Credit Scoring: A New Approach to Control Risk

Credit managers are in a subtle business, as their primary responsibility is controlling risk: Too much credit exposure leads to high default rates and large charge-off ratios; too little exposure leads to lost business and revenue. Fortuitously, credit managers have access to the statistical method of “credit scoring,” which provides the probability of a borrower‘s likelihood of default or delinquency. In other words, credit scoring (aka a scoring model or a scorecard) is a method of evaluating, and therefore controlling risk. Based on today’s gargantuan information – application data, personal and geo-demographic data, and historical credit bureau data – associated with potential borrowers, the data analysis builds the scoring model using an “inflexible” pre-specified (parametric, assumption-full) regression-based procedure, namely, the "old" classical standard ordinary least-squares (OLS) regression model. The working assumption that today’s big data fit the OLS model – which was formulated within the small-data setting of the day over 200 years ago – is not tenable. Accordingly, regression-based scoring models are not optimal.

The purpose of this article is to present a new approach to control risk. It is a non-statistical, machine learning method, a “flexible" nonparametric, assumption-free procedure that lets the data define the form of the model itself. A flexible, any-size data model that is self-defining clearly offers a potential for building a reliable, highly predictive model, which was unimaginable two centuries ago. Specifically, I introduce the GenIQ Model©, a flexible, any-size data method (with unique scalability) that lets the data, exclusive of anything else, define the credit scoring model.