Revisiting the Debt Sustainability Framework for Low-Income Countries
Author: International Monetary Fund
Publication Date:
01 Dec 2012
eISBN:
9781498341028
Language:
English
Keywords:
PP; risk rating; debt service; debt distress; interest rate; rate of return; external debt; debt projection; debt vulnerability; debt relief
Download PDF (1.7 MB)

Introduced in 2005, the joint IMF-World Bank Debt Sustainability Framework (DSF) is a standardized framework for conducting public and external debt sustainability analysis (DSA) in low-income countries (LICs). It aims to help guide the borrowing decisions of LICs, provide guidance for creditors‘ lending and grant allocation decisions, and improve World Bank and IMF assessments and policy advice. The framework was previously reviewed in 2006 and 2009. This paper provides a comprehensive review of the framework to assess whether it remains adequate in light of changing circumstances in LICs. It reviews the DSF‘s performance to date, presents the results of recent analytical work by IMF and World Banks staffs, and discusses a number of areas in which the framework could be improved.

Abstract

Introduced in 2005, the joint IMF-World Bank Debt Sustainability Framework (DSF) is a standardized framework for conducting public and external debt sustainability analysis (DSA) in low-income countries (LICs). It aims to help guide the borrowing decisions of LICs, provide guidance for creditors‘ lending and grant allocation decisions, and improve World Bank and IMF assessments and policy advice. The framework was previously reviewed in 2006 and 2009. This paper provides a comprehensive review of the framework to assess whether it remains adequate in light of changing circumstances in LICs. It reviews the DSF‘s performance to date, presents the results of recent analytical work by IMF and World Banks staffs, and discusses a number of areas in which the framework could be improved.

I. Introduction

1. A formal framework for conducting public and external debt sustainability analysis (DSA) in low-income countries (LICs) was put in place in 2005 (Box 1).1,2 The main objectives of the debt sustainability framework (DSF) are to:

  • Guide the borrowing decisions of LICs in a way that matches their financing needs with their current and prospective repayment ability, taking into account each country’s circumstances;

  • Provide guidance for creditors’ lending and grant allocation decisions to ensure that resources are provided to LICs on terms that are consistent with both progress toward their development goals and long-term debt sustainability;

  • Improve World Bank and IMF assessments, policy advice, and program design; and

  • Help detect potential crises early so that preventive action can be taken.

2. The objective of this paper is to review the DSF comprehensively and assess whether it remains adequate given the evolving needs of LICs. There is a large investment gap in LICs, particularly in infrastructure. Addressing it is critical to increasing potential growth. Aid is limited and LICs will likely rely increasingly on domestic resources and nonconcessional external borrowing. A wide range of DSA issues—some of which were raised in previous reviews of the framework—merit attention in this context.3 Should there be more emphasis on total public debt, and how can fiscal sustainability analysis be improved? Are the indicative thresholds on external public debt still appropriate? Does the DSF adequately reflect the growth dividend of public investment? Are risks around the baseline scenario, including contingent liabilities, adequately captured through stress testing? Can the DSA template be simplified to facilitate its use by country authorities?

The Debt Sustainability Framework for Low-Income Countries

The DSF, a standardized framework for analyzing debt-related vulnerabilities, was introduced in 2005 and reviewed in 2006 and 2009. Under the DSF, joint Fund-Bank DSAs are prepared for all PRGT-eligible, IDA-only countries. For PRGT-eligible countries that are not IDA-only, DSAs are prepared by Fund staff only.1

How the DSF works

The DSF consists of a set of indicative policy-dependent thresholds against which projections of external public debt over the next 20 years are compared in order to assess the risk of debt distress. Vulnerability to external and policy shocks is explored in alternative scenarios and standardized bound tests. The indicative threshold for each debt burden indicator depends on each country’s policy and institutional capacity, as measured by the World Bank’s Country Policy and Institutional Assessment (CPIA) index. The specific thresholds are as follows:

Debt Sustainability Framework: Indicative Policy-Dependent Thresholds

(Applicable to public and publicly guaranteed external debt)

Based on the assessment, one of four possible risk of debt distress ratings is assigned:

  • Low risk: All the debt burden indicators are well below the thresholds.

  • Moderate risk: Debt burden indicators are below the thresholds in the baseline scenario, but stress tests indicate that the thresholds could be breached if there are external shocks or abrupt changes in macroeconomic policies.

  • High risk: One or more debt burden indicators breach the thresholds on a protracted basis under the baseline scenario.

  • In debt distress: The country is already experiencing difficulties in servicing its debt, as evidenced, for example, by the existence of arrears.

The DSF also includes a public sector DSA, which assesses public domestic debt risks and overall fiscal sustainability. The risk of debt distress rating, however, is guided solely by an analysis of external public debt relative to the thresholds in the external DSA.

What is the CPIA?

The CPIA is an index of 16 indicators grouped into four categories: (1) economic management; (2) structural policies; (3) policies for social inclusion and equity; and (4) public sector management and institutions. Countries are rated on their current status in each of these performance criteria, with scores from 1 (lowest) to 6 (highest). The index is updated annually for all IDA-eligible countries, including blend countries.

1 Some PRGT-eligible countries are classified by the World Bank as middle-income countries. See http://data.worldbank.org/about/country-classifications/country-and-lending-groups.

3. This paper presents the results of recent analytical work and discusses a number of areas in which the framework could be improved. Its purpose is to seek the views of the Executive Boards on the merits of possible changes, bearing in mind that some options could have major operational implications, including, for instance, the way aid is allocated. The paper draws, where relevant, on the IMF staff paper on Modernizing the Framework for Fiscal Policy and Public Debt Sustainability Analysis, published earlier this year.4 The paper is organized as follows. Section II reviews the DSF’s performance to date and identifies areas that could be improved. Section III presents options for improving the analysis of total public debt and fiscal vulnerabilities. Section IV reconsiders the thresholds that guide the assessment of the risk of debt distress. Section V looks at improving the coverage of external debt. Section VI discusses ways to better capture the investment-growth nexus in the DSF. Section VII considers options for redesigning stress tests. Section VIII presents options for simplifying the framework. Finally, Section IX raises issues for Board discussion.

II. What are the Main Issues to Reconsider?

The DSF’s performance to date

4. Since the DSF was introduced in 2005, 367 DSAs have been produced for 73 different countries. Joint Fund-Bank DSAs have generally been produced for all IDA-only, PRGT-eligible countries on an annual basis. In some instances, more than one DSA was produced for the same country in a single year. In other cases, there are gap years where no DSA was produced. Since 2007, nearly all DSAs have contained both an external and a public sector DSA. Eighty-eight percent of DSAs have been published on the IMF or World Bank’s external website.

5. An explicit rating denoting the risk of external public debt distress has been assigned in all but 27 cases, most of which date back to the early years of the framework (Table 1). Nearly all cases of “in debt distress” have been pre-HIPC completion point countries with poor institutional capacity, as measured by the CPIA index (Table 2). Cases of high risk of debt distress have been evenly split between HIPCs and non-HIPCs with predominantly poor CPIA ratings. Most cases of low or moderate risk have been post-HIPC completion point and non-HIPC countries with medium to strong CPIA ratings.

Table 1.

DSA Risk Ratings by Year

Source: IMF

As of October 7, 2011.

Table 2.

DSA Risk Ratings by Country Characteristics

Source: IMF

6. Although suitably long data series do not exist to rigorously evaluate the accuracy of DSAs, a preliminary analysis suggests that DSA debt projections have not shown any evident bias. As illustrated in Figures 1 and 2, DSAs produced in 2006 and 2007 projected levels of external public debt to GDP in 2010 that fell short of actual 2010 levels in about half the cases and surpassed actual levels in the other half. In 60 percent of the cases, the difference between the actual level of debt in 2010 and the level projected in the baseline scenario of the 2006 or 2007 DSA was 10 percentage points or less. Large differences between actual and projected debt levels in HIPC cases reflect uncertainty about the timing of debt relief when the projections were made. These uncertainties, in turn, were related mostly to the timing of required policy actions. For non-HIPCs, the differences mostly reflect larger-than-anticipated macroeconomic shocks related to the global financial crisis.

Figure 1.
Figure 1.

PV of External Public Debt to GDP (2010) Difference between Projected and Actual Levels in Non-HIPCs 1/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/ Baseline projections made at the time of the 2006 or 2007 DSA, depending on availability.
Figure 2.
Figure 2.

PV of External Public Debt to GDP (2010) Difference between Projected and Actual Levels in HIPCs 1/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/ Baseline projections made at the time of the 2006 or 2007 DSA, depending on availability.

7. Since the DSF was introduced, debt sustainability in LICs has been broadly maintained. Only two non-HIPC LICs have experienced debt distress since 2005, both of which were already in debt distress at the time of their 2006 or 2007 DSA. An April 2010 IMF-IDA report concluded that, while the global financial crisis had had a significant impact on LIC debt vulnerabilities, the crisis was not expected to result in systemic debt difficulties across LICs, a finding corroborated by the lack of systemic evidence that debt vulnerabilities among LICs have intensified over the last 18 months.5

8. The DSF is used by a growing community of donors and lenders to help inform their financing decisions. Since 2005, IDA has used DSA risk ratings to determine the share of grants and loans in its assistance to each LIC (Box 2). Regional development banks have similarly geared their lending practices to DSA risk assessments.6 The Paris Club group of official creditors relies on DSAs in the context of debt restructurings under the Evian Approach, and member countries of the OECD Working Group on Export Credit and Credit Guarantees agreed in 2008 to take DSAs into account when providing official export credits to LICs.

