Policy Research Working Paper 9649 Economic Governance Improvements and Sovereign Financing Costs in Developing Countries Girum Abate Michael Brown Alex Sienaert Mark Thomas Country Credit Risk Department May 2021 Policy Research Working Paper 9649 Abstract Low- and middle-income country governments are increas- sovereigns still being relatively limited (although growing). ingly tapping the global debt capital markets. This is Better economic governance Country Policy and Institu- increasing the amount of finance available for development, tional Assessment scores are associated with better estimated but at a considerably higher cost than traditional external ratings and materially lower financing costs; on average, borrowing on concessional terms. Using a novel methodol- improvements that are sufficient to increase the Country ogy based on estimating sovereign credit ratings using the Policy and Institutional Assessment economic governance Moody’s scorecard, and examining the associations between indicator score by one point are associated with interest these ratings and the World Bank’s Country Policy and costs that are lower by about 40 basis points, even setting Institutional Assessment scores, this paper examines how aside the direct impact on ratings of better governance indi- making improvements in the quality of economic policies cators. There are many reasons why improving governance and institutions can help lower governments’ financing is a good thing. Among them is the potential payoff to the costs. This method aims to overcome the small-sample public purse—savings of $40 million or more on a standard problem due to the number of rated developing country $1 billion, 10-year bond. This paper is a product of the Country Credit Risk Department. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at gabate@worldbank.org, mbrown10@worldbank.org, asienaert@worldbank.org, and mthomas1@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Economic Governance Improvements and Sovereign Financing Costs in Developing Countries 1 Girum Abate, Michael Brown, Alex Sienaert, and Mark Thomas JEL Classification: F34, H63 Keywords: Economic governance, Debt Management, Sovereign Credit Ratings 1 The views expressed here are only ours and do not necessarily reflect those of the World Bank, its executive directors, or the countries they represent. Corresponding authors: gabate@worldbank.org, mbrown10@worldbank.org, asienaert@worldbank.org, mthomas1@worldbank.org 1. Introduction Since the 1990s, and with a sharp increase Figure 1: Number of countries included in the JP over the past decade, the governments of a Morgan emerging market bond index (EMBIG) growing number of low- and middle-income 80 countries have tapped the global debt capital 70 markets for financing. For example, the 60 number of countries included in a widely- followed index of emerging market 50 Eurobonds, the JP Morgan Emerging Market 40 Bond Index Global (EMBIG), has increased 30 from 14 around the time of its inception in 20 1992 (to track the returns of so-called “Brady 10 bonds”), to 38 at the end of 2007, and 73 as of 0 October 2020 (Figure 1). Reflecting their increasing participation in capital markets, a growing number of low- Source: JP Morgan and middle-income country governments are now rated by the major commercial credit rating agencies. For example, of the 138 countries classified as low- or middle-income economies by the World Bank in 2019, the number rated by Moody’s increased from 53 as of November 2009 to 84 as of November 2019 (Figure 2). Figure 2: The rising number of low- and middle-income countries with sovereign credit ratings (sovereigns rated by Moody’s since before 2009, shaded gray, and since 2009, shaded red) Notes: Date labels show rating commencement date Sources: Data from Moody’s Statistical Handbooks (Country Credit) The shift towards these governments becoming credit-rated entities and using debt market financing puts their commercial financing costs in focus. Hard currency borrowing at market rates is considerably costlier than borrowing on concessional terms from traditional bilateral and multilateral sources. The increased cost, however, may be unavoidable for countries whose government financing needs have grown beyond what can be covered by the traditional non- market sources, or by borrowing domestically in their own (often shallow) credit markets. The 2 higher costs of non-concessional external borrowing, notably by issuing bonds (the cost of which is the focus of this paper) may be justified where this debt finances investments with high social returns (for example, impactful infrastructure investments, or sound investments in education and health systems). As governments’ financing mixes tilt towards taking on more external debt at market rates, the importance of understanding how governments can minimize the costs increases. This paper examines the link between low- and middle-income country governments’ financing costs and the quality of their policies and institutions (broadly, ‘economic governance’). It builds upon Brown and Sienaert (2019) and extends it in four important dimensions. First, we consider a larger sample size consisting of 132 low- and middle-income countries and three sample periods (2006, 2012 and 2018) in contrast to Brown and Sienaert who consider 73 countries across two sample periods (2005 and 2018). Second, this paper uses panel data regression incorporating country and time-period fixed effects in the sample. Third, we evaluate the relationship between economic governance indicators and financing costs controlling for major macroeconomic indicators. Fourth, we assess how an exogenous improvement in macroeconomic indicators affects both government policies and institutions and estimated ratings concurrently. To do this, we expand the panel data on a 2-year interval as 2006, 2008, … 2018 (see Section 4.3). Our approach supposes that countries’ sovereign credit ratings are affected substantially by the quality of government policies and institutions, and that these ratings in turn are associated with differing financing costs (either causally, because they help determine borrowing costs [see next section for references and some plausible channels], or simply because they correlate with market risk premia). The analysis contributes to the literature on the determinants of sovereign financing costs by using a novel approach that evaluates the relationship between economic governance indicators and financing costs through the lens of estimated country credit ratings. 