the non
linearity exhibits a similar pattern but the threshold level now increases to around 60 percent
of GDP. In this region the point estimate for j is around 1.7. This implies an increase in
the debt to GDP ratio by a factor of 1.5 translates into an annual GDP growth rate lower
by 0.7 percent.22 Interestingly, the link between debt and growth becomes insignificant
at high debt levels when country fixed effects are included: it is largely because of time
invariant controls that high debt tends to be associated to in low growth. The Figure's
lower panel presents the kernel estimation results for the ratio of debt face value to exports.
As in Figure 1, controlling for fixed effects tends to substantially increase the range over
which a negative and significant relationship between debt and growth exists, from 150
to 200 percent of exports. A noticeable difference is that coefficients j do not exhibit a
similar downward pattern in OLS based kernel estimations. This might by explained by
crosscountry differences in export levels, which are captured imperfectly by the control
variables and, in particular, by trade openness. In fact, when fixed effects are controlled
for, the kernel estimation of debt to exports coefficients do exhibit a downward pattern
again.
The results for the ratios involving the present value of debt are reported on Figure 2.
The same phenomena are apparent: the estimated overhang range shifts to the right when
country specific intercepts are allowed for. OLS suggests a maximum in the debt Laffer
curve occurs when the present value of debt reaches around 30 percent of GDP, but it is
closer to 40 percent with country specific fixed effects. OLS estimates for the ratio to exports
are once again not downward sloping, but this is corrected when the kernel implements a
withingroup estimator. Interestingly, both present value measures also point to a positive
and significant effect of debt on growth at low debt levels, i.e. below 20 of GDP and 50
percent of exports.23
The debt relief initiative has been targeting especially low income economies, where
overhang issues are argued to be most prevalent. It is therefore of particular interest to
reproduce our analysis inside and outside of the sample formed by the 45 countries catego
rized as low income by the World Bank. Figures 3 and 4 report the corresponding estimates.
Even though the conclusions are somewhat weakened in reduced samples, the distinction
according to income levels is clearly relevant.24 While significantly negative coefficients
continue to prevail in the sample of low income countries, they are completely absent from
the complementary set of countries. This suggests the view that low income countries are
disproportionately affected by debt overhang problems is justified, at least on grounds of
growth effects. It remains to be seen whether the mechanisms at play are indeed consistent
with theory: comparing Figures 3 and 4 suggests the main reason why low income countries
may experience overhang is simply because they are more indebted on average.
22I.e. ln(1.5)*1.7
23The actual number of significant coefficients represented on the Figures cannot straighforwardly be
interpreted as reflecting the number of actual observations driving the evidence on debt overhang. Rather,
each significantly negative point in the kernel reflects a sample where overhang prevails, whose size equals
the bandwidth used in the estimator.
24Our initial sample splits roughly half way between the two income categories.
13
Figure 5 presents the debt Laffer curves as implied by the "double residuals" kernel
estimator. All estimations allow for country effects. Several results are worth pointing
out. First, estimates are rather imprecise. This is inherent to the estimator, and possibly
worsened by relatively small bandwidths.25 Imprecision is the reason why we privilege the
results implied by the "rolling window" lowess kernel. We adopt a prudent approach in this
paper and would rather investigate whether a negative coefficient in a growth regression
does indeed reflect overhang mechanisms, than perhaps mistakenly dismiss an insignificant
coefficient. That said however, the "double residuals" estimates are strikingly close to our
first set of results. In two cases, the coefficient estimates become significantly negative at
virtually identical levels of indebtedness: 60 percent of GDP for debt face value, and 35
percent for debt present value. Given the inefficiency of the estimator, it is remarkable
that such similar conclusions obtain. Abstracting from significance issues, minimum point
estimates for measures based on exports are actually reached for levels of indebtedness not
dissimilar to what is implied by the lowess estimator, i.e. when debt face value reaches
between 170 and 200 percent of exports, and its present value reaches 150 to 170 percent
of exports. In this last case, the estimator's low efficiency makes it impossible to ascertain
whether the threshold is a local extremum.
The kernel approach has three merits. First, it provides some support for a nonlinear
relation between debt and growth, or at least for deleterious effects of debt at high levels.
The evidence is general and not built on specific parametric assumptions. Second, in this
specific case, the approach illustrates the importance of timeinvariant country character
istics in jointly affecting economic growth and indebtedness: a Laffer curve prevails for a
much smaller set of countries, and for much higher debt levels, once country specific features
are accounted for. In other words, the negative relation is partly driven by omitted time in
variant (institutional?) variables which drive debt up but growth down. With fixed effects,
the relation between debt and growth becomes more elusive, perhaps partly because of the
available menu of estimators. This supports a dynamic, withincountry view of debt over
hang, whereby debt buildup opens the door to pathological overhang episodes irrespective
of the quality of the institutional environment. This of course does not mean institutional
quality does not affect the severity of an overhang episode. The next Section asks which
institutional variables appear to belong in that list. Third, the kernel approach provides
an objective criterion to isolate a sample of countries where debt overhang is estimated to
occur in sample. This is crucial for the event study we describe in Section 5.
