Lesson 10.3: RBI's expected credit loss framework and the 2027 transition
Learn how the RBI's expected credit loss rules work: three stages, provision floors, the 30-day trigger debate, scenarios and the 2027 transition.
By the end of this lesson you will be able to explain how the RBI's expected credit loss framework for banks differs from the old way of provisioning, place any loan in Stage 1, 2 or 3, calculate its provision from a model and from the regulatory floor, and weigh several economic scenarios into one number. You will also work out what the transition could cost a bank in capital, and know which parts of the rules are well established and which the public sources still disagree on. You should have read Lesson 1.1 (PD, LGD and EAD), Lesson 1.4 (defining default) and Lesson 10.2 on the general IFRS 9 staging model. The arithmetic is percentages and a little multiplication.
This lesson is built on research from public sources, listed at the end, and I have kept apart what several sources agree on from what only one source says. The RBI's own directions were not available to read while this lesson was prepared, so check any number you intend to rely on in the directions themselves. That caution is itself part of the lesson: regulation changes, secondary summaries disagree, and a credit analyst learns to go to the primary text.
The old rule and why it was changed
For many years, Indian banks provisioned by classification. A loan was standard, sub-standard, doubtful or loss, depending mainly on how long it had been overdue. A standard loan carried a small fixed provision, of about 0.4 per cent in many categories. When a loan crossed 90 days overdue it became a non-performing asset, or NPA, and the provision jumped. This is called an incurred-loss approach, because the bank provisions after a loss event has occurred.
The weakness is timing. A loan that is 85 days past due, in a borrower whose business has collapsed, and a loan that has never been late, both sat in the standard category with the same small provision. The big increase came at day 90, when the damage was already done. In a downturn, banks therefore recognised losses late and all at once, which is the worst moment to lose capital.
The expected credit loss approach, or ECL, moves the provision forward. The bank sets aside money for the losses it expects, loan by loan, as risk rises and not only after default. The RBI Governor said on 1 October 2025 that the framework was proposed from 1 April 2027, a date that several outlets report was kept in the final directions issued in April 2026. The three-stage design follows IFRS 9, the international accounting standard in force since 2018, which you met in Lesson 10.2.
The three stages
Every loan sits in one of three stages, and the stage decides both the length of the loss to be provided for and the size of the provision.
Stage · Meaning · Provision covers · How a loan gets thereStage 1 · No significant increase in credit risk since the loan was made · Expected loss over the next 12 months · Performing loans
Stage 2 · Significant increase in credit risk, but not yet defaulted · Expected loss over the whole remaining life · Triggered by a rise in risk, such as days past due or a PD that has moved up
Stage 3 · Credit-impaired · Lifetime expected loss, the loan treated as defaulted · Broadly the existing NPA test of 90 days past due
Two points are worth fixing in mind. First, the horizon changes at Stage 2. A Stage 1 loan needs the loss expected in the next year. A Stage 2 loan needs the loss expected until it is fully repaid, which for a three-year loan is three years of risk, and for a long home loan, far more. That jump in horizon is the main reason provisions rise sharply at the stage boundary. Second, Stage 3 stays tied to the 90-day rule. One source states plainly that the 90-day test for non-performing status is unchanged, and that the new checkpoint at Stage 2 affects provisioning only.
The trigger for Stage 2: a debate to follow
How does a loan get into Stage 2? Here the public sources do not agree, so this is worth studying as a case of reading evidence carefully.
Uniqus, an advisory firm, states that for term loans 30 days past due is a rebuttable presumption of a significant increase in credit risk, and for revolving facilities the presumption is 60 days of continuous out-of-order status. "Rebuttable" means the bank may argue the presumption does not apply to a particular loan, but the rebuttal must be approved by the board, backed by evidence and applied consistently. If the evidence is inadequate, the bank must stop rebutting.
JM Financial, a financial services firm, says Stage 2 will be "largely" 60 to 90 days overdue and gives no explicit 30-day trigger.
Both agree that qualitative signals also count: a rating downgrade, a change in pricing, or a worsening macroeconomic picture can move a loan to Stage 2 without any missed payment.
I cannot say from the material available which description matches the final directions exactly. What I can show is why it matters, using the example portfolio from Lesson 1.2, in which 4.4 per cent of loans are more than 30 days past due, 2.5 per cent more than 60 and 1.2 per cent more than 90. If Stage 2 begins at 30 days, it holds the loans between 30 and 90 days, which is 4.4 − 1.2 = 3.2 per cent of the book. If it begins at 60 days, it holds only 2.5 − 1.2 = 1.3 per cent. A different trigger puts two and a half times as many loans into the expensive stage.
