Is to Forgive to Forget? Sovereign Risk in the Aftermath of Private or Official Debt Restructurings

Abstract

We examine the link between sovereign defaults and credit risk by distinguishing between commercial and official debt and by taking into account the extent of the final restructuring events, which take place at the end of a default spell. We use a local projection-based approach, combined with propensity score weighting (Jordà and Taylor 2016), to estimate the average treatment effect of the final restructuring on our outcome variables of agency ratings and bond yield spreads. Our results show that the average treatment effect on ratings is negative (and positive for bond spreads) up to seven years following the final restructuring with private creditors, while the opposite holds for official creditors. Furthermore, our results are robust to using a panel analysis, which allows us to investigate the importance of the final haircut size. Specifically, we find that the rating (spread) variation (increase) is larger for cases with deeper haircuts. Therefore, we find evidence that official and private defaults have different costs and then may induce selective defaults.

Supplementary Information

The online version contains supplementary material available at 10.1057/s41308-023-00198-8.

Introduction

In the wake of adverse global shocks such as the Covid-19 pandemic or recent geopolitical crises, a prolonged economic slowdown looms (particularly for developing countries) and hence a series of sovereign debt restructurings are foreseeable in the coming years, including those with official creditors. As recently shown by Horn et al. (2020), official lending is much larger than generally assumed (see Fig. 1), often surpassing total private cross-border capital flows, especially in times of global turmoil when these flows generally shrink.1 Although official debt accounts for a substantial share of total sovereign debt (especially in developing countries) and is expected to rise in the future, there is still relatively little research on the relative treatment of official versus private defaults. Given the historical evidence on the interplay between the two, it is of interest to evaluate how financial markets, in particular bond markets and credit rating agencies, react to different restructuring episodes by distinguishing between private and official events.

Fig. 1.

Share of private and official debt

Despite a renewed interest in the role of official creditors (e.g., Lang et al. 2021; Mitchener and Trebesch 2021; Schlegl et al. 2019), not enough is yet known about the implications of debt restructurings involving official creditors. In particular, Lang et al. (2021) find that the official debt service suspension guaranteed by the Debt Service Suspension Initiative (DSSI) in May 2020 induced a larger decline in borrowing costs for countries eligible for this initiative, compared to similar but ineligible countries.2 Instead, what happens when countries exit a debt crisis and whether the related net present value variation leaves residual stigma in financial markets is a separate issue.

This paper aims to fill this gap by documenting the relationship between sovereign debt restructurings and a country’s credit risk, looking at how borrowing costs vary in the aftermath of a default.3 In particular, we take both an indirect and a direct measure of borrowing costs, namely agency ratings and bond yield spreads. Compared to bond spreads, credit ratings are available for a larger set of countries and are a reliable measure in times of crisis. Moreover, as a consequence of the Covid-19 crisis, credit-rating agencies are likely to be put under the spotlight, as is normally the case during downturns.4 We take restructurings—and not default—as our main explanatory variable. Restructurings typically take place at the end of a renegotiation spell, which may take years after the default occurs.5. Figure 2 describes the timeline we consider for our analysis. Given the scope of the paper, we distinguish between official and private restructurings. More specifically, official restructuring stands for agreements reached with official creditors (in the Paris Club of official creditors).6 In contrast, private restructuring denotes a restructuring deal with external private creditors (foreign banks and bondholders).

Fig. 2.

Crisis timeline

We should emphasize that this paper does not provide an evaluation of the effects of an official restructuring on a country’s overall (both official and private) financing conditions. Instead, our aim is to assess the ability of a defaulting country to tap into private capital markets in the aftermath of a debt crisis. For this reason, we consider two commonly used and relevant variables as outcomes following a default with either private or official creditors, acknowledging that they are both measures that are naturally “skewed” towards capturing the reactions of private creditors. Focusing on private market financing conditions allows us to compare more fairly the effects of private and official restructurings; at the same time, some of the differential effects of official vs private restructurings might arise from the way costs are being measured.

We add to previous works by comparing the rating outcome of official and private restructurings, hence primarily contributing to the empirical literature on official debt. To the best of our knowledge, it is the first time in this literature that the distinction between private and official deals, as well as the occurrence and magnitude of a default, is taken into account in the context of agency ratings and bond spreads. Our results may then provide some insight for the debate on the consequences of debt heterogeneity, which introduces the possibility for governments to operate selective defaults discriminating across investors [e.g., (Erce and Mallucci 2018; D’Erasmo and Mendoza 2021)].7

Sovereign credit ratings can be interpreted as a forward-looking summary indicator of macroeconomic and (often) political conditions, as these affect repayment prospects and tend to be highly correlated with borrowing costs.8 These measures explicitly pertain to a sovereign’s ability (and willingness) to service financial obligations to non-official (commercial) creditors. Hence, they are “biased” in favor of measuring the probability of default on debt owed to private creditors. Understanding how rating agencies and institutional investors evaluate the repayment ability towards official creditors is not straightforward. This depends on how visible official debt risk is and how rating agencies incorporate it into their rating models.

From official documentation, rating agencies seem to evaluate official risk only to the extent to which it can also affect the repayment prospects of government obligations to the private sector, due to the preferred creditor status associated with many official claims (see Dominion Bond Rating Services 2018; Standard and Poor 2019; Rating and Investment Information 2018; Moody's Investors Service 2018; Jianzhong 2019; Capital Intelligence 2018; Fitch Ratings 2019).9 In other words, official debt seems to be generally perceived as problematic, and hence adversely affecting sovereign rating, only to the extent to which arrears to official creditors may indicate growing financial distress and/or lack of willingness to pay, which eventually impacts private repayments as well. Moreover, the Paris Club includes a comparability of treatment clause, which aims to ensure a balanced treatment of the debtor country’s debt by all external creditors.10 Despite this caveat, we still believe that showing the heterogeneous treatment of creditors in the event of default is important as it could help to shed light on what precisely are the costs of default to a sovereign country.

Analyzing 130 final restructurings episodes over the 1990–2018 period, we use both the Adjusted Inverse Propensity-score Weighted (hereafter AIPW) estimator, which consists in a propensity-score-based method combined with local projections (Jordà and Taylor 2016), and standard panel data analysis. Since the choice to enter into a restructuring is contingent on the country’s economic conditions, this methodology allows us to explicitly model and account for the endogeneity of a default episode. The AIPW estimator proceeds in two stages: the first stage estimates a propensity score for each observation in the sample, while the second stage rebalances the sample and estimates the Average Treatment Effect (hereafter ATE) using a conditional local projection. Our results show that commercial and official defaults are associated with different outcomes: while the average treatment effect on ratings is negative (and positive for bond spreads) over the seven years following the final restructuring with private creditors, the opposite holds with official creditors.

In the second part of the paper, we use a panel analysis to explicitly take into account a measure of the severity of the debt crisis, such as the creditors’ loss (or haircut), as in Cruces and Trebesch (2013).11 On the one hand, default involving larger haircuts may entail more severe reputational costs. On the other hand, a channel of debt relief operates in the opposite direction (Krugman 1988). The overall impact is then theoretically ambiguous and remains an empirical question. Using monthly data on average ratings of eight rating agencies and 130 countries, we find that private defaults seem to involve some reputational costs up to seven years since the last agreement, while official defaulters may even benefit from the present value reduction. Using bond spreads as a dependent variable, we confirm the results of Cruces and Trebesch (2013) in the case of private haircuts, while we find that spreads fall for up to seven years after final official deals.

Thus, our main result is that private credit events are more costly than official ones when it comes to sovereign risk. Moreover, the rating (spread) variation (increase) is larger for cases with deeper haircuts, which are both new results.

Even if our results may depend on how rating agencies incorporate official risk into their rating models, they are important because they document that the costs of default vary with the amounts of debt and the type of creditors affected.12 In particular, the higher cost of private defaults is most likely driven by a less creditor-friendly negotiation process, which in turn results in higher economic uncertainty and more severe punishment from the creditors.13 On the other hand, official restructurings arranged within the “Paris Club umbrella” should guarantee a smoother approach in how deals are orchestrated, hence lowering the collateral damage of such an event.14 Moreover, while an official default often occurs without much media coverage, defaulting on private debt is more visible and hence more likely to result in a rating downgrade. Finally, new evidence (Horn et al. 2020; Schlegl et al. 2019) suggests that official lenders typically shoulder the burden for private creditors, which could explain why we find evidence of positive market sentiment in the aftermath of an official restructuring.

Sovereigns, being aware that the consequences of a default depend in important ways on who the defaulted creditors are and what bargaining power each creditor group has, may then decide to prioritize their repayments accordingly. These results are consistent with Schlegl et al. (2019), who find evidence of seniority for multilateral institutions but not for official (Paris Club) bilateral debt.15 What is more, private creditors seem to “free ride” on official ones: they are typically paid first and lose less than bilateral official creditors. Thus, the increase in borrowing costs we detect after private restructurings may explain why private creditors are typically paid first. To summarize, documenting this difference can help shed light on why countries default with certain creditors. As official lending is likely to increase and official debt sustainability is becoming again a topic of concern, understanding the difference between private and official deals will be even more important.16

The empirical literature on sovereign defaults has generally found that default costs are difficult to quantify and short lived.17 Only recently, thanks to novel measurement strategies of a country’s repayment record, persistent effects of default can be precisely detected, bringing the empirical results in line with the effects of a default according to the theory. This paper then contributes to the (empirical) literature on default costs. In particular, to the emerging literature on the characteristics and the economic relevance of debt restructurings, from both a private sector perspective (Asonuma and Trebesch 2016; Asonuma et al. 2021; Asonuma and Joo 2021; Forni et al. 2016; Kuvshinov and Zimmermann 2019; Meyer et al. 2019; Reinhart and Trebesch 2016; Schlegl et al. 2019; Schumacher et al. 2021; Trebesch and Zabel 2017) and an official sector perspective (Cheng et al. 2017, 2018a, 2018; Corsetti et al. 2018; Lang et al. 2021; Marchesi and Masi 2020, 2021; Reinhart and Trebesch 2016).18

The rest of this paper is organized as follows. Section 2 describes the data. Section 3 describes the inverted-propensity score local projection approach, and Sect. 4 provides the AIPW results. Section 5 presents the results of the panel analysis, while Sect. 6 contains a discussion of the findings, as well as some additional evidence on the importance of litigation costs. Section 7 concludes.

