## pp082217lic-dsf

## Source details

**Canonical URL:** [pp082217lic-dsf](https://www.imf.org/-/media/files/publications/pp/2017/pp082217lic-dsf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/pp/2017/pp082217lic-dsf.pdf.md)
- [Structured JSON version](/-/media/files/publications/pp/2017/pp082217lic-dsf.pdf.json)

---

### Overview
- The Debt Sustainability Framework for Low-income Countries (LIC DSF) is the cornerstone of assessments of risks to debt sustainability in LICs.
- Objective of proposed reforms: make the LIC DSF more comprehensive, more transparent, and simpler/easier to use while maintaining balance between early warning of debt distress and not unnecessarily constraining borrowing for development.
- Key date: August 22, 2017.

### Rationale for reform and performance diagnostics
- Financing landscape changes and identified gaps:
  - Increased borrowing from non-Paris Club creditors, domestic markets, and international bond markets.
  - Gaps: complexity, lack of transparency, limited tools to assess market-related risks, insufficient scrutiny of baseline macro projections and the investment-growth nexus.
- Weaknesses in characterizing debt evolution:
  - Near-term (one- and two-year) debt projections: median (absolute) unexpected change about 5 percent of GDP for both external and total public debt.
  - Five-year horizon: median deviation more than doubles for both external and total public debt.
  - About 40 percent of DSAs produced during 2007–10 contained unexpected changes in debt over a five-year horizon in excess of 15 percentage points of GDP; for small states this occurred in more than half of the DSAs.
  - Among DSAs with sizable unexpected changes in debt (>15 percentage points of GDP):
    - About 70 percent (external DSAs) underestimate debt outcomes.
    - About 80 percent (public DSAs) underestimate debt outcomes.
  - When moving from the 1- to the 7-year projection horizon, the share of DSAs underestimating debt outcomes rises from about 55 percent to more than 80 percent.
- Stress-test alignment and coverage gaps:
  - Standardized stress tests generally not well aligned with actual drivers of debt changes; primary balance shock under-calibrated.
  - Less than 50 percent of countries severely exposed to natural disasters used customized scenarios.
  - Commodity price shocks and contingent liabilities often not directly assessed where relevant.

### High-level package of proposed reforms
- Core architecture preserved: model-based results complemented by staff judgment.
- Main elements:
  - Move from CPIA-only classification to a composite indicator (CI) including CPIA, country growth, reserve coverage, remittances, and world growth.
  - Expand statistical model specification to include international reserves (scaled by imports), remittances (scaled by nominal GDP), and world growth; estimate model using LIC data only.
  - Introduce realism tools to assess plausibility of baseline macro projections (fiscal adjustment realism, investment-growth nexus).
  - Simplify debt indicators, thresholds, and standardized stress tests:
    - Debt burden indicators: Existing 5 → Reformed 4.
    - Thresholds: Existing 24 → Reformed 12.
    - Standardized stress tests: Existing 16 → Reformed 8.
    - Drop PV of external debt to revenues indicator.
  - Recalibrate and redesign stress tests to incorporate macro-linkages; halve standardized tests and introduce tailored scenario stress tests (natural disasters, commodity price shocks, market-financing shocks, contingent liability supplements).
  - Sharpen analysis of domestic debt vulnerabilities, market-financing pressures, and robustness within the moderate-risk category.
  - Staff Guidance Note to expand attention to even-handed application of judgment, including treatment of marginal/transitory breaches.

### Discount rate decision
- Proposal: keep the current five percent discount rate unchanged.
- Maintain a unified discount rate for the LIC DSF, the DLP, the NCBP, and the grant element calculator.
- Revisit the discount rate decision in future DSF reviews.

### Statistical model, identification of distress episodes, and classification
- Revised distress identification:
  - Include distress periods lasting one or two years (except when driven solely by arrears, which still require three consecutive years).
  - IMF-disbursement signal redefined to focus on large upfront disbursements: IMF disbursements during the first six months > 30 percent of quota (the 75th percentile since 1970).
  - Incorporate restructurings of public external debt held by private foreign creditors and outright defaults; treat multi-round HIPC restructurings as single episodes.
  - Revised identification raises episode count: Old definition 76 → New definition 98 episodes (1970–2015); median duration falls Old 18 → New 12 years.
- Expanded probit specification:
  - Adds reserves (and reserves^2), remittances, and world growth; model estimated on 80 LICs covering 1970–2014.
  - Key empirical regularities preserved: probability of debt distress positively correlated with indebtedness, negatively with CPIA, growth, reserves, remittances, and world growth.
  - Model performance: improved goodness-of-fit metrics (examples preserved): Pseudo R-squared examples 0.169; 0.174; 0.190; 0.184; Log-likelihood examples -150.1; -145.7; -116.8; -135.2; BIC examples 348.4; 339.5; 280.2; 317.9.
- New country classification:
  - Construct CI = weighted sum of CPIA, country growth, reserves, remittances, world growth with weights equal to average estimated coefficients.
  - Use cross-country distribution of the average CI over 2005–14; cutoffs:
    - Weak if CI < 25th percentile.
    - Medium if CI between 25th and 75th percentiles.
    - Strong if CI > 75th percentile.
  - CI calculated using latest five years of historical data and first five years of projections; CPIA forecast uses most recent value.
  - Require two consecutive signals where a country’s CI exceeds its classification cutoff to change classification.

### Realism tools for baseline projections and public investment
- Realism tool 1 — Decomposition of past and projected drivers of debt dynamics:
  - Show evolution of external and public debt to GDP ratios over DSA vintages (one-year and five-years ago) to identify marked changes and unexpected contributions.
- Realism tool 2 — Contrast projected fiscal adjustment with historical experience:
  - Compare projected headline primary fiscal adjustment over a three-year horizon to distribution observed in LICs requesting Fund programs.
  - Operational trigger: fuller discussion called for if projected primary fiscal adjustment over any three years exceeds 2½ percent of GDP (approximate top quartile of observed distribution).
- Realism tool 3 — Benchmarks for assessing growth assumptions (public investment and fiscal adjustment):
  - Simple growth-accounting decomposition: growth = contribution from government capital (β ΔG/G with β = 0.15) + εt.
  - Public investment efficiency parameterization (φF, φH) and government capital accumulation Gt+1 = (1 − δ) Gt + φF iGt with δ = 0.05.
  - Fiscal adjustment impact tool: fiscal multiplier m and persistence pt (default pt = 0.6); first-year full impact assumed one year after implementation, implementation-year impact assumed half.

### Stress testing framework: standardized and tailored tests
- Standardized tests (common to external and public DSA; dropped: GDP deflator stress test and all permanent shocks; historical scenario retained):
  - Retained shocks: (i) real GDP growth, (ii) export growth, (iii) primary balance, (iv) other flows (transfers and FDI), (v) exchange rate depreciation, (vi) contingent liabilities, (vii) combination shock.
  - Combination shock: half magnitudes of stand-alone shocks; incorporates macro interactions.
- Macro-linkages to be introduced (examples):
  - Export growth shock negatively impacts GDP growth.
  - Exchange rate shock raises inflation and net exports, mitigating FX depreciation impact on debt ratios.
  - Real GDP growth shock reduces inflation and the primary balance.
  - Primary balance shock increases commercial borrowing costs.
- Post-shock application rule:
  - Post-shock values set to historical average minus one standard deviation, or baseline projection minus one standard deviation—whichever is lower.
  - Contingent liability standardized shock recalibrated to 5 percent of GDP (from 10 percent).
- Tailored scenario stress tests (triggered where relevant; default settings based on median shock for LICs; country teams encouraged to customize):
  - Natural disasters:
    - Default: one-off shock to public debt of 10 percent of GDP in year 1; real GDP growth and exports lowered by 1.5 and 3.5 percentage points in the year of the shock; triggers: Group 1 (14 small-state LICs) and Group 2 (9 LICs meeting frequency and loss criteria).
  - Contingent liability supplement:
    - Trigger when significant public sector exposures not covered by public debt concept or financial sector risks exceed standardized shock.
    - PPP default scenario: permanent increase in external PPG debt equivalent to 35 percent of PPP capital stock in year 2 (PPP capital stock trigger > 3 percent of GDP; qualifies 26 countries).
    - SOE default scenario: one-off permanent increase of 2 percent of GDP in public debt from year 2 (qualifies 37 countries).
  - Commodity price shock:
    - Trigger: commodity exports ≥ 80 percent of merchandise exports (36 countries).
    - Default: commodity price gap shock to export-to-GDP in year 1, closing over 6 years; real GDP growth reduced 0.5 percentage points and fiscal revenues reduced 0.75 percentage points for each 10-percentage point contraction in commodity prices for first three years.
  - Market-financing stress test:
    - Trigger: frontier LIC definition / outstanding Eurobonds / meet PRGT market access criterion but not graduated (identifies 17 countries).
    - Default: 400 bps increase in cost of new external commercial borrowing (sustained 3 years), FX depreciation 15 percent, and shortening of maturities of new external commercial borrowing to 5-year maturity.
- Customized alternative scenarios permitted; DSF template enhanced to design customized scenarios.

### Assessment of total public debt and market-financing pressures
- Total public debt:
  - New methodology requires assessment of domestic debt risks; uses alternative identification of de facto domestic default (proxies) and a noise-to-signal (NTS) modelling approach for PV of total public debt benchmarks.
  - Benchmarks (PV of total public debt, rounded to nearest 5 percent):
    - 2017 review: weak 35; medium 55; strong 70; Type I error 0.24; Type II error 0.44; Loss function 0.31 (weight on Type I = 0.67).
- Market-financing early warning tool:
  - Benchmarks derived from NTS on frontier LICs and MICs (1995–2015). Breach rule:
    - GFNs above 14 percent of GDP AND EMBI spreads higher than 570 bps signal heightened liquidity risks; country teams expected to provide in-depth assessment of liquidity needs and creditor composition.

### Determination of mechanical risk signals, thresholds, and aggregation
- Mechanical signal rule:
  - Low risk: all debt burden indicators below thresholds under baseline and stress tests.
  - Moderate risk: any breach under stress tests but not under baseline.
  - High risk: at least one breach under baseline.
- Two methodological improvements:
  - Transparent policy choice for tolerance of missed crises vs false alarms: propose weight on Type I error ω = 0.67 (2:1 tolerance in favor of reducing missed crises).
  - Derive debt thresholds jointly in line with DSF aggregation rule (invert probit for each debt indicator using CI percentiles and chosen cutoff probabilities).
- Re-estimated external debt thresholds (Old → New by classification; exact values preserved):
  - PV of debt-to-GDP: Weak Old 30 → New 30; Medium Old 40 → New 40; Strong Old 50 → New 55.
  - PV of debt-to-exports: Weak Old 100 → New 140; Medium Old 150 → New 180; Strong Old 200 → New 240.
  - Debt service-to-revenue: Weak Old 18 → New 14; Medium Old 20 → New 18; Strong Old 22 → New 23.
  - Debt service-to-exports: Weak Old 15 → New 10; Medium Old 20 → New 15; Strong Old 25 → New 21.
- Simplification counts:
  - Debt burden indicators: Existing 5 → Reformed 4.
  - Thresholds: Existing 24 → Reformed 12.
  - Standardized stress tests: Existing 16 → Reformed 8.

### Predictive performance, back-testing, and judgment
- Predictive performance findings:
  - Existing mechanical framework: false alarms ~50 percent; missed crises ~20 percent.
  - Re-estimating thresholds using updated sample and existing methodology would increase false alarms ~60 percent and reduce missed crises ~8 percent (discussion varies by specification).
  - Proposed reforms: reduce false alarms by about 10 percentage points and improve ability to anticipate debt distress.
  - Table 2 summary (preserved examples):
    - Existing 2/: Type I and II errors estimated at 18 and 48 percent (note 2/).
    - Existing (updated) 3/: Type I error 0.22; Type II error 0.50; Loss function 0.31.
    - New 4/: Type I error 0.06; Type II error 0.68; Loss function 0.27; alternate Loss function 0.24.
    - Note 1/: Weighted sum of errors (weight on Type I error equal to 67 percent).
- Back-testing (mechanical simulations, sample of 67 LIC DSAs produced 2015–16):
  - Country classification (out of 67):
    - Unchanged: 40 (60 percent).
    - Upgraded: 22 (33 percent).
    - Downgraded: 5 (7 percent).
    - CPIA still accounts for 45 percent of the CI; most upgrades driven by large reserve positions and/or sizable remittances flows.
  - Mechanical external risk signals (out of 67):
    - Unchanged: 47 (70 percent).
    - Upgraded: 12 (18 percent).
    - Downgraded: 8 (12 percent).
- Staff judgment:
  - Since DSF inception, judgment applied in ~25 percent of cases, leading to upgrades ~80 percent of the time.
  - Judgment reduced false alarms vis-à-vis mechanical signals by about 10 percentage points.
  - Judgment typically used to override signals driven by marginal/transitory breaches; mechanical risk signal overridden in 48 percent of all marginal breaches (measured as a 5 percent deviation from the threshold).

### Implementation, capacity building, effective date, and operational implications
- Implementation pacing and materials:
  - Updated Staff Guidance Note (thorough revision, simplified) expected by end-December 2017.
  - DSF template to be re-programmed, simplified, streamlined, with automated composite indicator calculations and tailored scenario assumptions.
  - Training: external training missions increased from nine in FY 2017 to sixteen in FY 2018; four workshops in regional training centers in the Fall and nine training missions in first half of 2018; historical participation: during 2016–17, 175 representatives from 40 countries in one-week training; nearly 850 government officials used IMF MOOC.
- Effective date:
  - Proposed effective date for new framework: July 1, 2018.
  - Aim: allow teams to complete DSAs for all countries under new DSF prior to July 1, 2019 (cut-off for changes to country “traffic lights” under IDA’s fiscal 2020 allocations and IDA19 projections).
- Interactions and disclosure:
  - Greater engagement with country authorities on macro projections, tailored stress-test parameters, and technical/country-specific judgment; DSA write-ups to fully disclose and justify technical considerations and authorities’ views.
- Post-rollout support:
  - Continued training, technical assistance, and exploration of links with debt management tools (e.g., MTDS).

