IFRS 9 Scenario Implementation and ECL …

1 IFRS 9 Scenario Implementation and ECL Calculation for Retail PortfoliosAlaistairChan, Economist | Olga Loiseau-Aslanidi, Senior Economist and Risk ModelerIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 20172 Moody's Analyticsoperates independently of the credit ratings activities of Moody's Investors Service. We do not comment on credit ratings or potential rating changes, and no opinion or analysis you hear during this presentation can be assumed to reflect those of the ratings 9 Scenario and Retail Portfolio Strategy, October 24th, 20173 ECONOMISTC overs national and metropolitan economic issues across the Asia-Pacific ECONOMIST & RISK MODELERD esigns and implements models for stress-testing and Loiseau-AslanidiIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, scenarios in a Probability Weighted Macroeconomic Forecasting Credit Methodology ChallengesAgenda1 Economic scenarios in a Probability-Weighted EnvironmentIFRS 9 Scenario and Retail Portfolio Strategy, October 24th.

2 20176 AnentityshallmeasureECLofa financialinstrumentina waythatreflectsanunbiasedandprobability- weightedamountthatis determinedbyevaluatinga rangeofpossibleoutcomes. (5 . ) WhenmeasuringECL,anentityneednotnecessar ilyidentifyeverypossiblescenario. However,it shallconsidertheriskofprobabilitythata creditlossoccursbyreflectingthepossibili tythata creditlossoccursandthepossibilitythatnoc reditlossoccurs,evenif thepossibilityofa creditlossoccurringis verylow. (5 . ) Thismaynotneedtobecomplexanalysis. Insomecases,relativelysimplemodellingmay besufficient,withouttheneedfora largenumberofdetailedsimulationsofscenar ios. ( ) ..anentityneednotnecessarilyidentifyever ypossiblescenario. (5 . )Addressing IFRS 9 Forward-Looking AspectsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 20177 Forward Looking & Probability-Weighted Outcomes Requires expected credit losses (ECL) to account for forward-looking information Requires probability-weighted outcomes when measuring expected credit losses Estimates should reflect the possibility that a credit loss occurs and the possibility that no credit loss occursMacroeconomic modelling satisfies both requirements above Key Take-AwaysIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 20178 Example with 3 scenarios (s1, s2, s3) & (1, 2.)

3 , ) PeriodsECL Calculations Lifetime ExampleLifetime ECLECL(s1,t1)ECL(s1,t2)ECL(s1, )ECL(s2,t1)ECL(s2,t2)ECL(s2, )ECL(s3,t1)ECL(s3,t2)ECL(s3, )p(s2)p(s3)p(s1)IFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 20179On the Number of scenarios and Probability TargetsForward Looking Probability Weights2 Global Macroeconomic Forecasting ModelsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201711 Mix of Theory and DataMacroeconomic ForecastsData Quality and quantity of forecasts Easy to produce Not ideal for Scenario analysisTheory Quality of forecast and scenarios Complex Limited quantity of forecastsOur ModelsIntersection of purely data-and purely theory-based ModelsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201712 With Detailed Quantitative & Qualitative TestingStructural Forecast Model MethodologyExchange ratesInvestmentWages & salariesPopulationPricesGDPM onetary policy rateImportsGovernment ExportsGlobal GDPU nemployment rateConsumptionLabor forcePotential GDPB anking sectorImport prices10-yr yieldGlobal pricesEmploymentSpecification choice Theoretical reasoning versus statistical propertiesIn-sample equation fit R-squared, RMSE, information criteria Fitted values and residualsForecasting performance Back-testing: conditional and unconditional evaluation Benchmarking during important past episodesSensitivity to shocks Forecasts across scenarios Response to individual shocksIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201713 Core economic concepts.

4 National Accounts Balance of Payments Government Finance Industrial Production Producer Price Index Retail Sale IndexComprehensive Coverage More Than 1,800 Forecast Variables for 60+ CountriesCore economic concepts: Interest Rates Stock Markets Money Markets Labor Home Price Indices and many 9 Scenario and Retail Portfolio Strategy, October 24th, 201714 Key Features 30-year horizon, for baseline forecast plus up to eightalternative scenarios Coverage of more than 1,800 economic, financial and demographic variables Available for 60+ countries globally Forecasts updated monthly,history updated in real-time Fully documented model methodology; Scenario assumptions published monthly Model validation reports availableReasonable and Defensible Forecasts from Moody s AnalyticsMoody s Analytics ScenariosS1S2S3S4 BLS5S6S8 FEDPRAEBACUMOODYS ANALYTICS BASELINE + S1-S8 EXPANDED REGUALTORYCUSTOMIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201715 Weaker EconomyHealthier EconomyBaselineS3:Deep Recession1-in-10S4:Protracted Slump1-in-25 Alternative Economic ScenariosS2:Mild Recession1-in-4S1:Stronger Near-TermGrowth1-in-10 EmergingMarketsHard LandingSovereignDefaultShockIn line withRegulatoryGuidelines Custom Built ( ICAAP) Event-DrivenSimulation-Based1:100 1:25 1:20 1:10 1:4 Forecast 1.