Data

In the case of private creditors, we rely on the original data by Asonuma and Trebesch (2016), which provides information on the start and end date of defaults with private creditors. As a proxy for the severity of the debt restructuring, we consider the corresponding present value reduction or “haircut.” We rely on the original dataset by Cruces and Trebesch (2013) for the data on debt restructurings with foreign private creditors (i.e., commercial banks and bondholders). These data exclude debt restructurings that mainly affected domestic creditors. Focusing on foreign creditors makes sense for different reasons, one of which being that we wish to measure a financial market effect that is not heavily influenced by domestic events. This dataset provides a list of 187 distressed sovereign debt restructurings with external banks and bondholders that occurred between 1970 and 2013. It includes information on the amount of debt restructured, the face value reduction and a measure of debt relief (Preferred Haircut HSZ) computed by the authors considering the present value of both old and new debt instruments.

For official debt restructurings, we rely on the Paris Club data (collected by Cheng et al. (2017)), which contains 429 sovereign debt restructurings with the Paris Club between 1956 and 2015.19 Paris Club creditors may provide (official) debt treatments to debtor countries in the form of rescheduling (i.e., debt relief by postponement of debt service payments) or, in the case of concessional rescheduling, reduction in debt service obligations during a defined period (flow treatment) or as of a set date (stock treatment). The restructuring approach of the Paris Club has evolved over time. In the 1980s, negotiations took place on a case-by-case basis and focused on short-term liquidity problems, mostly implementing maturity extensions without nominal debt reduction. During the 1990s and 2000s, especially for low-income countries, restructurings became increasingly concessional, including debt stock cancellations. Specifically, as low-income countries are concerned, the possibility of a partial debt stock cancellation of non-ODA debt was gradually extended from 33% of the eligible debt in 1988 (Toronto Terms) to 50% in 1991 (London Terms) and 66% in 1994 (Naples Terms). In 1996, the World Bank and the IMF implemented the Heavily Indebted Poor Countries (HIPC) Debt Initiative, which was first strengthened in 1999, and, more recently, in 2005, when under the Multilateral Debt Relief Initiative (MDRI) multilateral institutions were encouraged to increase their specific contribution to debt reduction. Debt relief at the completion point under the HIPC Initiative is provided within the HIPC Exit Terms.

Following Cheng et al. (2017), by looking at the terms of treatment (reported in Table 3 of their paper), we were able to compute proxies of the present value reduction for official deals and to compare this value with the corresponding haircut measure in the case of private agreements (or Preferred Haircut HSZ) used by Cruces and Trebesch (2013).20 We should emphasize that due to important differences in the way in which private and official haircuts are computed, the comparison between the two can only be indicative and should be taken with caution.21

Table 3.

ATE on change in average ratings, private and official restructurings

Year	1	2	3	4	5	6	7
Panel A: Private restructurings
AIPW	− 0.67***	− 1.00***	− 1.96***	− 2.25***	− 1.35***	− 0.44*	− 0.76***
	(− 15.43)	(− 11.67)	(− 14.77)	(− 13.51)	(− 6.89)	(− 1.94)	(− 2.98)
Observations	14605	14605	13633	12656	11673	10697	9750
Panel B: Official restructurings
AIPW	0.33***	0.84***	1.12***	1.03***	0.64***	0.44*	0.65**
	(7.71)	(9.84)	(8.45)	(6.19)	(3.29)	(1.95)	(2.52)
Observations	14605	14605	13633	12656	11673	10697	9750

Table shows average treatment effect of private (Panel A) and official restructurings (Panel B) on change in average agency ratings. Standard errors are clustered at the country level. T statistics in parenthesis. The model uses predictors and controls for first and second stage listed in the Online Appendix B and controls for country fixed effects and time-varying heterogeneity. Significance levels: *0.10, **0.05, ***0.01

Our sample includes a maximum of 130 developing countries. Since the data on private debt restructurings are available only up to 2013, in the panel analysis, our year sample ends then. It includes 68 defaulting countries that experienced at least one debt crisis during the sample period and 62 non-defaulters. Among defaulters, 47 countries had both private and official debt restructurings, 14 countries had only official restructurings (through the Paris Club), while 10 countries experienced only private defaults.22 Table A1a, in online Appendix A, shows all countries and years, including a list of debt crisis episodes studied here.

Table 1 shows summary statistics on haircuts for different subperiods in the full sample of 264 restructurings.23 While the average haircut is about 34% over the full sample mean, looking at the three different subperiods, we detect a sizeable increase in this amount over time. Average haircut size is more than double during the last subperiod (2002–2013), as compared to the initial period (1970–1988), and about 20% higher with respect to the intermediate one (1989–2001).

Table 1.

Haircuts over time (in percent)

	Observations	Mean	SD	Min	Max
Private haircut
1970–1988	81	23	53	− 10	93
1989–2001	57	43	26	− 8	92
2002–2013	20	53	31	5	96
Official haircut
1970–1988	1	33	0	33	33
1989–2001	71	58	20	12	100
2002–2013	34	77	28	4	100

As official restructurings are concerned, we find that the average haircut over the entire period is about 64%, much higher than the corresponding average for private.24 Looking at the three different subperiods, we also find an increase in their size over time. The average haircut size during the last subperiod (2002–2013) is more than two times the average haircut implemented during the initial period (1970–1988) and almost double with respect to the average size of the intermediate period (1989–2001).

Figure 3 shows the evolution over time of both private and official haircuts (in percent). As can be seen, while private agreements were more common up to the mid-nineties, Paris Club deals have prevailed in more recent years. Moreover, the haircut size tends to be much higher under official deals.

Fig. 3.

Private and official haircuts over time (in percent)

Table 2 presents summary statistics of private/official haircuts according to a country’s income. As the number of countries is concerned, we find that middle-income countries tend to default more with both types of creditors, while low-income countries tend to benefit from the highest average haircuts.25 Finally, Fig. 4 reports the frequency by size of private and official haircuts (in percent). As can be seen, the distribution of the private haircuts is generally smoother, peaking around 90% during the HIPC initiative. On the other hand, official haircuts follow a multimodal distribution, where the multiple peaks correspond to the different Paris Club treatment terms.

Table 2.

Haircuts by country’s income (in percent)

	High income	Middle income	Low income
Haircut %
Private	27	33	53
Official	100	65	62
# of restructuring countries
Private	7	42	5
Official	1	22	9

Fig. 4.

Frequency by size of private and official haircuts (in percent). Notes: The distribution of the private haircuts is clustered around values of 40% or below, while also picking up parts of the HIPIC initiative with haircuts around 90%. On the other hand, official haircuts follow a multimodal distribution, where the multiple peaks correspond to the different Paris Club treatment terms, such as Toronto (33%), London (50%), Naples (67%), Lyon (80%) and Cologne (90%).

Dependent Variables

Our main proxy to measure a country’s creditworthiness is its sovereign long-term foreign-currency rating. As shown by Reinhart (2002), ratings predict defaults. Hence, this makes them an informative measure of creditworthiness for countries with severe payment problems. Moreover, ratings may also represent a ceiling for the credit rating of private companies from the respective country (Borensztein et al. 2013). Finally, they may also capture the private sector’s ease of access to foreign capital (Gehring and Lang 2020) as well as representing a good proxy for a country’s access to international financial markets.26

We retrieve monthly information via Bloomberg on eight rating agencies: CI, Dagong, DBRS, Fitch, JCR, Moody’s, R&I, and S&P. To analyze the dynamics around default times, we use data at a monthly frequency. We obtained an unbalanced panel, as each agency assigns ratings to a different set of countries over varying time periods. Table A2a, in the online Appendix A, shows the number of observations, countries and years for each agency. The pair-wise correlation between sovereign ratings from the eight credit rating agencies under analysis ranges from 0.869 (between Standard and Poor’s and Dagong Global) to 0.992 (between Fitch and Japan Credit Rating Agency) (see Table A2b). For our empirical analysis, all ratings have been translated to a 21-point scale. This means that we assign the highest value of 21 for an “AAA” rating. “C” and “D” in turn are translated into a value of one. We follow the translation described by Fuchs and Gehring (2017), which is reported in Table A3, in the online Appendix A.

Since the data on credit agencies are available for the full sample of countries only since 1990, our monthly data are organized in an unbalanced panel, including a maximum of 130 developing countries, over the years 1990–2018.27 We take the average rating of all eight agencies as the dependent variable. As a robustness check, in order to account for the possible influence of agency-country time-invariant characteristics (what is called the “home bias” in sovereign rating, see Fuchs and Gehring (2017)), we replicate the analysis using the agency-country pairs of ratings (dyadic data).28 As a further robustness check, we also estimate the effects on the average of only North American agencies (i.e., Moody’s, Fitch, Standard & Poor’s, Dominion Bond Rating Services) and the rating provided by Moody’s, that is the agency that supplies information for the highest number of countries.

