Credit Risk Modeling at a Glance
Credit risk modeling means building statistical models that answer three questions about any loan. How likely is the borrower to default? That is PD — probability of default. How much money would be at stake if they did? That is EAD — exposure at default. And what share of that money would the lender actually lose? That is LGD — loss given default.
Multiply all three together and you get expected loss. The Basel Committee writes this as EL = PD × EAD × LGD. This one formula sits underneath almost everything a bank's risk team does.
It decides the scorecard that approved your credit card. It sets the interest rate on a business loan. It shapes the loss provisions — the money a bank sets aside for expected loan losses — in an annual report. It even decides how much capital the bank must hold.
In India, this skill has just become urgent. On 27 April 2026, RBI (the Reserve Bank of India) issued final directions. Commercial banks must move to expected-credit-loss (ECL) provisioning from 1 April 2027, phased in through FY 2030–31.
Large NBFCs (non-banking financial companies) have already provisioned this way under Ind AS 109 since 2018–19. Every one of these institutions now needs people who can build, run and check credit risk models.
What Are PD, LGD and EAD?
PD, LGD and EAD are the three risk numbers at the heart of every modern credit risk model. Here is what each one means, in the Basel Committee's own plain-English framing. PD is the average percentage of borrowers in a rating grade who default within one year. EAD is the estimated amount outstanding at default — the balance already drawn, plus drawdowns likely to follow. LGD is the percentage of that exposure the bank might still lose once recoveries are counted (see the BCBS explanatory note on the IRB risk weight functions).
| Parameter | The question it answers | Measured as | Typical drivers |
|---|---|---|---|
| PD — Probability of Default | How likely is this borrower to default? | % over a one-year horizon | Credit history, income stability, leverage, bureau score |
| EAD — Exposure at Default | How much will be at stake if they do? | ₹ outstanding at default, including likely drawdowns | Sanctioned limit, utilisation behaviour, product type |
| LGD — Loss Given Default | How much of that will we fail to recover? | % of EAD lost after recoveries | Collateral quality, seniority, legal enforcement timelines |
A worked example makes the math concrete. Take an illustrative ₹50 lakh loan against property.
Say the model gives it a 2% PD for the next year. The full ₹50 lakh would be outstanding at default, so that is the EAD. After selling the collateral, the bank expects to lose 40% of the exposure — that is the LGD.
Expected loss = 0.02 × ₹50,00,000 × 0.40 = ₹40,000 per year. This ₹40,000 does not mean the bank predicts this one borrower will default. It is the average yearly loss across thousands of similar loans. That average is what flows into how the loan gets priced and provisioned.
One more distinction matters here. Expected loss is the average loss a bank can plan for, and pricing plus provisions cover it. Losses above that expected level are called unexpected losses, and regulatory capital exists to absorb those instead.
The Basel framework also sets a floor under model inputs: corporate and bank exposures carry a PD floor of 0.05% under CRE32 of the Basel Framework. In other words, no model can ever claim a borrower is risk-free.
How Does the IFRS 9 Expected Credit Loss Model Work?
IFRS 9 is the global accounting standard that made credit risk models compulsory for loss provisioning. It was issued by the IASB (International Accounting Standards Board) in July 2014, and took effect from 1 January 2018. Before this standard, banks used an "incurred loss" approach: a bank booked a loss only after it saw evidence of trouble.
The new standard flipped this to a forward-looking approach instead. Now banks must recognise expected credit losses at all times, using past events, current conditions and forecasts. They must also update the number at every reporting date (see the BIS Financial Stability Institute summary).
The standard sorts every loan into one of three stages. The stage a loan sits in decides two things: how big the provision must be, and how interest income gets recognised.
| Stage | Trigger | Provision required | Interest recognised on |
|---|---|---|---|
| Stage 1 | Loan originated or purchased; credit risk not significantly higher since origination | 12-month ECL | Gross carrying amount |
| Stage 2 | Significant increase in credit risk (SICR) — rebuttable presumption at 30+ days past due | Lifetime ECL | Gross carrying amount |
| Stage 3 | Credit-impaired | Lifetime ECL | Amortised cost (gross amount minus allowance) |
Two details separate candidates who have actually studied ECL from those who have just memorised a diagram. First, "12-month ECL" does not mean next year's expected cash shortfall. It is the slice of lifetime losses tied to a default happening in the next 12 months.
Second, the SICR (significant increase in credit risk) test tracks the change in default risk over the loan's life — that is, the change in PD. It does not track the change in the loss amount.
Both details come straight from the BIS Financial Stability Institute's summary of the standard. Interviewers use exactly this kind of nuance to filter candidates. It is also why lenders need modelers at all: someone has to estimate lifetime PDs, decide what counts as "significant" deterioration, and build macroeconomic forecasts into the numbers.
What Do RBI's New ECL Rules Mean for Indian Banks?
On 27 April 2026, RBI notified its final directions on expected credit loss provisioning. This is the biggest overhaul of Indian bank provisioning in decades. Banks currently provision under the old incurred-loss rules, called IRACP (Income Recognition, Asset Classification and Provisioning).
