The ECL Framework Can Foster Sustainable Asset Quality

Among critical factors, the sustainability of the banking system rests on asset quality. The risk-adjusted yield on the credit portfolio ultimately underpins bank profitability. It is worth asking why one bank lags behind others in managing asset quality when all other factors are identical and the credit risk management ecosystem is similar. The difference lies, among other things, in the quality of credit sourcing, impeccable appraisal methods, market intelligence, the intensity of technology-driven follow-up, timely recovery of dues, and promptness in supporting borrowers when business risks are beyond their control.

The test of a lender's credit risk management quality is how the loan asset performs over its lifetime and how predictable the credit risk is at different points in time, enabling the review of facilities and the setting of risk-based pricing that reflects changes in a borrower's credit risk profile.

This imperative for assessing future credit risk is enshrined in the expected credit loss framework, effective April 1, 2027, for which preparations to gather granular borrower data must begin now. The RBI has assessed the banking system's capacity in terms of data maturity, data integrity standards, and the level of technology required to adopt the ECL model. Before examining its details and implications, a brief look at how credit risk management tools have evolved will be useful.

Changing Classifications
During the era of manual banking, the RBI introduced the Health Code system in 1985 to classify credit risk across eight grades, from 'satisfactory' to 'loss'. Following the Narasimham Committee Report of 1991, international practices of income recognition, asset classification, and provisioning replaced it from 1992-93. The new norms revealed the true scale of the problem: the gross NPA-to-assets ratio of all public sector banks stood at 25% as of March 31, 1994, against 7.5% at end-March 2021

The substandard threshold moved from 180 days overdue in 1991-93, to 120 days in 1993-94, and has been fixed at 90 days since 2003-04. Provisioning follows: substandard assets attract 15%, rising to 25% on the unsecured portion; first-year doubtful assets attract 25% on the secured portion and 100% on the unsecured; those doubtful for one to three years attract 40% on the secured; beyond three years or classified as loss, 100%. Standard asset provisioning stands at 0.40%, with 0.25% for farm sector loans and 0.75% for commercial real estate.

As technology matured, external surveillance systems were introduced to reinforce credit risk assessment. TransUnionCIBIL was approved by the RBI as an External Credit Assessment Institution under Basel II and subsequently Basel III. Accredited ECAIs, including CRISIL, ICRA, India Ratings, Acuité CARE, and Brickwork Ratings India, provide ratings that banks use to assign risk weights for capital adequacy. CRILC, established in June 2014, requires institutions to notify the status of stressed borrowers to a central RBI database, serving as a reference point for borrower performance.

Banks were simultaneously required to develop internal monitoring systems for special mention accounts. SMA data is graded into SMA-0 (overdue 1-30 days), SMA-1 (31-60 days), and SMA-2 (61-90 days), after which accounts deteriorate into the substandard category, providing early warning of credit quality stress. Lenders must report SMA-2 accounts to CRILC within 30 days for exposures exceeding ₹10 million.

ECL Arrives
ECL is not a standalone reform. It is the culmination of four decades of progressively tightening asset quality standards, each triggered by the failures of the last. The health code system failed because it was subjective. IRACP was objective but backward-looking. ECL attempts to address the deepest problem: recognising risk before it becomes a loss, rather than after.

It does so by shifting away from the incurred-loss model, which recognised credit losses only after a default event, resulting in delayed provisioning and chronic underestimation of credit risk. Banks were essentially waiting for a borrower to fall before setting aside money for the loss, a reactive posture the 2008 global financial crisis exposed as dangerously inadequate.

ECL computation rests on three interlinked parameters: Probability of Default (PD), Loss Given Default (LGD), and Exposure at Default (EAD). Each captures a distinct element of credit risk; together they determine expected loss under both baseline and stressed macroeconomic conditions.

The ECL model divides credit exposures into three stages, each determining the time horizon for recognising expected losses and the quantum of provisioning. Stage 1 covers performing loans with minimal credit risk. Stage 2 applies when credit risk increases significantly. Stage 3 covers credit-impaired or defaulted assets. 

To guard against under-provisioning, the RBI has introduced product-wise prudential floors across asset classes including retail, corporate, MSME, agriculture, and real estate. Standard corporate and retail loans attract a minimum Stage 1 provision of 0.40%, rising to 5% in Stage 2, with materially higher provisioning mandated for Stage 3 assets depending on the duration of default.

ECL models must use representative datasets spanning at least one credit cycle, adjusted for current and forecasted macroeconomic conditions. Banks will need at least five years of historical data to run credible models. The RBI has also mandated robust governance: a committee of the board, including the CFO and CRO, must ensure data integrity throughout the ECL computation lifecycle and the complete independence of the internal model validation function.

Implementation Challenges
Implementing ECL is not merely a financial challenge. Banks will need robust historical data systems, granular data analytics, upgraded governance structures, and credit risk professionals capable of collecting and assessing that data, all entailing significant upfront costs, particularly for smaller banks.

The shift requires closer integration between finance and risk functions, and sustained investment in modelling capabilities. Banks may phase in the difference between ECL-based provisions and existing IRACP provisions over four years from 2027 to 2031, easing the capital impact. Most private-sector banks already carry contingent provisions; PSU banks do not, and the new Stage 2 minimum of 500 basis points, up from the current 40, will materially increase their annual provisioning run rate. That is the single most consequential near-term pressure point.

Poorly designed or overly optimistic model assumptions risk under-provisioning, defeating the purpose of the reform. Because the model is forward-looking, banks must make economic forecasts on growth, inflation, and unemployment, introducing subjectivity and consistency challenges across institutions.

The overall impact on capital adequacy is expected to remain manageable. But the ECL framework represents something more consequential than a technical upgrade to provisioning norms. For the first time, India's banking system will be required to price the future into the present. Banks that build that capability well will not merely comply; they will compete on its strength.