Category: Sovereign Debt

Angola, Pakistan, Egypt, Jordan, Argentina, El Salvador, Ecuador, and Belize are among the market-access countries most at risk of sovereign stress, according to the model presented below. Unsurprisingly, several advanced economies appear least at risk, including Norway, Ireland, Denmark, Singapore, the Netherlands, Luxembourg, Hong Kong, and Switzerland.

Earlier this year I published the high-level initial results of a sovereign debt stress tracker, based on a model developed by the International Monetary Fund for countries that it classifies as having access to international markets. The IMF presented this model as part of its update to its Debt Sustainability Framework for Market-Access Countries in 2021, claiming at the time that it had performed significant robustness checks to ensure forecast salience. Time will tell how useful this tool is in predicting sovereign debt strains, and, in any case, it should only be used in conjunction with other analytical approaches.

Heatmaps

Using the latest available data for 2023, the heatmaps below rank order countries by the probability of experiencing sovereign stress, as represented by the column farthest to the right. Neither the probabilities for the dependent variable nor any of the raw data readings for any of the independent variables is shown below. Instead, readers can see the percentile rank compared to the maximum value in each variable column, which is beneficial for visually detecting relative heat for each indicator.

Lighter colors represent more risk, while darker colors represent less risk. Independent variables with negative coefficients, i.e. are negative predictors of sovereign stress, have been reversed in order to ensure color scheme coherence. These include institutional quality, the current account, and international reserves.

The first heatmap below suggests that Angola, Pakistan, Egypt, Jordan, Argentina, El Salvador, Ecuador, and Belize are most at risk of experiencing sovereign debt strains. Looking across the independent variables for this group of countries:

• They generally suffer from high external public debt burdens and from relatively poor institutional quality, though Argentina and Jordan fare better on those measures, respectively.
• El Salvador is penalized relatively less on stress history, though this assumes spread widening in recent years remained under the IMF’s stress definition threshold (see “Model” section below).
• One-year changes in general government debt in Angola, Egypt, and Argentina point to potential risks.
• El Salvador, Jordan, and, to a lesser extent, Pakistan, appear to need some replenishing of their international reserve buffers.
• Angola and – to a lesser extent – Argentina are marked down for surging REERs.
• Pakistan and Egypt display relatively concerning public debt/revenue ratios.
• Jordan stands out for poor current account performance.
• Egypt, Jordan, and Ecuador exhibit high credit-to-GDP gaps, though several other countries fare worse on this measure.

Each value is divided by the maximal value in that column, resulting in its own empirical percentile. Each value shown is the percent of observations with that value or below it. Sources: IMF, WGI, WB, Bruegel, BIS, author’s calculations.

The second heatmap uses foreign currency general government debt to replace the external PPG debt indicator featured in the first heatmap (see explanation in “Data” section below). Neither of these indicators is ideal, as in both cases coverage for many countries is either lacking or data points are equal to zero. This is obvious in both heatmaps from the absence of dark-colored cells in the relevant column, meaning that many countries are zero. Overall country coverage on this variable is better in the first heatmap, but the second one provides value for countries where data is missing in the first one (e.g. Israel, Korea, Sweden).

The eight countries most at risk of sovereign stress in this second heatmap are the same as in the first one, albeit in a slightly different order and except for Mexico replacing Belize. On this latter point, FX general government debt data – sourced from the BIS (see “Data” section below) – is missing for Belize, conferring on it an unfair advantage over Mexico and other countries where data are present for this indicator. In the first heatmap, external debt data is present for both Mexico and Belize, with the latter appearing more at risk than the former.

Each value is divided by the maximal value in that column, resulting in its own empirical percentile. Each value shown is the percent of observations with that value or below it. Sources: IMF, WGI, WB, Bruegel, BIS, author’s calculations.