### Policy implications: DLP and NCBP
- The proposed methodology increases information value without creating additional data needs or immediate operational changes to Fund/Bank debt policies.
- DLP flexibility:
  - Upgraded risk rating could loosen or remove debt conditionality.
  - Downgraded ratings may still access DLP flexibility (loan-by-loan exceptions, nominal ceilings, PV ceilings where capacity adequate).
  - Staff would seek new understandings where debt conditionality changes are called for; Executive Board consideration for such revisions.
- NCBP treatment:
  - Options preserved for IDA-only non-gap countries and for grant recipients/MDRI recipients.
  - Choice between loan-by-loan exceptions and options for nominal ceilings or PV ceilings remain consistent with debt management capacity assessments.
  - IDA’s grant/loan allocation mix unaffected by rating changes until fiscal year 2020 (starting July 1, 2019).

### Expected benefits and trade-offs
- Outcomes expected:
  - Address technical criticisms; reduce false alarms while improving predictive power to signal debt distress events.
  - Incorporate additional country-specific information into classification and thresholds.
  - Provide richer dialogue on policy actions, debt-management trade-offs, and the investment–growth nexus.
  - Better realism checks, redesigned stress tests with macro-linkages, enhanced guidance on judgment, and attention to domestic debt and market-financing vulnerabilities.

*International Monetary Fund — EXECUTIVE SUMMARY, REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES: PROPOSED REFORMS, August 22, 2017.*

### EXECUTIVE SUMMARY

### EXECUTIVE SUMMARY

### Overview
- The Debt Sustainability Framework for Low-income Countries (LIC DSF) is the cornerstone of assessments of risks to debt sustainability in LICs.  
- Core architecture: model-based results complemented by staff judgment remains appropriate.  
- Objective of proposed reforms: make the LIC DSF more comprehensive, more transparent, and simpler/easier to use while maintaining balance between early warning of debt distress and not unnecessarily constraining borrowing for development.  
- Key date: August 22, 2017.

### Rationale for Reform
- The economic and financing landscape for LICs has evolved (e.g., increased borrowing from non-Paris Club creditors, domestic markets, and international bond markets), exposing LICs to a wider set of vulnerabilities, including market volatility.  
- Stakeholders and staff analysis identified gaps: complexity, lack of transparency, limited tools to assess market-related risks, and insufficient scrutiny of baseline macroeconomic projections and the investment-growth nexus.  
- The DSF has been reviewed previously (introduced in 2005; reviewed on three occasions, most recently in 2012).

### Proposed Reforms — High-level package
- Move away from relying exclusively on the CPIA as the measure for assessing debt-carrying capacity; adopt a composite measure based on an expanded set of economic variables that includes the CPIA.  
- Improve model specification and methodology to better identify debt distress episodes and enhance statistical accuracy in predicting debt distress.  
- Introduce new tools to assess plausibility of baseline macroeconomic projections (e.g., realism of projected fiscal adjustment and projected impact of public investment and fiscal adjustment on growth).  
- Introduce tailored scenario stress tests for risks of specific relevance (natural disasters, volatile export prices, market-financing shocks, contingent liability exposures).  
- Reduce the number of debt indicators, thresholds, and standardized stress tests:
  - One debt indicator—the present value of external debt to revenues—would no longer be used.
  - Number of thresholds would be reduced from 24 to 12.
  - Number of standardized stress tests would be halved.
- Sharpen analysis of domestic debt vulnerabilities; assess risks from shifts in market-financing conditions; analyze robustness of debt positions for countries at moderate risk of debt distress.  
- Staff Guidance Note to expand attention to ensuring even-handed application of staff judgment (including treatment of marginal and/or transitory breaches of thresholds).

### Discount Rate Decision
- Proposal: keep the current five percent discount rate unchanged.  
- Additional decision: maintain a unified discount rate for the LIC DSF, the DLP, the NCBP, and the grant element calculator.  
- Revisit the discount rate decision in future DSF reviews.

### Back-testing and Predictive Performance
- Back-testing suggests a balanced impact on the framework with significant improvements in predictive performance:
  - Mechanical risk signals would remain unchanged for most countries.
  - Slightly more mechanical upgrades than downgrades for remaining cases.
  - The reformed framework should significantly reduce the rate of false alarms (incorrectly predicting occurrence of debt distress).
  - The reformed framework should moderately decrease the rate of missed crises (failing to predict onset of debt distress).

### Operational and Analytical Enhancements
- Country-specific features (e.g., international reserve coverage) would play an expanded role as inputs and as drivers of tailored scenario stress tests.  
- Outputs would include a richer set of information on debt developments and vulnerabilities to better inform policy discussions and trade-offs between Bank and Fund staffs and country authorities.  
- Stress tests would be recalibrated to better reflect actual scale of shocks and would incorporate macro-linkages where appropriate.

### Implementation and Timing
- Implementation would be appropriately paced with updated guidance materials, templates, and training materials.  
- Training activities would be conducted ahead of implementation expected by the second half of 2018.

### Expected Benefits and Trade-offs
- The reforms aim to enhance the accuracy, transparency, and usability of the LIC DSF while simplifying aspects of the toolkit.  
- The proposed changes are designed to provide better early warning and richer diagnostics without unduly constraining borrowing for development.

*International Monetary Fund — EXECUTIVE SUMMARY, REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES: PROPOSED REFORMS, August 22, 2017.*

### 8. To help map out potentially useful reforms to the DSF, it is important to understand

### 8. To help map out potentially useful reforms to the DSF, it is important to understand first where it is performing well and not so well.

### Performance in Characterizing the Likely Evolution of Debt: some weaknesses
- Near-term (one- and two-year) debt projections: median (absolute) unexpected change about 5 percent of GDP for both external and total public debt.
- Five-year horizon: median deviation more than doubles for both external and total public debt.
- About 40 percent of DSAs produced during 2007–10 contained unexpected changes in debt over a five-year horizon in excess of 15 percentage points of GDP; for small states this occurred in more than half of the DSAs.
- Deviations of this scale particularly common in DSAs for countries currently at high risk of external debt distress.
- Directional bias in medium-term deviations:
  - Among DSAs with sizable unexpected changes in debt (larger than 15 percentage points of GDP):
    - About 70 percent (external DSAs) underestimate debt outcomes.
    - About 80 percent (public DSAs) underestimate debt outcomes.
  - When moving from the 1- to the 7-year projection horizon, the share of DSAs underestimating debt outcomes rises from about 55 percent to more than 80 percent.
- Median public debt-to-GDP projections across LICs for each of the last five DSA vintages show projections quickly reverting to a downward debt path even after significant short-term upward shifts.
- Decomposition of unexpected changes (5-year horizon):
  - Total public debt: in addition to the primary deficit, unanticipated positive residuals have contributed to sizable deviations (suggesting contingent liabilities materialization). Growth and exchange rate shocks less important.
  - External debt: evidence points to a much larger role of financial account flows driving unexpected changes in total external debt, suggesting larger-than-expected access to sources of finance (manifestation of changes in the financing landscape).
- Debt service indicators show relatively much smaller unexpected changes; share of DSAs with unexpected changes above 15 percentage points (in absolute value) ranges between 0 and 8 percent.

### Adequacy and Alignment of Stress Tests
- Standardized stress tests have generally not been well aligned with actual drivers of debt changes.
- Modeled vs. actual:
  - Simulated impact on debt from shocks to the primary balance is modest compared to actual impact of primary balance deviations.
  - Calibration issue: post-shock values set at one standard deviation below historical averages, assuming historical averages predict next 2–3 years. This assumption did not hold for the primary balance: actual primary surpluses (deficits) were consistently lower (larger) than historical averages during 2008–15.
  - Simulated impact from exchange rate shocks is large relative to actual impact; with correctly calibrated exchange rate shock magnitude the larger simulated impact suggests key macro interaction factors are not captured in stress testing.
- Role of specific stress tests in determining external risk rating:
  - Simulated shocks to exports, nominal depreciation, other flows (transfers and FDI), and a combination shock have played the main role in signaling risks of external debt distress.
  - The external financing alternative scenario and the real GDP growth and GDP deflator stress tests are rarely identified as sources of risk.

### Coverage Gaps in Stress Tests for Important Risks
- Natural disasters:
  - Less than 50 percent of countries severely exposed to natural disasters used a customized scenario to assess their associated impact on debt vulnerabilities.
- Contingent liabilities:
  - Few country teams directly assessed risks to public debt from realization of contingent liabilities. Examples of analysis: Nicaragua (2015) and Solomon Islands (2016).
- Commodity price shocks:
  - Not directly assessed in relevant cases. Over last two years, 8 of 11 LICs that experienced a downgrade in their risk rating were commodity exporters; the generic export shock scenario did not capture the full impact on export and fiscal revenues from price shocks.

### Evaluation of Debt Indicators and Thresholds: some not playing a role
- Debt service indicators:
  - Rarely determined risk signals of external debt distress: only in 8 percent of all DSAs were high risk signals exclusively informed by breaches of debt service thresholds.
  - Distance of debt service indicators to thresholds in run up to recent debt distress events remained significant; in some cases after sizable fiscal deviations and rising market risks, debt service thresholds were not breached.
- PV of external debt to fiscal revenues:
  - Appears redundant in determining external risk signals: only one case where this threshold was breached under the baseline without breaches in other debt stock indicators.
- PV of external debt to exports:
  - Signaled high risk of debt distress in about 80 percent of all cases.
- Implication: either debt service measures contain little information on signaling debt distress or relevant thresholds have been poorly estimated.

### Performance of New Features Added in the 2012 Review
- Remittances-augmented thresholds:
  - 13 out of 69 countries incorporated remittances as an integral part of the framework; of those, five saw upgrades in their risk ratings.
  - Extending remittances-augmented thresholds to all countries with remittances data available since 2014 would not have improved any risk rating and would have produced 14 downgrades.
  - Outcome driven by how remittances-augmented thresholds were derived in 2012 and highly-skewed distribution of remittances.
  - Countries that benefited had remittances flows as percentage of GDP and exports well above eligibility cutoffs (averages of 30 percent of GDP and 150 percent of exports).
- Probability approach:
  - Since operational in 2014, only nine out of 69 countries used it to inform risk ratings (leading to one final risk rating improvement).
  - Had it been more broadly applied to all DSAs since 2012, it would have upgraded risk ratings in 17 DSAs and downgraded risk ratings in 28 DSAs.

### Performance of Rules to Set Risk Ratings: false alarms and missed crises
- Mechanical framework overall produces a high rate of false alarms:
  - Existing framework yields a large rate of false alarms — about 50 percent of the cases in which debt distress is not observed.
  - Missed crises are more modest — about 20 percent of the cases in which debt distress is observed.
- Re-estimating debt thresholds with updated sample period and existing methodology would tighten thresholds, leading to:
  - False alarms about 60 percent.
  - Missed crises about 8 percent.
- Use of staff judgment:
  - Since inception of DSF, judgment applied to override mechanical risk signal in about 25 percent of cases, leading to an upgrade about 80 percent of the time.
  - Staff judgment reduced rate of false alarms vis-à-vis mechanical application by about 10 percentage points.
  - Judgment typically used to override signals driven by marginal and/or temporary breaches; mechanical risk signal overridden in 48 percent of all marginal breaches (measured as a 5 percent deviation from the threshold).
- Deeper risk disentanglement has been limited:
  - Moderate risk category conceals diverse vulnerabilities: number of years breaching thresholds under stress tests varies widely and breach sizes vary.
  - Feature added in 2012 for deeper assessment of total public debt and an overall risk rating used sporadically:
    - Since 2014, DSAs documented 26 countries with significant vulnerabilities from total public debt.
    - Only 10 countries reported an overall risk rating; only six of those provided an in-depth discussion of domestic public debt vulnerabilities.
    - Most DSAs did not provide discussion of vulnerabilities from rollover risks, increasing participation of non-residents in domestic local-currency bond markets, or structure of domestic public debt.

### Timeframe Issues
- DSF timeframe for considering risks appears excessive:
  - Breaches of debt burden indicators concentrated during early years of the projection horizon.
  - Debt forecasts have increasing underestimation errors as projection horizon is extended; even with better forecasts, significant uncertainty would remain.

### Case Studies
- Mixed effectiveness across country cases:
  - Successful signaling at least one year or more in advance in several cases (examples: Chad, Mongolia) capturing impact of expansionary fiscal policy and external debt service outlook.
  - Framework limited in other cases due to:
    - Need for stronger baseline debt projections (examples: Maldives, Djibouti).
    - Need for enhanced analysis of key risks in stress tests (examples: Sri Lanka, Samoa).
    - Need to expand assessment of broader risks (examples: Ghana, Central African Republic).
  - Annex II contains more details.

*Source: IMF staff analysis as presented in the provided chapter excerpt.*

### 20. A closer look at the technical approach underlying the DSF framework also reveals

### 20. A closer look at the technical approach underlying the DSF framework also reveals

### Major concerns with the existing DSF technical approach
- Core model and thresholds depend on how external debt distress episodes are identified; incorrect identification reduces model performance.
- Key methodological questions:
  - How well have debt distress episodes been defined?
  - How well specified and accurate is the core statistical model?
  - How are thresholds derived from the model?
  - What are limitations in the stress test methodology?

### Identification of external debt distress episodes (current limitations)
- Current approach focuses on severe distress by requiring distress signals to be observed for at least three consecutive years.
  - This ignores distress events (e.g., pre-emptive restructurings) that last less than three years.
- The DSF relies on three distress signals to identify external debt difficulties:
  - i) cumulative IMF disbursements from the General Resource Account (GRA)—under Stand-By Arrangements (SBA) and Extended Fund Facilities (EFF)—exceeding 50 percent of the member’s quota;
  - ii) restructuring of claims held by Paris Club creditors;
  - iii) accumulation of arrears on external PPG debt in excess of 5 percent of the outstanding stock of external PPG debt.
- Problems with specific signals:
  - IMF disbursements: using the 50 percent of quota cutoff can create false alarms or missed/late calls because many GRA arrangements breach the cutoff without debt difficulties.
  - Paris Club restructurings: restructuring deals typically come well after onset of debt problems; HIPC treatments are over-represented because multi-round restructurings between decision and completion points are counted as separate signals.

### Core statistical model specification (limitations and performance)
- Existing model relates probability of external debt distress to debt burden indicators and country characteristics but uses few variables.
- When the sample period was updated (seven extra years of data) and the identification of distress episodes revisited, model performance deteriorated sharply.
- Expanded specification recommended to improve predictive capacity (see later section).

