A credit risk framework is the structured set of policies, models, limits and controls a lender uses to identify, measure, monitor and report the risk that borrowers will not repay. At its core sit three parameters: Probability of Default (PD), Loss Given Default (LGD) and Exposure at Default (EAD).
Each parameter answers a different question. PD asks how likely a borrower is to default. LGD asks how much of the exposure the lender will lose after recoveries. EAD asks how much money will actually be outstanding on the day default happens. Taken one at a time, though, none of them tells a bank what it stands to lose. You only get that answer when you put all three together.
That combined number sits behind decisions every risk analyst deals with, such as loan pricing, provisions, capital buffers and credit limits. In India, it also now carries direct regulatory weight. In April 2026, the Reserve Bank of India (RBI) finalised directions that move commercial banks to an expected credit loss (ECL) model from 1 April 2027. So from that date, PD, LGD and EAD will drive provisioning at every bank the directions cover.
In this guide, we explain each parameter, combine them in the expected loss formula with a worked example, and then map the framework to Basel and IFRS 9. We close with a real default that shows what happens when the assumptions break.
Credit risk is the possibility that a borrower or counterparty fails to meet its obligations on the agreed terms. The Basel Committee on Banking Supervision defines it in much the same way in its 2000 Principles for the Management of Credit Risk. For most banks, loans are the largest source of this risk. However, it also arises from bonds, guarantees, letters of credit, derivatives and interbank lending.
Risk teams usually separate credit risk into three working categories:
Banks lend money that mostly belongs to depositors, and they run on thin equity relative to their loan books. So a loss estimate that is off by even a percentage point or two can consume a meaningful slice of capital. Measuring credit risk lets a bank price each loan properly, decide how much capital to hold, and set aside provisions before losses arrive.
On the surface, Indian banks look comfortable right now. According to the RBI’s Financial Stability Report (June 2026), the gross NPA ratio of scheduled commercial banks fell to 1.8% in March 2026, a multi-decadal low. The same report recorded bank credit growth of 14.5% during 2025–26. But periods of fast credit growth are exactly when measurement discipline matters. A low NPA ratio describes how past lending has performed. A credit risk framework has to estimate the losses still hidden in the loans being written today.
A credit risk framework turns credit risk from a vague concern into a managed, measurable process. Think of it as a bank’s operating system for lending. Models do the calculations, but governance, limits and reporting decide how anyone uses the results. Most frameworks rest on five connected components.
| Component | What it covers | Example in an Indian bank |
| Governance | Board oversight, roles, policies and the three lines of defence | A board Risk Management Committee approves the credit policy; an independent Chief Risk Officer oversees risk, including model risk |
| Risk appetite | How much credit risk the bank will accept, and where | Limits on unsecured retail as a share of the book, or a ceiling on exposure to one industry |
| Measurement | Rating systems, scorecards, PD, LGD and EAD models, stress tests | A logistic regression scorecard for personal loans and a rating model for mid-corporates |
| Monitoring | Early warning signals, watchlists and model performance tracking | Special Mention Account (SMA) tagging of overdue loans; quarterly Gini and stability checks |
| Reporting | Management information, regulatory returns and public disclosures | Board risk dashboards, RBI returns and Pillar 3 disclosures |
Measurement is the engine of the framework, but an engine without steering is a liability. Risk appetite sets the destination, for instance a target rating mix or a maximum expected loss rate. Governance then makes sure an independent team validates the models and that the business uses them as intended.
Monitoring closes the loop. If a borrower moves into the SMA-2 bucket (61–90 days overdue), the bank should review its rating and, by extension, its PD. Similarly, if a scorecard’s Gini coefficient drops, the model team must investigate before decisions degrade. Finally, reporting makes the whole process visible to the board and the regulator.
In short, a strong framework links every number to a decision and an owner. That link is why PD, LGD and EAD must integrate cleanly. If each team estimates its parameter in isolation, the combined loss figure will be inconsistent, and so will every decision built on it.
Every credit risk parameter answers one part of a simple question: how much could we lose on this loan? Below, we take each in turn.
Probability of Default is the likelihood that a borrower defaults within a given time horizon. Banks usually express it as a percentage. For example, a PD of 2% means that, out of 100 similar borrowers, about two are expected to default within the period.
Banks estimate PD in several ways:
The time horizon matters as much as the method. A 12-month PD covers default over the next year. In contrast, a lifetime PD covers default over the remaining life of the loan. Consider a five-year term loan to a manufacturer. Its 12-month PD might be 1%. If its risk stays roughly flat, the cumulative five-year PD comes to about 4.9%, and more if credit quality is expected to slip.
