Sovereign Credit Rating Explorer

A firm that assigns credit ratings to sovereign and corporate borrowers. This dataset uses three sovereign-focused agencies:

Standard & Poor's (S&P) — largest by coverage; uses AAA → D scale.
Moody's — uses Aaa → C scale (different notation, equivalent meaning).
DBRS Morningstar — smaller; skewed toward higher-income economies; assigns systematically higher scores (mean ≈ 41.1 vs ≈ 31 for S&P and Moody's).

Inter-agency correlation is very high (r ≈ 0.97 across all pairs), but agencies update at different speeds, causing short-term divergence after credit events.

The simple arithmetic mean of all available agency scores for a given country-year. If only one agency rates a country in that year, the composite equals that agency's score. If all three agencies rate it, the composite is their mean.

The composite is the primary outcome variable in this project. It abstracts over agency-specific biases but does not correct for them; use agency fixed effects in regressions when agency composition matters.

A rating agency's assessment of a sovereign government's ability and willingness to meet its debt obligations in full and on time. Ratings are expressed as letter grades (e.g. AAA, BB+) and are converted to a numeric 0–60 scale in this dataset for cross-agency comparability.

Ratings are event-driven — they are updated only when an agency judges a change is warranted, not on a fixed schedule. Between events, the outstanding rating remains in effect.

The change in a variable from year t−1 to year t. ΔM1 is the most commonly displayed delta in this project — it shows the year-on-year change in the M1 relative rating, decomposing whether improvement or deterioration was driven by a change in the country's absolute score, a change in the group benchmark, or both.

A reduction in a country's composite score from one year to the next. In this dataset, downgrades are larger in magnitude on average than upgrades (mean −6.1 points vs +4.4 points). The ordered model classifies a country-year as a downgrade if the composite score fell relative to the prior year.

The method used to convert sparse rating events into a balanced annual panel. Once a country receives a rating from an agency, that rating is carried forward to all subsequent years until a new rating event replaces it. This reflects the fact that an outstanding rating remains valid until explicitly revised.

Without forward-filling, the panel would have many missing values in years without rating events. The forward-filled value is the last observed rating, not an interpolation.

Net inflows of foreign direct investment as a percentage of GDP, sourced from the World Bank Development Indicators. Used as a macro covariate in panel regressions. Coverage ≈ 92% of panel rows.

Within-country FDI is not significantly associated with credit score changes in fixed-effects models (FDI coefficient ≈ 0, p > 0.1 in most specs), but is positively correlated with scores in pooled OLS (β ≈ +0.01***), reflecting the cross-sectional fact that wealthier, higher-rated countries attract more FDI.

GDP per capita in current USD (World Bank), log-transformed (base 10) for use in regressions. The log transformation reduces skewness and makes coefficients interpretable as semi-elasticities. In pooled OLS, log GDPpc has a coefficient of ≈ 9.2 (a 10× increase in GDP per capita is associated with a 9.2-point higher composite score).

A measure of inequality in composite scores within a peer group in a given year. A Gini of 0 means all members have identical scores; a Gini approaching 1 means one country holds all the score mass. Computed annually for all 59 peer groups.

The World Gini rose from 0.232 (2000) to 0.321 (2023) — a 38% increase — indicating growing divergence in global credit quality, driven by stable high-income sovereigns and deteriorating emerging/frontier markets.

A set of countries used as a benchmark for relative rating calculations. This dataset includes 59 groups, ranging from the 4-member GCC to the 150-member World group. Each country can belong to multiple overlapping groups (e.g. France is in EU, Eurozone, G7, G20, NATO, Schengen, Western Europe, and World simultaneously).

Only three partitions are mutually exclusive and collectively exhaustive (MECE): (1) World Bank income tiers, (2) broad continents, (3) African sub-regions within Africa. See Methodology §3 for the full list.

A rating at or above BBB− (S&P / DBRS) or Baa3 (Moody's), corresponding to a score of approximately 42 or above on the 0–60 scale. Below this threshold, a sovereign is classified as speculative grade (commonly "junk"). Investment-grade status affects the universe of institutional investors who can hold the debt, market access, and borrowing costs.

The average composite score of all peer group members excluding the focal country itself, in a given year. It is the benchmark used in all relative rating calculations (M1–M4).

Why exclude the focal country? If a country's own score is included in its benchmark, a country with a very high (or low) score will pull the mean toward itself, compressing its relative deviation toward zero. This creates a mechanical bias that understates how unusual the country's rating is, especially in small groups. LOO elimination removes this self-influence.

Practical effect: In small groups (e.g. GCC with 4 members), removing one country changes the mean substantially. In large groups (World, 150 members), the LOO mean is nearly identical to the simple group mean.

The phenomenon where a country's relative rating (M1/M3/M4) changes even when its absolute score is unchanged — because the peer group mean is moving. A country holding a constant absolute score of 33 while the world mean falls from 37 to 31 is improving in relative terms without any agency action. This is only visible through relative rating methods, not absolute scores. India is a notable example: its absolute score rose modestly (+8 pts, 2000–2025) while M1 nearly doubled, largely due to world mean deterioration.

