A credit rating is a relative, ordinal, through-the-cycle opinion on the probability of default and expected loss. It is not a buy/sell recommendation, not a point-in-time price, and (the agencies are emphatic) not a guarantee. Learn the scales first, because the whole market is wired to them.
The scales. Moody's runs Aaa → Aa1/Aa2/Aa3 → A1/A2/A3 → Baa1/Baa2/Baa3 → Ba… → B… → Caa… → Ca → C. S&P, Fitch, and India's CRISIL/ICRA/CARE run AAA → AA+/AA/AA− → A+/A/A− → BBB+/BBB/BBB− → BB… → B… → CCC/CC/C → D (default). The investment-grade / speculative-grade line falls between BBB−/Baa3 (the lowest IG) and BB+/Ba1 (the highest speculative, or "junk," or "high-yield"). That line is the most consequential boundary in the market because a vast pool of regulated capital may hold only IG. Short-term scales are separate (A1+…D in India; P-1… for Moody's) and map roughly, not exactly, to the long-term scale. Verify current symbology, which the agencies periodically revise.
Advertisement
How does a rating scorecard work?
The scorecard logic. Modern methodologies are largely transparent grids. A rating is built from two halves:
- Business risk: how safe are the earnings before you even look at the debt? Scale (bigger is safer), diversification (by product, geography, customer), competitive/market position (cost curve, moat, pricing power), and industry cyclicality and structure. A regulated utility scores strongly on business risk; a single-commodity, single-plant producer scores poorly however clean its balance sheet.
- Financial risk, the credit ratios: leverage (Debt/EBITDA), coverage (EBITDA/Interest), and above all the cash-flow ratios (FFO/Debt, RCF/Debt), plus financial policy (does management target a leverage band and mean it?) and liquidity (sources vs uses over the next 12–24 months).
Each factor gets a sub-score, the sub-scores are weighted (weights differ by sector: cash flow dominates for corporates, capital and asset quality for banks/NBFCs), and the weighted result maps to an indicated standalone rating. Worked example 5 builds one end-to-end.
Notching for support. The standalone credit profile is then adjusted for external support: a subsidiary of a strong parent may be notched up toward the parent (if the parent is likely to support it); a company reliant on a weak parent may be capped; a government-related issuer gets sovereign-support uplift (or is capped at the sovereign ceiling). India adds a specific vocabulary: "CE" (credit enhancement) ratings for instruments backed by a specific support structure (guarantee, escrow, pledge). The agency rates those on the structure, and they must be labelled so investors do not confuse a structured "AA(CE)" with a standalone "AA". India also separates issuer from instrument ratings (a senior secured instrument may rate above the issuer). Verify the current suffix conventions, which SEBI has changed over time.
What are the limits of credit ratings?
The limits, burn these in. Ratings lag: they are through-the-cycle by design and change slowly, so the market's spread often moves first (a bond trading at a distressed spread while still rated investment grade is the market telling you the rating is stale). Ratings face an issuer-pays conflict: the rated company pays the agency, a structural tension the analyst must hold in mind. And ratings can be catastrophically late at the cliff: IL&FS group entities held AAA ratings within weeks of defaulting to D in September 2018, the entire alphabet compressed into six weeks once the commercial paper stopped rolling. The lesson is not "ignore ratings" (they are a useful, comparable, information-rich prior) but "a rating is an opinion with a refinancing assumption inside it; read the ALM ladder, the maturity wall, and the spread yourself, and treat a rating that the market's pricing contradicts as a question, not an answer."
How do you analyse municipal and state bonds?
Everything above assumes a company. A large part of the world's rated debt is issued by entities that are neither a company nor a sovereign: cities, states, water utilities, toll authorities, port trusts. This is municipal and other non-sovereign government credit, and it is worth a section of its own because these issuers default rarely and, when they do, they default differently. The thing standing behind the bond is either a taxing power or a revenue stream, and never a business that can be sold. There is no enterprise value to distribute, no fulcrum security, and often no practical way to enforce a claim against an entity that must keep providing water.
