The new Definition of Default and credit risk models in the Covid-19 context.IRB and IFRS9 Validation.

ABI SUPERVISION, RISKS & PROFITABILITY, 22nd September 2020

1

The emergence of Covid-19Implications for credit risk management at banks…

NPL Strategy

Scenarios/Outlook

Approval

Early Warning+

Collection

PD, LGD and CCF

Classification

Loan Management

systems

Classification engines

Stress Testing+

Capital Plan

«Satellite» Models

Provisioning

!

► Policy review from Regulatory requirements► Dedicated process for SICR/UTP testing► Review of distressed restructuring process with

conjoint regard to the new DoD

Classification of moratoria

► Structuring, build and reconciliation of data environment for moratoria

► Implementation of process and model adaptations

► Adaptation of Regulatory and Accounting Disclosures for Industry focus and otherchannels of Covid-19 impact

Adaptation of credit risk management systems!

► Review of macroeconomic impacts► NPL flow volatility over a multi-year horizon► Impact of Moratoria on UTP flows and migrations

NPL Strategy review!

► Renewed IFRS9 Transitional Arrangements► Impacts on buffers and (Total) Capital Ratios► Alignment of Capital and Recovery Plan► Improvement of stress testing framework

Steering of Regulatory support!

!

► Re-definition of credit policies with Industry focus► Insertion of focus on moratoria in support of loan

approval and underwriting

Loan Policy and Procedures

Adaptation to Covid-19

Financial/Regulatory Reporting

► Staging and accounting of Moratoria► Policy review for IFRS9 scenarios and models► Management of Overlays

Adequacy of Provisioning framework!

► Consistency with Macro scenarios and (tempraray) re-weighting of Long-Run estimates

► Integration with internal models

Industry/Single-name Outlook!

► Calibration on current scenarios► Further industry focus on PD models

► possible view to EBA/ECB 2021 Stress Tests► LGD differentiation per Industry► Possible F-L adjustments for exposures under

moratoria

Covid-19 enhancements for IFRS9 and stress testing models!

► Integration of customer-level analysis with aggregate Forward-Looking view

► IFRS9► possible PIT adjustments from moratoria► possbile impacts of drawdowns on actual CCFs

► AIRB: upcoming impacts as per EBA GL on IRB modelling/CRR Art. 500

Covid-19 enhancements for PD and LGD models !

!

► Integration of moratoria and related monitoring► Integration of F-Looking view into loan

watchlisting

Strengthening of Early Warning

2

2016 2017 2018 2019 2020

Guidelines on the application of the definition

of default

RTS on materiality threshold of past due credit

obligation

Guidelines on PD/LGD estimation

RTS/GLs on Downturn period and Downturn LGD

CRR Art. 500

RTS on the assessment methodology for the IRB

Approach

ECB IRB Harmonisation

ECB Guidance to banks on non-performing loans

IRB Harmonisation1

New DoD

2

2-Step «Model Change»Application Package

IRB model target recalibration+

Selected EGIM adaptations

Supervisory Approval

AIRB Model recalibration

March 2018 Addendum

2021

TRIMI + TRIMIXGroundwork

Final Draft

2-Step Go-LiveSupervisory Approval

TRIM Guide Internal Models Guide. General Topics.

2022

The new Definition of Default…in progress on the revised EBA IRB Roadmap…

Consolidated EGIM.

1-Step «Model Change»Application Package

Assessment► Gap Analysis/Action Plan► Impact Analysis

«Best Effort» Recalibration► Historical reconstruction of new DoD► Model recalibration

Parallel Run

New DoD Go-Live

Supervisory Approval

IRB RWA Go-Live*

New DoD InternalValidation

AIRB As-Is

IRB model finalisation

Finalised IRB First Validation

* Non-mandatory deadline for selected Low-Default portfolios, targeted for final IRB harmonisation at 2023-end.

