Featured case studies
IRB models validation case studies
IRB model validation and ECB reporting
Background
- Client: Tier 1 European bank
- Context: the Bank implemented a new validation framework for IRB models and submitted the approval request to the ECB
- Problem: design, set-up, implementation and automation of the overall validation process (both internal validation and ECB reporting requirements); delivery of internal validation results and validation templates to the ECB
People
- Sponsor: Chief Risk Officer
- Stakeholders: ECB
Approach
- Methodology: we developed, implemented and automated the new internal validation framework leveraging best practice tests and methodology available in the MP; the ECB yearly validation reporting process has been implemented, in line with ECB requirements, and fully automated
- Delivery: first run of the new internal validation process and ECB reporting; delivery of internal validation results and validation templates to the ECB for all the IRB models/segments (PD, LGD, LGD DA, ELBE, CCF/EAD, Slotting)
Value delivered
- New internal and ECB validation framework approved, implemented and automated
- Validation module replaced, lowering the cost
IRB model validation
Background
- Client: Tier 1 European bank
- Context: the bank needed to enhance its internal validation framework for IRB models, integrating new tests and refining the thresholds for the traffic light approach; moreover, given the high number of models to be validated, a full automation of the process was required
- Problem: enhance the existing validation framework with best practices and benchmarks; automate the whole internal validation process
People
- Sponsor: Chief Risk Officer & Chief Lending Officer
- Stakeholders: ECB
Approach
- Methodology: we analysed and improved the existing validation framework via: replacement of unsuitable tests, inclusion of new tests to enhance the framework (e.g. PSI joint with Chi-Square, Anchor Point cycle-dependent, etc.); thresholds for traffic light approach have been defined/refined and customized on the portfolio; the process has been fully automated
- Delivery: first run of the new internal validation process; delivery of internal validation results to the ECB for all the IRB models/segments (PD, LGD, LGD DA, ELBE, CCF/EAD, Slotting)
Value delivered
- New internal validation framework approved, implemented and automated
- Validation reports delivered and approved
- Validation module replaced, lowering the cost
IFRS9 model validation case studies
IFRS9 model validation
Background
- Client: Asian Tier 1 bank
- Context: the bank developed all IFRS9 models in order to meet the regulatory requirements of the local Regulatory Authority aimed at starting the calculation of provisioning according to the international IFRS9 standards
- Problem: design, set-up, implementation and automation of the overall validation process for IFRS9 models; estimation of provisioning and reporting to the local Regulatory Authority
People
- Sponsor: Chief Financial Officer & Chief Risk Officer
- Stakeholders: Local Regulatory Authority
Approach
- Methodology: we developed, implemented and automated the internal validation framework leveraging best practice tests and methodology available in the MP (following, where relevant, the best practice for IRB models validation)
- Delivery: first run of the internal validation process and IFRS9 reporting
Value delivered
- Internal validation framework set-up, implementation and automation
- IFRS9 reporting process implemented
IFRS9 model validation
Background
- Client: Development Bank in Asia
- Context: the aforementioned Development Bank, part of the biggest Banking Group in that country, developed all IFRS9 models in order to meet the regulatory requirements of the local Regulatory Authority aimed at starting the calculation of provisioning according to the international IFRS9 standards
- Problem: design, set-up, implementation and automation of the overall validation process for IFRS9 models; estimation of provisioning and reporting to the local Regulatory Authority
People
- Sponsor: Chief Financial Officer & Chief Risk Officer
- Stakeholders: Local Regulatory Authority
Approach
- Methodology: we developed, implemented and automated the internal validation framework leveraging best practice tests and methodology available in the MP (following, where relevant, the best practice for IRB models validation)
- Delivery: first run of the internal validation process and IFRS9 reporting
Value delivered
- Internal validation framework set-up, implementation and automation
- IFRS9 reporting process implemented
PD model development case studies
Development of PD models
Background
- Client: large European corporate
- Context: the company wanted to set-up its risk management unit and planned the development of rating models for the SME portfolio in order to optimize its credit management process
- Problem: design, develop, validate and implement rating models for SME client alongside with training to the personnel and automation of the development process for next developments
People
- Sponsor: Head of Risk Management Unit
- Stakeholders: Risk Unit & Credit Unit
Approach