From 1 April 2027, covered banks move to a three-stage, IFRS 9-style ECL framework instead. The hit to profitability and capital is allowed to be phased in through FY 2030–31 (see KPMG India's summary of the directions).
Here are the essentials of the final framework:
- Who is covered. Commercial banks, excluding small finance banks, payments banks and local area banks — per KPMG's reading of the directions.
- Three stages, familiar logic. Stage 1 carries 12-month ECL. Stage 2 carries lifetime ECL, once credit risk has risen significantly. Stage 3 carries lifetime ECL too, once the loan is credit-impaired.
- A 30-day presumption. Credit risk is presumed to have risen significantly once payments are more than 30 days past due (this can be rebutted). The NPA (non-performing asset) definition stays at 90+ days overdue, unchanged.
- Prudential floors. RBI has set minimum provisioning floors, by product, for Stage 1 and Stage 2 exposures. These act as regulatory backstops — a model's output can never fall below them.
NBFCs (non-banking financial companies) make a useful contrast, because the larger ones have already lived through this transition. NBFCs that adopted Ind AS did so from FY 2018–19, if their net worth was ₹500 crore or more, and from FY 2019–20 for other covered NBFCs.
These NBFCs compute ECL under Ind AS 109 (the Indian accounting standard that mirrors the IFRS framework). RBI's March 2020 guidance added guardrails on top: board-approved ECL methods, documented assumptions, and an Impairment Reserve whenever the model's allowance comes in below IRACP provisioning.
RBI finalised ECL and Basel III credit-risk capital rules on the same day. It also issued Basel III standardised-approach directions for credit risk capital, effective from the same date of 1 April 2027.
India is using the standardised approach for capital, not internal models. Consulting firm Uniqus notes that IRB (internal ratings-based) model usage among Indian banks is near zero. That is what makes the ECL provisioning models the main event for hiring.
What Models Do Credit Risk Analysts Actually Build?
"Credit risk modeling" is not one model — it is a family of them. Each one answers a different business question: approval, monitoring, provisioning, capital or resilience. A typical Indian bank, NBFC or global capability centre runs all of them side by side. Most analysts specialise in just one or two:
- Application scorecards. These rank new applicants by default risk at the point of sanction. This is the classic logistic-regression scorecard behind instant loan decisions.
- Behavioural scorecards. These re-score existing borrowers using their repayment and card-usage behaviour. They power limit increases, collections queues and early-warning alerts.
- PD, LGD and EAD estimation models. These are the statistical engines behind ECL and pricing. They are calibrated to a lender's own default history and macro forecasts.
- ECL computation engines. These combine the three parameters across stages and scenarios into one provision number — the number auditors and RBI will scrutinise.
- Stress-testing models. These project portfolio losses under bad macro scenarios. RBI itself runs these at system level. Its June 2026 Financial Stability Report stress-tested 46 banks. The report projected the aggregate gross NPA ratio — bad loans as a share of all lending — could edge up from 1.8% to 1.9% by March 2028.
- Model validation. This is the independent second line that challenges everyone else's models. It is a hiring niche of its own — Indeed India showed around 800 postings matching "credit risk model validation" on 5 July 2026.
Which of these you end up building depends on where you sit. Banks and large NBFCs need the full stack. Big 4 teams validate and implement models for clients. Global capability centres run scorecards, stress tests and validation for parent banks abroad.
Which Tools and Skills Do Credit Risk Modelers Need?
Credit risk modeling sits at the crossroads of statistics, programming and regulation. You do not need all of it on day one. But the hiring bar is a working mix of the skills below:
- Python — now the default. On 5 July 2026, Indeed India keyword searches showed roughly 35,000 postings pairing Python with credit risk, against about 5,000 for SAS — a 7:1 gap. Treat these as broad portal counts, not a rigorous survey. But the direction is clear.
- SAS — still paying the bills. Plenty of bank model inventories and regulatory pipelines still run on SAS, which is why postings for it persist. Learning Python first, and staying SAS-aware, is the safest bet.
- SQL and Excel. Loan-book data lives in databases, so you need SQL to pull it. Committees read Excel, so you need that too. Both are assumed, not optional.
- Statistics you can actually defend. Logistic regression, discriminatory-power measures and calibration testing. Validators and auditors will ask why your model works, not just whether it does.
- Regulatory literacy. IFRS 9 / Ind AS 109 staging, RBI's ECL directions, and Basel capital basics. This is the vocabulary of every model documentation pack.
- Documentation and communication. Every model needs a paper trail a regulator can follow. Analysts who write clearly are the ones who get pulled into the important reviews.
Why Is Credit Risk Modeling a Growing Career in India?
Three forces are pulling in the same direction. First, there is a regulatory build-out. Every bank covered by the ECL directions must stand up compliant PD/LGD/EAD models, staging logic and validation before 1 April 2027 — and then keep maintaining them, permanently. Consulting firms, Big 4 practices and banks' own risk teams are all hiring for this.