Interpretation

Focusing on a country case helps illustrate ways to interpret the data in this model. Take Angola, as it appears most at-risk. Using heatmap (1), the brightest and thus most concerning data points are in the institutional quality, REER 3-year change, general government debt 1-year change, and external public and publicly-guaranteed debt columns. This suggests that the government and public sector more broadly are borrowing heavily, while prices and the exchange rate have also combined to rise quickly. Moreover, the institutions to set a good policy framework appear to be lacking. This is already a dangerous mix.

On the other hand, Angola scores well on its current account balance and international reserves variables. This is easily explained by the fact that the country is an oil exporter, thereby keeping its current account balance high and accumulating foreign reserves from the proceeds of these oil sales to buyers abroad.

While these oil exports provide Luanda with ample benefits, heavy reliance on a commodity-based export sector is also a double-edged sword. The result is often an appreciation of the exchange rate, making the economy less competitive for developing other industries: a classic case of Dutch Disease.

More concerning still is the presence of high inflation. The country’s surging REER variable already suggests that prices are probably rising, as the overall increase is unlikely to be due to nominal exchange rate dynamics alone. Increases in government debt suggest potential fiscal profligacy, which can lead to undesirably-high inflation, the presence of which is confirmed by a glance at recent Angolan statistics. The credit-to-GDP gap, which measures the deviation from trend of credit to the non-financial private sector as a share of GDP, is not particularly alarming in Angola, but may be high enough to also be contributing to the rising price level.

Angola also exhibits a high public debt-to-revenue ratio, which is worrying, given all the oil revenues that the country is seemingly raking in, suggesting that less borrowing and more fiscal discipline are likely needed. Recent sovereign stress is also a concern, indicating that, for all its natural resources, the government is unable or unwilling to pursue policies required to maintain macroeconomic stability.

Model

To recap, the model’s dependent variable is the probability of sovereign stress, which the IMF has detailed criteria for defining – running the gamut from outright default to a mere spread widening beyond a certain threshold. Regarding the independent variables:

• The first two represent how recently a country has experienced sovereign stress, and its government effectiveness and regulatory quality.
• Other explanatory variables are macroeconomic in nature, including current account balances, real effective exchange rates – which also capture price changes, credit gaps to the private sector, and international reserves.
• More specifically fiscal indicators include those on general government debt, foreign currency public debt, and public debt-to-revenue ratios.
• With the exception of REERs and debt-to-revenue, these macro-fiscal indicators are all expressed as a share of GDP.
• A global variable also features in the model, the VIX Index, which measures stock market volatility in the US, but is not presented in the heatmaps above, given its constance across countries.

Data

In the first iteration of the tracker, 2023 data was captured for 43 market-access countries, including both emerging markets-developing economies and advanced economies. Thanks to more available data for this year and refinements in data capture, coverage has been expanded to 82 MACs in these heatmaps.

Two similar heatmaps are presented in this article, with a difference in one of the independent variables and, as a result, slight changes to the overall results in the dependent variable. One of the IMF’s indicators is foreign currency public debt. In the first instance, the World Bank indicator for external public and publicly-guaranteed debt is used as the best available proxy for the IMF’s variable. While using this data from the World Bank remains the best possible option at this stage, there are some glaring omissions in coverage. For instance, the World Bank source suggests that Israel’s external PPG debt is equal to zero, which is clearly incorrect.

As a remedy to the World Bank’s data deficiencies, a second heatmap applies data from the Bank of International Settlements on foreign currency general government debt, as a proxy for this indicator in the same overall model. The BIS data does fill in some of the World Bank gaps – e.g. Israel – but in fact covers fewer countries than the first source. As such, the first heatmap is still preferable.

It is also worth noting that both the World Bank and BIS indicators differ from the IMF variable of foreign currency public debt. In the former case, external public debt differs somewhat from foreign currency public debt, even if virtually all external debt is in foreign currency. In the latter case, foreign currency general government debt excludes some types of debt that is covered under foreign currency public debt.

