## wpiea2019162-print-pdf

## Source details

**Canonical URL:** [wpiea2019162-print-pdf](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019162-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019162-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019162-print-pdf.pdf.json)

---

### II. Literature review — determinants of sovereign credit ratings
- Prior determinants identified: income per capita, GDP growth, inflation, external debt, level of economic development, and default history.
- Short-run determinants highlighted: changes in GDP per capita, GDP growth, government debt, and the fiscal balance.
- Long-run determinants highlighted: government effectiveness, external debt, level of foreign reserves, and default history.
- Noted heterogeneity: importance of economic variables varies across rating categories and country groups.
- This study’s contribution:
  - Focus on impact of the public debt-to-GDP ratio on sovereign credit ratings using a wide set of techniques, specifications, and country groupings.
  - Investigation of the nonlinear nature of the debt–rating relationship and differences across advanced economies (AEs) and emerging and developing economies (EMDEs).

### III. Dataset and empirical strategy
- Dataset description:
  - Public debt: general government gross and net debt (percent of GDP) from the IMF’s World Economic Outlook (WEO) database.
  - Sovereign credit ratings: Fitch Ratings, last rating assigned during the year.
  - Alternative rating proxy: Institutional Investor Index (0 worst to 100 best) from Institutional Investor, Inc.
  - Controls and other series:
    - GDP and inflation from WEO.
    - PPP GDP per capita from World Bank’s World Development Indicators.
    - 10-year U.S. interest rates and VIX from Bloomberg.
    - Sovereign bond spreads from JP Morgan’s EMBIG.
    - Export diversification indicator from Ding and Hadzi-Vaskov (2017).
  - Sample coverage:
    - Annual data 1998–2014 for 106 countries.
    - Country composition: 31 AEs and 75 EMDEs (11 Emerging Europe, 8 CIS, 9 Emerging Asia, 17 LAC, 9 MENA, 21 SSA).
- Descriptive patterns:
  - Negative relationship between gross debt and sovereign credit ratings across AEs, EMDEs, and EMDE regions.
  - Slope for AEs flatter (example: Japan: very high debt yet top ratings); within EMDEs slope negative across regions, somewhat flatter for SSA.
  - Positive relationship between GDP per capita (PPP) and sovereign credit ratings.
- Dominance analysis approach:
  - Based on the Panel OLS-Fixed Effect specification.
  - Result summary: public debt dominates other explanatory variables (GDP growth, inflation, VIX, interest rates, GDP per capita) in explaining credit ratings.
- Rating categorization:
  - Fitch’s 23 rating categories mapped to integer values 1 (worst) to 23 (best).
  - Aggregated groups:
    - Non-Investment Grade (NIG): grades DD to BB+ (categories 1 to 13) — mostly EMDEs.
    - Low-Investment Grade (LIG): lower half of investment grade — mix of AEs and EMDEs.
    - High-Investment Grade (HIG): A+ to AAA (categories 19 to 23) — mostly AEs.
  - Minimum of 30 country-year observations imposed in rolling regressions with five consecutive credit rating grades.
- Empirical methods:
  - Panel ordered probit: latent variable y*_{it} = β D_{it} + γ X_{it} + u_i + ε_{it}, where y* is rating category (1–23), D is gross or net debt-to-GDP, X control set, and u_i country fixed effects.
  - Panel OLS with country fixed effects: same specification treating rating grade as numeric dependent variable (1–23).

