ECL Calculation Overview
ECL (Expected Credit Loss) is a forward-looking (forecasting method) measure of potential losses a lender might incur due to a borrower's default on a financial obligation. It is part of credit risk modelling, introduced under IFRS 9 (International Financial Reporting Standard 9) to ensure timely recognition of credit losses in financial statements.
Purpose: Ensure institutions maintain adequate provisions for future defaults, thereby improving financial stability.
Frequency: ECL is calculated quarterly, based on both short-term (12 months) and lifetime horizons, depending on the risk stage of the exposure.
Mandate: IFRS 9 requires quarterly auditing of ECL.
ECL is calculated as:
ECL = EAD × PD × LGD
EAD - Exposure at Default
Is the outstanding amount at the time of default. It quantifies the amount of money that is at risk of being lost if the borrower defaults at a specific point in time. It provides a snapshot of the lender's total exposure.
This value is an amount
PD – Probability of Default
Estimates the likelihood of a customer defaulting.
This value is in probabilities i.e. the PD will be given as percentage value.
LGD – Loss Given Default
Reflects the loss incurred after accounting for recoveries.
Thus this value is also as a percentage calculated as 1 minus Recovery Rate.
Take for example below computation:
EAD = 10000 Cr INR
PD = 5% = 0.05
RecoveryRate = 60% = 0.60
LGD = 1 – 0.60 = 0.40
ECL = 10000 x 0.05 x 0.40 = 200 Cr INR
Stages for ECL
Under IFRS 9, ECL calculations are categorized into three stages based on Days Past Due (DPD).
Stage Definitions and Criteria:
ECL calculation is divided into three stages:
Stage 1: Customers with 0-29 DPD. Approximately 92-95% of customers fall into this category.
Stage 2: Customers with 30-89 DPD. This stage typically comprises around 4% of customers.
Stage 3: Customers with 90 or more DPD. ECL for Stage 3 is calculated by multiplying the Stage 3 EAD by the LGD. Since customers are already defaulted there is no need for calculating PD here. Approximately 1% of customers transition from Stage 2 to Stage 3.
Below is the example for calculating Stage wise PD:
PD Calculation Methods
Point-in-Time (PIT):
Uses data from the past typically 12 months.
Focuses on short-term changes in credit risk.
Through-the-Cycle (TTC):
Utilizes data from multiple historical multiple random windows (3-5 years).
Captures long-term credit risk patterns.
Weighted PD Example (Stage 1 PIT Calculation)
Loss Given Default (LGD) Calculation
Applies consistently across products or portfolios.
It is essentially dependent on expected recovery rates and its corresponding costs.
Formula:
Further Drill Down of PD using BSCORE Segmentations
Purpose: A Behavioural Scorecard is used to assess customer creditworthiness based on behaviour and historical data. Customers are segmented into bands (e.g., A, B, C, D, E) based on their risk levels. This segmentation based on behaviour scoring refines PD calculations. We Categorize Stage 1 customers into risk bands (A to E) and Stage 2 based on DPD buckets.
Stage 1 Segmentation Process:
Data Source: Combine on-book and off-book data.
Assigning Customer Segmented Bands from Behavioural Score Card:
A (Low Risk)
B, C, D (Medium Risk)
E (High Risk)
PD Calculation for Risk Bands:
Identify customers who default (90 DPD) within each grade.
Calculate defaulted exposure as a proportion of total exposure for that grade.
Compute a weighted average PD across all grades.
Stage 2 Segmentation:
Further drilled down by Segmenting into:
30-59 DPD (Early Delinquency)
60-89 DPD (Late Delinquency)
Calculate transition rates to default for each range.
Final ECL Calculations:
Strategies for Reducing ECL
Summary
ECL Components: PD, LGD, and EAD form the core calculation.
Stage 1-3 Segmentation:
Stage 1 (~92%): Low-risk customers (0-29 DPD).
Stage 2 (~4%): Medium-risk customers (30-89 DPD).
Stage 3 (~1%): High-risk customers transitioning to default (≥90 DPD).
Leverage Behavioural Score Card for Further Segmentation.
Reduction Strategies: Effective monitoring, borrower engagement, and leveraging advanced analytics significantly reduce ECL.
Comments