9. The DSF has enabled IMF and World Bank staff to integrate fiscal and debt issues more effectively into their analysis and policy advice. The framework has raised the profile of fiscal and debt issues in LICs through the annual frequency of DSAs, the improved quality and transparency of assessments, and the comparability of DSAs across countries. With the introduction of IDA’s Non-Concessional Borrowing Policy in 2006 and the review of the IMF’s external debt limits policy in 2009, DSAs were integrated into frameworks that offer countries flexibility to borrow on nonconcessional terms, depending on their debt vulnerability, debt management capacity, and capacity to manage public resources.7

The International Development Association’s Grant Allocation Framework

IDA’s grant allocation framework was adopted during the IDA14 Replenishment agreement in mid-2005. Its objective is to proactively mitigate the risks of external debt distress revealed by the DSF. Under the framework, IDA provides grants to countries facing a high probability of debt distress. Eligibility for IDA grants is limited to IDA-only countries. IBRD/IDA blend countries and “gap” countries are not eligible for grants, irrespective of their external debt situation.1

Grant eligibility is determined by the assessment of the country-specific risk of external debt distress emerging from DSAs conducted under the DSF. For countries assessed to be at a low risk of external debt distress, IDA provides its financing on standard IDA credit terms (40-year maturity, including a 10-year grace period, leading to a grant element of over 60 percent). For countries assessed to be at a moderate risk of external debt distress, IDA provides 50 percent of its financing on standard IDA credit terms and 50 percent on grant terms. Countries assessed to be “in debt distress” or at a high risk of external debt distress receive all of their assistance on grant terms. To mitigate equity and moral hazard concerns, the grant portion of a country’s allocation is discounted by 20 percent.

Nineteen countries at high risk of debt distress received their entire FY2011 allocation on grant terms. Of total IDA FY 2011 commitments of US$16.3 billion, 17 percent was provided on grant terms.

1 “Gap” countries are IDA-only countries with a GNI per capita that has been above the operational cut-off for IDA eligibility for more than two consecutive years.

To what extent can the framework be improved?

10. Although experience with the DSF to date suggests that it has performed relatively well, the question is whether it remains suitable in light of changing circumstances. Public finance in many LICs, as well as the range of available financing options, has changed significantly since the DSF was introduced. Debt relief under the HIPC Initiative and MDRI has lowered debt vulnerabilities—on a sustained basis (Figure 3)—and created new borrowing space. Many LICs are seeking to exploit this borrowing space to finance public investment and are relying increasingly on borrowing on nonconcessional terms. External public debt, though still the main component of overall public debt, is not as dominant as it once was, mainly as a result of debt relief. Domestic debt is likely to grow in importance as domestic savings increase and governments seek to develop domestic debt markets. LICs will face new risks as the universe of creditors and debt instruments continues to expand.

Figure 3.
Figure 3.

External Public Debt to GDP for HIPC Post-Completion Point Countries 1/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/ Countries that reached HIPC com pletion point in 2006 or earlier. The post-com pletion point period includes debt relief from MDRI.

11. A range of stakeholders have provided feedback on the DSF since the framework was introduced. The IMF and IDA Executive Boards recommended further work in certain areas when the DSF was reviewed in 2006 and 2009. With the benefit of several years of experience with the framework, users and outside observers have identified other elements that could be improved. The main issues that have been raised are as follows:

  • Improving the analysis of total public debt and fiscal vulnerabilities. The discussion of total public debt (both external and domestic) has tended to be less rigorous than the discussion of external public debt, reflecting both data limitations and the fact that the DSA risk rating is based exclusively on external public debt levels. With domestic debt playing an increasingly important role in some countries, public debt sustainability requires more attention.

  • Reconsidering the thresholds. Policy-dependent thresholds for external public debt are at the core of the DSF and guide the assignment of risk ratings. Do the thresholds remain accurate predictors of debt distress in light of more recent data? Should thresholds be formally adapted to take into account workers’ remittances?8 How can the framework make better use of country-specific information? Should there be thresholds for total public debt in addition to external public debt, and if so, should they inform the risk rating?

  • Improving the coverage of external debt. The DSF has traditionally focused on public external debt while paying less attention to private external debt. Does the latter merit closer scrutiny? In cases where private external debt is large and poses risks, should this be reflected in the risk rating?

  • Accounting for the impact of public investment on growth. The DSF has been criticized by some observers for being overly conservative in its assessment of the risk of debt distress, thereby constraining LICs from undertaking the borrowing necessary to finance growth-enhancing investments. This criticism is not new, but with the newly gained borrowing space after debt relief, the stakes appear to have increased as LICs seek to finance infrastructure projects critical for achieving development goals. While not an issue of DSA design per se, but rather a matter pertaining to the macroeconomic assumptions used in DSAs, the link between debt-financed investment and growth is integral to the quality of DSAs.

  • Redesigning stress testing. An often-heard criticism is that stress tests in DSAs are too mechanistic. Key macroeconomic variables (e.g., real GDP, exports, inflation) are shocked one at a time, without allowing for feedback between variables. While standardized stress tests facilitate cross-country comparisons necessary for operational purposes, they can also lead to situations where the most relevant shocks for a given country are not captured. What can be done to improve stress testing in DSAs?

  • Simplifying DSAs. Producing a LIC DSA can be a time-consuming, data-intensive exercise that competes with other priorities. One of the purposes of the DSF is to enable country authorities to produce their own DSAs. However, very few LICs use the current template, owing to its complexity and associated data challenges. Is there scope for simplifying the template? Would it be useful to divide the DSA into different modules that can be used flexibly, depending on the needs and the capacity of the user?

III. Improving the Analysis of Total Public Debt and Fiscal Vulnerabilities

12. Although the DSF recognizes the importance of domestic debt, the framework emphasizes the risks associated with external public debt. All LIC DSAs must include both an external DSA and a public sector DSA. The external DSA covers external debt contracted by the public and private sectors; the public sector DSA covers external and domestic debt contracted by the public sector.9 The risk of debt distress rating is based solely on an analysis of external public debt in the external DSA. The central role of external public debt in the DSF stems from the fact that, historically, external public debt has been the largest component of debt in LICs and the largest source of risk.

13. The analysis of total public debt has tended to be less thorough compared to the analysis of external public debt. The disparity in the analysis is most common—and to be expected—in countries where domestic public debt is negligible, or where data is unavailable. In countries where domestic public debt is relatively important and data is available, the discussion of total public debt is generally more detailed but remains overshadowed by the discussion of external public debt.

14. The changing context in LICs suggests a need to strengthen the analysis of total public debt and fiscal vulnerabilities. Although external public debt remains the largest component of debt in most LICs, domestic debt is becoming more prominent. Domestic debt carries benefits (e.g., development of local financial markets, no exchange rate risk) but also costs (e.g., crowding out of private investment, incentives for financial repression). Compared to external debt, domestic debt tends to be more expensive and have shorter maturities (Box 3). Since the introduction of the DSF in 2005, the share of domestic public debt in total public debt has increased from 19 percent to 29 percent on average across all LICs, largely reflecting external debt relief under the HIPC Initiative (Figure 4). Domestic debt as a ratio to GDP has remained flat on average, but this masks the fact that, in some countries, the level of domestic debt is already substantial (Figure 5) or is growing rapidly. In 11 LICs, the domestic debt-to-GDP ratio has more than doubled since 2006, albeit from low levels in some cases (Figure 6). During the global financial crisis, LICs relied heavily on domestic sources to finance larger fiscal deficits. This trend is expected to continue as domestic financial markets develop. In more advanced LICs, market development could be fueled by growing foreign investor interest in domestically-issued local currency debt.

Figure 4.
Figure 4.

Domestic Debt to Total Public Debt and to GOP

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations
Figure 5.
Figure 5.

Domestic Debt to GOP (2010) 1/2/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/ Based on most recent LIC DSAs produced for each country In some cases , 2010 figures are projections2/ Countnes shown are all PRG T-eligible
Figure 6.
Figure 6.

Domestic Debt to GOP (2006 vs. 2010) 1/2/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/ Based on most recent UC DSAs pro duced for each country. In some cases,2010 figures are projections.2/Countries shown are all FRGT-eligibte.

15. We outline below a number of ways to improve the analysis of total public debt and fiscal vulnerabilities in LIC DSAs. These echo many of the recommendations made in the IMF staff paper on Modernizing the Framework for Fiscal Policy and Public Debt Sustainability Analysis. That paper highlighted the need to improve the analysis of fiscal policy and public debt sustainability in the wake of large, unanticipated increases in public debt in many market-access countries (MACs). While circumstances in MACs differ from those in LICs, the LIC DSF, like the MAC framework, needs to adapt to a changing environment. Possible ways to strengthen LIC DSAs include:

  • Explicitly discussing the financing mix assumptions, between domestic and external debt on the one hand, and concessional and nonconcessional debt on the other. Currently, DSAs foresee a declining reliance on domestic debt in the medium term. Is this realistic? An assumption of continuous borrowing on highly concessional terms needs to be explained, particularly if a country has already begun to borrow nonconcessionally.

  • Giving greater scrutiny to assumptions about large fiscal adjustments. Fiscal adjustment in LICs is often rendered more difficult by the need to address large infrastructure gaps, pressures stemming from important social needs, and shallow tax bases that limit the scope for increasing revenue. For these reasons, a large fiscal adjustment assumed in the DSA needs to be well justified.

  • Assessing the realism of interest rate and growth assumptions. A strongly negative interest rate-growth differential has been a key benign force for debt sustainability in LICs. Given the sensitivity of debt dynamics to interest rate and growth assumptions, the realism of these assumptions should be systematically assessed. Domestic debt financing underlying the interest rate assumption should be explicitly discussed. Growth projections should try to capture the impact of public investment on growth while being mindful of historical trends. A significant acceleration in growth compared to past performance needs to be justified, if possible using model-based analysis (see Section VI).

Costs and Risks of Domestic Debt Financing

An analysis of the debt structure in 12 LICs shows that the domestic debt portfolio is subject to substantial refinancing and interest rate risks, as only few countries have been able to issue long-term domestic debt.

