Buy Now Pay Later (BNPL) is a short-term, interest-free consumer credit solution that is growing in popularity in the United States. However, none of the pure-play BNPL companies, including the BNPL giant, are profitable despite the explosive growth in sales. As the credit risk underwritten in BNPL is typically higher compared to other credit solutions, traditional statistical models used for credit risk modeling have to be adapted to his BNPL. Machine learning (ML) models have evolved significantly today. Better performance and better prediction of probability of default (PD) and expected credit loss (ECL).
In this blog, we discuss why BNPL has high credit risk and discuss approaches for better forecasting, especially how ML models can predict credit losses more accurately.
Significant credit risk at stake
While BNPL solutions are attractive to both merchants and consumers, BNPL providers are exposed to higher credit risk due to:
- Poor or no credit history: Although BNPL is offered to all segments of society, it is popular among those with poor or no credit history or who are unfamiliar with credit and the consequences of default. This customer segment poses a higher credit risk. According to a recent study by the CFPB (Consumer Financial Protection Bureau), his BNPL borrowers with access to conventional credit are more likely to incur large amounts of debt. The survey sample included consumers with at least one traditional line of credit. BNPL borrowers tend to have higher financial difficulties, higher delinquency rates on traditional credit products, and lower credit scores than non-BNPL borrowers.
- behavioral risk: Customers are generally tempted to buy more than they can afford due to the ability to make instant credit decisions. This impulse purchase increases the risk of default if the customer does not manage their finances properly.
The above risks should be considered when projecting credit losses. The impact of these risks can be better understood by incorporating the following indicators into credit loss calculations:
- Macroeconomic factors tend to affect large segments of the population. It helps predict the repayment capacity of BNPL borrowers, especially those with poor or no credit history.
- Customers tend to leave footprints in their attitudes and actions in social media conversations. Tracking such behavior on social media can help detect high-risk customers.
Impact of macroeconomic factors
BNPL providers typically initiate a soft credit pull for credit decisions. A customer’s credit trade line and credit default data is one of the key determining factors. Apart from these, macroeconomic factors can also be important determinants in assessing credit risk. For example, if the unemployment rate is expected to rise, it may be reflected as “increased credit risk” in credit loss estimates.
Considering these factors, researchers built a machine learning model to calculate the ECL of BNPL portfolios. They found that by including macroeconomic factors, they were able to predict credit losses more accurately.
Social Media to Assess Behavioral Risk
A customer’s ability to repay may change during the term of the BNPL contract due to personal, emotional and psychological factors such as loss of family members. Social media can be viewed as a powerful channel for gaining such insights about your customers. Customers’ attitudes and behaviors are also influenced by their peers within their social circle. A study conducted to determine the impact of social media behavior on default probability predictions revealed that social media behavior data produced more accurate results.
This research can also be extended to BNPL. The social media data below can be used to build ML models specifically for BNPL. This helps determine any changes in a BNPL customer’s behavior while calculating his ECL for the duration of the contract.
- Number of social media platforms used by customers
- Total number of followers and demographics, profiles you follow
- Active Posts, Content Quality, Behavior on Posts
- nomination or sponsorship
- Group behavior regarding credit and payments (delinquency or default)
lastly
With the increase in losses reported by BNPL providers, it is very important to build robust mechanisms to more accurately predict expected credit losses. This helps improve the credit decision-making process. Financial data is augmented with non-financial data such as social media behavior and macroeconomic factors to accurately predict credit losses. Using the above machine learning models for prediction allows for more accurate credit risk provisioning.
