Guide · Debt markets · Updated 2026-09-23

A credit rating is an opinion about default, and the migration record is what turns the letter into a number

Most people read a rating as a verdict: AAA is safe, BBB is acceptable, anything lower is junk. The regulator's own description is narrower. SEBI's frequently asked questions on rating agencies, issued on 2 March 2026, call a rating the agency's opinion on the likelihood of a debt obligation being repaid in full and on time, and therefore an assessment of the probability of default. An opinion about a probability is only as good as its track record, and every registered agency has to publish that track record in a standard format. This guide reads the letter through it.

An opinion on a scale the regulator wrote down

The symbols are not the agencies' own. SEBI standardised them in a circular of 15 June 2011, and the definitions now sit in Annexure 2 of its Master Circular for Credit Rating Agencies of 11 July 2025 (SEBI/HO/DDHS/DDHS-POD2/P/CIR/2025/101). Each agency prefixes its name to the same symbol, and every definition is about one thing: the timely servicing of financial obligations. The long-term scale covers securities with an original maturity above one year; a parallel short-term scale runs from A1 to A4 for maturities up to a year.

Read the definitions down the column and the vocabulary changes at one line. From AAA to BBB they grade safety. From BB down they grade risk of default. That line, BBB- and above, is what the market calls investment grade, and rules key off it: agencies must report every rating that crosses it in either direction each half-year (master circular, para 27.3.1.3), and a debt fund can segregate an issuer's paper when it is downgraded below it, taking the most conservative rating where agencies differ (SEBI's Master Circular for Mutual Funds of 20 March 2026, para 5.5.2), the mechanism explained in the guide to side-pocketing.

D is not a forecast. The master circular defines default as non-payment of interest or principal in full on the pre-agreed date, and requires it to be recognised at the first instance of delay (para 27.4.1.5). A coupon paid one day late is a default even if it is paid in full the following week. The scale therefore measures timeliness, not what a lender eventually recovers, which is why a separate expected-loss scale, EL 1 to EL 7, exists for infrastructure instruments (para 5.6).

A notch ranks inside a category and has no default rate of its own

Plus and minus are modifiers, available from AA to C, and SEBI defines them only relatively: they reflect the comparative standing within the category. Counting AAA and D, which take no modifier, the long-term scale has twenty grades. SEBI's FAQ adds two corrections worth keeping: the minus sign carries no negative connotation, and AA- is stronger than A+, because the category outranks the modifier.

The default-rate and transition formats SEBI prescribes (Annexures 27 and 28) are laid out by category, AAA through C, so no mandated disclosure gives AA- a default rate of its own, and RBI's risk-weight table for banks ignores the modifier too. The notch counts where regulators measure movement: a downgrade of three notches or more between two consecutive rating actions is a sharp rating action (para 27.3.3.2), reported for investment grade to the stock exchanges and depositories each half-year. A to BBB is three notches and counts; A to BBB+ is two and does not.

The next compression is still a proposal. A SEBI consultation paper of 13 August 2026 would make issuers and online bond platforms show a six-level, colour-coded Credit Risk-o-Meter beside each debt security, following the lowest rating where there are several. Comments closed on 3 September 2026 and, as at 23 September, no final circular has been issued.

The issuer pays, and the rules are built around that fact

SEBI's FAQ says it directly: in India the debt-issuing entity pays for the credit rating, a practice termed the issuer-pays model. The conflict follows from the structure rather than from anyone's intent. The party whose debt is judged chooses the agency, pays the fee, may approach more than one agency and decides whether to accept the result, and the agency's revenue depends on being chosen. The investor has no contract with the agency at all (FAQ, question 12), so a reader who relies on a rating has no claim on it.

None of these rules removes the conflict. They make its effects visible, and two matter most to a reader. The non-accepted ratings list exposes shopping. The published default record exposes drift, because an agency that hands out letters too generously shows it later in the default column of its own matrix. An issuer can choose its agency. It cannot choose that agency's history.