The primary relative rating measure. Defined as a country's composite score squared, divided by the current-year leave-one-out group mean:

The squaring amplifies differences at the upper end of the distribution (where most rated countries cluster) and ensures M1 is always non-negative. M1 is unbounded above — a top-rated country in a low-rated group can have M1 well above 100.

M1 captures both genuine rating changes and latent relative drift (passive improvement/deterioration as the group mean moves). See Methodology for full details.

The ordinal rank of a country within its peer group in a given year, expressed as a raw integer (e.g. 47 out of 150). A higher M2 means a higher score relative to peers. M2 is non-parametric and does not require assumptions about the score distribution, but is not comparable across groups of different sizes (use M4 for that).

Identical to M1 but uses the prior year's LOO mean as the denominator:

M3 is most informative at crisis onset: because the prior-year benchmark has not yet absorbed the group's deterioration, M3 is systematically more optimistic than M1 immediately after a shock. During recovery, M3 lags M1 as the old (poor) benchmark persists. M3 is undefined in the first year of a country-group series.

The within-group percentile rank, normalised to the unit interval [0, 1] by dividing M2 by the group size. M4 = 0.5 means the country is exactly at the median; M4 close to 1 means near the top.

M4 is comparable across groups of different sizes, unlike M2. It saturates near (but never exactly at) 1 for the top-rated country because the group size N includes all members.

A partition of all 151 countries into non-overlapping groups that together cover every country exactly once. Only three groupings in this dataset are MECE over all countries: (1) the four World Bank income tiers, (2) the six broad continents, (3) the five African sub-regions within Africa. All other groupings (G20, NATO, BRICS, etc.) intentionally overlap. Do not include multiple overlapping group dummies in the same regression specification — they will be collinear.

The empirical tendency for countries with high (or low) credit scores to experience subsequent changes in the opposite direction. In the ordered logit results, the lagged score coefficient is −0.019*** — each additional point of current score reduces the probability of an upgrade and increases the probability of a downgrade in the next period, consistent with mean reversion. This reflects rating agencies' tendency to anchor on a long-run equilibrium.

A model for an ordinal outcome — here, the annual rating transition: downgrade (−1), stable (0), or upgrade (+1). The ordered logit and probit produce nearly identical results (AIC ≈ 4560 vs 4562), as expected for symmetric distributions. Estimated on N = 2,922 country-year transitions, 2000–2025.

A qualitative signal attached to a rating indicating the likely direction of a future rating change: Positive, Stable, Negative, or Watch (imminent change). Outlooks are recorded in the raw data but are not used in computing composite scores or relative ratings in this project, as they are not fully comparable across agencies.

The main dataset is a balanced panel with one observation per country per year (2000–2025), after forward-filling. The panel has 3,323 country-year rows across 144 countries with at least some macro covariate coverage. Panel regressions use two-way fixed effects (country + year) with standard errors clustered at the country level.

See Latent Relative Drift.

A country's position within its peer group expressed as a fraction of the group. The Explorer displays two versions: M2 (raw integer rank) and M4 (normalised to 0–1). The 50th percentile (M4 = 0.5) means the country is at the group median.

A measure of a country's credit standing within a peer group, as opposed to its absolute score on the 0–60 scale. Four relative rating methods are implemented (M1–M4). Relative ratings can change even when absolute scores are unchanged — for instance, when the peer group mean deteriorates, a country with a fixed score improves in relative terms. See Methodology for the full definitions.

The numeric representation of a letter credit rating on a uniform 0–60 scale, where 60 = AAA/Aaa (best) and 0 = D/SD/C (default). The scale is constructed so that each full letter-grade notch corresponds to approximately 3 points (e.g. AA+ ≈ 57, AA ≈ 54, AA− ≈ 51). The 0–60 scale is applied uniformly across S&P, Moody's, and DBRS after mapping each agency's notation to the equivalent grade.

A failure by a government to meet its debt obligations (missed payment, restructuring, or distressed exchange). Reflected in the rating scale as D or SD (Selective Default) at the bottom of the 0–60 range (score ≈ 0–6). Rating agencies typically downgrade to this level on or around the event date.

A year-on-year change in composite score. Classified as: Upgrade (score rose), Stable (no change), or Downgrade (score fell). In the dataset: ~17% upgrades, ~70% stable, ~13% downgrades. Stability is the dominant outcome because credit scores are highly persistent — agencies change ratings infrequently.

An increase in a country's composite score from one year to the next. Upgrades average +4.4 points, smaller in magnitude than downgrades (−6.1 points), consistent with the stylised fact that agencies are more cautious about upgrading than downgrading (asymmetric conservatism). There have been 802 upgrades vs 604 downgrades in the full 1949–2026 history.

The share of variance in the dependent variable that is explained by the regressors after removing country and year means (i.e. after partialing out fixed effects). In Model A (growth + FDI only), R²(within) ≈ 0 — macro flows explain almost none of the year-to-year variation in a country's score, once its average level is controlled for. By contrast, Pooled OLS R² = 0.90, reflecting that almost all variation is between countries (structural differences in wealth).

The broadest peer group, consisting of all 150 rated countries. Used as the global benchmark for the "World LOO mean" — the average composite score across all countries, excluding the focal country. Relative ratings against the World group measure a country's standing in the global distribution rather than within a regional or political subgroup.