Macroeconomic scenarios in ECL models are how a lender brings the future into the provision. Instead of relying on past losses alone, the model projects PD, LGD and EAD under several economic paths, usually base, upside and downside, then weights the results by probability. That weighted figure becomes the allowance. The same scenarios also feed SICR assessment, so they affect both how much you provide and which stage a loan sits in.
Picture a provisioning team in the last week of a quarter. The economics desk revises the GDP forecast downward. Meanwhile, unemployment assumptions move. Then someone asks whether the Stage 2 numbers will change, and someone else asks whether the board pack is already printed. That scramble is what forward-looking provisioning looks like in practice.
It’s also about to get more common in India. RBI’s final expected credit loss directions for banks take effect on 1 April 2027, which is now under six months away. NBFCs reporting under Ind AS have lived with this for years, but banks are about to join them.
This post covers how macroeconomic scenarios in ECL models work, how they connect to staging, and what Indian and international regulators actually expect.
The old incurred-loss approach waited for evidence of a loss. IFRS 9, however, asks a different question: given what we know today, including forecasts, what do we expect to lose?
Paragraph 5.5.17 of IFRS 9 is the anchor. It says an ECL estimate must reflect an unbiased, probability-weighted amount found by evaluating a range of possible outcomes. It must also reflect the time value of money. Finally, it must use reasonable and supportable information, available without undue cost or effort, about past events, current conditions and forecasts of future economic conditions.
That last phrase is why scenarios exist. A model fitted only on last year’s defaults will tell you what happened, but it won’t tell you what a rate cycle, a fuel price spike or a slowdown in hiring is likely to do next.
There’s a second reason that gets less attention. A single forecast, even a good one, isn’t a range of outcomes. Credit losses don’t rise and fall in a straight line with the economy, so one “most likely” path tends to understate the average loss. We’ll come back to that in the worked example.
You don’t need thirty variables to build macroeconomic scenarios in ECL. You need a handful that have a defensible link to defaults in your own portfolio, and a forecast you can actually source.
A useful test for each candidate:
The table below shows common starting points. These are hypotheses to test on your data, not a rule set.
| Portfolio | Variables often tested | Why they’re tested |
| Home loans | Interest rates, house price index, unemployment | Affordability and collateral value |
| Unsecured retail | Unemployment or wage growth, inflation, GDP growth | Household cash flow stress |
| MSME and SME | GDP growth, credit growth, sector output indicators | Revenue and working capital pressure |
| Corporate | GDP growth, commodity or input prices, sector indicators | Margin and cash flow sensitivity |
One practical warning. India has relatively short and noisy default histories for some segments, and the pandemic years distort almost everything. So if a variable only “works” because of 2020, treat it with suspicion.
Most of the effect usually comes through PD, but you shouldn’t ignore the other two.
LGD. Recoveries move with the cycle. Collateral values fall in downturns and workouts take longer. Beta regression is a common choice for modelling LGD with macro drivers, because LGD is bounded between 0 and 1.
EAD. For term loans this is mostly contractual. For revolving products, such as credit cards and overdrafts, drawdown behaviour near default changes with stress, so credit conversion factors can be scenario-dependent.
A fair question is whether a regression-based satellite model is overkill for a small portfolio. Sometimes it is. In that case, a simpler approach, such as scalar adjustments to PD calibrated on a stable relationship, can be easier to govern. Even so, what you can’t do is skip the link entirely and hope the overlay covers it.
Most lenders run a base case, an upside and a downside. Some add a severe downside to make sure the tail is represented. IFRS 9 doesn’t prescribe a number, but it asks for a range that’s unbiased and reflects the possible outcomes.
Three design choices matter more than the number of scenarios:
Weights are the uncomfortable part. Assigning 50% to base, 20% to upside and 30% to downside is a judgement, and reasonable people will disagree. My view is that the honest approach is to document the logic, show a sensitivity table (what happens to ECL if the downside weight moves by 10 points), and revisit weights each reporting cycle. A bank that can defend its weights is in a better position than one that claims false precision.
This example is ILLUSTRATIVE. The numbers are invented to show the arithmetic, not taken from any portfolio.