### IV. Empirical results — main findings
- A. The negative relationship between debt and credit ratings
  - Higher public debt is associated with worse sovereign credit ratings.
  - Quantification: an increase in the debt ratio by 10 percent of GDP is associated with almost half a notch lower credit rating (panel OLS-FE result consistent with ordered probit).
  - Controls behave as expected:
    - Higher GDP per capita and higher real GDP growth → better ratings.
    - Higher inflation → worse ratings.
    - VIX and US interest rates included to control for global factors.
  - GDP per capita is highly significant and does not fully account for AE vs EMDE differences — nonlinear debt–rating relation matters.
- B. Uncovering non-linearity across credit rating grades
  - Rolling ordered probit (5-grade windows): coefficient pattern shows an asymmetric U-shape across the rating spectrum.
    - Debt impact weakest for very low grades, somewhat higher for the very best grades, and highest for middle grades.
  - Interpretation:
    - Very weak-rated countries: limited space for downgrades; other factors may dominate.
    - Best-rated countries: fiscal discipline matters but strong institutions may dampen debt’s effect relative to middle graders.
    - Middle graders (close-to-investment / LIG): most sensitive to debt increases.
  - Marginal probabilities (within 5-grade windows):
    - In middle range (broadly LIG): marginal probability ≈ -0.005; a debt increase by 10 percent of GDP → about a 5 percent higher probability of being placed into a worse category (and 5 percent lower probability of being placed into a better category) within the 5-grade window.
    - For lower ratings (NIG): effect smaller and close to zero for the lowest ratings.
    - For higher ratings (HIG): effect around a 3 percent change in probability for a 10 percent change in debt.
  - Rolling panel OLS-FE regressions confirm asymmetric U-shape:
    - In the middle range (LIG): a debt increase by 10 percent of GDP → decline in rating of almost ½ of a notch (about 10–15 percent of one standard deviation).
    - For HIG: about ¼ of a notch for a 10 percent of GDP debt increase.
    - For NIG: about 1/6 of a notch (smaller impact), eventually close to zero at the lowest ratings.
  - Institutional Investor Index results reproduce the asymmetric U-shape:
    - A debt increase of 10 percent of GDP → decline in the Institutional Investor Index by up to 2 ½ units (about 10–15 percent of one standard deviation) for middle-range ratings.
    - Effect halved for best rating grades and smaller for worst ratings.
- C. The non-linearity explains differences across AEs and EMDEs
  - Pooled regressions (without separating by rating-category) show more negative estimated debt coefficients for AEs than EMDEs.
  - Explanation: nonlinear debt impact across rating grades combined with uneven distribution of AEs and EMDEs across rating categories.
  - Within the same rating category, the effect of debt on ratings is similar for AEs and EMDEs (example: within LIG, estimated coefficient for AEs (-0.09) similar to EMDEs (-0.1)).
  - Selected numeric estimates (panel regressions):
    - Full sample gross debt coefficient: -0.0393***.
    - AEs gross debt coefficient: -0.0533***.
    - EMDEs gross debt coefficient: -0.0235***.
  - Regional evidence: EMDE regions with higher average ratings (Emerging Asia, Emerging Europe, MENA) tend to have steeper debt-rating slopes than regions with lower average ratings (CIS, SSA), consistent with nonlinearity.

### V. Robustness checks
- Replacing gross debt with net debt:
  - Main results hold: negative relationship persists.
  - Key patterns preserved: effect larger for AEs than EMDEs in pooled samples but similar within rating categories; strongest effect for LIG, followed by HIG, then NIG.
  - Institutional Investor Index consistent: 10 percent of GDP net debt increase → up to 2 ½ units decline for middle-range ratings.
- Controlling for export diversification:
  - Inclusion leaves main results virtually unchanged.
  - Export diversification associated with better credit ratings in several specifications.
- Lagged dependent variable:
  - Including lagged dependent variable does not overturn negative impact of public debt; ordered probit specifications retain negative coefficients, most strongly significant.
- Alternative estimator:
  - Results robust to using ordered logit instead of ordered probit.

### Key quantitative findings (selected)
- Sample and coverage:
  - Period: 1998–2014.
  - Countries: 106 (31 AEs, 75 EMDEs).
  - Rating scale used: 23 Fitch categories mapped to values 1–23.
- Debt–rating impact magnitudes for a 10 percent of GDP increase in debt ratio:
  - Middle-range (LIG): almost ½ of a notch lower credit rating.
  - High-rated (HIG): about 1/4 of a notch.
  - Lowest-rated (NIG): about 1/6 of a notch.
  - Marginal probability effect in LIG: marginal probability ≈ -0.005 → ≈ 5 percent higher probability of being placed in a worse grade within 5-grade windows.
  - Institutional Investor Index: decline up to 2 ½ index units for middle-range ratings (≈ 10–15 percent of one standard deviation).
- Representative coefficient estimates (panel OLS-FE, Table 4 Panel A):
  - Full sample gross debt coefficient: -0.0393***.
  - AEs gross debt coefficient: -0.0533***.
  - EMDEs gross debt coefficient: -0.0235***.