Finally, we take the monthly average secondary market bond stripped yield spread from J.P. Morgan’s EMBI Global (EMBIG) for each country as a secondary dependent variable.29 EMBIG spreads have been used to proxy foreign currency borrowing costs of both governments and the private sector in emerging market economies. Due to data availability, this sample is restricted to 47 countries from 1993 to 2018. Among the 47 countries covered by the EMBIG, 23 are countries that restructured their debt, while the other 24 are non-defaulters.30 Table A2c, in the online Appendix A, shows the correlation between the (average) agency rating and bond spread in the reduced sample, while Table A4 presents some summary statistics.

AIPW Estimation

Our baseline estimates rely on a local projection methodology developed by Jordà and Taylor (2016) to account for the endogeneity of a sovereign default. Using a propensity score-based method, combined with local projections (Jordà 2005), we find the average treatment effect of the end of a debt crisis on our outcome variables over a 7-year period.

Calculating the average, unbiased, effect of a sovereign default on ratings would require comparing two contrasting scenarios: one where we can measure the change in ratings following a debt restructuring event, and one where we measure the change in ratings when no such event has occurred, ceteris paribus. If the decision was fully exogenous, we could simply compare the average change in ratings of defaulters versus non-defaulters. However, the choice to enter into a restructuring with either private or official creditors is endogenous to a number of observable and non-observable factors influencing ratings. Furthermore, it is difficult to pinpoint the direction of the effect, as falling ratings are just as likely to signal a default as they are to be a consequence of defaults. With this methodology, we accept the endogeneity of default, and instead attempt to explicitly model and account for it.

The technique was applied to the area of sovereign debt distress by Kushinov and Zimmerman (2019), who estimate the effect of defaults on GDP. Following Asonuma et al. (2016) and Rho and Saenz (2021), we first estimate the probability of being under a debt restructuring with private or official creditors. Then, if the decision is modeled correctly, we can re-balance the sample as if the decisions were taken at random (Jordà and Taylor 2016; Kushinov and Zimmerman 2019).

In the second stage, we use the average rating and the monthly bond spread as the potential outcome variables, as described in the previous section. The AIPW estimator gives us an unbiased estimate for the average treatment effect of a final restructuring on sovereign credit ratings. We define as final restructurings those that were not followed by another restructuring vis-a-vis private or official creditors within (at least) the subsequent four years. Local projections have the attractive property of being free of structural constraints that would instead be imposed on a parallel VAR model. Therefore our ATE response varies nonlinearly over the forecast horizon. In the scope of this paper, we apply this methodology to cases of defaults with private and official creditors in order to compare the differential effects on sovereign credit ratings and bond spreads.

Identification

The methodology is divided into two stages. First, we model the probability of being under a debt restructuring by estimating a propensity score for each element in our sample. We evaluate the use of different discrete choice models. First, a “one-type” logit model with outcome equal to one for either private or official restructuring years. A “two-type” or multinomial logit, where the outcomes are private restructuring years in one case and official ones in the other. Finally a “three-type” model which considers an additional outcome when restructuring years are for both private and officials. As in Asonuma et al. (2016), we screen between the models selecting the one reporting the lowest Akaike information criterion (as reported in Table B3, in the online Appendix B). This leads us to select the first binary model. The propensity score is then the likelihood of a default as predicted by the logit model:

$$\begin{array}{c}\hfill {\mathit{PD}}_{i,t}=\mathrm{\Lambda }(\beta ,Z_{i,t-1})\end{array}$$ 1

where $\mathrm{\Lambda }$ is the logistic distribution function and Z is a vector of macro and political control variables lagged by one year. Our predictor variable set is based on Asonuma et al. (2016) and Rho and Saenz (2021). In particular, our predictors of choice include bank credit to GDP, a banking crises dummy, general gov. gross debt to GDP, a dummy for high inflation, openness, general gov. primary balance, real GDP growth, reserves to GDP, the US effective federal funds rate and the share of past months under default for a given country as well as income group dummies. The standard errors are clustered at the country level (Jordà and Taylor 2016; Kushinov and Zimmerman 2019). The estimated ${\widehat{\mathit{PD}}}_{i,t}$ is then the predicted default probability for country i at time t conditional on our set of predictor variables.

Then, the second stage re-balances in order to create a synthetic sample where the default decision is as good as random. Using our logit estimates, we can estimate the extent of the non-randomness in our sample. Specifically, a highly endogenous default would be predictable based on observables and have a high ${\widehat{\mathit{PD}}}_{i,t}$, while a highly endogenous control country would have a low ${\widehat{\mathit{PD}}}_{i,t}$. We assign the weights $1/{\widehat{\mathit{PD}}}_{i,t}$ to the defaulter (treatment) group and $1/(1-{\widehat{\mathit{PD}}}_{i,t})$ to the non-defaulter (control) group. Given the re-balanced sample, the average treatment effect will then be the difference between the average weighted potential outcomes of defaulters and non-defaulters across our sample.

Table B2, in the online Appendix B, reports the estimated coefficients from the first stage for our baseline model in column 1. Levels of debt are important for predicting the probability of being under a debt restructuring, while reserves to GDP are negatively correlated with a debt crisis but not significant, there is strong evidence of path dependency, and measures of systemic financial risk increase the probability of being under a debt restructuring.

Figure 5 shows the receiver operating characteristic (ROC) curve for the first stage. The ROC curve plots the true positive rates against the false positive rates, and we can interpret the area under the curve (AUC statistic) as the predictive ability of the model. An AUC statistic equal to 0.50 means that the covariates have no predictive ability, while a value equal to 1 corresponds to perfect predictive ability. Our model returns a AUC of 0.88, confirming its predictive ability.

Fig. 5.

AIPW first stage ROC curve. Notes: The receiver operating characteristic (ROC) curve plots the true positive rates against the false positive rates, such that the area under the curve (AUC statistic) indicates the predictive ability of the model. Under the null that the covariates have no predictive ability, the AUC is equal to 0.50, and perfect predictive ability corresponds to an AUC statistic of 1. Our first stage for estimating the probability of being in a debt crisis under our baseline sample of monthly data returns an AUC of 0.88

Figure 6 shows that the matching procedure generates a control group that is similar enough to the treated group. Indeed, most observations of treated and control groups range in the same level of the estimated probability of being under a debt restructuring.

Fig. 6.

Post-matching estimated propensity score. Notes: Boxplots of the estimated propensity scores, after the matching procedure, between countries that are under a debt crisis (treated countries) and countries that are not under a debt crisis (untreated countries), for our baseline monthly dataset.

The potential outcome, which is modeled in the second stage, is the change in the outcome following the end of a restructuring as measured with a local projection (Jordà 2005):

$$\begin{array}{c}\hfill \mathrm{\Delta }{y}_{i,t+h}=\alpha +\beta Z_{i,t-1}+{\gamma }_{j}F{C}_{i,t}+{\delta }_{j}F{R}_{i,t}+{\eta }_{i,h}+{\tau }_t+u_{i,t+h},\text{ }h=1,\dots ,7.\end{array}$$ 2

Here $\mathrm{\Delta }{y}_{i,t+h}$ is the cumulative conditional forecast of the change in outcome from time t to $t+h$, and h is our forecast horizon spanning up to seven years. We consider as outcome both the credit rating and bond spread measured in country i. We take as time t for the treatment the period belonging to the year of the final private (or official) restructuring, therefore estimating the changes in ratings/bond spreads following the end of a debt crisis. Furthermore, because we are interested in evaluating stigma effects, and not just the mechanical co-movements of ratings over the restructuring period, as in Cruces and Trebesch (2013), we exclude the years of the crisis.

$F{C}_{i,t}$ is a dummy equal to one when a country has finalized its final private (or official) restructuring and $F{R}_{i,t}$ denotes the corresponding amount of final private (or official) haircut. We include both private and official restructurings in the same specification, therefore accounting for both types of events given the overlap between the two. $Z_{i,t-1}$ is a vector containing macroeconomic and political control variables, lagged by one year as in the first stage. ${\eta }_i$indicates country fixed effects and ${\tau }_t$ denotes time fixed effects, which allows us to control for common trends. This way, we can also account for global factors that might have influenced the simultaneous dating choice of debt restructuring events (e.g., Baker or Brady plan in the two periods, 1985–1988, or 1989–1994). Finally, $u_{i,t+h}$ is the error term. As in the first stage, we cluster the standard errors at the country level, as the treatment occurs at the country level.

As control variables are concerned, we mainly rely on the specification by Cruces and Trebesch (2013). Therefore, in order to capture the sovereign’s domestic economic performance, we include public debt to GDP, the general government net lending/borrowing, GDP real growth, reserves to GDP, inflation rate (based on consumer prices), current account and the ICRG political risk index. Furthermore, we include the amount of IMF net lending to control for the possibility that the different results, between private and official agreements, may depend on additional financing from the IMF that are associated with official restructurings. While an IMF program is a sine qua non condition for Paris Club creditors to provide relief, not all private restructurings were associated with IMF programs. Table B1, in the online Appendix B, provides a detailed description of each variable and its source.

