Risk identification using alternate data and potential applications in credit decisions

Abstract

Traditionally, financial institutions rely on bureau information such as risk scores, past delinquencies, and payment behavior to make credit decisions including aspects such as approvals, line assignments, and lending attributes such as interest rate and tenor. However, this creates challenges when calculating risk for consumers without documented credit history or in emerging markets such as the Indian subcontinent where credit penetration is still relatively low compared to developed economies.

This creates an opportunity for using alternative methods to identify risk and approve credit and loan characteristics for these emerging markets and customer segments. This paper highlight some of the alternative sources of data financial institutions can use in their underwriting strategy to augment bureau information, incorporate smart algorithms into their credit decision engines, and better measure risk when traditional data is not readily available.

Keywords – credit card, alternate data, credit risk, underwriting

1. Introduction

There are an estimated 2.8 billion credit cards in use worldwide. 1.06 billion are in the United States alone, serving over 170 million Americansⁱ. Roughly 70% of the eligible US population carry a credit card, with an average household debt-to-GDP ratio of 79%². In contrast, 3% of the Indian population has a credit cardⁱⁱⁱ, and an average household to GDP ratio of 35.8%ii with an estimated CAGR of 25%^iv. Tapping into this market presents an opportunity for financial institutions to use alternate sources of data to make smart credit decisions to mitigate risk and optimize revenue.

There is significant population who have either no credit history or too little credit history to generate a credit score, an issue affecting an estimated 45 million Americans^v.. To overcome this, first-time credit grantors can base their cutoff scores on market research, third-party data, and other alternative sources to assess credit worthiness.

Traditional data usually means data from a credit bureau, a credit application, or a lender’s own files on an existing customer. Alternative data is everything else.

Alternative data helps banks to supplement their existing credit risk decision frameworks to include this underserved population, enabling banks to make informed decisions and helping individuals without credit history join the mainstream financial ecosystem. Using alternative data is becoming a necessity for banks to compete with fintech competitors who are aggressively leveraging it for business expansion. These fintechs have a head start from combining APIs with AI tools to come up with predictive models that gauge customer creditworthiness and risk potential, using them to aggressively expand their books. Some of them have gone further to create an alternative credit scoring process when traditional sources present too little information to be able to create a score. This can be a win-win situations for lenders and consumers, as financial institutions can increase their customer base while underserved individuals can secure better terms by participating in the traditional credit system^vi.

2. Problem statement and scope

Credit services such as mortgages, auto or personal loans, and credit cards require consumers to have a documented credit history. However, an estimated three billion individuals worldwide such as immigrants, young adults, the unbanked, and the underbanked often cannot provide much supporting data^vii. This poses a problem for banks to extend their services to credit for worthy borrowers who do not have an established credit history. There are an estimated 3 billion adults worldwide who don’t have credit and so don’t have credit records.

The amount of time between updates for these traditional reports is fairly lengthy. While it takes 30-45 days for FICO score and credit score across all three major credit bureaus to refresh, alternative data is more up-to-date and can be collected in real time without compromising credit scores due to frequent or hard inquiries.

Alternative data is the financial information that isn’t typically collected by credit reporting agencies or provided by customers while seeking credit. It covers a wide spectrum of sources ranging from bank account cash flow analysis, credit card usage patterns, and checks into payday loans, rent or utility bill payments, and other areas. Public records including education and employment background, asset ownership, online and social media activities are other notable sources. Other data such as travel history and spending on entertainment or dining can also give an indication of financial health and the stability of the consumer. Digital attributes, such as the device used when applying for credit online or the digital journey the customer took to reach the point of application, can also shed light on a person’s creditworthiness. While none of them make a strong case individually, aggregating these details can identify qualified individual who might otherwise be dismissed as too risky.

Alternative data can be applied in areas including:

Alternative data on a customer’s digital habits can also be used in lending decisions.

• Individuals with inadequate or no credit history: New consumers who have relocated internationally are often seen as having too little credit history for lenders to make credit decisions.
• Low bureau score bands: Lenders using a bureau score as a filtering criteria issue blanket rejections for consumers below the cutoff. The margin population is cutoff sensitive. While expanding the exposure by liberalizing credit policies, alternate data can be used to find opportunities to segregate risk into marginal segments.
• Evaluating incremental opportunities for cross-sells: For consumers on books and adequate credit history, alternate data can be used to make decisions regarding including credit line increases, APR adjustments, line optimizations, and buy-now-pay-later plans.

Alternative data on a customer’s digital habits can also be used in lending decisions. Around half of all credit card offers are now made digitally, and 73% of all card applications are received digitally. In other words, people receive a card offer in the mail, but more of them now apply online or on a mobile phone rather than mailing in the application^viii.