The first cut is between the two structures. A general obligation bond is backed by the issuer's full faith, credit and taxing power. There is no specific project and no dedicated revenue; the promise is that the issuer will levy what it takes. Analysis is therefore about the tax base and about willingness. You look at the size, diversity and trajectory of the local economy, the concentration of the top ten taxpayers, employment and population trends, and the debt burden per head and against assessable property. Then you look at what competes with debt service for the same tax rupee, which in practice means pensions and other post-employment obligations, because those are politically far harder to cut than a bond payment is to delay.
A revenue bond is backed only by the net revenues of a defined enterprise, and the taxing power is explicitly not pledged. Analysis is therefore about that enterprise. Is demand essential or discretionary? Can the issuer set its own prices or must it ask a regulator? How far do revenues exceed debt service, and what is the flow of funds, meaning the contractual order in which cash pays operating costs, senior debt service, reserves and subordinate debt? That order is the muni version of the priority of claims waterfall, except that it runs every month rather than once at the end.
Those two cuts produce a more useful third one, between tax-backed credits and enterprise-revenue credits. It is the better sort because the legal form often misleads. A general obligation of a city whose tax base is one factory is closer in risk to a revenue bond than to a diversified state's general obligation, and an essential-service revenue bond with an unregulated rate covenant can be stronger than either.
Two covenants do most of the work in a revenue-bond indenture, and they are the muni analogue of the maintenance covenants covered in debt covenants. The rate covenant obliges the issuer to set prices at a level that produces net revenues of at least a stated multiple of debt service, typically 1.10 to 1.25 times, and it is the strongest protection in the structure because it converts a shortfall into a legal duty to raise the tariff. The additional-bonds test stops the issuer from diluting you, by requiring that new senior debt clear a coverage test, usually 1.25 times, on historical or projected revenues.
Work one. A synthetic water and sewer system (illustrative) has gross revenues of $148,000,000 and operations and maintenance costs of $92,500,000, so net revenues available for debt service are $55,500,000. Senior-lien annual debt service is $34,800,000 and a subordinate lien takes another $8,600,000. Senior coverage is 55.5 ÷ 34.8 = 1.594828 times, comfortably above a 1.20 rate covenant, and net revenues could fall 24.7568% before that covenant is breached. All-in coverage is 55.5 ÷ 43.4 = 1.278802 times, and against a 1.10 all-in covenant the cushion is only 13.9820%, which is the number to quote. Now stress it: at gross revenues of $132,000,000 net revenues fall to $39,500,000 and senior coverage drops to 1.135057 times, below the covenant. The rate covenant then bites in the issuer's favour and in yours. Restoring 1.20 times needs net revenues of $41,760,000, an increase of $2,260,000, which is a tariff rise of 1.7121% of stressed gross revenue. That is why an essential-service revenue bond with a rate covenant and no regulatory veto is a different instrument from one that must petition a commission for every increase.
India's analogue has three layers and the vocabulary is different. State development loans are the states' own market borrowings, auctioned by the Reserve Bank of India, carrying no explicit central-government guarantee and trading at a spread of a few tens of basis points over the central G-sec of matching tenor. They are the closest thing to a general obligation. The spread across states is compressed relative to their fiscal dispersion, which is a market judgement about implicit support rather than about state finances (check the current levels and any change in the framework). Municipal bonds proper are issued under the SEBI issue and listing regulations for municipal debt securities, usually with an escrow of specified revenues and a debt-service reserve, and they remain a small market by value. Public-sector enterprise and infrastructure paper carries whatever explicit or implicit sovereign or state support its structure creates, which is where this section's notching and the CE label do their work.