We run the above regression for each point in horizon h on the re-balanced sample and reach the desired ATE:

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill AT{E}_h& =\frac{1}{n}\sum _i\sum _t\{[\frac{(y_{i,t+h}-y_{i,t})(F{C}_{i,t})}{P{D}_{i,t}}-\frac{(y_{i,t+h}-y_{i,t})(1-F{C}_{i,t})}{1-P{D}_{i,t}}]\hfill \\ \hfill & -\frac{(F{C}_{i,t}-P{D}_{i,t})}{P{D}_{i,t}(1-P{D}_{i,t})}[(1-P{D}_{i,t})m_1^h(Z_{i,t-1},F{R}_{i,t},{\theta }_1^h)+P{D}_{i,t}{m}_0^h(Z_{i,t-1},F{R}_{i,t},{\theta }_{0h})]\}\hfill \end{array}\end{array}$$ 3

Here $y_{i,t+h}-y_{i,t}$ is the estimated cumulative conditional forecast from our local projections, and $F{C}_{i,t}$ is the dummy used to distinguish between defaulters and non-defaulters and $P{D}_{i,t}$ are the estimated propensity scores from the first stage. The first part is a standard inverse score weighted estimator of the ATE. Intuitively, this is like a group-means comparison between defaulters and non-defaulters, with the difference that we correct for allocation bias of the treatment by modeling for it with the propensity score, afterwards inverting it to achieve a random distribution. The second part is an adjustment term consisting of the weighted average of the two regression estimators. The purpose of the adjustment term is to stabilize the estimator as the propensity score gets close to the extremes (0 or 1) and therefore alleviates the need to truncate the weights. Hence $AT{E}_h$ is the average treatment effect of final restructuring computed over the seven-year horizon.

The AIPW estimator has a number of features that make it suitable for calculating the dynamic effects and for the estimation under endogeneity issues. The combination of local projections and propensity score weighting is doubly robust, in that the estimator will be unbiased as long as either of the stages is specified correctly. The underlying idea is that the predictor set in the first stage, and the control set in the second stage, should be expansive enough to explain as much variation in sovereign default decisions as possible.31

Results

We now present the ATE estimates, starting with credit rating as the dependent variable, while in the next section, we use bond spread as the dependent variable.

Credit Ratings

To identify post-crisis episodes, we focus on final restructurings only and we exclude observations during crisis years. This will allow us to compare our outcomes with respect to “normal times,” as agency rating would mechanically improve the assessment of a country, once it exits from default status. Table 3 shows the ATE in the case of private and official restructurings. In the case of private agreements, the estimates indicate a persistent negative effect of a final restructuring on agency ratings. While in the first years, there is an average drop by less than one notch in our scale of agency ratings, by the third year, the drop in agency ratings increases to more than one notch. The effect peaks after four years with a 2.25 drop in agency ratings. Notably, the ATEs are negative and significant for all the seven years in the analysis.

The dynamics of ratings in a post-crisis setting suggest that a private restructuring likely implies a long-lasting, reputational effect on the sovereign defaulter. Clearly, this is influenced by the size of haircuts imposed on creditors, which is the reason why we control for severity of default. The second-stage local projection used in the estimation of this ATE does well in forecasting the change in agency ratings both in the short and long term, with the R-squared going from 18 to 60%.32

As official deals are concerned, we find, on average, an increase in ratings, as the ATE is always positive after the final restructuring. As before, the results are significant for every year considered, and the effect peaks after 3 years, when the expected change in ratings is of 1.12 notch with respect to the base year.

Positive spillover effects seem to dominate following a restructuring with official creditors. The ATEs from Table 3 are plotted in Fig. 7. As can be seen, the dynamic response of agency ratings following the final restructurings for both event types is persistent for all the years of our estimates. Moreover, the differences between private and official defaults persist when we consider dyadic data as opposed to the average rating as well as the average of only North American agencies (i.e., Moody’s, Fitch, Standard & Poor’s, Dominion Bond Rating Services), and the rating provided by Moody’s. Table B6 reports the estimated ATE for the dyadic setting and the alternative outcome variables.33

Fig. 7.

Year-by-year ATE, ratings. Notes: Graphs show AIPW average treatment effect estimates for each h-step ahead forecast of change in average agency ratings following the end of a private restructuring in one case, and official restructuring in another

Bond Spread

This section now presents the results of the AIPW methodology applied to the monthly average secondary market bond stripped yield spread (EMBIG). Given the direct connection between ratings and spread, we expect our results to mirror those on ratings.

Table 4 shows the computed ATEs in the case of both private and official agreements, while Table B5, in the online Appendix B, reports the coefficients from the second stage. As above, the estimated local projection includes controls, country, and time fixed effects, thereby estimating the ATE of the conditional forecast of bond spreads for h-steps ahead. Following the final private restructuring, the ATE is large. One year after the event, we find an increase of about 204 basis points in the spread with respect to the base year. The effect peaks after three years, when the spread is 1393 bp with respect to the base period, after which this change in spread falls. Finding such results for secondary market yields reveals that the aforementioned reputational effects are felt on markets as well as being perceived by credit rating agencies. Even after the end of a debt crisis involving private creditors, investors’ sentiment remains sour for sovereign debt instruments.

Table 4.

ATE on change in bond spread, private and official restructurings

Year	1	2	3	4	5	6	7
Panel A: Private restructurings
AIPW	203.66***	260.76***	1393.04***	140.84**	181.06***	482.68***	168.79***
	(8.14)	(5.49)	(20.11)	(2.11)	(3.71)	(10.77)	(4.42)
Observations	5382	4833	4307	3785	3337	2923	2528
Panel B: Official restructurings
AIPW	− 257.33***	− 337.88***	− 247.44***	− 231.94***	− 329.48***	− 273.35***	− 346.96***
	(− 10.33)	(− 7.13)	(− 3.60)	(− 3.48)	(− 6.76)	(− 6.12)	(− 9.11)
Observations	5382	4833	4307	3785	3337	2923	2528

Table shows average treatment effect of private (Panel A) and official restructurings (Panel B) on change in monthly average country yield spread over US Treasury bonds (EMBIG stripped spread) measured in basis points (bp). Standard errors are clustered at the country level. T statistics in parenthesis. The model uses predictors and controls for first and second stage listed in the Online Appendix B and controls for country fixed effects and time-varying heterogeneity. Significance levels: *0.10, **0.05, ***0.01

On the contrary, following an official restructuring, the change in spread with respect to the base year is constantly falling, where in the first period, the spread falls by a little more than 257 basis points, and then falls consistently over the forecast horizon. The ATEs from Table 4 are plotted in Fig. 8. As we can see, the dynamic response of bond spread following the end of a restructuring episode for both event types is persistent for all the years in our estimates.

Fig. 8.

Year-by-year ATE, bond spreads. Notes: Graphs show AIPW average treatment effect estimates for each h-step ahead forecast of change in monthly average country yield spread over US Treasury bonds (EMBIG stripped spread) measured in basis points (bp) following the end of a private and official restructuring.

These results mirror those obtained when considering the Institutional Investor’s index as the dependent variable and using the Synthetic Control Method instead of the AIPW (see Marchesi and Masi 2020). On the other hand, our results contrast with those of Reinhart and Trebesch (2016), who document a strong increase in average ratings (for emerging markets) when private agreements follow a debt relief. Reinhart and Trebesch also find that despite the substantial relief obtained, ratings in advanced economies did not recover after the war official debt forgiveness of 1934.

In summary, consistently with Schlegl et al. (2019), we find that defaulting on private debt is highly visible and hence more likely (than official crisis) to result in a rating downgrade. On the other hand, official lenders may shoulder the burden for private creditors, which is one explanation for why following official restructurings we find evidence of positive market reaction. We return to these points in greater detail in Sect. 6.

As previously mentioned, different reactions are likely to depend on the different terms of the restructurings with private with respect to official creditors. In particular, the higher cost of large defaults is most likely driven by a less creditor-friendly negotiation process, which in turn results in higher economic uncertainty and more severe punishment from the creditors. On the other hand, official restructurings that are arranged within the “Paris Club umbrella” are supposed to guarantee a relatively smoother approach to the way in which deals are actually orchestrated with respect to private ones, hence lowering even further the collateral damage of a default. In the next two sections, we provide some additional evidence on the importance of restructuring size, as well as on the role of litigation costs in the case of private agreements.

Restructuring Size

The AIPW methodology comes with the advantages of overcoming endogeneity and being free of structural constraints. However, it does not allow us to evaluate whether the effect changes at different levels of restructuring size. As we mentioned in the introduction, restructurings involving higher haircuts may entail more severe reputational costs. On the other hand, the channel of debt relief operates in the opposite direction. Since higher haircuts reduce government’s debt substantially, such debt reduction may allow countries to exit a situation of debt overhang and improve economic prospects, as described by Krugman (1988). To evaluate such an effect, we use ordinary least squares to distinguish the rating variation associated with the default per se from that associated with the amount of the debt affected, i.e., “occurrence” versus “magnitude.” Sections 5.1 and 5.2 present the results obtained by using the dependent variables credit ratings and bond spreads, respectively.

Credit Ratings

As mentioned in Sect. 4.1, in order to consider post-crisis outcomes, we exclude observations during crisis years, and take up to seven years after the final haircut, to capture the existence of persistent effects.34 We estimate a model that includes country fixed effects, period-fixed effects, and cluster the standard errors at the country level. We, therefore, control for unobserved effects that vary at the country and period level, substantially reducing concerns over endogeneity. Ordinary least squares treat the dependent variable as cardinal. This implies that the difference between an “AA” and an “AA+” rating, for example, is the same as between “BB” and “BB+.”35 The regression equation then is:

$$\begin{array}{c}\hfill c_{i,t}=\alpha +\beta Z_{i,t-1}+{\gamma }_{j}F{C}_{i,t-j}+{\delta }_{j}F{R}_{i,t-j}+{\eta }_i+{\tau }_t+u_{i,t},j=1,\dots 3,4\&5,6\&7\end{array}$$ 4

where $c_{i,t}$ represents the credit rating in country i at time t. $F{C}_{i,t-j}$ is a dummy equal to one when a country has finalized its final private (official) restructuring and $F{R}_{i,t-j}$ denotes the corresponding amount of private (official) haircut, and Z is a vector containing the control variables (lagged by one year). ${\eta }_i$ and ${\tau }_t$ denote agency-country pair and time fixed effects, respectively. Finally, $u_{i,t}$ is the error term.