### Limitations in deriving debt thresholds
- Thresholds are derived individually for each of the five debt indicators without regard to information in other debt indicators; this contradicts the DSF aggregation rule (which signals risk if any indicator breaches its threshold) and may introduce a downward (conservative) bias.
- The framework lacks a transparent policy choice for tolerating missed crises (type I error) and false alarms (type II error):
  - Present framework averages results over a range of weights for type I error from 50 to 75 percent; some observers view this as too conservative and potentially biasing thresholds downward.
- Country classification relies solely on CPIA scores (policies and institutions), omitting other predictive factors.

### Stress testing framework limitations
- Modeled shocks lack macro-linkages among key variables, contributing to miscalibration (notably of the exchange rate shock).
- Shocks are not consistently applied across external and total public debt DSAs:
  - Example: the primary balance shock has been considered only in public DSAs, not in external DSAs, despite fiscal deteriorations affecting external public debt via higher external borrowings and borrowing costs.
  - Some simulated shocks to external public debt do not inform assessment of total public debt vulnerabilities.

### Market-financing and liquidity risks not explicitly addressed
- Frontier LICs have increasingly accessed market and non-concessional financing, changing public debt profiles (shorter maturities, more diverse creditor base).
- Market access exposes countries to more frequent spikes in financing needs, particularly from bonds with bullet amortization.
  - Since 2005, 14 LICs have issued 29 Eurobonds worth US$20 billion; of the 29 issuances, 25 or US$17 billion had bullet payments.
  - Principal repayments of sovereign external debt (in percent of exports of goods and services) by frontier LICs are projected to exceed those of the 17 largest EMs over the next five years (reference to IMF-WB, 2015).
- External debt service indicators alone may be insufficient to capture liquidity risks (gross financing needs—GFNs—can be high); country teams have at times relied on analyses outside the LIC DSF (examples: Ghana and Sri Lanka).

### Overarching principles guiding proposed reforms
- Preserve the DSF’s core architecture: model-based results complemented by judgment.
- Ensure balance: provide early warnings of debt distress without generating multiple false alarms.

### Summary of proposed reform objectives
- Correct weaknesses and gaps in the framework.
- Adapt the framework to evolving LIC circumstances.
- Make the framework more transparent and simpler to use.
- Integrate key country-specific information into classification and risk assessment.
- Introduce realism tools to test key macroeconomic projection assumptions.
- Drop features that add little practical value and add tools for deeper risk assessment and firmer judgment application.

### Key proposed reforms (high-level)
- Update the approach to identify debt distress episodes.
- Expand specification of the core statistical model to include key country-specific fundamentals.
- Integrate the model changes into country classification and risk assessment.
- Introduce realism tools for macro projections and stress testing.
- Simplify debt indicators and thresholds, and redesign stress tests with macro-linkages.
- Add tools to assess domestic debt vulnerabilities, market-financing pressures, and diversity of vulnerabilities within moderate-risk countries.
- Enhance guidance for the application of judgment, including on marginal/transitory breaches and on severe domestic-debt vulnerabilities.

### A. Strengthening the statistical model for predicting debt distress
- Purpose: identify key factors affecting probability of debt distress and derive debt thresholds consistent with different debt-carrying capacities.
- Revised procedure for identifying distress episodes:
  - Include distress periods that last only one or two years (except when the shorter episode is driven solely by occurrence of external arrears).
  - Redefine criteria for distress signals:
    - IMF disbursement identification refocused on large upfront financing disbursements.
    - Incorporate information from new debt restructuring databases.
    - Improve treatment of timing of debt restructurings.
- Expanded model specification:
  - Add proxies of capacity to repay:
    - international reserves scaled by imports,
    - remittances scaled by nominal GDP,
    - proxy for global shocks: world growth.
  - New model estimated using data only from LICs (previous model included MIC data).
- Variables considered but not included (did not improve predictive performance):
  - real GDP per capita, current account balance/GDP, foreign direct investment/GDP, a measure of the country risk premium, global interest rates, commodity prices, terms of trade, measures of natural disasters and conflicts.
  - Trade openness was explored but performed worse than remittances in the specification; remittances-only model performs better.
  - Estimating separate models for small states was technically infeasible due to limited distress episodes; judgment and tailored scenario stress tests can address small-state vulnerabilities.

### B. Classifying countries by debt-carrying capacity (reforms)
- Existing approach: classification relies exclusively on the CPIA and is backward-looking.
- Reformed approach:
  - Construct a composite indicator covering:
    - CPIA,
    - country growth,
    - reserve coverage,
    - remittances,
    - world growth.
  - Weights for the composite indicator are given by the estimated coefficients of these variables in the model.
  - Use the cross-country distribution of the average composite indicator over a 10-year period (2005–14) to establish classification cutoffs:
    - Weak if composite indicator < 25th percentile of distribution.
    - Medium if composite indicator between 25th and 75th percentiles.
    - Strong if composite indicator > 75th percentile.
  - Fit countries to classification using the country-specific composite indicator.
  - Integrate forward-looking elements:
    - Calculate country-specific composite indicator based on latest five years of historical data and the first five years of projections.
  - Require two consecutive signals where the country’s composite indicator exceeds its classification cutoff to change classification (mitigates undue volatility).
  - Anchoring on a 10-year average attenuates sensitivity to cyclical fluctuations.

### Reform implications summarized (from the comparative table)
- Core debt distress model:
  - From: identifies only severe episodes; few country-specific explanatory variables.
  - To: enhanced methodology identifying all distress episodes; expanded specification with key country-specific fundamentals.
- Country classification:
  - From: relies exclusively on CPIA; backward-looking.
  - To: composite measure (CPIA, growth, reserve coverage, remittances, world growth); incorporate forward-looking elements.
- Realism tools:
  - From: support stronger baseline projections and implement classification (e.g., realism of projected fiscal adjustment and investment-growth nexus).
- Debt indicators and thresholds:
  - From: complex (five debt indicators and 24 thresholds); thresholds derived individually introducing conservative bias.
  - To: simplification to four debt indicators and 12 thresholds; thresholds derived jointly in line with aggregation rule (eliminating a source of conservative bias).
- Standardized stress tests:
  - From: 16 stress tests lacking macro-linkages and non-common testing across external and public DSA.
  - To: 8 common re-calibrated and re-designed stress tests across external and public DSA, incorporating macro-linkages.
- Tailored stress tests:
  - Continue to evaluate country-specific scenario risks (e.g., natural disasters).
- Assessment of other potential risk factors:
  - From: tools to assess domestic debt vulnerabilities.
  - To: tools to assess domestic debt vulnerabilities, market-financing pressures, and diversity of vulnerabilities in moderate-risk countries.
- Enhanced guidance for judgment:
  - From: guidance on marginal/transitory breaches.
  - To: guidance also on severe domestic-debt vulnerabilities and exposure to external market-financing pressures, among other factors.

*Source: pp082217lic-dsf - 20. A closer look at the technical approach underlying the DSF framework also reveals*

### 35. The proposed new classification methodology has significant advantages over relying

### 35. The proposed new classification methodology has significant advantages over relying

### Advantages of the new classification methodology
- The new approach would give the overall framework greater predictive power. Ignoring the rich information brought in by the additional country-specific variables would misclassify countries and raise both missed crises and false alarms.
- Threshold effects—the fact that a small change in the CPIA can produce large changes in debt thresholds under the current framework—would be mitigated. The several components of the new composite indicator—whose co-movements are not perfectly synchronized—and the long-term averages used for its computation would make it less likely that a movement to another classification category would be due to a change in only one variable.
- The determinants of the country-specific composite indicator would be shown in the DSA output to facilitate understanding of what is driving the classifications and where any changes have come from.
- Forecasts of the additional variables in the composite indicator are routinely produced in the WEO database and individual DSAs. Recognizing its slow-moving nature, the forecast for the CPIA rating would consist of its most recent value.
- The use of forward-looking information would allow country authorities to understand the relationship between their policy framework and their debt-carrying capacity, enhancing engagement. More generally, classification would be more transparent, as the additional country-specific information is objective and readily available to country teams, country authorities, and other stakeholders.

### Alternative approaches considered
- Staff considered an alternative to the debt threshold approach to classification (probability threshold approach) but does not see it as viable.
- Conceptual comparison:
  - Probability threshold approach: uses country-specific information more efficiently to predict debt distress and avoids threshold effects implied by grouping country-specific information into three classification categories.
  - Debt threshold approach: conditional on a sound classification of countries’ debt-carrying capacity, has operational advantages:
    - (i) rules out situations where statistical models would imply extreme debt thresholds for countries with very weak or very strong fundamentals (which could imply zero borrowing space for some LICs and implausibly-high debt limits for others);
    - (ii) allows debt targets in Fund-supported programs, the Fund’s DLP and the Bank’s NCBP to be more directly mapped to debt thresholds than to probability cutoffs;
    - (iii) debt thresholds are more intuitive and easier to interpret than some maximum level for the probability of debt distress.
- Note on threshold effects: small changes in predictors can lead to discrete jumps in debt thresholds (example: CPIA 3.24 → threshold 30 percent; CPIA 3.26 → threshold 40 percent, per IMF (2012), Appendix 1).

### C. Improving the Realism of Baseline Projections — Overview
- To support stronger baseline debt forecasts and implementation of the new classification methodology, the new DSF would include realism tools (already included in the DSF for market access countries).
- Purpose of realism tools:
  - Provide point of comparison for forecasts (country history, cross-country experience, relationships from economic theory).
  - Inform users where important drivers of macroeconomic baseline debt projections deviate markedly from experience (optimistic or pessimistic bias).
  - Highlight key assumptions underpinning projections to focus discussion; tools are not prescriptive.

### Realism tool 1 — Decomposition of past and projected drivers of debt dynamics
- DSF users would be shown the evolution of projections of external and public debt to GDP ratios over DSA vintages (one-year and five-years ago).
- The tool would provide summary charts to help identify and scrutinize marked changes in historical and projected drivers of debt dynamics (illustrative country shown in Figure 8).
- Practical implications:
  - A high past contribution of unexpected primary deficits would caution against projecting an excessive reliance on fiscal adjustment.
  - The tool would shed light on differences between historical and projected contributions of the current account and FDI flows to external debt, which may signal potential optimism or pessimism in projected reserve accumulation.

### Realism tool 2 — Contrast projected fiscal adjustment with historical experience
- Comparison group focuses on LICs requesting a Fund-supported program.
- The tool would present the distribution of observed headline primary fiscal adjustment over a three-year horizon, against which a country’s projected primary fiscal adjustment would be compared.
- Operational trigger:
  - A fuller discussion of the credibility of the fiscal path would be called for if the projected primary fiscal adjustment over any three years during the projection horizon exceeds, say, 2½ percent of GDP, which is approximately equal to the top quartile of the distribution of observed primary fiscal adjustment (Figure 9, illustrative case).
- Note: Ideally would use cyclically-adjusted primary balances, but these are difficult to compute for LICs given uncertainty in output gaps; measuring headline primary balances may include exogenous drivers (e.g., natural resource projects).

### Realism tool 3 — Benchmarks for assessing growth assumptions (public investment and fiscal adjustment)
- Assessing the investment-growth nexus:
  - The DSF would use a simple growth accounting framework to decompose projected growth rates into: (i) contribution from changes in government capital stock due to public investment dynamics, and (ii) contribution from other sources (Annex IV methodology).
  - The two sources of growth could be compared with historical data and past projections to trigger deeper discussion of underlying assumptions (illustrative case in Figure 10.a).
  - A formal assessment of public investment’s impact on growth is beyond the DSF’s scope; more detailed analysis may be done outside DSF (IMF’s Debt-Investment-Growth model; World Bank’s Long-Term Growth model).
- Assessing the impact of fiscal adjustment on growth:
  - Negative growth surprises have been identified as the main factor derailing fiscal consolidations.
  - The tool would show the impact of planned fiscal adjustment on growth projections under a range of plausible fiscal multipliers and persistence parameters, allowing comparison with the baseline projected growth path (Figure 10.b; Annex IV methodology).
  - While uncovering the growth path consistent with a neutral fiscal stance can be useful, it is challenging for LICs because borrowing capacity is very weak in absence of fiscal adjustment.

### D. Determining Mechanical Risk Signals — Methodology and Rules
- Key output: mechanical risk signal derived by comparing baseline and stress-test debt projections with debt thresholds, using an aggregation rule and risk-weighting scheme (weight on missed crises vs false alarms).
- Simple rule for mechanical risk signal:
  - Low risk: all debt burden indicators below corresponding thresholds under baseline and stress test scenarios.
  - Moderate risk: any breach under stress tests but not under baseline.
  - High risk: at least one breach under baseline.
- Two methodological improvements to threshold setting:
  - Introduce a transparent policy choice for the tolerance of missed crises and false alarms: proposed to be set at 2:1 (recognizing importance of strong early warning framework). The existing complex averaging produces an effective weight around this level.
  - Derive debt thresholds for the three classification categories in a manner consistent with the DSF’s aggregation rule to efficiently rebalance the role of debt burden indicators per their capacity to signal debt distress (see Annex III).

### Simplification of mechanical signal generation and stress testing
- The aggregation rule remains unchanged, but:
  - Number of thresholds, debt burden indicators, and standardized stress tests would be considerably reduced.
  - Reduced standardized stress tests allow room for tailored scenario stress tests to better capture key risks.
  - The procedure for generating mechanical risk signals would be considerably simplified while introducing a more comprehensive stress testing framework (Figure 11).

### Debt burden indicators and thresholds — Streamlining changes
- Redundancies to be addressed:
  - Drop the indicator on the present value of external debt-to-fiscal revenues: including or excluding this indicator does not affect missed crises and false alarms; historically it has not played a role in signaling high risk of debt distress.
  - Discontinue remittances-augmented thresholds: remittances will be included directly in the statistical model informing country classification, removing the need for the indirect approach.
  - Shorten projection horizon for mechanical risk signals from 20 to 10 years:
    - Evidence shows threshold breaches are concentrated in first five years and high uncertainty exists in long-term projections.
    - New DSF template would continue to report 20 years of projections; staff judgment on 10–20 year period could override the mechanical risk signal, focusing attention on determinants of long-term breaches.