The World Bank and the IMF are providing technical assistance to help LICs improve debt management through the formulation of medium-term debt management strategies (MTDS). An MTDS complements the LIC DSA and helps operationalize a country’s debt management objectives by outlining cost-risk tradeoffs in meeting the government’s financing needs and payment obligations.1

An analysis of 12 MTDSs elaborated by authorities of IDA-eligible countries in 2010–11 helps define the cost and risk indicators that characterize the domestic debt portfolio of LICs (see table). In the two years considered, the median interest rate on domestic debt was 9 times higher than on external debt. The sample median of the average time to maturity of the domestic debt portfolio was only 2.5 years, compared to an average time to maturity of more than 13 years for external debt, reflecting the latter’s concessional component. In addition, the share of domestic debt maturing within one year was much higher compared to external debt. Although most of the domestic debt was issued at fixed rates, the short maturity structure implies that most interest rates would refix every year.

Cost and Risk Indicators of the Debt Portfolio in 12 IDA-Eligible Countries

article image
Source: Medium-term debt management strategies prepared during 2010–11.
1

Interest payments in 2010 or 2011 divided by the debt stock at the end of the previous year, in local currency.

2

Average of the years of repayment w eighted by the share of principal payments in the debt portf olio.

3

Domestic (external and total) debt maturing in one year in percent of domestic debt (external and total), respectively.

4

Average time until all principal payments in the debt portf olio become subject to a new interest rate.

5

Domestic (external and total) debt refixing in one year in percent of domestic debt (external and total), respectively.

6

Percent of debt issued at a f ixed rate, f or any maturity.

The current debt structure of LICs is subject to substantial refinancing risk. Few LICs have been able to issue domestic debt at maturities longer than 10 years. As domestic markets in LICs are usually shallow, with few participants (mainly from the banking sector), a short maturity structure increases refinancing and interest rate risks, particularly in the event of negative shocks to the economy, or if government domestic borrowing needs increase abruptly.

Developing the domestic debt market through the introduction of instruments with longer maturities and expanding the domestic creditor base could lower refinancing and interest rate risks, but also comes at a cost for governments, depending on the liquidity and inflation risks perceived by domestic debt market participants.

1 See “Developing a Medium-Term Debt Management Strategy (MTDS)— Guidance Note for Country Authorities,” February 24, 2009.

  • Strengthening the analysis of contingent liabilities. Data on contingent liabilities in LICs tends to be scarce. The current framework includes one stress test—a 10-percent-of-GDP increase in debt-creating flows—that resembles a generic contingent liability shock. Where information is available, a more country-specific approach may be warranted to capture contingent liabilities arising from state-owned enterprises (SOEs), public-private partnerships (PPPs), and weaknesses in the financial sector. 10

  • Considering tail risks. Stress tests in the DSF are intended to capture the most likely risks to debt sustainability. Enhancing the analysis of tail risks—that is, low probability events with potentially severe consequences, such as the impact of severe crisis or climate-related risks—could complement the current focus on more likely risks. Since tail risks represent events that, by definition, are unlikely to occur, they should not inform the risk rating. Nevertheless, the sudden increase in debt levels in many advanced economies in recent years is a reminder that tail risks should not be ignored altogether.11

  • Expanding the analysis to cover all obligations of the public sector. The DSF covers, inter alia, public and publicly-guaranteed external debt and thus should include debts from the central government, regional and local governments, the central bank, and public enterprises. While most borrowing in LICs is done at the central government level, some regional and local governments as well as SOEs have been able to either borrow externally or issue debt domestically. DSAs should strive to incorporate all debt, including domestic arrears, which can be substantial but are often not reported. However, given data limitations and capacity constraints, initial efforts should focus on encompassing debt of entities that may pose the largest fiscal risk.

  • Assessing the risks associated with the public debt profile. For more advanced LICs with a high level of debt contracted on market terms, both the level and the profile of the debt are relevant when assessing vulnerabilities. In such cases, a fuller discussion of the risks linked to debt structure characteristics—including maturity, currency composition, and the creditor base—is warranted. This analysis would consider cost and risk indicators of the domestic debt portfolio, as discussed in Box 3.

16. Similar to the approach adopted for MACs, the depth and breadth of the analysis should be tailored to identified country-specific risks. Under such a risk-based approach, the DSA should focus on those vulnerabilities deemed most relevant by country teams for which information is available. Adding benchmarks for total public debt would help frame the discussion and indicate when to conduct a more detailed assessment of public debt vulnerabilities. This idea is discussed in more detail in the next section.

IV. Revisiting Thresholds

A. Current Thresholds on Public and Publicly-Guaranteed External Debt

17. Thresholds for public and publicly guaranteed (PPG) external debt rest on empirical analyses by Kraay and Nehru (KN) and IMF and World Bank staff (Staff 2004).12 Thresholds for the present value (PV) of debt to GDP, debt to exports, and debt to revenue were calibrated using Staff 2004, while thresholds for debt service to exports and debt service to revenue were calibrated using KN. Both KN and Staff 2004 used a probit model to explain the probability of distress on PPG external debt as a function of (i) a country’s debt burden; (ii) the quality of a country’s policies and institutions, as measured by the CPIA index; and (iii) real GDP growth.13

18. The methodologies used by KN and Staff 2004 to estimate the probability of debt distress, while broadly similar, differed in certain technical respects (Box 4). In particular, KN and Staff 2004 defined debt distress and non-distress episodes differently, resulting in different estimation samples and different frequencies of debt distress episodes within the samples. KN used a more stringent definition of debt distress and had fewer debt distress episodes in their sample. Nevertheless, the studies came to similar conclusions: countries with higher debt burdens are more likely to experience debt distress, and countries with strong policies can sustain a higher debt burden than those with weak policies.

Comparing Previous Methodologies Used to Calibrate Thresholds for PPG External Debt

The calibration of debt burden thresholds based on Staff 2004 and KN differ in certain technical aspects: First, different definitions of debt distress and non-distress episodes were used. Staff 2004, which used data on LICs only, defined a non-distress episode as at least three consecutive years in which the stock of arrears to official creditors was smaller than five percent of the total debt stock. All other years were defined as debt distress episodes. The KN study, which used data for both LICs and middle-income countries, relied on three distress signals: arrears on PPG external debt to all creditors (official and private), a Paris Club rescheduling, and IMF GRA financing on a commitment basis. KN characterized a debt distress episode as a period of three years or longer during which at least one of the three distress signals was observed. A non-distress episode was defined as a non-overlapping five-year period without any debt distress signal. As a result of these definitional differences, the unconditional probability of debt distress (i.e., the frequency of debt distress episodes observed in the sample) was higher in Staff 2004 than in KN.

Second, the choice of the probability of debt distress used to calibrate the thresholds differed.

In KN, the probability of debt distress was roughly in line with the unconditional probability of debt distress (25 percent). By contrast, Staff 2004 used a probability of debt distress (around 20 percent) that corresponded to the median PV of debt-to-GDP ratio observed in LICs immediately prior to an outbreak of debt distress (43 percent of GDP).

Thresholds under the DSF

(Applied to external public and publicly-guaranteed debt)

article image

B. Re-estimating Thresholds for PPG External Debt

19. IMF and World Bank staff updated the thresholds for PPG external debt based on more recent data, and using a single methodology applied consistently across all debt burden indicators (Annex 1). This exercise provided staff with the opportunity to address several technical issues, including controlling for the presence of middle-income countries in the estimation sample, using debt service data on PPG external debt instead of total external debt, and using consistent definitions of debt distress and non-distress episodes for all regressions. As in the current DSF, thresholds were calculated for three values of the CPIA (3.25, 3.5, and 3.75) associated with weak, medium, and strong performance in terms of the quality of policies and institutions.

20. The thresholds were calibrated using different methods to fix the probability of debt distress. As in previous analysis, the calibration of the thresholds rests on the historical values of the variables used in the analysis as well on the choice of the probability of debt distress. Staff derived thresholds using three different concepts of probability of debt distress: (1) the unconditional probability of debt distress; (2) the probability of debt distress corresponding to the median value of the relevant debt burden indicator immediately prior to an outbreak of debt distress; and (3) the probability of debt distress that minimizes the number of missed crises and false alarms.14 The first two methods replicate what was used in KN and Staff 2004, respectively. The third method, preferred by staff, balances the two possible types of errors produced by the model, thus ensuring that the resulting thresholds are neither too permissive nor unduly conservative. Under this method, the probability of debt distress ranges from 13 to 15 percent, depending on the debt burden indicator. The other two methods yield probabilities ranging from 11 to 16 percent and lead to thresholds that are broadly similar.15 The full results of all three methods are presented in Annex 1.

21. The re-estimated thresholds are roughly in line with the current DSF thresholds, with the exception of the threshold for debt service to revenue. Table 3 shows re-estimated thresholds based on staff’s preferred method of minimizing the number of missed crises and false alarms. The re-estimated thresholds for the PV of debt to GDP and debt service to exports are fairly close to the current ones. The re-estimated thresholds for the PV of debt to exports are slightly higher than in the DSF, while the thresholds for the PV of debt to revenue are slightly lower. In the latter two cases, the differences are not large enough to warrant a change to the thresholds, in staff’s view. However, the re-estimated thresholds for debt service to revenue are significantly lower (in percentage terms) than current thresholds. Staff proposes lowering these thresholds to 18, 20, and 22 percent from the current values of 25, 30, and 35 percent. The results across debt burden indicators are robust to different measures of governance and macroeconomic shocks. Notwithstanding the proposed adjustment to the debt service-to-revenue thresholds, the re-estimated thresholds strongly support the main conclusions of KN and Staff 2004—namely, that countries with higher debt burdens are more likely to experience debt distress, and countries with strong policies can sustain a higher debt burden than those with weak policies. Staff estimates that only 2 out of 66 countries would receive a higher risk rating (moving from medium to high) if the proposed debt service to revenue thresholds were applied.16

Table 3.

Re-estimated External Public Debt Thresholds

(Applied to external public and publicly-guaranteed debt)

article image

Thresholds currently in the DSF.

Updated thresholds calibrated using the probability minimizing type I and II errors.