As part of tracking the probability of credit stress among 112 sovereign issuers, one of the variables of interest in the IMF’s model is the credit-to-GDP gap. This indicator matters not only because of its predictive power for sovereign credit events but for many other reasons as well, including monetary policy transmission and government borrowing. When analyzed in conjunction with other data, credit gaps are useful barometers for detecting the presence of credit bubbles, economic over- or under-heating, and policy distortions such as financial repression and fiscal dominance.

Credit gaps are derived from observations of credit extended to the private sector as a percentage of GDP, and then a statistical technique, usually Hodrick-Prescott filtering, is used to smooth out the data points in the time series as a way to measure an underlying trend. While there are some shortcomings to this approach, including the arbitrary nature of the smoothing parameters used to identify the trend, it helps ascertain whether the cyclical component of lending is above or below long-term expected processes.

A credit-to-GDP gap is thus an actual observation at time t minus the trend in the same time period. As such, a positive credit gap is one in which lending is above trend, whereas a negative one is below. The credit gap coefficient in the IMF model is positive, as is the case in other models in the financial crisis academic literature, meaning that higher values are associated with an increased likelihood of sovereign debt strains. Looking at quarterly data through end-2022¹2023 data will be presented in future posts on credit gaps. from the Bank of International Settlements on 43 countries plus the Euro Area, this first set of charts presents the actual credit-to-GDP ratios in blue, the smoothed trend in yellow, and the credit-to-GDP gaps in green.

In the chart above, credit gaps in this mix of developed (DMs) and major emerging market economies (EMs) are currently mostly negative. This makes sense given the tighter monetary policy stances around the world to combat inflation, with EM central banks having begun their rate-raising cycles²Some EM central banks are so far ahead of DM that they have already begun cutting rates in 2023. before their DM peers. A spike in credit gaps can also be observed in 2020 as policymakers worldwide lowered interest rates and facilitated the extension of credit as part of emergency measures to mitigate economic scarring at the height of the pandemic.

The BIS credit gap data is extremely useful for this set of countries, which, after all, comprise the world’s largest economies, and all the more so because it is available on a quarterly basis. The BIS describes its credit-to-GDP ratio as capturing total borrowing from all domestic and foreign sources by the private non-financial sector.³https://www.bis.org/statistics/about_credit_stats.htm However, other sources are needed for measuring credit to the private sector in other countries, and, thankfully, the World Bank has a similar indicator: domestic credit to the private sector by banks.⁴ https://databank.worldbank.org/metadataglossary/jobs/series/FS.AST.PRVT.GD.ZS The World Bank data series has far broader country coverage than the BIS data, thus opening vast additional swathes of the world to analytical coverage.

In contrast to the BIS data with its inclusion of both domestic and foreign credit to the non-financial private sector, the World Bank indicator appears to only include domestic sources of financing. Moreover, the BIS data appears to include sources of non-bank financing, unlike the World Bank data. Taken together, these two differences likely explain much of the discrepancy between these two datasets. Further, the World Bank data appears to only be available at a yearly frequency, thus requiring the BIS quarterly data to be transformed to yearly averages for the purposes of comparison.

The charts below present BIS data in blue and World Bank data in yellow, in yearly form through end-2022 in both cases. As seen above, the BIS provides credit gap and trend data alongside its credit ratios and uses a one-sided Hodrick-Prescott filter with the smoothing parameter λ set to 400,000 for this quarterly data. The World Bank only provides its credit ratios on a standalone basis, meaning that the trend and credit gap need to be estimated independently. This is simple enough for one country, and thankfully panel statistical techniques enable scalability for quick estimation across a large number of countries and years. As such, trends and credit gaps are derived from the World Bank’s credit ratios using a two-sided HP filter with λ = 100, the recommended setting for annual data.