Non-payment leaves a mark in the data itself. Failing to pay the fee for surveillance is one of the master circular's own examples of non-cooperation (para 11.1). The rating then carries the suffix ISSUER NOT COOPERATING, shortened to INC below, and must be downgraded to non-investment grade once non-cooperation has lasted more than six months (para 11.9.1). Whether a study keeps those issuers in its pools changes some of its numbers threefold, as the section on pools shows.

Outlook and watch say which way the letter may move, not how likely default is

Both are standardised in the master circular's chapter on press releases. An outlook is the agency's view of the expected direction of the rating in the near to medium term, and every rating gets one of three: Stable, Positive or Negative (para 10.1.3). The exceptions are short-term ratings, the C and D categories, ratings already on watch, securitisation pools and mutual fund credit-quality ratings under monthly surveillance (para 10.1.4). A rating watch is the view of the expected direction in the short term, phrased as Rating Watch with Positive, Developing or Negative Implications (para 10.1.5). Developing, meaning the direction itself is open, exists only as a watch.

SEBI fixes the words, not a clock: each agency publishes its own policy for placing ratings on watch (para 8.2.3), and that policy governs how long a watch may run. What SEBI does fix is the surrounding disclosure. Every review press release carries the three-year rating history of the issuer's securities (para 10.1.6), a section on rating sensitivities stating, in numbers where possible, what would trigger an upgrade or a downgrade (para 10.2), and a section on liquidity (para 10.1.9). Where the rating assumes support from a parent, group or government, the analytical-approach section is to name the supporter and the rationale (para 10.1.8.1), the point on which the guide to the BHARAT Bond index turns.

The limitation for a reader is structural. The mandated matrices are split by category only, so a BBB on Negative outlook sits in the same row as a BBB on Stable. If outlooks carry information, the average row understates how likely the first is to move down and overstates it for the second, and the sensitivities section is where that adjustment gets its facts.

What every agency must publish, and when

The disclosure rules are the reason a rating can be read at all. Chapter III of the master circular sets them out in three layers: a standing benchmark, half-yearly lists and annual statistics.

Three definitions do most of the work. Default is recognised at the first missed day. A withdrawn rating stays in the default cohorts until the cohort ends or the instrument matures, and debenture trustees must keep reporting delays to the agency for the instrument's life even after withdrawal (para 27.4.1.2(b)), which closes the route of dropping a rating just before trouble. And averages are weighted by the number of ratings in each static pool, so a crowded year counts for more than a thin one.

The PD benchmarks are the one place SEBI attaches numbers to letters, and they are ceilings built from confidence bounds, not expectations: in the long-run table a BBB category may show a one-year default rate of 3.3 per cent before it breaches. They say when the regulator would treat an agency's categories as having failed, which is a different question from what a rating should be expected to do, and a later section sets them against a far stricter number.

One agency's one-year matrix, and what five years of it look like

The newest study examined here, called study A in the figures, is the annual default and rating transition study for fiscal 2026 of one of the large Indian rating agencies, published in July 2026. It covers the agency's long-term ratings through monthly static pools over fiscal 2016 to fiscal 2026 (121 monthly cohorts, the count the study says aligns it with SEBI's norms), at category level, with non-cooperating issuers removed from the pools. Its one-year matrix is reproduced below as published.

Each row is where issuers that began a year in that category ended it. The diagonal is stability: 99.03 per cent of AAA ratings were still AAA a year later, 91.64 per cent of BBB. The D column is the one-year default rate, and it matches the study's separately published one-year cumulative default rates to the second decimal, a useful check that the two tables come from the same pools.

SEBI's statistics stop at three years. To see five, the matrix has to be multiplied by itself. The chance of going from BBB to BB in two years is the sum, over every category the issuer could have occupied at the end of the first year, of the chance of getting there multiplied by the chance of going on from there to BB. Repeating that four times, with default treated as a state nobody leaves, gives the five-year matrix.