22. The sensitivity of the results to definitional changes is a reminder that judgment should be used when interpreting breaches of thresholds. As discussed in Annex 1, re-estimated thresholds vary to some extent depending on the definition of debt distress and non-distress episodes. While the overall results largely validate existing thresholds, they also highlight that a balance should be struck between due attention to debt levels rising above thresholds and the need for judgment when assessing the risk of distress. A marginal or temporary breach of a threshold may not necessarily imply a significant vulnerability. Conversely, a near breach should not be dismissed without careful consideration.

C. Including Remittances in External Debt Thresholds

23. Debt burden indicators discussed in the previous section focused on the typical measures of repayment capacity (GDP, exports, and revenues). However, remittances can also affect the probability of debt distress by enhancing a country’s capacity to repay its external debt.17

24. At the time of the last review of the DSF, Executive Directors agreed that remittances should be taken into account when assigning risk ratings.18 Following that decision, remittances were incorporated into the analysis without a formal re-estimation of the thresholds. Specifically, modified debt burden indicators—the PV of PPG external debt to the sum of GDP and gross remittances, the PV of PPG external debt to the sum of exports and gross remittances, and debt service to the sum of exports and gross remittances—were included in the analysis. The inclusion of remittances in the denominator lowers the debt burden indicators. Mirroring this decrease, indicative thresholds for countries with significant remittances were lowered by 10 percent. The new thresholds allowed countries with large remittances to carry higher levels of debt without breaching the indicative thresholds.

25. Staff re-estimated the basic econometric models, taking remittances explicitly into consideration.19 Due to the limited availability of data on remittances, the number of observations in the estimation sample was reduced significantly, thus affecting the results. Since it would be misleading to compare the re-estimated thresholds presented in the previous section (full sample) to ones incorporating remittances (small sample), staff compared thresholds excluding remittances to thresholds with remittances using a common (small) sample.

26. Explicitly including remittances in the model results in lower debt burden thresholds. After controlling for sample size, the threshold for the PV of debt to the sum of GDP and remittances is approximately 10 percent lower than the corresponding threshold without remittances, while export-based thresholds with remittances are roughly 20 percent lower than their counterparts without remittances (Table 4). This suggests that the current DSF threshold for the PV of debt to the sum of GDP and remittances is appropriate, while the export-based thresholds with remittances included should be 20 percent lower than the corresponding thresholds without remittances, instead of 10 percent lower.

Table 4.

Including Remittances: Impact on External Public Debt Thresholds

article image

27. For countries with large remittances, staff proposes to adjust the thresholds to reflect the updated analysis. Specifically, the export-based thresholds should be lowered by

20 percent compared to their counterparts without remittances, while the current threshold for the PV of debt to the sum of GDP and remittances would remain unchanged. For countries where remittances are large and represent a reliable source of foreign exchange, adjusted thresholds could be used, resulting in the possibility of carrying higher level of debts without breaching the thresholds. As under the current guidelines, judgment should be used when determining the impact of remittances on the external risk rating. This judgment should be informed, inter alia, by recent developments in remittances and their prospects for the future.20 The retroactive application of the proposed remittance-adjusted thresholds would lead to minimal changes in the assessment of the risk of debt distress.21

D. Use of Country-Specific Information

28. The current methodology to calibrate thresholds leads to a manageable number of thresholds, but at the cost of losing country-specific information. The methodology defines thresholds for each of the five debt burden indicators considered in the DSF (a “one-debt-variable-at-a-time” approach). Thresholds are derived by fixing real GDP growth at the sample mean for all countries, and by using one of three CPIA cutoff values (3.25, 3.5, or 3.75) corresponding to a country’s policy performance classification (weak, medium, or strong). Using discrete CPIA cutoffs gives rise to CPIA threshold effects, whereby a small change in a country’s CPIA score near the boundary of two policy performance categories results in a large shift in debt burden thresholds. The implied debt distress tolerance (as represented by debt burden thresholds) may thus be too high or too low depending on the divergence between a country’s actual CPIA score and the CPIA cutoff value used. The current methodology also makes no use of country-specific growth information in setting debt burden thresholds. The paragraphs below consider whether greater use of country-specific information is warranted.

29. The case for using country-specific growth rates is not compelling. Real GDP growth is included in the underlying model as a proxy for macroeconomic shocks to the economy. Recent growth performance is not a robust predictor of future shocks or of a country’s ability to sustain higher levels of debt in the medium term. Since country-specific growth rates already enter the framework through their impact on debt dynamics, using them to calibrate the thresholds would also risk overstating the impact of growth on debt sustainability, and could impart a pro-cyclical bias to the policy.

30. The benefits of using country-specific CPIA information have to be weighed against the operational complexity it would entail. Country-specific thresholds could be derived using country-specific CPIA scores rather than discrete cutoff values. This would more closely relate debt distress thresholds to country-specific assessments of institutional and policy quality. It would also eliminate CPIA threshold effects. However, country-specific thresholds imply a large number (5 times the number of countries) and a wide range of thresholds rather than the current set of 15. In discussions on the initial design of the DSF, the IMF and IDA Executive Boards had expressed concerns about an initial proposal by staff that generated a wider dispersion of thresholds than the figures that were ultimately agreed.22 Subsequently, the Boards also expressed little support for a proposal to mitigate CPIA threshold effects by increasing the number of thresholds from 15 to 20. Directors opted for an alternative approach that they deemed more manageable and easier to implement.23

31. An alternative approach for incorporating country-specific information would be to express thresholds in terms of the probability of debt distress (see Annex 1). In this case, the DSA would focus on the evolution of the probability of debt distress over time rather than the evolution of various debt burden indicators, thus more transparently conveying the probabilistic nature of the DSF. In addition to eliminating CPIA threshold effects, such an approach would allow for richer specifications and a fully country-specific assessment of risk. However, extensive econometric analysis has so far failed to yield an evidently superior approach to the one-debt-variable-at-a-time approach underlying the current framework. Given this, it is not clear that the potential benefits of a probability-based approach outweigh the costs involved in replacing a framework that has become widely adopted and understood by development partners and country authorities. On balance, therefore, staff does not advocate changing the basic architecture at this time.

32. Instead, staff sees substantial merit in strengthening the use of judgment based on country-specific information when assessing the risk of debt distress. This approach is also in line with the original intent of the DSF to provide a practical tool to help detect and measure potential debt-related vulnerabilities, rather than mechanically generate risk ratings. Staff proposes to reinforce this principle and to foster its more consistent application by developing clearer guidance on the use of judgment. For example, for countries with minor or temporary breaches of thresholds, staff would advocate taking into consideration whether their CPIA score is significantly higher than the cutoff value used to calibrate the threshold when determining the rating. Staff also proposes to carry out further analytical work on probability thresholds and the use of specifications that combine different debt measures, to assess whether such analysis could usefully complement the information provided by the five debt-burden indicators, particularly in cases where the latter give conflicting indications on the risk of debt distress. If it proves useful, the analysis based on probability thresholds could be incorporated as part of the toolkit for capturing country-specific information in the assessment of risks of debt distress, in which case staff would seek to reflect this in the revised guidance note.

E. Benchmarks for Total Public Debt

33. Introducing benchmarks for total public debt, in addition to existing thresholds for external public debt, could further strengthen public sector DSAs. In the absence of benchmarks, public sector DSAs have tended to focus somewhat simplistically on the projected path of public debt, often without drawing firm conclusions about the risk of debt distress. By comparison, in external DSAs, thresholds have anchored the analysis and helped country teams make judgments about the risk classification, based not only on the path but also the level of debt. In addition, thresholds have facilitated discussions with country authorities by clearly demarcating “danger zones” that are easily understood by broad audiences.

34. IMF and World Bank staffs have previously argued that benchmarks for total public debt are difficult to derive. The main empirical difficulty has been the absence of reliable and comprehensive historical data on domestic debt stocks (including arrears to suppliers) and associated debt service. A key conceptual challenge is determining when a domestic debt distress event has occurred. When governments face serious domestic debt difficulties, they often do not overtly default on domestic debt, but rather resort to other actions such as inflating debt away or financial repression. Domestic debt is also a highly heterogeneous concept across countries, which complicates cross-country comparisons.

35. While these considerations are still valid, the need to improve the analysis of public debt and the availability of new debt data present an opportunity to estimate benchmarks. Using the same probit model that underpins the thresholds for external public debt, IMF staff derived a set of policy-dependent benchmarks for public debt to GDP (Box 5 and Annex 2). This work draws on a new database on public debt compiled by the IMF’s Fiscal Affairs Department and on recent studies and surveys documenting episodes of domestic debt default. The results suggest that public debt to GDP in excess of about 40 to 70 percent in PV terms, or 50 to 75 percent in nominal terms, signals heightened debt vulnerabilities, with the actual benchmark varying according to the CPIA score. Similar to the case of external public debt, the regressions clearly show that the higher the level of total public debt, or the lower the quality of policies and institutions, the higher the risk of debt distress.

Estimation of Indicative Public Debt Benchmarks

The probit model used to derive thresholds on PPG external debt was expanded to include domestic debt. A sample was compiled covering the period 1971–2007 and containing 155 countries, of which 79 were classified as middle-income countries (MICs) at end-2007.1,2 Debt distress signals and debt distress episodes are defined as follows:

  • External debt distress signals are defined as (i) PPG external arrears in excess of five percent of external PPG debt; (ii) a Paris Club restructuring;3 and (iii) IMF GRA financing disbursed in excess of 50 percent of quotas.

  • Domestic debt distress signals are defined as instances of outright sovereign defaults – that is, the failure by the sovereign to meet a principal or interest payment on the due date (within a specified grace period). This includes debt restructurings (e.g., converting a note to a different currency of less than equivalent value, rescheduling payments at less favorable terms, discounting an obligation). Defaults are identified using studies by Reinhart and Rogoff, Standard and Poor’s, and Moody’s, as well as IMF staff reports.4,5 Data on domestic arrears is largely missing and thus was not systematically included in the definition of a domestic debt distress signal.

