Credit card business implications for comprehensive capital analysis & review (CCAR)

This paper is designed to help big financial institutions meet US regulatory compliance with respect to Comprehensive Capital Analysis and Review (CCAR). This information is suited especially to domestic or international banks doing business in USA. I designed a Credit Card CCAR Loss Modelwhich is designed to measure risk and forecast expected credit losses under different economic scenarios

Objective: This model is required for regulatory purposes to support capital planning and ensure that the financial institution has sufficient capital to continue lending to support real economic activities while meeting their financial obligations, even under the stressed and severely stressed economic conditions.

Data: I have taken historical month end outstanding credit card balance data of a large US financial institution which has significant number of credit cards in its assets holdings. The data span 7 years and then 1.6 years of data all used for validating the previous model results. I used open and closed-with-balance loans from January 2007 to September 2012 and validated on an out-of-time sample consisting of October 2012 to March 2014 data.

Methodology: I used an Expected Loss framework, which consists of three component models: (i) the model of Probability of Default (PD), the model of Exposure at Default (EAD), and the model of Loss Given Default (LGD).  For the PD component, I created model segments using FICO score and Payment Ratio.  I developed different segments for Consumer and Business cards.  For each segment, I applied a two-step estimation approach by calculating economically neutral PD predictions using long-term average transition matrices, followed by econometric modeling to quantify the economic impact using macroeconomic variables.  For the EAD component, I created model segments using Credit Line, Credit Balance and Time elapsed from the Observation Month.  Similar to the PD model, Consumer and Business Cards have separate segments. For the LGD component, after much research I abandoned the segmentation scheme because test results showed no significant improvement in predictive accuracy.

Results: In summary, the Credit Card CCAR Loss Model is a segment level approach for evaluating and forecasting PD, EAD and LGD model components. By combing the PD, EAD and LGD components, I provided an economically sensitive forecast to expected losses and required capital for meeting CCAR standards. Under stressed and severely stressed economic situations the model predicts higher defaults and losses in credit cards payments by borrowers, hence the need for additional capital by financial institutions.