Case study
IFRS 9 expected credit loss models
PD, LGD and EAD models, Basel capital calculations and the data pipelines behind them, for a large multi-entity portfolio of a global banking group.
- Client
- MUFG
- Period
- Mar 2022 – Aug 2023
- Role
- Consultant, then Senior Consultant, Data Science & Strategy at Management Solutions
- Where
- Dallas, on-site and remote
Context
My last engagement at Management Solutions was in Dallas, for the U.S. operations of MUFG. The objective was the expected credit loss (ECL) models under IFRS 9 for a large, multi-entity portfolio within the group’s banking hub.
Challenge
Build the full chain of credit loss models (probability of default, loss given default and exposure at default) on data that lived in multiple source databases, in different languages.
My role
I built the models and led the data work behind them, produced the reporting on their results, and managed one team member.
What I did
- Probability of Default (PD) and recovery rate. Built both with logistic regression, deriving Loss Given Default as LGD = 1 − recovery rate.
- Exposure at Default (EAD). Designed a two-stage model: a logistic regression predicting whether exposure would be zero or positive, followed by a regression model estimating the amount.
- Capital. Calculated unexpected losses and capital requirements using the Basel regulatory formulas.
- Data. Much of the invisible work was here. I led the cleansing and centralization of data from multiple source databases in different languages, and built the ETL processes that fed the loss models.
- Reporting. Produced regulatory and management reports on the model results.
Results
- A complete set of IFRS 9 ECL components (PD, LGD and EAD), with Basel unexpected-loss and capital calculations built on them.
- A centralized, multi-source data pipeline feeding the loss models.
- Regulatory and management reports on the results.
- In September 2022, during this engagement, I was promoted to Senior Consultant.
I left Management Solutions in August 2023 with a solid foundation in risk modeling, banking regulation and direct work with executives at major U.S. banks.
Technologies
- Logistic regression
- Regression modeling
- ETL
- IFRS 9
- Basel