Predictive analytics for consumer lending: Managing portfolio risk

Predictive analytics is the art and science of analyzing historical data to create mathematical models that can be used to gather business insights for effective decision-making. The utility of predictive models, however, goes far beyond the predictive capabilities of the model. The key to maximizing the utility of the predictive models lies in the interpretation of the results (insights) and the ability to apply these insights in the business context. In this article, we will discuss two opportunities in which lending institutions could harness the capabilities of predictive models.

Predictive models as risk management tools

In the lending industry, predictive models make for natural risk management tools. Predictive models attempt to quantify a borrower’s creditworthiness by placing them on a numerical scale, rather than broad categories such as, “good” or “bad”. By quantifying a borrower’s creditworthiness, a lender is able to place each potential borrower into more robust categories like the one shown below. The gives the lender, the opportunity to perform more detailed analysis of the entire portfolio and construct a variety of policies that are customized to maximize profits.

      Borrower Spectrum

Predictive models for Risk Diversification


With a hypothetical economic downturn looming, a lending institution can adjust overall portfolio risk and disburse loans conservatively.

With a two-category system (“good” or “bad”), the institution foregoes the opportunity to disburse more loans products to qualified candidates. Borrowers who are classified as barely “bad” borrowers are not left out. Using a quantitative risk index allows lending institutions to make decisions at the margin, more effectively resulting in the disbursement of loans to larger number of potential borrowers based on a quantified risk.

Predictive models can be used effectively for risk diversification. For example, predictive models which are constructed with geographic features opens up the opportunity for a lending institution to diversify across multiple geographical locations. These models can also serve as a hedge against regional catastrophes that could include natural disasters and economic disaster.  Another model could enable categorization by income levels or industry verticals to hedge against governmental policies that could affect individuals based on their income and the nature of their professions.

Identify growth areas using predictive models

Predictive models can also help a lending institution manage growth. As a result of the saturated state of the lending industry, it is critical to identify areas for growth. Business Management PlanPredictive models created can provide data-driven recommendations for growth.

Developing business management tools using predictive models requires the combination of business data and the predicted results. In short, the model predictions are used to create innovative data visualizations that take into account the lending institution’s current and forecasted state which are in turn analyzed to produce targeted policies and growth plans.

A predictive model can provide the insights required to manage and monitor controlled growth of the business. If a model includes, both, seasonal and geographic features, then it is capable of finding patterns that take both these factors into account. It might be able to forecast the season and the state at which growth is highly likely. For example, it might suggest the summer months are the best time to target the state of California as a nascent market opportunity. Depending on the frequency of data updates, this could even be done in weekly intervals. In addition, by taking into account other features in the predictive model, a more specific growth plan can be identified. For example, with the appropriate features, a predictive model can suggest the best state, time, and income levels to target for growth (i.e. construct policies to target individuals in California during the summer who make $80,000 to $100,000).

The accuracy and utility of predictive models is highly dependent on the amount of data available and the features it contains. Though a problem decades ago, this dilemma is hardly an issue today as the amount of data being collected seems to multiply exponentially over time. The flexibility and compactness of predictive models allows them to be packaged into fully functional platforms and applications that provide the user to manage the business from anywhere using a variety of devices. Imagine being able to run your business effectively on your smartphone while relaxing at a beach!


Leave a reply

Your email address will not be published. Required fields are marked *