Platform Engineering

Anindya Roy

Credit Assessment & Social Data — Part 2: Using Social Indicators

Posted by Anindya Roy on 22 October 2018

process, credit, strategy, banking

Following my previous post (Part 1 - Traditional Credit Rating) you hopefully now know your credit score and you have mastered the art of managing your credit score. Now brace yourself — what if I tell you that you probably shouldn’t be too excited about it? Read on.

Problems With Traditional Credit Scoring

The traditional credit rating model has a fundamental problem – it tries to assess a borrower based on his / her financial history. What happens when someone doesn’t have a financial history? In many developing countries, a large section of the population still remains outside the formal financial system, without any financial history. It is hard to assign a credit score to such individuals. Note that income is not a factor in credit scoring so there is no correlation between their low income and their creditworthiness.

Consider another scenario. A fresh graduate with a new job is applying for a credit card and doesn’t have a financial history in the credit file. What is the likelihood that the person will be a good borrower and will pay back future debts? It is hard to predict with no existing financial precursor for the person, and it is a big risk for the lender. Will the lender refuse to lend outright? It may choose to, but then it would have the downside of losing out on potential new customers. Lenders want to offer their products to greenfield customers, but only to ‘good’ borrowers.

So how can lenders be sure?

Using Social Data to Assess Creditworthiness

In today’s digital world, information is abundant and all around us. Social media contain a vast source of information about individuals in modern society. Social information refers to data which is obtained from internet and mobile based services including social networking sites (Facebook, Twitter, LinkedIn etc.), blogs, wikis, online forums, discussion groups, and media-sharing sites. The information contained in these channels are easily accessible and can be easily analysed to derive opinions about consumers. Such information, outside the limited set of their available financial information, tells lenders more about their potential customers. This is a key trend which increasingly aids decisions about credit worthiness of a consumer during the underwriting process.

The Effectiveness of Social Data1

Penetration:

Over the last decade, global availability and improved speed of internet connections, improved software and penetration of handheld smart mobile devices have accelerated the penetration and usage of social media. They have continued to gain widespread acceptance from all demographics in varying degrees. Some widely available statistics (as of Jan 2018) below prove their popularity:

  • Facebook has an estimated 2.2 billion active monthly users
  • YouTube has an estimated 1.5 billion active monthly users
  • WhatsApp and Facebook Messenger each has an estimated 1.3 billion active monthly users
  • Instagram has an estimated 800 million active monthly users
  • Twitter has an estimated 330 million active monthly users
  • LinkedIn has an estimated 260 million active monthly users

There are many more other popular social media. These eye-popping numbers go to prove the acceptance, popularity and penetration of social media in our social and economic lives.

Ease of Accessibility:

A third of the world’s population interact through these channels, generating a vast amount of data daily about their lives, preferences, demographics, abilities, affiliations — almost everything! This information, in most cases, is free and shared voluntarily. The information trail left behind by netizens is easily accessed and processed using powerful data science tools, which help create a comprehensive profile about consumers.

Using this abundantly available data, financial institutions are discovering new ways of working. Globally we have started to see interesting examples of using social data like:

  • Penetrating new customer segments which have limited financial history
  • Understanding customer needs through analytics and thereby providing a differentiated customer experience
  • Strengthening the current underwriting processes using additional social data points and thereby reducing losses from bad credit
  • Preventing fraud by using social data to cross reference against loan application information
  • Developing a competitive edge through an overall improved operational process

Research on Social Data

Researchers are now paying more attention to the use of social data for credit scoring. A few of these studies are highlighted below:

  • Daniel and Grissen (2015) analysed behavioural patterns from mobile phone data to conclude that banks can reduce credit defaults by 41% while still accepting 75% of applicants by analysing mobile phone data. The parameters used were age, gender, top-up & depletion patterns, mobility, pattern of handset use, strength & diversity of social network connections, intensity & distribution over space & time, loan size and loan term
  • Masyutin (2015) analysed data from Russia’s most popular social network to distinguish between solvent and delinquent debtors of credit institutions. He proved that social indicators are better predictors of fraud, and are ideal to supplement the classical credit scoring model. The parameters from social data used were age, gender, marital status, number of children, time since last visit, number of subscriptions, time since first post, number of user’s posts with photos, number of user’s posts with videos, major things in life and their major qualities
  • Ntwiga (2016) analysed social data to predict likely defaulters in a loan portfolio, by considering the time dependency and cyclical interdependencies among individuals and studying a set of social parameters which are age, gender, trust, interactions, risk factor, sociability, relationship strength, distrust, private data and return on private data.
  • Wei, Yildirim, Bulte and Dellarocas (2015) compared scoring accuracy of multiple models with and without individual’s social network information. They concluded that considering individual’s network information is beneficial, as there is an above average chance of individuals socially interacting with others having similar credit worthiness.

Emergence of Social Credit Rating Companies

Aligned with the increasing popularity of using social data for credit scoring, there is a breed of emerging companies all around the globe. Some of the most popular companies who provide credit score using information available on the internet are:

https://www.lenddo.com
https://www.neoverify.com
http://www.friendlyscore.com
https://www.affirm.com
https://www.kabbage.com
https://www.kreditech.com

There are many more upcoming companies exploiting this emerging area of financial lending. Websites used by these companies are the popular social websites including Facebook, Twitter and LinkedIn. They apply different algorithms on the data to analyse them and to provide a rating. The algorithms are proprietary to the companies’ business models.

Challenges of Using Social Data

Although a popular and emerging area, social data mining also presents us with challenges, some of the most common ones are:

  • Entity Resolution – Searching and extracting online information could be inaccurate, simply because there could be multiple entities having the same search parameters like name, last name, first name, etc. Hence there are challenges in identifying the correct real world person or entity from a set of multiple virtual matches. Users can apply probabilistic approaches in such cases and they must use the right model fit for purpose.
  • Social Awareness & Reputation Management – With increasing penetration of social media in our lives, we are becoming more socially aware and educated. Many people control their own information about what and how to project themselves, in order to maintain a ‘favourable’ social status. This means the virtual selves only project a fractional and amplified favourable portion of their real selves.
  • Information Extraction – Although techniques exist to extract human characteristics from messages posted on social media, the extraction of information from online channels is a challenge due to many factors like privacy settings, use of colloquial or informal texts, language etc. It requires sophisticated data extraction tools and their constant improvement to keep up with the evolving complexities.
  • Correlation and Algorithm – There is further research1 (examples: Dewing, 2012; Dubois et al., 2011) which show that the impact of social media on how people interact and share information is still not fully understood. We are not completely certain which of the available social parameters (or a combination thereof) accurately correlates with credit worthiness. There are opportunities to further assess the network variables of individuals using accurate mining tools and to handle the inherent risks therein.

Using social data alone for predicting credit worthiness is still evolving as an alternative model, hence collection and application of social data must be used with caution. Although applying social data tells us much more than established formal, financial models alone – there are areas where more research is required to eliminate uncertainties.

Watch out for part 3 of this series which goes beyond financial credit rating.

References:

Consumer Lending Using Social Media Data

 

If you like what you read, join our team as we seek to solve wicked problems within Complex Programs, Process Engineering, Integration, Cloud Platforms, DevOps & more!

 

GET IN TOUCH!

Leave a comment on this blog: