Credit Scoring: Which Personality Traits Predict Credit Risk?

This is a commentary by Javier Frassetto, Vice President of Modeling and Data Science at FICO EFL Global . EFL recently announced its merger with Lenddo. Our work with EFL and Lenddo is part of the FICO's Financial Inclusion Initiative .

Lenders rely on credit ratings to assess consumer risk, and credit ratings are based on credit data. But what happens if a candidate is new to credit?

EFL offers financial institutions a different way to assess creditworthiness and promote financial inclusion: by understanding the personality. Everyone has a personality that can help us understand their risk profile, making universal EFL assessment – we can score points.

To date, EFL has shared very little of the traits that characterize our credit models. Today, we share knowledge on two key features of risk assessment: impulsivity and deferred gratification.

An Alternative to Credit Data

For over 10 years, EFL has developed a credit assessment using psychometric and behavioral data that allows financial institutions to better understand candidates and make better credit decisions. The evaluation evolves and improves constantly. It started as a multiple choice test taken with a pen and paper and became a gamered digital assessment. The most recent version has a median completion time of 15 minutes and ranges from 6 to 10 sections.

We face two main challenges using alternative data for credit assessment:

Predictive power: The data provided must be able to tell the difference between good and bad borrowers.
Evaluation Time: As time goes on, the attention of candidates decreases and productivity decreases for financial institutions.

Because of these challenges, the design of the EFL assessment is based on three guiding principles:

Maximum time limit – the content of the evaluation must be designed to have a maximum time associated with each exercise. This ensures a predictable schedule for evaluation, easing the operational burden for financial institutions.
Rich in data – each exercise needs to capture multiple traits, relying heavily on metadata and interactions to capture as much information in as little time as possible.
Text – the content should be designed with minimal text to allow illiterate candidates to easily interact with the assessment.

The EFL evaluation captures more than 25 personality traits. The most relevant are locus of control, fluid intelligence, impulsiveness, confidence, delayed gratification and consciousness. These features allow us to identify the applicants who are likely to repay their loans.

Let's take a look at two examples.


Impulsivity, the trait of acting on impulse without thinking, is measured in many exercises within the EFL assessment. A clear way to see impulsivity in evaluation is the speed match exercise. In this module, borrowers play an interactive game that consists of quickly clicking on a series of images and selecting whether or not the image matches the previous image. (Original design of the module inspired by Sternberg et al., 2013 and Morrison et al., 2015).

<img class="alignleft size-full wp-image-32771" src="" alt=" Screen capture of the EFL test "width =" 716 "height =" 476 "/>

(This is only one example: in the actual exercise, the plaintiff only sees one picture but not if the choice was correct.)

Applicants are asked to rate 20 images in a specific time window. This generates an environment in which the candidate must decide whether he should move faster and make mistakes, or move more slowly without making mistakes but not complete the 20 images.

EFL data showed that candidates who move faster with an impulse to complete the application and get false answers are riskier borrowers than those who take longer to respond with fewer errors. The figure below shows that impulsive candidates have higher failure rates than other candidates. The relationship between the trait and the defect remains stable and the division between the groups can be as high as three times. (The data come from two different financial institutions in two countries where the EFL program was deployed: the first graph comes from a Latin American country and the second from an African country.)

<img class="alignleft size-full wp-image-32772" src="" alt=" EFL test results "width =" 737 "height =" 375 "/>

Deferred gratification

EFL measures deferred gratification, or the ability to delay something now in order to get something better in the future, through the financial compromise module. This module primarily evaluates the customer 's discount rate by measuring the amount of money that she would be willing to forgo receiving an immediate payment rather than a deferred payment. Candidates demonstrate their financial preferences by making several selections between an immediate hypothetical reward and a larger deferred reward. (Module designed in partnership with Hal Hershfield (UCLA) Initial module design and estimates by Weber (2007), Weber et al (2013) and Meier and Sprenger (2010).)

<img class="alignleft size-full wp-image-32773" src="" alt=" Screen capture of the EFL test "width =" 716 "height =" 476 "/>

Discount rates are measured in different rounds for different delays. The time discount rate (as measured in the module) has been correlated to things like saving rates, (non) smoking, healthy diets and other behavior-oriented behaviors. self.

Data from different EFL deployments show that candidates with higher discount rates are riskier than those with lower discount rates. Figure 2 shows how this exercise was applied in two different populations. The first graph refers to an African population and the second to a Latin American population. The refresh rates and the default value follow the same relationship and the split between groups can reach 2.1 times.

<img class="alignleft size-full wp-image-32776" src="" alt=" EFL test results "width =" 826 "height =" 422 "/>

Opening the door to credit

Most temporal characteristics are not measured in a single fiscal year as part of the EFL credit assessment, but in many across the assessment. These two examples intuitively show how different perspectives of a candidate can actually be used to predict credit risk.

EFL models generally rely on 10-12 specific traits. We have found that some populations reveal more information about themselves than others, which means that our credit models work differently across segments; However, when financial institutions are able to assess enough candidates, a strong predictive model can certainly be built.

By assessing impulsivity and deferred gratification, two universally available traits, we can improve financial inclusion and open the door to funding for millions of people around the world – people who do not have from credit history but who certainly have a personality to mark.

Leave a Reply