### Predictive performance and thresholds — Findings and quantitative results
- Streamlining improves predictive performance relative to existing benchmarks:
  - Under proposed reforms, probability of false alarms is significantly reduced (by 10 percentage points), while model’s capacity to anticipate debt distress is improved.
  - Re-estimating thresholds using existing model and updated data (1970–2014) shows existing re-estimated model explains new episodes poorly; tighter thresholds increase mechanical rate of false alarms to about 70 percent without overall predictive gains (see Annex III).
- Simplifications summary (counts preserved from figures):
  - Debt burden indicators: Existing 5 → Reformed 4
  - Thresholds: Existing 24 → Reformed 12
  - Standardized stress tests: Existing 16 → Reformed 8
- Table 2 statistics (Based on DSF's aggregation rule):
  - Existing 2/ (performance of the existing framework in predicting debt distress episodes as identified by the new approach): Type I and II errors are estimated at 18 and 48 percent, respectively (note 2/).
  - Existing (updated) 3/ (performance of re-estimated existing model on new sample 1970–2014): Type I error 0.22; Type II error 0.50; Loss function 1/ 0.31.
  - New 4/ (performance of the reformed framework): Type I error 0.06; Type II error 0.68; Loss function 1/ 0.27; alternate Loss function 0.24 shown for another metric.
  - Note 1/: Weighted sum of errors (weight on type I error equal to 67 percent).
  - Note 3/: Re-estimated existing model drove up false alarms to about 70 percent in earlier discussion.
- Re-estimated external debt thresholds (Table 3): Old vs New values by Country Classification
  - PV of debt-to-GDP:
    - Weak: Old 30 → New 30
    - Medium: Old 40 → New 40
    - Strong: Old 50 → New 55
  - PV of debt-to-exports:
    - Weak: Old 100 → New 140
    - Medium: Old 150 → New 180
    - Strong: Old 200 → New 240
  - Debt service-to-revenue:
    - Weak: Old 18 → New 14
    - Medium: Old 20 → New 18
    - Strong: Old 22 → New 23
  - Debt service-to-exports:
    - Weak: Old 15 → New 10
    - Medium: Old 20 → New 15
    - Strong: Old 25 → New 21
- Interpretation:
  - Re-estimated thresholds imply some additional room to borrow if countries manage debt service well.
  - Debt stock thresholds are maintained or increased for all countries; weak and medium debt-carrying capacity countries face lower debt service thresholds.
  - Because the new classification scheme upgrades classifications after bringing in key fundamentals, not all countries effectively face lower debt service thresholds.

*REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES: PROPOSED REFORMS — INTERNATIONAL MONETARY FUND*

### 47. The new framework would apply a smaller set of standardized shocks in the external

### pp082217lic-dsf - 47. The new framework would apply a smaller set of standardized shocks in the external

### Standardized stress tests: scope and purpose
- A smaller common set of standardized stress tests would be applied to both external and public DSAs to eliminate inconsistencies in the risk assessment.
- The GDP deflator stress test and all permanent shocks would be dropped (the “historical” scenario would remain as a realism check).
- The retained standardized stress tests would include shocks to:
  - (i) real GDP growth
  - (ii) export growth
  - (iii) primary balance
  - (iv) other flows (transfers and FDI)
  - (v) exchange rate depreciation
  - (vi) contingent liabilities
  - (vii) a combination shock
- The combination shock: reflects multiple simultaneous shocks; it applies half the magnitudes of all stand-alone shocks, and incorporates individual macro interactions as assumed under each individual shock.

### Re-calibration and redesigned interactions
- Main macro-variable interactions to be introduced:
  - (i) the export growth shock would negatively impact GDP growth
  - (ii) the exchange rate shock would positively affect inflation and net exports, which would mitigate the impact of a sharp FX depreciation on debt ratios
  - (iii) the real GDP growth shock would negatively affect inflation and the primary balance
  - (iv) the primary balance shock would increase commercial borrowing costs
- Point of application of shocks:
  - For all stress tests, post shock values would be set to the historical average minus one standard deviation—or the baseline projection minus one standard deviation—whichever is lower for the relevant periods.
  - This change aims to reduce the likelihood of underestimation of primary balance shocks and improve the accuracy of all stress test shocks.
- Contingent liability shock redesign:
  - The existing contingent liability shock (the “other debt-creating flows” stress test) would be re-calibrated to 5 percent of GDP (from 10 percent).
  - The 5 percent calibration is aimed at capturing potential liabilities stemming from financial sector vulnerabilities.
  - A tailored shock could expand scope where relevant to capture larger financial sector risks or exposures not covered under the debt concept in the DSA.
- Note on empirical basis:
  - Modeling of elasticities for interactions draws on an event study and on empirical evidence from the existing literature (see Annex I).
  - The 5 percent of GDP calibration is based on the average increase in debt-to-GDP ratios observed for 44 banking crisis episodes for LICs since the 1980s, as in Laeven and Valencia (2013).

### Tailored scenario stress tests: triggers and examples
- New tailored scenario stress tests would be introduced and triggered for countries assessed as exposed to particular risks (default settings based on the median shock for LICs; country teams encouraged to customize).
- Tailored tests would reflect risks common to groups of LICs; some countries may face more than one test. Examples:
  - Natural disasters:
    - Required for LICs identified as particularly vulnerable to natural disasters (see IMF, 2016).
    - Default magnitudes: a one-off shock to public debt (capturing fiscal impacts), and real GDP and export growth decline in the year of the shock, calibrated using data covering natural disasters during 1950–2015.
    - Country teams encouraged to adjust parameters to country-specific circumstances and explain adjustments, including any baseline-embedded disaster impacts.
  - Contingent liabilities supplement:
    - Triggered when significant elements of the public sector are not covered by the public debt concept used in the DSA (e.g., other parts of general government, PPP exposures, or unaccounted state enterprise exposures) or when financial sector risks to the sovereign balance sheet exceed the standardized contingent liability shock.
    - Scenario: a one-time increase in the debt ratio scaled to the size of potential exposures in sectors not covered by the country-specific public debt concept.
    - This test would help eliminate a disincentive to achieve full and transparent disclosure of public sector contingent liabilities.
  - Commodity export price shock:
    - Triggered for LICs where commodity exports represent at least 80 percent of merchandise exports.
    - Proposed scenario: capture impact of a sudden one standard deviation decline in commodity export prices (using the distribution underlying WEO forecasts), with macro interactions incorporated, based on staff event analysis and recent studies (IMF, 2015c and Aslam and others, 2016).
    - The 80 percent cutoff captures 36 countries (see Annex I).
  - Market-financing stress test:
    - Triggered for LICs with access to market-based financing.
    - Tailored combination shock to capture stressed external commercial borrowing conditions from a deterioration in global risk sentiment: increases in the cost of new external commercial borrowing, temporary nominal depreciation, and shortening of maturities of new external commercial borrowing.

### Customized alternative scenarios and DSF template enhancements
- Customized alternative scenarios would remain permitted to assess idiosyncratic risks not covered by standardized and tailored tests (e.g., delays in megaprojects, active conflict scenarios).
- The DSF template would be enhanced to better allow users to fully design customized scenarios.
- The revised Staff Guidance Note would address good practice in handling such risks, including those related to conflicts.

### Analyzing other potential risk factors: overview
- Staffs are called to examine risks not captured in the core statistical model: domestic debt vulnerabilities, risks to the balance of payments and government exposure from elevated private external debt, and risks from shifts in market sentiment for LICs with a high share of public debt on market terms.
- The new framework would continue this approach and include new tools for:
  - Assessing public debt-related risks
  - Assessing market financing-related risks
  - Summarizing diverse risk characteristics of countries at moderate risk of debt distress via an automated DSF heat map

### Total public debt: methodology and benchmarks
- The new DSF would require assessment of risks from domestic debt levels through analysis of total public debt, given increasing importance of domestic debt markets and non-resident participation.
- Methodological enhancements:
  - An alternative identification procedure to capture de facto domestic default is used, based on two proxies (see Annex III).
  - A noise-to-signal (NTS) modelling approach is used to derive benchmarks for the PV of total public debt, contrasting with the probit models used in the 2012 LIC DSF review.
- Benchmarks:
  - The newly-estimated benchmarks for the PV of total public debt are not substantially different from existing benchmarks (Table 4).
  - The framework would signal:
    - High risk of public debt distress if any of the four external debt burden indicators or the public debt indicator breach their corresponding threshold/benchmark
    - Moderate risk if thresholds/benchmark were breached in stress tests
    - Low risk if thresholds/benchmark were not breached in either the baseline or stress test scenarios
  - Benchmarks were rounded up to align the strong classification category with the MAC DSA high-risk benchmark.

### Market-financing pressures: early warning tool
- Rationale:
  - Existing debt service indicators and the market stress test provide information, but historical experience suggests they are insufficient alone to capture liquidity risks for countries with market access under the baseline scenario.
- New tool:
  - Assesses whether estimated gross public financing needs (GFNs, as a share of GDP) and prevailing EMBI spreads point to risks of debt distress.
  - Noise-to-signal approach applied to external debt distress episodes in a sample of frontier LICs and MICs between 1995 and 2015 produces estimated benchmark levels for GFNs to GDP and for bond spreads.
  - The two indicators tend to move together in the run-up to debt distress.
- Benchmarks and signal rule:
  - A breach of both benchmarks—GFNs above 14 percent of GDP and EMBI spreads higher than 570 bps—would signal heightened liquidity needs under adverse external market conditions, increasing rollover risks and the likely need to seek exceptional financing.
  - In such circumstances, country teams would be expected to provide an in-depth assessment of liquidity needs and creditor base composition, as input into final risk-assessment conclusions.

### Moderate risk category: “space to absorb shocks”
- A tool would assess how far a country’s debt burden indicators are from crossing thresholds under the baseline—“distance to breaching thresholds” or “space to absorb shocks”.
- Classification for countries rated moderate risk:
  - “Limited space to absorb shocks”:
    - If a median-size observed shock would lead to threshold breaches under the baseline, excluding marginal and/or temporary breaches.
    - Median observed shock estimated at 20 percent of the threshold for debt stock indicators, and 12 percent for debt service indicators.
  - “Substantial space to absorb shocks”:
    - If threshold breaches would not occur under the baseline except for shocks in the top quartile of observed distribution.
    - Top quartile observed shocks estimated at 40 percent of the threshold for debt stock indicators, and 35 percent for debt service indicators.
- The assessment would illuminate robustness of debt positions for moderate-risk countries; it would not have operational implications for Fund/Bank debt policies.

### The application of judgment
- Mechanical risk signals provide a first cut for external risk rating, but staff judgment can modify ratings; any judgment must be justified in the DSA write-up.
- Factors that could justify adjustment of external risk rating:
  - Breaches of thresholds assessed to be marginal and/or transitory in nature—staff expected to avoid mechanical application of risk rules in such cases; a technical specification will be developed for the Staff Guidance Note.
  - Severity of domestic debt vulnerabilities—e.g., where non-residents hold a sizable share of domestic government debt; such concerns should be flagged in the summary risk assessment even if not warranting rating adjustment.
  - Exposure of the sovereign to external market-financing pressures—could justify adjustment if sufficiently severe; would warrant mention even if not sufficient to adjust the external risk rating.
  - Other country-specific considerations not captured by the statistical framework—examples include availability of sizable public financial assets, conflict risks, long-term considerations (e.g., vulnerability to climate change), availability of insurance-type arrangements, and collateralized financing arrangements.
- The Staff Guidance Note, prepared by IMF and WB teams, would discuss considerations for these factors in final external risk determinations.

*Source: REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES: PROPOSED REFORMS (excerpts provided).*

### 63. Having conducted the various tests and assessments discussed above, and brought in

### pp082217lic-dsf - 63. Having conducted the various tests and assessments discussed above, and brought in

### Summary risk-assessment outputs
- Staffs are expected to provide:
  - A rating of the risk of external debt distress using categories: low, moderate, high, or in debt distress.
  - A rating on the overall risk of debt distress where vulnerabilities linked to public domestic debt levels are a serious concern: low, moderate, high, or in debt distress.
  - A full discussion of the main risks to this assessment, including factors such as data coverage, macroeconomic uncertainty, policy implementation risks, and global factors.
- The summary risk assessment should:
  - Explain how staffs reached their conclusion, covering results from mechanical features and deviations from mechanical risk signals.
  - Comment on the evolution of debt risks and vulnerabilities over time and identify key risks/mitigating factors that could shift the risk assessment going forward.

### Back-testing results — country classification and mechanical signals
- General notes on back-testing:
  - Mechanical back-testing results cannot be taken as literal predictions because they use a vintage of macro data and projections.
  - Most DSAs (about 95 percent) used in the simulations were produced during 2015–16.
  - Country classifications were informed by macro data and projections from the 2016 Fall WEO.
  - Mechanical risk signals do not factor in full customization of tailored stress tests nor staff judgment, which has been applied in about 25 percent of DSAs since DSF inception.
- Country classification (out of 67 countries for which LIC DSAs are currently produced):
  - Unchanged: 40 countries (60 percent of the sample).
  - Upgraded: 22 countries (33 percent).
  - Downgraded: 5 countries (7 percent).
  - Note: CPIA still accounts for 45 percent of the CI; most upgrades driven by large reserve positions and/or sizable remittances flows.
- Mechanical external risk signals (out of 67 countries):
  - Unchanged: 47 countries (70 percent of the sample).
  - Upgraded: 12 countries (18 percent).
  - Downgraded: 8 countries (12 percent).
  - Aggregate distribution across risk signal categories remains broadly unchanged relative to the existing framework.
- Assessment of other risk factors from back-testing:
  - Public debt: Broadly parallels current DSF; re-estimated total public debt benchmarks are broadly unchanged and most assessments under the existing framework (about 90 percent of the cases) are reaffirmed.
  - Market-based risks: Two countries out of 17 LICs with market access would be flagged. Five countries do not breach any benchmark; in other cases either only one benchmark is breached or assessment is inconclusive due to unavailability of EMBI spreads.
  - “Space to absorb shocks” within moderate risk category:
    - Number of countries currently rated as moderate risk: 32.
    - Moderate with limited space: 16.
    - Moderate: 10.
    - Moderate with substantial space: 6.
    - Of the 16 flagged as “with limited space to absorb shocks,” 7 already have a mechanical high risk signal.

