I. Introduction

Climate change poses an existential threat to the world economy. With the global average surface temperature rising by 1.1 degrees Celsius since 1880, the frequency and severity of climate shocks—ranging from heatwaves and droughts to hurricanes and coastal flooding—have intensified across the world (Figure 1). Looking ahead, extreme weather events are projected to worsen as the global annual mean temperatures increase by as much as 4 degrees Celsius over the next century (IPCC, 2007; Stern, 2007; IPCC, 2014).² The economic consequences of climate change will be felt across the world, but the extent of potential vulnerability depends on the size and composition of economies, the resilience of institutions and physical infrastructure, and the capacity for adaption and mitigation.

Weather Anomalies Across the World

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Source: NOAA.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Weather Anomalies Across the World

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Source: NOAA.

• Download Figure
• Download figure as PowerPoint slide

Weather Anomalies Across the World

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Source: NOAA.

• Download Figure
• Download figure as PowerPoint slide

While there is a growing body of literature pointing to significant negative effects of climate-related shifts in the physical environment on economic growth (Gallup, Sachs, and Mellinger, 1999; Nordhaus, 2006; Dell, Jones, and Olken, 2012), research on how climate change can affect sovereign default risk is absent. In Cevik and Jalles (2020), we show that climate change vulnerability and resilience have significant effects on government bond yields and spreads, after controlling for conventional macroeconomic factors, especially in developing countries. In this paper, we uncover another layer of empirical information by estimating the impact of climate change on the probability of sovereign default in a group of 116 countries over the period 1995– 2017, taking advantage of a comprehensive dataset on sovereign defaults compiled jointly by the Bank of Canada and the Bank of England and a dataset of climate change vulnerability and resilience developed by the Notre Dame Global Adaptation Institute (ND-GAIN).³

This paper contributes to the literature by looking beyond previous studies and providing robust empirical evidence on the relationship between climate change and sovereign defaults, which impose large welfare losses, as countries are cut off from capital markets and creditors typically suffer large losses. We employ alternative estimation methodologies and control for account conventional determinants of sovereign default risk. Using a large panel of 116 countries over the period 1995–2017, we find that climate change vulnerability and resilience have significant effects on the probability of sovereign debt default, especially among low-income countries. That is, countries with greater vulnerability to climate change face a higher likelihood of debt default compared to more climate resilient countries. We also find that climate change resilience has a similarly significant negative impact on the probability of debt default. That is, countries that are more resilient to climate change have a lower risk of default relative to countries with greater vulnerability to climate change. These findings remain robust to a battery of sensitivity checks, including alternative measures of debt default, model specifications, and estimation methodologies.

The econometric evidence presented in this paper has policy implications, especially for developing countries that are relatively more vulnerable to risks associated with climate change. While climate change is an inevitable reality across the world with increasing temperatures, changing weather patterns, melting glaciers, intensifying storms and rising sea levels, the negative coefficient on climate resilience shows that enhancing structural resilience through cost-effective mitigation and adaptation, strengthening financial resilience through fiscal buffers and insurance schemes, and improving economic diversification and policy management can help cope with the consequences of climate change for public finances and thereby reduce the likelihood of sovereign default.

The remainder of this paper is organized as follows. Section II provides an overview of the related literature. Section III describes the data used in the analysis. Section IV introduces the salient features of our econometric strategy. Section V presents the empirical results, including a series of robustness checks. Finally, Section VI offers concluding remarks with policy implications.

II. A Brief Overview of the Literature

This paper draws from two prolific threads of the literature: determinants of sovereign debt defaults and the macroeconomic impact of climate change. First, most studies on sovereign defaults acknowledge and find strong evidence that macroeconomic factors matter, including the volatility of growth and terms-of-trade, as well as political and institutional variables (Catão and Sutton, 2002; Reinhart et al., 2003; Catão and Kapur, 2004; Manasse and Roubini, 2009; Panizza et al., 2009; Reinhart and Rogoff, 2009; Hilscher and Nosbusch, 2010; Cohen and Valadier, 2015). In particular, Aguiar and Gopinath (2006) and Arellano (2008) link sovereign defaults to exogenous income shocks and asset position changes. Likewise, Sturzenegger and Zettelmeyer (2006) find that debt defaults are triggered by an adverse terms-of-trade shock, an economic recession in creditor countries, and an increase in the cost of borrowing.