A second distinction is through-the-cycle (TTC) versus point-in-time (PIT). Basel capital rules anchor PD to long-run average one-year default rates, which pushes banks toward stable TTC estimates. IFRS 9, however, requires PIT estimates that reflect current and forecast economic conditions. For a deeper look at modelling choices, see our guide to PD estimation methods
Loss Given Default is the share of the exposure a lender loses once a borrower defaults. It is the mirror image of the recovery rate, so LGD = 1 − recovery rate. If a bank recovers 60% of a defaulted loan, its LGD is 40%.
In practice, recoveries can take years, and they cost money. So a proper workout LGD discounts every future cash flow back to the default date. It also subtracts legal, collection and asset-sale costs. Three factors drive LGD most strongly:
Regulators also require downturn LGD. Defaults rise in recessions, and collateral values tend to fall at the same time. That is why Basel requires LGD estimates that reflect economic downturn conditions, not average years. Under the Foundation IRB approach, supervisors set LGD directly: 45% for senior unsecured claims under Basel II and 75% for subordinated claims. The 2017 Basel III reforms lowered the senior unsecured figure to 40% for non-financial corporates, while keeping 45% for exposures to banks and other financial institutions. We explore recovery data in detail in our article on LGD recovery rates
Exposure at Default is the amount a bank expects to be owed at the moment a borrower defaults. For a fully drawn term loan, EAD is close to the outstanding principal plus accrued interest. Revolving facilities are where it gets harder.
Consider a cash credit or overdraft line. Borrowers in distress tend to draw down unused limits before they default. As a result, EAD includes part of the undrawn amount, using a Credit Conversion Factor (CCF):
EAD = \text{Drawn} + CCF \times \text{Undrawn}
Suppose a company has a ₹10 crore cash credit limit and has drawn ₹6 crore. With a CCF of 50%, EAD is ₹6 crore + 50% × ₹4 crore, which equals ₹8 crore. Banks with enough history estimate CCFs from how past defaulters drew down their lines. Others apply supervisory CCFs under standardised or foundation approaches.
Derivatives need a different method again. Their exposure depends on future market values, so banks model potential future exposure. Under Basel III, banks without internal model approval use the standardised approach for counterparty credit risk (SA-CCR) for this.
One common mistake is treating EAD as a fixed number. In reality, EAD changes with amortisation schedules, prepayments and borrower behaviour. This matters most for lifetime ECL, where the bank must project exposure for every future period.
The three parameters combine in a single multiplication. This is the expected loss formula:
EL = PD \times LGD \times EAD
The logic is easy to follow. PD captures how often loss happens, LGD captures how severe it is, and EAD captures how large the exposure is. Multiply frequency, severity and size, and you get the average loss the bank should expect.
Let us return to the ₹10 crore cash credit line from the EAD section. Assume the borrower is a mid-sized manufacturer with a security package of receivables and plant. The figures below are illustrative.
| Step | Parameter | Calculation | Result |
| 1 | EAD | ₹6 crore drawn + 50% CCF × ₹4 crore undrawn | ₹8 crore |
| 2 | PD (12-month) | From the borrower’s internal rating grade | 2.0% |
| 3 | LGD | Recoveries net of haircuts and costs, discounted | 40% |
| 4 | Expected loss | 2.0% × 40% × ₹8 crore | ₹6.4 lakh |
| 5 | EL rate | ₹6.4 lakh ÷ ₹8 crore | 0.8% of EAD |
The 0.8% EL rate is useful on its own. It tells the relationship manager that the loan’s pricing needs at least 0.8% a year just for credit losses, on top of funding, operating and capital costs.
Now apply a downturn. Suppose the PD rises to 5% and LGD rises to 50% as collateral values fall. Expected loss becomes 5% × 50% × ₹8 crore, or ₹20 lakh. That is more than three times the base case. In reality, EAD would often rise as well, because stressed borrowers draw down their limits. That is why you can’t estimate the parameters in silos: in a downturn, they tend to get worse together.
Expected losses also add up. Summing EL across every loan gives the portfolio’s expected loss, which the bank can compare with provisions and pricing income.
Expected loss is an average. Real losses, however, swing above and below it from year to year. Unexpected loss (UL) measures that volatility, especially the severe tail when many borrowers default together.
An insurance analogy helps. An insurer sets premiums to cover average claims. It then holds capital for the bad year when claims spike. Banks work the same way. Pricing and provisions absorb expected loss, whereas capital absorbs unexpected loss.
In practice, most errors come from parameters that do not match each other:
Regulators use the same three parameters for two different purposes. Basel uses them to set capital for unexpected loss. IFRS 9, in contrast, uses them to set accounting provisions for expected loss.
Under the Basel IRB approach, approved banks feed their own estimates into a supervisory risk-weight formula. In the Foundation IRB approach, banks estimate PD, while supervisors prescribe LGD and the conversion factors used for EAD. In the Advanced approach, banks estimate all three. The formula then calculates capital to cover unexpected loss at a 99.9% confidence level over one year. Expected loss sits outside this capital charge. Instead, banks compare it with their provisions, and any shortfall reduces capital.