### Concluding remarks — main empirical conclusions and interpretation
- Main empirical conclusions:
  - Higher public debt lowers the probability of being placed in a better credit rating category for both gross and net debt and for both AEs and EMDEs.
  - The negative relation is nonlinear across rating grades:
    - Strongest effect in middle range (LIG).
    - Smallest effect at lower end (NIG).
    - Intermediate effect at upper end (HIG).
  - Nonlinear relation combined with uneven AE/EMDE distribution across grades explains apparent AE–EMDE differences.
  - Within the middle rating grades, the negative effect of debt on ratings is very similar for AEs and EMDEs.
- Quantified effects reiterated:
  - Ordered probit/logit rolling regressions: for LIG, 10 percent of GDP debt increase → about 5 percent higher (lower) probability of being placed into a worse (better) category within a 5-grade window.
  - OLS regressions: 10 percent of GDP debt increase → decline in rating of almost ½ of a notch for middle range; about ¼ of a notch for highest grades; close to zero for lowest ratings.
- Robustness and alternative specifications:
  - Results robust to alternative dependent variables, gross vs net debt, alternative ratings groupings, and ordered probit vs ordered logit.
  - Alternative groupings (e.g., placing all A grades in LIG; splitting NIG into HNIG and LNIG) preserve the central finding that LIG shows the highest sensitivity to debt.
- Interpretation and suggested channels (speculative, for future research):
  - Middle-range countries treat current debt as a key indicator of public-finance health, hence stronger rating effects.
  - Best-rated countries benefit from strong institutions; debt increases less likely to materially affect ratings.
  - Weakest grades suffer from multiple deficiencies where other factors may matter more than debt.
- Illustrative implications (based on ratings as of May 28 2019):
  - EMDEs in the lower half of investment grade (LIG) would experience similar effects as AEs in same rating group.
  - NIG countries would experience a smaller effect of debt on ratings than LIG countries.

### Annex I. Dominance Analysis — key statistics and matrix
- General dominance statistics (Panel A):
  - Gross debt: 0.20 0.53 1
  - GDP growth: 0.04 0.10 3
  - Inflation: 0.01 0.02 5
  - VI X: 0.00 0.00 6
  - US interest rate: 0.03 0.07 4
  - GDP per capita: 0.10 0.27 2
- Complete dominance designations (Panel B) — matrix interpretation:
  - Gross debt row: 0 1 1 1 1 1 (gross debt designated to dominate GDP growth, Inflation, VI X, US interest rate, and GDP per capita).
  - GDP per capita row: -1 1 1 1 1 0 (GDP per capita designated to dominate GDP growth, Inflation, VI X, and US interest rate; dominated by Gross debt).
  - VI X and US interest rate rows show multiple -1 entries indicating they are dominated by other variables in the baseline specification.
- Notes:
  - Calculations based on baseline panel regression specification for the full sample with country-specific fixed effects.
  - Values of 1 designate dominance by the row-marked variable; values of -1 designate dominance by the column-marked variable.

### Annex II. Debt–Ratings Relationship Across Regions — presentation notes
- Regions covered: Advanced Economies, Emerging Europe, CIS, Emerging Asia, LAC, MENA, SSA.
- Presentation:
  - Table and Figure present average ratings per region and coefficient estimates from region-restricted regressions analogous to column 1 of Table 4.
  - Figure axis/scale annotations:
    - Vertical axis range shown: -0.07, -0.06, -0.05, -0.04, -0.03, -0.02, -0.01, 0.00
    - Horizontal axis tick labels shown: 10 12 14 16 18 20
    - Horizontal axis label: Average rating
    - Vertical axis label: Coefficient estimate

*Source: IMF staff analysis (wpiea2019162-print-pdf: Sections II–V, Annex I, Annex II).*

### Section III describes the dataset and sets the empirical strategy. Section IV presents the main

### wpiea2019162-print-pdf - Section III describes the dataset and sets the empirical strategy. Section IV presents the main

### II. Literature review — determinants of sovereign credit ratings
- Prior literature identifies key determinants: income per capita, GDP growth, inflation, external debt, level of economic development, and default history (Cantor and Packer (1996); Afonso (2003)).
- Short-run determinants: changes in GDP per capita, GDP growth, government debt, and the fiscal balance (Afonso, Gomes, and Rother (2011)).
- Long-run determinants: government effectiveness, external debt, level of foreign reserves, and default history (Afonso, Gomes, and Rother (2011)).
- Heterogeneity: the importance of economic variables varies across rating categories and country groups (Bissoondoyal-Bheenick (2005); Boumparis et al. (2015)).
- This study’s contribution:
  - Focuses on the impact of the public debt-to-GDP ratio on sovereign credit ratings using a wide set of techniques, specifications, and country groupings.
  - Investigates the nonlinear nature of the debt–rating relationship and differences across advanced economies (AEs) and emerging and developing economies (EMDEs).