4.3 The Role of Institutions
In this Section, we seek to identify which institutions tend to jointly explain high debt
and low growth. We focus on a subsample of observations where OLS estimates predict
a significantly negative relation between debt and growth, and augment the specification
with institutional controls. Relevant institutional arrangements are those which affect the
significance of the debtgrowth estimates.26
25The results reported correspond to a bandwidth of 200 observations. We experimented with up to
300 observations, without noticeable change in the bootstrapped standard errors. Similarly, increasing the
number of repetitions beyond 10,000 has little effect on the bands.
26We also experimented with augmenting our kernel with relevant institutional controls. Predictably, the
resulting relation resembled a midpoint between the OLS and the FixedEffects curves.
14
Without fixed effects, the kernel estimates point to a significantly negative coefficient
when debt face value ranges between 30 and 250 percent of GDP.27 We focus on the sub
sample formed by this range of indebtedness, and ask what institutional controls best mimic
the inclusion of fixed effects in Figure 2, i.e. act to weaken the estimated coefficient on debt.
We consider the benchmark OLS regression and augment it with a single institutional
variable at a time. As most of the institutional variables are not available for the early part
of the sample, we take a timeless perspective and use country means for all the institutional
indicators. This approach, consistent with the objective of uncovering specific fixed effects,
is valid under the assumption of a high degree of persistence in institutional variables.
Formally, we estimate
yit+1  yit = Dit + Xit0 + Ii + t + it
where Ii is a timeinvariant institutional variable.
Results are reported on Table 4 for the Kaufmann, Kraay and Mastruzzi (2004) syn
thetic institutional indexes, which include voice and accountability, political rights, cor
ruption, government effectiveness and rule of law. A striking result appears. Government
effectiveness and rule of law knock all significance out of the coefficient on debt. The same
is however not true of any time invariant control, as introducing the other three KK in
stitutional variables leaves the link between debt and growth virtually unchanged. These
findings are consistent with theory. Debt overhang is less likely to occur with more effective
governments and within a better legal and contractual environment. It might still happen,
but will do so at higher levels of indebtedness.
Table 5 reports similar results with ICRG average indexes. An index of bureaucratic
quality is the only variable that affects significantly the debt and growth relationship. Nei
ther the level of democratic rights, nor the occurrence of conflicts or ethnic tensions alter
the negative link between debt and growth.
5 Debt Overhang: An Event Study
We use the nonparametric results to identify countryyears where an overhang episode
is estimated to have happened in sample. In other words, we isolate a panel of country
years where we know the link between debt and growth to be negative. If the mechanisms
underpinning debt overhang are to be observed anywhere in available data, it is bound to
be in this sample where we know debt has withincountry deleterious effects on growth. In
choosing these samples, we opt for prudence. We investigate the possibility of overhang as
soon as one of our estimator points to a negative and significant debtgrowth relation, rather
than dismissing the argument if and when the estimates are not unanimous. In fact, since
the "double residuals" estimator tends to lack efficiency, we focus on the thresholds implied
by the simple rolling window, lest we mistakenly reject overhang phenomena. (Fixed effects)
kernel estimation results imply the following threshold levels for our various measures of
indebtedness:
27This excludes the nine countryobservations year with highest debt to GDP ratio in our sample. Inclusion
of these extreme values results in insignificant coefficient estimates in Figure 1. Chances are these are outliers,
with ratios of debt in excess of 500 percent of GDP, or even sometimes 1,000 as Nicaragua in the late 90s.
15
Ratio Threshold
Total Debt to GDP 60%
Total Debt to Exports 200%
Present Value of Debt to GDP (PVY) 40%
Present Value of Debt to Exports (PVY) 140%
We seek to characterize the dynamic response of investment, policy and the terms of
borrowing before, during and after the onset of an overhang episode. Thus, a definition for
an overhang episode is called for, that distinguishes situations where debt is continuously
high except for one or two exceptionally high growth years for instance, or where debt is
on the whole low, but passes above the threshold for a few years in a row. We arbitrarily
label an overhang episode a sequence of at least eight consecutive years above the threshold,
following five consecutive years below.28 Imposing at least eight years above the threshold
rules out configurations where a high debt to GDP ratio only reflects a decline in real
GDP during a business cycle recession. Requiring five years below screens out countries
permanently located in a high debt trap.29 The notion of an "overhang episode" should
be understood in a hypothetical way, as the objective of this Section is precisely to assess
whether these episodes exhibit a pattern consistent with overhang theories.
We follow a standard procedure.30 First we identify countryyears constituting overhang
episodes. Second, we demean all the variables of interest, controlling for both time and
country averages.31 Third, we average the resulting series across all overhang episodes. The
average path of each variables and the standard error band is displayed on a sixteenyear
window going from t = 5 to +8, where t = 1 corresponds to the onset of the overhang
episode. We track the responses of investment, macroeconomic policy and the terms of
borrowing. The results of the study are presented in Figures 6 to 12, and are discussed over
the next few sections. Each Figure reports the responses of a variable of interest under the
four alternative definitions of an overhang episode, according to either TOD to GDP, TOD
to exports, PVY to GDP or PVY to exports ratios.
