Methodology note
Split ratings: median, lowest, highest or provider priority
A split rating is when agencies rate the same bond differently, and a credit limit needs one grade. The policy must say how to choose: the median, the lowest, the highest, or the first provider in a priority order. With ratings of A, Baa3 and BB+, the median is BBB- and the lowest is BB+.
The four policies
- Median: the middle of three ratings, and the lower of two. This is the shipped default.
- Lowest: the most conservative reading.
- Highest: the most permissive reading.
- Provider priority: the first agency in a stated order.
The organisation names its rating providers and the field each writes to. With none named, the rule's own rating field is the single source.
One scale for every agency
Grades are compared on the long-term S&P and Fitch ladder from AAA to D. Moody's symbols are aliased onto it, so Baa3 sits with BBB-.
Anything else is unparseable and disclosed, never guessed. Every result states the policy it was judged under, and whether the organisation stated it or a provisional default applied.
Pitfall: the median of two
With two ratings there is no middle, so "median" means the lower one. A book of bonds rated by two agencies is, under a median policy, judged on the lower rating.
Some policies state outright that the lower rating applies. Then choose lowest, and say so in the rule, so the intent survives a third agency being added.
In the example
Bond X is rated A, Baa3 and BB+. Under the median policy it resolves to BBB-, meets a BBB- floor, and the book holds nothing below the minimum.
Switch the organisation's policy to lowest and Bond X resolves to BB+. Its 15% then breaches a 10% cap on holdings below the floor.
Worked example
Bond X, rated A / Baa3 / BB+, resolves to BBB- under the median policy and meets a BBB- floor; under a lowest policy it would count below the floor.
Rule template:credit_quality minimumGrade: BBB-
Bounds: hard max 10%
| Holding | Market value | RATING_SP | RATING_MOODYS | RATING_FITCH |
|---|---|---|---|---|
| Bond X | 150 | A | Baa3 | BB+ |
| Gilt holdings | 850 | AA | Aa3 | AA- |
| Rating providers | Split ratings | Watch / outlook notching |
|---|---|---|
| SP (RATING_SP), MOODYS (RATING_MOODYS), FITCH (RATING_FITCH) | median | off |
| Group | Status | Value | Bounds | Headroom to the hard bound | Code |
|---|---|---|---|---|---|
| Credit quality | Compliant | 0% | ≤ 10% | 10% | WITHIN_LIMITS |
WITHIN_LIMITS: The measured value sits inside every approved bound.
Computed at build time by the Acclimetry rule engine (backend commit 8197063); the same code runs in the product.
Questions
Can one rule use a different split-rating policy?
In the registry the split-rating policy is set for the organisation, so it applies to every credit rule. The unrated treatment is the setting a rule can state for itself.
What happens to a rating the engine cannot read?
It is unparseable and disclosed on the result, never guessed. Fix the value at its source rather than mapping it by hand.