Take a single Stage 2 loan with exposure at default of ₹10,00,000. To keep it simple, ignore discounting. Suppose the model produces these scenario inputs:
| Scenario | Weight | Lifetime PD | LGD | ECL (EAD × PD × LGD) |
| Base | 50% | 6% | 40% | ₹10,00,000 × 6% × 40% = ₹24,000 |
| Upside | 20% | 4% | 35% | ₹10,00,000 × 4% × 35% = ₹14,000 |
| Downside | 30% | 12% | 50% | ₹10,00,000 × 12% × 50% = ₹60,000 |
Probability-weighted ECL:
Two comparisons are worth making.
First, the base case alone gives ₹24,000. The weighted figure is ₹8,800 higher, roughly 37% more. The downside scenario is not just more likely to hurt, it hurts disproportionately, and a single-scenario approach misses that.
Second, a shortcut some teams try is to average the inputs first. Weighted PD is 7.4% (0.5 × 6 + 0.2 × 4 + 0.3 × 12). Weighted LGD is 42% (0.5 × 40 + 0.2 × 35 + 0.3 × 50). Multiplying gives ₹10,00,000 × 7.4% × 42% = ₹31,080. That’s ₹1,720 lower than the correct answer, because PD and LGD both rise together in the downside. The right order is to compute ECL per scenario, then weight.
This is the part many ECL articles skip, and it’s where scenarios and staging meet. For a full treatment, see our module on macroeconomic scenarios and SICR assessment. [LINK: IFRS 9 course page, anchor #macroeconomic-scenarios-and-sicr]
Stage 2 applies when credit risk has increased significantly since origination. IFRS 9 doesn’t define “significant”, but it does include a rebuttable presumption that a loan more than 30 days past due has crossed the line (paragraph 5.5.11). Most lenders also use quantitative triggers, such as a lifetime PD comparison against origination, and qualitative ones such as watch-list status.
Here’s the link. If your SICR test compares the lifetime PD now with the lifetime PD at origination (IFRS 9 also allows a 12-month PD as an approximation where it’s a reasonable proxy), and that PD is scenario-weighted, then a weaker outlook raises the current PD. As a result, more loans cross the threshold. They then move from 12-month ECL to lifetime ECL. Your allowance rises through two doors at once: higher PDs, and more loans measured over a longer horizon.
Because of this, a cliff effect can appear. A small change in the forecast leaves a cluster of loans just past the trigger, so the provision jumps. Two things help:
The frameworks agree on the principle and differ on the detail. Here’s the short version, followed by a comparison table.
IFRS 9 sets the core requirement. Paragraph 5.5.17 asks for unbiased, probability-weighted estimates that use reasonable and supportable forward-looking information. Staging and the 30-days-past-due presumption sit in paragraphs 5.5.9 to 5.5.11.
Basel Committee (BCBS). The 2015 Guidance on credit risk and accounting for expected credit losses sets supervisory expectations for banks applying ECL accounting. It covers governance, validation and the use of forward-looking information, including macroeconomic factors, in credit risk practices.
EBA (European Union). The EBA’s Guidelines on credit institutions’ credit risk management practices and accounting for expected credit losses (EBA/GL/2017/06) translate this into expectations for EU banks. They push institutions to consider multiple forward-looking scenarios, paying particular attention to the non-linear relationship between scenarios and credit losses, and they also cover governance and documentation.
US CECL (ASC 326). The US standard takes a different route. It recognises lifetime expected losses from day one, so there are no stages and no SICR test. Forecasts are used over a reasonable and supportable period, after which estimates revert toward historical experience. The standard neither requires nor prohibits multiple scenarios, so some US lenders run a single forecast and others weight several.
India has two tracks, and until now they’ve been separate.
Ind AS 109 is closely aligned with IFRS 9, so the same ECL model, staging and forward-looking requirements apply. Ind AS has applied to NBFCs and asset reconstruction companies above RBI’s applicability thresholds since 1 April 2019. RBI’s 13 March 2020 guidance for Ind AS implementation by NBFCs and ARCs (RBI/2019-20/170) adds a prudential layer. Entities hold impairment allowances as Ind AS requires, but they also compute provisions under RBI’s asset classification norms. Where the Ind AS allowance is lower, the shortfall is appropriated from profit into an “Impairment Reserve”, which doesn’t count toward regulatory capital.
Scheduled commercial banks have so far provisioned under RBI’s income recognition, asset classification and provisioning (IRAC) norms. These are driven by days past due and asset classification rather than forecasts. Those norms stay in force until the new ECL Directions take effect.