The erosion is steady and uneven by grade. An AAA loses almost nothing to default within five years in this record, but 4.6 per cent of AAA issuers are AA or lower by then. An A keeps its category 70.3 per cent of the time, is higher 17.6 per cent and lower 11.7 per cent, with 0.48 per cent in default. A BBB has a 15.2 per cent chance of sitting below investment grade at the five-year mark and a 3.0 per cent chance of having defaulted. For anyone who marks a bond to market, or holds it under an investment-grade mandate, the migration figure, about 5 times the default figure for a BBB, is the one that bites first.

The yearly increments show the mechanism. For a BBB they rise, from 0.43 per cent in year one to 0.75 in year five, because downgrades keep feeding the riskier rows. For a B they fall, from 8.60 to 4.79, because survivors are partly upgraded and the pool left to default shrinks. One letter therefore carries a different shape of risk over time, not just a different level.

The product assumes a rating has no memory, and the measured record disagrees

Multiplying one-year matrices is a model with three assumptions: that an issuer's next move depends only on its current category and not on how it got there, that every year behaves like the average year, and that every issuer in a category behaves like every other. The study also publishes measured two- and three-year cumulative default rates from the same pools, so the model can be checked where the record exists.

It undershoots in every category that recorded a default. The product gives 1.56 per cent for a BBB at three years against a measured 1.97; 8.48 for BB against 9.70; 22.36 for B against 27.28; and for A, 0.25 against 0.58, less than half. A second large agency's transition and default study for fiscal 2025, study B, which publishes observed three-year matrices alongside its one-year matrix (ten annual pools ending March 2025, unadjusted version), gives a harsher verdict: its own one-year matrix multiplied to three years puts BBB default at 5.0 per cent against 9.1 observed, and BB at 19.7 against 35.6.

The published tables cannot apportion the gap, but they identify its candidates. A three-year pool has to begin at least three years before the window closes, so the measured three-year figure leans on the earlier years while the one-year average also includes the benign recent ones; the first study reports fiscal 2025 as its lowest default year in seventeen. Pools also shed issuers whose ratings are withdrawn, and a longer horizon sheds more of them. And ratings may carry memory: studies of international rating histories have long reported that an issuer recently downgraded into a grade is more likely to be downgraded again than others already in it, which a one-step model cannot represent, and the Indian formats are not split by prior rating, so that effect cannot be isolated here. The practical reading follows: use a measured multi-year figure wherever one exists, and treat the five-year products above, the only five-year numbers the disclosures allow, as more likely to understate the historical experience than to overstate it.

Who is in the pool decides the number

Every figure so far rests on a choice that no headline shows: which issuers the pool contains. The fiscal 2026 study publishes its tables twice, once with non-cooperating issuers removed and once with them kept, and the two describe different worlds. With them removed, 4.07 per cent of BBB ratings fell to BB within a year. With them kept, 10.86 per cent did, because a rating left in non-cooperation for six months has to be moved below investment grade. Kept in, they enlarge the B pool from 1,70,605 issuer-months to 7,97,256 and cut its one-year default rate from 8.60 per cent to 2.92. Multiplied to five years, the kept-in matrix puts a BBB below investment grade 37.6 per cent of the time, against 15.2 per cent on the other. Even the direction of the model's error changes: on the kept-in pool the three-year product overshoots below investment grade, 6.06 per cent for BB against a measured 5.09.

SEBI has taken a side. Its own transition format removes withdrawn and non-cooperating ratings, but the master circular now says that excluding them "might not depict a true picture of stability" (para 27.4.3.2), so for listed securities agencies must also publish the version that keeps them, with a column showing how many ratings were withdrawn. SEBI describes most non-cooperating issuers as unlisted and small. Both studies pool bank-loan ratings with securities, and the first attributes the recent shift in its own rating mix mainly to rising non-cooperation in sub-investment grade after banks raised the exposure levels at which they require an external rating. The lower rows of an all-ratings study describe a different population from the listed bonds a retail investor can buy, which is why the listed-securities versions exist.

Two further adjustments sit inside the agencies' own studies. The second study offers an adjusted version that removes defaults it calls isolated, including those of a large infrastructure-finance group; in that version no AAA defaulted within one year or within three, while in its unadjusted version 0.2 per cent of AAA ratings defaulted within a year and 1.2 per cent within three. The first study leaves out what it calls operational defaults, where the rating was reinstated at once. Both choices are disclosed and both can be argued, but the regulatory disclosure keeps non-cooperating issuers and isolated defaults in, which is the reason to read it alongside any study.

