What Is the Probability of Default Benchmark?

The probability of default benchmark is the yardstick SEBI makes every credit rating agency publish and then be measured against. A rating is a statement about default risk, and until there is a common default-rate scale, one agency's AA and another's AA are not comparable claims. The framework sits at para 26 of the Master Circular for Credit Rating Agencies dated 11 July 2025, in the chapter on reporting and disclosures.

What is the probability of default benchmark for?

Para 26.1 states the objective plainly: to enable investors to discern the performance of a CRA against a standardised PD benchmark scale. The benchmark is not the agency's own observed default record. It is the reference level, and the agency's observed default study is the number that gets compared to it.

That separation is the useful part. A default study on its own tells you what happened in one agency's book. Set against a common benchmark, it tells you whether that agency's ratings behaved the way that rating category is supposed to behave.

The AAA, AA and A levels

Para 26.2.6 fixes the benchmark for the three highest categories, subject to unexpected legal events or mitigating circumstances affecting default rates, with permitted tolerance levels. It runs one sub-paragraph per category: 26.2.6.1 for AAA, 26.2.6.2 for AA, 26.2.6.3 for A.

Para 26.2.6 sets no benchmark for the blank cells. The pattern in what it does set is that the number of horizons covered shrinks as the category gets lower: three for AAA, two for AA, one for A.

The pattern is that tolerance widens, and attaches at a shorter horizon, as the category gets lower: 1 per cent at three years for AAA, 2 per cent at two years for AA, 3 per cent at one year for A. Para 26.2.9 notes the benchmarks and tolerance levels may be re-indexed from time to time, so the numbers above are a dated reading rather than a permanent fixture.

How the numbers are built

Para 26.2 sets out the method, and the choices in it are worth understanding because they change what the benchmark measures.

The base approach. A marginal default rate approach, using a monthly static pool, over the last 10-year period.

Short-run versus long-run. The two are the same data treated with different confidence:

• Short-run benchmarks may account for spikes caused by economic cycles or unforeseen events, and so may carry a wider band. They are computed on a 99.7 per cent confidence interval over the weighted average of one, two and three-year default rates for the last 10 years.
• Long-run benchmarks iron out economic cycles because they cover a longer tenure, and so may be narrower. They are computed on a 95 per cent confidence interval over the same weighted average default rates.

Both are adjusted, wherever required, to achieve ordinality. Ordinality is the property that a lower rating category should not show a lower default rate than a higher one; where the raw data breaches that, the benchmark is adjusted rather than published inverted.

Withdrawals and non-cooperation. Para 26.2.4 requires adjustment for rating withdrawals, and for securities the rating stays in the computation until the cohort completes or the instrument matures, whichever is earlier. Para 26.2.5 keeps ratings of non-cooperative issuers in the cohort, under the category in which the instrument is currently rated. Neither a withdrawal nor an issuer-not-cooperating tag lets a rating quietly exit the denominator.

The counting rules that stop double-counting

Paras 26.2.7 and 26.2.8 handle the case where one issuer has many rated instruments, which would otherwise let a single default register several times.

• Non-structured instruments. Instruments of an issuer with equal seniority and the same rating are not included separately. Instruments with different seniority levels are included as separate instances, subject to a cap of three instances across all rating categories put together.
• Structured instruments. Tranches issued by a trust with the same seniority and the same rating are not counted separately. Tranches of differing seniority are counted separately, with a cap of three tranches per rating category per issuer, stated as a guard against under-estimating default rates where there are many tranches of differing seniority but the same rating.

The instance caps cut both ways, and para 26.2.8 says which way SEBI was worried about: understating, not overstating.

Where the benchmark is published

Para 26.2.10 requires the standardised PD benchmarks to be disclosed on each CRA's website for long-term and short-term instrument ratings, on a consolidated basis across all financial instruments the CRA has rated. Para 26.2.11 adds that CRAs may review their rating methodologies to align them with the benchmarks.

So the benchmark is a per-agency website disclosure rather than a single number SEBI publishes centrally. Comparing agencies means reading each one's disclosure.

Reading the rating record alongside it

The probability of default benchmark is a calibration tool, not a view on any single issuer. The issuer-level facts live in the agency's own press releases and disclosures:

• How to read a rating rationale is where a specific rating action is explained.
• Rating outlook versus rating watch distinguishes the two forward-looking qualifiers that sit on a rating.
• A no default statement is the monthly issuer confirmation that feeds the default record this benchmark is built from.
• The post default curing period governs how long a defaulted issuer stays marked as such.
• An unsolicited credit rating is the category that carries its own separate disclosure treatment.

Flock reports the filings themselves, each stamped with its date and linked back to the exchange or regulator that published it. What any of it means for you is your call to make.