New bad loan provision measures will tell banks apart under ECL regime

RBI's ECL regime will shift the focus from recognising bad loans to predicting them, making risk management a key differentiator for banks

The Reserve Bank of India (RBI) on April 27 issued a Master Direction on Expected Credit Loss (ECL) provisioning, ending three decades of rule-based provisioning. Banks until now provisioned in fixed proportions after a loan went bad, depending on how long it remained unpaid. Henceforth, banks must provision before they incur a loss. That means forecasting future losses from loan health, economic stress scenarios and recovery assumptions. Earnings may get hit, as the regulator scrutinises both the analytics and the judgmental overlay every quarter. 

The markets will start telling banks apart based on how well their provisions anticipate stress. Provisions that surge only after defaults materialise — or are released just before — raise questions no earnings call can answer clearly. This happens in public, every quarter. Boards need to own the provision number, not just sign off on it. Treating ECL as a compliance exercise is a risk in itself. 

While risk management, finance, compliance and analytics teams will be busy implementing the RBI’s norms, there are three things boards and chief executive officers need to ask them to ensure success. 

Is the Significant Increase in Credit Risk (SICR) threshold calibrated to Indian stress conditions? SICR triggers where a loan shifts from Stage I, which carries low risk and minimal provisioning, to Stages II and III, which carry substantially higher lifetime provisions. The 30-days-overdue approach is common and broadly compliant, but it is not the right answer across all loan types or conditions. European banks learned this during the pandemic when loosely calibrated SICR triggers produced provision surges and regulatory censure — the European Systemic Risk Board (ESRB) had flagged the vulnerability as early as 2019. India’s provisioning floor reduces some of that risk. The harder question for any board is whether its product-wise SICR thresholds would have triggered during the stress periods India went through over the past decade. 

Are we blindly following model outputs? The economic scenarios and probability weights driving ECL numbers need board buy-in. The ESRB found weights should hold steady through normal business cycles and shift only when something genuinely exogenous hits. Overlays on model outputs are legitimate, since no model captures everything. The problem is what happens without governance around them. An ECB review of 51 European banks in July 2024, at 95 per cent statistical confidence, found banks applied larger overlays when profits were strong and pulled back when under stress — exactly backwards. Several faced supervisory action. The failure was not modelling capacity but the absence of any rules when overlays got applied and when they came off. Boards that allow provisions to drift with each quarter’s commercial narrative add volatility without adding foresight. 

Do we have dedicated data and infrastructure capability to run this? ECL models need at least five years of loan-level data across a full credit cycle. The 2018-20 non-banking financial companies stress period and the pandemic restructuring wave both need to be in the dataset. A World Bank survey of International Financial Reporting Standards 9 supervisors found data quality was the single biggest implementation failure in every jurisdiction studied. India adds a complication. Unlike markets that adopted the Advanced Internal Ratings-based approach for capital estimation, Indian banks had little reason to build skills in loss-given-default, lifetime loss value, or macro-projection-based loss estimation. Such capabilities exist in pockets but are not yet widespread enough to deliver the provision precision markets will eventually price. 

Banks are in various stages of preparedness to meet the regulatory deadline of April 2027. While most are expected to meet the deadline, they should strive to assure regulators and markets of the precision, stability and quality of their provision measure. The provision number will increasingly be read as a proxy for the quality of the bank’s risk management and markets will price it accordingly.

The writers are, respectively, BCG’s managing director and senior partner (head of risk Asia Pacific), and principal, data science