Using Macroeconomic Variables in Credit Modelling

As financial institutions learned during the subprime crisis, the odds of defaulting on a loan do not stay static but vary with economic conditions. And as data scientists, we should remember that a model can only reflect the dataset it was built on. A model trained on data from a certain period will implicitly contain the economic conditions of that period.? The standard methodology for a scorecard in credit modelling is logistic regression, where we predict whether an applicant will default on a loan and the probability is then translated into a score for each applicant. A higher score indicates a lower likelihood of default.

We collect as much information on the applicant as possible to make the most accurate prediction. Still, an overlooked source of data is the state of the economy. Broad economic indicators measure an economy's overall health and performance. Examples of these variables include Gross Domestic Product (GDP), unemployment rate, inflation rate, and consumer sentiment.?

To address the logistic regression shortcomings, we add spicier ingredients to the basic recipe, weights-of-evidence to address non-linear trends and outliers in the data and reject inference to adjust for through-the-door applications. But there is a missing ingredient; if no broader economic variables are available in the modelling process, the assumption is that the past and future economic conditions are the same.

But we know changes can happen rapidly, and? Covid can appear from nowhere and change everyone’s ability to repay their loans. We should thus add a dynamic component to our static scorecard. The methodology compares the macroeconomic variables at the time of application with the predicted probability of default and describes the time-varying part with the macroeconomic variables. Banks introduced the macroeconomic scorecards to address Basel II capital adequacy standards in that if a crisis occurs, does the bank have access to enough capital? But any business can use this technology to do scenario planning.

Adapting a scorecard for macroeconomic conditions, we take the complete economic cycle of a variable and convert it into a quantile, and compute the correlation of the variable to the probabilities of our scorecard, assuming both are normally distributed. The effect of changing economic conditions is then quantifiable on the changes in the likelihood of defaulting on the complete portfolio of loans or if more data is available on any specific sector.

The goal is not to predict future macroeconomic conditions but to accurately describe the current credit risk that a lender is taking and how changing economic conditions will affect it. Narratives are great marketing tools, but the quantitative analysis is far more useful for decision-makers.?

In conclusion, incorporating macroeconomic variables allows a business to adapt credit scorecards dynamically to changing economic conditions allowing more agile decision-making.