Two regulators now put numbers on the same letters, and for BBB they differ eightfold

From 1 April 2027, under RBI's Directions for commercial banks on the capital charge for credit risk, standardised approach (RBI/DOR/2026-27/397, issued on 27 April 2026), a bank's risk weight on a rated corporate exposure starts from a base weight for the rating category and is then adjusted by the rating agency's own record. Paragraph 27(2) tells banks to take the latest average one-year observed default rate that each domestic agency publishes for each category and compare it with RBI's reference ranges in Table 21. At or below the range, the base weight applies. Above it, the weight moves one bucket higher; for AAA, to 30 per cent. Where a borrower has two ratings, the adjustment is run on both and the higher weight applies; with three or more, on the two most favourable (para 30). Unsolicited ratings are ignored (para 29) and an INC rating floors the weight at 100 per cent, rising to 150 per cent after six months (para 31).

Set the two regulators' numbers side by side and the middle of the scale is where they part. For BBB, SEBI's long-run benchmark tolerates a one-year default rate up to 3.3 per cent; RBI's reference range stops at 0.40, about one eighth as much, and from A to BB SEBI's figure is 8 to 15 times RBI's ceiling. Only at AAA and AA, where SEBI's one-year benchmark is zero, is SEBI's the stricter. The two answer different questions, the first whether an agency's categories have broken down and the second what a letter should cost a bank in capital, but across the middle of the scale the regulator of banks has now set the far stricter bar, and it applies that bar agency by agency.

The fiscal 2026 study shows how fine the line can be. Its BBB one-year default rate is 0.43 per cent on the pool that removes non-cooperating issuers and 0.39 per cent on the pool that keeps them, either side of 0.40. That is not a prediction of how any agency's ratings will be weighted: the second study, in the version published in July 2026, says RBI indicated that a separate guidance note would set out how the observed default rate is to be computed, and that none had been issued. It is a demonstration that the pool definition, which most readers never see, can now decide between a 75 and a 100 per cent risk weight, a third more capital for the same loan.

Price is a different question

SEBI's FAQ is explicit that a rating is not a recommendation to buy, hold or sell, and is not intended to measure liquidity, pre-payment, interest rate, secondary market loss or exchange loss risk. The mechanism is in how a bond is priced: fixed cash flows discounted at a government yield plus a spread, as the guide to bond price and yield sets out. A rating speaks to one input of that spread, the chance of default. It is silent on recovery after a default, on how easily the bond can be sold and on where government yields go, and a long AAA bond loses price whenever yields rise, with no rating action at all, by the amount duration and convexity computes.

The arithmetic of expected loss shows how small the default component can be. Expected loss per year is roughly the default probability times the share lost in default, so at the fiscal 2026 study's 0.43 per cent one-year rate for BBB, every ten percentage points of loss given default add about 4.3 basis points a year. Whatever a BBB bond yields beyond that pays for risks the rating does not address, and the rating does not say whether it is enough. Speed differs as well: between 91.6 and 99.0 per cent of each investment-grade category kept its category over a full year in that study, so a rating moves rarely and in steps while a price can move in any session.

What the letter is for

Read properly, a rating is a pointer to a row. The letter says which row of an agency's matrix an issuer sits in; the row, with its stability, its migration and its default column, is the actual statement; the outlook, the watch and the sensitivities say whether this issuer is typical of its row; and the pool definition says which population the row describes. Different holders then read different columns. A holder to maturity needs the D column over the life of the bond. A fund with an investment-grade limit needs everything to the right of BBB, and the segregation rules if an issuer crosses the line. A bank from April 2027 needs the agency's observed default rate against RBI's table.

What the letter cannot do is carry its evidence. That lives in the agency's disclosures, archived for ten years, and in the studies the agencies publish each summer. Reading a rating through its record rather than as a verdict is method, and it is the method taught here.