### III. Dataset and empirical strategy
- Dataset description
  - Public debt: general government gross and net debt (percent of GDP) from the IMF’s World Economic Outlook (WEO) database.
  - Sovereign credit ratings: Fitch Ratings, last rating assigned during the year.
  - Alternative rating proxy: Institutional Investor Index (0 worst to 100 best) from Institutional Investor, Inc.
  - Controls and other series:
    - GDP and inflation from WEO.
    - PPP GDP per capita from World Bank’s World Development Indicators.
    - 10-year U.S. interest rates and VIX from Bloomberg.
    - Sovereign bond spreads from JP Morgan’s EMBIG.
    - Export diversification indicator from Ding and Hadzi-Vaskov (2017).
  - Sample coverage:
    - Annual data 1998–2014 for 106 countries.
    - Country composition: 31 AEs and 75 EMDEs, of which 11 Emerging Europe, 8 CIS, 9 Emerging Asia, 17 LAC, 9 MENA, and 21 SSA.
- Descriptive patterns
  - Negative relationship between gross debt and sovereign credit ratings across AEs, EMDEs, and EMDE regions.
  - Slope for AEs flatter (e.g., Japan: very high debt yet top ratings); within EMDEs slope negative across regions, somewhat flatter for SSA.
  - Positive relationship between GDP per capita (PPP) and sovereign credit ratings.
- Dominance analysis
  - Based on the Panel OLS-Fixed Effect specification (section III.E, Table 2 column 2).
  - Result: public debt dominates other explanatory variables (GDP growth, inflation, VIX, interest rates, GDP per capita) in explaining credit ratings (see Annex I).
- Rating categorization
  - Fitch’s 23 rating categories mapped to integer values 1 (worst) to 23 (best).
  - Three aggregated groups:
    - Non-Investment Grade (NIG): grades DD to BB+ (categories 1 to 13) — mostly EMDEs.
    - Low-Investment Grade (LIG): lower half of investment grade — mix of AEs and EMDEs.
    - High-Investment Grade (HIG): A+ to AAA (categories 19 to 23) — mostly AEs.
  - Minimum of 30 country-year observations imposed in rolling regressions with five consecutive credit rating grades.
- Empirical methods
  - Panel ordered probit:
    - Latent variable specification: y*_{it} = β D_{it} + γ X_{it} + u_i + ε_{it}, where y* is the country’s credit rating category (1–23), D is gross or net debt-to-GDP, X control set, and u_i country fixed effects.
  - Panel OLS with country fixed effects:
    - Same specification as the ordered probit but treating rating grade as a numeric dependent variable (1–23).

### IV. Empirical results — main findings
A. The negative relationship between debt and credit ratings
- Ordered probit and fixed-effects panel results (Table 2):
  - Higher public debt is associated with worse sovereign credit ratings.
  - Quantification: an increase in the debt ratio by 10 percent of GDP is associated with almost half a notch lower credit rating (panel OLS-FE result consistent with ordered probit).
- Controls behave as expected:
  - Higher GDP per capita and higher real GDP growth → better ratings.
  - Higher inflation → worse ratings.
  - VIX and US interest rates included to control for global factors.
- GDP per capita is highly significant and does not fully account for AE vs EMDE differences — nonlinear debt–rating relation matters.