  • Debt distress episodes are defined as non-overlapping periods in which either an external or domestic debt distress episode occurs. External debt distress episodes are defined as a period lasting three years or more in which at least one of the external debt distress signals is observed. Any instance of domestic default constitutes a domestic debt distress episode. Normal time episodes are defined as non-overlapping periods of three consecutive years in which in which no public debt distress signal (external nor domestic) is observed. A total of 99 public debt distress episodes are observed. Most of the episodes are linked to either external events or domestic events that coincide with external ones.

Detailed results of this analysis can be found in Annex 2. The table below shows the implied policy-dependent public debt sustainability thresholds.6 These thresholds were calculated based on probabilities that minimize type I and type II errors for the mean of the public debt ratios.

Indicative Policy-Dependent Public Debt Benchmarks

article image
1 LICs are defined as IDA-only countries. A country’s income classification can change within the sample period. For example, a country can move from LIC to MIC status as it graduates from IDA.2 See IMF Fiscal Affairs Department’s Historical Public Debt Database for a description of data sources and methodology. Public debt is expressed in percent of GDP and is defined as gross general government debt, although in many cases (especially before 1980) only central government data are available.3 Paris Club restructurings are assumed to last three years, with the exception of completion point treatments, which are assumed to last only one year.4 See Reinhart and Rogoff (2009), Standard and Poor’s Sovereign Defaults and Rating Transition Data reports, and Moody’s Sovereign Default and Recovery Rates report.5 For countries not included in the aforementioned sources, IMF staff reports were used to determine any instance of domestic d ebt defaults.6 Thresholds were also estimated for public debt to revenue, but the results were not robust and therefore are not presented.

36. While the estimated benchmarks represent progress over past econometric efforts, the results should be interpreted with caution. The results are sensitive to the definition of debt distress and normal time episodes (i.e., the length of each). Despite improvements in the data, comprehensive data on domestic arrears still does not exist, which implies that the stock of domestic debt could be underestimated and debt distress events overlooked. Another reason to interpret the results with caution is the lack of homogeneity in the coverage of the public sector debt across countries. Moreover, the underlying probit did not perform significantly better when domestic debt was included as an explanatory variable, reflecting the fact that, in most cases, domestic debt events have coincided with external debt events.

37. The empirical analysis carried out by staff could, however, serve to define reference points for triggering a deeper discussion of total public debt. As noted earlier, these benchmarks could be used to determine when to conduct deeper analysis, thus providing an anchor to help frame the analysis. When total public debt moves toward, or exceeds, the relevant benchmarks over the projection horizon, a detailed discussion of potential risks to debt sustainability arising from high public debt levels would be expected— particularly in cases where external public debt indicators remained below their respective thresholds.

38. A key question is whether the assessment of total public debt should inform the risk rating. The risk rating is arguably the most important outcome of the DSA: it facilitates cross-country comparisons, provides critical guidance to lenders and donors on the appropriate terms of financing, and has implications for the grant-loan allocation decision of IDA and other multilateral institutions and for IMF program design through the Fund’s debt limits policy. The public sector DSA allows staff to provide detailed analysis of public debt, including domestic debt. That analysis can lead to a different conclusion from the assessment of external public debt only. However, this overall assessment of debt vulnerability does not currently affect the risk rating.24 The growing prominence of domestic debt in some countries, and the need to strengthen public sector DSAs, raises the question of whether the risk rating should continue to be determined by an assessment of external public debt only. If external public debt is low but domestic public debt is high, the external risk rating could send a misleading signal about overall debt sustainability.

39. Staff believes that risk ratings derived from external DSAs could be supplemented in cases where the public sector DSA identifies significant vulnerabilities. For the majority of LICs, where external public debt remains dominant, the risk rating would continue to be assigned by comparing the projected path of external public debt indicators against policy-dependent thresholds. For those LICs with total public debt moving toward or exceeding benchmarks, country teams would be expected to conduct in-depth analysis to determine the extent of domestic debt vulnerabilities. In the event that the analysis uncovered significant domestic debt vulnerabilities, an additional risk rating providing the overall assessment of debt vulnerability would be assigned. If the IDA and IMF Executive Boards agree with this approach, staff will develop detailed guidance.

40. The additional risk rating would not be a substitute for the risk rating on external public debt. Governments with high domestic debt vulnerabilities would need to design macroeconomic and structural policies to reduce these vulnerabilities and to avoid the negative consequences of excessive domestic debt on the economy. Maintaining or increasing access to concessional financing can be an important element to help governments implement the required policies. For this reason, the assessment of the risk of external debt distress would continue to inform the financing decisions of IDA, while the additional risk rating on the overall assessment of debt vulnerability would inform the macroeconomic and structural policy dialogue with country authorities.

V. Improving the Coverage of External Debt

41. External DSAs capture both public and private external debt, but in practice the analysis has focused almost exclusively on public external debt. This reflects both the dominant share of public external debt in total external debt in most LICs as well as limited information on private external debt. Nevertheless, an increase in external private investor interest in LICs, including for the financing of infrastructure, begs the question of whether a greater focus on private external debt is warranted. Increasing levels of private sector external debt from a low base, while generally a positive sign of growing business activity, could increase external debt vulnerabilities. High levels of private external debt could create balance of payments pressures by competing with the public sector for foreign exchange and could also increase exposure to risks stemming from the accumulation of contingent liabilities.

42. While private external debt is unlikely to pose an immediate concern in most LICs, some exceptions apply. Private sector external debt has remained broadly stable across LICs, compared to a growing trend in MICs (Figure 7). In many LICs, private external debt is negligible, or data is unavailable. In a sample of 70 LICs, only half report any private external debt in their most recent DSA. There are, however, some cases where private external debt is already substantial in relation to GDP (Figure 8), and one can expect the number of such cases to rise in the coming years.

Figure 7.
Figure 7.

Private External Debt to GDP 1/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/LICs are defined as PRGT-eligible countries. The sample includesonly those LICs where private external debt is reported. MICs are defined according to the World Bank's Global Development Finance classification, and excludes those countries that overlap with PRGT-eligible countries.
Figure 8.
Figure 8.

Private External Debt to GDP at end -2010 1/2/

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Source: IMF staff calculations1/ Countries shown are all PRGT-eligible.2/ Madagascar’s external debt-to-GDP ratio is based on estimates from the 2008 DSA.

43. The presence of high levels of private external debt in only a few LICs suggests that a country-specific approach would be appropriate. To assist teams in monitoring private external debt dynamics, additional charts could be added to the standard template showing, for instance, the pace of accumulation of private external debt or the path of total external debt. Where private external debt is significant, the DSA should discuss the risks to overall external debt sustainability, as exemplified in a few recent cases.25 In the event that the risks associated with private external debt were judged to be significant, they would be reflected in the additional risk rating denoting the overall assessment of debt vulnerability. As stated earlier, this overall assessment of debt vulnerability would not be a substitute for the risk rating on external public debt.26

VI. Strengthening the Analysis of the Public Investment and Growth Nexus

44. To achieve accelerated and sustained growth, LICs will require much higher investment, particularly in infrastructure. By raising productivity and encouraging private investment, closing the present large infrastructure gap could substantially increase rates of per capita income growth. According to a World Bank report, the cost of addressing Sub-Saharan Africa’s infrastructure needs is estimated at around US$93 billion a year, equivalent to 15 percent of the region’s GDP, or 22 percent of GDP for the region’s low-income countries.27 This raises the twin challenges of investing efficiently in infrastructure to get the biggest possible growth dividend and financing that investment in a sustainable manner.

45. In November 2011, the G20 leaders committed to help scale up and diversify sources of financing for infrastructure in LICs, particularly in Sub-Saharan Africa. Leaders endorsed the recommendations of the High Level Panel on Infrastructure to (i) increase the funds that multilateral development banks (MDBs) dedicate to facilities to help prepare and finance investments; (ii) build an enabling environment for private and public infrastructure financing, especially for regional projects; and (iii) improve access to funding, notably through the strengthening of local intermediaries and financial markets, and through more effective use of MDB capital, including the use of credit enhancement and guarantee instruments.

46. A recurring criticism of the DSF is that it does not adequately capture the benefits of debt-financed public investment. Proponents of scaling up public investment maintain that productive investment, while increasing debt ratios in the short run, can lead to higher growth, revenues, and exports—and therefore to lower debt ratios—over time. Some argue that LIC DSAs, by failing to take sufficiently into account the assets and future income that public investment may generate, lead to overly pessimistic risk assessments, which in turn discourage potential investors while constraining how much LICs can borrow in accordance with the Fund’s external debt limits policy and the Bank’s external non-concessional borrowing policy.

47. IMF and World Bank staffs have long recognized the importance of gaining a better understanding of the public investment-growth nexus. The issue was discussed in detail by the IMF and IDA Executive Boards in 2006 and 2009. Staff presented evidence that, despite the difficulties in assessing the growth effect of public investment, recent DSAs had attempted to take into account such effects. Although suitably long data series did not exist to systematically evaluate the criticism that growth projections in DSAs were too conservative, staff analysis comparing actual versus projected GDP growth for the period 2004–2008 did not reveal a tendency to underpredict growth in countries with high levels of public investment. Executive Boards saw the need for more work and agreed that analyzing the investment-growth nexus required a country-specific approach, using a broad range of indicators, supplemented with model-based approaches where appropriate.

48. Work on the investment-growth nexus is ongoing and goes beyond the scope of the DSF. Financing investments to help countries achieve higher and sustained growth is at the core of the World Bank development model. The growth dividend from additional debt-financed investment depends on a number of country-specific factors, including the quality of public investment, the crowding-in effect of public investment for private investment, and the capacity of the government to increase revenues to repay the initial debt.

49. IMF staff has developed a dynamic general equilibrium model that analyzes the linkages between public investment and growth and the implications for debt sustainability. The model is designed to include features and shocks that are common in LICs. It incorporates a production function with private and public capital, so productive government spending can raise output directly and crowd in as well as crowd out private investment.28 It takes into account potential inefficiencies in translating a dollar of public investment into a dollar of public capital, as well as absorptive and capacity constraints. It allows for government concessional and nonconcessional borrowing and states explicitly the fiscal policy reactions that may ensure debt sustainability. The framework was piloted in Togo and presented to the IMF’s Executive Board in the context of Togo’s 2011 Article IV consultation (Box 6). In Togo’s case, the model showed that a gradual investment path was preferable, to reduce inefficiency losses due to capacity constraints and to allow time for reforms in public financial management.

