PD/LGD/EAD Framework for Expected Credit Loss Modelling — A Technical Guide for Indian Banks and NBFCs

Expected Credit Loss modelling under Ind AS 109 is, at its core, a valuation exercise — it is the determination of the present value of expected cash shortfalls on a financial instrument over its contractual life, probability-weighted across multiple scenarios and forward-looking in its use of information. The three foundational parameters — Probability of Default (PD), Loss Given Default (LGD), and Exposure at Default (EAD) — are not accounting estimates in the traditional sense. They are credit risk model outputs that require statistical methodology, historical data calibration, forward-looking scenario construction, and professional judgment in their assembly and application. For Mumbai-based banks, housing finance companies, and large NBFCs reporting under Ind AS, the ECL model is both a regulatory capital tool and a financial reporting instrument, and its quality is scrutinised simultaneously by statutory auditors, the RBI inspection team, and NFRA’s oversight mechanism.

The Probability of Default measures the likelihood that a borrower will fail to meet its contractual obligations within a defined time horizon. For Stage 1 exposures — performing loans with no significant increase in credit risk since origination — the PD is calculated over a twelve-month horizon. For Stage 2 exposures — performing loans where credit risk has increased significantly since origination — the PD is calculated over the remaining life of the instrument, producing a lifetime PD that is substantially higher than the twelve-month PD for longer-duration instruments. The identification of significant increase in credit risk — the staging trigger — is the most consequential judgment in the entire ECL framework, because movement from Stage 1 to Stage 2 triggers a step-change from twelve-month to lifetime ECL that can dramatically increase provisioning requirements.

Building PD, LGD, and EAD Models That Withstand RBI Inspection and Auditor Scrutiny

In practice, staging is governed by a combination of quantitative triggers — typically a specified deterioration in PD since origination, or crossing an absolute PD threshold — and qualitative indicators such as restructuring, forbearance, or watch-list status. RBI’s asset classification framework (SMA-0, SMA-1, SMA-2, NPA) provides a regulatory overlay that Indian banks and NBFCs must reconcile with the Ind AS 109 staging criteria. The reconciliation is non-trivial: a loan can be in SMA-1 (31-60 days overdue) under RBI classification while simultaneously being in Stage 1 under Ind AS 109 if the credit risk has not increased significantly since origination, or in Stage 2 if it has. Conversely, a restructured loan that is current on payments may be in SMA-0 under RBI classification but in Stage 2 under Ind AS 109 because the restructuring itself is evidence of significant credit deterioration. Managing this dual classification consistently and defensibly is a major operational challenge for Mumbai-based lenders.

The Loss Given Default measures the portion of the exposure that will not be recovered after the borrower defaults, expressed as a percentage of the exposure at default. LGD is driven primarily by collateral value and enforceability, the seniority of the lender’s claim in the borrower’s capital structure, and the legal and practical recovery timeline. For secured real estate loans — the dominant collateral class in Mumbai’s banking and NBFC sector — LGD modelling requires assumptions about the current and future market value of the collateral, the cost of enforcement under SARFAESI or IBC, the time to recovery, and the discount rate applied to the recovery cash flow. LGD for Mumbai residential mortgage collateral in a base case scenario might be in the range of 20-35%, but under a stress scenario — where property values have declined materially, enforcement is delayed by litigation, and the forced sale discount is elevated — LGD can approach 60-70%. The spread between base case and stress LGD is the primary driver of the sensitivity of the ECL model to macroeconomic conditions.

The Exposure at Default estimates the outstanding balance of the instrument at the point of default, including any undrawn committed facilities that the borrower is contractually entitled to draw down before defaulting. For revolving credit facilities, overdrafts, and partially drawn term loans, EAD requires a credit conversion factor that estimates the proportion of the undrawn commitment that will have been drawn at the point of default. For Indian corporate borrowers, the draw-down behaviour before default is an empirically observed phenomenon — borrowers under financial stress typically maximise utilisations of available credit lines before a default event — and the credit conversion factors applied in Indian ECL models must be calibrated to Indian observed default data rather than borrowed from international models without adjustment.

