By definition, OR is a field that is focused on using mathematics and all its offspring to help us make better decisions in the real world, making it a little ironic that so many practitioners seem to focus on very theoretical (and mostly deterministic) topics. But for those of us that decide to swap the robe and mortarboard for a suit and tie, the world is really your oyster when considering the skillset developed during your university tenure. Every business in every industry needs to make decisions, and many of the more capable ones are using analytical solutions to support in this decision making. And that’s where we come in.
I joined FNB as part of the graduate programme in January 2015 and was placed in the Consumer Credit Scoring team. I was a little apprehensive, as most of my colleagues were from an actuarial or business mathematics background (the BMI stream at North-West University is a big feeder into South African banking analytics) and here I was with a degree no one seemed to know about. But when I took a minute to sit back and look at what we were doing it became obvious that credit scoring is an excellent application of OR in the banking world. It comes down to using mathematical (statistics is just applied mathematics, after all) techniques to help make better decisions.
As a lender, it seems logical that you would want to lend money only to those that you feel confident would pay you back. In the old days, you’d have to present your case to a senior bank representative (like the bank manager) and they would take everything into consideration and judgementally decide on whether to grant you the line of credit. This is obviously a rather time-consuming process and personal bias would often sway their decision. Today, with all the computing power available to us, a more automated system was developed known as credit scoring. The goal of credit scoring is to provide a measure of how risky a customer is when it comes to providing them a line of credit by using historical data to try and predict their future behaviour. We can achieve this by creating a “profile” of the customer consisting of variables obtained from their demographic data or behavioural history, for example their age (demographic), their ratio of savings to expenses (behavioural) and their current outstanding balances on current credit accounts (historical). Using this profile, we assign a score to each variable according to how, in the historical data, it appeared to affect their ability to repay for their credit. As a simple example, we might find that younger customers would default more often than older customers, so we assign you a better score the older you are. All these scores are summed up into a single credit score which is then translated to a probability of default (PD), which is essentially the odds of someone with that score failing to repay their credit. The scores themselves are determined as part of the output of the chosen predictive modelling technique, where logistic regression is often the tool of choice. I won’t go into technical details, but I do recommend reading Naeem Siddiqi’s Credit Risk Scorecards: Developing and Implementing Intelligent Credit Scoring for a well-explained rundown of the entire process.
Once a credit score and PD has been calculated for a customer, it can be used to make several decisions based on the bank’s risk appetite. They can decide that everyone below a certain score is automatically declined, they can scale interest rates to your score, leading to higher rates for riskier scores, and for fixed-term credit like a personal loan they can even cap how long you get to repay based on your score. The entire credit strategy employed by a bank revolves around this number, making it the key operational decision-making tool and an important topic in the field of OR.
It is rather unfortunate that credit scoring is so underrepresented in the OR community, with maybe a small handful of related topics presented in even the larger conferences. While it does not provide you with some deterministic and optimal solution like most OR methodologies, being able to make quicker and better decisions that provide a significant business benefit is definitely worthy of my personal OR stamp of approval.
PS: As a bit of a thought experiment, if you were a bank and you could develop a scorecard that perfectly predicts whether a customer would default, would you use it?