Social Data Can Enable Better Credit Scoring

Lending institutions try to create as many new loan accounts as possible, while at the same time keeping their risks manageable. To do this, they have to factor in credit scores. Theoretically, financial analysts in banks look at the credit score before they approve loans, or set other factors like loan duration and the interest rate. Practically, it’s not as simple as it sounds. There are several flaws in the traditional credit score systems.

Most of the credit score systems don’t consider those who don’t have credit cards. Considering that most companies are currently targeting millennials, a third of whom never have, and might not even apply for credit cards, these are people that don’t have scores. Millennials reportedly have increasing buying power as time goes by, and as their disposable income increases. Therefore, it’s important that lending institutions consider revising their scoring protocols to capture this population.

Besides millennials, there are other people who earn very good salaries and are frequent spenders, but this doesn’t show up in their credit history. There’s also the category of people who earn a lot of money but barely spend any of it.

A good scoring process shouldn’t only be used to determine the creditworthiness of an individual. In fact, there are lots of other applications to which such reports can be put to use.

Rethinking the Concept of Credit Scores

There are different factors that differentiate loan applicants such as their background, age, socioeconomic status, and even the reason for taking a loan. Other than that, most people don’t understand the way credit score companies work. In fact, a lot of people are not even sure whether they are eligible for loans or not.

Considering these factors and the reality that the majority of the millennials barely consider financial products that affect their credit scores, like mortgages and credit cards, it might be time for financial institutions to consider other methods of rating credit scores.

Having worked with different financial institutions in the past, it’s indeed possible to create products that give credit scores a personal appeal. These are products that consider the individual’s social appeal in aggregating their creditworthiness.

It’s also possible to improve the current scoring intelligence systems and improve the ability to predict delinquency by up to 6%, which is a good way to increase revenue. In this report, we will look at ways of improving credit scoring systems.

Enhanced Credit Score

This scoring method builds on the conventional credit score systems. It optimizes the scores by adding new information from external sources that can help make predictions more accurate for different reasons.

To illustrate this, let’s take the example of an individual who engages in high-risk behavior, and is very flamboyant. This person poses a higher risk to lending institutions than someone who earns the same amount but lives a reserved life. This introduces the concept of a behavioral credit score. Behavioral credit scores can be built on social media or from public data.

Public Data

There are lots of resources out there that can be used to track the spending patterns of a customer. Buying stuff like a car or a house is public information. In fact, the FTC provides a lot of information through their respective websites, data.gov and usaspending.gov. These archives include data like school enrolment, consumer spending activity, American housing survey and lots of other information from which insight into spending habits can be deduced. There are other rare cases when spending information can be available in the public domain, such as when the P2P lending firm Lending Club published their loan data on Kaggle.

Social Data

One of the best ways of finding out more information about an individual is to check social media and study their lifestyle choices. This might provide some information on the possibility of defaulting on loans. It can also help lending institutions reduce the risk of default. The information that’s available on social media can help these institutions determine the spending and behavioral traits of individuals, hence present a realistic creditworthiness profile.

At the moment, many startups are using data from social media to determine the creditworthiness of their customers. We have witnessed an increase in financing startups and credit scoring companies that leverage their data on social network data. This has also seen a lot of low-income consumers come on board, creating opportunities for a lot of people who would have otherwise found it very difficult to get loans.

A Combined Behavioral Credit Score

By combining public and social data, it’s possible to create a powerful behavioral credit profile using machine learning algorithms to identify and predict trends. These can help determine high-risk loan accounts early on, and come up with relevant measures, such as offering debt-restructuring mechanisms to such customers.

Behavioral credit scores can also be used to create an accurate estimation of the customer’s risk exposure, which will go a long way in reducing the risk of bad debt. This is also a good way of determining the probability of recovering defaulted loans. Therefore, lenders are able to optimize their efforts on collection by increasing their efforts on loan recovery.

Enhanced Credit Scoring Models

Having looked at the sources of data that will be used for enhanced credit scores, we must also understand how these credit score models are run. There are different models, including neural networks, decision trees, and scorecards.

Before you settle on the right model that will meet your needs, you should consider other factors like how easy it is to understand and use the model, and whether you can easily justify its application to your scenario.

Another important factor that you must address for whichever model you use is how accurate its predictive ability is. Can you use the scores it presents to make reject or accept decisions as suggested by the model? The outright winner, therefore, will be a model that meets the purpose for which it’s built, and the set of data that it validates.

• Scorecards

Scorecards are the most basic, traditional credit scoring models you can come across. Scorecards are tabular forms with questions that the applicants must answer. The questions are characteristics, while the possible answers the applicant will choose are attributes.

A characteristic, in this case, might be the applicant’s age, while the respective attributes will be a range of age groups where the applicant falls into.