As explained above, the advantage of including both official and private restructurings in the same specification is that it allows us to detect their effects and avoid an omitted variable bias. Moreover, we are also able to distinguish the rating variation associated with the default per se from that associated with the haircut size (“occurrence” versus “magnitude” ). The list of control variables is the same described in Sect. 3.1. Table C1, in the online Appendix C, provides a detailed description of all our variables.

Table 5 presents the results obtained by considering the size of the final private and official haircuts. In columns 1–2 of Table 5, we include haircut size, expressed in percentage points, up to seven years after the final restructuring (with and without control variables, respectively). Column 2 shows that a one percentage point increase in the private haircut size is associated with a decrease of about 0.04 notch in the credit rating in year one after the final haircut. This implies that a final haircut of about 40%, roughly corresponding to the mean for our sample, can be associated with a decrease of about 1.6 notches in year one.

Table 5.

Private and official haircut and average rating, OLS

	(1)	(2)	(3)	(4)	(5)	(6)
Final private haircut (− 1)	− 0.046***	− 0.039***			− 0.013	− 0.021
		(− 2.919)	(− 3.185)		(− 0.548)	(− 1.504)
Final private haircut (− 2)	− 0.035***	− 0.029***			− 0.004	− 0.020
		(− 2.841)	(− 2.726)		(− 0.218)	(− 1.398)
Final private haircut (− 3)	− 0.025***	− 0.018***			− 0.002	− 0.008
		(− 2.985)	(− 2.985)		(− 0.110)	(− 0.751)
Final private haircut (− 4 & 5)	− 0.020***	− 0.017***			− 0.003	− 0.012
		(− 2.661)	(− 2.921)		(− 0.224)	(− 1.229)
Final private haircut (− 6 & 7)	− 0.015**	− 0.013***			− 0.008	− 0.013
		(− 2.274)	(− 2.936)		(− 0.639)	(− 1.529)
Final official haircut (− 1)	− 0.002	0.017***			0.016	0.019*
		(− 0.410)	(3.746)		(0.908)	(1.786)
Final official haircut (− 2)	− 0.003	0.010**			0.021	0.018
		(− 0.615)	(2.328)		(1.387)	(1.558)
Final official haircut (− 3)	− 0.005	0.004			0.017	0.014
		(− 1.034)	(0.912)		(1.373)	(1.354)
Final official haircut (− 4 & 5)	− 0.002	0.007			0.016*	0.016*
		(− 0.596)	(1.468)		(1.800)	(1.878)
Final official haircut (− 6 & 7)	0.001	0.007**			0.015***	0.018***
		(0.318)	(2.341)		(3.503)	(3.447)
Final priv. haircut dummy (− 1)			− 2.392***	− 2.324***	− 1.868**	− 1.301**
			(− 3.733)	(− 3.445)	(− 2.007)	(− 2.060)
Final priv. haircut dummy (− 2)			− 1.801***	− 1.432**	− 1.650**	− 0.528
			(− 3.576)	(− 2.524)	(− 2.182)	(− 0.715)
Final priv. haircut dummy (− 3)			− 1.230***	− 0.868**	− 1.184*	− 0.570
			(− 3.252)	(− 2.565)	(− 1.753)	(− 0.953)
Final priv. haircut dummy (− 4 & 5)			− 0.928***	− 0.721**	− 0.802	− 0.215
			(− 2.853)	(− 2.369)	(− 1.298)	(− 0.409)
Final priv. haircut dummy (− 6 & 7)			− 0.574**	− 0.449*	− 0.273	0.020
			(− 2.077)	(− 1.901)	(− 0.514)	(0.049)
Final off. haircut dummy (− 1)			− 0.274	1.170**	− 1.191	− 0.087
			(− 0.485)	(2.414)	(− 0.910)	(− 0.099)
Final off. haircut dummy (− 2)			− 0.555	0.545	− 1.705	− 0.548
			(− 1.038)	(1.100)	(− 1.549)	(− 0.609)
Final off. haircut dummy (− 3)			− 0.626	0.163	− 1.522*	− 0.714
			(− 1.448)	(0.378)	(− 1.932)	(− 1.005)
Final off. haircut dummy (− 4 & 5)			− 0.501	0.152	− 1.359**	− 0.788
			(− 1.426)	(0.378)	(− 2.469)	(− 1.463)
Final off. haircut dummy (− 6 & 7)			− 0.248	0.175	− 1.050***	− 0.832***
			(− 1.118)	(0.654)	(− 3.698)	(− 3.220)
GDP real growth (− 1)		0.025		0.028		0.028
		(1.152)		(1.247)		(1.267)
Primary balance to GDP (− 1)		0.006		0.005		0.005
		(0.246)		(0.225)		(0.245)
Current account to GDP (− 1)		− 0.018		− 0.018		− 0.017
		(− 1.416)		(− 1.393)		(− 1.325)
Reserves to GDP (− 1)		0.003		0.004		0.003
		(0.314)		(0.367)		(0.262)
Public debt to GDP (− 1)		− 0.041***		− 0.041***		− 0.041***
		(− 3.522)		(− 3.366)		(− 3.445)
Inflation (− 1)		1.352		1.411		1.375
		(0.505)		(0.534)		(0.517)
Political risk (− 1)		0.147***		0.141***		0.144***
		(5.352)		(5.102)		(5.149)
IMF net loans (− 1)		− 0.125		− 0.114		− 0.126
		(− 1.135)		(− 1.039)		(− 1.142)
Constant	12.370***	5.350**	12.441***	5.613**	12.435***	5.415**
	(17.358)	(2.211)	(17.341)	(2.323)	(17.331)	(2.233)
Observations	20,409	13,296	20,409	13,296	20,409	13,296
R− squared	0.121	0.415	0.135	0.410	0.140	0.420
Number of id	124	85	124	85	124	85
Period FE	Yes	Yes	Yes	Yes	Yes	Yes

This table shows coefficients of an unbalanced panel data regression with OLS fixed effects at the country-period level. The dependent variable is the monthly average agency rating. Country and year fixed effects are included. Standard errors are clustered at the country level; t statistics are in parentheses. Significance levels: *0.10, **0.05, ***0.01

In the case of an official agreement, a one percentage point increase in an official haircut is associated with an increase of about 0.02 notch in the credit rating, in year one after the restructuring. Hence, a haircut of about 60% (the mean for our sample) can be associated with an increase of about 1.2 notch, in year one. These results are economically relevant both in the case of private and official restructurings. In turn, in columns 3–4, we include only the dummy indicating the occurrence of the private and official restructuring, while the last two columns contain the full specification (with and without control variables). While all these results are reported for comparison, we mostly base the discussion on the fully specified model of column 6.

To be able to comment these results, however, it should be kept in mind that the coefficients shown in the fully specified model have to be interpreted conditionally, as in any interaction model. The best way to interpret the findings of Table 5 is to look at Fig. 9a, b, which shows the expected variation in agency ratings conditional on the private and official haircut size. In other words, we plot the marginal effect ${\delta }_j{R}_{i,t-j}+{\gamma }_j$ from Eq. 4 above. The different panels correspond to the number of years after the restructuring, and the dotted lines show 90% confidence bands. The effects are calculated from the complete specification (column 6). Aside from an easier interpretation, this joint estimate and the resulting graphs are important because the high correlation between C and R makes it complicated to draw inference about individual effects, but facilitates inference about their sum (see Cruces and Trebesch 2013).36

Fig. 9.

a Expected effect on agency rating for different levels of private haircut. Notes: Each graph shows the marginal effect of private haircut on average agency ratings, for different haircut sizes and at different lag lengths. The dashed lines show 90% confidence bands. The effects are calculated using the coefficients from Table 5, column 6. The rating contraction after a restructuring is statistically significant for levels of haircut at which the upper confidence band is below the zero horizontal line. b Expected effect on agency rating for different levels of official haircut. Notes: Each graph shows the marginal effect of official haircut on agency rating, for different haircut sizes and at different lag lengths. The dashed lines show 90% confidence bands. The effects are calculated using the coefficients from Table 5, column 6. The rating increase after a restructuring is statistically significant for levels of haircut at which the lower confidence band is above the zero horizontal line.

The bottom line of Fig. 9a is that private haircuts are negative and statistically significant for years one to seven after the final agreement. This can be seen because the upper confidence band is always below the zero horizontal line for every haircut size greater than 20% (the mean of this sample being around 40%). The reduction in credit rating associated with haircut size is also economically substantial, especially for years one and two after the final agreement.

In the case of official deals, as in Fig. 9b, the rating increase of a restructuring is statistically significant for levels of haircut at which the lower confidence band is above the zero horizontal line. In years one to three after the final agreements, haircuts greater than 60% (corresponding to the mean of this sample) can be associated with significantly higher ratings. At lag 4 & 5, the rating increase can be significant only for haircuts greater than 80%, while at lag 6 & 7 the effect is never significant.

The results are also robust to a dyadic set-up in which we take into account the dyadic relationship between agency-country pairs, at least as time-invariant factors are concerned (as in column 1 of Table C2, in the online Appendix C).37 The results are also robust to the inclusion of further variables to control for the presence of omitted variable bias, such as the number of years the chief executive has been in office, total population (in log), and per capita GDP (as in column 3 of Table C2).38 The results also hold when using an ordered logit model for the discrete 21-step end-of-month rating, which accounts for the bounded nature of the dependent variable (as in columns 2 and 4 of Table C2). Finally, they are also robust to using as the dependent variable the average of only North American agencies (i.e., Moody’s, Fitch, Standard & Poor’s, Dominion Bond Rating Services) as opposed to only Moodys ’ (columns 5–6 of Table C2). Taken together, this is strong evidence pointing to a significant difference between the effects on credit ratings from private and official restructuring events.

In conclusion, the (private) haircut size seems to involve some reputational costs and the correlation between private restructuring and agency credit rating is negative for years one to seven after the final restructuring episode. These results are consistent with Meyer et al. (2019), who document that the decline in investor returns is much smaller for low-haircut cases (i.e., lower than the median value) and with Asonuma et al. (2021), who find that post-default restructurings are associated with a decline in bank credit, an increase in lending interest rates, and a higher likelihood of triggering a banking crisis (especially in the case of preemptive agreements). Finally, they are also in line with Gennaioli et al. (2014), who show that the spillovers of a default, on domestic and foreign banks, are larger the higher the haircut.

The opposite holds in the case of official agreements, where agency ratings generally improve, and the more so, the larger the haircut.39 As previously mentioned, there is a trade-off concerning the effect of debt reduction: a positive debt relief effect and a negative reputational effect. This evidence then suggests that while for private defaulters the negative reputational effects dominate, for official defaulters positive (debt relief) spillovers prevail (as in Arslanalp and Henry (2005)).40 However, the results in this section should be taken with caution, as identification is difficult and we cannot claim any causal effect but only strong conditional correlations. In the next section, we will consider a more direct measure of borrowing costs (bond spreads) as in Cruces and Trebesch (2013).

Bond Spread

In this section, we estimate equation 4 by taking the bond (EMBIG stripped) spread for each country as the dependent variable. The results are presented in Table 6. As in the previous section, in columns 1–2 of Table 6, we include the final haircut size, expressed in percentage points, up to seven years after the final agreement (with and without control variables). Column 2 shows that a one percentage point increase in haircut is associated with bond spreads that are about 4 bp higher in year one after the final restructuring. Thus, a restructuring with a final haircut size of about 40% (the mean in our sample of private haircuts) can be associated with 160 bp higher in year one. In the case of an official agreement, a one percentage point increase in the final official haircut is associated with a decrease of about 5 bp in the first year after the restructuring. This implies that a restructuring with a final haircut of about 60% (mean in our sample of official haircuts) can be associated with a reduction of almost 240 bp in the first year after the official agreement. These results are hence sizable both in the case of private and official deals.

Table 6.

Private and official haircut and bond spread, OLS

	(1)	(2)	(3)	(4)	(5)	(6)
Final private haircut (− 1)	7.213***	5.483***			6.420**	2.658
		(4.397)	(4.469)		(2.021)	(0.793)
Final private haircut (− 2)	4.745***		3.811***		6.186**	3.648
		(3.975)	(3.468)		(2.544)	(1.382)
Final private haircut (− 3)	4.088***	2.852**			5.354**	5.220**
		(3.467)	(2.439)		(2.359)	(2.148)
Final private haircut (− 4 & 5)	3.186***	2.335***			7.012***	5.175**
		(3.704)	(2.672)		(4.039)	(2.575)
Final private haircut (− 6 & 7)	0.714	0.782			7.787***	5.104***
		(0.791)	(0.873)		(5.066)	(3.170)
Final official haircut (− 1)	− 6.099***	− 3.864***			− 7.074***	− 3.964***
		(− 6.917)	(− 4.101)		(− 5.864)	(− 2.812)
Final official haircut (− 2)	− 7.726***	− 5.110***			− 8.974***	− 6.177***
		(− 6.539)	(− 4.959)		(− 5.102)	(− 3.415)
Final official haircut (− 3)	− 5.824***	− 4.433***			− 4.889**	− 3.252
		(− 3.891)	(− 3.302)		(− 2.414)	(− 1.483)
Final official haircut (− 4 & 5)	− 5.537***	− 4.772***			− 5.235***	− 4.364**
		(− 4.596)	(− 3.973)		(− 3.075)	(− 2.172)
Final official haircut (− 6 & 7)	− 4.320**	− 1.984			− 3.434	− 1.277
		(− 2.461)	(− 0.990)		(− 1.562)	(− 0.473)
Final priv. haircut dummy (− 1)			282.817***	249.980***	16.249	143.651
			(3.449)	(2.958)	(0.107)	(0.752)
Final priv. haircut dummy (− 2)			165.635***	151.506**	− 84.230	5.254
			(2.616)	(2.225)	(− 0.702)	(0.038)
Final priv. haircut dummy (− 3)			132.791**	70.706	− 72.835	− 120.083
			(2.153)	(1.230)	(− 0.654)	(− 1.144)
Final priv. haircut dummy (− 4 & 5)			69.092	49.256	− 192.207**	− 133.768
			(1.463)	(1.030)	(− 2.174)	(− 1.366)
Final priv. haircut dummy (− 6 & 7)			− 47.559	− 19.131	− 313.644***	− 184.746**
			(− 1.179)	(− 0.461)	(− 4.109)	(− 2.274)
Final off. haircut dummy (− 1)			− 402.231***	− 235.900***	95.154*	43.102
			(− 3.239)	(− 2.871)	(1.759)	(0.671)
Final off. haircut dummy (− 2)			− 237.451*	− 195.862**	209.638**	93.816
			(− 1.686)	(− 2.194)	(2.170)	(0.912)
Final off. haircut dummy (− 3)			− 236.661**	− 206.217**	12.591	− 23.764
			(− 2.287)	(− 2.543)	(0.151)	(− 0.221)
Final off. haircut dummy (− 4 & 5)			− 213.239***	− 162.998**	47.605	40.920
			(− 2.895)	(− 2.476)	(0.696)	(0.459)
Final off. haircut dummy (− 6 & 7)			− 200.692**	− 106.753	− 86.382	− 69.284
			(− 2.226)	(− 1.207)	(− 1.317)	(− 0.836)
GDP real growth (− 1)		10.664***		11.582***		10.081***
		(5.659)		(5.772)		(5.064)
Primary balance to GDP (− 1)		− 2.332		− 0.959		− 0.564
		(− 0.665)		(− 0.269)		(− 0.163)
Current account to GDP (− 1)		− 1.301*		− 1.954***		− 1.310
		(− 1.917)		(− 2.635)		(− 1.648)
Reserves to GDP (− 1)		− 0.132		− 0.078		− 0.163
		(− 1.095)		(− 0.644)		(− 1.392)
Public debt to GDP (− 1)		− 15.440***		− 17.510***		− 16.195***
		(− 2.790)		(− 3.210)		(− 2.952)
Inflation (− 1)		− 9.207***		− 10.291***		− 9.077***
		(− 3.061)		(− 3.266)		(− 2.960)
Political risk (− 1)		− 8.369***		− 7.612**		− 6.954**
		(− 2.936)		(− 2.516)		(− 2.402)
IMF net loans (− 1)		29.047		27.687		30.778
		(1.391)		(1.390)		(1.476)
Constant	595.586***	926.318***	597.325***	848.260***	631.377***	851.802***
	(7.247)	(4.661)	(5.959)	(3.948)	(6.905)	(4.184)
Observations	5,369	4,189	5,369	4,189	5,369	4,189
R− squared	0.350	0.447	0.332	0.440	0.367	0.456
Number of panel	47	35	47	35	47	35
Year FE	Yes	Yes	Yes	Yes	Yes	Yes

This table shows coefficients of an unbalanced panel data regression with OLS fixed effects at the country-period level. The dependent variable is the monthly average country yield spread over US Treasury bonds (EMBIG stripped spread) measured in basis points (bp). Country and year fixed effects are included. Standard errors are clustered at the country level, t statistics are in parentheses. Significance levels: *0.10, **0.05, ***0.01?

In columns 3–4, as before, we include only the dummy indicating the occurrence of the final private/official default, while the last two columns contain the full specification, which confirm the relationship between private haircut and subsequent spreads for years four to seven after the final restructuring. In particular, Fig. 10a, b, which are based on the full specification, shows the mean increase in bond spreads associated with the final private restructuring for different levels of haircut and at different lag lengths. The main message of Fig. 10a is that final restructurings with haircuts above 40% can be associated with significantly higher spreads from one to seven years after the restructuring.41 For further illustration, suppose that haircuts increase by one standard deviation (21 pt in this sample); this implies spreads that are 109 bp higher in years 4 and 5 after the final restructuring, and 107 bp higher in years 6 and 7. These results are economically relevant and quite similar to those obtained by Cruces and Trebesch (2013) in the case of private deals.

Fig. 10.

a Expected effect on bond spread for different levels of private haircut. Notes: Each graph shows the marginal effect of private haircut on bond spreads, for different haircut sizes and at different lag lengths. The dashed lines show 90% confidence bands. The effects are calculated using the coefficients from Table 6, column 6. The spread increase after a restructuring is statistically significant for levels of haircut at which the lower confidence band is above the zero horizontal line. b Expected effect on bond spread for different levels of official haircut. Notes: Each graph shows the marginal effect of official haircut on bond spreads, for different haircut sizes and at different lag lengths. The dashed lines show 90% confidence bands. The effects are calculated using the coefficients from Table 6, column 6. The spread decrease after a restructuring is statistically significant for levels of haircut at which the upper confidence band is below the zero horizontal line

Finally, as official restructurings are concerned, Fig. 10b shows that official haircuts above 30% (the mean of this sample being around 60%) can be associated with significantly lower spreads from one to seven years after the final official restructuring (at lag 6 &7 the effect is statistically significant only for smaller haircuts, that between 10 and 50%). More specifically, an increase of official haircut by one standard deviation (39 pt in this sample) implies spreads that are 172 bp higher in years 4 and 5 after the final restructuring (while the coefficient is not significant at lag 6 &7). This result is consistent with the recent findings of Lang et al. (2021), who show that countries benefiting from the Debt Service Suspension Initiative (DSSI) experience a larger decline in bond spread compared to similar but ineligible countries. As the DSSI is a NPV-neutral debt service suspension, we actually find that an official debt relief does not generate stigma, even when it is associated with an NPV reduction.42

In summary, as in Cruces and Trebesch (2013), we find that controlling for both the occurrence and the magnitude of default is crucial to detecting a more lasting link between debt default and borrowing costs. Most importantly, private (official) restructurings are generally associated with lower (higher) ratings and higher (lower) spreads up to seven years after the final restructuring. What is more, the rating (spread) decline (increase) is larger for cases with deeper haircuts. Hence, the trade-off concerning the effects of sovereign debt restructurings seems to be associated with opposite outcomes for private and official defaulters. For the former, negative (reputational) spillovers seem to prevail, while for official defaulters, the positive spillovers of a debt reduction are more important (as in Arslanalp and Henry (2005)).

As rating and spread represent indirect and direct measures for borrowing costs, our results suggest that default costs may vary with the restructuring terms and the relative treatment of official versus private creditors. Our results point to the importance of the way in which debt restructurings are orchestrated, in line with the distinction between “excusable and inexcusable” (Grossman and Van Huyck 1988) and “hard” and “soft” defaults (Trebesch and Zabel 2017). In the next section, we provide some further discussion on this issue, as well as some additional results pointing to the importance of the restructuring terms.

Discussion of the Results

This section provides further discussion on the reasons behind the differences in sovereign risk in the aftermath of a final restructuring with private as opposed to official creditors. As we mentioned in introduction, the higher cost of private defaults is most likely driven by a less creditor-friendly negotiation process, while Paris Club official restructurings are supposed to guarantee a relatively smoother approach in the way in which deals are actually orchestrated. Hence, we focus on three reasons explaining this difference. The first reason is the greater overall visibility of private deals, as opposed to official ones. As previously mentioned, official defaults occur without much media coverage and hence are less likely to result in collateral media damage. On the other hand, defaulting on private debt is highly visible and more likely to result in a rating downgrade. For example, recent highly publicized cases of private restructurings (read Greece 2010 and Argentina v. NML Capital) indicate that private restructurings are considerably more influential for financial markets.

A second reason may depend on the circumstance that official lenders, at least to some extent, shoulder the debt burden for private creditors, as suggested by new evidence from both Horn et al. (2020); Schlegl et al. (2019), which could explain why we find evidence of positive market sentiment in the aftermath of an official restructuring.43 In order to disentangle this possible effect, we first distinguish in our sample of restructuring episodes between “pure” official restructurings (official restructurings occurring independently of others) and “twin” official restructurings involving both private and official creditors. In the case of our bond spreads sample, however, we consider a looser definition of “pure” and also include those official events that were anticipated by a preemptive private restructuring (Asonuma and Trebesch 2016).44 This is for two reasons: on the one hand, we have too few cases of pure official restructurings due to the reduced size of the bond spread sample. On the other hand, we believe our original motivation stands for grouping them, as these two types of events have in common the fact of not being related to an actual private default.

Figures 11 and 12 (and accompanying Table B7 in the online Appendix B) show the ATE for the two different samples. Within Table B7, while panels A and B show ATEs of pure vs. twin official restructurings on changes in agency ratings, panels C and D show average treatment effects of the extended sample of “pure” (i.e., pooling together pure official restructurings with those anticipated by a preemptive private deal) vs. twin restructurings on changes in bond spreads. Interestingly, we find different results in the case of rating and spread. As can be seen, the change in rating, with respect to the base year, is increasing under both types of classifications. However, such variation seems more pronounced in the case of pure restructurings, suggesting that agency ratings evaluate more positively the exit from an official agreement when not overlapping with a private debt restructuring. On the other hand, when considering bond spreads, the decrease is greater when private restructurings are also taking place. Since bond spreads mainly reflect (forward looking) market sentiment of private creditors, some free-riding by private creditors—and/or an implicit subsidy from official bilateral creditors—may explain the more positive market reaction at times of financial turmoil.45

Fig. 11.

Year-by-year ATE of “pure” versus “twin” official restructurings, ratings. Notes: Graphs show AIPW average treatment effect estimates for each h-step ahead forecast of change in agency ratings following the end of either a “pure” official restructuring (i.e., official restructurings not preceded by other events, or at most following a pre-emptive private restructuring) or a “twin” official restructuring (i.e., countries with both a private and official restructuring)

Fig. 12.

Year-by-year ATE of “pure” versus “twin” official restructurings, bond spreads. Notes: Graphs show AIPW average treatment effect estimates for each h-step ahead forecast of change in bond spreads following the end of either a “pure” official restructuring (i.e., official restructurings not preceded by other events, or at most following a pre-emptive private restructuring) or a “twin” official restructuring (i.e., countries with both a private and official restructuring)

The last reason may depend on the different relationship between debtors and official or private creditors. More generally, contrary to official defaults, the relationship between debtors and private creditors may vary a lot across crises (and sometimes even during the same default episode). As illustrated by Trebesch and Zabel (2017), there are striking differences across debt crisis events. On the one hand, there are cases such as Russia during the 1990s, Ecuador 2008–2009 or Argentina 2002–2005, in which governments opted for a unilateral payment moratorium and sometimes even refused to negotiate with their foreign banks and bondholders. On the other hand, there are debt crises that were solved in a consensual manner, with close creditor consultations and little (or no) missed payments (examples may include Ukraine in 1999–2000 and Uruguay in 2003).

What is more, recent evidence shows that disruptive private creditors litigation on external debt has been increasing over time.46 More specifically, as recently shown by Schumacher et al. (2021), the existence of litigation costs has strengthened the hands of private external creditors and raised the cost of default for debtors. These authors find that legal disputes in the US and the UK disrupt government access to international capital markets. To the extent that litigation increases the default costs but involves only external debt held by private creditors, these findings may help us to understand the increase in borrowing costs in the aftermath of a default with private creditors.47

In order to provide some evidence of the importance of litigation costs for sovereign risk, in the next section, we extend our framework to empirically test how litigation costs may increase sovereign risk both during a debt crisis and in the aftermath of a default.

Litigation Costs

Schumacher et al. (2021) provide new data on litigation costs, building, in particular, three measures of litigation. The first indicator is a dummy equal to one in those years in which a sovereign faces one or more pending creditor lawsuits. The second indicator is coded as one if one or more creditors attempt to seize assets of the respective sovereign. Finally, the third indicator is a variable built as a share of litigation to GDP, which is constructed by using the available information on case amounts and then summing the amounts at a country-year level. In this section, we use this last measure and consider either all the data on litigation (to GDP) occurring throughout the default period, or only the final one, that is the case amount associated with the debt crisis exit.

Since litigation costs are available at the country year level, we take as dependent variable the Institutional Investor’s creditworthiness index (Institutional Investor Magazine), which was published twice a year since 1979 (up to 2016) in the March and September issues of the Institutional Investor Magazine.48 We take annual observations (i.e., yearly averages of these bi-annual data) of this variable. This rating is based on information provided by economists and sovereign risk analysts at leading global banks and securities firms. The rating grades each country on a scale from 0 to 100 and is available for 178 countries over the period 1979–2016.49 Unfortunately, we cannot directly control for litigation in our baseline specifications, as litigation can be observed both during the debt crisis and after the final restructuring. As in our baseline model, we exclude observations during crisis years in order to focus on the post-default period.

The sample of countries is the same as that used in the previous sections, while the data now go from 1990 to 2010, as litigation costs are available only up to 2010. More specifically, we estimate the following two equations:

$$\begin{array}{c}\hfill I{R}_{i,t}=\alpha +\beta Z_{i,t-1}+\gamma C_{i,t}+\delta L_{i,t}+\zeta (C_{i,t}\ast L_{i,t})+{\eta }_i+{\tau }_t+u_{i,t}\text{,}\end{array}$$ 5

and:

$$\begin{array}{c}\hfill I{R}_{i,t}=\alpha +\beta Z_{i,t-1}+{\gamma }_{j}F{C}_{i,t-j}+{\lambda }_{j}F{L}_{i,t-j}+{\eta }_i+{\tau }_t+u_{i,t}j=1,\dots 3,4\&5,6\&7\end{array}$$ 6

where $I{R}_{i,t}$ represents the annual Investor rating of a country i, at year t. $C_{i,t}$ is a dummy equal to one for every year of the default spell, while $L_{i,t}$ denotes the size of litigation to GDP. $F{C}_{i,t-j}$ is a dummy equal to one in the last year of the private debt crisis, while $F{L}_{i,t-j}$ denotes the amount of litigation to GDP associated with the end of the default spell. X is a vector containing the control variables (lagged one year). ${\eta }_i,$ and ${\tau }_t$ denote country and year dummies, respectively. The list of controls is the same as the one described in the previous sections.