5.1 Overhang Countries
We list all overhang countries, as well as the onset date in Appendix A. Using the ratio
of PVY to GDP, we identify 37 episodes of debt overhang in our sample of 87 low and
middle income countries. Of these, 23 are in Africa, 12 in Latin America and only 2 in
Asia.32 The mean real per capita GDP at the inception of each debt overhang episode is
$877, the poorest being Ethiopia ($111) and the richest Venezuela ($3500). The historical
concentration of episodes overrepresents the eighties (24 episodes). This selection partly
derives from our definition of overhang episodes.
28We experimented with imposing five or ten years after the threshold, without substantial changes in the
conclusions.
29We also permit one year spent above or below the threshold during the event years. Thus, four out of
five years spent below the threshold, or seven out of eight years above the threshold is still considered a
relevant event.
30See for instance Henry and Arslanalp (forthcoming).
31For interest rates, we actually demean the spread with US tenyear Treasury Bills.
32The Asian cases are the Philippines (1985) and Syria (1986).
16
Using total debt to GDP, we find an almost identical number of episodes (36), which
also reflects the predominance of African Countries (24). The average per capita income in
this sample is slightly lower ($800). There are 25 countries that exhibit overhang episodes
according to either debt measure.33 The lists are overall similar when considering definitions
based on the ratio of PVY (TOD) to exports, with 38 and 39 countries, respectively. These
include large countries such as Argentina, Brazil or India that do not experience overhang
episodes according to debt to GDP ratios. In addition, geographic and time coverage tend
to be more balanced according to these criteria.
5.2 The Response of Investment
The common hypothesis derived from overhang theories is that countries experience a re
duction in investment once they reach a high enough debt level, either directly through an
anticipation of the costs associated with a potential default or as a response to a deteriora
tion of policies as debtor countries lose incentives to follow sound macro policies. Figure 6
provides some support for this hypothesis.
For all measures of indebtedness, investment follows a clear and significant downward
trend over the fourteen years considered around the event. Investment is (significantly)
above or around its mean in the preoverhang period, but significantly below afterwards.
In addition, in three out of four cases, investment actually builds up prior to the overhang
date, which argues against the possibility that an investment slump actually predates the
overhang and explains the debt buildup. Investment does not fall in earnest until the
overhang date, or a couple of years thereafter. For instance, when indebtedness is measured
by the face value of debt as a proportion of GDP, investment does fall precipitously at
t = 0, as predicted by theory. For alternative measures based on debt face value to exports
or debt present value to GDP, investment actually increases slightly at the overhang date,
before collapsing to its lowest level one or two years later.34
5.3 The Response of Policy
Figures 7 and 8 plot the typical response in government expenditures and inflation before
and during an overhang episode. Here the evidence is more mixed. Government expendi
tures show no systematic pattern, while we do observe a clear and significant increase in
inflation during the overhang period in three out of four cases. The response of inflation is
in all cases occurring at positive values of t: entering the overhang zone tends to be associ
ated with increasing inflation, as it would if price stability became less of a policy priority.
The absence of any response in government expenditures might actually reflect a policy
deterioration, as a sound macroeconomic response to debt buildup might be to attempt
and reduce the level of public expenditures.35
33For these countries, the timing of episodes may change slightly accross the two debt measures but usually
by no more than two years
34The number of observations used to compute Figure 6 can be different from the total number of events,
as investment data are not available everywhere debt or growth data are. This is true throughout the event
study.
35In fact, one of our criterion suggests such a contraction in spending in the overhang years.
17
Figure 9 plots the response of the CPIA index which reflects a (subjective) World Bank
ranking of the overall quality of economic policy. The response in CPIA is close to mirroring
inflation, possibly because it figures prominently in the list of ingredients World Bank
economists use to provide an assessment of the quality of overall policy. In three out of
four cases, policy deteriorates after the overhang. While these results are less unanimous
than those quantifying an investment effect, and also more difficult to interpret, they do
provide some support to the hypothesis that overhang episodes coincide with a noticeable
deterioration in macroeconomic policies.
5.4 The Response of the Terms of Borrowing: Interest Rates and Com
mitments
Figures 10 plots the dynamic response in the interest rate spread, measured by the (de
meaned) difference between local rates as implied by Global Development Finance sources,
and the yield on a tenyear Treasury Bill. Interestingly, the onset of debt overhang appears
to be characterized by a fall in spreads. What is more, in all cases this easing of borrowing
conditions tends to follow a tightening, with interest rates actually increasing prior to the
event threshold. This runs exactly contrary to the theoretical prediction that an overhang
problem appears because of prohibitive borrowing terms. Here, debt seems to become more
concessional as the overhang zone is reached.