From stage to provision
A provision under ECL is the expected loss, computed as you learned in Lesson 1.1: PD × LGD × EAD. The stage decides which PD to use. For Stage 1, the PD is for the next 12 months. For Stage 2, the PD covers the loan's remaining life. For Stage 3, the loan has defaulted, so the PD is 100 per cent and the provision is the expected loss given default on the outstanding amount.
Take an unsecured retail term loan with ₹8 lakh outstanding and an LGD of 55 per cent. These numbers are illustrative.
Stage 1. The 12-month PD is 3 per cent. ECL = 0.03 × 0.55 × ₹8,00,000 = ₹13,200, which is 1.65 per cent of the exposure.
Stage 2. The loan's risk has risen. The lifetime PD over the remaining life is 14 per cent. ECL = 0.14 × 0.55 × ₹8,00,000 = ₹61,600, which is 7.7 per cent of the exposure.
Stage 3. The loan has defaulted, PD is 100 per cent. ECL = 1.00 × 0.55 × ₹8,00,000 = ₹4,40,000, which is 55 per cent of the exposure.
Predict first: by how many times does the provision rise when this loan moves from Stage 1 to Stage 2? Compare ₹13,200 with ₹61,600 before you read on.
The provision rises 4.7 times. Nothing about the borrower's willingness to repay changed overnight. The increase comes from a longer horizon and a higher PD, and it shows why the stage boundary is the most sensitive point in the whole framework.
Floors: the regulator's backstop
Models can be wrong, optimistic or untested. To guard against that, the RBI sets floors, minimum provisions that a bank must hold whatever its model says. The provision is the higher of the model's figure and the floor.
The reported floors, from Uniqus's reading of the final directions' product table, are as follows. Stage 2 floors are 5 per cent for corporate, MSME, retail (secured and unsecured), farm credit and loans to banks and non-bank lenders. Lower Stage 2 floors apply to housing loans and gold loans at 1.5 per cent, loans secured by commercial real estate at 2.5 per cent, and loans backed by term deposits at 0.4 per cent. Restructured loans still classed as standard carry 5 per cent in Stage 1 and 10 per cent in Stage 2. The current provision on standard assets, for comparison, is about 0.4 per cent. CRISIL describes the Stage 1 floors as broadly similar to the existing norms.
The floor works differently depending on the product. For our unsecured loan, the Stage 2 floor of 5 per cent of ₹8 lakh is ₹40,000. The model's figure of ₹61,600 is higher, so the model binds and the bank provides ₹61,600. If the bank's model had produced only ₹24,000 (3 per cent), the floor would bind and the provision would be ₹40,000. The floor therefore acts as a quality check on the models: when it binds often, that is a signal that the model is too optimistic or the book is safer than the regulator assumes.
Weighing economic scenarios
The old approach looked backwards, at how long loans had been overdue. ECL must be forward-looking. JM Financial reports that banks must use probability-weighted estimates across several macroeconomic scenarios, such as a base, an optimistic and a stress case. The reasoning is that a single forecast can be wrong in either direction, and an average over plausible futures is a better estimate of the loss than any one path.
The method is a weighted average. Take the Stage 1 loan above, with the 12-month PD of 3 per cent in the base scenario. Suppose the bank defines three scenarios with weights and PDs as follows.
Scenario · Weight · 12-month PDBase · 55% · 3.0%
Upside · 15% · 2.2%
Downside · 30% · 5.5%
The weighted PD is 0.55 × 3.0 + 0.15 × 2.2 + 0.30 × 5.5 = 1.65 + 0.33 + 1.65 = 3.63 per cent. The ECL is 0.0363 × 0.55 × ₹8,00,000 = ₹15,972, against ₹13,200 on the base case alone, an increase of 21 per cent. The downside scenario carries more weight than the upside, and bad outcomes hurt more than good ones help, so the weighted figure is above the base.
Scenario work is where credit risk meets economics. The bank needs forecasts for variables such as growth, inflation and interest rates, and a model that links them to PDs. Lesson 10.4 builds that link. For now, notice that the choice of weights is a judgement, and regulators expect it to be documented and reviewed. A bank that always puts a large weight on the upside will provide too little in bad times.
