Estimator 9 (IFRS 9 ECL)

1. portfolio segmentation

Receivables’ portfolio is segmented on the basis of both empirical judgement as well as statistical characteristics estimated through statistical moments of default rates’ pattern.

Receivables from counterparties based in different countries must be segmented separately as their default rates will be correlated to the macro-economic parameters of their domiciled countries.

Segments having similar range of statistical moments i.e. mean, standard deviation, skewness and kurtosis are merged.

2. MEVS IDENTIFICATION

Historical data of maximum number of relevant macro-economic variables is taken and tested for Correlation & Concordance with portfolios’ historical default rates.

Macro-economic variables that qualify minimum thresholds of correlation & concordance are taken for while other non-qualifying variables are discarded.

Qualified macro-economic variables are further tested for R-Square (Model Fit) and P-Values (Predictive Capability) for default rates pattern through regressive analytics.

Macro-economic variables that qualify minimum thresholds of model fit and predictive capability are taken while other variables are statistically rejected.

Identified macro-economic variables are empirically back tested to ascertain statistical validity.

This step is conducted for each segment of receivables’ default rates as identified in portfolio segmentation process.

3. ECONOMIC SCENARIOS

Annual projected data for final MEVs is taken and YOY Intercepts and Slopes of annual data points are estimated. Annual data points are converted in quarterly data points through Interpolation using estimated YOY intercepts and slopes and YOY rate of changes is estimated for projected MEVs to Stationarize data.

Stationarized MEVs projected data is combined with historical standardized data points to create data stream. Factor Analysis is conducted on data stream to estimate Eigenvalues & Eigenvectors and Principal Components are identified using Eigenvalues & Eigenvectors through Principal Component Analysis.

Primary Principal Component is identified on the basis of Explained Cumulative Variance and Regressive Analytics are conducted to test Statistical Fit & Predictive Accuracy of DRs using Principal Components.

Base, Down and Up scenarios are identified by mapping movements of DR with the increase in Principal Component using Visual Cutoff Points on scree plot. Probabilistic weights are estimated for each scenario as Realized Probabilities based on Cutoff Points adjustments in historic default rate pattern.

Regressive Analytics are conducted to test Statistical Fit & Predictive Accuracy of DRs under base, down and up scenarios

Slopes & Intercepts of base, down and up scenarios are used to project DR changes under 3 scenarios (IFRS 9 requires minimum of 2 scenarios).

4. Pit pd simulations

The process of point in time probability of default requires combining non-parametric approach with parametric approach under IFRS 9 standard modeling.

Net Flow Rates are estimated from historical default rate pattern under non-parametric approach and adjusted for its standard deviation.

Economic overlay is created on historic net flow rates by combining macro-economic scenarios through exponential distribution which estimates point forward probabilities of default under each scenario.

Final point in time probabilities of default for each segment of receivables portfolio are estimated as weighted average of scenario’s probabilistic weight and scenario driven probability of default.

This is essentially a copula approach which substitutes transition matrix which is extremely data intensive otherwise.

5. lgd simulations

PIT collateral values are estimated by extrapolating the negative/positive Slope of valuation points over Euclidean Space for collateralized exposures using estimated accelerators.

Segment-wise total outstanding exposure and total collateral values are used to calculate Exposure Weighted LTVs for each portfolio segment.

Historic Recovery Rates are used to estimate Mean Portfolio Recovery Rates and Standard Deviation of Portfolio historic recovery pattern.

Lower Bound of LGD Distribution is fixed at 0% and Upper Bound is fixed at 100%.

Beta Distribution is used to estimate LGD for each segment with collateral and Implied LGD Generator is used for clean exposures.

6. Ead estimations

Total exposure under each portfolio segment is divided into 3 buckets as per IFRS 9 guidelines.

All gross exposure within each segment that is less than 30 days past due (DPD) is summed under Bucket I together with any off balance sheet exposure.

All gross exposure within each segment that is more than 30 days past due but less than 89 days past due is summed under Bucket II together with any exposure where risk has materially changed since origination.

All net exposure within each segment that is more than 89 days past due is summed under Bucket III. Net exposure is estimated by reducing gross exposure to the extent of collaterization after applying prudent haircuts.