2. Literature Summary A sovereign’s market borrowing cost is affected by the premium required to compensate investors for the risk that it defaults on its debt service obligations (the credit risk premium), and this premium is in turn affected by the quality of the country’s economic governance (institutions and policies). While this claim is not open to serious doubt, it remains the subject of a large and growing literature attempting to clarify and quantify the mechanisms at work. The rating actions of commercial credit rating agencies impact sovereign debt prices, but the channels through which they do so, and magnitudes, are contested. An overview of the literature is provided by Kiff, Nowak and Schumacher (2012), who cite evidence that rating changes can have significant impacts on debt prices (for example, when countries are moved between “investment grade” and “speculative grade” ratings). They also provide a useful typology of the theories of why ratings can be expected to be relevant for market prices, namely through the channels of (i) providing specialized information on default risk probability that affects investors’ decisions and hence pricing, (ii) providing certification services (e.g. characterizing bond default risks in ways that are relevant for index inclusion and hence impact liquidity and pricing), and (iii) monitoring services, whereby rating agencies’ visible ratings and intentions to change ratings (credit watch procedures) act as focal points for market actors and issuers alike, affecting their behaviors and pricing (e.g. by prompting actions by issuers to avoid downgrades). For the purposes 3 of this paper, however, what matters is simply that bond ratings are correlated with yields where we infer the extent to which better economic governance is associated with lower yields through estimated ratings. The quality of policies and institutions (economic governance) plausibly affects sovereign borrowing costs. A leading paradigm in the literature on the determinants of sovereign bond spreads is that they are driven by a combination of “push” (global interest rates, liquidity and risk appetite factors) and “pull” (country-specific) factors (see, for example, Haque, Bogoev & Smith, 2017, Eichengreen and Mody [1998]). Among the pull factors, the analytical focus tends to be on macroeconomic and financial indicators linked to default risk and hence risk premia, rather than on economic governance and institutional quality (Eichengreen and Mody [1998]). Clearly, however, macroeconomic performance is partly a function of the quality of a country’s institutional and policy frameworks: the better the quality of economic governance, the more stable and high-performing an economy is likely to be, and the lower the credit premium will be that investors demand to compensate for default risk. From a credit risk perspective, another useful paradigm is to characterize default risk as a function of weaknesses in a country’s willingness and ability to service its debts (Toksöz, 2014). Economic governance quality likely affects both factors There is empirical evidence that economic governance matters for sovereign debt prices. For OECD countries, Crifo, Diaye and Oueghlissi (2017) find that governments’ environmental, social and governance (ESG) performance is associated with lower borrowing costs, though with a magnitude only one-third that of financial metrics. They infer that investors use non-financial ratings as a supplement to financial ratings when making investment decisions. It seems reasonable that economic governance could be a more prominent factor in investment decisions in weaker economic governance settings, if marginal returns to credit quality from better economic governance are diminishing. Relatedly, several papers find links between political instability and corruption, and sovereign borrowing costs (e.g., Connolly [2007]). 2 This paper bases its analysis on estimated country credit risk ratings, even for countries which do not have actual public credit ratings from the major commercial rating agencies. The idea of estimating ratings for all developing countries is not novel; for example, Canuto, Mohapatra & Ratha (2011) estimate “shadow sovereign ratings” for developing countries using a regression- based methodology. These authors emphasize that the shadow ratings they obtain for many unrated governments are not at the bottom of the ratings range, which is consistent with our findings. However, to our knowledge this is the first paper to estimate country credit risk ratings for a large sample of low- and middle-income countries and investigate their association with economic governance. It does so based on an implementation of Moody’s scorecard-based rating methodology. 3 3. Data and Methodology We develop a set of ‘estimated ratings’ for low- and middle-income countries using the rating methodology published in 2018 of one of the three major commercial credit rating agencies, 2 Baldacci et al. (2011) also document that political factors are significant determinants of bonds spreads in emerging markets. 3 The advantage of Moody’s scorecard methodology is that it is transparent, easy to understand, the data are easily replicable, and it is straightforward to validate the estimated ratings against the available actual ratings. 4 Moody’s (Moody’s Investors Service, 2018). Moody’s ratings are informed by a scorecard that evaluates four factors: (i) economic strength, (ii) institutional strength, (iii) financial strength, and (iv) event risk. Each factor includes quantitative subfactors (such as growth rates, inflation, or public debt) that are sorted into scores ranging from “Very High +” to “Very Low –” based on a distribution described in the rating methodology. For example, an average annual growth rate above 4.5 percent scores as “Very High +” while an average growth rate below 0.5 percent scores as “Very Low –”. The subfactor scores are aggregated to generate overall factor scores for economic, institutional and financial strength and event risk. The scorecard then produces numerical ratings corresponding to the alphanumeric rating scale published by Moody’s, with larger numbers indicating weaker ratings, from 1 (Aaa, lowest credit risk) to 21 (C, lowest rating, typically indicating default with little if any potential recovery) (Table 1). Moody’s uses the scorecard as guide for the rating committees which determine its published sovereign credit ratings. To create the estimated ratings, we referenced the published 2018 methodology to construct a live replica of Moody’s scorecard. Factor scores for economic and fiscal strength are calculated using the quantitative subfactors scored as explained in the methodology. The score for each factor is also subject to adjustments, which reflect Moody’s subjective judgement in some cases or quantitative adjustments in others (such as the share of foreign currency debt in public debt). Of seven adjustments applied to economic and financial strength, two are subjective “other” adjustments. We were able to automate two of the remaining five adjustments using data (debt trend and foreign currency debt). The final three adjustments, which we were unable to automate, we maintained as neutral for all countries. We modified Moody’s assessment of institutional strength to avoid creating endogeneity problems in our regression analysis. As of 2018, Moody’s used three Worldwide Governance Indicators (WGI) to account for 75 percent of its institutional assessment, along with inflation for the remaining 25 percent. One of the WGI indicators, government effectiveness, includes World Bank CPIA scores as a subcomponent. Since we use CPIA