Expected Loss Models: Methodological Approach to IFRS9 Impairment & Validation Framework Jad Abou Akl 30 November 2016 2016 Experian Limited. All rights reserved. Experian and the marks used herein are trademarks or registered trademarks of Experian Limited. Other products and company names mentioned may be the trademarks of their respective owners. No part of this copyrighted work may be reproduced, modified, or distributed in any form or manner without prior written permission of Experian Limited. Experian Public.
Agenda IFRS9 Impairment Models Expected Loss Models Challenges & Ideas Case Study Impairment Models Validation 2
The Components of IFRS9 Models IAS 39 Drawbacks Enhancement in IFRS 9 Integrated Through No assessment of credit risk increase The credit risk is assessed periodically and the impact is reflected in credit losses Stage classification: all exposures will be classified into risk class 1, 2 or 3 Provisions are recognised when it is too late, only after the default occurs Provisions must recognize the future expected loss at an early stage Expected loss models: to be applied for the entire portfolio Provisions are not sensitive to credit risk increase Provisions will be sensitive to significant increase in credit risk Provisions must cover lifetime expected losses when significant increase in risk is detected Impairment models are backward looking and not sensitive to economy change Impairment models have to be forward looking and integrate the future impact of economy change Macro economic modelling: will be considered for future loss calculation 3
IFRS 9 Impairment Impairment Models Calculation Process Flow Experian methodology for IFRS9 impairment calculation provides several alternatives, depending on the level of advancement of IRB practice in the Bank. The scheme below shows the modeling flow for banks with advanced IRB models Credit Loss Calculation PD, LGD, EAD Basel Models Model adjustments Macro economic models Risk Stage Classification IFRS9 compliant PD, LGD, EAD Stage 1: 12-month Expected losses Stage 2 & 3: Lifetime expected losses 2016 Experian Limited. All rights reserved. 4
Impairment Models Basel Models Adjustment Basel PD Models TTC: Through the cycle 12 month PD Principle of conservatism Based on the historical long run average IFRS 9 PD Models PiT: point in time Lifetime PD for stage 2 & 3 Accurate estimation of risk Forward looking: sensitive to economic conditions Basel LGD Models Many alternatives for discount rate Indirect costs to be includes Principle of conservatism Downturn impact to be added IFRS 9 LGD Models Effective interest rate on the loan must be used Only costs directly related to the loan Accurate estimation of loss rates Forward looking: sensitive to economic conditions Basel Exposure & Maturity Models Contractual maturity can be used CCF is needed for revolving exposures Principle of conservatism TTC: Through the cycle estimation IFRS 9 Exposure & Maturity Models Expected maturity is a must for revolving exposures for lifetime loss calculation Prepayment models are also to be used for loans Accurate estimation of CCF factors Forward looking: sensitive to economic conditions 5
Impairment Models Expected Loss Calculation Basel Model Adjustments Macro economic Models 1 Expected maturity of the credit FiT parameters year 1 FiT parameters year 2 FiT parameters year n 2 Expected loss for stage 1 PD 1 LGD 1 EAD 1 PD 2 LGD 2 EAD 2 PD n LGD n EAD n Expected Loss Year 1 Expected Loss Year 2 Expected Loss Year n 3 Lifetime Loss The lifetime expected loss (ECL) is the sum of yearly discounted loss for the remaining expected lifetime of the credit 6
Challenges & Ideas (1/2) Challenges Our approach Tested Banks having only scoring models - Future looking economy-sensitive score model is developed - Default rates are used to calibrate the score into PiT future PD - Two options for LGD: regulatory or accounting loss models Banks with no risk models or scorecards - Default rates are used to deriveeconomy sensitive pool level PDs - Two options for LGD: regulatory or accounting loss models Robust economic model not easy to develop - Several modeling alternatives have been successfully applied: Multi-linear regression Error correction models 7
Challenges & Ideas (2/2) Challenges Our approach Tested Future in time LGD is very hard to model - LGD is a multi-year figure. It makes it difficult to find correlations with yearly economic indicators - Instead of modelling LGD, decompose into several components that can be easily linked with the economy: House price index Marginal recovery rates Default cure ratese The Bank policies & procedures have changed over time, same for the credit portfolio composition - The history should be assessed taking into consideration the internal adjustments in the bank - Vintage analysis helps separate the loss fluctuations coming from internal & Exogenous (economy) factors 8
Case Study Tier 1 Bank in Romania Case study on personal loan (PL) portfolio of a major Romanian Bank Average loan duration 36 months-from 6 up to 180 months Loans with very broad vintages-from 1990s up to 2015 The portfolio has loans with different currencies The portfolio is risky showing high PD and LGD values The model was developed as follows: The future PD is calculated with migration matrix The macro economic model is developed on the Romanian Macro-economic data, following the EMV method 9
Axis Title Case Study Tier 1 Bank in Romania Future PD Estimation Results PD Macro Forecast 7,00 6,00 5,00 4,00 3,00 2,00 1,00 0,00 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 PD 6,36 5,96 5,24 4,22 3,62 3,62 3,60 3,58 3,41 3,40 3,42 3,37 3,40 3,60 3,61 3,49 10
Case Study Tier 1 Bank in Romania Expected Loss Results Risk stage Nb Accounts % Total EL Total Balance Av EL Ratio 1 89,539 76.77% 4 Mio 350 Mio 1.14% 2 21,739 18.64% 31 Mio 119 Mio 26.05% The EL decreases to 13.46% if the 12-month EL is used instead of lifetime. A 100% adjustment. 3 5,352 4.59% 30 Mio 43 Mio 69.76% TOTAL 116,630 100% 65 Mio 512 Mio 12.69% The EL ratio is very high since the portfolio is very risky. The performing LGD reaches up to 65%. 11
Case Study Tier 1 Bank in Romania Expected Loss Results Calibrated on Turkish market To calibrate the results on the Turkish market levels, we decreased the average performing LGD of the portfolio to around 30%. The number is chosen based on our experience for Personal loan LGD in the Turkish banking sector. Risk stage Nb Accounts % Total EL Total Balance Av EL Ratio 1 89,539 76.77% 2 Mio 350 Mio 0.57% 2 21,739 18.64% 14 Mio 119 Mio 11.7% 3 5,352 4.59% 26 Mio 43 Mio 60.4% The EL ratio is more reasonable, when using 30% performing LGD averages TOTAL 116,630 100% 42 Mio 512 Mio 8.2% 12
Agenda IFRS9 Impairment Models Expected Loss Models Challenges & Ideas Case Study Impairment Models Validation 13
Validation of IFRS9 Models Framework The framework consists of quantitative and qualitative elements Validation of IFRS9 Models Quantitative Validation Data Integrity Qualitative Validation Backtesting Benchmarking Model Design Model output Model Documentation Model design & assumptions Performance & calibration Population & model stability IFRS9 models specific tests Economic models, triggers for stage allocation 2016 Experian Limited. All rights reserved. Experian Public. 14
Validation of IFRS9 Models Challenges Challenges Regulatory Approach Details The approval body: National regulator or External audit company Still to experience the regulator approach: points of attentions, show stoppers Best Practice Complexity of the Model Data & Sample Duration IFRS9 is a new topic, building the best practice both for model development and validation will take a while The standards are not addressing requirements in details Risk stage classification: the regulation is not comprehensive PD, LGD & EAD adjustments Macro economic models 12-month and lifetime losses Validating the economic models requires a long period of data ideally covering a complete economic cycle Validating the lifetime loss models cannot be performed with the regular back-testing due to absence of sufficient years of data 2016 Experian Limited. All rights reserved. Experian Public. 15