A natural explanation for this rests in possible changes in the composition of debt, as if
private investors exit the market, and are replaced by an increasing share of multilateral
agencies lending at concessional rates. But this would suggest debt relief would have hardly
any easing impact on the conditions at which highlyindebted countries can borrow, and
indeed would if anything worsen the terms of borrowing. Figures 11 and 12 plot the average
dynamic path of (the value of) new commitments arising from the private and the official
sectors, respectively. Private commitments fall precipitously in all cases for nonnegative
values of t; what is more, the overhang is preceded by a buildup of private lending, so
that our event study is not merely capturing a trend "rush for the exit" amongst private
investors. In stark contrast, official commitments increase in value after the threshold.36
The change in the composition of debt with the onset of overhang is actually not phrased
out in any theory that we are aware of, but it is presumably what exonerates highly
indebted countries from having to face exorbitant borrowing conditions. We later present
evidence that the servicing of debt actually falls at positive values of t as well, which is also
consistent with the terms of borrowing becoming increasingly concessional in the overhang
zone. Importantly, this also means that debt relief could actually increase debt service and
have the type of crowding out effects on investment that are customarily ascribed to a high
debt burden.
6 Robustness
We check robustness along two important dimensions. First, we verify whether the dynamics
we identify are not caused by global shocks, for instance to interest rates, that would act to
36 Changes in private lending are actually close to one order of magnitude larger than official ones.
18
lower investment, particularly in highly indebted economies forced to dedicate a large share
of their resources to servicing debt. Second, we exclude from our sample all the countries
that experienced sizeable rescheduling in our sample.
6.1 Interest Rate Shocks
If t = 1 tends to correspond on average with a period of increasing interest rates worldwide,
it is possible that the fall in investment that we capture should be a mere manifestation
that increasing debt service makes it particularly hard for highlyindebted economies to
invest. In addition, rising interest rates could also account for falling output (and perhaps
exports), and so explain a sudden jump in debt ratios. And since we use interest rate
spreads, we do not capture a world increase in rates. Here we offer two rebuffals to this
alternative scenario. First, we provide evidence that debt service actually does not rise
during our average event (which is consistent with falling interest rates). Second, we provide
sample splits showing that for similar levels of indebtedness, different countries display
different investment responses .Heterogeneous responses for a given debt level rule out
the possibility that our evidence stems from indiscriminate crowding out of investment,
mechanically caused by high debt service.
Figure 13 plots the time path of debt service. There is no evidence that the debt burden
increases for positive values of t. In two cases, debt service increases in the first years of the
event, but if anything it turns downward in the overhang zone, undoubtedly thanks to the
concessional borrowing conditions we document in the previous section. This is inconsistent
with the view that would ascribe the observed fall in investment to mechanical crowding
out effects.
On Figure 14, we plot the response of investment for two subsamples of events, according
to the enforcement of property rights as measured by Acemoglu, Johnson and Robinson
(2001).37 Arguably, property rights are relevant theoretically, since they may capture the
ability creditors have to monitor and sanction debtors' behavior, and thus they may reflect
the gravity of an overhang problem. And indeed, the fall in investment is clearly subdued
in countries with good enforcement, between two and three times smaller than in the rest of
the overhang sample. While investment tends to fall as well even with good property rights,
the relevance of the actual overhang date is much less clear. The response of investment is
much more severe with poor property rights, which is consistent with theory, and rules out
the possibility that our event study merely captures the chronology of a world recession.
Finally, Figure 15 reports the time path of investment in two samples, characterized by
the World Bank classification of low income countries.38 Consistent with the nonparametric
results, debt overhang appears to set mostly in low income countries, where investment falls
by the largest proportion in most cases.39 An explanation of our results based on a global
recession would have difficulties accounting for these differential responses.
37Low enforcement countries are ones with grades of 1 or 2, whereas high ones take values 3, 4 or 5.
38We have between 21 and 23 low income countries in our sample of events, depending on the criterion
used to identify the event.
39In results available upon request, we also investigate the response of interest rates in both samples. The
split continues to be relevant, with most fall in interest rates occurring in the low income countries.
19
6.2 Rescheduling and Default Episodes
While the bias this would create is ambiguous, it is possible that some of our results are
influenced by the debt crisis of the 1980's, and the associated wave of debt rescheduling
programs. Our data reflect restructuring programs, and debt ratios may be falling from high
levels because of rescheduling agreements, rather than because countries grow themselves
outside of a debt spiral. If anything, this would bias our results against finding evidence for
debt overhang, since we would mistakenly exclude from our sample a country with a debt
history that does not fit our criterion because it goes through rescheduling episodes.
In Figure 16, we omit the period from 1979 to 1984 from our sample, and therefore
characterize our chronology on the basis of overhang episodes outside the range customarily
associated with the debt crisis. The response of investment is virtually unchanged.40 Finally,
in Figure 17, we omit from our study all rescheduling episodes targeting more than 5
percent of debt face value. This is meant to ensure the debt ratios in the sample we end
up focusing on are not perturbed by punctual restructuring agreements. As expected,
investment continues to fall markedly at the overhang date.41 These results suggest that
the presence of rescheduling episodes in our benchmark data tends if anything to obscure
the main results of the paper.42
7 Conclusion
We provide nonparametric evidence supporting a debt Laffer curve among 87 developing
economies. Overhang sets in when the face value of debt reaches 60 percent of GDP or 200
percent of exports, or when the present value of debt reaches 40 percent of GDP or 140
percent of exports. Then, initial debt tends to be associated with subsequently low growth.