3. Methodology

Alternative data where information can be directly linked to the financial conduct an individual is relatively intuitive to incorporate into credit decisions. Cashflow information, such as data from consumer deposits, card accounts, or small business accounting software, is one of the most promising options for improving automated underwriting as it provides a detailed and timely picture of how applicants manage their finances than traditional credit reports. In a study by FinRegLab, cash flow scores were directly found correlated with FICO scores and delinquencies^ix. Thus, cashflow data provides meaningful predictive power among populations and products where traditional credit history is not available or reliable. Therefore, segmenting the population using cashflow variables can separate risk and be used in credit decisions in conjunction with existing scores or in the absence of credit history.

Similarly, using attributes such as utility bills can be assessed to assess credit worthiness. In fact, agencies such as Experian and UltraFICO have come with a credit boosting mechanism where consumer can link their utility bills to the platform to provide an adjusted FICO score. ^{x, xi}

When alternative data is not directly linked to consumer financial conduct, a more hypothesis-driven methodology and control framework is suggested to improve decision making while assessing risk and potential revenue to be realized. The risk scorecards can learn from the test hypothesis and control framework to improve the alternate data scorecards based on these attributes. Some of the hypothesis are listed below:

Hypothesis 1: The type of smartphone or device used by a consumer may reflect the financial state of the consumer and can be used in credit decision making

Hypothesis 2: Social media metadata may be segmented population in clusters to make granular bins of risk separation

Hypothesis 3: Source of origination of digital applicants, such as third-party websites, email, or social media can be used to assess risks and the potential revenue to be realized

The hypothesis test can be evaluated to measure their impact on separating risk. While the traditional risk scorecard can give larger risk segments, supplementing this information with alternative data can provide much more granular risk segmentation scorecard.

The credit decision engine can iteratively learn from the hypothesis and can mitigate risk while optimizing revenue for the creditor.

4. Example case study

EXL helped a global bank in the APAC market optimize their credit cards initial line assignment strategy using alternate data. The results were resulted in increased revenue by 120% and an increase in risk-adjusted return by 190%. Before the optimization exercise, the institution was using internal risk variables to provide the initial line to customers and was not able to segregate populations based on revenue segments. EXL developed a strategy where untapped alternate data sources were utilized to target profitable customers within the risk appetite. Some of alternate data sources and fair lending regulations . This shows regulators realize the importance of alternate data sources in credit decisions.

However, data privacy regulations require institutions must receive customer permission to use alternative data in credit decisions. This creates a challenge as consumers may be unwell to give third parties permission to receive their data. Even when this permission is granted alternative data is often highly distributed and unstructured, adding another layer of complexity. Lenders will require API technology, effective aggregation tools, dynamic analytics capabilities, and a transparent scoring methodology for alternative data to be reliable vi.

With so much alternative data available, it becomes important to have a robust quality control mechanism for a reliable scoring mechanism. In cases where this data is implemented into credit decision engines, there needs to be a timely, reliable flow of data for a consistent and transparent approach to assess creditworthiness.

6. Conclusion

There are certainly incremental benefits for institutions, consumers and the economy at large for using alternative data to make effective, transparent decisions, especially for underserved customer segments who have thin or no credit history. Having another level of granularity in credit decision strategies will help lenders realize higher profits. Alternative data may never replace the formal credit sourcing system, but it can supplement existing risk scorecards to better mitigate risk and further realize higher revenues.

References

i Credit Card Statistics [Updated August 2021] Shift Processing, cfpb_consumer-credit-card-market-report_2019.pdf (consumerfinance.gov)
ii Households Debt to GDP - Countries - List (tradingeconomics.com)
iii Credit Card Usage in India Statistics & Facts - 2022 (findly.in)
iv India Credit Card Market Size, Share, Trend & Forecast 2025 | TechSci Research
v Using alternative data to evaluate creditworthiness | Consumer Financial Protection Bureau (consumerfinance.gov) vi The alternative data revolution in banking - Fintech News
vii Using Alternative Data In Credit Risk Modelling (fico.com)
viii Key Trends Reshaping Credit Card Marketing (thefinancialbrand.com)
ix Fact Sheet: Cash-Flow Data In Credit Underwriting - FinRegLab
x Experian Boost - Improve Your Credit Scores Instantly for Free
xi Introducing the UltraFICO™ Score | Ultrafico
xii alternative-data-use-credit-underwriting.pdf (ncua.gov)

Written by:

Darpan Jain
Vice President, US & APAC Banking Analytics

Amit Dhankar
AVP, Analytics

Kanishk Dabakra
AVP, Analytics