Second, there is a growing literature on the economic and financial effects of climate-related shifts in the physical environment.⁴ Starting with Nordhaus (1991; 1992) and Cline (1992), aggregate damage functions have become a mainstay of analyzing the climate-economy nexus. Although identifying the macroeconomic impact of annual variation in climatic conditions remains a challenging empirical task, Gallup, Sachs, and Mellinger (1999), Nordhaus (2006), and Dell, Jones, and Olken (2012) find that higher temperatures result in a significant reduction in economic growth in developing countries. Burke, Hsiang, and Miguel (2015) confirm this finding and conclude that an increase in temperature would have a greater damage in countries that are concentrated in geographic areas with hotter climates. Using expanded datasets, Acevedo and others (2018), Burke and Tanutama (2019) and Kahn and others (2019) show that the long-term macroeconomic impact of weather anomalies is uneven across countries and that economic growth responds nonlinearly to temperature. In a related vein, it is widely documented that climate change by increasing the frequency and severity of natural disasters affects economic development (Loyaza and others, 2012; Noy, 2009; Raddatz, 2009; Skidmore and Toya, 2002; Rasmussen, 2004), reduces the accumulation of human capital (Cuaresma, 2010) and worsens a country´s trade balance (Gassebner and others, 2010).

There is, however, no existing research on the relationship between climate change and sovereign default. The closest line of research concerns the impact of climate change on asset prices. Cevik and Jalles (2020) show that climate change vulnerability and resilience have significant effects on government bond yields and spreads, especially in developing countries. In a similar vein, Bansal, Kiku, and Ochoa (2016) and IMF (2020) find that the risk of climate change—as proxied by temperature rises—has a negative effect on asset valuations, while Bernstein, Gustafson, and Lewis (2019) show that real estate exposed to the physical risk of sea level rise sell at a discount relative to otherwise similar unexposed properties. Likewise, focusing on the U.S., Painter (2020) find that counties more likely to be affected by climate change pay more in underwriting fees and initial yields to issue long-term municipal bonds compared to counties unlikely to be affected by climate change.

III. Data Overview

We use several sources to construct a panel dataset of annual observations covering 116 countries over the period 1995–2017.Economic and financial statistics are assembled from the IMF’s International Financial Statistics (IFS) and World Economic Outlook (WEO) databases, and the World Bank’s World Development Indicators (WDI) database. The data on sovereign defaults is drawn from a comprehensive database developed by the Bank of Canada in partnership with the Bank of England. The database provides estimates of the nominal value of government debt in default, including obligations owed to official and private creditors by country and debt type and aggregated globally (Beers and de Leon-Manlagnit, 2019). Consistent with the literature, the database considers a default when debt service is not made on the due date or within a specified grace period, when payments are not made with the time frame specified under a guarantee, or, absent an outright payment default, when creditors incur material economic losses on the sovereign debt they hold due to the government action. Accordingly, an episode of sovereign default is defined as a binary variable a country has non-missing and non-zero value in a given year in the database.⁵

The main explanatory variables of interest are vulnerability and resilience to climate change as measured by the ND-GAIN indices, which capture a country’s overall susceptibility to climate-related disruptions and capacity to deal with the consequences of climate change, respectively.⁶

The composite indices are based on 45 indicators, of which 36 variables contributing to the vulnerability score and 9 variables constituting the resilience score. Vulnerability refers to “a country’s exposure, sensitivity, and capacity to adapt to the impacts of climate change” and comprise indicators of six life-supporting sectors—food, water, health, ecosystem services, human habitat and infrastructure. Resilience, on the other hand, assesses “a country’s capacity to apply economic investments and convert them to adaptation actions” and covers three areas— economic, governance and social resilience—with nine indicators.⁷

Figure 2 shows the time profile and box-whisker plots for both the climate change vulnerability and resilience indices for the entire sample and income group, respectively. Although the ND-GAIN indices show improvements in climate change vulnerability and resilience in recent years, there is significant heterogeneity across countries. For example, while the mean value of climate change resilience is 33.7, it varies between a minimum of 0.2 and a maximum of 71.3. It is also clear from the data that advanced economies are much less vulnerable to climate change than developing countries. It is important to highlight that the time-series variation in the ND-GAIN indices reflect the changes in countries’ levels of vulnerability and resilience (which are not necessarily forward-looking), not from the changes in the projected vulnerability and resilience to physical risks associated with climate change.

Climate Change Vulnerability and Resilience

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Source: ND-GAIN; Bloomberg; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Climate Change Vulnerability and Resilience

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Source: ND-GAIN; Bloomberg; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Climate Change Vulnerability and Resilience

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Source: ND-GAIN; Bloomberg; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Aggregate pictures, however, hide marked heterogeneity across countries that should not go unnoticed. Figure 3a compares the climate change vulnerability index in 1995 with that in 2017. We can see that Canada, Australia, some parts of South America and Asia improved the situation, while Sub-Saharan Africa remained relatively unchanged over the past two decades. In Figure 3b, we do the same for the climate change resilience index. It is interesting to observe a slight deterioration in the case of the US and in some countries in Sub-Saharan Africa, but improvements in Europe, Russia and other parts of South East Asia as well as South America.

Climate Change Vulnerability Across the World in 1995 vs 2017

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Note: color scheme for less (blue) to more vulnerable to climate change (red).Source: ND-GAIN; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Climate Change Vulnerability Across the World in 1995 vs 2017

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Note: color scheme for less (blue) to more vulnerable to climate change (red).Source: ND-GAIN; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Climate Change Vulnerability Across the World in 1995 vs 2017

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Note: color scheme for less (blue) to more vulnerable to climate change (red).Source: ND-GAIN; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Climate Change Resilience Across the World in 1995 vs 2017

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Note: color scheme for less (red) to more resilient to climate change (blue).Source: ND-GAIN; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Climate Change Resilience Across the World in 1995 vs 2017

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Note: color scheme for less (red) to more resilient to climate change (blue).Source: ND-GAIN; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Climate Change Resilience Across the World in 1995 vs 2017

Citation: IMF Working Papers 2020, 231; 10.5089/9781513560403.001.A001

Note: color scheme for less (red) to more resilient to climate change (blue).Source: ND-GAIN; authors’ calculations.

• Download Figure
• Download figure as PowerPoint slide

Following the literature, we introduce a set of control variables following recent papers studying the determinants of sovereign default, including the level and growth rate of real GDP, consumer price inflation, government debt as a share of GDP, financial development as measured by domestic credit to the private sector as a share of GDP, the terms-of-trade index, the real effective exchange rate (REER) index.⁸ There is a significant degree of dispersion across countries in terms of climate change vulnerability and resilience as well as macroeconomic, fiscal and financial performance, as presented in Appendix Table A1.

IV. Empirical Strategy and Results

We empirically investigate the long-term impact of climate change on sovereign defaults. Based on a bivariate characterization, we estimate logistic regressions to assess the probability of debt default and test the impact of climate change, while controlling for conventional determinants of sovereign defaults identified in the literature. In the binary choice model, the dependent variable takes the value 1 in a given year if a country is in default, and the value of 0 otherwise (non-occurrence of default). Accordingly, we estimate the following baseline model:

in which βs are the parameters to be estimated, Climate denotes the measures of climate change vulnerability and resilience, X is a vector of other exogenous variables, and Φ(⋅) is the logistic function. Using disaggregated indices rather than the overall ND-GAIN index allows us to estimate the separate effects of climate change vulnerability and resilience on the risk of sovereign default. The structural model associated with model (1) can be written as:

where the dependent variable, $D_{it}^{*}$, is a binary variable taking the value of 1 if there is sovereign default in country i and time t; Climate_it-1 denotes the measures of climate change vulnerability and resilience country i and time t-1; X_i,t-1 is a set of control variables including government debt, real GDP per capita, real GDP growth, consumer price inflation, a measure of financial development, the terms-of-trade index, and the REER. All independent variables are lagged by one year to avoid the problems of simultaneity and endogeneity. The η_i and μ_t coefficients denote the time-invariant country-specific effects and the time effects controlling for common shocks that may affect the probability of sovereign default across all countries in a given year, respectively. s_it is an idiosyncratic error term that satisfies the standard assumptions of zero mean and constant variance. To account for possible heteroskedasticity, robust standard errors are clustered at the country level.

As a baseline, the logistic model of sovereign default is estimated for the full sample and separately for emerging market economies and low-income countries during the period 1995–2017. Table 1 presents the results for climate change vulnerability, which is found to have a statistically and economically significant effect on the probability of sovereign debt default. The coefficient on climate change vulnerability ranges between 0.122 and 0.407 depending on the sample coverage, but always remaining positive and statistically significant. This means that greater vulnerability to climate change is associated with higher risk of sovereign default. According to our benchmark model based on the full sample, a one percentage point increase in climate change vulnerability is associated with an increase of 0.41 percent in the probability of debt default.

^{Table 1.}

Climate Change Vulnerability and Sovereign Defaults—Logit Model

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

^{Table 1.}

Climate Change Vulnerability and Sovereign Defaults—Logit Model

Specification	1	2	3
Dependent Variable	default	default	default
Sample	All	EME	LIC
Climate vulnerability	0.4071** (0.173)	0.0279* (0.016)	0.1223*** (0.021)
Real GDP growth	-0.0326 (0.025)	-0.0013 (0.008)	-0.0417 (0.027)
Real GDP per capita (t-1)	-3.0271*** (0.884)	-0.1468*** (0.025)	0.0457 (0.057)
Inflation (t-1)	-0.0118** (0.006)	-0.0001 (0.004)	0.0436** (0.018)
Government debt (t-1)	0.0365*** (0.008)	0.0188*** (0.002)	0.0041 (0.003)
Financial development (t-1)	0.0219*** (0.008)	-0.0209*** (0.003)	-0.0111** (0.005)
Terms-of-trade (t-1)	-1.1646* (0.700)	-0.4710 (0.296)	-0.5482 (0.379)
REER (t-1)	-2.3086*** (0.697)	0.2474 (0.402)	-0.6166 (0.733)
Number of countries	116	61	55
Number of observations	952	1,026	875
Pseudo R-squared	0.4260	0.1390	0.3220

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

View Table

^{Table 1.}

Climate Change Vulnerability and Sovereign Defaults—Logit Model

Specification	1	2	3
Dependent Variable	default	default	default
Sample	All	EME	LIC
Climate vulnerability	0.4071** (0.173)	0.0279* (0.016)	0.1223*** (0.021)
Real GDP growth	-0.0326 (0.025)	-0.0013 (0.008)	-0.0417 (0.027)
Real GDP per capita (t-1)	-3.0271*** (0.884)	-0.1468*** (0.025)	0.0457 (0.057)
Inflation (t-1)	-0.0118** (0.006)	-0.0001 (0.004)	0.0436** (0.018)
Government debt (t-1)	0.0365*** (0.008)	0.0188*** (0.002)	0.0041 (0.003)
Financial development (t-1)	0.0219*** (0.008)	-0.0209*** (0.003)	-0.0111** (0.005)
Terms-of-trade (t-1)	-1.1646* (0.700)	-0.4710 (0.296)	-0.5482 (0.379)
REER (t-1)	-2.3086*** (0.697)	0.2474 (0.402)	-0.6166 (0.733)
Number of countries	116	61	55
Number of observations	952	1,026	875
Pseudo R-squared	0.4260	0.1390	0.3220

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

The estimation results for climate change resilience, presented in Table 2, confirm that investing in adaptation and mitigation helps lower the probability of sovereign default in our sample of countries during the period 1995–2017. The coefficient on climate change resilience has a statistically and economically significant coefficient, ranging between -0.324 and -0.402 according to the model specification. In other words, countries that are more resilient to climate change face a lower risk of sovereign default relative to countries with greater vulnerability to risks associated with climate change. According to our benchmark specification, a one percentage point increase in climate change resilience is associated with a decrease of 0.33 percent in the likelihood of debt default.

^{Table 2.}

Climate Change Resilience and Sovereign Defaults—Logit Model

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

^{Table 2.}

Climate Change Resilience and Sovereign Defaults—Logit Model

Specification	1	2	3
Dependent Variable	default	default	default
Sample	all	EME	LIC
Climate resilience	-0.3327*** (0.072)	-0.3240*** (0.077)	-0.4017* (0.208)
Real GDP growth	-0.0307 (0.025)	-0.0383 (0.026)	-0.0224 (0.083)
Real GDP per capita (t-1)	-3.2212*** (0.755)	-3.5713*** (0.923)	-1.9671 (1.566)
Inflation (t-1)	-0.0087* (0.005)	-0.0102** (0.005)	-0.0337 (0.032)
Government debt (t-1)	0.0366*** (0.008)	0.0323*** (0.011)	0.0423*** (0.013)
Financial development (t-1)	0.0364*** (0.009)	0.0352*** (0.012)	0.0379** (0.016)
Terms-of-trade (t-1)	-1.2074* (0.707)	-0.9742 (0.809)	-2.5955** (1.203)
REER (t-1)	-1.8861*** (0.696)	-2.7440*** (0.895)	-0.0229 (1.071)
Number of countries	116	61	55
Number of observations	952	694	258
Pseudo R-squared	0.4398	0.4165	0.5095

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

View Table

^{Table 2.}

Climate Change Resilience and Sovereign Defaults—Logit Model

Specification	1	2	3
Dependent Variable	default	default	default
Sample	all	EME	LIC
Climate resilience	-0.3327*** (0.072)	-0.3240*** (0.077)	-0.4017* (0.208)
Real GDP growth	-0.0307 (0.025)	-0.0383 (0.026)	-0.0224 (0.083)
Real GDP per capita (t-1)	-3.2212*** (0.755)	-3.5713*** (0.923)	-1.9671 (1.566)
Inflation (t-1)	-0.0087* (0.005)	-0.0102** (0.005)	-0.0337 (0.032)
Government debt (t-1)	0.0366*** (0.008)	0.0323*** (0.011)	0.0423*** (0.013)
Financial development (t-1)	0.0364*** (0.009)	0.0352*** (0.012)	0.0379** (0.016)
Terms-of-trade (t-1)	-1.2074* (0.707)	-0.9742 (0.809)	-2.5955** (1.203)
REER (t-1)	-1.8861*** (0.696)	-2.7440*** (0.895)	-0.0229 (1.071)
Number of countries	116	61	55
Number of observations	952	694	258
Pseudo R-squared	0.4398	0.4165	0.5095

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

These effects of climate change vulnerability and resilience remain robust to the inclusion of other control variables, including foreign reserves, external debt, budget balance, or political and institutional factors as suggested by Kohlscheen (2007), for which we obtain coefficients that are as expected and broadly comparable to the findings in previous studies looking at determinants of sovereign defaults (not shown but available upon request). Nevertheless, we perform several sensitivity checks to validate the robustness of our baseline empirical results.

First, the probability of sovereign default is estimated using the alternative probit model. These results, presented in Table 3 both for climate change vulnerability and resilience, confirm the impact of climate change on sovereign default risk. However, these results should be read with caution as probit models do not render themselves well to the fixed-effects treatment due to the incidental parameter problem (Wooldridge, 2002).

^{Table 3.}

Climate Change and Sovereign Defaults—Probit Model

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

^{Table 3.}

Climate Change and Sovereign Defaults—Probit Model

Specification	1	2	3	4	5	6
Dependent Variable	default	default	default	default	default	default
Sample	all	all	EME	EME	LIC	LIC
Climate vulnerability	0.0568*** (0.005)		0.0216** (0.009)		0.0637*** (0.010)
Climate resilience		-0.0281*** (0.004)		-0.0169*** (0.005)		-0.0270*** (0.005)
Real GDP growth	-0.0052 (0.005)	-0.0029 (0.005)	-0.0009 (0.005)	-0.0012 (0.005)	-0.0147 (0.012)	-0.0202 (0.014)
Real GDP per capita (t-1)	-0.0594*** (0.011)	-0.0557*** (0.012)	-0.0843*** (0.014)	-0.0862*** (0.015)	0.0537* (0.028)	0.0291 (0.028)
Inflation (t-1)	0.0005 (0.002)	-0.0009 (0.002)	0.0002 (0.002)	-0.0002 (0.002)	0.0259*** (0.009)	0.0260*** (0.009)
Government debt (t-1)	0.0075*** (0.001)	0.0084*** (0.001)	0.0118*** (0.001)	0.0125*** (0.001)	0.0012 (0.001)	0.0016 (0.001)
Financial development (t-1)	-0.0107*** (0.001)	-0.0122*** (0.002)	-0.0117*** (0.002)	-0.0102*** (0.002)	-0.0128*** (0.003)	-0.0088*** (0.003)
Terms-of-trade (t-1)	0.1049 (0.123)	0.2935** (0.122)	0.3388* (0.175)	0.4068** (0.182)	-0.0709 (0.176)	-0.2497 (0.195)
REER (t-1)	0.1194 (0.184)	0.2876 (0.197)	0.1815 (0.226)	0.1741 (0.229)	0.0146 (0.373)	-0.1239 (0.344)
Number of countries	116	116	61	61	55	55
Number of observations	1,901	1,901	1,026	1,026	875	875
Pseudo R-squared	0.2143	0.1818	0.1298	0.1313	0.2859	0.3131

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

View Table

^{Table 3.}

Climate Change and Sovereign Defaults—Probit Model

Specification	1	2	3	4	5	6
Dependent Variable	default	default	default	default	default	default
Sample	all	all	EME	EME	LIC	LIC
Climate vulnerability	0.0568*** (0.005)		0.0216** (0.009)		0.0637*** (0.010)
Climate resilience		-0.0281*** (0.004)		-0.0169*** (0.005)		-0.0270*** (0.005)
Real GDP growth	-0.0052 (0.005)	-0.0029 (0.005)	-0.0009 (0.005)	-0.0012 (0.005)	-0.0147 (0.012)	-0.0202 (0.014)
Real GDP per capita (t-1)	-0.0594*** (0.011)	-0.0557*** (0.012)	-0.0843*** (0.014)	-0.0862*** (0.015)	0.0537* (0.028)	0.0291 (0.028)
Inflation (t-1)	0.0005 (0.002)	-0.0009 (0.002)	0.0002 (0.002)	-0.0002 (0.002)	0.0259*** (0.009)	0.0260*** (0.009)
Government debt (t-1)	0.0075*** (0.001)	0.0084*** (0.001)	0.0118*** (0.001)	0.0125*** (0.001)	0.0012 (0.001)	0.0016 (0.001)
Financial development (t-1)	-0.0107*** (0.001)	-0.0122*** (0.002)	-0.0117*** (0.002)	-0.0102*** (0.002)	-0.0128*** (0.003)	-0.0088*** (0.003)
Terms-of-trade (t-1)	0.1049 (0.123)	0.2935** (0.122)	0.3388* (0.175)	0.4068** (0.182)	-0.0709 (0.176)	-0.2497 (0.195)
REER (t-1)	0.1194 (0.184)	0.2876 (0.197)	0.1815 (0.226)	0.1741 (0.229)	0.0146 (0.373)	-0.1239 (0.344)
Number of countries	116	116	61	61	55	55
Number of observations	1,901	1,901	1,026	1,026	875	875
Pseudo R-squared	0.2143	0.1818	0.1298	0.1313	0.2859	0.3131

Note: Robust standard errors reported in brackets. A constant is included in each regression, but not shown in the table. *** p<0.01, ** p<0.05, * p<0.1

Second, we estimate the logistic model for the probability of sovereign default on different types of debt and creditors, including foreign currency-denominated loans and bonds and domestic currency-denominated debt, and debt extended by private creditors, multilateral institutions and the Paris Club, an informal group of 22 bilateral official lenders. These results, presented in Appendix Tables A2-A3, are broadly consistent with the baseline findings, although a small number of rare events reduces statistical significance.

