KeyBank Analytics and Quantitative Modeling Rotational Analyst Program Interview Case Studies
As a lender, KeyBank offers clients access to money in the form of loans or lines of credit with the understanding that some customers may fail to repay it. In order to maintain a profitable business, this expected loss must be factored into the credit decision-making process by pricing the risk into the interest rate charged to each client.
To this end, many Quantitative Analysts at KeyBank assist in the design, testing, implementation, and validation of models to assess credit risk characteristics of our borrowers and predict expected loss.
Option 2 focuses on the concept of delinquency. Delinquency rate is a pivotal credit risk metric used to assess the quality and health of KeyBank’s credit portfolios. As an example, a customer who has paid all prior months’ loan bills on time is considered current on their obligations. However, once that customer misses a payment for the first time they become delinquent. As the portion of the portfolio considered delinquent grows, investors and managers quickly take notice.
In this case study, you are tasked with predicting which customers from this month’s pool of “current” customers will become delinquent next month.
Requirements Develop an analytical system or model to predict which Home Equity Loan (HELOAN) customers will become delinquent the following month. Please describe your method in sufficient detail so the interviewers fully understand your process. This can include your complete code, or a thorough description of data treatments and coefficients.
Data Overview You are provided an Excel document including two fictionalized data sets related to Home Equity Loans (HELOAN) for which you will forecast delinquency. The two datasets, Borrower and Loans, are accompanied by a Data Dictionary to define any fields you may be unfamiliar with. The banking terminology follows generally accepted definitions which can also be found online. Six months of data are included, and your approach should be built such that it generalizes well to any given month. The data fields included are listed below: