Providing credit to self employed and micro businesses in emerging markets is difficult. Usually these kinds of businesses cannot provide the required documentations needed by lenders to correctly evaluate the lending risk. One solution to overcome this lack of information is to form lending groups. The peer pressure and the shared liability in such groups leads to better loan performances also in difficult market conditions. This lending model is successfully being applied worldwide.
While internal analytics and/or credit bureau data can lead to satisfactory scoring performance for individual borrowers, the risk for group lending is more complex to model. Standard analytics lead here to rather poor scoring results.
About our Client
Our client (Group Lending MFI – GL MFI) is one of numerous lenders in its home market, Mexico. Out of a total population of more than 130 million in Mexico, 60% are unbanked and it is estimated that more than 27 million of Mexican adults are using micro credits –loans smaller than USD 500. Since smaller MFIs are currently facing less regulatory requirements, a dynamic credit market with high demands for credit products has evolved.
GL MFI has been set up in 2017 and currently counts more than 50,000 active clients with thousands of credit applications every month. The average loan size for short term loans (up to 4 months) is USD 3,000 per group which is split between usually 5-9 members. GL MFI sticks out from its competitors with high growth rates in its lending business. The underwriting process relies on agents as the focal point for credit applications and financial transactions. Right from the beginning of this new business initiative, GL MFI made efforts to centralize decision making and storing of data.
In order to safeguard these achievements and to facilitate future growth, GL MFI was looking for improvements in its credit underwriting.
Paretix was contracted to upgrade the current decision-making process. Our suggested solution concentrated on the improvement of the scoring algorithms as well as generate credit offers automatically. Early in the process, we realized that most of the value could come from improvements in the current credit decisioning process which relied on internal analytics in combination with credit bureau data.
The fundamentals of credit scoring are well published and have been researched for decades. While the availability of affordable computational power has led to a leap in the theoretical modeling performance, actual gains are only seen in extremely large data sets which are not common in credit. The performance gain had to be achieved by other means…
When we first analyzed the behavior of the groups over time, the changing composition of the group was what caught our interest. We realized that by concentrating our efforts on finding a better solution for this segment, we could make the impact that our client was looking for.
A traditional scoring approach focused on individuals will not make use of the rich information contained in the composition of the group. According to our findings, up to 60% of the explanatory power is provided by ‘group cohesion’ characteristics such as group composition and joined credit history. Using these characteristics for scoring not only increases the overall performance but also greatly reduces the dependency on external data (e.g. credit bureau).
Analyzing the social relationships in group credit and using the contained information value for scoring, provides an opportunity to tap into new markets. Potential borrowers that have no documented income and no physical assets in their possession can use their social capital (relations with other market participants) to qualify for credit offers.
In the future, our Paretix Instant Lending App, which can access data from smartphones, will unlock this social capital even for customers new to lending.
The implementation of the algorithms to create and score subsets is challenging. The Paretix Instant Lending Platform provides the ideal environment for implementing such procedures and at the same time to profit from the additional features like integrated back office, product configuration module and flexible ETL tools to connect to the different data sources. In addition, it features a fully featured dashboard which provides access to KPIs in real-time.
By integrating the Paretix Instant Lending Platform into the credit origination process, the full analytical potential will be made available to each single credit decision. The built-in machine-learning capabilities ensure that performance is kept over time with low maintenance requirements. Upgrading the platform with the Paretix Lending App in the future, will enable instant loan decisions and disbursement possible while further reducing the need to access external data sources.
After integrating the Paretix solution into the application process, overall defaults are expected to go down by 20% for a similar acceptance level. The overall model performance for segments with dynamic group compositions reached a 10% higher GINI.
It is expected that following implementation, the overall turnaround time for a credit application will go down from several days to several hours.