B. Uncovering non-linearity across credit rating grades
- Rolling ordered probit (5-grade windows) results (Figure 4):
  - Coefficient pattern: asymmetric U-shape across rating spectrum.
  - Debt impact weakest for very low grades, somewhat higher for the very best grades, and highest for middle grades.
- Interpretation:
  - Very weak-rated countries: limited space for downgrades and other factors (institutions, credibility) may dominate.
  - Best-rated countries: fiscal discipline matters, but strong institutions may dampen debt’s effect relative to middle graders.
  - Middle graders (close-to-investment / LIG): most sensitive to debt increases.
- Marginal probabilities (Figure 5):
  - Within 5-grade windows, probability shifts from debt increases concentrated more strongly for middle-range grades.
  - Quantification:
    - In the middle range (broadly LIG), marginal probability is about -0.005: a debt increase by 10 percent of GDP is associated with about a 5 percent higher probability of being placed into a worse category (and 5 percent lower probability of being placed into a better category) within the 5-grade window.
    - For lower ratings (NIG) the effect is smaller and eventually close to zero for the lowest ratings.
    - For higher ratings (HIG) the effect is around a 3 percent change in probability for a 10 percent change in debt.
- Rolling panel OLS-FE regressions (Figure 6):
  - Confirm the asymmetric U-shape: debt penalizes middle graders most.
  - Quantification (panel OLS-FE):
    - In the middle range (LIG), a debt increase by 10 percent of GDP is associated with a decline in rating of almost ½ of a notch (about 10–15 percent of one standard deviation).
    - For HIG: about ¼ of a notch for a 10 percent of GDP debt increase.
    - For NIG: about 1/6 of a notch (smaller impact), and eventually close to zero at the lowest ratings.
- Institutional Investor Index (Figures 7–9):
  - Close correlation between Institutional Investor Index (0–100) and Fitch ratings.
  - Rolling fixed-effects panel regressions with the Institutional Investor Index as dependent variable reproduce the asymmetric U-shape.
  - Quantification:
    - A debt increase of 10 percent of GDP is associated with a decline in the Institutional Investor Index by up to 2 ½ units (about 10–15 percent of one standard deviation) for middle-range ratings.
    - Effect halved for best rating grades and smaller for worst ratings.

C. The non-linearity explains differences across AEs and EMDEs
- Ordered probit and panel regressions (Tables 3–5) — key pattern:
  - On average, estimated debt coefficients are more negative for AEs than for EMDEs in pooled regressions that do not separate by rating-category.
  - This appears counterintuitive but is explained by:
    - Nonlinearity of debt’s impact across rating grades (largest negative effect in LIG).
    - Uneven distribution of AEs and EMDEs across rating categories (AEs concentrated in LIG and HIG; EMDEs concentrated in LIG and NIG).
  - Within the same rating category, the effect of debt on ratings is similar for AEs and EMDEs.
    - Example (textual): within LIG, estimated coefficient for AEs (-0.09) is very similar to EMDEs (-0.1) (ordered probit result discussed in text).
- Selected numeric estimates (Panel regressions, Table 4 Panel A):
  - Full sample gross debt coefficient: -0.0393***.
  - AEs gross debt coefficient: -0.0533***.
  - EMDEs gross debt coefficient: -0.0235***.
  - Note: p-values shown in parentheses in tables; significance levels indicated as *** p<0.01, ** p<0.05, * p<0.1.
- Regional evidence:
  - EMDE regions with higher average ratings (Emerging Asia, Emerging Europe, MENA) tend to have steeper debt-rating slopes than regions with lower average ratings (CIS, SSA), consistent with nonlinearity.

### V. Robustness checks
- Replacing gross debt with net debt (Tables 6–9):
  - Main results hold: negative relationship between debt and credit ratings persists.
  - Key patterns preserved:
    - Effect of debt larger for AEs than EMDEs in pooled samples but similar within rating categories.
    - Strongest effect remains for LIG, followed by HIG, then NIG.
  - Institutional Investor Index results consistent: 10 percent of GDP net debt increase associated with up to 2 ½ units decline in the index for middle-range ratings.
- Controlling for export diversification:
  - Inclusion of a diversification indicator leaves main results virtually unchanged.
  - Export diversification is associated with better credit ratings in several specifications.
- Lagged dependent variable:
  - Including a lagged dependent variable to account for persistence of ratings does not overturn the negative impact of public debt on credit ratings; ordered probit specifications retain negative coefficients, most strongly significant.
- Alternative estimator:
  - Results robust to using ordered logit instead of ordered probit.