50. IMF staff intends to pilot the framework in five more countries over the next year. The model would be particularly useful in countries considering different financing options to finance large investment projects. To facilitate broader use in DSAs, IMF staff intends to develop a user-friendly apparatus in the coming months. As staff gains more experience with this tool, an explicit assessment of the trade-offs between the usability and the complexity inherent to the calibration and use of dynamic general equilibrium models will be needed before mainstreaming this approach can be envisaged.

51. To inform lending operations and policies for economic and social development, World Bank staff has developed a broad set of analytical tools and instruments that touch upon the growth-investment nexus. The Bank’s analytical work takes a comprehensive approach to development, with growth and investment being key elements along with poverty alleviation and social inclusion. To inform medium- and long-term government development policies for reducing poverty and achieving the Millennium Development Goals (MDGs), the Bank has developed the Maquette for MDG Simulations (MAMS), which quantifies investment levels needed to meet the MDGs and estimates their impact on growth (Box 6). Another example is the Spatial Approach model, created to help countries assess their proposed infrastructure investment plans by identifying priorities and formulating an adequate sequencing of projects.

52. These various models are complementary. A single model cannot accommodate the heterogeneity of country-specific circumstances with regard to growth and investment. Different analytical tools are needed to inform the macroeconomic projections underlying the DSA, and country teams should choose the tools that best suit each particular case. The IMF’s model is focused on producing macroeconomic frameworks for DSAs that explicitly incorporate public investment and growth linkages. It incorporates fiscal reaction functions and produces trajectories for growth, the real exchange rate, and debt stocks that can be compared to DSF thresholds. MAMS focuses on the linkages across different sectoral investments and looks at the implications for growth and development outcomes. The Spatial Approach model estimates the rate of return of groups of investment projects from a micro-sectoral perspective. The available tools thus offer complementary views on the investment-growth nexus and could be used jointly. Where appropriate, future DSAs will include specific references to analytical work carried out by IMF and the World Bank staff to anchor expected growth rates, taking into account planned public investments.

Modeling the Links Between Public Investment and Economic Growth

The model developed by IMF staff was applied to Togo in the context of the Article IV consultation (Appendix III) to evaluate the authorities’ investment plan, worth 192 percent of GDP over 10 years. The model was used to simulate scenarios and to assess the current DSA projections. The model showed that with good structural conditions—rates of return of 25 percent, efficiency of public investment similar to that of the average Sub-Saharan African country, and high capacity to collect user fees and to tax revenues—this surge could be self-financing and could increase output by more than 8 percent in the long run. But the model also highlighted formidable transition problems, with total debt exceeding 55 percent of GDP and required tax pressure increasing by 2 to 4 percent of GDP in the medium term. Nonconcessional borrowing could smooth the adjustment but the strategy would be risky: high cost of borrowing raises the stakes as small changes in rates of return or in the efficiency of public investment can lead to unsustainable debt dynamics. Overall, the model suggested that a gradual investment path is preferable, to reduce inefficiency losses due to capacity constraints and to allow time for reforms in public financial management.

The Togo pilot highlighted that judgment is critical when applying the model. The model allows policymakers to quantify the investment-growth nexus, but requires a careful choice of parameters and scenarios. In this sense, the model helps apply empirical information where available and makes explicit the assumptions underlying the projections, furthering discussions internally and with stakeholders.

Country studies based on the Maquette for MDG Simulations (MAMS) have been carried out by the World Bank and its development partners in 16 LICs since 2008, mainly in Sub-Saharan Africa. MAMS is a dynamic computable general equilibrium model that includes fiscal spending across sectors (including infrastructure) and financing options (taxation, domestic and foreign borrowing, and foreign aid). Economic performance is measured by the evolution of macroeconomic indicators such as GDP, the budget, and the balance of payments, and other indicators such as poverty or MDG targets.

The Spatial Approach model, presented in the 2009 World Development Report, has been applied to the Republic of Congo and the Democratic Republic of Congo (DRC). By geo-referencing data on productive sectors and infrastructure networks, the analysis portrays the economic geography of a country and allows for evaluating the returns associated with existing and proposed investment packages, as well as the synergy effects of creating spatially coordinated bundles of infrastructure.

In the DRC case, the Spatial Approach indicated that infrastructure investment would lead to higher growth rates. The model estimated that priority public infrastructure projects would raise real GDP growth by 0.7 percentage points on average during the construction phase 2009–13, mainly through the impact on domestic demand, and by 0.2 percentage points following completion of the projects, reflecting gains in total factor productivity associated with improved public infrastructure. The higher growth projections have been incorporated in the baseline scenario of the DRC’s annual LIC DSAs since 2009. The model highlighted a large variation in economic rates of return across different investments, including cross-border investments, underscoring the importance of project screening.

53. The available tools could also be used to enhance the analysis of the growth dividend of regional projects. The macroeconomic frameworks underpinning the DSAs are country-specific in nature, and as such may not appropriately reflect the growth impact generated by large regional infrastructure projects. The Spatial Approach can estimate the economic return from synergies between cross-country investment projects. The other models could be expanded to take these positive externalities explicitly into account, thereby allowing for a more comprehensive analysis of the expected costs (in terms of debt accumulation) and benefits (in terms of higher growth path) of such projects and of alternative financing options.

VII. Redesigning Stress Tests

54. The DSF features a series of stress tests used to assess the impact of shocks that could result in a significant deterioration in the debt outlook. The baseline scenario is based on explicit macroeconomic assumptions deemed to be the most likely outcome, taking into account the authorities’ intended policies. Stress tests, consisting of alternative scenarios and bound tests, illustrate the sensitivity of baseline debt projections to changes to key assumptions. The alternative scenarios entail permanent changes to key assumptions—for example, setting variables at historical levels, or assuming less favorable financing terms. Bound tests show the impact of temporary adverse deviations in key assumptions, with the size of the shock calibrated to match each country’s historical experience. Standardized stress tests simplify the analysis, facilitate cross-country comparisons, and ensure a degree of consistency in the assessment of the risk of debt distress across countries. In addition to these standardized stress tests, country teams are encouraged to design customized scenarios to highlight key country-specific risks.

55. Stress tests were originally calibrated to illustrate the degree of uncertainty surrounding debt projections. The bound tests were calibrated to yield roughly a 25 percent probability of shock occurrence at a 10-year horizon based on stochastic simulations for a representative PRGT-eligible country.29 The 10-year horizon was intended to strike a balance between the uncertainty of long-term projections and the desire to capture debt service on loans with long maturities and grace periods.

56. The DSF’s broadly satisfactory track record suggests that stress tests have met their main objective. A comparison of actual levels of debt in 2010 to projections made in LIC DSAs conducted in 2006 and 2007 reveals that in only 7 out of 60 cases did the actual level of debt in 2010 exceed the level projected by the most extreme stress test.30 Comparing the baseline scenario to an alternative based on a country’s historical record provides a useful “reality check,” drawing attention to cases where the underlying macroeconomic assumptions may be overly optimistic. Bound tests are designed to help identify key vulnerabilities and gauge the impact of remedial policy options.

57. Stress tests have been criticized for being too standardized and lacking interaction between key variables. For example, the bound test that simulates a one-time 30 percent permanent depreciation of the domestic currency has no impact on exports or the current account balance. Furthermore, the persistence of shocks is constrained to be the same across countries even though the dynamic adjustment process is generally believed to depend on various country-specific attributes (the exchange rate regime being a prime example).

58. Possible methodological refinements must consider limitations imposed by data availability and the need to maintain some degree of cross-country comparability. One way to address criticisms of the current framework would be to estimate country-specific dynamic interaction between key variables and the covariance between shocks using vector autoregressive (VAR) models (Annex 3). Such models have been used to assess public debt sustainability in a number of advanced and emerging market countries.31 However, several issues limit the widespread application of such methods across all LICs. First among these is the lack of adequate data.32 Moreover, estimates can be sensitive to model specification and the sample period used, and may be misleading in cases where there have been structural shifts (for example, in the conduct of fiscal and monetary policy and the exchange rate regime), which tend to be frequent in LICs.

59. One way to address the issue of data limitations would be to estimate dynamic linkages between variables using panel data. Pooling observations across countries would increase the degrees of freedom required to attain reasonably reliable estimates, while maintaining some degree of cross-country comparability. Such an approach has the added benefit of reducing the resource intensity of the empirical exercise. A more granular approach could entail estimating VARs for different groups of countries based on economic characteristics, such as the exchange-rate regime or the dependence on a specific commodity export. This would not preclude country teams from using alternative model specifications

tailored to capture country-specific attributes.33

60. Staff proposes introducing methodological refinements of stress tests on an experimental basis, which would enable users to become familiar with new techniques. The innovations would provide country teams with a broader menu of options available to enrich their analysis, but would not have a formal role in determining the risk rating. The innovations would not be incorporated into the standard template, in keeping with the objective of simplification, as discussed below.

VIII. Simplifying the DSA Template

61. The review of the DSF provides an opportunity to look for ways to simplify the LIC DSA template. Feedback from staff indicates that producing a DSA is a resource-intensive exercise, largely due to the complexity of the template and the extensive data input requirements. According to a staff survey, the time it takes to produce a DSA can range from a few weeks to well over a month, depending on the country case and the experience of the staff member with the template. Another indication of the template’s complexity is the fact that few country authorities are able to use it to produce their own DSAs, despite many workshops provided by IMF and World Bank staffs over the years.

62. The template has already undergone numerous changes since the DSF was introduced. Major changes include the unification of the external DSA and public sector DSA templates and the introduction of additional tools to better incorporate remittances into the analysis. Overall, these changes have increased the template’s complexity. Introducing new features could enhance the analysis of debt sustainability but risks increasing the template’s complexity even further.