These thresholds apply within countries, that is accounting for countryspecific institutional
arrangements. This does not mean institutions do not matter for debt and growth. In
particular, we find that government effectiveness, the rule of law and bureaucratic quality
all act to limit debt buildup while encouraging economic growth. We provide direct tests
of the theoretical conjecture that high debt worsens incentives. We find that investment
collapses in the overhang zone, and the conduct of economic policy deteriorates observably.
However, spreads fall. This is due to official lenders taking over from private creditors, and
extending loans at concessional rates. Borrowing conditions do not become exorbitant with
debt overhang because the high interest rates private creditors would impose on overhung
creditors do not happen in equilibrium. Our results suggest that debt relief might have a
stimulating effect on investment, and possibly on economic policy, but would if anything
result in worse borrowing conditions, and thus possibly in a rising debt burden.
40As are the responses of economic policy, interest rates and commitments, which are available upon
request for the sake of brevity.
41As indeed do rates and private commitments. Official commitments and inflation, in turn, increase
markedly. These results are available upon request.
42In results available upon request, we plot the time paths of GDP growth and the terms of trade, as they
could both affect the dynamics of the debt ratios we examine. There is some evidence that the terms of trade
worsen and growth decelerates somewhat prior to the event date. But both tend to recover quickly and rise
throughout the actual overhang dates. This is consistent with the notion that negative terms of trade shocks
or a recession may actually trigger debt overhang. More importantly it suggests the fall in investment and
economic policy deteriorating tend to happen in a relatively mild macroeconomic environment.
20
References
[1] Acemoglu, D., James A. Robinson and Simon Johnson, 2001, "The Colonial Origins of
Comparative Development: An Empirical Investigation", American Economic Review,
91:13691401.
[2] Aghion P. and P. Bolton,1997, "A Theory of TrickleDown Growth and Development",
Review of Economic Studies, 64:151172
[3] Alesina A. and G. Tabellini, 1989, "External Debt, Capital Flight and Political Risk,"
Journal of International Economics, 27:199220
[4] Barro, R., 1979, "On the Determination of Public Debt," Journal of Political Economy,
87: 94122
[5] Barro, R. and X. SalaIMartin, 1995, Economic Growth, MIT Press.
[6] Clements, B., R. Bhattacharya and T.Q. Nguyen, 2003, "External Debt, Public Invest
ment and Growth in Low Income Countries", IMF Working Paper 03/249.
[7] Cohen, D, 1993, "Low investment and Large LDC Debt in the 1980's", American
Economic Review, 83:437:449
[8] Cohen D., 1995, "Large External debt and (slow) Domestic Growth: a Theoretical
Analysis," Journal of Economic Dynamics and Control 19:11411163
[9] Cohen, D., and J. Sachs, 1986, "Growth and External Debt Under Risk of Debt Re
pudiation", European Economic Review, June, 30:529550
[10] Dikhanov, Y., 2004, "Historical Present Value of Debt in Developing Economies: 1980
2002", manuscript, The World Bank.
[11] Doppelhofer, G., R. Miller and X. SalaIMartin, 2004, "Determinants of LongTerm
Growth: A Bayesian Averaging of Classical Estimates (BACE) Approach", American
Economic Review, 94(4): 81335.
[12] Easterly, W., 2001, "Growth Implosions, Debt Explosions and My Aunt Marylin: Do
Growth Slowdowns Cause Public Debt Crises?", World Bank Policy Research Working
Paper No.2531.
[13] Hauk, W., and R. Wacziarg, 2004, "A Monte Carlo Study of Growth Regressions",
NBER Technical Working Paper #T0296.
[14] Henry, P. and S. Arslanalp, forthcoming, "Is Debt Relief Efficient?", Journal of Fi
nance.
[15] Kaufmann, D. A. Kraay and M. Mastruzzi, 2004, "Governance Matters III: Governance
Indicators for 19962002", World Bank Economic Review.
[16] Krugman, P, 1988, "Financing vs. Forgiving Debt Overhang",Journal of Development
Economics,29, 253268
[17] Levine, R., and D. Renelt, 1992, "A Sensitivity Analysis of CrossCountry Growth
Regressions," American Economic Review, 82(4):94263.
21
[18] Obstfeld M & K. Rogoff,1996, Foundation of International Economics, MIT Press.
[19] Patillo, C., H. Poirson and L. Ricci, 2002, "External Debt and Growth", IMF Working
Paper 02/69.
[20] Patillo, C., H. Poirson and L. Ricci, 2003, "What are the Channels through which
External Debt Affects Growth?", Brooking Trade Forum, 2003:259274.
[21] Piketty, T. 1997, "The Dynamics of the Wealth Distribution and the Interest Rate
with Credit Rationing", Review of Economic Studies, 64:173189
[22] Robinson, P., 1988, "Rootn consistent semiparametric regressions, Econometrica,
56:931954.
[23] Velasco, 1997, "A Model of Endogenous Fiscal Deficit and Delayed Fiscal Reform",
Fiscal Institutions and Fiscal Performance, James Poterba and Jurgen Von Hagen
Eds., University of Chicago Press.
[24] Warner, A, 1992, "Did the Debt Crisis cause the Investment Crisis", Quarterly Journal
of Economics, 1161:1186
[25] Windmeijer, F., 2005, "A finite sample correction for the variance of linear efficient
twostep GMM estimators", Journal of Econometrics, 126(1):2551.