### Discount rate for the LIC DSF
- Importance: DSF uses present-value debt indicators; discount rate choice affects measured debt burden and risk.
- Historical methodology and adjustments:
  - 2005: discount rate set at USD CIRR of 5 percent; rule to adjust by 100 bps if 6-month average USD CIRR deviated by at least 100 bps for a 6-month period.
  - Adjusted to 4 percent in 2009 and 3 percent in 2012.
  - 2013: methodology changed to long-term average (10-year average of the USD CIRR) plus a margin; discount rate set at 5 percent.
- Two methodologies assessed in review:
  - Existing CIRR methodology: applying the existing methodology today, the discount rate would be 4.75 percent.
  - LICs’ nominal per capita income-growth linkage: using median nominal dollar GDP growth rate for LICs at 7 percent and median population growth rate at 2.2 percent implies a discount rate of 4.8 percent.
  - Note on medians: The median dollar nominal GDP growth rate is 7 percent when based on (i) 5 years of history and 5 years of projections, (ii) 10 years of history and 10 years of projections, or (iii) 5 years of history and 10 years of projections, using data submitted in the latest LIC DSAs. The median population growth rate is 2.2 when based on (i) 5 years of history and 5 years of projections or (ii) 10 years of history and 5 years of projection, using WEO data.
- Proposal:
  - Keep the current 5 percent discount rate.
  - Revisit this decision in future DSF reviews.
  - Maintain a unified discount rate for the LIC DSF, the debt limits policy (DLP), the non-concessional borrowing policy (NCBP), and the grant element calculator to avoid operational complications from different PV calculations.

### Issues in implementation, capacity-building, and timeline
- Expected interactions and disclosure:
  - Greater engagement with country authorities under the new country classification approach to discuss macro projections and policy frameworks that determine classification.
  - Customization of tailored stress tests will require discussion with country authorities on appropriate scenario parameters.
  - Technical/country-specific considerations underlying application of judgment and analysis of other risk factors will be fully disclosed and justified in DSA write-ups after discussion with country authorities; DSA write-up will incorporate authorities’ views.
  - Increased policy dialogue on debt management capacity and strategy where additional room to borrow exists but debt service risks require attention.
- Support for implementation:
  - Updated materials to support implementation (expected to be available by end-2017):
    - Thorough revision and simplification of the Staff Guidance Note; expected issuance by end-December 2017.
    - DSF template to be re-programmed, simplified, streamlined, with automated features; composite indicator calculations and assumptions for tailored scenario shocks to be automated and transparently reported.
  - Interactions with country authorities through three channels:
    - Joint IMF-WB high-level seminar during the 2017 Annual Meetings.
    - Technical discussions with country counterparts during Fund/Bank missions.
    - Expanded training program: with donor support, external training increased from nine missions during FY 2017 to sixteen during FY 2018, including four workshops in regional training centers during the Fall and nine training missions during the first half of 2018.
      - Historical participation: During 2016–17, 175 representatives from 40 countries participated in one-week training sessions; nearly 850 government officials familiarized themselves with the online LIC DSF materials through the IMF Massive Open Online Course on debt sustainability and debt management.
- Effectiveness date and operational timing:
  - Proposed effective date for new framework: July 1, 2018.
  - Predicated on updated Staff Guidance Note and DSF template finalized by end of December 2017.
  - Effectiveness date aimed to allow teams to complete DSAs for all countries under the new DSF prior to July 1, 2019 — the cut-off date for determination of changes to country “traffic lights” under the IDA grant allocation framework applicable to IDA’s fiscal 2020 allocations and IDA19 projections.
  - Missions conducted before DSF enters into effect will produce DSAs based on the old framework; timing of final mission determines whether new or old framework applies when multiple missions are needed.
- Ongoing support after rollout:
  - Continued training and technical assistance to support capacity building and integration of debt management strategies in DSAs.
  - Staff will explore options to link the DSF with debt management tools, in particular the Medium-term Debt Management Strategy (MTDS) analytical tool.

*REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES: PROPOSED REFORMS — excerpt pp082217lic-dsf*

### 77. The flexibility embedded in the DLP and NCBP should help smooth any implications

### 77. The flexibility embedded in the DLP and NCBP should help smooth any implications arising from changes in final risk ratings in the wake of this review.

### DLP: implications and embedded flexibility
- If a country’s risk rating is upgraded, the country would face looser debt conditionality (or no debt conditionality) under a Fund-supported program.
- If a country’s risk rating is downgraded, it could still benefit from the flexibility embedded in the current DLP:
  - A country downgraded from moderate to high could still access non-concessional resources to finance critical development projects or to implement debt management operations that improve the overall debt profile.
  - A country downgraded from low to moderate would generally continue benefiting from the flexibility of the DLP as it provides room for non-concessional borrowing (provided the borrowing does not lead to a deterioration in the moderate risk rating).
- Where the risk rating of a member under an existing arrangement changes as a result of the new DSF, staff would seek to reach new understandings when changes in debt conditionality are called for by the DLP.
- Such revisions to debt conditionality would be considered by the Executive Board in the subsequent staff report.

### NCBP: treatment for IDA-only non-gap countries and options preserved
- Similar considerations apply under the NCBP for all IDA-only non-gap countries that are either grant recipients in the current fiscal year or MDRI recipients, irrespective of their risk rating of debt distress.
- All countries will continue to be provided with a choice between loan by loan exceptions to the NCBP.
- Countries at low or moderate risk of debt distress are also provided the option to seek nominal ceilings on new non-concessional external PPG borrowing.
- Countries at low or moderate risk of debt distress with adequate capacity may also opt for a PV ceiling on total new external PPG borrowing.
- The options available to countries will remain consistent with the conclusion of the joint IMF-WB debt management capacity assessment.
- IDA’s grant/loan allocation mix will not be affected by the changes in country risk ratings resulting from the application of the new DSF until the fiscal year 2020 (starting July 1, 2019).
- Note: Each year, IDA determines its grant/allocation mix for the next 12 months based on the external risk ratings available by end-June.

### Conclusion: expected benefits of proposed DSF reforms
- The proposed reforms imply a significant overhaul of the DSF, bringing significant advantages to all stakeholders.
- Expected outcomes:
  - Address several technical criticisms of the existing framework.
  - Reduce an excessive rate of false alarms (with improved predictive power to signal debt distress events).
  - Incorporate additional key country-specific information into country classification and derivation of thresholds, improving customization of risk assessments to country conditions.
  - Provide a basis for a richer dialogue on how policy actions affect debt-related risks.
  - New realism checks and stress test reforms, enhanced guidance on the application of judgment, and greater attention to domestic debt vulnerabilities and market-financing risks will allow a more informed dialogue on risks and trade-offs for national policy-makers.

### Issues for discussion (questions posed to Directors)
- Do Directors agree that an expanded set of country-specific information (including reserve coverage and remittances flows) should be used to underpin the assessment of countries’ debt-carrying capacity in the framework?
- Do Directors see merit in introducing realism tools to promote a deeper understanding of key assumptions underlying baseline macroeconomic projections, including the assumed relationship between public investment and growth, and to support stronger debt projections?
- Do Directors agree with the need to streamline the mechanical framework, by reducing the number of debt indicators, debt thresholds and standardized stress tests, and to rebalance the debt thresholds?
- Do Directors support the proposed enhancements to the stress testing framework, including the recalibration and new features of standardized stress tests (i.e., macro-interactions) as well as the inclusion of tailored scenario stress tests to assess key risks facing LICs?
- Do Directors see a need for a better assessment of broader risks stemming from high domestic debt levels and market-financing pressures, and a tool for characterizing countries in the moderate risk category in terms of its “space to absorb shocks”?
- Do Directors see merit in enhancing the Staff Guidance Note to ensure a more uniform and transparent application of judgment?
- Do Directors support the proposed timeline for the implementation of the framework?

### Appendix I — Overview of the LIC DSF: core methodology and mechanics
- A model explaining the probability of external debt distress is at the core of the methodology underpinning the DSF:
  - Given an approach to identify external debt distress episodes, the probability of experiencing such events is explained through probit models controlling for debt burden indicators, the strength of institutions and policies—measured by the CPIA—, and country growth.
- Country classification and thresholds:
  - Countries are classified per their borrowing capacity in three categories: weak, medium, and strong performers. This categorization is exclusively based on the CPIA.
  - Once countries are classified, debt thresholds derived for each category from the probit models are assigned.
- Mechanical determination of preliminary risk signals:
  - Forecasts for five debt burden indicators are produced using baseline macro projections and a battery of stress tests: PV of PPG external debt to GDP, exports, and fiscal revenues; and PPG external debt service to exports and fiscal revenues.
  - Aggregation rule for mechanical risk signals:
    - (i) if none of the forecasted debt burden indicators exceed their corresponding thresholds under the baseline and stress test scenarios, the DSF would signal a low risk of debt distress;
    - (ii) if all baseline forecasts are below their thresholds but at least one forecast exceeds its threshold under the stress test scenarios, the DSF would signal a moderate risk;
    - (iii) if at least one baseline debt forecast exceeds its threshold, the DSF would signal a high risk;
    - (iv) significant or sustained breach of thresholds, actual or impending debt restructuring negotiations, or the existence of arrears would generally suggest that a country is in debt distress.
- Staff judgment:
  - The mechanical risk signals are combined with staff judgment to incorporate country-specific considerations and technical elements (e.g., whether breaches are marginal and/or one-off) that cannot be captured by the core model to make a final determination of the risk rating of external debt distress (Low/Moderate/High/In Debt Distress).
- Features incorporated in the 2012 review:
  - i) use of remittances-augmented thresholds, only applied to countries receiving sizable remittances;
  - ii) introduction of the “probability approach”, an alternative technical methodology, to be applied at “borderline cases”;
  - iii) qualification of risks stemming from total public debt or private external debt.
- Definitions and criteria referenced:
  - Countries eligible to incorporate remittances into the DSF mechanical analysis have remittances flows greater than 10 percent of GDP and greater than 20 percent of exports of goods and services. Both ratios are measured on a backward-looking, three-year average basis.
  - A borderline case is defined as one where the largest breach, or near breach, of a threshold under any scenario falls within a 10-percent band around the threshold.

*Review of the Debt Sustainability Framework for Low Income Countries: Proposed Reforms (selected content).*

### 3. Setting adverse shocks at one standard deviation remains appropriate. Shocks of this

### 3. Setting adverse shocks at one standard deviation remains appropriate. Shocks of this

### Rationale for one standard deviation shocks
- A one standard deviation adverse shock from the mean yields a post-shock value at about the 16th percentile for variables that follow normal distributions.
- Shocks of this magnitude have a chance of about 16 percent to occur each year, or around once every six years on average.
- This time horizon coincides with the length of business cycles in major economies; the average length of business cycles in the US is about 5¾ years.
- Note: distributions of economic variables tend to have fatter tails than normal distributions, so the likelihood of shocks of magnitudes at or above one standard deviation could be slightly higher than 16 percent.
- Note: Some studies suggest business cycles may be shorter in developing countries; for such countries, one standard deviation shocks would represent more severe shocks than those observed in a typical business cycle.

### Empirical assessment of standardized stress tests (LIC DSAs 2008–15)
- Method: For LIC DSAs produced since 2008, staff tested whether post-shock values in stress tests (second and third year of projections; first year for FX shock) correspond to about the 16th percentile of realized distributions.
- Findings (post-shock value percentiles vs outturns; Actual < Post-Shock Value / Actual > Post-Shock Value):
  - GDP Growth Shock: 18.2 / 81.8 (in percent)
  - Exports Growth Shock: 19.9 / 80.1 (in percent)
  - Primary Balance Shock: 31.4 / 68.6 (in percent)
  - Non-Debt Flows Shock: 15.3 / 84.7 (in percent)
  - Exchange Rate Shock (annualized): 16.6 / 83.4 (in percent)
- Data sources: LIC DSAs during 2008–15 and Fund staff calculations. For the exchange rate shock, WEO data covering 1990–2015 was used.

### Primary balance shock: identified weakness
- The primary balance shock appears too small relative to other shocks.
- There is about a 31 percent chance that the observed primary deficit was higher than the post-shock value used in the stress test scenario (Actual < Post-Shock Value = 31.4 percent).
- Interpreting under normal distribution: this corresponds to only about ½ standard deviation below the sample mean, indicating the primary balance shock has been milder than envisaged.

### Exchange rate shock magnitude
- The exchange rate shock in DSF is a 30 percent depreciation in one year; to compare with other shocks that normally last two years, this is annualized to about 15 percent annually.
- Using LIC data over 1990–2015, a 15 percent nominal depreciation lies at about the 17th percentile of the actual distribution of exchange rate changes, which is in line with other shocks.

### Bias from historical averages and proposed correction
- Post-shock values in DSA stress tests are set at one standard deviation below historical averages; this assumes 10-year historical averages and standard deviations are good predictors for projection years.
- For primary balance, the assumption fails: outturns of primary balances were lower than historical averages in about 2/3 of observations.
- On average, the actual primary balance to GDP ratio was 1.2 percentage points lower than historical averages.
- Table comparing actuals to historical averages and to baseline projections (in percent):
  - GDP Growth: Actual < Historical Average 46.8; Actual > Historical Average 53.2; Actual < Baseline Projection 52.8; Actual > Baseline Projection 47.2
  - Exports Growth: Actual < Historical Average 57.8; Actual > Historical Average 42.2; Actual < Baseline Projection 48.9; Actual > Baseline Projection 51.1
  - Primary Balance: Actual < Historical Average 67.6; Actual > Historical Average 32.4; Actual < Baseline Projection 56.2; Actual > Baseline Projection 43.8
  - Non-Debt Flows: Actual < Historical Average 40.3; Actual > Historical Average 59.7; Actual < Baseline Projection 47.5; Actual > Baseline Projection 52.5
- Predictive power comparison (Median Forecast Error; Std. Dev. of Forecast Error):
  - GDP Growth — Historical Average: Median Forecast Error 0.3; Std. Dev. of Forecast Error 5.5. Baseline Projection: Median Forecast Error -0.1; Std. Dev. of Forecast Error 4.5.
  - Exports Growth — Historical Average: Median Forecast Error -2.5; Std. Dev. of Forecast Error 21.6. Baseline Projection: Median Forecast Error 0.3; Std. Dev. of Forecast Error 19.3.
  - Primary Balance — Historical Average: Median Forecast Error 1.1; Std. Dev. of Forecast Error 4.6. Baseline Projection: Median Forecast Error 0.4; Std. Dev. of Forecast Error 4.5.
  - Non-Debt Flows — Historical Average: Median Forecast Error 0.4; Std. Dev. of Forecast Error 7.0. Baseline Projection: Median Forecast Error 0.1; Std. Dev. of Forecast Error 7.9.
- Proposed effective approach:
  - Set post-shock values at one standard deviation below the historical average, or the projected value under the baseline, whichever is lower.
  - Rationale: minimizes bias from historical averages while limiting bias from overly optimistic baseline projections.
  - Result: applying this approach places the post-shock value of primary balance at about the 16th percentile of the distribution of actual outturns, aligning it with other shocks.