New UTP Triggers

Pillar II Calendar ProvisioningPillar I NPE CoverageExpectations for NPE Coverage

3

Basel vs. IFRS9…beyond IRB models…

Stress TestingLGD models

Basel

IFRS9

► Long-Run PD

► Likely Range of Variability

► Forward-Looking Lifetime PD

Basel

IFRS9

► IRB Ratings

► Distressed Restructurings

► 90DPD counters

►D(PD) Thresholds for SICR detection

► Equivalent Minimum rating downgrade schemes

► Forbearance

► 30DPD counters

Basel

IFRS9

Basel

IFRS9

► “Satellite” models for PD, LGD and ELBE/Stage 3 LGD

► Prospective Stage Transition calculation

► Prospective LGD calculation

► Prospective ECL calculation

Common data and risk modelling layer for IRB, ICAAP/SREP and IFRS9 approaches

Coverage and rationalisation of overarching model management process

Tightening IRB Standards

►Standards for New DoD in the models RDS

►Art.500/IW adjustments for LGD

►MoCs

►Calibration of ELBE and LGD “In Default”

Validation and Assurance

► Increased scope for Internal Validation in all risks to Capital

►ECB Validation Reporting

► IFRS9 validation pre-requisite to Assurance

►Model Risk Management

NPL Management

►Backtesting of “individual assessment” LLPs

►Cascading of NPL Plan on Accounting framework

►CRR/SREP Calendar Provisioning

… to Business Impact

► Planning/budgeting

► NPL Strategy

► Loan Underwriting

► Early Warning

► Long-Run/Downturn LGD/LGDD

► LGD of “Massive Disposals”

► PIT ELBE

► Forward-Looking Lifetime LGD

► Sale Scenarios

► IFRS Discounting

► IFRS Costs

Staging AllocationPD models

4

The Validation Continuum…to Internal Control.

IRB Systems

Base Scenario

Stress Scenario

Validation Framework

Stress Testing

Accounting framework

Capital Adequacy framework

IFRS9

Reperformingand Sensitivity

Macroeconomic scenarios

Forecasting data and Systems

Validation Continuum

IFRS9

CRR/EBA/EGIMModel Design

IRB Model Backtesting

Credit risk “actuals”

“Satellite” model Backtesting

Lifetime metrics backtesting

IFRS9/GPPCModel Design

Default and modelling requirements are cascaded through Regulatory and Accounting applications into the related ValidationContinuum, providing an overarching framework for controls of model design, backtesting and IT implementation testing.

Classification

AIRB Models

Credit Processes

Stress/Downturn calibration

Use Test

RWA data and Systems

Prospective Impairment

“Satellite” models backtesting

Fwd-Looking calibration

SICR backtesting

5

Credit risk modelling and validationAn ongoing convergence path in the market

Model recalibrationfrom „New

DoD“ Application

Re-calibration of LGD modelsbased on

historical new DoD and Incomplete Workouts

Initial AIRB Application

Redevelopment of PD/LGD/EAD modelsunder

New EBA requirements for Default and IRB Harmonisation

Internal Validation for AIRB Model Change Applicationfrom

New Definition of Default and IRB Harmonisation

Internal Validation of IFRS9 model updatesstemming from

Covid-19 adjustments and AIRB Model Change

AIRB Validation for„New DoD“ Application

IFRS9 ModelValidation

6

IFRS9 Impairment ValidationGovernance, processes and models through Covid-19

IFRS9 Impairment Validation

• Emergence of several new diverse market practices

• Validation cycles vs. Fast Closings

• Sale Scenarios vs. IRB CRR treatment of Massive Disposals

• Covid-19 impacts:

• Increasing relevance of Management Overlays

• Macroeconomic scenarios: cliff effect in levels and volatility

• ECL and Staging:

• new focus on Industries

• SICR sensitivity to FLI

• (retention of) approaches for mitigating pro-cyclicality

Current Industry Challenges

ii. Design of models and scenarios

iii. Data and IT implementation

iv. Reperforming and Backtesting

Governance and Supervision

Process and Control Framework

1. IFRS9 Parameters2. Staging Allocation

Process3. Lifetime ECL

Calculation

i. Documentation

The impact of the Covid-19 emergence on the Impairment framework has far-reaching implications, beyond uncertainty reflectedin forward-looking estimates, as volatility induced and initial mitigation applied are increasingly needing to be managed in a «newnormal» setting.