[26] Yatchew, A., 2003, SemiParametric Regressions for the Applied Econometrician, Cam
bridge University Press
22
Table 1 : Present Value of Debt to GDP and Growth: threeyear average panel
1 2 3 4 5 6
[OLS] [OLS] [FE] [FE] [GMM system] [GMM system]
Net Present Value of Debt [PVY]/GDP (avg) 0.606 0.19 0.397
[2.66]*** [0.63] [0.47]
Net Present Value of Debt [PVY]/GDP (initial) 0.484 0.217 0.496
[2.42]** [0.88] [0.65]
Initial GDP per Capita 0.674 0.598 6.981 6.691 1.933 1.913
[3.01]*** [2.68]*** [8.46]*** [8.22]*** [2.04]** [2.11]**
Population Growth 21.165 21.329 1.449 0.965 20.304 25.419
[1.33] [1.31] [0.08] [0.05] [0.58] [0.63]
Secondary Schooling 1.312 1.277 0.021 0.115 4.93 4.996
[4.94]*** [4.79]*** [0.03] [0.16] [3.07]*** [3.21]***
Terms of Trade Growth 0.017 0.016 0.011 0.01 0.038 0.058
[1.28] [1.23] [0.99] [0.84] [1.07] [1.21]
Trade Openness 1.013 0.935 3.643 3.47 1.21 0.987
[2.24]** [2.07]** [4.43]*** [4.23]*** [1.14] [1.11]
Observations 604 604 604 604 604 604
Sargan Pvalue 0.62 0.41
Serial Correlation (second order) Pvalue 0.97 0.82
* significant at 10%; ** significant at 5% ***significant at 1%
Absolute value of t statistics in brackets
Note: [GMM System]: Two System Estimator with Small Sample Windmejer (2005) Robust Correction
[PVE]= PV data from B.Easterly; [PVY]: PV data from Yuri Dikhanov
All regressions include time effects
23
Table 2 : External Debt and Growth: threeyear average panel
1 2 3
[OLS] [FE] [GMM system]
Present Value of Debt [PVY]/Exports (avg) 0.591 0.93 0.397
[2.46]* [2.96]** [0.46]
Present Value of Debt [PVY]/Exports (initial) 0.445 0.325 0.735
[1.96]* [1.12] [0.89]
Present Value of Debt [PVE]/Exports (avg) 0.454 1.468 1.21
[1.55] [3.78]*** [1.33]
Present Value of Debt [PVE]/Exports (initial) 0.175 0.691 0.873
[0.62] [1.99]** [0.89]
Total Outstanding Debt/Exports (avg) 0.562 1.207 0.178
[1.88]* [3.44]*** [0.17]
Total Outstanding Debt/Exports (initial) 0.408 0.593 0.206
[1.45] [1.92]* [0.17]
Present Value of Debt[PVE]/GDP (avg) 0.559 0.4 1.292
[1.95]* [1.08] [1.26]
Present Value of Debt [PVE]/GDP (initial) 0.388 0.01 0.606
[1.48] [0.03] [0.54]
Total Outstanding Debt/GDP (avg) 0.607 0.378 0.364
[2.05]* [1.11] [0.30]
Total Outstanding Debt/GDP (initial) 0.541 0.024 0.296
[2.04]** [0.08] [0.31]
* significant at 10%; ** significant at 5%; *** significant at 1%
Absolute value of z statistics in brackets
Note: [GMM System]: Two System Estimator with Small Sample Windmejer (2005) Robust Correction
[PVE]= PV data from B.Easterly; [PVY]: PV data from Yuri Dikhanov
Set of Control Variable identical to Table 1
All regressions include time effects
24
Table 3 : External Debt and Growth: 5year average panel
1 2 3
[OLS] [FE] [GMM system]
Present Value of Debt [PVY]/GDP (initial) 0.208 0.561 0.323
[0.99] [2.26]* [0.54]
Present Value of Debt [PVE]/GDP (initial) 0.118 0.28 0.215
[0.47] [0.89] [0.39]
Total Outstanding Debt/GDP (initial) 0.269 0.053 0.236
[1.19] [0.18] [0.41]
Present Value of Debt [PVY]/Exports (initial) 0.011 0.267 0.051
[0.05] [1.02] [0.08]
Present Value of Debt [PVE]/Exports (initial) 0.298 0.067 0.147
[1.12] [0.20] [0.25]
Total Outstanding Debt/Exports (initial) 0.007 0.437 0.168
[0.03] [1.42] [0.26]
* significant at 10%; ** significant at 5%; *** significant at 1%
Absolute value of z statistics in brackets
Note: [GMM System]: Two System Estimator with Small Sample Windmejer (2005) Robust Correction
[PVE]= PV data from B.Easterly; [PVY]: PV data from Yuri Dikhanov
