To build upon two previous articles that unpack the recent Mercator Advisory Group white paper Credit Scoring, Fintech, and Consumer Loans: Why AI Scoring Models Do Not Replace the FICO Score, PaymentsJournal sat with Brian Riley, director of the Credit Advisory Services Practice at Mercator Advisory Group, to hear more about how the industry-leading FICO credit scores are the most reliable measure of creditworthiness. Fairness and Objectivity in Credit Scoring Financial institutions must have accurate metrics to make decisions, control risk, and assess credit quality.

Since 1989, the FICO Score has relied upon factual data to rank risk, drawing upon information furnished by creditors. The underlying information comes from five data points: loan repayment history, the amount owed, length of credit history, recency of new credit applications, and type of credit history. The FICO Score uses the precise sources of information to provide an accurate, consistent, and fair measure that spans all facets of collateralized and uncollateralized consumer credit.

“The FICO Score sticks to the facts that regulators govern. It does not attempt to bring in casual or social elements. The score creates a relative ranking based on the risk of the account,” Riley said.

“No matter the customer’s background, a 660 means the same thing anywhere in the United States, for any borrower. So do a 520 FICO Score and an 800 FICO Score. ” FICO’s approach has two key advantages.

First, the data used in computing the scores is straightforward and regulated to ensure it is inherently unbiased against any individual or group. Second, the calculation of FICO Scores has been tested for decades and is transparent. FICO’s transparency contrasts with newcomers to the credit scoring industry, such as UpStart, which uses AI-powered systems that are effectively black boxes in calculating credit scores.

Such scores can arouse suspicion due to their murky origins. Machine learning shows promise in consumer credit, and there is evidence of artificial intelligence evolving into the space. While there may be substance, the models rely on hype or unregulated data that might be misleading or unfair.

Other models consider data used in calculating FICO Scores but seek to step outside traditional boundaries with data elements such as college education, social media presence, and previous purchases. These models aim to open the underwriting gate and bring in the credit invisible, the underbanked, or the credit impaired. However, these plans carry the danger of introducing bias and creating a credit-rating system that is impossible for people to understand and even harder to justify.

A transparent credit-rating system is essential. When a loan request is rejected, the applicant warrants an explanation. This not only is good business but also is required by various regulations, such as Fair Lending and Fair credit reporting.

Transparency is a fundamental component of the FICO Score, yet many alternative models miss the mark. Bias in Credit Scoring Over the past months, the use of certain alternative data in credit scoring has sparked pushback from policy leaders. These events sparked the introduction of a recent bill in the House that calls for the Consumer Financial Protection Bureau to assess the use of educational data by consumer lenders in their underwriting processes, publicize that assessment, and report its findings and recommendations for addressing potential disparities to Congress.

In contrast to some fintech AI models, the FICO Score has complied with fair-lending requirements for decades. Fair-lending regulators have found that the FICO Score shows no prediction bias against protected classes. In comparing persons with the same likelihood of repayment or default, the model did not score individuals in these protected groups lower than individuals in the general population.

In an environment where racial equity concerns carry a high focus, credit ratings that prove fair over across decades ought to be the gold standard. Lack of Transparency in Credit Score Calculation is a Problem As noted earlier, companies like Upstart use machine learning algorithms, which are difficult for mere mortals to understand. Highly flexible machine learning algorithms often have limited transparency.

Understanding a variable’s contribution to a prediction, how the variables interact with each other, and why the algorithm may have deemed the variable important is often extremely difficult. When these algorithms are particularly complex, the term “black box” suggests that the algorithm lacks clarity and the predictions are indefensible or inexplicable. Given that fair-lending laws and federal regulations require a lender to clearly explain loan rejections, companies that use machine learning algorithms to produce credit scores may be in a precarious legal position.

The inherent weakness, lack of transparency, and legal ramifications may be why the stock prices of companies such as Upstart have tanked recently. This indicates a lack of market trust in their underlying business models. Credit Scoring and the Inevitable Recession Considering the coming recession, companies need to rely on credit scoring that is dependable and innovative.

FICO has been in business for decades and has established a persistent, ubiquitous risk assessment metric. Upstart companies do not have data yet on how their model works in a recession, so they are effectively untested in such environments. Now is not the time for a bank to base its credit risk assessment on nascent, untested models.

Furthermore, FICO is an industry-leading company that has been the first to market with tools that subtly consider additional data in their models. To prevent lenders and consumers from taking on more risk than they can manage, the FICO Score is slowly expanding to allow relevant data points to complement furnished data to the three major credit bureaus (Experian, Equifax, and TransUnion). “There is going to be a horizon where the change takes place, and don’t expect it to be rapid, but expect it to be very thoughtful,” Riley said.

A current example of the volatility of alternative scoring can be seen in recent Securities and Exchange Commission (SEC)  filings by Oportun, a fintech lender that uses a proprietary score to address the unscored population. In a recent investor report, the firm notes that they helped establish credit histories for 1 million people, through their artificial intelligence scoring model.   While this is an exciting claim, it is interesting to note that the average Annual Percentage Rate (APR) for loan products is at the high end of the spectrum, with personal loans at an average APR of 32.

3, followed by Secured Personal Loans at 29. 1%, and credit cards at 29. 8%.

  These high interest rates are important facets of their credit acceptance model for embracing the unscored and indicative of the risk associated with AI scoring.   In contrast to the credit card APR at Oportun, the Federal Reserve reports that the average APR for accounts assessed interest in May 2022 was 15. 13%, nearly half the rate charged by Oportun.

  High-interest rates are necessary when considering loan losses.   At Oportun, Annualized Net Charge-Off Rates for the six months ending June 30, deteriorated from 7. 5% in 2021 to 8.

8% in 2022, and now, as the United States faces the threat of persistent inflation, loan losses trend towards the firm’s peak levels, which in 2020 hit 9. 8% Riley provided the example of rent and mortgage payments in various parts of the country to illustrate the FICO Score’s absorption of relevant data. A Chicago renter and a Sioux Falls homeowner might receive different credit scores, but both can demonstrate responsible, on-time payments related to their housing.

These and other similar factors appear in different versions of the FICO Score: FICO 8: The most widely used version of the standard credit scoring model, using the five primary metrics as its core rubric for credit scoring from 300 to 850. FICO 9: This version features adjustments to the treatment of medical collection accounts, rental history, and third-party collections. FICO 10/10T: A more predictive version of the FICO Score, using the previous two years of credit activity for accounting for potential future risk.

FICO 10T: The T stands for “trended data,” which is included in this score version. UltraFICO: An opt-in credit model that helps individuals boost their FICO Score by evaluating personal banking information and behavior, including cash on hand, frequency of transactions, and history of positive balances. FICO XD: An alternative credit score created in conjunction with LexisNexis and Equifax that can evaluate borrowers with little to no credit history based on utility, cable, and phone bills.

By gradually and strategically applying a mix of data to bolster risk profiles, various versions of the FICO Score can bring new consumers into the fold without sacrificing the regulatory oversight that makes it such a sound scoring standard. Count on the FICO Score to continue evolving safely and responsibly, maintaining its essential integrity while reflecting the realities of the modern world. Register to download the FICO complimentary whitepaper - Credit Scoring, FinTech, and Consumer Loans: Why AI Scoring Models Do Not Replace the FICO Score First Name (required) Last Name (required) Your Email (required) Company (required) Title (required) Δ

By PaymentsJournal
Sep 01, 2022 00:00
Original link