Consistent with the inclusion of foreign sources of credit, the BIS credit ratios are usually higher than those from the World Bank, especially in many European countries, e.g. Luxembourg and Belgium. Elsewhere, the figures track more closely, as is the case with Japan, Malaysia, and the UK. The US and China also fell into this category, but the datasets have diverged in recent decades for those countries. Surprisingly, there are also a few countries where the World Bank ratio exceeds the BIS reading, despite the former excluding foreign credit sources, with South Africa and the US standing out most prominently from this perspective.

The point of comparing the two datasets is to use the BIS as a benchmark to get a sense if the World Bank data is at least somewhat aligned with the former and it is any good for predictive purposes. Certainly, the similar characteristics of the actual and trend data above are a positive sign. As for the credit gaps themselves, the BIS and World Bank figures are presented below. While there are large differences in most countries, there are also similar processes at work in many countries, e.g. the United Kingdom, Malaysia. The BIS credit gaps appear to be more volatile than those of the World Bank, which could be explained by the former’s inclusion of foreign lending: capital flows of the portfolio variety, which includes debt, are prone to sudden stops and starts.

To simplify further, the difference in the BIS and World Bank credit gaps, where the former is subtracted from the latter (difference = WB – BIS), features in the chart above. Ideally, the data readings would all be horizontal lines at zero or at least resemble a stationary process hovering above and below zero. While some countries do have these features – Sweden, the UK, the US, and Switzerland, among others, a large cohort exhibits some sort of bias. A statistical test of this difference in credit gaps across this panel of countries over these years would likely reject the notion that the difference is equal to zero. Nevertheless, the World Bank domestic credit to private sector by banks indicator seems fit for purpose, particularly given the large role that domestic banks play in credit provision in most economies.

Future posts will expand further on the importance of credit gaps and present broad country coverage of World Bank credit gap data.

• 1
  2023 data will be presented in future posts on credit gaps.
• 2
  Some EM central banks are so far ahead of DM that they have already begun cutting rates in 2023.
• 3
  https://www.bis.org/statistics/about_credit_stats.htm
• 4
  https://databank.worldbank.org/metadataglossary/jobs/series/FS.AST.PRVT.GD.ZS

Introducing a sovereign stress tracker covering 100+ countries, based on the IMF’s Debt Sustainability Framework for Market-Access Countries. The model used in this analysis suggests that sovereign debt strains are lower in 2023 than they were in either 2022 or 2020 for this group of countries. MACs comprise all economies that are lower-middle income and above, including many emerging economies and all advanced economies.

Market-Access Countries

In 2021, the IMF released its new Debt Sustainability Analysis framework for Market-Access Countries, in line with its differentiation between MACs and low-income countries. The reasons given for distinguishing between these two groups is that MACs generally have significant access to international capital markets, whereas LICs rely on concessional resources to fulfill their external financing needs.

As such, the Fund has a separate approach to debt sustainability analysis for LICs, which is beyond the scope of this tracker. The strict definition is that countries eligible for the IMF’s Poverty Reduction and Growth Trust, which is an interest-free concessional financing tool, are treated as LICs, whereas the rest are considered MACs.

Geographic coverage

Overall, 140+ countries and territories were included in this analysis, but results were only obtained for 112,¹Angola, Albania, United Arab Emirates, Argentina, Armenia, Antigua & Barbuda, Australia, Austria, Azerbaijan, Belgium, Bulgaria, Bahrain, Bahamas, Bosnia & Herzegovina, Belarus, Belize, Bolivia, Brazil, Barbados, Brunei, Botswana, Canada, Switzerland, Chile, China, Colombia, Costa Rica,  Cyprus, Czechia, Germany, Denmark, Dominican Republic, Algeria, Ecuador, Egypt, Spain, Estonia, Finland, Fiji, France, Gabon, United Kingdom, Georgia, Equatorial Guinea, Greece, Guatemala, Hong Kong SAR, China, Croatia, Hungary, Indonesia, India, Ireland, Iran, Iraq, Iceland, Israel, Italy, Jamaica, Jordan, Japan, Kazakhstan, St. Kitts & Nevis, South Korea, Kuwait, Lebanon, Sri Lanka, Lithuania, Luxembourg, Latvia, Morocco, Mexico, North Macedonia, Malta, Mongolia, Mauritius, Malaysia, Namibia, Nigeria, Netherlands, Norway, New Zealand, Oman, Pakistan, Panama, Peru, Philippines, Poland, Portugal, Paraguay, Qatar, Romania, Russia, Saudi Arabia, Singapore, El Salvador, Suriname, Slovakia, Slovenia, Sweden, Eswatini, Seychelles, Syria, Thailand, Trinidad & Tobago, Tunisia, Turkey, Ukraine, Uruguay, United States, Venezuela, Vietnam, South Africa given insufficient data availability in around 30 cases. The analysis is based on a multivariate model, meaning that a missing data point for a single variable across all years makes it impossible to derive a final measurement for the country in question, resulting in exclusion.

The calculated probabilities of sovereign stress for the 112 countries do not cover all years, unfortunately. For instance, there are only results for 43 countries in 2023, given less availability of annual data and/or forecasts for the current year. Data coverage will be improved in future iterations of the tracker.

All countries included are either high, upper middle, or lower middle income countries, with few exceptions, such as Syria, which the World Bank reclassified as a LIC in 2018. There is also some debate as to whether Venezuela constitutes an UMIC or a LMIC, though it is treated as a LMIC here.

Model

The IMF claims that extensive testing demonstrates that its new MAC DSF is much better at accurately predicting sovereign debt distress. Predictive analysis is based on a multivariate logit model developed by Fund staff. Passing the required data into the model provides a probability that a sovereign borrower experiences debt stress:

Multivariate logit model specification

Results

In addition to the dotplot chart above, a further way to view the broad results from this analysis of 112 countries is as a boxplot, presented below. I fully acknowledge that this data is unbalanced, given the limited number of data points in 2023 and also in the early 2000s – as can be seen in the first chart above – compared to better country representation in the middle years of the sample. More charts are presented in the next section below in order to address this issue.

As can be seen in the data, in 2023 there appears to be less systemic sovereign stress among MACs as compared to previous years, particularly 2022 and 2020. Future posts will provide granular details and heatmaps at the country level.

• 1
  Angola, Albania, United Arab Emirates, Argentina, Armenia, Antigua & Barbuda, Australia, Austria, Azerbaijan, Belgium, Bulgaria, Bahrain, Bahamas, Bosnia & Herzegovina, Belarus, Belize, Bolivia, Brazil, Barbados, Brunei, Botswana, Canada, Switzerland, Chile, China, Colombia, Costa Rica,  Cyprus, Czechia, Germany, Denmark, Dominican Republic, Algeria, Ecuador, Egypt, Spain, Estonia, Finland, Fiji, France, Gabon, United Kingdom, Georgia, Equatorial Guinea, Greece, Guatemala, Hong Kong SAR, China, Croatia, Hungary, Indonesia, India, Ireland, Iran, Iraq, Iceland, Israel, Italy, Jamaica, Jordan, Japan, Kazakhstan, St. Kitts & Nevis, South Korea, Kuwait, Lebanon, Sri Lanka, Lithuania, Luxembourg, Latvia, Morocco, Mexico, North Macedonia, Malta, Mongolia, Mauritius, Malaysia, Namibia, Nigeria, Netherlands, Norway, New Zealand, Oman, Pakistan, Panama, Peru, Philippines, Poland, Portugal, Paraguay, Qatar, Romania, Russia, Saudi Arabia, Singapore, El Salvador, Suriname, Slovakia, Slovenia, Sweden, Eswatini, Seychelles, Syria, Thailand, Trinidad & Tobago, Tunisia, Turkey, Ukraine, Uruguay, United States, Venezuela, Vietnam, South Africa