Second, there are global capability centres (GCCs — offshore hubs that global companies run in India). The Risk Management Association of India counts more than 1,500 GCCs across Bengaluru, Hyderabad, Pune, Mumbai and Chennai. Credit risk analyst, stress-testing and risk-reporting roles are among their most in-demand profiles.
Taggd's Decoding Jobs 2026 research adds a striking number. BFSI GCCs (banking, financial services and insurance) are only about 10% of GCCs by count, yet they employ 33% of India's GCC workforce. The same research forecasts BFSI hiring to rise 8.7% in FY 2025–26.
Third, you can see the demand in live hiring. On Indeed India on 5 July 2026, credit-risk and model-validation postings were live across Mumbai, Bengaluru, Hyderabad, Delhi and Chennai. The employers included JPMorganChase, Goldman Sachs, UBS, Citi, Nomura, Standard Chartered, DBS Bank, SMBC, KPMG, TCS, Genpact, Tata Capital and Aditya Birla Group.
The asset-quality backdrop makes the case sharper, not weaker. Gross NPAs of Indian banks touched a multi-decadal low of 1.8% in March 2026, per RBI's Financial Stability Report released on 30 June 2026 (coverage via ThePrint/PTI). Clean loan books are not a reason to shrink risk teams. They are what disciplined measurement plus regulation actually produce — and the new ECL regime raises that bar again.
On pay: AmbitionBox estimates credit risk analysts in India earn ₹12.7–14 lakh per year (1.8k salaries, updated 2 July 2026). Glassdoor's estimate is close, at ₹12.25 lakh average total pay (452 salaries, July 2026). We break the full picture down by experience, employer type and skill premium in our credit risk analyst salary in India guide.
How Do You Learn Credit Risk Modeling?
This learning path is more structured than most finance skills, because the target roles are clearly defined. Here is a sequence that works for both students and working professionals:
- Step 1 — Credit fundamentals. Learn how lenders assess borrowers before you try to model them. This means the 5 Cs, financial-statement red flags, and security and seniority. Our credit analysis guide covers this ground.
- Step 2 — Statistics you will actually use. Probability, logistic regression, discrimination and calibration. You need enough to defend a scorecard, not to publish research papers.
- Step 3 — Python and SQL on lending data. Use Pandas for data prep, scikit-learn for estimation, and SQL to pull the loan book. This is where portal demand is: 35,000 Python-tagged credit-risk postings vs 5,000 for SAS on Indeed India, 5 July 2026.
- Step 4 — Build a PD scorecard end to end. One portfolio project, from data cleaning to validation charts, beats ten certificates in interviews.
- Step 5 — Layer in ECL mechanics. Add staging, SICR triggers, 12-month vs lifetime ECL, and prudential floors. This regulatory context is what turns a data scientist into a credit risk modeler.
If you want that path compressed, with faculty support, QuintEdge's Credit Risk Modeling course is self-paced and built around exactly this progression. It covers PD/LGD/EAD models and ECL computation in Python, on realistic lending data. Pair it with FRM coaching if you also want the certification signal: the FRM curriculum covers credit risk measurement theory that complements the hands-on build.
Frequently Asked Questions About Credit Risk Modeling
Credit risk modeling uses data and statistics to estimate how likely a borrower is to default, and how much a lender would lose if they did. Banks and NBFCs use these models to approve loans, price interest rates, set loss provisions under ECL rules, and hold the right amount of regulatory capital.
PD (probability of default) is the likelihood a borrower defaults, usually measured over one year. EAD (exposure at default) is the amount likely to be outstanding when default happens, including expected future drawdowns. LGD (loss given default) is the percentage of that exposure the lender ultimately loses. Multiply the three together and you get the expected loss on a loan.
From 1 April 2027. RBI issued final expected credit loss directions on 27 April 2026, after a draft in October 2025. These cover commercial banks other than small finance banks, payments banks and local area banks, and allow the financial impact to be phased in through FY 2030–31.
The larger ones do. NBFCs that implemented Ind AS did so from FY 2018–19, for those with net worth of ₹500 crore or more, and from FY 2019–20 for other covered NBFCs. They already compute expected credit losses under Ind AS 109, with RBI guardrails like board-approved methodologies and an Impairment Reserve where ECL provisions fall below IRACP levels.
Increasingly, yes. Indeed India keyword searches on 5 July 2026 showed about 35,000 postings mentioning Python with credit risk, against roughly 5,000 for SAS — a 7:1 gap. These are broad keyword counts, not a rigorous survey, but the gap is real. Many bank systems still run SAS, so knowing both, plus SQL, is the safest combination.
Yes — demand is strong and growing. Every large bank must build and check ECL models before April 2027. Bigger NBFCs already run them, and global banks staff large risk teams in Indian GCCs. AmbitionBox estimates credit risk analysts earn ₹12.7–14 lakh per year (1.8k salaries, July 2026); our credit risk analyst salary guide has the full breakdown.
Very. The FRM curriculum covers credit risk measurement — PD, LGD, EAD, credit VaR (value at risk) and counterparty risk — in real depth, and bank risk teams recognise the certification. Pair that theory with hands-on model building in Python or SAS, and you cover both halves of what interviewers test: concepts and implementation.