Set of Control Variable identical to Table 1
All regressions include time effects
25
Table 4: Kaufman and Kray Controls
1 2 3 4 5 6
[OLS] [OLS] [OLS] [OLS] [OLS] [OLS]
Total Outstanding Debt/GDP (initial) 0.771 0.771 0.839 0.59 0.581 0.792
[1.82]* [1.82]* [2.02]* [1.55] [1.43] [2.02]*
Initial GDP per Capita 0.407 0.526 0.648 0.967 0.911 0.923
[1.63] [1.93]* [2.53]* [3.96]** [3.65]** [3.69]**
Population Growth 62.015 53.03 48.783 24.355 32.314 39.386
[2.19]* [1.78]* [1.73]* [0.95] [1.14] [1.55]
Secondary Schooling 1.027 1.095 1.224 1.034 1.064 1.059
[3.01]** [3.14]** [3.49]** [3.17]** [3.15]** [3.28]**
Trade Openness 0.925 0.929 0.422 0.487 0.261 0.429
[1.51] [1.51] [0.65] [0.85] [0.44] [0.74]
Voice and accountability 0.3
[0.94]
Political Statbility 0.724
[2.92]**
Government Effectiveness 2.383
[6.55]**
Rule of Law 1.896
[4.85]**
Corruption 2.111
[4.88]**
Observations 250 250 250 250 250 250
Robust t statistics in brackets
* significant at 10%; * significant at 5%; ** significant at 1%
26
Table 5: International Country Risk Guide Controls
1 2 3 4 5 6
[OLS] [OLS] [OLS] [OLS] [OLS] [OLS]
Total Outstanding Debt/GDP (initial) 0.912 0.763 0.912 0.887 0.908 0.991
[1.80]* [1.52] [1.79]* [1.76]* [1.80]* [1.92]*
Initial GDP per Capita 0.109 0.283
[0.40] [1.07]
Population Growth 62.641 63.279 62.649 60.607 66.274 64.203
[2.12]* [2.18]* [2.12]* [1.99]* [2.16]* [2.21]*
Secondary Schooling 0.477 0.495 0.476 0.432 0.38 0.572
[1.35] [1.38] [1.36] [1.13] [1.08] [1.56]
Trade Openness 1.162 0.936 1.165 1.234 1.369 1.214
[1.48] [1.17] [1.47] [1.50] [1.74]* [1.52]
bureaucratic quality 0.569
[1.76]*
democratic accountability 0.007
[0.03]
internal conflict 0.052
[0.34]
external conflicts 0.295
[1.73]*
ethnic tensions 0.202
[1.12]
Observations 211 211 211 211 211 211
Robust t statistics in brackets
* significant at 10%; * significant at 5%; ** significant at 1%
27
Figure 1: Kernel Estimates  Debt Face Value
Debt over GDP  OLS Debt over GDP  FE
.5
.5
0
0
5.
5.
1
1
5.1
5.1
2
20 30 40 50 60 70 80 20 30 40 50 60 70 80
DebtMidpoint for Estimation Debt Midpoint for Estimation
Debt over Exports  OLS Debt over Exports  FE
.5
.5
0
0
5.
5.
1
1 5.1
5.1 2
100 150 200 250 300 350 100 150 200 250 300 350
DebtMidpoint for Estimation Debt Midpoint for Estimation
28
Figure 2: Kernel Estimates  Debt Present Value
Debt PV over GDP  OLS Debt PV over GDP  FE
.5 2
0 1
5. 0
1 1
5.1 2
10 15 20 25 30 35 40 45 10 15 20 25 30 35 40 45
DebtMidpoint for Estimation Debt Midpoint for Estimation
Debt PV over Exports  OLS Debt PV over Exports  FE
.5
1
0 .5
5. 0
1 5.
5.1 1
40 60 80 100 120 140 160 180 200 40 60 80 100 120 140 160 180 200
DebtMidpoint for Estimation Debt Midpoint for Estimation
29
Figure 3: Kernel Estimates  Low Income Countries
Debt over GDP  FE Debt over Exports  FE
1 .5
0
0
5.
1
1
5.1
2
40 45 50 55 60 65 70 75 80 200 220 240 260 280 300 320 340 360
DebtMidpoint for Estimation Debt Midpoint for Estimation
Debt PV over GDP  FE Debt PV over Exports  FE
1
1
0
0
1
1
2
2
18 20 22 24 26 28 30 32 34 36 38 90 100 110 120 130 140 150 160 170 180
DebtMidpoint for Estimation Debt Midpoint for Estimation
30
Figure 4: Kernel Estimates  Non Low Income Countries
Debt over GDP  FE Debt over Exports  FE
1 .6
.4
.2
.5
0
2.
0
4.
30 32 34 36 38 40 42 44 46 48 50 120 130 140 150 160 170 180 190 200
DebtMidpoint for Estimation Debt Midpoint for Estimation
Debt PV over GDP  FE Debt PV over Exports  FE
41. 51.
21. 1
1 .5
.8 0
.6 5.
