Engineering credit data has become imperative considering the increasingly risky and competitive lending environment. In order to improve decisioning, credit providers often leverage bureau summary variables which are generally calculated from credit bureau data sources. This includes credit account data, deeds office information, collections and microlending data.
Bureau information gaps
However, there are gaps in the current set of most bureau summary variables. Certain types of behaviour are not captured and this results in improper risk management. A “risk paradox” can result where, for example, a reverse ranking where the observed risk is not in line with the expected risk. Consequently, credit risk models are of limited used due to this deficiency in the variables.
This project aimed to understand and fill the gaps in bureau summary variables. The Matogen Applied Insights (MAI) team reprocessed and imputed the raw payment strings received from the credit bureaus to create new variables that examine consumer behaviour at different overlapping and non-overlapping periods relative to each other. More than a hundred new variables were scoped with a focus on delinquency. The data scientists examined and visualised variable lists focusing on utilisation, payments, balances and instalments to illustrate a customer’s trajectory over time. Traditionally, customers displaying the same delinquency at a specific point would have been assigned the same risk. With additional insights, it is clear that while one might be improving, the other could be deteriorating. This process of engineering credit data makes it possible to assign different risk categories to customers.
Insights from engineered variables
As such, MAI was able to compensate for the shortcomings in traditional credit bureau data and identify payment behaviour trends over different periods. Bureau summary variables is an innovative way of using credit bureau data, which is generally not used to scrutinise historic data from more than one point in time.