Advanced IRB Approach

Maintaining Models, Validation and Re-calibration

Agenda Program
divider graphic
Prague, NH Hotel Prague
divider graphic
Understanding of Risk Parameters
Risk Prediction Data Mart
Regulatory and Functional Requirements
Central Definition of the Default Indicator
Model Validation and Re-calibration
Benchmarking and Low Default Portfolios
Practical Applications and Maintenance of Models
This course is part one of a two course series. It concentrates on rating system use and maintenance aspects, including validation and re-calibration, while the second course concentrates on the creation of rating models.

The topics in this first course relate more to the Business-As-Usual activities of a credit risk management team, while model creation often takes the form of projects, which may sometimes be outsourced to third parties. Having said that, it is strongly advised for everybody working in this field to have a good understanding of all rating system aspects.

This course starts with an overview of the components of individual risk prediction and explains the risk parameters PD, LGD and EAD. After that, various uses of risk predictions are explained. Special attention is given to credit approval topics and risk weighted asset calculation. To end the introductory part, an overview of the tasks involved in maintaining prediction quality is given.

The focus then shifts to the organisation of the historical data that is at the center of all risk prediction. Various functional distinctions that the data mart must account for are laid out and a general structure of the modelling and validation tables that form the end product of the data integration efforts is given. It is in this part that we cover the central definition of the default indicator. Also the various types of characteristics that are typically available to describe a case are discussed here.

On the basis of a risk prediction data mart the quality of rating models can then be assessed. We cover a variety of widely used measures of key model quality aspects, such as calibratedness, discriminatory power and stability and discuss their interrelatedness and their interpretation in the light of economic cycles. Regulatory requirements and qualitative aspects of rating system validation are covered as well.

Finally, models may be re-calibrated without performing a full model re-build. We discuss mathematical re-calibration techniques as well as the rationale behind re-calibration.

Program of the seminar: Advanced IRB Approach

The seminar timetable follows Central European Time (CET).

09.00 - 09.15 Welcome and Introduction

09.15 - 12.00 Use and Maintenance of Individual Risk Predictions

  • Risk parameters
    • PD
    • LGD
    • EAD
  • Use of individual risk predictions
    • Credit approval
    • Unexpected loss / regulatory risk capital
    • Credit risk reporting
    • Portfolio forecasting
    • Limit setting
    • Collection
    • Marketing
  • Model maintenance
    • Histories of individual characteristics and performance
    • Prediction quality monitoring
    • Re-calibration
    • Re-modeling
    • Re-implementation
    • Model use and decision monitoring
    • Documentation and administration
    • Organisation
  • Case studies
    • Setting Cutoff
    • Risk weighted asset calculation

12.00 - 13.00 Lunch

13.00 - 16.30 The Risk Prediction Data Mart

  • Modeling and validation tables
  • Data mart architecture
    • PD/LGD/EAD
    • Retail/Corporate
    • Product
    • Application Scoring/Behavioural Scoring
    • Account/Customer
    • Accepts/Rejects
    • Characteristics/Performance
    • Yearly/Monthly/Daily
    • Raw values / re-calculated values
  • Regulatory and functional requirements
    • Data quality, internal and external consistency
    • Retro-calculation / re-mapping
  • Characteristics
    • Retail characteristics
      • Application characteristics
      • Behavioural characteristics
    • Corporate characteristics
    • Partial scores and composite models
  • Time Stamps
  • Segment indicator
  • Default indicator
    • Default event definition
    • Application scoring default indicator
      • Maturation curves
      • Prediction horizon
      • Immature cases
      • Rejected cases
    • Behavioural scoring default indicator
  • LGD and EAD
  • Case studies
    • Application scoring default indicator
    • Behavioural characteristic calculation

09.00 - 12.00 Validation and Re-calibration

  • Validation vs. monitoring
  • Validation vs. re-calibration vs. re-modeling
  • Regulatory rating system validation requirements
  • Qualitative validation
    • Model design
    • Data
    • Use test
    • Organisation
    • The validation protocol
  • Benchmarking and low default portfolios
  • Backtesting
    • Measures of model quality
    • Score level vs. characteristic level
    • Groupings
  • Calibratedness
    • Definitions of default rate
    • Binomial test
    • Hosmer-Lemeshow
    • Spiegelhalter

12.00 - 13.00 Lunch

13.00 - 16.30 Validation and Re-calibration (continued)

  • Power measures
    • Gini, ROC,KS
    • Information Value
    • Confidence intervals for power measures and curves
  • Stability measures
    • Plots over time
    • Stability Index and Default Stability Index
    • Power Stability
  • Other measures
    • Monotonicity
    • Herfindahl concentration
    • Migration index
  • Interrelationship of measures
  • Economic cycles
    • PIT vs TTC
  • Conservatism
  • Re-calibration
    • Bayes transformation to a central tendency
    • Ongoing re-calibration
    • Pool-based rating systems and pool-wise re-calibration
  • Case studies
    • PD validation and re-calibration

Evaluation and Termination of the Seminar

Training catalogue in PDF
arrow-up icon