7

IFRS9 Impairment ValidationValidation and model challenge for Covid-19 impacts

Lifetime PD

Lifetime LGD

IFRS9 modelling approaches

Static treatment of Moratoria

Cure assumptions

Management Overlays on F-L PD

F-L framework for Loss Given Liquidation

Validation and Model Challenge

▪ Projection of segment-level moratoria portfolio shares

▪ Interval estimation of expected default flow for Industry-levelmoratoria rates

▪ Extrapolation testing of PD «satellite» models at segment/industrylevel against modelling of segmented default rate time series

▪ Testing significance of Industry-level differentiation of cure rates

▪ Testing for correlation between Corporate PDs and CRE price Indices

MacroeconomicScenarios

Interpolation of baseline scenario forecasts

Re-weighting of multiple macroeconomic scenarios

Overweighting of «Long Run» FLI

▪ Re-allocation of calendar-year forecasts by Accounting forecasting period

▪ Comparison test between «Favourable» scenario and «Consensus» pre-Covid scenarios

▪ Comparison Test between «Long Run» parameters and parametersconditional on macroeconomic «Long Run» trends

8

Diagnostics

Model Challenge

The Way ForwardMachine Learning in credit risk model validation

The EY Solution for ML-aided validation covers a full spectrum of model challenge activities, enabling process automation with AImodules for challenging the development of a tailored suite of model type and parameters consistent with model Tiering.

Reference Data Set

Validation Assumptions

Model under assessment

Challenge Approach

AutomatisedClosed-Form approaches

AutomatisedMachine-Learning

approaches

Challenger Model

Performance comparison

Inp

ut

Mo

de

l C

ha

lle

ng

e

Automation and Machine Learning in credit risk model validation

Variable clustering performed with Supervised Learningtechniques, enabling maximization of heterogeneity of(e.g.) PD/LGD across grades, constrained to satisfyvalidation assumptions.

Enhanced challenger variable selection with SupervisedLearning techniques, enabling an efficient multi-variableapproach to catch non linear relationship at validationand development stages.

Challenger model estimates based on a wide range ofSupervised Learning techniques, driven by alternativeaccuracy/interpretability trade-offs consistent withModel Tier, including Classification and Regression Trees(CART), Bagging and Random Forest.

Wide set of generalised performance KPIs in line withSupervisory guidance and bank’s own validationpractices, including performance metrics andassumptions testing.

VariableSelection

Clustering

Ou

tpu

t

ECB Validation Reporting

Model Tier

Assumptions Testing

Bank Backtesting

KPIs

9

Machine Learning in credit risk model validationM-L Challenge for LGD Validation at a primary IRB Bank

-75% FTE compared to standard model validation

process

+293 bps of Goodness-Of-Fit Index compared to challenged

model

The use of the EY Solution has enabled a significant reduction in effort and a significant gain in performance KPIs used for dialogue with the Supervisor, highlighting feasible improvements in developed models under validation

+284 bps of discriminatory power compared to challenged

model

Process Automation within the model validation module

Constrained model validation

+Process Automation

+Machine Learning

Highly streamlined LGD validation process through use of Process Automation and

Machine Learning techniques

Challenger estimates from on a wide range of Supervised Learning techniques, driven by

model/parameter-specific accuracy/interpretability trade-offs

Selected highly interpretable Machine Learning techniques highlight room for further improvement for the challenged model

10

-80% FTE compared to standard model validation

process

20+ automatically filled reporting templates

covering

30+ modelsfor all AIRB

segments/parameters

100% coverage of Supervisory requirements

Process Automation within model backtesting

Consistency with Supervisory Instructions

+Process Automation

Highly automated Internal Validation and ECB Validation Reporting through Process

Automation

Automatically filled reporting templates covering Supervisory instructions on an end-to-end basis

Room for further efficiency gain in backtesting standardization over the full spectrum of credit

risk models

Machine Learning in credit risk model validationValidation Reporting Backtesting at a primary IRB Bank

The use of the EY Solution for ECB Validation Reporting has enabled a significant reduction in effort and critically reduced therequired elapsed for anticipated testing of alternative model/parameter specifications

Thank You

EY | Assurance | Tax | Strategy and Transactions | Consulting

About EY

EY is a global leader in assurance, tax, strategy, transaction and consulting

services. The insights and quality services we deliver help build trust and

confidence in the capital markets and in economies the world over. We develop

outstanding leaders who team to deliver on our promises to all of our

stakeholders. In so doing, we play a critical role in building a better working

world for our people, for our clients and for our communities.

EY refers to the global organization, and may refer to one or more, of the

member firms of Ernst & Young Global Limited, each of which is a separate

legal entity. Ernst & Young Global Limited, a UK company limited by guarantee,

does not provide services to clients. Information about how EY collects and

uses personal data and a description of the rights individuals have under data

protection legislation are available via ey.com/privacy. For more information

about our organization, please visit ey.com.

For more information, please visit

www.ey.com

© 2020 EY Advisory S.p.A.

All Rights Reserved