### Shock duration assessment
- An event analysis using WEO data (1995–2015) compared DSF shocks (10-year historical average minus one standard deviation) with distribution of outturns to evaluate shock duration.
- Result: a shock duration of about 2 years remains broadly appropriate.

### Macro-linkages in stress tests
- Event studies identified interactions among key macro variables: FX nominal depreciation, real GDP growth, GDP deflator, and export growth.
- Events defined to remove outliers: e.g., growth shock events where actual GDP growth falls between a historical average of minus 1.25 and 0.75 standard deviations.
- Elasticities estimated from events:
  - Inflation decreases with an elasticity to real growth of 0.6.
  - Real GDP growth falls with an elasticity to exports of 0.8.
  - Exchange rate pass-through to the GDP deflator is estimated at 0.3 in the year of the shock.
- Other interactions are sourced from the literature (for example, Aisen and Hauner (2008) on domestic borrowing costs–primary balance interaction).
- Staff focuses on first-round effects rather than introducing all interactions under a given shock to avoid scenarios becoming equivalent to combined shocks.
- Proposed macro-interactions and re-calibrated standardized stress tests persistence (years) and elasticity notes:
  - Ex-B1. Real GDP growth — 1.3
  - Ex-B2. Exports — 1.3
  - Ex-B4. Other flows — 1.8
  - Ex-B6. Depreciation (>30%) — 1.9
  - Pub-B2. Primary balance — 1.8
  - Source: Fund staff calculations.

### Tailored stress tests
- (Section heading present; no further text supplied in the source content.)

*Source: REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES—ANNEXES (selected text).*

### 12. The design of tailored stress tests rests on a variety of specialized data sources. To

### 12. The design of tailored stress tests rests on a variety of specialized data sources.

### a) Natural disasters
- Trigger:
  - Group 1: Small-state LICs identified as vulnerable to natural disasters in IMF (2016). These are 14 countries: Comoros, Dominica, Grenada, Kiribati, Maldives, Micronesia, Samoa, São Tomé and Príncipe, Solomon Islands, St. Lucia, St. Vincent and the Grenadines, Tonga, Tuvalu, and Vanuatu.
  - Group 2: LICs that meet a frequency (around 2 disasters every 3 years) and economic losses (above 5 percent of GDP per year) criteria, based on the EM-DAT database during 1950-2015. These criteria lie between the 75th and 90th percentile of the respective distributions. It identifies 9 countries: Bangladesh, Haiti, Honduras, Nepal, Madagascar, Mozambique, Myanmar, Nicaragua, and Tajikistan.
- Default scenario:
  - Size: A one-off shock to public debt of 10 percent of GDP in the first year of projection. This corresponds to the median change in the public debt to GDP ratio one year after the natural disaster from its pre-shock level, across all episodes with measured economic losses of at least 5 percent of GDP.
  - Interactions: Real GDP growth and exports are lowered by 1.5 and 3.5 percentage points, respectively, in the year of the shock, based on staff event analysis comparing the median growth during the year when the natural disaster took place and the median growth over the preceding 10 years, across the same sample used to identify the size of the shock (Table AI.7).
- Notes:
  - Pandemics were excluded in the analysis for determining the default scenario settings given their rare and infrequent nature. Nevertheless, the tool could be customized to assess their impact where relevant.
  - Example distribution percentiles: the median, 75th and 90th percentiles of the distribution of economic losses across all LICs is equal to 0.3, 5 and 10 percent of GDP, respectively.
- Event analysis sample statistics (Sample of episodes with economic losses > 5 percent of GDP):
  - Real GDP growth (median): Event year 2.2; 10-year average 3.7; Sample size 47.
  - Export growth (median): Event year 7.4; 10-year average 10.8; Sample size 45.

### b) Contingent liability supplement
- Purpose: Supplement the re-calibrated standardized contingent liability shock with a tailored shock when there are unaccounted dimensions of the non-financial public sector in the debt concept reported in the DSA (e.g., other parts of general government, PPP exposures, or unaccounted SOE exposures) or when financial sector risks to the sovereign balance sheet exceed the standardized shock.
- General design:
  - The general government component and financial sector risk supplement would be defined by DSF users based on country-specific circumstances (default shocks would be set to zero).
- PPP shock:
  - Trigger: PPP capital stock above 3 percent of GDP (equivalent to the median across LICs based on the WB’s Private Participation in Infrastructure database over the period 1990–2015). This qualifies 26 countries: Benin, Bhutan, Burundi, Cape Verde, Cambodia, Rep. of Congo, Cote d'Ivoire, Djibouti, Dominica, Ghana, Grenada, Honduras, Lao PDR, Liberia, Mali, Mauritania, Mozambique, Nepal, Nicaragua, São Tomé and Príncipe, Senegal, St. Vincent and the Grenadines, Tajikistan, Togo, Uganda, and Zambia.
  - Default scenario: A permanent increase in external PPG debt equivalent to 35 percent of the country’s PPP capital stock in the second year of projection (proxying for the present value of direct and potential future fiscal costs from PPP distress and/or cancellations). This estimate models a shock to ¼ of the PPP portfolio (assuming that the rest of the portfolio is not highly correlated) plus 10 percent of penalties.
- SOE shock:
  - Trigger: Based on a Fund staff survey conducted in 2016, LICs subject to this shock are those whose responses indicated a partial coverage or unavailable data of SOE external guaranteed debt. The new DSF template would continue monitoring SOE liabilities to inform this trigger going forward. This qualifies 37 countries: Afghanistan, Bangladesh, Benin, Cape Verde, Cambodia, Cameroon, Rep. of Congo, Djibouti, Dominica, Eritrea, Gambia, Haiti, Kyrgyz Republic, Lao PDR, Lesotho, Liberia, Madagascar, Malawi, Maldives, Mali, Mauritania, Moldova, Nepal, Papua New Guinea, Samoa, São Tomé and Príncipe, Senegal, Sierra Leone, Solomon Islands, Somalia, South Sudan, Sudan, Uganda, Uzbekistan, Yemen, and Zimbabwe.
  - Default scenario: One-off permanent increase of 2 percent of GDP in public debt, starting in the second year of projection, based on the median SOE PPG external liabilities.

### c) Commodity Price Shock
- Trigger:
  - LICs whose commodity exports represent at least 80 percent of merchandise exports, based on data from UNCTAD (36 countries): Afghanistan, Benin, Burkina Faso, Cameroon, Central African Republic, Chad, Dem. Rep. of Congo, Rep. of Congo, Cote d'Ivoire, Eritrea, Ethiopia, Ghana, Guinea, Guinea-Bissau, Guyana, Kiribati, Lao PDR, Malawi, Maldives, Mali, Mauritania, Micronesia, Mozambique, Myanmar, Papua New Guinea, Rwanda, Solomon Islands, South Sudan, Sudan, Sierra Leone, Somalia, Tanzania, Timor-Leste, Yemen, Zambia, and Zimbabwe.
- Default scenario:
  - Size and duration: Export-to-GDP ratio is shocked by a commodity price gap in the first year of projection, which closes over 6 years. The price gap is defined as the difference between the baseline commodity price in the first year of projection and the lower end of the 68 percent confidence interval (equivalent to a minus one SD) from RES commodity price forecast distributions for fuel (WTI crude oil price projections, Figure AI.1) and non-fuel (wheat and copper). The price gap for fuel and non-fuel exports is multiplied by their respective shares in total commodity exports.
  - Interactions:
    - Real GDP growth is reduced by 0.5 percentage points, and fiscal revenues-to-GDP are reduced by 0.75 percentage points in each of the first three years of projections for each 10-percentage point contraction of commodity prices. This gap converges to the baseline in 6 years. The size and duration of these responses were informed by analysis of episodes of commodity price busts in a sample of 34 commodity-intensive LICs during 1990-2015.
    - GDP deflator is reduced by the impact of the commodity price gap in the first year of the shock, converging to the baseline in 6 years.
- Notes:
  - Event analysis procedure: (i) construct a country-specific commodity price index as the weighted average of food, raw materials, metals and fuel using export shares; (ii) extract cyclical component using HP filter (longer sample 1960-2030 used); (iii) identify episodes where cyclical component falls below trend for at least two consecutive years and exceeds one country-specific standard deviation over 1990–2015; (iv) study responses of GDP growth and government revenues around bust episodes.
  - Commodity cycles tend to be persistent per staff event analysis and IMF (2015c).

### d) Market-financing shock
- Trigger:
  - LICs that either: (i) are identified as “frontier” according to the prevailing IMF definition; (ii) have outstanding Eurobonds; and (iii) meet the market access criterion for graduation from the PRGT but have not graduated due to short-term vulnerabilities. These criteria identify 17 countries.
- Countries meeting criteria (by indicator columns Frontier LICs / Past Bond issuers / PRGT Market Access Criteria):
  - Bangladesh (✓)
  - Cabo Verde (✓)
  - Cameroon (✓)
  - Rep. of Congo (✓✓)
  - Cote d'Ivoire (✓✓✓)
  - Ethiopia (✓)
  - Ghana (✓✓✓)
  - Honduras (✓)
  - Kenya (✓✓)
  - Maldives (✓)
  - Mozambique (✓✓)
  - Papua New Guinea (✓)
  - Rwanda (✓)
  - Senegal (✓✓)
  - Tanzania (✓✓)
  - Uganda (✓)
  - Zambia (✓✓)
  - Notes: 1/ As defined in the IMF (2014) LIDC report. 2/ These are countries that meet the market access criteria as in IMF (2015), but did not graduate from PRGT given their short-term vulnerabilities.
- Default scenario:
  - Size and duration: Based on a staff event analysis of the median responses around external debt distress episodes during 1995-2015, the scenario consists of a 400 bps increase (sustained for 3 years) in the cost of new external commercial borrowing, FX depreciation equivalent to 15 percent, and shortening of maturities of new commercial external borrowing (to 5-year maturity, with grace periods adjusted according to the type of instrument).

### Annex II — Select Case Studies (key findings relevant for stress-test design)
- Successes: The DSF has been successful in signaling impending debt difficulties in several country cases: Sri Lanka (downgraded to high risk of debt distress in 2008, a year ahead of a request for a Fund program with exceptional access), Chad (rated at high risk of debt distress since 2012, debt restructuring in 2015 and sizable external arrears), and Mongolia (downgraded to high risk in 2015, subsequent debt service difficulties and a 2017 request for official financial support and a debt exchange).
- Areas needing strengthening:
  - Securing stronger baseline debt projections: Optimism in macro projections (growth and fiscal adjustment) impeded early warnings in some cases (e.g., Ghana, St. Lucia, Dominica, Maldives). Public investment scaling-up introduced forecast risk (e.g., Djibouti and Ghana, 2015).
  - Stress tests: Existing standardized stress tests did not capture relevant risks in some countries (e.g., market-financing risks in Sri Lanka and Ghana; lack of tools for natural disasters, commodity price shocks, contingent liabilities prevented frequent analysis in vulnerable countries such as Samoa and Malawi).
  - Determination of risk ratings: In some cases, identified risks did not appropriately inform risk ratings (e.g., Maldives 2015; Ghana’s high public GFN did not inform risk rating until 2014). Steady erosion of debt positions within a moderate risk category may be under-emphasized (e.g., Central African Republic 2010–12, Samoa 2010–12, Ghana 2010–13).
  - Factors not captured by the model: False alarms or crises that did not occur linked to framework limitations in accounting for country-specific factors (e.g., Sao Tome and Principe or Burundi) or special country features (e.g., post-civil conflict cases like Afghanistan).
- Illustrative missed-call examples:
  - Ghana (2012-2014): Crisis triggered by accumulation of large fiscal and external imbalances, policy slippages, external shocks, rising interest cost, weakening net international reserves, exchange rate depreciation and rising public debt. DSAs prior to downgrade emphasized risks from growing GFN, shift in financing mix, foreign investor participation in domestic debt, and Eurobond issuance, but standard stress tests were unable to anticipate the large exchange rate depreciation or primary balance shock. Proposed DSA review responses include realism tool to flag inconsistent macro projections, strengthening size of standardized primary balance shock, and tighter revised debt service thresholds.
  - Malawi (2010-2012): Chronic BOP difficulties, shortages of foreign exchange and critical imports, increased government expenditure while external grants fell and revenue deteriorated, leading to increased public debt and devaluation. The 2010 DSA flagged vulnerability to a shock on tobacco prices and simulated a permanent one-third fall in tobacco prices projecting a breach in the debt-to-exports ratio, but did not anticipate the decline in external grants that, together with the tobacco price drop, led to significant public debt deterioration.

*Source: Review of the Debt Sustainability Framework for Low-Income Countries — Annexes (Fund staff calculations and text).*

### Introduction of a commodities price stress test to

### Introduction of a commodities price stress test to formally model the debt and growth outlook for commodity exporters under adverse prices

### Key reforms and tools introduced
- Introduction of a commodity price stress test to formally model the debt and growth outlook for commodity exporters under adverse prices.
- Realism tool to flag inconsistency in key macro projections, including on ambitious growth projections and fiscal consolidation plans.
- Incorporation of signal from total public debt benchmark in the external risk rating.
- Redesign of standard stress test framework, in particular having the primary balance shock factored in the assessment of the external risk of debt distress.
- Revised debt service thresholds are now tighter to reflect evolving financing landscape.
- A new market risk module proposed to suggest early warning benchmarks on gross financing needs to better capture rollover risks.

### Country case: Mozambique (2009–2014)
- Context:
  - Mozambique was under the PSI when it requested a 12-mo. External Shock Facility to forestall a projected large decline in reserves due to reduced FDI and other capital flows following the global financial crisis.
  - Government significantly scaled up investments financed by nonconcessional loans during this time.
- Developments and risks:
  - By 2015, the economy suffered fiscal slippages and a series of external shocks resulting in a large depreciation.
  - Debt service to export was projected to almost double in three years.
  - A Standby Credit Facility was requested to supplement the PSI.
  - Although the risk rating remained low from 2009 to 2012, the debt projection was continually revised upward due to the heavy public investment schedule.
  - Breaches of the PV of debt/GDP threshold for 8 years in the most extreme scenario were deemed temporary.
  - In 2013, downgrade to moderate risk due to (i) lower discount rate, (ii) increase in NCB to finance public investment, and (iii) deterioration in the current account due to natural gas exploration.
  - In 2015, the risk rating remained moderate, but risks from nonconcessional borrowing were highlighted.
  - During this time, even though the medium term debt service projection was continually revised up, none of the debt service indicators breached their thresholds.
- Lessons:
  - While breaches to the PV debt/GDP ratios appeared in every DSA, the use of judgement for retaining the low risk rating may have unduly delayed the downgrade to moderate risk, thereby providing a false sense of comfort regarding the true extent of debt vulnerabilities.