One warning about reading the state-loan curve. Its shape is not mostly a credit statement, and treating it as one will mislead you. The FI1.03 treatment of preferred habitat is the right lens, and that lens is about curve segmentation. Insurers and pension funds with long rupee liabilities buy the long end because their liabilities live there, not because they have formed a view on any particular state, and the resulting demand can flatten or invert a segment of the curve while nothing about the borrowers has changed. Read a spread move across the whole state complex as a flow event first and a credit event second, and reserve the credit reading for a state whose spread moves against its peers.
How is sovereign credit rated?
A sovereign is the one borrower that cannot be sued into liquidation, and sovereign credit therefore splits at a point no corporate analysis has. Ability to pay is arithmetic: the debt stock, the deficit, growth, the interest bill and the currency the debt is written in. Willingness to pay is a political judgement about whether a government that can service its debt will choose to. It is the reason sovereign defaults cluster around elections, regime changes and moments when the political cost of austerity exceeds the political cost of default. Most sovereign defaults in history were choices made by governments with assets rather than by governments with none, which is why a purely fiscal model of sovereign risk has never worked well.
The consequence for an analyst is that there is no waterfall to compute. Restructuring a sovereign is a negotiation among creditors with no bankruptcy court and no enforceable priority, mediated by collective-action clauses in the bond documents and by whichever official-sector lender is prepared to fund the gap. What you are estimating is not a recovery from an estate; it is the haircut a government expects to be able to impose and still return to the market.
That split explains the market's most confusing convention, the local-currency and foreign-currency rating pair. A government that borrows in its own currency controls the printing press, so it can always nominally repay, and its local-currency rating is usually at or above its foreign-currency rating. A government that borrows in a currency it cannot print must earn or borrow that currency, so foreign-currency debt is a genuine hard constraint. M7.04 states the principle as "denomination is destiny," and the rating pair is that principle written into a scale. The gap between the two ratings is a measure of how much of the sovereign's protection comes from monetary sovereignty rather than from solvency, and the honest reading is that repaying in printed money is a default on purchasing power that the rating scale does not measure.
Agency sovereign scorecards are built from four factor groups, weighted and then adjusted. Economic strength covers income per head, size, growth and diversification. Institutions and governance strength covers policy credibility and effectiveness, rule of law, the track record on inflation targets, and data transparency. Fiscal strength covers debt to gross domestic product, the interest bill against revenue, and the maturity and currency profile of the stock. Susceptibility to event risk covers political, banking-sector, external-liquidity and government-liquidity risk, and it usually acts as a cap rather than an average, because a sovereign is only as strong as the shock it cannot survive. Score four factors at 3, 4, 5 and 4 on a scale where 1 is strongest, weight them equally, and the weighted score is 4.000000 while the binding constraint is the 5 on fiscal strength, which is the number the committee argues about.
The event-risk group deserves its own note, because it is where sovereign analysis differs most from corporate analysis and where the numbers are easiest to find and hardest to interpret. Four sub-risks sit inside it. External liquidity asks whether foreign-currency obligations falling due in the next twelve months are covered by reserves and expected inflows, and the standard reading is reserves against short-term external debt plus the current-account deficit. Banking-sector risk asks what a bank rescue would cost the budget, because a banking system large relative to output converts a private balance-sheet problem into a sovereign one overnight. Government liquidity asks whether the maturity profile forces the state into the market at a bad moment, which is the distressed debt maturity wall at national scale. And political risk asks whether the institutions that would have to deliver an adjustment can survive delivering it. None of these averages with the others; the weakest one caps the rating, which is why the factor group behaves like a constraint rather than a score.
Compute the fiscal-strength input rather than asserting it, using M7.04's debt arithmetic without re-deriving it. Take debt at 82.0000% of gross domestic product, a nominal effective interest rate on that debt of 7.2000%, nominal growth of 9.8000%, and a primary deficit of 1.5000% of output. The automatic dynamics term (r − g) ÷ (1 + g) × d is −1.941712%, so debt falls by 0.441712 points to 81.558288% despite the primary deficit, and the debt-stabilising primary balance is a deficit of 1.941712%. That is the benign regime, and it is doing all the work. Now let a confidence shock take the effective rate to 12.5000% with growth unchanged. The automatic term flips to +2.016393%, debt rises 3.516393 points in the first year, and holding it flat now requires a primary surplus of 2.016393% of output, a swing of 3.516393 points of gross domestic product from where the budget actually is. Nothing about the country changed except the rate at which it borrows. That single number is why the willingness question and the ability question are not separable in practice: the rate a sovereign pays is partly a market judgement about its willingness, and that judgement then changes what it is able to do.
How do credit scoring models work?
Credit scoring models are the last thing on this list, and they are two different objects wearing one name. An agency scorecard and a statistical credit score both compress a balance sheet into a single figure. A scorecard is a stated weighting of factors chosen by analysts, published in a methodology document, and adjusted afterwards by a committee; Worked example 5 builds one. A statistical score is fitted: its weights are estimated from a sample of firms that did and did not fail, and nobody adjusts it afterwards. The first is an argument you can disagree with clause by clause. The second is a summary of what separated survivors from failures in one historical sample, which is a much narrower claim than it looks.
The Altman Z-score (Edward Altman, 1968) is the canonical fitted version, and computing it once shows what such a model contains. In its original public-manufacturer form it is Z = 1.2X₁ + 1.4X₂ + 3.3X₃ + 0.6X₄ + 1.0X₅. The five variables are working capital, retained earnings and earnings before interest and tax, each over total assets, then the market value of equity over the book value of total liabilities, and finally sales over total assets. Take a synthetic listed manufacturer with total assets of ₹4,250 crore and total liabilities of ₹2,890 crore. Its working capital is ₹180 crore, retained earnings ₹615 crore, earnings before interest and tax ₹388 crore, market capitalisation ₹2,460 crore and sales ₹5,180 crore. The five terms are 0.050824, 0.202588, 0.301271, 0.510727 and 1.218824, summing to a Z of 2.284233. That sits in the grey zone between the 1.81 distress cut and the 2.99 safe cut, which is the honest answer and also an uncomfortable one, because a grey-zone score is precisely the case a committee wanted the model to settle.
Before the criticisms, be clear about what scores are legitimately for, because they are used seriously by people who know all of the following. A score is a screening and ranking device: run it across two hundred names and it tells you which twelve to read first, which is a real saving and a real edge over reading alphabetically. It is a monitoring trigger: a score that crosses a band boundary is a reason to re-open a file, in the way the covenant headroom in distressed debt is. And in a bank it is a calibrated input to regulatory capital, where an internal model maps a score to a probability of default estimated on that bank's own portfolio, which is a different and much more demanding exercise than reading a published cut-off. The word doing the work there is calibrated: an uncalibrated score is an ordering, not a probability, and treating a raw Z as though it were a default rate is the most common misuse of the whole family.
Three things follow, and they matter more than the number. First, a fifth of this score is a share price: the market-value term contributes 22.3588% of the total, so a 30% fall in the stock, with no change to a single financial statement, drops Z to 2.131015. A model with the market inside it will always look prescient in hindsight, because it moves when the market does. Second, the cuts are cohort-dependent. Altman's coefficients were fitted on US manufacturers of a particular era, and applying them unadjusted to an Indian services company is a category error however clean the arithmetic; use the score as a ranking device inside a comparable set, not as an absolute verdict. Third, and most important, a score predicts a default, not a return. Those are different questions, and confusing them is the standard error. Take a bond in a score band with an assumed one-year default probability of 4.5000% and a loss given default of 55.00%: expected loss is 247.5000 bp. If the bond trades at a 410.0000 bp spread, expected loss is 60.3659% of the spread and the remaining 162.5000 bp is the risk and liquidity premium in spreads for credit. The weakest credit in a set can be the best investment in it, if it is priced for a worse outcome than you expect, and the safest credit can be the worst. A default model ranks survival. Only a comparison of spread against your own expected loss ranks return.