AUC (Area Under Curve)

To assess the creditworthiness of potential borrowers, financial institutions use various credit scoring models tailored to their unique needs.

But how can you ensure that the chosen model objectively determines the probability of default? In other words, how can its discriminatory power be evaluated?

To achieve this, creditors often turn to the metric known as the Area Under the Curve (AUC)​. 

‍

AUC and ROC curve: understanding key concepts

What is the Area Under the Curve (AUC)? This is a metric widely used in credit risk assessment. It evaluates the discriminatory power of a credit scoring model, i.e., how accurately the model distinguishes reliable borrowers from those prone to default.

The Area Under the Curve is closely linked to another concept — the ROC (Receiver Operating Characteristic) curve.

‍

The ROC curve visualizes the model's performance at various threshold levels. It takes into account the True Positive Rate (TPR) and the False Positive Rate (FPR).

The area under the ROC curve (AUC) demonstrates the likelihood that the model will assign a higher score to TPR compared to FPR.

In the context of lending, this can be interpreted as follows: AUC reflects the objective ability of a scoring model to correctly classify defaulters and reliable applicants.

In practice, it looks like this:

‍

AUC range

Area Under the Curve is expressed as a numerical value from 0 to 1. It can take the following values:

• AUC = 1.0. This represents a perfectly accurate model that always correctly classifies borrowers.
• 0.5 < AUC < 1.0. The model has some discriminatory power. The higher the coefficient, the greater this power.
• AUC = 0.5. The model operates at the level of random selection.

Based on the obtained values, creditors rank scoring model effectiveness as follows:

‍

Increasing the discriminatory power of scoring models

Credit organizations strive to achieve the most accurate assessment of applicants. To do this, it is necessary to enhance the efficiency of the credit scoring models in use.

So, how to boost predictive power of your credit model?

It is advisable to assess potential borrowers using not only traditional data (such as credit history) but also alternative data.

Consumer digital footprints are deservedly considered the most informative in this situation. For example, their activity on various web resources, online payments, and so on.

This approach to risk assessment positively impacts the scorecards predictive power. This is demonstrated by the RiskSeal’s diagram below:

Here, the following trend is observed:

• For scoring models using only data from credit bureaus, AUC = 0.68. This means the discriminatory power will be classified as “low.”
• In the case of relying only on non-traditional data, AUC = 0.7. This means the predictive power of the model slightly exceeds that of traditional scoring models. It can be considered “acceptable.”
• Combined models show the highest AUC = 0.73. This is the optimal value for lending in conditions where sufficient borrower data is available.

It can be concluded that using alternative credit scoring platforms that assess applicants' creditworthiness based on a combination of data is most effective for the lender.

‍

Related articles

Improving Credit Score Modeling With Digital Footprint Analysis

The 7 Lending Trends Reshaping Credit Scoring

How to Improve Credit Scoring Using Digital Footprints

Top 10 Alternative Credit Scoring Platforms for Lenders

How to Assess the Effectiveness of Credit Scoring Models