63. Adopting a modular approach could address the tradeoff between sophistication and ease of use. Some technical aspects of the template could be simplified without sacrificing any functionality. In particular, staff is considering: (i) introducing a core module consisting of a simplified macroeconomic framework input sheet, a baseline scenario, and basic stress tests;34 and (ii) grouping customized scenarios and additional stress tests in optional modules.35

64. A simpler template with the aforementioned characteristics would allow country authorities to more easily produce their own DSAs and enhance discussions between country authorities and staff. The core DSF module as described above would be more intuitive and would ensure consistency of the macroeconomic framework with the underlying economic and financing assumptions. Its output would be limited to a set of standard tables and charts. IMF and World Bank staffs could continue to use most or all of the modules, depending on the complexity of the case, thus avoiding a loss of sophistication and accuracy.

65. Staff proposes that full joint DSAs be produced every three years, with lighter joint updates in the interim years. This would reduce the resource intensity of the DSA exercise through more streamlined written analysis, while still providing an in-depth look at debt sustainability when warranted.36 A full DSA would still be expected whenever there were significant changes to the global environment or to the authorities’ policy stance, particularly if these changes entail a change in the risk rating.

66. A simpler template and greater focus on identified country-specific risks would help reduce the resource implications of the enhancements proposed in previous sections. Under a risk-based approach, the depth of the DSA would depend on the extent of identified risks. In countries where public debt and fiscal vulnerabilities are of concern or external private debt is significant, additional resources may be needed to conduct in-depth analysis. In countries where such risks are deemed low, a risk-based approach ensures that resources are not unnecessarily diverted from the analysis of more pressing matters. In countries where scaling up public investment is envisaged, it may be appropriate for country teams to use one of the models developed by IMF and Bank staffs to evaluate the impact of investment on growth. In countries where public investment is relatively small, a model-based approach would not be recommended.

IX. Issues for Discussion

  • Do Directors agree to maintain the indicative policy-dependent thresholds defined in terms of the various debt burden indicators and to introduce revisions to the thresholds for debt service to revenue and for the PV of debt to the sum of exports and remittances?

  • Do Directors agree with the need for strengthening the analysis of total public debt and fiscal vulnerabilities in DSAs? Do they agree to introduce an additional risk rating, which would complement the assessment of external public debt, in cases where there are significant vulnerabilities related to domestic public debt or private external debt?

  • Do Directors agree that country-specific information should be more systematically taken into account when assessing the risk of debt distress?

  • Do Directors support ongoing efforts by IMF and World Bank staffs to develop models and tools that will strengthen the treatment of investment-growth linkages in DSAs?

  • Do Directors support the inclusion, on an optional basis, of a new stress test reflecting dynamic linkages between macroeconomic variables?

  • Do Directors see merit in developing a simplified DSA template, built around a baseline scenario and simple stress tests, to facilitate its use by country authorities?

Annex 1. External Public Debt Thresholds

Overview of previous methodologies used to derive external debt thresholds

The DSF thresholds for public and publicly guaranteed (PPG) external debt are based on studies by Kraay and Nehru (KN) and IMF and IDA Staff (Staff 2004).1 The calibration of these thresholds comprises three main steps:

  • (i) Identification of debt distress and non-distress episodes on the basis of ‘signals’ of external debt servicing difficulties such as arrears, Paris Club reschedulings, and IMF GRA financing.

  • (ii) Estimation of a parsimonious econometric model (probit) to explain the incidence (probability) of debt distress. The probit model takes the following form:
    P(debt distress)=Φ(β1*debt burden+β2*governance+β3*shock+β4*other)(1)

    where “debt distress” is a binary variable taking the value of 1 if the country experiences debt distress and zero otherwise; Φ is the cumulative distribution function (CDF) of the standard normal distribution; “debt burden” is a measure of indebtedness (PV of debt or debt service) scaled by a measure of repayment capacity (GDP, exports, or government revenue); “governance” is a measure the quality of policies and institutions (the World Bank’s CPIA index); “shock” is a proxy for macroeconomic shocks to the economy (real GDP growth); and “other explanatory variables” in Staff 2004 included GDP per capita and a dummy variable for Africa.2

  • (iii) Calibration of indicative debt burden thresholds. This is achieved by fixing in equation (1) the values for the probability of debt distress, governance, and macroeconomic shock, and solving for the debt burden. In the DSF, the probability of debt distress was set between 18–22 percent, depending on the debt burden indicator.3
    Threshold=Φ−1(P(debt distress))−β^2*governance−β^3*shock−β4*otherβ^1(2)

While the studies share the same basic methodology, they differ significantly in several technical respects. Staff 2004 focuses on LICs and considers only arrears on PPG external debt to official creditors as a signal of debt distress. It defines a non-distress episode as at least three consecutive years in which there is no signal of debt distress (i.e., arrears). All other individual years are defined as debt distress episodes. In contrast, the KN study uses data for both LICs and MICs and relies on three distress signals: arrears on PPG external debt to all creditors (official and private), Paris Club reschedulings, and IMF GRA financing on a commitment basis. KN characterizes a debt distress episode as a period of three years or longer in which at least one of the three distress signals is observed, while a non-distress episode is a non-overlapping five-year period showing no distress signal. Table A1 summarizes the main differences between the two studies.

Table A1.

Description of Methodologies for Estimating DSF Thresholds

article image

Despite these technical differences, the studies find similar results. Empirical results indicate that countries with higher debt burdens are more likely to experience a debt distress episode, and countries with strong policies can sustain a higher debt burden than those with weak policies. These findings are robust to alternative model specifications, including a broader set of explanatory variables. Furthermore, the KN study finds that the basic model has a good out-of-sample predictive power.

The DSF used these empirical analyses to introduce indicative debt burden thresholds. In particular, thresholds were calibrated for five debt burden indicators: PV of debt to GDP, PV of debt to exports, PV of debt to revenue, debt service to exports, and debt service to revenue. The calibration of thresholds for the PV of debt indicators mainly followed the econometric results reported in Staff 2004, whereas those for the debt service indicators used estimates from KN. In line with the robust empirical result that policies matter, specific debt burden thresholds were calibrated for CPIA scores associated with strong (3.75), medium (3.5), and weak (3.25) performers. Thus, a country with strong policies can have debt burden thresholds higher than those of a country with weak policies, acknowledging that the former can carry more debt than the latter. Probabilities ranging from 18 percent to 22 percent, depending on the debt burden indicator, were used to calibrate the thresholds. The probabilities were essentially based on the median PV of debt-to-GDP ratio in LICs immediately preceding an outbreak of debt distress, as observed in the Staff 2004 sample, and were fairly close to the unconditional probability (i.e., the frequency) of debt distress episodes found in the KN study.4 The resulting thresholds were broadly consistent with the results of other studies of the impact of debt on macroeconomic performance (Cohen (1997) and Pattillo, Poirson and Ricci (2002)).

There are three main areas where the technical analysis underlying the DSF thresholds could be improved:

  • Debt distress and non-distress episodes are not uniformly defined across the debt thresholds since thresholds on PV of debt are calibrated using Staff 2004 while thresholds on debt service are calibrated using KN. Lack of uniformity in the definition of episodes implies that the probabilities (18–22 percent) underlying the DSF thresholds refer to the occurrence of events that are rather different, and therefore direct comparisons can be misleading. For example, a country breaching the thresholds on PV of debt (calibrated using Staff 2004) would have a probability of 18–22 percent (or larger) of experiencing distress in the next year, while a country breaching the thresholds on debt service (calibrated using KN) would have a probability of 18–22 percent (or larger) of experiencing protracted distress over the next three or more years.

  • Debt service thresholds (calibrated using KN) apply to LICs but were calibrated on the basis of regressions estimated using a sample of LICs and MICs without controlling for differences between the two groups. LICs tend to borrow on more concessional terms than MICs, resulting in higher debt service obligations for the latter. As a result, the lack of control for both country groups could lead to estimation biases.

  • Debt service thresholds (calibrated using KN) apply to the debt service on PPG external debt but were calibrated on the basis of regressions estimated using the debt service on total external debt (private and public). The current thresholds on PPG external debt service could be inflated by the debt service on private external debt.

Staff has re-estimated the econometric models underlying the LIC-DSF using updated data and a single methodological framework to identify debt distress and non-distress episodes. The new Staff 2011 estimations: (i) rely on a sample of developing countries (LICs and MICs) over 1970–2007; (ii) use a consistent methodology to identify debt distress and non-distress episodes; (iii) control for the presence of MICs in the database; and (iv) use debt service on PPG external debt rather than total external debt. The new estimations, therefore, address the three areas discussed above: the probabilities of experiencing debt distress refer to episodes defined homogenously across regressions and thresholds, the regressions permit calibrating thresholds applicable to LICs only, and the explanatory variables in the regressions coincide with those used in the DSF.

The Staff 2011 estimations: description of episodes and basic statistics

Staff focuses on three indicators related to exceptional external financing to signal whether a country is experiencing debt distress: (i) the accumulation of arrears on PPG external debt in excess of five percent of the PPG external debt stock outstanding; (ii) a rescheduling of obligations due to Paris Club creditors; or (iii) the disbursement by the IMF of GRA resources exceeding 50 percent of IMF quota.5

A debt distress episode is defined as a period lasting three or more years in which at least one distress signal is observed. As in KN, by imposing a three-year minimum duration requirement on distress episodes, the analysis captures severe and persistent debt service difficulties, and rules out temporary events.6 A non-distress episode is defined as a non-overlapping three-year period during which none of the distress signals is observed.7 Accordingly, the duration of distress episodes is in line with KN while the duration of non-distress episodes is in line with Staff 2004. Debt burdens are taken from the World Bank’s Debt Reporting System, in which the PV of debt is calculated by discounting the stream of debt service denominated in original currencies using time-varying currency-specific commercial interest reference rates (CIRRs), as in KN.8,9 Staff uses the debt service data from the World Development Indicator (WDI) database, which is on a paid basis, and adjust data by the accumulation of arrears to estimate service on a due basis. Series for real GDP growth, nominal GDP, exports and government revenues were constructed by merging information from different databases: WDI, Global Finance Statistics (GFS), United Nations (UN), and World Economic Outlook (WEO). The database covers LICs and MICs.10 The World Bank’s CPIA is used as a measure of the quality of policy and institutions.11

Staff uses a sample of 130 countries, of which 61 are LICs, and identifies 105 debt distress episodes and 654 non-distress episodes. Compared to KN, the current exercise shortens the duration of non-distress episodes (from five to three years), thus increasing their number in the sample. Compared to Staff 2004, the current exercise lengthens the duration of debt distress episodes (from one to three years), thus reducing their number in the sample. As a result, the frequency of debt distress episodes—and hence the unconditional probability of debt distress—is lower than in previous studies.