### Key quantitative findings (selected)
- Sample and coverage:
  - Period: 1998–2014.
  - Countries: 106 (31 AEs, 75 EMDEs).
  - Rating scale used: 23 Fitch categories mapped to values 1–23.
- Debt–rating impact magnitudes:
  - A 10 percent of GDP increase in debt ratio is associated with:
    - Almost ½ of a notch lower credit rating on average for middle-range (LIG) countries.
    - About 1/4 of a notch for high-rated (HIG) countries.
    - About 1/6 of a notch for lowest-rated (NIG) countries.
  - Marginal probability effect in LIG: marginal probability ≈ -0.005; 10 percent of GDP debt increase → ≈ 5 percent higher probability of being placed in a worse grade within 5-grade windows.
  - Institutional Investor Index: 10 percent of GDP debt increase → decline up to 2 ½ index units for middle-range ratings (≈ 10–15 percent of one standard deviation).
- Representative coefficient estimates (panel OLS-FE, Table 4 Panel A):
  - Full sample gross debt coefficient: -0.0393***.
  - AEs gross debt coefficient: -0.0533***.
  - EMDEs gross debt coefficient: -0.0235***.

### Conclusions (empirical implications)
- Public debt is a central determinant of sovereign credit ratings and dominates several standard macro-financial controls in explaining ratings variation.
- The relationship between public debt and sovereign creditworthiness is nonlinear across the rating distribution:
  - Middle-rated countries (LIG) are most sensitive to debt increases.
  - Very low-rated countries show limited additional downgrade sensitivity to debt increases.
  - High-rated countries are sensitive but less so than middle-rated countries.
- Apparent differences in debt sensitivity between AEs and EMDEs are driven by:
  - The nonlinear debt–rating relationship across rating grades; and
  - Uneven distribution of AEs and EMDEs across rating categories.
- Main results are robust to alternative debt measures (net debt), additional controls (export diversification), accounting for rating persistence (lagged dependent variables), and alternative estimators (ordered logit).

*Source: IMF staff analysis (wpiea2019162-print-pdf: Sections II–V).*

### conclusion for the panel regressions with gross and net debt, respectively.

### Concluding remarks (panel regressions with gross and net debt)

### Main empirical conclusions
- Higher public debt lowers the probability of being placed in a better credit rating category. This result holds when public debt is measured in gross and in net terms and applies to both AEs and EMDEs.
- The negative relation between debt and ratings is nonlinear and depends on the rating grade:
  - The effect is strongest in the middle range of rating grades (broadly encompassing the low investment grades, LIG).
  - The effect is smallest at the lower end of the ratings range (non-investment grades, NIG).
  - The effect is of intermediate size for the upper end of the ratings range (high investment grades, HIG).
- The nonlinear relation—combined with the uneven distribution of AEs and EMDEs across rating grades—explains the apparent difference in the effect of debt on ratings for AEs and EMDEs even when controlling for income level and macroeconomic variables. In the middle range of rating grades (lower range of investment grade), the negative effect of debt on ratings is very similar for AEs and EMDEs.

### Quantified effects (from ordered probit/logit and OLS)
- Ordered probit and logit rolling regressions:
  - For countries with middle ratings (LIG group), a debt increase by 10 percent of GDP is associated with about 5 percent higher (lower) probability of being placed into a worse (better) category within a window of 5 adjacent grades.
  - The effect is smaller for lower ratings within the NIG group (and eventually close to zero for the lowest ratings).
  - For the best ratings in the HIG group, the effect is about 3 percent change in probability.
- OLS regressions:
  - A debt increase of 10 percent of GDP is associated with a decline in rating of almost ½ of a notch for the middle rating range.
  - The decline is smaller for lower ratings (close to zero for the lowest ratings).
  - For the highest rating grades, the decline is about ¼ of a notch.

### Robustness and alternative specifications
- Results are robust to:
  - Alternative dependent variables.
  - Gross versus net debt definitions.
  - Alternative ratings groupings (e.g., moving A+ to keep all A grades together; splitting NIG into HNIG and LNIG).
  - Choice of ordered probit versus ordered logit: ordered logit rolling regressions produce a highly nonlinear pattern consistent with probit results.
- Alternative ratings grouping findings:
  - When all A grades are placed in LIG, the middle group (LIG) still has the highest effect, while HIG and NIG tend to experience lower effects.
  - Splitting NIG into HNIG (all “B-BBB” grades) and LNIG (grades below “B-”) suggests the effect for NIG is weaker at the lower end (LNIG) and stronger at the higher end (HNIG); limited LNIG sample size restricts some probit estimation and renders panel results indicative.

