Leading U.S. commercial bank
Banking and financial services
Business need addressed:
Capacity to replace outdated Generation 2 PD, EAD, and LGD models for retail and wholesale portfolios with new Basel II AIRB-compliant models
Built new, more accurate models that are Basel II AIRB-compliant
In addition to meeting federal regulatory compliance obligations and improving the bank’s solvency and stability, the more robust models resulted in more accuracy, better model accuracy, and greater ability to measure risk weighted assets
A $120 billion commercial bank holding company headquartered in the Northeastern U.S. was using a set of Generation 2 Probability of Default (PD), Exposure at Default (EAD), and Loss Given Default (LGD) models built in 2007, all of which were performing poorly. The Gen 2 models were neither Basel II nor Advanced Internal Rating Based (AIRB) compliant, nor did they have the ability to factor in macroeconomics, seasonal impact, or the downturn-period effect. The qualitative factors used in the Gen 2 models were generic and not specific to the bank’s retail and wholesale portfolios, which include retail mortgage, home equity, auto, credit card, small and medium enterprises (SMEs), small business owners (SBOs), and wholesale commercial and industrial (C&I), commercial real estate, construction, and commercial individual loans.
The bank and its subsidiaries have approximately 1,400 branches, more than 3,500 ATMs, and more than 18,000 employees operating in 12 states with non-branch retail and commercial offices in more than 30 states. The bank needed a new generation of models (Gen 3) that would meet new minimum capital requirements and comply with new federal regulatory mandates in Basel II and AIRB. Genpact’s role was to help the bank build Basel II and AIRB-compliant PD, EAD, and LGD models. In addition, Genpact was assigned calibration and validation projects covering multiple portfolios and built a retail and wholesale model monitoring framework.
Before the start of Genpact’s engagement, the bank had performed a gap analysis that identified the shortcomings of the Gen 2 model and the need for a Gen 3 model.
At the beginning of the engagement, Genpact performed a thorough root cause analysis using Lean and Six Sigma principles. Validation analysis—including model accuracy, Kolmogorov-Smirnov test (KS), Gini (income distribution measurement), and population stability index (PSI) statistics—determined that the Gen 2 PD, EAD, and LGD models had very low accuracy, could not discriminate well between healthy and bad accounts, and were not capturing relevant dimensions of different portfolios.
Since the Gen 2 models were not performing as desired, a more robust generation of models that were Basel II and AIRB-compliant and more accurate and could incorporate a broader range of statistical information had to be built.
As an interim solution, Genpact built calibration models until the new Gen 3 models were approved and implemented. In total, Genpact helped the client build more than 20 Gen 3 PD, EAD, and LGD models over a three-year period.
This modeling effort resulted in Gen 3 models that were more accurate thanks to the Gini Index, KS, and regression analysis (R square), and the incorporation of all applicable variables, including data from the recent recession and latest business cycle. Since this was a long-term project, Genpact used the following roadmap:
Genpact built Gen 3 models that were better than the Gen 2 models in terms of accuracy and other key statistics such as Gini, KS, and R square. The new models included business-focused variables and were approved by an expert panel.
Examples include the following:
Home equity loan:
- PD accuracy (error) moved from –64% to 2% from Gen 2 to Gen 3
- LGD model R square moved from 33% to 41% from Gen 2 to Gen 3
Home equity credit line:
- PD accuracy (error) from –70% to –0.3% from Gen 2 to Gen 3
- LGD model R square from 5% to 25% from Gen 2 to Gen 3
Genpact’s initial two-year engagement has been regularly renewed. Genpact has now extended its scope beyond the current team and works closely with the model validation and model monitoring teams as well.
For more information, contact: firstname.lastname@example.org and visit,genpact.com/what-we-do/capabilities/analytics/financial-services-analytics/risk-management