According to the World Bank, as of 2017, 1.7 billion adults around the world have no access to financial services[i]. This issue isn’t necessarily isolated to impoverished countries. According to the FDIC, in 2017, nearly one in five Americans were underbanked, meaning 24.2 million households went without common financial services like access to credit cards or loans[ii]. The reasons for this may vary, ranging from people not having enough money to keep an account, people’s mistrust in banks, or simply living in a geographical area not in close proximity to a bank[iii]. Ultimately underbanked individuals are characterized by a “thin” or insufficient credit history, which effectively blocks their access to many formal financial services. Altogether, this situation has led to new approaches to promote financial inclusion among the underbanked. Credit card companies, for example, now offer “credit-building cards” with low fees that can be used by just about anyone. Fintech innovations in the form of peer-to-peer lending, financial robo-advisors, and mobile banking, are a few more examples. Another interesting approach that is gaining popularity is the use of alternative data to assess individuals’ creditworthiness. This article will expand upon various alternative credit scoring techniques including payment histories, personal data, behavioral information, and psychometrics-based analysis.
Non-Credit Payment History
One type of alternative data that is widely considered to be important is non-credit payment data. Non-credit payment data refers to an individual’s payment history of utility bills, rent, phone plans and more. The idea behind this method is that bill payments reflect payment responsibilities in general, and may be indicative of future loan payments as well. In other words, people who pay their bills on time might be more inclined to responsibly manage credit obligations. For example, since 2016, FICO, the leading measurer of consumer credit risk in the United States, has been using a score called ‘FICO XD’ which is derived from cable, cell phone, and utility payments in its final credit scorecard[iv]. Other companies are following suit with new credit bureau scores, such as Experian’s Boost, which “boosts” traditional credit score based on utility and mobile phone bills data[v], or the multi-bureau sponsored ‘VantageScore,’ which started considering telecom, utility, and rent data already in 2006[vi]. Ultimately, many credit-reporting companies have found success in using payment history information to predict good future borrowers, and this trend will likely continue.
Non-Financial Personal Data
Another type of alternative credit data is related to non-financial personal information. Personal data, such as employment history, education, personal assets, local public records and even affiliated social groups can be indicators of creditworthiness[vii]. Indeed, some financial institutions have found personal professional data such as these to be a good supplement to traditional scoring methods, especially when the latter are scarce. There may be some regulatory constraints that come into play regarding the use of this type of data in some countries. Specifically, lenders should take caution against building scoring models that rely on datasets that might skew toward certain demographic groups over others, and should be aware of the potential adverse discrimination their model might have against certain protected groups[viii]. Nevertheless, these issues clearly depend on the way the data is used, and do not preclude their potential effectiveness.
People’s behaviors online leave a detailed digital footprint, and an enormous amount of data that may be relevant for credit scoring. Take, for example, the length and time spent on different types of websites visited, social media interactions, and geo-location tracking. More specifically, whether a website is accessed through an ad or search, whether it was done at 3am or 3pm, and whether it was accessed from home, work or abroad, might be reflective of some behaviors that are relevant to credit. In addition, a person’s activity on social media, such as posting and commenting, and the content of this activity might be indicative of traits that are relevant for creditworthiness. Indeed, many thousands of such data points are measurable, and today’s machine learning and AI technologies have the computer power to evaluate their predictive value of loan payments like never before possible. It is therefore not surprising that credit reporting agencies and fintech lenders such as Germany’s online lender Kreditech, and Silicone Valley based Branch, are leveraging this vast data to better assess loan applicants. As mobile and internet use increase, it seems that methods such as these will become more interesting to the credit underwriting.
Whereas behavioral data might be inferentially related to some traits that are relevant to creditworthiness, psychometric data attempts to measure such traits more directly. Psychometric data is typically captured in brief questionnaires which inquire about a borrower’s behavioral preferences, and whose algorithms tap traits such as trustworthiness and responsibility. Psychometrics may be considered a kind of proxy to the elements of a traditional loan applicant interview for assessing a borrower’s character. In this way, psychometric data helps to bring back some of the ‘personal aspects’ of loan decisioning, in a structured and scalable way, and is often not limited to geography, demographic or credit history. Innovative Assessments (IA) is an example of a fintech in this space, whose scores are designed to supplement both traditional and alternative financial-based data. Psychometric based credit scoring is starting to gain traction, and shows promise for augmenting credit models among the underbanked.
In summary, as technology advances, so does the amount of benefit it offers for underbanked people everywhere. Although the methods mentioned above provide new ways for measuring creditworthiness, the actual implementation of these technologies is of fundamental importance. In particular, is the consideration of how a lender might apply multiple approaches per circumstance. For instance, a tiered approach might be an effective option in some cases. For example, a lender might start by examining a credit application using traditional bureau scores, and then ‘escalate’ to other alternative methods as necessary, based on preliminary risk assessments, and the information (or lack thereof) available. In addition, it is worth mentioning that alternative technologies do have challenges and raise important issues that have yet to reach a consensus opinion. Consumer privacy, adverse impact, and justifiable/explainable decisions are among these issues. It will be exciting to see how these technologies and the industry as a whole develop in the coming years.
Co-written by Cole Gallagher——————————————————————————————————————————————