- Methodology: we designed, estimated, tested and implemented three rating models for SME portfolios (according to the company segmentation); we used the dedicated models library of the MP and built models in line with best practice
- Delivery: rating models for SME delivered and implemented, allowing an automated estimation process (to a significant extent) for the next re-developments to be carried out by the company
Value delivered
- Full set of rating models to be used for managerial processes
- Estimation process engineering and training to personnel
Development of PD models
Background
- Client: consortium managing models IT infrastructure and services for a significant number of small and medium-sized regional banks
- Context: the Consortium aimed at centrally developing managerial PD, LGD and CCF/EAD models to be implemented used by the member banks for their managerial processes (including IFRS9 and Pillar 2)
- Problem: design, develop, validate and implement all the set of Credit Risk models for all portfolios of the member banks
People
- Sponsor: Head of Analytics and IT
- Stakeholders: member banks
Approach
- Methodology: we designed, estimated, tested and implemented credit risk models (typical Pillar 1 models) for all the portfolios of the member banks, developing and/or calibrating them by bank/group; we used the dedicated models library of the MP and built models in line with best practice and typical regulatory requirements
- Delivery: models for all portfolios and member banks have been delivered (including implementation) and are up and running for daily processes, provisioning calculation and Economic Capital estimation
Value delivered
- Full set of models for IFRS9, Pillar 2 and managerial processes implemented
- Models customized for banks or group of banks
Machine Learning model development case studies
Development of ML transactional score
Background
- Client: Tier 1 European bank
- Context: the bank aimed at enhancing its IRB models for one portfolio with an additional module, based on a new data source and innovative methodology (ML models on transactional data)
- Problem: design and develop the transactional module via Machine Learning techniques, to be integrated later as a modular add-on to the “core” components of the IRB PD model for Small Business; being an IRB model, all the typical requirements for IRB models development and validation had to be met (including interpretability)
People
- Sponsor: Chief Risk Officer & Chief Lending Officer
- Stakeholders: ECB
Approach
- Methodology: we designed and implemented a new module for IRB models, leveraging the proprietary libraries of Prometeia Modeling Platform dedicated to transactional score and ML models; the module uses very granular information of current accounts and credit card transactions of SB clients and has been developed by building over 35,000 variables and using XGBoost; an intense work for the model interpretability has been done
- Delivery: model delivered to the Bank and implemented for credit processes purposes, model submitted to the ECB for IRB approval (the switch-on/switch-off component allows to neutralize the new module and apply only with the core IRB modules in case of issues with the ECB due to ML techniques used during the development)
Value delivered
- Sharp increase in Accuracy Ratio (+10%) wrt the standard IRB model
- Model much more reactive (especially during the Covid-19 pandemic, where typical models do not work properly
- Submission to the ECB
Development of ML transactional score
Background
- Client: Tier 1 European bank
- Context: the Bank aimed at enhancing its IRB models for the Private Individuals portfolio with an additional module, based on a new data source and innovative methodology (ML models on transactional data)
- Problem: design and develop the transactional module via Machine Learning techniques, to be integrated later as a modular add-on to the “core” components of the IRB PD model for Private Individuals; being an IRB model, all the typical requirements for IRB models development and validation had to be met (including interpretability)
People
- Sponsor: Chief Risk Officer & Chief Lending Officer
- Stakeholders: ECB
Approach
- Methodology: we designed and implemented a new module for IRB models, leveraging the proprietary libraries of Prometeia Modeling Platform dedicated to transactional score and ML models; the module uses very granular information of current accounts and credit card transactions of PI clients and has been developed by building over 30,000 variables and using XGBoost; an intense work for the model interpretability has been done
- Delivery: model to be delivered to the Bank and implemented for credit processes purposes, model to be submitted to the ECB for IRB approval (the switch-on/switch-off component allows to neutralize the new module and apply only with the core IRB modules in case of issues with the ECB due to ML techniques used during the development)
Value delivered
- Significant increase in Accuracy Ratio
- More reactive model
- Submission to the ECB
Meet the QuanTech network
The network is good way for specialists to engage in discussions on topics of interest and increase the amount of connections and knowledge. There are opportunities to participate in exciting projects and create new methodologies, prototypes and tools.