### Country case: Mongolia (2009–2015)
- Developments over the period:
  - Macro projections were consistently optimistic compared to actual fiscal outturn and mining sector developments (timing of mining projects and export revenues), resulting in significant forecast errors in most debt stock indicators.
  - Baseline assumptions were continually revised leading to risk rating deterioration—to moderate in 2014, high in 2015.
  - By the 2015 DSA, baseline assumptions differed greatly from the previous DSA, reflecting more realistic fiscal and mining revenue assumptions, and the borrowing required to meet the BOP gap.
  - Unanticipated large debt issuances by the Development Bank of Mongolia contributed to the downgrades.
- Hints in the DSA and problems:
  - In 2014, the team prepared a DSA with a weak and strong policy scenarios. Significant risks to external debt sustainability were illustrated by the weak policy scenario, but the more optimistic scenario was taken as the baseline.
  - Alternative scenarios illustrated a steadily rising trend of public debt ratios, but this was not used to inform final risk ratings.
  - Specific problems identified:
    i) Baseline oversight: Projections did not envision delays in mining projects, or fiscal slippages vis-a-vis the newly legislated fiscal framework.
    ii) Unanticipated shocks: Off-budget borrowing accounted for much of the increase in total public debt. Limited debt coverage limits the team's ability to account for these risks.
    iii) Inability to factor in fiscal risks in external risk rating: Lack of primary balance shock analysis in the external DSA limited the link between public DSA signals and external risk ratings.
- Resulting recommended changes (as echoed for other cases):
  - Realism tool to flag inconsistency in key macro projections, including ambitious growth projections and fiscal consolidation plans.
  - Redesign of standard stress test framework, with the primary balance shock factored in the assessment of the external risk of debt distress.
  - Incorporation of signal from total public debt benchmark in the external risk rating.
  - Introduction of a commodity price stress test for commodity exporters.

### Annex III: Update of the Core Debt Distress Model — identification and model enhancements
- Objective:
  - Reforms to the DSF aimed at improving predictive performance, covering: revised methodology for identifying episodes of external debt distress; enhanced specification of the core debt distress model; improving the framework for determining external risk ratings; new approach to estimate total public debt benchmarks.
- Revised methodology for identifying external debt distress episodes:
  - No longer requires observing distress signals for at least three consecutive years to identify debt distress episodes; episodes lasting one or two years included, except when solely driven by arrears (which still require at least three consecutive years).
  - An episode of debt distress is defined as a period in which at least one distress signal is observed; a non-distress episode remains a non-overlapping three-year period with none of the distress signals.
- Expanded set of distress signals:
  - Includes restructurings of public external debt held by private foreign creditors and instances of outright defaults.
  - Stand-alone restructurings of commercial debt help identify crises where market financing reliance rises; outright defaults help predict onset when preceding other distress signals.
- Better treatment of IMF disbursement signals:
  - Proposed criterion focuses on large upfront financing disbursements measured as IMF disbursements during the first six months of a Fund-supported program that are larger than 30 percent of quota, including both GRA and PRGT disbursements.
  - This cutoff is twice as large as the current “access norm” to Fund resources and corresponds to the 75th percentile of disbursements to LICs during the first six months of Fund-supported programs under the GRA and PRGT since 1970.
- Better treatment of debt restructurings:
  - Debt difficulties leading to restructurings are assumed to start in the year preceding the restructuring deal (negotiation phase).
  - Multi-round restructurings under the HIPC Initiative are treated as a single episode.
- Outcomes of the revised identification methodology:
  - Revised methodology identifies a total of 98 debt distress episodes in LICs between 1970 and 2015 (compared to 76 under the existing methodology).
  - Of these 98 episodes: about a third was triggered by IMF signals, a third by the occurrence of external arrears, and a third by debt restructurings, defaults or a combination of signals.
  - Ten episodes of external debt distress are identified since 2008 under the proposed new methodology, while only one would be captured under the existing methodology.
  - Relaxing the duration assumption results in an average duration of debt distress episodes of just over a decade (compared to 18 years under the existing methodology).
- Tabular summaries (reported values preserved):
  - Signal counts and durations:
    - Number of episodes: Old definition 76; New definition 98.
    - Of which triggered by IMF disbursements: Old 7; New 35.
    - Of which triggered by Arrears: Old 51; New 32.
    - Restructurings: Old 15; New 22.
    - Some combination of the above: Old 3; New 8.
    - Median duration in years: Old 18; New 12.
  - IMF-disbursement criterion specifics:
    - Disbursements in the first 6 months > 30 percent of quota.
    - Cutoff corresponds to the 75th percentile of disbursements to LICs during the first six months of Fund-supported programs under the GRA and PRGT since 1970.

### Enhanced specification of the core debt distress probit model
- Rationale:
  - Expanded specification warranted because the existing model predicts debt distress less well under the revised identification methodology.
- Additional explanatory variables:
  - International reserves (scaled by imports), included with a linear and a quadratic term to capture non-linearities.
  - Remittances (scaled by nominal GDP).
  - World growth as a proxy for global shocks.
- Data and estimation:
  - New model estimated in a dataset that pools multiple episodes for each country.
  - Sample covering 80 LICs during 1970–2014.
- Empirical findings (summary statistics and relationships preserved):
  - The likelihood of debt distress is positively correlated with the level of indebtedness, and negatively correlated with the quality of institutions and policies (measured by the CPIA), and with other country-specific factors (country growth, reserves, remittances).
  - Favorable external conditions (world growth) exert an important impact on the probability of debt distress.
  - Inclusion of world growth tends to reduce the statistical significance of domestic growth, reflecting business cycle synchronization.
  - Reserves’ non-linear term has the expected positive sign, implying that above certain coverage, additional accumulation of reserves contributes less to reducing the probability of debt distress (with diminishing marginal contribution up to about 6 months of imports).
- Model performance:
  - The new specification outperforms the existing one in terms of goodness of fit (R-squared, log-likelihood, and BIC), despite a smaller number of observations resulting from removing the MICs sample.
- Selected regression outputs (preserved):
  - Observations in summary table columns: 409; 403; 343; 380.
  - Pseudo R-squared: 0.169; 0.174; 0.190; 0.184.
  - Log-likelihood: -150.1; -145.7; -116.8; -135.2.
  - BIC: 348.4; 339.5; 280.2; 317.9.
  - Coefficient signs and significance (examples from table):
    - Debt burden indicator coefficients: 1.541***; 0.359***; 3.745***; 3.541***.
    - CPIA: -0.400***; -0.381***; -0.362***; -0.395***.
    - World growth: -12.40***; -14.09***; -13.84***; -13.75***.
    - Reserves and Reserves^2 show negative linear and positive quadratic terms with varying significance across specifications.
  - Statistical significance notation preserved: ***, **, * denote significance at 1%, 5%, and 10%, respectively, based on robust standard errors.

*Source: pp082217lic-dsf - Introduction of a commodities price stress test to*

### 11.      The new model survives relevant robustness checks. Staff subjected the new model to

### 11. The new model survives relevant robustness checks

### Robustness checks and main findings
- Subsample 1980-2014: Staff re-estimated the model excluding the 1970s. All coefficients retain their expected sign but CPIA and remittances lose statistical significance in one model each.  
- Adding real GDP per capita: The variable comes with the wrong sign and is not generally significant. The coefficients on CPIA and on the other variables remain statistically significant and are somewhat larger than in the baseline.  
- Adding trade openness:
  - In specifications without remittances, openness was statistically significant.
  - Openness tends to lose statistical significance when jointly considered with remittances, reflecting high correlation.
  - Baseline model with only remittances has a better predictive performance (see footnote indicating better predictive performance refers to a lower in-sample weighted sum of type I and II errors).
  - An alternative openness definition (exports + imports net of FDI, scaled by GDP) weakens remittances’ significance in some models: remittances no longer significant in two models (PV of debt-to-GDP and debt service-to-revenue); openness is not significant in any model.
- Replacing remittances with broader FX income (exports + remittances): Results comparable to baseline; added variable not significant in one model (debt service-to-exports).
- Replacing world growth with a proxy for country risk premium (Moody’s corporate Baa spread over the U.S. 10-year treasury):
  - Estimated coefficient has expected sign.
  - Statistically significant in specifications that include debt service indicators but not in those including debt stock variables.
  - When included together with world growth, the risk premium is not statistically significant, while world growth remains economically and statistically relevant across the four probit regressions.
- Adding an indicator of conflicts (dummy for armed conflicts): Variable has the expected sign in three of four models but is not statistically significant.

### Representative probit regression results (selected coefficients and metrics)
- Debt burden indicator coefficients (examples from regressions reported):
  - 1.912***, 1.763***, 1.436***, 1.541*** (first block)
  - 0.319***, 0.335***, 0.294***, 0.359*** (PV of debt-to-GDP block)
  - 4.877***, 3.174***, 4.739***, 3.745*** (PV of debt-to-exports block)
  - 3.398***, 3.768***, 3.704***, 3.541*** (Debt service-to-revenue block)
- Selected control variable coefficients (examples):
  - CPIA: -0.603***, -0.572***, -0.532***, -0.400***; -0.421***, -0.593***, -0.475***, -0.381***; -0.365***, -0.409***, -0.416***, -0.362***; -0.448***, -0.461***, -0.430***, -0.395*** (across blocks)
  - Domestic growth: -6.136***, -4.414***, -4.547***, -3.081*; -7.447***, -3.886***, -4.256***, -2.853*; -7.386***, -4.500***, -2.724, -3.001; -5.550***, -3.455***, -2.553, -1.942 (across blocks)
  - Reserves: -4.223*** (with Reserves^2 3.953*); -4.591*** (Reserves^2 4.582**); -3.696** (Reserves^2 3.743); -3.699*** (Reserves^2 3.683*) (examples)
  - Remittances: -2.235**, -2.282***, -1.635*, -1.934** (examples)
  - World growth: -12.40***, -14.09***, -13.84***, -13.75*** (examples)
- Sample sizes and fit metrics (examples):
  - Observations: 740, 893, 455, 409 (first block)
  - Pseudo R-squared: 0.188, 0.156, 0.108, 0.169 (first block)
  - Log-likelihood: -238, -276.4, -183.1, -150.1 (first block)
  - BIC: 509, 586.8, 390.7, 348.4 (first block)
- Significance notation: ***, **, * denote significance at 1%, 5%, and 10%, respectively, based on robust standard errors.

### Improving the framework for deriving external risk ratings — approach and algorithm
- Motivation: Debt thresholds should be derived consistently with the DSF’s aggregation rule (any single breach of a debt threshold under baseline signals high risk; any single breach under stress signals moderate risk). Previous 2012 approach derived thresholds individually and could introduce conservative bias.
- Proposed algorithm (five steps, with key definitions and parameters preserved):
  Step 1: Estimate a probit model for each debt burden indicator:
    - P(Debt Distress) = Φ(αj + γj dj + Σk βj,k Xk)  
    - dj: one of the four debt burden indicators (PV of debt to GDP, PV of debt to exports, debt service to revenues, debt service to exports).  
    - Xk: non-debt explanatory variables (CPIA, country growth, reserves, squared reserves, remittances, world growth).
  Step 2: Construct a composite indicator (CI):
    - CI = Σk β̅k X̅k (where β̅k is the average slope coefficient across the four probit regressions and X̅k is a 10-year average prior to the episode).
    - Benchmarks for strong/medium/weak are set at the 75th/50th/25th percentiles of the CI distribution over 2005-14.
  Step 3: Invert each probit given a cutoff probability pj* to obtain debt thresholds:
    - d̅j(pj*) = (Φ−1(pj*) − αj − CI) / γj
    - Evaluate at CI percentiles to obtain thresholds for strong/medium/weak categories.
  Step 4: Define DSF-consistent prediction rule:
    - Signal high risk (ŷ = 1) if any dj > d̅j(pj*) for ANY debt indicator j; signal low risk only if ALL dj < thresholds.
  Step 5: Select cutoff probabilities pj* to minimize a prediction loss function penalizing Type I and Type II errors:
    - Loss Function = ω Type I error + (1 − ω) Type II error
    - Weight on Type I error set at ω = 0.67.

### Implications of re-estimated thresholds and predictive performance
- Re-estimated thresholds:
  - The new framework would maintain or increase debt stock thresholds for all countries.
  - Countries retaining weak or medium classification under the new classification would see lower debt service thresholds (the present framework had set debt service thresholds at typically non-binding levels).
- Predictive performance:
  - The proposed changes improve predictive performance relative to the existing framework.
  - The rate of false alarms is significantly reduced by 10 percentage points while DSF’s capacity to anticipate debt distress is improved.
  - A model specification unchanged but re-estimated on the updated sample (1970–2014) with a new distress-definition methodology would imply significantly tighter thresholds, much higher false alarm rate (~70 percent) and very low missed crises (~6 percent).
- Other context:
  - Existing approach (Box AIII.1) used individual threshold derivation: probit estimation on 1970-2007 LIC+MIC sample; optimal cutoff probabilities chosen by minimizing weighted sum of Type I and II errors across various weights (type I weight ranged from 0.5 to 0.75); thresholds inverted with fixed CPIA cutoffs (weak = 3.25, medium = 3.50, strong = 3.75) and LIC average growth.

*Source: pp082217lic-dsf - 11. The new model survives relevant robustness checks. Staff subjected the new model to — REVIEW OF THE DEBT SUSTAINABILITY FRAMEWORK FOR LOW INCOME COUNTRIES—ANNEXES, INTERNATIONAL MONETARY FUND*

### 16.      Countries would continue to be classified as weak, medium or strong based on an

### pp082217lic-dsf - 16.      Countries would continue to be classified as weak, medium or strong based on an 

### Expanded country classification based on a country-specific CI
- Countries are classified as weak, medium, or strong based on an expanded measure of capacity to repay—their country-specific CI—moving away from relying exclusively on the CPIA.
- The CI is defined as the weighted sum of the non-debt determinants of debt distress (CPIA, country growth, reserves, squared reserves, remittances, world growth), where the weights are given by the average estimated coefficients across the probit models.