For the reason mentioned earlier, our final robustness exercise consists in estimating a rare-events logit (or relogit) model. In a logistic regression, the Maximum Likelihood estimates are consistent but only asymptotically unbiased. The basic problem is having a number of units (default episodes) in a panel that has no events. This means that the country-specific indicators corresponding to the all-zero countries perfectly predict the zeroes in the outcome variable (King, 2001). This is a well-known phenomenon in the statistical literature (for an overview see Gao and Shen, 2007). The simplest way of dealing with this problem is decreasing the rareness of the event of interest, which is not possible in our setting. Alternatively, we use the bias correction method—the relogit estimator—proposed by King and Zeng (2001).⁹ The relogit estimator for dichotomous dependent variables provides a lower mean square error in the presence of rare events and can be defined as follows:

with i = 1, …, N; t = 1, …, T, where $\mathrm{\Phi }(\cdot )=\frac{1}{1+e^{-1(Z{\prime }_{it}\vartheta )}}=\frac{1}{1+e^{-(\gamma +{\beta }_{1}C\lim ate+{\beta }_{2}X)\prime }}$ γ,β₁ and β₂ are the vectors of the parameters to be estimated, andbold> Φ(·) is the logistic function.

The parameters can be estimated by maximum likelihood, and the variance of the estimated coefficients can be expressed as $Var\left(\hat{\vartheta }\right)={(Z\prime VZ)}^{-1}$, where V is a diagonal matrix, with diagonal entries equal to Φ(·) · [1 – Φ(·). In the case of rare events, Φ(·) will be generally small. However, as pointed out by King and Zeng (1999a, 1999b, 2001), the estimates of Φ(·) and Φ(·) · [1 – ·] among observations that include rare events (in our case, for which D = 1) will be typically larger than those among observations that do not include rare events (i.e., for which D = 0). Consequently, their contribution to the variance will be smaller, rendering additional ‘rare’ events more informative than additional ‘frequent’ events. Therefore, we follow King and Zeng (1999a, 1999b) and correct for the small sample and rare events biases and estimate a relogit model where the sampling design is random or conditional on Z’_it. These results, presented in Appendix Table A4, validate the baseline findings and remain robust to the inclusion of additional control variables such as government revenues and exports.

V. Conclusion

Climate change is an existential risk. While there is a growing body of literature on the economic consequences of climate change, research on the link between climate change and sovereign default is nonexistent. We fill this gap in the literature by estimating the impact of climate change vulnerability and resilience on the probability of sovereign debt default. Using a large panel of 116 countries over the period 1995–2017, we find that climate change vulnerability and resilience have significant effects on the probability of sovereign debt default, especially among low-income countries. That is, countries with greater vulnerability to climate change face a higher likelihood of debt default compared to more climate resilient countries. Our empirical results also indicate that climate change resilience has a similarly significant negative impact on the probability of sovereign debt default. That is, countries that are more resilient to climate change have a lower risk of debt default relative to countries with greater vulnerability to climate change. These findings remain robust to a battery of sensitivity checks, including alternative measures of debt default, model specifications, and estimation methodologies.

The econometric evidence presented in this paper has policy implications, especially for developing countries that are relatively more vulnerable to risks associated with climate change. While climate change is an inevitable reality across the world with increasing temperatures, changing weather patterns, melting glaciers, intensifying storms and rising sea levels, the negative coefficient on climate resilience shows that enhancing structural resilience through cost-effective mitigation and adaptation, strengthening financial resilience through fiscal buffers and insurance schemes, and improving economic diversification and policy management can help cope with the consequences of climate change for public finances and thereby reduce the likelihood of sovereign default. In particular, low-income countries with limited fiscal capacity could benefit from debt-for-nature swaps designed to mobilize resources for investments in environmental conservation measures while reducing the debt burden (Hansen, 1989).