Current credit bureau analytics are based on slowly varying consumer characteristics, and are therefore not as relevant for tactical risk management decisions. This project aims to construct forecasting models to improve the classification rates of credit-card-holder delinquencies and defaults. Using account, transaction, loan, customer, and credit bureau data provided by a commercial bank, we apply machine learning algorithms such as random forests, logistic regression, and support vector machines and investigate models across products and time. We will then evaluate the model performances using metrics such as out-of-sample forecasts, ROC curve, and contingency table. We will present the models for applications in consumer credit and macroprudential risk management.

I worked at Yahoo on entity reconciliation and clustering using natural language processing and machine learning methods. I also worked at Bloomberg R&D on enhancing the pricing engines and revamping the graphical user interface for commodity fair value curves. For the past two years I worked on a UROP with Professor Lo and evaluated the pharmaceutical and biotechnology industries returns and risks.