14 16 18 20 22 24 26 28 30 65 70 75 80 85 90 95 100 105 110
DebtMidpoint for Estimation Debt Midpoint for Estimation
31
Figure 5: "Double Residuals" Kernel Estimates
Debt over Exports
Debt over GDP
4
3
3
2
2
1
1
0
25.8 27.0 28.4 30.2 32.1 34.0 37.3 39.6 40.8 42.3 44.0 46.5 49.3 51.6 56.7 59.4 62.6 65.6 68.7 73.0 77.7 0
110.6 121.0 130.5 135.6 142.5 151.1 166.8 182.7 190.7 193.4 203.9 213.7 227.4 240.9 249.9 262.7 282.9 300.5
1 1
2 2
3
3
4
4 Debt PV over Exports
Debt PV over GDP
3
3
2
2
1
1
0 0
11.4 12.4 13.2 14.0 15.3 16.2 17.4 18.0 20.2 22.1 24.5 26.3 27.2 29.0 30.6 33.3 35.5 37.3 39.9 41.9 50.5 54.3 63.2 68.8 72.3 76.8 82.4 88.8 96.8 102.7 109.2 115.4 124.8 132.8 141.7 149.5 163.0 16
1 1
2
2
3
3
4
4
5 32
Figure 6: The Response of Investment
dx px
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1. 1.
2. 2.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1. 1.
2. 2.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
33
Figure 7: The Response of Policy (I): Government Expenditures
dx px
.2 .2
.1 .1
0 0
1. 1.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
.2 .2
.1 .1
0 0
1. 1.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
34
Figure 8: The Response of Policy (II): Inflation
dx px
.5 .5
.3 .3
.1 .1
1. 1.
3. 3.
5. 5.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
.5 .5
.3 .3
.1 .1
1. 1.
3. 3.
5. 5.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
35
Figure 9: The Response of Policy (III): CPIA
dx px
5.1 .15
.1 .1
.05 .05
0 0
50. 50.
1. 1.
51. 51.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
.15 5.1
.1 .1
5.0 .05
0 0
50. 50.
1. 1.
51. 51.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
36
Figure 10: The Response of Borrowing Conditions: Interest Rate
dx px
1.5 1.5
1 1
.5 .5
0 0
5. 5.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
51. 51.
1 1
.5 .5
0 0
5. 5.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
37
Figure 11: The Response of New Commitments: Private Creditors
dx px
2 2
1.5 1.5
1 1
.5 .5
0 0
5. 5.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
2 2
1.5 1.5
1 1
.5 .5
0 0
5. 5.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
38
Figure 12: The Response of New Commitments: Official Creditors
dx px
.6 .6
.4 .4
.2 .2
0 0
2. 2.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
dy py
.6
.4 .6
.4
.2 .2
0 0
2. 2.
5 4 3 2 10 1 2 3 45 6 7 8 910 5 4 3 2 10 1 2 3 45 6 7 8 910
Time Time
39
Figure 13: Robustness (I) Total Debt Service
dx px
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1. 1.
2. 2.
5 4  3  2 10 1 2 3 45 6 7 8 910 5 4  3 2  10 1 2 3 4 5 6 7 8 910
T im e T im e
dy py
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1. 1.
2. 2.
5 4  3  2 10 1 2 3 45 6 7 8 910 5 4  3 2  10 1 2 3 4 5 6 7 8 910
T im e T im e
40
Figure 14: Robustness (II) : Property Rights and The Response of Investment
All Countries Low Property Rights High Property Rights
d x d x d x
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5 4  3  2  10 1 2 3 45 6 7 8 91 0
T im e T im e T i m e
p x p x p x
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5 4  3  2  10 1 2 3 45 6 7 8 91 0
T im e T im e T i m e
d y d y d y
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1.
2. 2. 1.
2.
 5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5 4  3  2  10 1 2 3 45 6 7 8 91 0
T im e T im e T i m e
p y p y p y
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5 4  3  2  10 1 2 3 4 5 6 7 8 91 0  5 4  3  2  10 1 2 3 45 6 7 8 91 0
T im e T im e T i m e
41
Figure 15: Robustness (III) : LowIncome Countries and The Response of Investment
All Countries Low Income Countries Other Countries
d x d x d x
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 45 6 7 8 91 0
T i m e T i m e T i m e
p x p x p x
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 45 6 7 8 91 0
T i m e T i m e T i m e
d y d y d y
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 45 6 7 8 91 0
T im e T im e T i m e
p y p y p y
.6 .6 .6
.5 .5 .5
.4 .4 .4
.3 .3 .3
.2 .2 .2
.1 .1 .1
0 0 0
1. 1. 1.
2. 2. 2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 4 5 6 7 8 91 0  5  4  3  2  10 1 2 3 45 6 7 8 91 0
T im e T im e T i m e
42
Figure 16: Robustness (IV) : Excluding Events between 1979 and 1984 The Response of
Investment
d x p x
.6 .6
.5 .5
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1.2. 1.2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0  5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0
T im e T im e
d y p y
.6 .6
.5 .5
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1.2. 1.
2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0  5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0
T im e T im e
Figure 17: Robustness (V) : Excluding Rescheduling Episodes
d x The Response of Investment p x
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1. 1.
2. 2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0  5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0
T im e T im e
d y p y
.4 .4
.3 .3
.2 .2
.1 .1
0 0
1. 1.
2. 2.
 5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0  5  4  3  2  1 0 1 2 3 4 5 6 7 8 9 1 0
T im e T im e
*countryyear with rescheduling of at least 5% of total external debt
43