The applicant earns points for each answer, and points are awarded according to the level of risk exposure. If the applicant’s aggregate scores meet a certain threshold, their application can be considered for acceptance.

This system is generally built to prevent loan default. It cannot learn, and therefore, it’s not recommended in case the ultimate goal is to increase the loan book while reducing the risk of loan default.

• Regression and Decision Trees

Lenders also like to use regression and decision trees. However, they are not always the best, especially because they are not future-proof. These models are not effective over the long term, and the quality of information collected in them is eroded over time. Decision trees are built on decision rules that are very easy to understand, which means that they are often ripe for fraudulent manipulation.

The best way is to consider using generic models, like neural networks. However, for these to be efficient, they demand a lot of features. Neural networks combine different characteristics to determine the credit scores.

They offer superior predictive accuracy than scorecards and decision trees. However, the problem with neural networks is that it might not be easy to explain the rationale behind a given score, which makes it a difficult decision making through this model.

Neural networks, therefore, are more realistic in case you need information regarding averages, over information that requires insight into the specifics of a particular case.

Social Network Scoring in Developing Nations

Lenders in developing nations have been using data from social networks to determine the creditworthiness of their prospective borrowers. Most of these countries don’t have enough infrastructure that contains data on individual borrowers. As a result, credit history is more of a mirage.

You will notice that in most of these countries, the internet revolution is yet to take place, or where it’s already happening, it’s at a very slow pace. Therefore, the banks don’t have the resources needed to collect relevant and sufficient information on the borrowers.

However, the underbanked and unbanked in these countries have access to social media, one of the most important tools for credit scoring. There are special smartphone packages that are provided by companies like Facebook, for example, who have a unique mobile data plan in Thailand where users can access the social network.

Another example is Lenddo that offers credit assessment services in the Philippines, Colombia, and Mexico. The Lenddo system is heavily leveraged on the information that the borrowers have on their social networks.

A peek into the kind of success that microfinance institutions have enjoyed in such countries makes a strong case for an attempt to do the same with social networks. What companies need to do is to learn more about the social dynamics of the community, and then use this knowledge confidently to come up with feasible solutions. According to one of the local retired bankers, the concept of social network scoring is similar to systems that have been in place hundreds of years before credit bureaus came into being.

Since companies like Kreditech, Kiva, and Lenddo ventured into the market, a lot of people have taken advantage of this to start or fund their small businesses, pay for health care, and school fees.

Impact of Social Fragmentation on Credit Scores

Once your information is in the hands of the credit company, they analyze it from different angles. For example, people in your network that you barely communicate with will hold less weight than those you communicate frequently with, and the lenders will, therefore, make predictions based on this.

If most of your friends have better credit scores than you do, your risk exposure to lenders might be fairly lower. When users fragment their networks and relationships online in this way, the accuracy of such credit scores becomes ambiguous.

Other than that, these scores can provide an accurate reflection of the borrower’s risk exposure because of the ego network they belong to. The problem with this is that as the ego networks increase, they become smaller in size, which means they hold less information on every member. This makes predictions from the networks inaccurate.

According to experts, however, network manipulation by adding or removing people depending on their credit scores might not always work for you. This is because the emphasis is not on who your friends are but on the nature of your shared connections. The fact that you are friends with someone on social media might not hold much value when determining your social credit score.

Impact of Social Fragmentation

Of course, some people will try altering their networks online so that they appear more attractive to lenders. There are some ramifications to such traits. This is, in essence, a form of discrimination. Over time, some people will be ostracized and eliminated from certain networks. There’s also the possibility of discrimination based on the personal effort someone seems to be putting in improving their scores.

There lies a risk in being extremely choosy with your social connections. Obviously, the more selective you are, the smaller your circle will be. This eventually reveals a low credit score, which makes you a less favorable borrower.

On the other hand, we also have individuals who have poor credit scores but don’t seem to be making any effort to improve their current standing. This can result in disengagement by their peers, as they wouldn’t want to be associated with someone whose scores are very low.

Some people might struggle, and as a result, be more selective in the information they reveal to lenders. At the same time, the lenders are also aware of the fact that there might be a lot of people who are genuinely good but are unable to access lending facilities.

Conclusion

FICO scores are currently the most popular credit risk assessment protocol in the US. However, social network scoring is increasingly providing the underbanked and unbanked population with the opportunity to access capital that they would otherwise not have enjoyed earlier on.

Owing to the lack of reliable infrastructure, lenders in such communities might have a difficult time determining the credit risk that the borrowers expose them to. However, social network scoring might prove to be quite an asset and a practical alternative too.

The best course of action, therefore, would be to use any available data to support this model in assessing the creditworthiness of individuals.

The best summary for this concept is the proverb noscitur ex sociis, which loosely translates in English to you are the company you keep.