It's time to ‘Like’ Social Scoring
Thus it makes difficult for loan providers to assess consumer’s credit risks. Often lack of information is the main reason of applicants‘ rejection by loan providers.
On the other hand most of these customers have accounts in social networks and most of them would like to provide their online information in order to improve their chances of receiving a loan.
That is why Creditinfo has conducted case study to understand how useful data from social networks could help in predicting of customer’s creditworthiness.
All of us know the main concept of prediction of risk using credit bureau data: customers with past payment delinquencies probably will show negative behavior in future, young customers are most risky then older customers, unemployed customers are more risky than those with permanent jobs. Social data open us access to new predictors, which we haven’t observed before. What can we say about customer’s risks if this person travels a lot? Speaks 5 languages? Has a big family? Plays online games all day long?
Amazingly, there is a strong correlation between customer’s social data and his/her credit risk. Based on 300k of profiles from social networks Creditinfo team has developed a scorecard aimed to predict customer’s credit risks. In contrast with the classic credit bureau scorecard, which draws attention to payment history and bureau inquiries, social model emphasizes personal features, achievements and social connections. Chart 1 provides comparison of main variable clusters used by both types of models
Social networks reflect person’s lifestyle. For instance, variables from cluster Hobbies/likes helps to discover more about customer’s preferences: whether a customer is interested in sport, reading, travelling, cooking and other activities or prefer to spend time online playing games and watching videos. Person’s social openness gives loan providers chance to know better their customer in order to provide most suitable services. But also it helps banks to verify that an applicant is actually the person who he claims to be and prevent third party fraud.
The following charts show trends of variables from clusters Hobbies/likes, Communication culture and Usage level. Red columns represent number of observation in each variable category and black line is corresponding level of risk.
The model is represented as a scorecard and efficiency analysis was performed on a development sample. Model Gini – 44.1% and K-S statistic – 0.33. It also shows monotonous Bad Rate in score intervals. Out-of-time efficiency (population after rating was launched) – 51.0% Gini.
Social score is a powerful tool for evaluation of customers’ creditworthiness, which could be used as addition to classic scoring procedure.
You would wonder – is it not a kind of Big Brother? Well, statistics show, that more than 80% of all Facebook profiles are open to public and people nowadays share their successes and interesting events with pleasure. Some of the financial institutions, mainly online lenders, have already understood the value of understanding social aspects of their customers. When there is no historical credit data at all – any data is an advantage. A little piece of trust could bring customers loyalty, which will pay off in a form of growing portfolio and self-promotion in social networks, where we all live now.