The results confirm key findings from earlier studies: countries in debt distress tend to have higher debt levels than countries not experiencing distress, and higher capacity countries are able to carry higher levels of debt. Table A2 reports summary statistics disaggregated by types of episode (distress and non-distress) and country groups (LIC and MIC). Summary statistics are calculated using values corresponding to the year prior to the outbreak of a debt distress episode, and to the first year of a non-distress episode.12 The median value of PV of debt-to-GDP ratio is 30 percent in debt distress episodes and 17 percent in non-distress episodes. A similar pattern is observed for the other debt indicators. This finding is also evident when looking at the frequency distribution of debt burden indicators (Figures A2 and A3, for the PV of debt-to-exports and debt service-to-exports ratios, respectively), which show that debt distress episodes are typically characterized by higher debt burden than non-distress episodes. Figure A2 and A3 also show that non-distress episodes for weak performers typically occur at lower debt levels compared to medium and strong performers (i.e., a larger proportion of non-distress episodes can be found at lower debt level for weak performers).

Figure A2.
Figure A2.

Relative Frequency Distribution of the PV of Debt-to-Exports Ratio

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Figure A3.
Figure A3.

Relative Frequency Distribution of the Debt Service-to-Exports Ratio

Citation: Policy Papers 2012, 098; 10.5089/9781498341028.007.A001

Table A2.

Summary Statistics for Debt Distress and Non-distress Episodes

article image
1/

Sample median for the debt burden indicator, governance, and real GDP growth.

LICs and MICs also exhibit different debt levels between distress and non-distress episodes (Table A2). These differences suggest the econometric estimations should control for country groups if used to calibrate debt thresholds applicable to LICs only.

The new Staff 2011 estimations and calibration of debt thresholds

Staff first investigates the determinants of debt distress by formulating a probit model as in Staff 2004 and KN.13 The basic model specification includes a debt burden, the CPIA score, and the real GDP growth as covariates.14 Interactive dummy variables are included to allow the coefficient associated with the debt burden to vary between LICs and MICs.15

Estimation results are reported in Table A3. All estimated coefficients have the expected sign and most of them are statistically significant. The (estimated) probability of debt distress increases with the level of debt, whereas it decreases with the quality of policies and institutions and real GDP growth. In the regressions including PV of debt-to-exports and debt service-to-revenue ratios, the interactive dummy variables are significant and thus indicate differences in the coefficients associated with the debt burden in LICs and MICs. This allows calibrating debt thresholds for LICs that are not biased by the presence of MICs in the estimation sample.

Table A3.

Estimation Results and Calibration of Thresholds

(Basic Specifications)

article image
1/

*** p < 0.01, ** p < 0.05, * p < 0.1

Indicative debt burden thresholds are calibrated by fixing values for the following variables: (i) a probability of experiencing a debt distress episode; (ii) a CPIA score—specifically, the value characterizing a weak, medium, or strong performer; and (iii) a real GDP growth rate.

As in the current DSF, thresholds are calibrated for three values of the CPIA (3.25, 3.5, and 3.75) associated with weak, medium, and strong performance in terms of quality of policies and institutions. For simplicity, real GDP growth is set to its sample average for LICs, which can be interpreted as the growth rate in the absence of macroeconomic shocks. Calibrated thresholds are reported for three different debt distress probabilities:

(i) The unconditional probability of debt distress. This is the frequency of debt distress experienced by LICs in the estimation sample (for each regression). It depends therefore on the definition of distress and non-distress episodes. This is the approach that was used to calibrate debt service thresholds in the DSF.

(ii) The probability associated with the median debt burden indicator. This probability is calculated by evaluating equation (1) at the median debt burden indicator for LICs in the year prior to a debt distress, and using the median CPIA and real GDP growth for LICs. Intuitively, this approach therefore produces debt burden thresholds that are similar to the median debt burden before a crisis (adjusted for changes in the CPIA). This approach is also conceptually consistent with the one used to calibrate thresholds in Staff 2004.

(iii) The probability that minimizes the occurrence of type I and type II errors. This is the preferred approach of staff. A type I error denotes the failure to predict a debt distress episode (i.e., a missed crisis) whereas a type II error refers to the failure to predict a non-distress episode (i.e., a false alarm).16 Hence, this probability is the one that performs optimally (on average) when evaluating the in-sample forecasting properties of the estimated probit model. It simultaneously minimizes the number of missed crises and false alarms produced by the model, thus ensuring that the thresholds are neither too permissive nor unduly conservative.

Overall, debt burden thresholds calibrated with the adjusted probabilities are broadly in line with the DSF, except for the debt service-to-revenue threshold.

Robustness of the basic specifications

Staff assessed the robustness of estimation results and calibrated thresholds by using alternative sample periods, proxy variables for governance and macroeconomic shocks, and definitions of non-distress periods (Table A4). Results suggest the basic specifications are robust in terms of point estimates and statistical significance of the explanatory variables. In addition, numerical values of the debt thresholds calibrated using the alternative regressions are similar to those obtained from the basic specifications.

Table A4.

Estimation Results and Calibration of Thresholds

(Robustness of the Basic Specifications)

article image

Staff re-estimated the probit model excluding the 1970s, on the basis that in the early 1970s the integration of LICs and MICs into international financial markets was relatively limited, and that the debt crisis in the 1980s and the waves of financial liberalization during the 1990s might have implied structural breaks with respect to the past.17 Most of the estimated coefficients and calibrated thresholds are similar to the basic specifications. Staff also re-estimated the probit model substituting alternative measures of governance and macroeconomic shocks, respectively, for the CPIA score and real GDP growth. Table A4 reports regressions including four selected variables: (i) the World Bank’s sub-CPIA index specific to macroeconomic and debt management, which assesses government policies affecting debt and repayment capacity directly; (ii) the International Country Risk Guide (ICRG) index; (iii) the deviation of real GDP growth from the three-year historical average, which controls for differences in growth rates among countries and is consistent with the notion that a shock would deteriorate the current growth outcome vis-à-vis the recent performance of the economy; and (iv) the growth rate of export receipts, which captures external shocks. Estimated coefficients associated with debt burden indicators, governance, and macroeconomic shocks are always significant. The calibrated thresholds, in addition, are close to those resulting from the basic specifications, with a few exceptions in the regressions, including the PV of debt-to-exports ratio.18

Finally, two alternative definitions of non-distress episodes are considered, in which the length of these episodes is extended from three (in the basic specifications) to four and five years, respectively. Since periods of non-distress last longer, the characterization of the dependent binary variable in the probit model changes accordingly: the number of non-distress episodes decreases, the frequency of distress episodes increases, and the size of the estimation sample drops significantly. Table A4 reports regressions including the two alternative definitions. Increasing the length of non-distress episodes leads to a higher probability used to calibrate thresholds (as the frequency of debt distress in the estimation sample increases). For example, for the PV of debt-to-GDP ratio, the probability minimizing type I and II errors in the basic specification is 13.5 percent, increasing to 19.5 percent and 24.2 percent when the length of non-distress episodes is increased to four and five years, respectively. Estimated coefficients for governance and macroeconomic shocks are similar to those in the basic specifications, whereas estimated coefficients for debt burden indicators are higher.

Remittances

Remittances constitute a source of income and foreign exchange for a country, along with its GDP and exports. It has been argued that remittances should therefore be added to measures of repayment capacity when assessing the risk of debt distress. To address formally the impact of remittances on debt distress probability, staff estimated a probit model including GDP plus gross remittances, and exports plus gross remittances, as denominators of the debt burden indicators.

Because of limited availability of data on remittances, comparability between the regressions excluding and including remittances requires controlling for possible changes in the sample size. For example, the sample for the regression on the PV of debt-to-GDP ratio has 315 observations when including remittances compared to 740 observations when remittances are excluded.19 For that reason, directly comparing the re-estimated thresholds presented in the previous section (full sample) to the ones which incorporate remittances (small sample) may be misleading. Staff therefore compared thresholds excluding remittances (small sample) to thresholds for debt burden indicators with remittances (small sample).

Adding remittances to GDP or exports results in higher estimated coefficients associated with debt burdens vis-à-vis the regressions excluding remittances. Consequently, the corresponding calibrated thresholds are lower than those found in the basic specifications, which excludes remittances (Table A5). Such a result implies that, for a country adding remittances to GDP or exports in the debt burden indicators, the predicted probability of debt distress would be lower only if the amount of remittances is sufficiently large to offset the lower threshold. In other words, the country can carry a higher debt level only if it receives a large flow of remittances.

Table A5.

Estimation Results and Calibration of Thresholds with Remittances

article image

Probability thresholds

The current methodology to calibrate thresholds is relatively simple. As mentioned above, once debt distress and non-distress episodes are identified, a probit model is estimated:

P(debt distress)=Φ(β1*debt burden+β2*governance+β3*shock+β4*other)(1)

Thresholds are then calibrated by solving equation (1) for debt burden and fixing the values for the probability of debt distress, governance, and macroeconomic shock:

Threshold=Φ−1(p(debt distress))−β^2*governance−β^3*shock−β^4* otherβ^1