5 10 Moody s Analytics ScenariosIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201716 How Do we Gauge Severity and Probability of a Particular Scenario ?From Severity to Probability of OccurrenceHistorical analysis: Compute historical mean and standard deviations of variables of interest; GDP growth rate Compare scenarios in terms of standard deviation Assume normality to estimate probabilities Drawback: history only one realization of the stochastic process; not many periods to compareDifferent metrics: Real GDP growth over a period Real GDP start-to -trough Unemployment rate start-to -troughSeverity: Probability that particular economic statistic is less than in a given Scenario -> IFRS 9 IFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201717 Shocks and ScenariosPoints of Calibration Severity ( peak to trough of real GDP, peak unemployment rate) Probability (where does severity fall in distribution of outcomes?)

6 Methods of Calibration Historical analysis: empirical distribution of a given metric Example: a histogram of the unemployment rate, or real GDP growth rates Drawbacks: Limited data samples, reflects an unconditional distribution Monte Carlo simulation Estimate a stochastic model ( VAR of growth, inflation and interest rates) Simulate forecast paths and compute distribution of 5-year growth rates. Assign probabilities based on the distribution of simulated paths IFRS 9 convert continuous CDF probabilities to discrete PDFs. 10th percentile low stress 50th percentile baseline 90th percentile high upsideIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201718 Scenario Calibration: Discrete Scenario Growth %, Annualized avg, 10,000 Simulations over a 5-yr PeriodSource: Moody s : 10% CDF downsideS3: 30% discrete prob= 10% (s3 CDF) + 20% (1/2 distance s3-bl)BL: 50% CDF baselineBL: 40% discrete probS1: 30% discrete prob= 10% (s1 CDF) + 20% (1/2 distance s1-bl)S1: 10% CDF upsideIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 2017194567800040812162024 BaselineUpsideDownsideSources: Australia Bureau of Statistics, Moody s AnalyticsAustralia, Scenario ForecastsUnemployment rate, %IFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201720 Sources.

7 National Bureau of Statistics, Moody s AnalyticsChina, Scenario ForecastsGDP, % change yrago24681012141600040812162024 BaselineUpsideDownsideIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201721 Sources: National Bureau of Statistics, Moody s AnalyticsThailand, Scenario ForecastsExports, 2002 THB bil, SAAR2,0004,0006,0008,00010,00012,0000004 0812162024 BaselineUpsideDownsideIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201722 Web Based Interface, Office Add-In, or APIView, Manipulate and Automate Delivery of Data in a Variety of Formats 3 Retail Credit Methodology ChallengesIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201724 Retail Credit Methodology ChallengesEnsuring Consistency with Stress Testing, ICAAP & Pricing Models PIT PD Lifetime PD/PD Term Structure PIT LGD Lifetime LGD/LGD Term Structure Discounting (EIR)

8 Lifetime EAD/EAD Term Structure Discounting Prepayment Threshold Definition Intra-stage Movements IFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201725 Dynamic panel data models, cluster-based combined with vintage/cohort analysis Well suited to capture lifecycle aspect of credit behavior and bring macroeconomic factors explicitly Alternative methods can be tested for default rates or losses, such as Roll -rate/transition approaches Loan or obligor-level logistic regression models Survival analysis models, or simpler top-down time-series modelsRetail Credit Methodology ChallengesEnsuring Consistency with Stress Testing, ICAAP & Pricing ModelsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201726 Historical data used to build the econometric equations Optimal variable search methods leveraged to select the macroeconomic best drivers into the model Historical data integrity is imperative, and the main dependency Granularity, frequency, consistency, completeness, and quality Limitations can put constraints on the type of modelling techniques that can be utilizedRetail Credit Methodology ChallengesEnsuring Consistency with Stress Testing, ICAAP & Pricing ModelsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201727 Selected macroeconomic variables tested to bring the forward-looking dimension PDs and secured LGDs explicitly connected to macro data EAD and unsecured LGDs often modelled independently of economic factors Explicit.

9 Closed-form equations, ensuring 100% transparency in our approach Detailed model documentation and validation Feedback from validation team & auditors incorporated Model governance, re-calibration updates, monitoring, use test .Retail Credit Methodology ChallengesEnsuring Consistency with Stress Testing, ICAAP & Pricing ModelsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201728 Leading Example: Vintage Panel-Data Structure0246810 Delinquency rate (%)020406080 Age-on-books (months)0246810 Delinquency rate (%)020406080 Age-on-books (months)0246810 Delinquency rate (%)020406080 Age-on-books (months)Hypothetical Data (over age)Lifecycle Effect (over age)Vintage Quality (over age)Business Cycle Effect (over time)IFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201729 Leading Example: Vintage Panel-Data StructureIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201730 Macroeconomics of Credit Risk.

10 Data InputExamples of Economic and Internal DriversEconomic Data (Macro & Regional)Portfolio Data LabourMarket IndicatorsEmployment, Unemployment, Wage/Salary Growth Housing Market IndicatorsHome Prices, Home Sales, Housing Starts, Permits Financial Market IndicatorsPolicy Rate, Debt-Service Ratios, Revolving Consumer Debt Economic PerformanceGDP Growth, Disposable Income Growth Industry/Sector DriversSales, Employment, Pricing Observed PerformanceDelinquencies, Defaults, Losses, Prepayments Credit Score Cut-offs Credit Applications LTVs Payment-to-Income Marketing Activity Collection TreatmentsIFRS 9 Scenario and Retail Portfolio Strategy, October 24th, 201731 Credit Risk Model Building Example Variable Selection AlgorithmAll combinations of size kfrom vect