Case Study — Micro Financing
Credit Risk Analysis
Credit risk modeling refers to data-driven risk models that calculate the chance of a borrower defaulting on a loan (or credit card).
Traditional loan approval methods include looking at a credit score, collateral, inquiry of the borrower via agencies, etc. But with the advent of the internet, that captures the lifestyle of the users, and the ease of running manipulations on them using Big Data Tech has expedited the loan approval process efficiently.
The loan approval process in large banks/corporations is not the topic of discussion here. Since, they have sophisticated agencies and models that grant loans to people with considerable collateral, credit score, etc. Rather, this article focuses on how credit risk analysis can be done in micro-credit unions(micro-financing).
Understanding Micro Financing
Microfinancing is a banking service that is provided to unemployed, low-income individuals, daily wage workers, or people who have no other access to financial services.
It gives the impoverished people an opportunity to become self-sufficient.
People who live on less than Rs 200 a day and attempt to save, borrow or acquire credit from loan sharks falling into the never-ending whirl of debt, are usually the customers of microfinance, which helps credit them loans with ethical lending practices.
Most of the micro-financing operations happen in developing countries like India, Bangladesh, Indonesia, etc.
And of course, like conventional lenders, micro financiers charge interest on loans, which are far lower than the conventional banks
The credit can be as low as 500 Rupees.
Data to analyze the creditworthiness of the borrower
The data to understand the borrower can be broadly categorized into two.
I. Conventional Features
II. Beyond the Conventional Features (Alternate Data)
The combined use of both conventional data and the data from alternative sources can enhance decision-making.
Conventional Features
Data that is inquired from the borrower or can be an inference from his geolocation.
Age
Marital Status
Number of Dependants
Assets
Expenses
Earnings in the last few months (since the income is irregular)
Number of previous credits
Number of defaults on previous loans
Climatic/Political/Economic conditions of the place
Beyond the Conventional Features
These are the features that can be extracted from public sites, social networks, mobile recharges, bill payments, etc, which help in upping the game for understanding the financial lifestyle of the borrower.
The study of psychometrics, which is primarily related to individuals’ knowledge, abilities, personality traits, etc, can lead to some significant key drivers for credit risk assessment. This can also be crucial info in the case of first-time borrowers.
Some of the borrower aspects that can be exploited are
Places visited
Friends circle
Likes/Dislikes
Followings
Followers
Online Shopping
Search History
Bill Payments
Clickstream Data, the way customer moves in your page, where they click and how long they take on a page
And many more…
Major Challenges
The sheer volume of the alternate data to be scraped and extract the insights out of them.
— This can be taken care of by using the concepts of Big Data
It is unlikely that a borrower would share all his social network details, bill payments, mobile browser details, etc, either voluntarily or upon a request. This poses a major challenge in gathering the alternative data of the borrower.
— There are few techniques with which this can be achieved. You will know them as you go through this article.
Insights on Alternate data done by Fintech
- Folks who use their true names in their email addresses are better bets than those who jumble common nouns and numbers.
- Shoppers who click paid online ads before ordering are twice as likely not to pay for what they buy as those who spent time on price-comparison sites first.
- T-Mobile customers are less of a credit risk than Gmail users, who are less than Yahoo! email users.
These indicate how valuable the alternate date insights can be whence exploited
Extracting borrowers’ alternate data
Social Mining: Using Gmail/Phone Contacts
- Facebook, Twitter, Snapchat, and many other social sites have an incorporated functionality that allows you to see whether your Gmail/Phone Contacts are using these services.
- Well… Although this integration is intended for personal use, it can be utilized for your business too.
- All you are supposed to do is save the Gmail/Phone number in your respective Gmail Account/ Phone with which you are signing up on the social platform.
- This built-in functionality of the social outposts will allow you to track/follow/befriend your customers. Which is intrinsically mapping your saved contacts in your account with their respective social accounts.
Online History Mining: Using cookies
- What is a cookie? A cookie is information saved by your web browser. When you visit a website, the site stores a cookie so it can recognize your device in the future. Later if you return to that site, it can read that cookie to remember you from your last visit. It keeps track of you over time, which helps customize your browsing experience, or deliver targeted ads.
- Cookies can be placed by the site you visit or by the ones who are partnered with the website.
- Over time, these companies may develop a detailed history of the types of sites you frequent, and this information can help understand the lifestyle of the borrower.
Finale
There are enough data sources to build an ML mechanism that will be able to extract sensible & usable information from the ocean of data. With large, unstructured data sets, the right use of ML & AI can help identify data patterns that relate specifically to credit risk.
The credit industry can acknowledge the importance of financial inclusion and the lending gap it needs to close. Credit information companies have started seeking credit scoring models that include alternate data to establish a credit score. Tapping the volcano of this hidden and never-tapped data should be a priority in micro-financing institutions.
Now that we have got a hold of both, the conventional and the alternate data, all we need to do is put forth the steps involved in building a typical classification type ML Model.
It seems just like a quiz that has simple questions based on common daily situations. The way a person responds can give a view of the borrowers’ creditworthiness and loan repayment intent.