Forward-looking macroeconomic overlays are the most judgmentally intensive element of Ind AS 109 ECL modelling. The standard requires that ECL estimates incorporate forward-looking information — including macroeconomic scenarios — to the extent that such information is available without undue cost or effort. In practice, this means constructing multiple macroeconomic scenarios (typically three: base, optimistic, downside) with probability weights assigned to each, and calculating scenario-conditional ECL estimates that are then probability-weighted to produce the reported ECL. The macroeconomic variables most relevant to Indian bank and NBFC portfolios include GDP growth, unemployment rates, property price indices (particularly Mumbai Metropolitan Region indices for real estate-secured portfolios), benchmark interest rates, and sectoral stress indicators. The relationship between these macro variables and the PD, LGD, and EAD parameters in the ECL model — the transmission mechanism — must be modelled explicitly and calibrated to Indian historical data, which for many institutions requires constructing proxy relationships where direct historical evidence is limited.

The macroeconomic overlay construction for Indian banks and NBFCs presents a data challenge that distinguishes the Indian ECL modelling context from its international counterparts. In developed markets — the UK, the US, the Eurozone — banks can draw on decades of granular credit performance data, including through multiple credit cycles, to calibrate the statistical relationships between macroeconomic variables and PD and LGD parameters. In India, the credit data infrastructure is less mature. Credit bureau data from CIBIL, CRIF, and Experian covers the retail segment reasonably well, but corporate credit data suffers from definitional inconsistencies in default classification across regulatory periods, limited public information on recovery outcomes, and the confounding effect of regulatory forbearance measures — particularly around the COVID-19 restructuring frameworks — on the observed default rate time series.

The result is that Indian ECL modellers must exercise significant professional judgment in constructing the macroeconomic transmission mechanism — the quantitative relationship between, say, GDP growth and corporate PD — with less empirical backing than their counterparts in developed markets would enjoy. This judgment must be documented and defended in the ECL model governance framework. Where empirical calibration is not feasible from domestic data alone, a combination of international empirical evidence (adjusted for Indian structural differences) and expert judgment (from senior credit officers with Indian market experience) provides a defensible proxy approach.

The scenario design and probability weighting for the three-scenario ECL overlay is the step that attracts the most audit scrutiny in Indian bank and NBFC ECL models. Auditors assess whether the scenario definitions are genuinely distinct — whether the downside scenario differs materially from the base in its assumptions about GDP growth, interest rates, property prices, and credit spreads — and whether the probability weights assigned to each scenario are consistent with forward-looking market signals rather than calibrated to minimise provisioning. For Mumbai-based large NBFC boards and audit committees who must approve the scenario framework and weighting annually, the ECL governance process is a substantive risk management exercise, not a checkbox. The independent valuer or credit risk specialist who advises on ECL model methodology brings an outside view that helps prevent the model from being optimised around desired provisioning outcomes rather than honest assessment of forward credit risk.

At Harshul Mangal & Associates, our IBBI Registered Valuer practice (Reg. No. IBBI/RV/16/2025/16044) includes ECL model review and collateral LGD valuation support for banks and NBFCs undergoing Ind AS 109 implementation — providing the independent, professionally accountable assessment that statutory auditors require when testing management ECL estimates.

For further reading on the regulatory framework governing this area, refer to the RBI Scale-Based Regulation and Basel III framework for NBFCs, which provides the primary regulatory foundation for the analysis discussed here.

Our Due Diligence Services covers the full range of assignments described in this post. If you need professional valuation assistance, we would be pleased to assist. You can reach out to us here or write to harshulmangal.ca@gmail.com.

Model Validation and Governance for Ind AS 109 ECL Models

The model validation requirements for Ind AS 109 ECL models have become progressively more demanding as both regulators and auditors have developed deeper expertise in ECL methodology. RBI’s guidance on model risk management — which, while formally directed at banks, is increasingly referenced by auditors of large NBFCs — specifies that every quantitative model used for financial reporting must be subject to independent validation before it is deployed and periodically thereafter. Independent validation means validation by a party that had no involvement in model development — typically an internal model risk management function for large institutions, or an external validator for smaller entities that lack the internal capability. The validator must test the model’s conceptual soundness, data integrity, parameter estimation methodology, and output stability under a range of input perturbations.

The back-testing of PD and LGD estimates — comparing the model’s predictions against actual default experience in the subsequent period — is the empirical heart of ECL model validation. For Indian banks and NBFCs with limited historical default data, back-testing is necessarily constrained by sample size, and the results must be interpreted with appropriate statistical humility. A model that was calibrated on the pre-2016 data, before the RBI-mandated Asset Quality Review that reclassified large volumes of restructured loans as NPAs, will systematically understate PD for corporate exposures because the base period included a period of regulatory forbearance that compressed observed default rates. Recognising and correcting for this historical data bias is one of the most important exercises in ECL model validation for institutions with legacy models built on pre-AQR data.