India follows a different path for capital. The RBI requires banks to calculate credit risk capital under the Standardised Approach, and it has not permitted IRB models for this purpose. Its revised Standardised Approach directions, issued in April 2026, also take effect on 1 April 2027.
The IASB issued IFRS 9 in July 2014, and it became effective on 1 January 2018. India’s equivalent is Ind AS 109, which larger NBFCs already apply. Under IFRS 9 expected credit loss rules, each exposure falls into one of three stages:
The RBI has now brought this logic into prudential rules. Its Commercial Banks – Asset Classification, Provisioning and Income Recognition Directions, 2026, dated 27 April 2026, apply from 1 April 2027. They cover commercial banks, excluding small finance banks, payments banks and local area banks. The directions also set prudential floors, so modelled ECL cannot fall below minimum provisioning levels.
| Feature | Basel IRB | IFRS 9 / Ind AS 109 ECL |
| Purpose | Regulatory capital | Accounting provisions |
| Loss covered | Unexpected loss | Expected loss |
| PD type | Through-the-cycle, 12-month | Point-in-time, 12-month or lifetime |
| LGD | Downturn LGD | Best-estimate LGD across scenarios |
| Macroeconomic view | Long-run average | Probability-weighted forward-looking scenarios |
| Applies in India | Not permitted for capital | Ind AS NBFCs now; commercial banks from April 2027 |
Lehman Brothers filed for bankruptcy on 15 September 2008. The BIS Quarterly Review (December 2008) records what happened next in the CDS market. On 10 October, an ISDA auction among 14 dealers set the recovery value of Lehman bonds at just 8.625%. The same BIS box notes that DTCC counted about $72 billion of CDS contracts referencing Lehman, with roughly $6 billion in net settlements.
Put that in the language of our three parameters. A recovery of 8.625% implies a market LGD of about 91% for senior unsecured bondholders. Foundation IRB, by contrast, assigns 45% to senior unsecured claims on financial institutions. To be fair, 45% is a portfolio-level calibration, not a forecast for any single name. Still, a gap that wide shows how far one default can stray from the average.
The PD signal was weak as well. Lehman still carried investment-grade credit ratings in the days before its filing, so rating-based PDs signalled little danger. Meanwhile, counterparties’ EAD on derivatives and funding lines shifted rapidly as markets moved.
One caveat is worth stating. The auction price reflects market value shortly after default, not the final workout recovery. Even so, the lesson for any credit risk framework is clear: PD, LGD and EAD deteriorate together in a crisis, so stress testing must move them together as well.
A credit risk framework works only when its parts connect. PD tells you how likely default is, LGD tells you how much you lose, and EAD tells you how much is at stake. Together, the expected loss formula turns them into a number a bank can price, provide for and capitalise against. With the RBI’s ECL directions taking effect in April 2027, these skills are now core requirements for risk roles in Indian banking.
What is a credit risk framework? A credit risk framework is the set of policies, models, limits and controls a bank uses to manage lending risk. It covers governance, risk appetite, measurement, monitoring and reporting. Its measurement core estimates Probability of Default, Loss Given Default and Exposure at Default, then combines them to calculate expected credit losses.
What is the expected loss formula? Expected loss equals Probability of Default multiplied by Loss Given Default multiplied by Exposure at Default (EL = PD × LGD × EAD). For example, a ₹8 crore exposure with a 2% PD and 40% LGD has an expected loss of ₹6.4 lakh. Banks cover expected loss through pricing and provisions.
How does the Basel IRB approach differ from IFRS 9 expected credit loss? Basel IRB uses PD, LGD and EAD to calculate regulatory capital for unexpected loss, with through-the-cycle PDs and downturn LGD. IFRS 9 uses the same parameters to calculate accounting provisions for expected loss. It relies on point-in-time, forward-looking estimates and applies 12-month or lifetime ECL depending on the exposure’s stage.
When does the RBI expected credit loss framework apply to Indian banks? The RBI issued its final ECL-based directions on 27 April 2026, effective 1 April 2027. They apply to commercial banks, excluding small finance banks, payments banks and local area banks. Banks must stage exposures and estimate 12-month or lifetime ECL using PD, LGD and EAD, subject to prudential floors.
Explore Dexlab Analytics’ Credit Risk Modeling certification program to build PD, LGD, and EAD models from scratch, work through IFRS 9 ECL frameworks, and learn model validation techniques used by practicing risk teams.
.
Credit and Market Risk, Credit Risk, credit risk analysis, Credit Risk Analytics And Modeling, credit risk analytics training, Credit risk certification India, Credit risk modeling course India, Credit Risk Modelling, Credit Risk Modelling Using SAS, Credit scorecard modeling, Expected credit loss ECL, IFRS 9 modeling course India, Logistic regression scorecard, PD LGD EAD modeling, Python credit risk modeling
Comments are closed here.