What the transition costs
All of this adds up to a one-time change when the framework starts on 1 April 2027, as the rules replace the old provisions. CRISIL estimates a one-time net impact of up to 120 basis points on the banking system's core capital ratio, from a gross impact of up to 170, the gap being provisions banks already hold. Banks can spread the cost over four fiscal years, setting the extra provisions against retained earnings.
To see what 120 basis points means, take a stylised bank with risk-weighted assets of ₹10,00,000 crore and a core capital ratio of 14 per cent, so core capital is ₹1,40,000 crore. A fall of 120 basis points, or 1.2 percentage points, is 1.2 per cent of ₹10,00,000 crore, which is ₹12,000 crore of capital. Spread evenly over four years, that is ₹3,000 crore a year, or 30 basis points of risk-weighted assets a year. If the bank earns and retains enough, the hit is absorbed without cutting its ratio below the minimum. If it cannot, it must raise capital or slow lending.
How big is the Stage 2 part of that cost for a typical bank? CRISIL notes that loans in the early overdue buckets are only about 2 to 2.2 per cent of the banking system, so the step from about 0.4 to 5 per cent affects a small share of the book. For a ₹1,00,000 crore loan book, with 2.1 per cent in Stage 2, the extra provision is 1,00,000 × 0.021 × (0.05 − 0.004) = ₹96.6 crore, about 9.7 basis points of the book. A bank with 8 per cent in Stage 2 would need about 37 basis points. The deep dive on this site, "RBI's expected credit loss rules: what changes for banks", works through this in a simulator.
Two further points from the sources. CRISIL says credit costs are likely to rise structurally to some extent, because delinquent assets that are not yet NPAs now carry higher provisions. And some of the transition cost may be offset by capital released through revised risk weights that apply from the same date, which the reported figures do not include.
What the framework demands from analysts
The staging rule, the floors and the scenarios all require data and models that many banks are still building. Uniqus lists several challenges, and each one corresponds to a topic in this track.
Day-1 fair valuation. Banks must value their entire loan book at fair value on 1 April 2027, with the difference taken to opening retained earnings. Uniqus calls it the weakest area of preparedness.
Point-in-time PD. The provision should reflect today's conditions, so a through-the-cycle PD must be adjusted or replaced. Lifetime curves need default data across a full credit cycle, which many banks do not have.
LGD from scattered recoveries. Recovery data is held in legal, restructured and written-off pools. Workout recoveries for corporate loans can take over five years. Proxy curves and external benchmarks will be needed at first.
EAD for undrawn limits. The rules cover off-balance-sheet exposures such as unused limits, and standard conversion factors are unlikely to be enough for revolving facilities.
A borrower-level view. Stage 3 is judged per borrower, so a bank with separate product systems needs a combined view before go-live.
Governance. Independent model validation, a model inventory, and overlays documented as formal adjustments and not as informal memos.
If you recall Lesson 1.2, none of this is new in kind. It is the same loan data, the same definition of default and the same PD, LGD and EAD, now with a legal deadline. This is why the framework matters so much to anyone building a career in credit risk.
A staging rule in code
A first version of a staging rule can be written in a few lines. It is only a screen. A real bank adds PD-based triggers and judgement.
python
import numpy as np
import pandas as pd
loans = pd.readcsv("loans.csv") # columns: loanid, dpd, producttype, ead, ratingdowngrades
def stage(row):
if row["dpd"] >= 90:
return 3
# 30 days past due is a rebuttable presumption for term loans (as reported)
if row["dpd"] >= 30 or row["ratingdowngrades"] >= 2:
return 2
return 1
loans["stage"] = loans.apply(stage, axis=1)
floors = {1: 0.004, 2: 0.05} # reported floors for most loans; check the directions
loans["floorprov"] = np.where(loans["stage"] == 3, np.nan,
loans["stage"].map(floors) loans["ead"])
loans["stage"].valuecounts(normalize=True) # share of the book in each stage
The last line is the first thing a risk committee will ask for: what share of the book is in each stage, and how is that changing month by month? Pair it with the vintage and migration methods from Module 4, which show which loans are likely to cross the boundary next.
Common mistakes
Treating the floor as the provision. The floor is a minimum. If the model gives a higher figure, the model decides.
Confusing the 30-day Stage 2 checkpoint with the 90-day NPA rule. Reports say the 90-day definition of an NPA is unchanged. Do not describe a Stage 2 loan as an NPA.
Using a through-the-cycle PD for Stage 1 and Stage 2 without adjustment. Provisions should reflect current conditions, so the PD must be point-in-time or scaled with evidence.
Forgetting off-balance-sheet exposure. Unused limits and guarantees now attract provisions.
Choosing scenario weights to hit a number. Weights are judgements that must be documented and reviewed. A habit of optimistic weights understates provisions.
Relying on a secondary summary. Numbers from commentary can differ from the directions. Check the primary text.
Key takeaways
Under the old incurred-loss approach, provisions followed the days-past-due class and jumped at 90 days. Under ECL, provisions build as risk rises, before default.
There are three stages: 12-month expected loss in Stage 1, lifetime expected loss in Stage 2, and the defaulted loan in Stage 3, which stays tied to the 90-day rule.
A reported 5 per cent floor applies to most Stage 2 loans, up from about 0.4 per cent, with lower floors for housing, gold and some secured loans. The provision is the higher of the model and the floor.
Sources differ on whether 30 days past due is a rebuttable trigger for Stage 2, and the choice changes the size of Stage 2 severalfold.
Provisions must be forward-looking, weighted across economic scenarios, so the weights and the scenario PDs are judgements that need governance.
CRISIL estimates a one-time net hit of up to 120 basis points on banks' core capital, spread over four fiscal years from 1 April 2027, with a structural rise in credit costs afterwards.
Check yourself
1. A loan has a 12-month PD of 2 per cent, an LGD of 50 per cent and an exposure of ₹6 lakh. What is the Stage 1 provision?
0.02 × 0.50 × 6,00,000 = ₹6,000.
2. A loan of ₹8 lakh moves to Stage 2, where the floor is 5 per cent. What is the floor in rupees, and how much more is it than the 0.4 per cent standard-asset provision?
5 per cent is ₹40,000 and 0.4 per cent is ₹3,200, so the floor is ₹36,800 higher.
3. A term loan is 35 days past due and the bank has no approved rebuttal. Which stage does the reported presumption put it in, and is it an NPA?
Stage 2. It is not an NPA, because that still needs 90 days past due.
4. A bank weights a 3 per cent PD scenario at 70 per cent and a 6 per cent PD scenario at 30 per cent. What is the weighted PD?
0.70 × 3 + 0.30 × 6 = 2.1 + 1.8 = 3.9 per cent.
5. A model gives a Stage 2 provision of 3 per cent and the floor is 5 per cent. What does the bank provide, and what does the gap suggest?
It provides 5 per cent, because the floor binds. A model that falls below the floor is probably too optimistic.
Try this
Build a mini ECL calculation for a small imaginary book. Set aside forty-five minutes.
1. Invent 12 loans with an outstanding balance, days past due, a product type and a 12-month PD. Make at least two loans over 30 days past due and one over 90.
2. Assign each loan a stage using the screen in the simulator, and note any loan where you would argue for a rebuttal and why.
3. Compute the provision for every loan using PD × LGD × EAD, with a lifetime PD for Stage 2 loans that you choose, and an LGD for each product. Then apply the floor and take the higher figure.
4. Add up the provisions by stage. Compare the total with what the old rule would have required, using 0.4 per cent for standard loans and a higher figure of your choice for the NPA loan.
5. Define three scenarios with weights and recompute one Stage 1 loan with the scenario simulator. Write down how much a heavier downside weight changes the total.
6. Write a short note to a risk committee: where the book sits by stage, what drives the largest provisions, and one judgement you made that could change the result.
A good result is a one-page table and a short note that a colleague could audit, with every assumption labelled as illustrative. The next lesson looks at the modelling behind the PD, building point-in-time PDs, lifetime curves and scenario links.
Sources
- [Business Standard: ECL framework proposed to be implemented from April 1, 2027, RBI Governor](https://www.business-standard.com/amp/finance/news/ecl-framework-proposed-to-be-implemented-from-april-1-2027-rbi-guv-1251001005111.html)
- [CRISIL Ratings: For banks, ECL norms to have one-time net impact of up to 120 bps](https://www.crisilratings.com/en/home/newsroom/press-releases/2026/04/for-banks-ecl-norms-to-have-one-time-net-impact-of-up-to-120-bps.html)
- [Uniqus: The implementation reality check](https://uniqus.com/the-implementation-reality-check/)
- [JM Financial: RBI unveils final ECL framework for banks](https://www.jmfinancialservices.in/blogs-and-articles/rbi-unveils-final-ecl-framework-for-banks)
Tags: Credit Risk Analytics Expected credit loss RBI Provisioning Regulation