scores as independent variables in the regression analysis (discussed below), we completely removed the WGI indicators from the scorecard and allowed inflation to account for 100 percent of the institutional assessment. Of the two adjustments to the score, we used Moody’s data to automate the “Track Record of Default” and set the manual “Other” adjustment to neutral. Our modified version of the scorecard generates slightly less accurate estimated ratings, but removes the risk of our regression results being affected by a mechanical link between estimated ratings and CPIA governance measures. The combination of economic, institutional and fiscal strength provides Moody’s assessment of “government financial strength,” which Moody’s adjusts further with a fourth factor, event risk. 4 Event risk is a max function that takes the riskiest value from an assessment of four subfactors: political risk, liquidity risk, banking sector risk and external vulnerability risk. While the methodological paper lays out the calculations, in practice Moody’s assessment of event risk utilizes more qualitative judgement that is difficult to automate. As such, our approach to event risk held the score as “medium” for all countries. While this overstated risks for some countries and understated it for others, testing indicated this generated the most even distribution of errors 4 See Appendix D for a demonstration of Moody’s scorecard methodology along with list of the variables used in the scorecard. 5 in estimated credit ratings. An approach that more fully replicates Moody’s event risk scoring is an area of future research to consider. The accuracy of the simulated scorecard is dependent on proper data inputs. We fed data from the IMF World Economic Outlook into our model for macroeconomic and fiscal variables. Moody’s uses World Economic Forum competitiveness scores and WGI (control of corruption, rule of law and government effectiveness) for competitiveness and governance assessments. As discussed above, we omitted the WGI data. For the assessment of public external debt as a share of public debt, we used Moody’s own values (from the Country Credit Statistical Handbook) where possible. If Moody’s did not cover a country, we used data from the World Development Indicators “external debt stocks, public and publicly guaranteed” to calculate the ratio. In a few instances, where neither Moody’s nor the WDI had data, the figure was manually extracted from IMF publications to complete the data set. Our final data set created estimated ratings for 2006, 2008, 2010, 2012, 2014, 2016 and 2018. See Appendix D “Scorecard data sourcing” for a line-by-line review of the scorecard and data used. For countries with actual Moody’s credit ratings, our estimated ratings track official ratings closely. As of end-2018, Moody’s rated 82 of the 137 countries classified by the World Bank as low- or middle-income, allowing us to compare our estimated ratings for accuracy. Estimated ratings are within two notches of the actual rating 61 percent of the time. Gaps of 4 notches or more occurred only in only 29 percent of cases. Overall, based on regressing the Moody’s ratings on our estimated ratings, the estimated and actual ratings have a pairwise correlation of 0.8 with 89 percent of the actual ratings within 4 notches of the best-fit line. There is no sign of any directional bias in the estimates (Figure 3). Table 1: Mapping of numerical ratings Figure 3: Actual and estimated Moody’s ratings corresponding to published Moody’s ratings 1 Aaa 11 Ba1 2 Aa1 12 Ba2 3 Aa2 13 Ba3 Non-Investment Grade Investment grade 4 Aa3 14 B1 5 A1 15 B2 6 A2 16 B3 7 A3 17 Caa1 8 Baa1 18 Caa2 9 Baa2 19 Caa3 10 Baa3 20 Ca 21 C Having constructed our dependent variable (estimated ratings), we turned next to the data capturing the quality of economic governance, policies, and institutions. For this, we used the World Bank’s Country Policy and Institutional Assessment (CPIA) scores. These assess how well countries’ policy and institutional frameworks function in support of growth, poverty reduction, 6 and the effective use of development assistance. 5 The assessment scores countries on a 1-6 point scale with 0.5 unit increments across 16 criteria divided into four clusters (Table 2). Table 2: Summary of World Bank Country Policy and Institutional Assessment (CPIA) criteria Abbreviation CPIA criteria Comments ECON A. Economic management This cluster score aggregates the scores for its constituent indicators 1-3 below, to capture the extent to which the country’s overall economic policy framework supports growth and development. MACR 1. Monetary and exchange Higher scores mean policy and institutional frameworks better-supporting rate policies internal and external macroeconomic balances, and better flexibility to adjust to shocks. FISP 2. Fiscal policy Higher scores mean that fiscal policies are better at stabilizing the economy over the cycle and following shocks, and more effective at providing public goods. DEBT 3. Debt policy and Higher scores mean that debt management strategy and implementation is more management conducive to debt sustainability and helps to minimize financing risks to the budget. STRC B. Structural policies This cluster score aggregates the scores for its constituent indicators 4-6 below, to capture the extent to which the country’s structural policies support growth and development. TRAD 4. Trade Higher scores mean that the policy framework is more conducive to global integration in goods and services, covering both trade restrictiveness (tariff and non-tariff barriers) and trade facilitation. FINS 5. Financial sector Higher scores mean that policies and regulations are more supportive of financial sector development in terms of fostering financial stability, the efficiency and extent of resource mobilization through the financial sector, and the level of access to financial services. BREG 6. Business regulatory Higher scores mean that the legal, regulatory, and policy environment is more environment helpful in supporting private business in investing, creating jobs, and becoming more productive. SOCI C. Policies for social This cluster score aggregates the scores for its constituent indicators 7-11 inclusion/equity below, to capture the extent to which the country’s policies concerning social inclusion and equity support growth and development. GNDR 7. Gender equality Higher scores mean that policies and institutions foster more equal access for men and women to human capital development, productive and economic resource, and legal status and protections. PRES 8. Equity of public resource Higher scores indicate that the pattern of public expenditures and revenue use collection is more oriented towards supporting the poor and is consistent with national poverty reduction priorities. HRES 9. Building human Higher scores indicate that national policies and public and private sector resources service delivery support better access to, and quality of, health- and education- related services. PROT 10. Social protection and Higher scores mean that social protection and labor policies more effectively labor help protect poor and vulnerable households and workers from risks, support their incomes, and promote human capital development and income generation. ENVR 11. Policies and institutions Higher scores indicate that environmental policies and institutions are more for environmental effective in fostering the protection and sustainable use of natural resources and sustainability the management of pollution. PUBS D. Public Sector This cluster score aggregates the scores for its constituent indicators 12- Management and 16 below, to capture the extent to which the country’s overall public sector institutions management and the quality of its legal institutions support growth and development. PROP 12. Property rights and rule- Higher scores indicate stronger property and contract rights, a better legal and based economic judicial system, and lower crime and violence governance FINQ 13. Quality of budgetary and Higher scores indicate more comprehensive and credible national budgets which financial management embed policy priorities, along with more effective financial management resulting in better budgetary implication, and time and accurate fiscal reporting, including transparent audits and follow-up actions. REVN 14. Efficiency of revenue Higher scores indicate better tax policy and administration, encompassing mobilization effective revenue mobilization with as little distortion of economic activity as possible, the stability, clarity and predictability of tax laws, and the extent to which tax administration is efficient, rules-based and transparent. 5 This paragraph and Table 2 are based on the document “CPIA 2011 Criteria”, available at: http://siteresources.worldbank.org/IDA/Resources/73153-1181752621336/CPIAcriteria2011final.pdf. 7 PADM 15. Quality of public Higher scores indicate that the civil service performs better in terms of managing administration its operations and human resources and implementing and managing regulations. TRAN 16. Transparency, Higher scores indicate stronger accountability and oversight mechanisms, more accountability and and better access by civil society to information which is relevant to the oversight corruption in the public of the executive, decreased influence of vested interests in policies and resource sector allocation, and better integrity in the use of public resources. The CPIA scores capture the World Bank’s assessment of the quality of countries’ governance. There is evidence that they are a good predictor of future economic growth (Gonzalez & Nishiuchi, 2018). Based on this, and on the fact that commercial agencies’ sovereign credit rating methodologies consider many governance-related metrics, our prior is that estimated ratings will be associated with the CPIA scores. Improvement or deterioration in macroeconomic management and the quality of economic governance are captured in factor two of Moody’s scorecard: Institutional Strength. As such, strengthening the quality of economic governance in a country can be linked to changes in a country’s credit rating and, by extension, its borrowing costs in global capital markets. Changes in macroeconomic indicators can also affect both economic governance and ratings simultaneously as discussed in Section 4.3. As of the end of 2018, 25 out of 73 low income countries had official Moody’s ratings – up from just eight a decade before, but still a small group. To circumvent this small-sample limitation for regression analysis, we expand the number of countries in our sample to 132 by estimating ratings for the years 2006, 2012 and 2018. The estimated ratings for all 132 countries with the corresponding CPIA scores between 2006, 2012 and 2018 constitute our panel data. 6 We include as controls the major macroeconomic correlates of sovereign risk such as GDP per capita, real GDP growth, the government debt burden, current account balance and consumer price index (CPI). GDP per capita (PPP) and government debt burden data are from the World Bank Development Indicators (WDI) for the years 2006, 2012 and 2018. Real GDP growth, inflation, and the current account balance are extracted from the IMF World Economic Outlook (WEO) database. The dependent variable is the estimated Moody’s rating for the corresponding years. 4. Results a. Regression analysis We conduct regression analysis of the relationship between borrowing costs and economic governance improvements using a panel data set of 132 low- and middle-income countries during 2006, 2012 and 2018. The full list of countries is given in Table A1 in the appendix. The data set includes information on World Bank’s CPIA, consisting of four clusters of indicators: Economic Management (ECON), Structural Policies (STRC), Polices for Social Inclusion/Equity (SOCI), and Public Sector Management and Institutions (PUBS). 7 The CPIA data for low income countries are publicly available while the CPIA data for middle income countries are confidential. 6 The CPIA data are available at: https://datacatalog.worldbank.org/dataset/country-policy-and-institutional- assessment. The latest, 2017 CPIA indicators were published in 2018 based on the latest data available at that time, so it is appropriate to match these with 2018 estimated ratings. CPIA data for some of the middle-income countries in our sample are confidential. 7 The reason for conducting regression analysis at the CPIA cluster-score level, as opposed to individual CPIA scores, is described in the appendix. 8 Table 3 summarizes the descriptive statistics of Moody’s Table 3: Descriptive statistics estimated rating, CPIA indicators and the macroeconomic Variable Mean Std.dev variables. The average rating is 11.68, corresponding to Ba3 – Rating 11.68 3.29 three notches below investment grade. Table 4 reports the ECON 3.53 0.71 pairwise correlation between estimated ratings, governance STRC 3.44 0.59 indicators and macroeconomic variables. The Economic PUBS 3.17 0.54 Management (ECON) CPIA cluster score has a correlation of SOCI 3.42 0.54 0.519 with the ratings, with higher scores being associated with GDP 8.67 0.96 lower (i.e., better) ratings. The links between ratings and Debt 3.61 0.98 macroeconomic indicators appear generally weaker, with the BOP -3.9 10.61 exception of debt level (“Debt”). CPI 6.46 7.39 Growth 4.82 7.42 Table 4: Pairwise correlation coefficients Rating ECON STR PUBS SOCI GDP Debt BOP CPI Growth Rating 1.000 ECON -0.519 1.000 STR -0.351 0.682 1.000 PUBS -0.359 0.653 0.788 1.000 SOCI -0.299 0.646 0.756 0.832 1.000 GDP -0.376 0.309 0.515 0.499 0.514 1.000 Debt 0.500 -0.402 -0.141 -0.061 -0.072 -0.122 1.000 BOP -0.262 0.054 -0.063 -0.091 -0.060 0.260 -0.310 1.000 CPI 0.400 -0.278 -0.324 -0.323 -0.216 -0.103 0.200 0.012 1.000 Growth -0.075 0.176 0.030 0.052 0.127 0.019 -0.127 0.173 -0.055 1.000 To formally assess the relationship between borrowing costs and economic governance improvements, we estimate the following cross-country regression equation: = + + + , (1) where is estimated ratings for country in year , is CPIA score (either for ECON or PUBS) and is the coefficient on the CPIA scores, is macroeconomic indicator (GDP per capita, CPI, balance of payments, real GDP growth and government debt) and the scalar α and the vector γ are estimated coefficients. and , respectively, are country and time period fixed effects. The country fixed effect is an unobserved time-invariant term that measures time-invariant features of countries (such as geographical factors); is a time-period (year) fixed effect controlling for variables that are constant across countries but vary over time (such as variables capturing global economic conditions). Using both the country and time fixed effects eliminates bias from both kinds of unobservable fixed factors. 9 Regression results are shown in Table 5. Column I shows the pooled OLS regression, column II and III show panel regressions with year and country fixed effects, respectively. Column IV shows the panel regression with both year and country fixed effects, and column V is the panel regression with both year and country fixed effects including macroeconomic indicators. Amongst the CPIA cluster scores, only ECON (Economic Management) is consistently statistically significant. It also consistently has the expected, negative sign: higher scores are associated with lower estimated rating numbers (stronger ratings). The cluster score PUBS (Public Sector Management and Institutions) is statistically significant in three of the five specifications, but there is no robust evidence of a well-defined relationship with ratings, as signs and coefficient estimates are not significantly different from zero in the other two panel data specifications. The results of the panel regression with both year and country fixed effects, and macroeconomic controls, shown in column V, point to GDP growth being significant determinants of ratings. Compared to the specification without macroeconomic controls (shown in column IV), the estimated coefficient for the ECON cluster score is considerably smaller, -0.96 vs. -2.00, but remains statistically significant. We infer that the quality of economic management is associated with countries’ income and debt levels, but may also affect ratings independently of these associations. Taking the coefficient of ECON in specification V as the most conservative estimate of the impact of economic governance improvements on ratings, the estimated effect is economically significant: an increase in the economic governance score results in rating improvement. In the next section, we conduct a qualitative check of this result – that economic governance matters for ratings, even controlling for macroeconomic factors – by examining how actual ratings have changed over time in cases where there have been large changes in CPIA scores. 10 Table 5: Regression results I II III IV V ECON -1.307*** -2.509*** -1.746*** -2.008*** -0.962** (0.279) (0.330) (0.462) (0.468) (0.550) STR -0.085 0.055 1.318 0.988 0.964 (0.449) (0.520) (0.864) (0.862) (0.958) SOCI -0.065 1.202*** 0.545 1.146 2.207** (0.505) (0.597) (0.976) (0.995) (1.049) PUBS -2.211*** -0.975 -1.946** -1.755 -2.808** (0.568) (0.649) (1.122) (1.108) (1.256) GDP -1.272 (1.006) Debt 0.024 (0.007) BOP 0.004 (0.020) Growth -0.104** (0.045) CPI 0.025 (0.031) R2 0.432 0.322 0.807 0.815 0.854 Year FE No Yes No Yes Yes Country FE No No Yes Yes Yes Notes: The dependent variable is the estimated Moody’s rating (higher number indicates lower credit rating). Column I shows pooled OLS regression, column II and III are, respectively, panel regressions with year and country fixed effects. Column IV is panel regression with both year and country fixed effects, and column V is panel regression with both year and country fixed effects including macroeconomic indicators. The regressions include a constant. Standard errors in parentheses. *** p<0.01, ** p<0.05. The sample period is 2006, 2012 and 2018 across 132 countries. 11 b. Assessing actual rating changes for countries with large changes in CPIA scores We will now check whether the associations Figure 4: Histogram of publicly available CPIA ECON described above are detectable in how score changes (first to last available year, 2005-18) countries’ CPIA scores and actual ratings change over time. For this dynamic analysis, we are again constrained to the 2006-2018 period, to those countries whose CPIA scores are publicly available, and to those countries which had Moody’s ratings over the period. We consider countries’ cluster scores for economic management (ECON) specifically, the score that is estimated robustly to be statistically significant on the basis of the above regression analysis. The distribution of ECON score changes, from the first available to the last available score from 2006 to 2018, ranges from -3.7 to 1.7 (Figure 4). Figure 4 suggests that unusually large ECON score changes over this period can be considered decreases of 1 point or more or increases of 0.5 points or more. By this measure, 25 countries experienced ‘large’ changes in their economic management scores during the period. Table 6 lists these countries and summarizes their actual sovereign ratings trajectory. Of the 25 countries, 15 had no rating at any time during this period (grey-shaded rows in Table 6), so we cannot judge how their ratings responded to large improvements or deteriorations in ECON CPIA scores. Of the remaining 10 countries, nine experienced Moody’s rating developments in the expected direction on the basis of their (relatively large) shifts in CPIA ECON scores over the period. In six cases, there were strong improvements in ECON scores, and these resulted either in rating upgrades over the period, or first-time ratings following the start of the trend-improvement in CPIA scores, or some combination of these; we posit that first-time ratings generally occur when countries’ have developed a sufficiently positive policy and economic management track record to achieve global market access. In three cases, there were strong deteriorations in ECON scores. Commensurately, there were corresponding rating downgrades in two of these (Mozambique and Pakistan). In one case (Maldives), there were no downgrades, but we mark this case as still being consistent with expectations, because the period during which CPIA scores deteriorated (2006- 2010) was well before it acquired a rating for the first time (2016), suggesting these developments were not linked. This leaves only one case which is anomalous: Tajikistan saw its CPIA ECON score slide from 3.8 in 2011 to 3.0 in 2016, yet the following year (2017) it still acquired a rating for the first time. Notably, however, this first-time rating was a low B3, consistent with the deterioration in economic governance scores weighing heavily on the credit. 12 Overall, the results of this more qualitative analysis agree with those from the regressions. CPIA score changes for economic management generally correspond with rating changes in the expected direction. They are not deterministic, as a range of other factors also decide countries’ rating trajectories, introducing considerable noise when attempting to discern the independent impact of economic governance quality changes. But these changes do appear to matter, albeit with caveats about the small available sample size. Of the 10 countries for which CPIA scores of interest changed significantly (six improved; four deteriorated), and which had ratings, ratings moved in the expected direction in all but one case. Table 6: The rating performance of countries experiencing large changes in CPIA scores (Color key: orange = rating changes in line with expectations, grey = no info, red = anomalous) 5 Change in Rated Country graduated to Rating changes over Acquired a rating Rating movement ECON continually IBRD (causing CPIA CPIA change period / # Country between 2005 and consistent with ECON score, since 2005 or scores to stop being since initiation / last 2018 score change? 2005-18 earlier available before 2018) available score +1 notch to Ba3 in 1 Angola 0.5 Yes, in 2010 Yes (in 2013) Yes 2011 2 Comoros 0.5 No No No N/A N/A +1 notch to Ba3 in 3 Cote d'Ivoire 1.7 No Yes, in 2015 No Yes 2015 4 Georgia 0.5 No Yes, in 2010 In 2013 +1 notch to Ba2 Yes 5 Guinea 0.8 No No No N/A N/A 6 Haiti 0.5 No No No N/A N/A 7 Kosovo 0.5 No No No N/A N/A Yes (CPIA deterioration came more than 5 years 8 Maldives -1.2 No Yes, in 2016 No Rating unchanged (B2) before 1st rating, stable thereafter) 9 Mauritania 0.7 No No No N/A N/A 10 Moldova 0.7 Yes +1 notch to B3 in 2010 Yes Yes, in 11 Mozambique -1.3 No No -5 notches to Caa3 Yes 2013 12 Myanmar -3.7 No No No N/A N/A 13 Pakistan -1.0 Yes No No -1 to B3 Yes 14 Rwanda 0.5 No Yes, in 2016 Rating unchanged (B2) Yes 15 Somalia -1.8 No No No N/A N/A South 16 -1.8 No No No N/A N/A Sudan 17 St. Lucia -1.0 No No No N/A N/A No (CPIA slipped within a few years of first 18 Tajikistan -1.2 No Yes, in 2017 No Rating unchanged (B3) rating, and no subsequent downgrade) 19 Timor-Leste 0.5 No No No N/A N/A 20 Togo 1.2 No No No N/A N/A 21 Tonga 0.7 No No No N/A N/A 22 Tuvalu -2.5 No No No N/A N/A 23 Uzbekistan 0.7 No Yes, in 2018 Rating initiated at B1 Yes Yemen, 24 -1.8 No No No N/A N/A Rep. 25 Zimbabwe 1.3 No No No N/A N/A c. Robustness checks In this section, we reexamine the overall robustness of our main result. A potentially important concern is that both estimated ratings and CPIA scores are functions of macroeconomic variables, resulting in the endogeneity of CPIA in our regression framework. To circumvent this issue, we 13 consider a regression framework where estimated ratings and CPIA ECON (the CPIA score of key interest based on the regression results) are both functions of lagged macro variables and examine how an exogenous improvement in macro shifts both CPIA ECON and estimated ratings concurrently. For this, the data is expanded on a 2 year interval from 2006 to 2018 (2006, 2008…2018) and a system GMM is employed. 8 Table 7: Alternative regression results Variables Dep var: Estimated ratings Dep var: CPIA (ECON) Lag GDP -0.581** 0.158** (0.250) (0.076) Lag growth -0.195 0.037 (0.272) (0.0634) Lag CPI 0.293 0.0554 (0.210) (0.054) Lag bop -0.108 0.004 (0.038) (0.010) Lag debt 0.0024 0.004 (0.018) (-0.004) Notes: The dependent variable for the first and second column, respectively, is Moody's estimated ratings (higher number indicates lower credit rating) and CPIA measured by ECON. Standard errors in parentheses. *** p<0.01, ** p<0.05. The sample period is 2006, 2008, ..., 2018 across 132 countries. The regressions include a constant. The results are reported in Table 7. The first column reports the results where the dependent variable is the estimated ratings while the second column reports the result with CPIA ECON itself as a dependent variable. GDP per capita appears to have a significant effect both on estimated ratings and CPIA ECON simultaneously. Further, we conduct two additional exercises. First, we consider an alternative regression specification that allows for lags of CPIA ECON, to consider potential impacts of the lagged CPIA ECON score on contemporaneous estimated ratings. Second, we relate estimated ratings to contemporaneous CPIA ECON and macro (GDP per capita and bop) with GDP as an endogenous regressor (where GDP in turn is related to lagged CPIA) to allow for potential impacts of CPIA on macro. Table C.1 in the appendix reports the results. Our main result does not change with this alternative specifications i.e, better economic governance scores are associated with better credit ratings. 8 We use lagged explanatory variables as instruments in the GMM estimation, see Jarmuzek and Lybek (2018), Gonzalez and Nishiuchi (2018) for similar approaches. 14 5. Policy Implications There is an association between countries’ quality of economic management (as measured by the ECON CPIA cluster score) and their sovereign credit ratings. The magnitude of the relationship varies across regression specifications, but in the specification with country and time fixed effects and macroeconomic controls, the estimated coefficient is around 0.9. We take this as a conservative estimate of the average effect of an economic governance score change on ratings. That is, on average, a 1-point increase (decrease) in CPIA economic governance score yields a 0.9 notch improvement (deterioration) in the rating. To evaluate the financing cost implications, Figure 5 shows EMBIG spreads for countries in the EMBIG index for which we have estimated ratings. The positive relationship between spreads and higher rating scores (worse ratings) is clear. Table 8 shows the results of regressions of spreads on ratings. The dependent variable is the EMBIG spread, and the independent variable is the estimated Moody’s rating (where a higher number indicates a worse credit rating). Column I is a pooled OLS regression, columns II and III are, respectively, panel regressions with year and country fixed effects, and column IV includes both country and year fixed effects. Using the most conservative specification that allows for both country and time-period fixed effects (column IV), on average, a 1-notch improvement in the estimated rating is associated with a 40-bps reduction in spread. Figure 5: EMBIG spreads and Moody’s Table 8: Regression of EMBIG spreads on ratings estimated ratings I II III IV Rating 38.37*** 40.05*** 15.21 33.014*** (4.99) (5.29) (10.19) (8.82) R2 0.299 0.445 0.703 0.804 Year FE No Yes No Yes Country No No Yes Yes FE Notes: The figure shows the relationship between the average EMBIG spread (computed for the years 2006, 2012, and 2018 for each country) and average Moody’s rating (computed for the years 2006, 2012 and 2018; higher numbers indicates credit ratings). The table shows regression results where the dependent variable is the EMBIG spread, and the independent variable is the estimated Moody’s rating (where a higher number indicates a lower credit rating). Column I is pooled OLS regression, columns II and III are, respectively, panel regressions with year and country fixed effects, and column IV includes both country and year fixed effects. The regressions include a constant. Standard errors in parentheses. *** p<0.01, ** p<0.05. The main policy implication is that making improvements in the quality of economic governance, as recognized by the CPIA indicators, can plausibly lead to substantial reductions in governments’ external financing costs. On average, improvements which are sufficient to improve a CPIA economic governance indicator score by 1 point are associated with a 0.9 notch improvement in the estimated rating. This, in turn, is associated with a 36-basis point (40bp×0.9) reduction in spread. For a $1 billion bond with a 10-year tenor, for example, this would reduce nominal interest costs by $36 million over the life of the bond. This conclusion requires nuancing. First, the relationship between CPIA score changes and rating changes is by no means deterministic. As is apparent from the regression analysis, CPIA indicators 15 pick up only some factors which drive countries’ ratings. Economic governance improvements that move a country’s CPIA score may or may not have a material impact on its sovereign credit rating, depending on its unique circumstances and credit drivers. Similarly, there will be variation in how quickly (if at all) a country’s rating responds to economic governance improvements. The rating might respond quickly when the measures address economic governance shortcomings that are deemed to be a critical rating weakness, but only slowly in other cases where there are also other rating-critical factors at play. In some countries, economic governance improvements are likely to filter into economic performance and sovereign debt dynamics quickly and tangibly, and in others less so, for example due to other, unresolved bottlenecks in institutions and the economy. Particularly large deviations from the average effect of economic governance improvements on yields could be driven by differences in countries’ initial conditions and discontinuities affecting market access and costs. The analysis is based on estimated ratings and how these typically map to financing costs. But imagine a country which gains global capital market access for the first time, for example by acquiring a rating and issuing its first Eurobond. Making economic governance improvements as reflected by an improved CPIA score could be critical to achieving market access, which in turn could have a big impact on the government’s financing possibilities and costs. In such cases, the stakes for economic governance improvements could be much higher than those suggested by the average effects that have been our focus here. Similarly, countries may face important discontinuities, for example becoming included or excluded from global bond indices because of crossing the investment-/speculative-grade rating divide. Finally, global financial market conditions are fluid, resulting in variations over time in risk premia and hence in the payoff to countries from improvements that reduce their credit risk. The results reported above are based on differences in yields across ratings during the study period (2005- 2018). In future, risk premia may widen (increasing the payoff for risky countries of reducing investors’ risk perceptions), or compress (potentially reducing this payoff). Averaged across time, we can expect risk premia to be significant, and so to generate significant potential payoffs from measures which reduce country risk. However, the size of these payoffs will likely vary over time, which could affect governments’ incentives to undertake difficult reforms to lower financing costs. This paper has used a novel approach – constructing estimated Moody’s ratings – to argue that in developing countries, economic governance improvements can lower governments’ external financing costs. This is not a surprising conclusion, given the links between sound policies and institutions and economic performance and development, and thus on credit ratings and market- determined government debt risk premia. Data limitations preclude making precise statements about which economic governance improvements can deliver the largest reduction in financing costs. Even the average effects we report should be understood as indicative estimates. While clearly a limitation in terms of the statistical power we achieve with this approach, this research outcome is also intuitive: country circumstances vary widely, so there will be large, idiosyncratic differences in which breakthroughs in policies and institutions will be more fruitful – to support sustainable development in general and to lower market financing costs to government in particular. These results have practical implications. For policy makers, tackling economic governance weaknesses can deliver immediate and tangible gains to the government bottom line. There could 16 be quick wins, where policy makers are able to identify reforms, or strengthen implementation of existing policy and institutional frameworks, which deliver improvements to ratings and bond yields. Ultimately, the benefit of making such improvements is of course less about the direct impact on financing costs than their potential power to drive economic development. Still, reducing financing costs constitutes a visible, tangible, near-term benefit, which could strengthen reform incentives to spend political capital on difficult reforms. For some countries, working towards achieving a better credit rating, or indeed acquiring a rating and external financial market access for the first time, can be a useful focal point for policy. Where appropriate given country circumstances, this could be complemented by using estimated ratings, as used in this paper, to identify areas for policy action based on a high likelihood that the rating will be sensitive to improvements. 17 Appendix A: Table A.1 Countries No. Country No. Country No. Country 1 Afghanistan 46 Gambia, The 91 Pakistan 2 Albania 47 Georgia 92 Papua New Guinea 3 Algeria 48 Ghana 93 Paraguay 4 Angola 49 Grenada 94 Peru 5 Argentina 50 Guatemala 95 Philippines 6 Armenia 51 Guinea 96 Poland 7 Azerbaijan 52 Guinea-Bissau 97 Romania 8 Bangladesh 53 Guyana 98 Russian Federation 9 Belarus 54 Haiti 99 Rwanda 10 Belize 55 Honduras 100 Samoa 11 Benin 56 India 101 Senegal 12 Bhutan 57 Indonesia 102 Serbia 13 Bolivia 58 Jamaica 103 Seychelles 14 Bosnia and Herzegovina 59 Jordan 104 Sierra Leone 15 Botswana 60 Kazakhstan 105 Solomon Islands 16 Brazil 61 Kenya 106 South Africa 17 Bulgaria 62 Kiribati 107 Sri Lanka 18 Burkina Faso 63 Kyrgyz Republic 108 St. Kitts and Nevis 19 Burundi 64 Lao PDR 109 St. Lucia 20 Cabo Verde 65 Lebanon 110 St. Vincent and the Grenadines 21 Cambodia 66 Lesotho 111 Sudan 22 Cameroon 67 Liberia 112 São Tomé and Principe 23 Central African Republic 68 Madagascar 113 Tajikistan 24 Chad 69 Malawi 114 Tanzania 25 China 70 Malaysia 115 Thailand 26 Chile 71 Maldives 116 Timor-Leste 27 Colombia 72 Mali 117 Togo 28 Comoros 73 Marshall Islands 118 Tonga 29 Congo, Dem. Rep. 74 Mauritania 119 Trinidad & Tobago 30 Congo, Rep 75 Mauritius 120 Tunisia 31 Costa Rica 76 Mexico 121 Turkey 32 Croatia 77 Micronesia 122 Turkmenistan 33 Côte d'Ivoire 78 Moldova 123 Uganda 34 Djibouti 79 Mongolia 124 Ukraine 35 Dominica 80 Montenegro 125 Uruguay 36 Dominican Republic 81 Morocco 126 Uzbekistan 37 Ecuador 82 Mozambique 127 Vanuatu 38 Egypt, Arab Rep 83 Myanmar 128 Venezuela, RB 39 El Salvador 84 Namibia 129 Viet m 40 Equatorial Guinea 85 Nepal 130 Yemen, Rep 41 Eritrea 86 Nicaragua 131 Zambia 42 Eswatini 87 Niger 132 Zimbabwe 43 Ethiopia 88 Nigeria 44 Fiji 89 North Macedonia 45 Gabon 90 Panama 18 Appendix B: Reasons for conducting regression analysis at the CPIA cluster level The estimated Moody’s ratings are correlated with a Table B.1 Pairwise Correlation Coefficients RATING MACR FISP DEBT PROP FINQ REVN PADM TRAN number of CPIA indicators, RATING 1.00 -0.42 -0.51 -0.30 -0.38 -0.52 -0.36 -0.47 -0.40 many of which are strongly MACR -0.42 1.00 0.54 0.63 0.63 0.53 0.68 0.80 0.72 correlated with one another FISP -0.51 0.54 1.00 0.44 0.66 0.76 0.51 0.76 0.53 (see table B.1). The pairwise DEBT -0.30 0.63 0.44 1.00 0.55 0.40 0.82 0.77 0.89 correlations with the estimated PROP -0.38 0.63 0.66 0.55 1.00 0.57 0.64 0.87 0.68 Moody’s ratings (RATING) FINQ -0.52 0.53 0.76 0.40 0.57 1.00 0.44 0.66 0.45 which stand out are the scores REVN -0.36 0.68 0.51 0.82 0.64 0.44 1.00 0.81 0.91 for fiscal policy (FISP) and the PADM -0.47 0.80 0.76 0.77 0.87 0.66 0.81 1.00 0.88 quality of budget and financial TRAN -0.40 0.72 0.53 0.89 0.68 0.45 0.91 0.88 1.00 management (FINQ). Many CPIA indicators are also strongly correlated with other indicators, as is to be expected given that the scores capture different, but usually related, dimensions of the quality of policy making and institutions. For example, the correlation between FISP and FINQ, the two indicators most correlated with RATING, is a very strong 0.76. Regression-based estimates of the magnitude of the link between specific CPIA scores and ratings are not sufficiently precise. The CPIA scores are clearly far from orthogonal (as shown by the table), causing multicollinearity which prevents individual parameters from being estimated reliably. For example, taking the two most-strongly correlated CPIA scores with the estimated rating, FISP and FINQ, it is apparent that both are strongly associated with RATING, and with each other (see figure B.1). Consequently, a regression containing both FISP and FINQ as explanatory variables would be unlikely to be able to distinguish between their independent effects. In addition, both the estimated ratings (RATING) and the CPIA scores are ordinal numbers not continuous variables. An ordered choice model could account for the dependent variable being an ordinal ranking, but is not appropriate when both the dependent and independent variables are ordinal. Recognizing these challenges to statistical inference of the associations with specific CPIA indicators, we limit the analysis to the level of the CPIA cluster scores: ECON, STRC, SOCI and PUBS. Figure B.1: FISP and FINQ Both Have Strong Associations with RATING, and They are also Strongly Correlated with Each Other (Scatter Plots with Linear Best-Fit Lines) 20 20 4.8 4.4 18 18 4.0 16 16 3.6 14 3.2 14 RATING RATING FINQ 2.8 12 12 2.4 10 2.0 10 1.6 8 8 1.2 6 0.8 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 6 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 0 1 2 3 4 5 FISP FISP FINQ 19 Appendix C: Additional robustness test results Table C.1: Additional robustness results I IV ECON -1.338*** -1.656*** (0.281) (0.338) STR -0.358 0.434 (0.434) (0.439) SOCI 1.039** 0.618 (0.539) (0.593) PUBS -0.6476843 0.039 (0.588) (0.637) GDP -0.486** (0.228) BOP -0.050*** (0.018) ECON 0.154*** (0.058) R2 0.119 -- Notes: The dependent variable is the estimated Moody’s rating (higher number indicates lower credit rating). Column I reports panel regressions of ratings on two year lags of CPIA, Column II reports regression of ratings on two year lags of CPIA and contemporaneous macro. Column II reports results of a GMM estimation where ratings are related to CPIA and macro (GDP per capita and bop) with GDP per capita as an endogenous regressor. The ECON coefficients in the bottom panels of Column II is the impacts of lag CPIA (ECON) on GDP per capita. Standard errors in parentheses. *** p<0.01, ** p<0.05. The sample period is 2006, 2008, 2010..., 2018 across 132 countries. 20 Appendix D: Scorecard data sourcing Sample Country Scorecard Factor 1 - Economic Strength Data Source Growth Dynamics Average Real GDP Growth (t-4 to t+2) WEO Database Volatility in Real GDP Growth (t-9 to t) WEO Database WEF Competitiveness Index World Economic Forum Scale Nominal GDP (US$ bn) WEO Database National Income GDP Per Capita (PPP, US$) WEO Database Adjustments Credit Boom Set to neutral Other - Manual Adj. Set to neutral Factor 2 - Institutional Strength Data Source Institutional Framework Government Effectiveness Index Omitted from calculation Rule of Law Index Omitted from calculation Control of Corruption Index Omitted from calculation Policy Credibility Inflation (t-4 to t+2) WEO Database Inflation Volatility (t-9 to t) WEO Database Adjustments Track Record of Default - Manual Adj. Moody's default study Other - Manual Adj. Set to neutral Economic Resilience Score (Combination of Factor Scores for Economic and Institutional Strength) Factor 3 - Fiscal Strength Data Source Debt Burden Gen. Gov. Debt/GDP WEO Database Gen. Gov. Debt/Revenue WEO Database Debt Affordability Gen. Gov. Interest/Revenue WEO Database Gen. Gov. Interest/GDP WEO Database Adjustments Debt Trend (Debt/GDP: t-4 vs. t+1) WEO Database Gen. Gov. FC Debt/Total Debt Moody's or WDI database Other Public Sector Debt - Manual Adj. Set to neutral Sovereign Wealth Fund - Manual Adj. Set to neutral Other - Manual Adj. Set to neutral Government Financial Strength (Combination of Factors 1 & 2 with Fiscal Strength) Factor 4 - Susceptibility to Event Risk Set to "Moderate" Indicative Rating (Combination of Factors 1 & 2 & 3 with Event Risk) 21 References Baldacci, E., S. Gupta, and A. Mati (2011). Political and fiscal risk determinants of sovereign spreads in emerging markets. Review of Development Economics 15, 251–263. Brown, M. & Sienaert, A., (2019) Governance improvements and sovereign financing costs in developing countries. World Bank, No . 14 Canuto, O., Mohapatra, S. & Ratha, D. (2011) ‘Shadow Sovereign Ratings’, World Bank Poverty Reduction and Economic Management (PREM) Economic Premise Note, No.63 Connolly, M. (2007) ‘Measuring the Effect of Corruption on Sovereign Bond Ratings’, Journal of Economic Policy Reform, 10:4, 309-323 Eichengreen, B. & Mody, A. (1998) ‘What Explains Changing Spreads on Emerging-Market Debt: Fundamentals or Market Sentiment?’, NBER Working Paper No.6408 Gonzalez, C. & Nishiuchi, T. (2018) ‘Revisiting the Relevance of the World Bank’s Country Policy and Institutional Assessment (CPIA) on Economic Growth’, MTI Global Practice Discussion Paper No.9 Hanusch, M. & Vaaler, P. (2015) ‘Credit Ratings and Fiscal Responsibility’, MFM Global Practice Discussion Paper No.4 Haque, T., Bogoev, J., & Smith, G. (2017) ‘Push and Pull: Emerging Risks in Frontier Economy Access to International Capital Markets’, World Bank MFM Global Practice Discussion Paper No.17 Moody’s Investors Service (2018) ‘Rating Methodology: Sovereign Bond Ratings’ Toksöz, M. (2014) ‘Guide to Country Risk: How to identify, manage and mitigate the risks of doing business across borders’. 22