### CI calculation: data and methodology
- The CI would be calculated using the latest five years of historical data and the first 5 years of projections; CPIA forecasts use its most recent value due to slow-moving nature.
- CI formula (notation preserved from source):
  - 퐶퐼푖 = ∑푘=1^6 훽̅푘 X̅푖,푘
    - where k denotes the control variables,
    - 훽̅푘 denotes the average coefficient for the k control across the four probit models,
    - X̅푖,푘 denotes the 10-year average of each control, covering the latest 5 historical and first 5 projections years.

### Country classification steps and cutoffs
- Steps:
  - Calculate the CI for each country i using the weighted sum described above.
  - Compare the country CI with CI cutoffs, estimated as the 25th (퐶퐼25) and 75th (퐶퐼75) percentiles of its distribution in the last ten years of the regression sample (2005-2014).
  - Classification rule:
    - Weak if 퐶퐼푖 < 퐶퐼25
    - Medium if 퐶퐼25 ≤ 퐶퐼푖 ≤ 퐶퐼75
    - Strong if 퐶퐼푖 > 퐶퐼75
- CI distribution illustration (values preserved as in source):
  - CI distribution:
    - 25th percentile = 2.6975th percentile = 3.05
    - 50th percentile = 2.86
  - Contribution to the CI across LICs (Min / Max per source):
    - World growth 0.47 / 0.47
    - CPIA 0.76 / 1.54
    - Reserves 1/ 0.03 / 1.03
    - Remittances 0.00 / 0.31
    - Country growth -0.01 / 0.24
  - Note: 1/ Including reserves' linear and non-linear terms.

### New approach for estimating total public debt benchmarks
- Rationale:
  - Data constraints on domestic arrears led staff to use signals beyond outright defaults, including de facto domestic defaults proxied by inflation spikes and/or persistent negative real interest rates indicative of financial repression.
- Proxies and definitions:
  - Staff selected 18 different definitions of domestic debt distress:
    - 16 definitions based on negative real interest rates (RIR): four unconditional, 10 control for the size of domestic debt, two control for the size of domestic-debt-to-revenue ratio.
    - Two definitions based on episodes of inflation coinciding with large government financing by the central bank but not with significant supply or demand shocks.
  - Table AIII.7 (summary from source) — Total episodes and triggers (preserved labels):
    - Total episodes: 116 81 12 89 18 0 5 3 8 86 69 76 71 25 89 11 18 36 64 34 53 5 (as presented in source table format)
    - Of which triggered by:
      - Outright default: 18 18 19 20 19 19 19 20 21 19 20 19 21 22 23 (as presented)
      - Financial repression: 9 7 6 2 10 9 7 1 6 1 3 4 6 9 4 6 9 2 6 3 4 6 2 2 3 1 2 (as presented)
      - Both: 11 0 0 0 0 0 0 0 0 10 0 0 0 0 0 1 0 0 0 (as presented)
    - (Table values preserved as presented in source; formatting reflects source aggregation.)
- Noise-to-signal (NTS) approach for PV of total public debt:
  - Rather than estimating a separate probit for total public debt, staff used an NTS approach to ensure consistency with the external block and avoid conflicting composite indicators.
  - Five-step NTS strategy:
    - Define an indicator of overall public debt distress by combining external and domestic debt distress episodes.
    - Assign countries the same classification as in the external block.
    - Impose an expanded aggregation rule using joint information from five debt burden indicators (the four from the external block plus PV of total public debt-to-GDP).
    - Define Type I and Type II errors as in the external block.
    - Jointly choose the three debt benchmarks for total public debt for the weak/medium/strong categories to minimize the loss function.
  - Aggregation rule operational detail:
    - Framework signals high risk of debt distress (no high risk) if any (none) of the five debt burden indicators breaches (does not breach) its corresponding threshold/benchmark.

### Benchmarks for the PV of total public debt (exact values preserved)
- Table AIII.8: Existing and Proposed Benchmarks for the PV of Total Public Debt
  - 2017 review 1/:
    - weak 35
    - medium 55
    - strong 70
    - Type I error 0.24
    - Type II error 0.44
    - Loss function 0.31
    - Note: 1/ Based on NTS approach, using 0.67 weight on type I error. Benchmarks are rounded up to the nearest 5 percent.
  - 2012 review 2/:
    - weak 38
    - medium 56
    - strong 74
    - Type I error 0.19
    - Type II error 0.52
    - Loss function 0.31
    - Note: 2/ Based on the probit model and weight on type I error ranging from 0.5 to 0.75.
- Finding: The new benchmarks are slightly lower across classification categories and imply a lower rate of false alarms compared to the existing framework at the time of the 2012 review.

### Data sources, sample, and coverage (Box AIII.2 — preserved details)
- Data sources used to construct debt burden indicators and distress episodes include:
  - World Bank’s Debt Reporting System (external debt stock), World Development Indicators (WDI), International Debt Statistics (IDS), IMF’s WEO, Government Finance Statistics (GFS), Fund staff compilations, Abbas and Christensen (2009), Abbas and others (2010), Panizza (2008), Reinhart and Rogoff (2010), IFS, Fund’s Finance Department, Global Development Finance (GDF), Bank of Canada Database on Sovereign Defaults, Das and others (2011), Cruces and Trebesch (2013), Standard and Poor’s, Catão and Milesi-Ferretti (2014), EM-DAT, UCDP/PRIO, World Bank’s Migration and Remittances Data, Penn World Tables, Maddison Statistics.
- Sample:
  - Final sample comprises 80 LICs and covers 45 years (1970–2014).
  - LICs are defined as IDA-only countries as in the 2012 review; countries that graduate (reverse graduate) from IDA status are treated as LICs if they spent at least half of the sample years under that status; otherwise excluded.
- Data construction notes:
  - PV of external debt is calculated by discounting debt service using time-varying discount rates (CIRRs).
  - External debt service data on a paid basis adjusted by accumulation of arrears to estimate due-basis debt service.
  - For remittances, missing observations approximated using nominal GDP growth of major source country; bilateral remittances data from World Bank used to identify major source country.
  - To maximize CPIA observations, missing CPIA data in early 1970s taken from staff’s 2012 estimations fitted using a regression with three covariates.

### Reserve measurement for currency union members (Box AIII.3 — preserved findings)
- Fifteen LICs are members of three currency unions:
  - ECCU: Dominica, Grenada, and St. Vincent and the Grenadines;
  - CEMAC: Cameroon, Central African Republic, Chad, and Republic of Congo;
  - WAEMU: Benin, Burkina Faso, Cote d’Ivoire, Guinea-Bissau, Mali, Niger, Senegal, and Togo.
- Conceptual considerations:
  - Union-wide reserve pooling or country-level reserves may be appropriate depending on union and country circumstances (financial architecture, liquidity allocation, synchronization of shocks, potential free-riding).
- Empirical approach and robustness:
  - For probit estimation, staff used members’ imputed reserves due to data constraints; alternative proxy of regional reserve coverage had secondary impact on probit model and no meaningful effect on composite indicator weights.
  - For practical classification, union-wide reserve coverage is generally appropriate, but exceptions exist where members effectively lose access to the pool; in those cases a switch to imputed reserves for classification may be warranted and detailed in the Staff Guidance Note.
- Alternative model specification table (selected preserved coefficients and statistics):
  - Debt burden indicator regressions include coefficients (significance preserved) such as:
    - GDP/Exports/Revenue/Exports: 1.614***, 0.361***, 3.717***, 3.591***
    - CPIA: -0.396***, -0.380***, -0.355***, -0.388***
    - Domestic growth: -3.120*, -2.926*, -3.051, -1.989
    - Reserves_CU: -4.242***, -4.458***, -3.983**, -3.876***
    - Reserves_CU^2: 4.069**, 4.520**, 4.183*, 4.019*
    - Remittances: -2.288**, -2.355***, -1.648*, -1.943**
    - World growth: -12.42***, -14.07***, -13.48***, -13.76***
  - Observations: 410, 404, 343, 381 (by specification)
  - Pseudo R-squared: 0.166, 0.166, 0.191, 0.184
  - Log-likelihood: -151.0, -147.2, -116.6, -135.4
  - BIC: 350.1, 342.4, 279.9, 318.4
  - Significance legend: ***, **, * denote significance at 1%, 5%, and 10%, respectively, based on robust standard errors.

*Source: Fund staff calculations as presented in the chapter excerpt.*

### Annex IV. Realism Tools for Assessing Baseline Growth

### Annex IV. Realism Tools for Assessing Baseline Growth Projections

### A. Public Investment Scaling-up

- Purpose and scope
  - Tool goal: provide a simple Excel-based diagnostic to assess realism of baseline growth projections in the context of public investment scaling-up.
  - Not a substitute for more sophisticated models (examples cited: IMF’s Debt-Investment-Growth model; World Bank’s Long-Term Growth model).

- Core idea
  - Takes projected paths of public investment and growth as given.
  - Decomposes projected growth rates into:
    - contribution of the increase in the government capital stock due to public investment, and
    - contribution from other sources (denoted εt).
  - Enables explicit comparison of implicit country-team assumptions on public investment contribution with historical data or reference points.

- Production function and growth accounting
  - Aggregate production function assumed isoelastic in government capital:
    - Yt = Gt^β F(At, Kt, Ht)
    - where Yt is real GDP; Gt is government capital; β is the output elasticity of government capital; F(·) is a function of productivity At, private capital Kt, and human capital Ht.
  - Growth decomposition:
    - (Yt − Yt−1)/Yt−1 = β (Gt − Gt−1)/Gt−1 + εt
    - First term: contribution of changes in government capital to growth.
    - Second term εt: contribution of all other factors and productivity.
  - Benchmark parameter:
    - β = 0.15 (based on cited literature).

- Government capital accumulation
  - Accumulation equation:
    - Gt+1 = (1 − δ) Gt + φF iGt
    - Parameters:
      - δ (depreciation rate) = 0.05
      - φF (public investment efficiency) — can be calibrated using FAD estimates
      - iGt is the public investment path
  - From projected Gt path, compute growth rates of the government capital stock and the implied growth contribution β ΔG/G.

- Historical vs future investment efficiency
  - Efficiency defined as fraction of a dollar of government investment that turns into government capital.
  - Incorporation:
    - replace projected future investment with φF iGt
    - scale down historical capital stock series by factor φH
    - Efficiency parameters satisfy 0 < φF ≤ 1 and 0 < φH ≤ 1
  - Analysis depends only on the ratio φF / φH.
  - Benchmark assumption: φF = φH = 1; departures (especially implying efficiency improvements) require careful justification.

- Data requirements
  - Projected paths of public investment and growth.
  - Estimate of stock of government capital (FAD database available for 170 countries).
  - Estimate of output elasticity of government capital (β).
  - Estimate/calibration of public investment efficiency (FAD estimates; IMF, 2015a referenced).

- Limitations and extensions
  - Decomposition captures only direct effects of government capital changes on growth; does not capture endogenous responses of productivity or private factors.
  - Country teams may use more sophisticated models cited in the text to capture richer channels.

### B. Fiscal Adjustment Impact on Growth

- Purpose and rationale
  - Tool to illuminate impact of planned fiscal adjustment on growth projections under a range of plausible fiscal multipliers, allowing comparison with baseline projected growth.
  - Motivation: overly optimistic growth projections can undermine adjustment plans; negative growth surprises identified as main factor derailing fiscal consolidation.

- Core mechanics
  - The tool recovers the underlying projected growth path absent fiscal adjustment:
    - Underlying projected growth = Projected growth (including fiscal adjustment) + change in growth due to fiscal adjustment
  - Change in growth due to fiscal adjustment depends on assumed fiscal multiplier size and persistence.

- Fiscal multiplier modeling
  - Impact multiplier m defined as percent change of real GDP from a one percentage point adjustment in the structural primary balance (as percent of GDP) during the first year of full impact.
  - Timing assumptions:
    - First year of full impact is set as the year after implementation.
    - Impact during the implementation year itself is assumed to be half of the impact multiplier.
  - Persistence captured by AR(1) process:
    - mt = m* pt−1
    - pt is the autocorrelation coefficient; default value pt = 0.6.

- Outputs and use
  - Once underlying projected growth is uncovered, the tool can calculate framework-consistent projected growth paths assuming different values for the fiscal multiplier.
  - Enables comparison of baseline projected growth with alternative paths under plausible multipliers to assess realism and risks to fiscal consolidation.

*Source: Annex IV. Realism Tools for Assessing Baseline Growth Projections, extracted from the provided IMF content unit.*

### 6. The proposed methodology would increase the information value of the DSF without

### 6. The proposed methodology would increase the information value of the DSF without 

### Impact on Fund and Bank debt policies
- The proposed methodology would increase the information value of the DSF without creating additional data needs and operational implications to Fund/Bank debt policies.
- The Fund’s Debt Limits Policy and the World Bank’s Non-Concessional Borrowing Policy will be unaffected by the characterization of “space to absorb shocks” for countries in the moderate risk category.
- The rules under the two policies relevant for countries at moderate risk of debt distress would continue to apply, regardless of whether countries are characterized as having limited or substantial space.
- This tool is a counterfactual exercise, and does not constitute a view on the likelihood of shocks occurring that could lead to a rating downgrade.

### Threshold characterization and adjustments
- Threshold formula representation shown:
  - (1-X)*Threshold
  - (1-Y)*Threshold
  - Moderate with limited space
  - Moderate
  - Moderate with substantial space
- Note: For the PV debt/GDP and PV debt/exports thresholds, X is 20 percent and Y is 40 percent.
- Note: For debt service/exports and debt service/revenue thresholds, X is 12 percent and Y is 35 percent.

### Key implications for DSF use
- The methodology differentiates within the moderate risk category by characterizing countries as having either limited or substantial space to absorb shocks, enhancing the DSF’s informational content.
- No change in operational application of existing Fund and World Bank policy rules for countries at moderate risk of debt distress.
- The distinction is explicitly a scenario/counterfactual tool rather than a probabilistic forecast of shock occurrence or rating changes.

*Source: Review of the Debt Sustainability Framework for Low Income Countries — Annexes*

---


_Source: https://www.imf.org/-/media/files/publications/pp/2017/pp082217lic-dsf.pdf_