Moody's Corporation · NYSE:MCO
Moody's — Moody's Analytics and KBA webinar (May 2021)
45 slides from Moody's's conference presentation, May 2021
Slide 1: Forward-looking Risk Measurement
Forward-looking Risk Measurement A Moody’s Analytics Presentation Armen Mirzoyan, Associate Director, MA Predictive Analytics Unit Gega Todua, Assistant Director, MA Predictive Analytics Unit May 2021
Slide 2: Overview
1 Overview Forward-Looking Risk 2
Slide 3: Regulatory Origins of Forward-Looking Risk
Regulatory Origins of Forward-Looking Risk IFRS 9 Requirements “ An entity shall measure ECL of a financial instrument in a way that reflects an unbiased and probability-weighted amount that is determined by evaluating a range of possible outcomes. ” (5.5.17) “ When measuring ECL, an entity need not
Slide 4: Key Take-Aways
Key Take-Aways 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 th
Slide 5: Macroeconomic Forecasting
2 Macroeconomic Forecasting Forward-Looking Risk 5
Slide 6: Coverage and Key Features
2.1 Coverage and Key Features Forward-Looking Risk 6
Slide 7: Moody's Analytics Global Forecast Coverage
Moody’s Analytics Global Forecast Coverage April 2021 Included in the Global Model (73 countries) Additional forecasts (IFRS9) (28 countries) Sources: Moody’s Analytics Forward-Looking Risk 7
Slide 8: Key Features
Key Features Collaborative Access and Integration Develop scenarios individually or collaboratively in a real-time, multi-user environment. Integrate forecasts into your workflow seamlessly through our API and Excel Add-In. Comprehensive Coverage Create scenarios for 101 countries and 10 regional ag
Slide 9: Moody's Global Macroeconomic Model
2.2 Moody’s Global Macroeconomic Model Forward-Looking Risk 9
Slide 10: Global Macroeconomic Model
Global Macroeconomic Model Provides Globally Linked Forecasts Linkages in the model allow for global shock propagation and contagion effects, and help ensure scenario consistency » Trade flows (exports reflect partner imports) » Financial markets (stock prices and bond yields) » Prices (exchange rat
Slide 11: Modelling Approach
Modelling Approach Each Country-model is a Mix of Theory and Data Theory » Quality of forecast and scenarios » Complex » Limited quantity of forecasts Our Models Intersection of purely data- and purely theory-based models Data » Quality and quantity of forecasts » Easy to produce » Not ideal for sce
Slide 12: Country-model Methodology Overview
Country-model Methodology Overview Detailed Quantitative & Qualitative Testing Specification choice » Theoretical reasoning versus statistical properties In-sample equation fit » R-squared, RMSE, information criteria » Fitted values and residuals Forecasting performance » Back-testing: conditional a
Slide 13: Rigorous Equation Development Process
Rigorous Equation Development Process Identify Equation Model Equation Equation Equation for Estimation Integration Approval Implementation Development •Specification Testing •Model owner •Production team implements •Revised indicator •Variable selection •Test inclusion of examines equation equation
Slide 14: Equations Designed to Balance Theory & Empirics
Equations Designed to Balance Theory & Empirics Variable Specification suggested by economic theory draws on… Unemployment rate Okun's Law Labor Force Participation rate & demographics Private consumption expenditure Keynesian consumption function / Euler equation Public consumption expenditure Baum
Slide 15: Kenya GDP Components – Model Drivers
Kenya GDP Components – Model Drivers GDP Identity Real Demand Real Net Exports Real Potential Output Gap (Lending Output Gap GDP Rate - Monetary Aggregate External Shock Growth Policy rate) Demand (e.g., U.S. GDP) Expectations Forward-Looking Risk 15
Slide 16: Baseline Forecasting
Baseline Forecasting Real GDP, % chg year ago Real GDP, % chg year ago 15.0 21.0 10.0 US 16.0 China 11.0 5.0 6.0 0.0 1.0 -5.0 -4.0 -10.0 -9.0 2007Q1 2012Q1 2017Q1 2022Q1 2028Q2 2033Q2 2007Q1 2012Q1 2017Q1 2022Q1 2027Q1 2032Q1 Real GDP, % chg year ago Real GDP, % chg year ago 15.0 14.0 10.0 Eurozone
Slide 17: Scenario Generation
2.3 Scenario Generation Forward-Looking Risk 17
Slide 18: Phases of Scenario Workflow
Phases of Scenario Workflow Credit risk (PD, LGD, EAD, ECL) and market risk instruments forecast Global Market and Credit Combined Key Assumptions Macroeconomic Risk Models Forecast x Model z(x,y) (x,y,z) y Macroeconomic series forecast Forward-Looking Risk 18
Slide 19: Scenario Generation Using Moody's Analytics Global Model
Scenario Generation Using Moody’s Analytics Global Model Forecasts for 70+ countries used by 780 Clients worldwide Standard Scenarios Custom Scenarios Scenario Studio Moody’s Analytics Off-the- Partially or Fully Specified Web-based Scenario Shelf Scenarios Economy Assumptions Building Application E
Slide 20: Pillars of Scenario Generation
Pillars of Scenario Generation I II III Severity Narrative Transmission Quantitative representation of “How Determines overall nature of the Global linkages in models transmit favorable/adverse is given scenario” scenario and guides the exact path shocks across countries Ensures that scenarios are o
Slide 21: Scenario Calibration: Discrete Scenario Prob.
Scenario Calibration: Discrete Scenario Prob. GDP Growth %, Annualized avg., 10,000 Simulations over a 5-yr Period BL 800 0.50 Severity S3: 23% Weight S1: 23% Weight 40% Weight 700 600 S3: 10% S1: 90% S0: 7% Weight 500 downside severity S4: 7% Weight severity 400 S4: 4% 300 downside severity S0: 96%
Slide 22: Scenario Forecasting
Scenario Forecasting U.S. GDP, 2012 bil. USD China GDP, 2015 Bil. CNY 148,000 S1 S1 23,000 138,000 S3 S3 Baseline 128,000 21,000 Baseline 118,000 19,000 108,000 98,000 17,000 88,000 15,000 78,000 18 19F 20 21F 22F 23F 24F 25F 26F 27F 28F 18 19F 20 21F 22F 23F 24F 25F 26F 27F 28F Euro Zone Inflation,
Slide 23: Scenario Studio
2.4 Scenario Studio Forward-Looking Risk 23
Slide 24: Elements of Forecast Integrity
Elements of Forecast Integrity Specifically: » Up-to-date » No errors High-quality Sound » Long time series data model Comprising: » Temporally consistent » Process – The steps taken in the production » Accurately calculated of a forecast » Platform – the tools Sound used to implement the assumpt- L
Slide 25: Technology-enabled forecasting
Technology-enabled forecasting A cloud platform enables a distributed process with a globally linked model Installed-software world Cloud platform world » Sequential economic linkages » Simultaneous economic linkages » Simplified international interactions » Sophisticated international interactions
Slide 26: Scenario Studio is …
Scenario Studio is … » A secure web application for scenario forecasting » Facilitates collaborative forecasting » Hosts several Moody’s Analytics models – Global, U.S., sub-national » Supports rigorous forecast governance processes » Enables model customization Forward-Looking Risk 26
Slide 27: Forward-looking Risk
3 Forward-looking Risk Forward-Looking Risk 27
Slide 28: Linking Scenarios to Risk Measures
Linking Scenarios to Risk Measures IFRS9, ICAAP, IRRBB, Stress Testing, Business Planning Key Risk Models • Macroeconomic data • Expected Credit Loss • Baseline scenario • Capital Assessment forecasts • Credit risk • Change in Net Interest • Alternative scenarios • Market risk Income forecasts • Int
Slide 29: Expected/Stressed Loss Calculation Framework
Expected/Stressed Loss Calculation Framework Macroeconomic Scenario Forecast Minimum 3 scenarios (baseline, upside & downside) Scenario Probability Weights Macro-Conditioned, Point-in-Time, Forward-Looking Probability of X Loss Given X Exposure at X Discount Factor = Expected Credit Default Default
Slide 30: PD Modelling Techniques and Approaches
PD Modelling Techniques and Approaches Model Types Vary by Need Vintage-level Account-level Portfolio-level Panel-data modelling approach PD is forecasted using customer and PD directly linked to macroeconomic where the data is split by vintages loan characteristics, and drivers using time series te
Slide 31: PD Vintage-level Approach
PD Vintage-level Approach PD= f( Lifecycle , Quality of Vintage , Forward-looking Indicator ) 0.1 0.16 0.1 0.09 0.14 0.09 0.08 0.08 0.12 0.07 0.07 0.1 0.06 0.06 PD PD PD 0.05 0.08 0.05 0.04 0.06 0.04 0.03 0.03 0.04 0.02 0.02 0.01 0.02 0.01 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8
Slide 32: PD Vintage-level Approach: Mortgages Example
PD Vintage-level Approach Mortgages Example 1. Lifecycle 2. Quality of Vintage 3. Forward-looking Indicator Final PD Model 0.035 HPI YoY Growth Default Rate Fitted Values Baseline 0.02 0.030 History Baseline Upside Downside Fitted Values Upside Fitted Values Downside 0.018 Predicted Lifecycle 0.025
Slide 33: PD Vintage-level Approach: Credit Cards Example
PD Vintage-level Approach Credit Cards Example 1. Lifecycle 2. Quality of Vintage 3. Forward-looking Indicator Final PD Model 0.014 Private Consumption (YoY Growth) 0.012 Default Rate Fitted Values Baseline 0.012 History Baseline Upside Downside Fitted Values Upside Fitted Values Downside Predicted
Slide 34: Account-level PiT PD
Account-level PiT PD Mapping IFRS9 PDs to IRB PD The vintage PD model is used to map the shape of the PD curves to an account-specific risk metric in order to obtain consistent PD levels based on the latest available credit risk information. 3% 0.03 Baseline Upside Downside Prob. - Weighted Prob.-we
Slide 35: Portfolio-level Modelling
Portfolio-level Modelling » If a portfolio has restricted aggregate 0 – 29 DPD 30-59 DPD 60-89 DPD 90-119 DPD information available such as number of 0-29 DPD 95.37% 2.13% 0.69% 1.81% accounts in DPD buckets, outstanding 30-59 DPD 77.57% 1.82% 0.64% 19.97% balance in DPD buckets, etc.: 60-89 DPD 43.
Slide 36: Optimal Variable Searching Tool
Optimal Variable Searching Tool User friendly Performs the Variable Selection Algorithm for as much as 25 potential drivers in one step. Flexible and customizable Allows the user to specify the model as desired. Functional Exports pre-selected models and estimation statistics to an Excel file. Makes
Slide 37: OVS Customizable Features
OVS Customizable Features » Target variable » Scenarios » Potential drivers and expected signs » Maximum number of drivers in the final model » Maximum number of lags for drivers » Estimator and estimation options: any built-in estimator in the software » Correlation coefficient threshold. Default v
Slide 38: Dynamic Credit Risk Model-Building
Dynamic Credit Risk Model-Building Variable selection Best Subset for scenario-conditional Variable forecasting Selection Algorithm 𝑪𝑪𝒌𝒌𝒎𝒎 k: right-hand-side variables m: potential economic & internal drivers All combinations of size k from vector m Pairwise Correlations (optional) Exclude…
Slide 39: Optimal Variable Searching Tool
Optimal Variable Searching Tool Web Application » Allows to run OVS on your own data » No R installation needed » Runs in browser » Easy-to-use, no code involved » 3 menu items on the sidebar with the last one showing OVS results 1. File upload – upload a file with data in CSV format – upload the ap
Slide 40: OVS Tool Web Application
OVS Tool Web Application Variables Selection 2. Variable selection – Choose one target variable – Select fixed explanatory variables (optional) – Select potential drivers from the remaining variables – Specify the explanatory variables for which positive coefficient is required (optional) – Specify
Slide 41: OVS Tool Web Application
OVS Tool Web Application Parameter Set-up & Results 3. Parameter set-up & OVS results – Choose the maximum number of drivers that can be included in the model – Specify a p-value threshold for testing significance of explanatory variables – Input path to a file where you want to export the OVS resul
Slide 42: Significant Increase in Credit Risk
Significant Increase in Credit Risk Account-level Reporting Lifetime PD Account-level Origination Lifetime PD » To measure the change in risk since initial d Distance to assess for threshold recognition, we examine the proportional 7.0% difference between 6.0% – the lifetime PD at the reporting date
Slide 43: LGD Design Approaches
LGD Design Approaches Balance and Recoveries By Assumption For a facility i, time t and workout period w: LGD of 50-60% for PF, 30-40% for RE and 65-75% for CC; fully insured products usually get LGD of 5-10%. 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑖𝑖,𝑡𝑡 − 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑖𝑖,𝑡𝑡+𝑤𝑤…
Slide 44: Prepayment Model with Macro Overlay
Prepayment Model with Macro Overlay UK Mortgage, Loan-level Model Customer, Loan Characteristics and Prepayment Rate Scenario Forecast Macroeconomic Factors » Modelling prepayment factor Customer Characteristics: First-time interpreted as the Buyer, Employment Status, Primary –Baseline – Moderate Re
Slide 45: Contact Us: Economics & Business Analytics Offices
Contact Us: Economics & Business Analytics Offices West Chester, EBA-HQ London Prague Singapore +1.610.235.5299 +44.20.7772.5454 +420.22.422.2929 +65.6511.4400 121 North Walnut Street, Suite 500 One Canada Square Washingtonova 17 6 Shenton Way West Chester PA 19380 Canary Wharf 110 00 Prague 1 #14-0
Moody's — Moody's Analytics and KBA webinar (May 2021) on 6ix