RBI’s regulatory package of 27 March 2020 allowed a payment moratorium. For Ind AS lenders, the open question was whether moratorium accounts should automatically move to Stage 2. The IASB noted in March 2020 that IFRS 9 doesn’t require a mechanistic approach to SICR, and the Basel Committee’s April 2020 statement stressed the same flexibility. As a result, lenders had to reassess PD and LGD for pandemic impact instead of applying a mechanical rule.
After a draft in October 2025, RBI issued the final Reserve Bank of India (Commercial Banks – Asset Classification, Provisioning and Income Recognition) Directions, 2026 on 27 April 2026, effective 1 April 2027. In summary, they:
The Directions define expected credit loss as the weighted average of credit losses under different scenarios, with each scenario’s probability as the weight. So scenario weighting is part of the final text, not just the draft.
| Framework | Who it applies to | Forward-looking requirement | Staging approach | Status / notes |
| IFRS 9 | IFRS reporters globally | Unbiased, probability-weighted, reasonable and supportable information (para 5.5.17) | 3 stages: 12-month vs lifetime ECL | In force; supervisors add guidance (BCBS, EBA) |
| US CECL (ASC 326) | US GAAP reporters | Reasonable and supportable forecasts, then reversion | None; lifetime ECL from day one | In force; multiple scenarios permitted, not required |
| Ind AS 109 | Ind AS-reporting entities, including NBFCs and ARCs above applicability thresholds | Same as IFRS 9 | 3 stages | In force; RBI requires Impairment Reserve if Ind AS ECL is below IRACP provisions |
| RBI ECL Directions | Scheduled commercial banks (certain categories excluded) | Forward-looking; probability-weighted across scenarios | 3 stages with product-wise prudential floors | Issued 27 Apr 2026; effective 1 Apr 2027 |
| RBI existing IRAC norms (banks, until transition) | Scheduled commercial banks | Not forecast-based | Asset classification by days past due | Replaced by the ECL Directions from 1 Apr 2027 |
The practical takeaway for Indian banks: scenario design, independent forecasts, weighting and documentation, which were nice-to-haves, become audit and supervisory topics.
Most ECL problems that reach auditors and supervisors aren’t mathematical. Instead, they’re about who decided what, and whether it’s written down.
A workable governance setup includes:
Long-running overlays are a warning sign. For example, if an adjustment has been in place for six quarters, the model probably needs fixing.
If you’re building or validating these models, the useful skills sit across macroeconomic variable selection, PD, LGD and EAD modelling, staging and SICR, and scenario weighting. DexLab’s Advanced Certificate in IFRS 9 Modeling covers these areas, including a final ECL case study using anonymised bank data. For the section most relevant to this post, see macroeconomic scenarios and SICR assessment
Related reading: PD estimation methods , What is SICR? , Expected credit loss calculation .
What are macroeconomic scenarios in ECL models? They’re internally consistent forecasts of variables such as GDP growth, unemployment and interest rates, used to project how PD, LGD and EAD may behave. Lenders typically run a base case plus upside and downside paths, estimate ECL under each, and weight the results by probability to get one allowance.
How many scenarios does IFRS 9 require? IFRS 9 doesn’t set a number. It requires an unbiased, probability-weighted amount reflecting a range of possible outcomes (paragraph 5.5.17). Three scenarios is a common starting point. The real test is whether your set captures the non-linear way losses rise in a downturn.
What is the difference between Ind AS 109 and RBI norms? Ind AS 109 uses a forward-looking, three-stage ECL model for Ind AS-reporting entities such as eligible NBFCs. RBI’s existing norms for banks rely on asset classification and days past due. NBFCs compare the two and park any Ind AS shortfall in an impairment reserve. Banks move to ECL from 1 April 2027.
How does IFRS 9 differ from CECL? Both are forward-looking, but IFRS 9 uses three stages: 12-month ECL for performing loans and lifetime ECL after a significant increase in credit risk. CECL, under US ASC 326, recognises lifetime expected losses from day one with no staging. IFRS 9 also stresses probability-weighted outcomes explicitly.
How do macroeconomic scenarios affect Stage 2 classification? Scenarios change the projected lifetime PD, and SICR tests often compare that with PD at origination. A weaker outlook can push more loans past the threshold, moving them from 12-month to lifetime ECL. That’s why scenario choices affect staging, not only the size of the allowance.
What are management overlays in ECL? Overlays are adjustments to model output when the model can’t capture a risk, such as a new shock or a data gap. They need a documented rationale, an approver, a quantified size and a plan to remove them or build them into the model. Left open-ended, they hide model weakness.
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.