To investigate more carefully the importance of litigation for sovereign risk, contrary to previous specifications, we now consider both the duration of the debt crisis and up to seven periods after the end of default. Considering the scope of litigation within the crisis period is crucial because it helps quantify the determinants of a negative drop in ratings, while looking at the effect after the crisis retains the same interpretation as in our original analysis (that is, the reputational effects of a default event). Thus, we apply our baseline specification from the start of the debt crisis, and using duration data for private (Asonuma and Trebesch 2016) debt crises. As above, we then include up to seven-year lags of both the occurrence and the magnitude of final litigation.

The results are presented in Table 7. In columns 1–2, we control for both the duration of a private debt crisis, and the amount of litigation to GDP (expressed in percentage points) on private debt, with and without control variables, respectively. In columns 3–4, we include final private litigation to GDP (that is associated with the end of the debt crisis) up to seven years after the end of the default spell, with and without control variables, respectively.

Table 7.

ATE following “pure” or “twin” official restructurings

Year	1	2	3	4	5	6	7
Panel A: “Pure” official restructurings, rating
AIPW	0.34***	1.27***	1.60***	1.86***	1.16***	1.38***	1.44***
	(4.50)	(8.90)	(7.83)	(8.13)	(4.53)	(4.79)	(4.44)
Observations	8945	8945	8369	7793	7206	6622	6046
Panel B: “Twin” official restructurings, rating
AIPW	0.38***	0.98***	1.28***	1.24***	0.82***	0.53**	0.79***
	(8.23)	(11.07)	(9.65)	(7.65)	(4.38)	(2.51)	(3.33)
Observations	12441	12441	11613	10792	9965	9145	8341
Panel C: “Pure” official restructurings, bond spread
AIPW	− 288.61***	− 329.99***	129.16	142.12	− 10.05	− 395.11***	− 647.64***
	(− 5.84)	(− 3.09)	(0.73)	(0.71)	(− 0.06)	(− 4.70)	(− 8.87)
Observations	2757	2451	2162	1880	1648	1420	1203
Panel D: “Twin” official restructurings, bond spread
AIPW	− 356.89***	− 380.52***	− 344.49***	− 338.58***	− 460.72***	− 335.75***	− 485.21***
	(− 10.82)	(− 6.02)	(− 3.86)	(− 4.10)	(− 8.32)	(− 6.82)	(− 11.82)

Notes: Panels A, C, and E show average treatment effects of a “pure” official restructuring (i.e., official restructurings not preceded by other events, or at most following a preemptive private restructuring) on changes in average agency ratings and bond spreads, respectively. Panels B and D show average treatment effects of “twin” restructurings (i.e., countries with both official and private restructurings). T statistics in parenthesis. The model uses predictors and controls for first and s

econd stage listed in the Online Appendix B. Significance levels: *0.10, **0.05, ***0.01

As our variables of interest are concerned, during the debt crisis, we can observe that prolonged private debt crises are associated with a significant contraction of the Investor’s rating of about 4.6 to 5.1 points in the Investor’s rating per year, depending on the specification (all coefficients are significant at the one percent level). As the average duration of a private default is about 8 years in this sample, this result implies that the average GDP loss associated with private default is about 40 points in total. The coefficient of the interaction between the scope of litigation and duration is also negative and significant at the one percent level. The size of the coefficients goes from 2.5 to 3.2, implying a total contraction in the Investor’s rating of about 8 points due to litigation costs.

After the end of the debt crisis, in column 4, we find that both the coefficients denoting the lags of the end of the default spell and those denoting the lags of the litigation size are always negative and generally significant up to five years after the final agreement. The best way to interpret these findings, however, is to look at Fig. 13, which shows the expected variation in Investor rating conditional on the litigation size. The different panels correspond to how many years after the end of the default spell rating is being measured, and the dotted lines show 90% confidence bands. The effects are calculated from the specification of Table 7, column 4. The rating decrease due to litigation costs is statistically significant for levels of litigation at which the lower confidence band is below the zero horizontal line. We can observe a significant decrease in Investor rating for any size of litigation from one to three years after the end of default. After four to five years, the effect is not significant, while after six to seven years ratings start to improve.

Fig. 13.

Expected effect on Investor rating for different levels of litigation scope. Notes: Each graph shows the marginal effect of litigation size on Investor rating for different litigation sizes and at different lag lengths. The dashed lines show 90% confidence bands. The effects are calculated using the coefficients from Table 8, column 4. The rating decrease due to litigation costs is statistically significant for levels of litigation at which the lower confidence band is below the zero horizontal line

In summary, we find that ratings decrease with litigation costs in the aftermath of a default. To the extent that litigation costs characterize deals with private creditors only, these results may explain the different outcomes in terms of sovereign risk of private versus official deals.

Table 8.

Private default, Investor’s index and litigation, OLS

	(1)	(2)	(3)	(4)
End of private default dummy (− 1)			− 2.322**	− 4.284***
			(− 2.070)	(− 4.132)
End of private default dummy (− 2)			− 1.040	− 3.247***
			(− 0.973)	(− 3.384)
End of private default dummy (− 3)			− 0.407	− 2.047**
			(− 0.371)	(− 2.088)
End of private default dummy (− 4 & 5)			0.347	− 0.968
			(0.343)	(− 1.057)
End of private default dummy (− 6 & 7)			1.321	0.855
			(1.313)	(0.993)
Final litigation scope (− 1)			0.771	− 1.327
			(0.353)	(− 0.648)
Final litigation scope (− 2)			3.210*	1.367
			(1.959)	(1.096)
Final litigation scope (− 3)			2.573	0.643
			(1.508)	(0.506)
Final litigation scope (− 4 & 5)			1.287	1.814
			(0.401)	(0.886)
Final litigation scope (− 6 & 7)			1.670	1.049
			(1.028)	(1.008)
Litigation scope	− 0.905*	− 0.015
	(− 1.879)	(− 0.047)
Private default duration	− 5.132***	− 3.109**
	(− 2.892)	(− 2.318)
Private default duration $\times$ Litigation scope	− 3.164***	− 2.740***
	(− 4.367)	(− 4.762)
Constant	31.015***	4.178	28.980***	0.344
	(24.680)	(0.957)	(27.328)	(0.076)
Observations	1897	1323	1830	1357
R-squared	0.598	0.699	0.585	0.715
Number of Countries	98	84	98	77
Controls	No	Yes	No	Yes
Country FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes

This table shows coefficients of an unbalanced panel data OLS regression with fixed effects at the country-year level. Standard errors are clustered at the country level. The dependent variable the Institutional Investor’s creditworthiness index (Institutional Investor Magazine). t statistics are in parentheses. Significance levels: *0.10, **0.05, ***0.01

Conclusions

This paper studies the relationship between sovereign debt default and a country’s creditworthiness by considering the depth of a debt restructuring and distinguishing between commercial and official sovereign debt agreements.

We analyze 130 final restructurings, of 130 countries, over the period 1990–2018, and we consider agency ratings and bond spreads as indirect and direct measures of borrowing costs, respectively. Using both the adjusted inverse propensity-score weighted estimator and a standard panel data analysis, we find a lasting relationship between debt default and credit risk. More specifically, this paper provides evidence of the heterogeneous effect of final private and official restructurings on borrowing costs: (i) private events are more costly (in terms of higher sovereign risk) than official ones; (ii) the rating (spread) variation (increase) is larger for cases with deeper haircuts.

Our results point to the importance of the way in which debt restructurings are orchestrated, in line with the distinction between “hard” and “soft” defaults (Asonuma and Trebesch 2016; Trebesch and Zabel 2017). To the extent that Paris Club deals may represent an example of a “soft” default, this evidence suggests that they are associated with better outcomes in terms of borrowing costs.

To conclude, we find further evidence for the heterogeneity of the economic impact of debt restructurings, confirming that official and private defaults may have different costs and then induce selective defaults. Debtor countries, being aware that the consequences of default depend on who the defaulted creditors are, may then decide to prioritize their repayments accordingly. In particular, these results are consistent with Schlegl et al. (2019), who find that Paris Club creditors bore higher losses (haircuts) than private creditors, over the years 1970–2015. The increase in borrowing costs we detect after private restructurings may then explain why private creditors are paid first and lose less than official ones. As a number of debt restructurings, including those with official creditors, become more likely over the next years, it will become crucial to consider who is going to bear the actual costs of sovereign debt renegotiation.

Supplementary Information

Below is the link to the electronic supplementary material.

Biographies

Silvia Marchesi

is Professor of Economics at the University of Milano Bicocca and research fellow at the Center for European Studies (CeFES) and at the Centro Studi Luca d’Agliano. She holds a PhD in Economics from the Unversity of Warwick. Her research interests include Macroeconomic Development and Sovereign Debt.

Tania Masi

is assistant professor at the “G. d‘Annunzio” University of Chieti-Pescara and research fellow at the Center for European Studies (CefES). She holds a Ph.D. in Economics from the University of Verona. Her research interests are in the fields of Development Economics, Political Economy and Impact evaluation.

Pietro Bomprezzi

is a Ph.D. candidate at the department of Economics at the University of Milan-Bicocca. He has been a visiting Ph.D. at the Strategy, Policy and Review department of the IMF and external collaborator for the Joint Research Centre of the EU Commission. His research interests are in International Finance, Development Economics and International Organizations.

Declarations

Conflict of interest

On behalf of all authors, the corresponding author states that there is no conflict of interest.