### Interpretation and suggested channels
- Possible explanation for nonlinearity (left for future research; described as speculative):
  - Countries in the middle range (lower half of investment grade) treat current debt as an important indicator of public-finance health—hence debt increases have stronger rating effects.
  - Countries with the best credit ratings enjoy the strongest institutional frameworks and are perceived as most creditworthy; debt increases are less likely to materially affect ratings because markets expect offsets in the future.
  - The weakest grades suffer from multiple deficiencies (lack of policy credibility, weak institutions), which may matter more for ratings than debt levels.
- Noted exceptions have plausible explanations:
  - Some AEs have very high credit ratings despite high debt (considered safe havens).
  - Some developing economies have both low debt and low ratings owing to limited market access.

### Illustrative country-group implications (based on ratings as of May 28 2019)
- EMDEs in the lower half of the investment grade range (LIG)—examples listed in the source—would experience similar effects as AEs in the same rating group (e.g., Italy, Portugal, Spain).
- Countries in the NIG group (examples listed in the source) would experience a smaller effect of debt on ratings than LIG countries.

*Source: conclusion and robustness sections (ordered probit/logit and panel regressions) from the provided IMF content unit.*

### Annex I. Dominance Analysis

### Annex I. Dominance Analysis

### General dominance statistics
- Table: Panel A. General Dominance Statistics (values as presented)
  - Gross debt: 0.20 0.53 1
  - GDP growth: 0.04 0.10 3
  - Inflation: 0.01 0.02 5
  - VI X: 0.00 0.00 6
  - US interest rate: 0.03 0.07 4
  - GDP per capita: 0.10 0.27 2

- Notes:
  - Calculations based on baseline panel regression specification for the full sample with country-specific fixed effects.

### Complete dominance designations (Panel B)
- Dominance matrix (row variable dominates column variable when value = 1; column variable dominates row variable when value = -1)
  - Variables (rows, columns): Gross debt, GDP growth, Inflation, VI X, US interest rate, GDP per capita
  - Matrix entries (rows × columns):
    - Gross debt: 0 1 1 1 1 1
    - GDP growth: -1 0 1 1 0 -1
    - Inflation: -1 -1 0 1 0 -1
    - VI X: -1 -1 -1 0 0 -1
    - US interest rate: -1 0 0 0 0 -1
    - GDP per capita: -1 1 1 1 1 0

- Interpretive points from the matrix:
  - Gross debt is designated to dominate GDP growth, Inflation, VI X, US interest rate, and GDP per capita (row entries = 1).
  - GDP per capita is designated to dominate GDP growth, Inflation, VI X, and US interest rate (row entries = 1) but is dominated by Gross debt (column entry = -1).
  - VI X is designated as dominated by Gross debt, GDP growth, Inflation, and GDP per capita (row entries = -1).
  - US interest rate is designated as dominated by Gross debt and GDP per capita (row entries = -1 for those columns; zeros with some others).

- Notes:
  - Values of 1 designate dominance by the row-marked variable, and values of -1 designate dominance by the column-marked variable.
  - Calculations based on baseline panel regression specification for the full sample with country-specific fixed effects.

### Annex II. Debt–Ratings Relationship Across Regions

### Regional coverage and presentation
- Regions shown:
  - Advanced Economies
  - Emerging Europe
  - CIS
  - Emerging Asia
  - LAC
  - MENA
  - SSA

- Figure/Table presentation:
  - The Table presents average ratings per region based on the observations included in the baseline specification, as well as coefficient estimates based on separate regressions as the one in column 1 of Table 4, but restricting sample to each region.
  - The Figure presents average ratings per region based on the observations included in the baseline specification, as well as coefficient estimates based on separate regressions as the one in column 1 of Table 4, but restricting sample to each region.

- Axis/scale annotations visible in the Figure:
  - Vertical axis range shown: -0.07, -0.06, -0.05, -0.04, -0.03, -0.02, -0.01, 0.00
  - Horizontal axis tick labels shown: 10 12 14 16 18 20
  - Horizontal axis label: Average rating
  - Vertical axis label: Coefficient estimate

*Source: Annex I and Annex II, wpiea2019162-print-pdf - Annex I. Dominance Analysis*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019162-print-pdf.pdf_