By Alex Johnson
3 Fintech News Stories
#1: Enabling Discernment Without Permitting Discrimination
A group of 20 banks and credit unions are collaborating on ways to increase access to credit for marginalized borrowers:
The collaboration, Underwriting for Racial Justice, aims to create new lending criteria for lower-income U.S. borrowers who have historically faced higher barriers to obtaining financing.
Under the two-year pilot program, which launched last month, participating lenders will be able to exchange best practices on how to gauge a borrower’s ability to repay. The lenders will also submit data about loans they make to individuals from marginalized communities, which will then be shared with other members of the group.
The program seeks to minimize the use of traditional credit scores in loan decisions. Instead, the use of artificial intelligence technology will “enable lenders to say yes to more borrowers without increasing risk,” said Laura Kornhauser, CEO and co-founder of the fintech company Stratyfy, which is developing the program’s tech platform.
At the end of 2020, the homeownership rate for Black Americans was 43.4%. It was 51.1% for Hispanic Americans, 61.7% for Asian Americans and 72.1% for whites, according to an analysis conducted by the National Association of Realtors.
Black and Hispanic mortgage applicants were more likely than others to be rejected, the analysis found.
Two things about this specific project stand out to me.
First, it sounds like the 20 banks and credit unions participating are all pretty small:
Financial institutions that are participating in the program include Berkshire Bank, the New Orleans Firemen’s Federal Credit Union and the Chehalis Tribal Loan Fund, which is a community development financial institution.
That’s a big difference from some of the government-driven initiatives mentioned above, which were able to recruit much larger participants.
Second, the focus of this initiative is on novel and increased usage of machine learning rather than the use of alternative data, which has, historically, been the lever that financial institutions have relied on to responsibly increase access to credit. The basic idea here is that while automated, rule-based underwriting (which most banks use, at least to a degree) has been effective in eliminating a lot of negative (often unconscious) bias from the lending process, it also left the process a bit too rigid:
Berkshire [Bank] also extends loans to borrowers who have gone through bankruptcies and credit-related judgments, as well as those who have been convicted of what Dixon described as victimless felony crimes.
“That is not a no-go — you can still be eligible,” she said.
Also participating in the Underwriting for Racial Justice program is the Washington State Employees Credit Union.
The credit union’s underwriting team has worked to understand the difference between “perceived risk and actual risk,” said David Puszczewicz, the credit union’s director of community homeownership development.
What Berkshire Bank and Washington State Employees Credit Union are talking about in these quotes is the importance of discernment – the ability to find nuanced reasons to say yes when the standard data and rules are saying no.
The challenge is enabling discernment without permitting discrimination. Human underwriters aren’t capable of it. Beneficial State Foundation and Stratyfy are betting that machine learning will be.
#2: FedNow Paranoia
Banking industry analysts are worried about the impact that FedNow might have on bank deposits:
The Federal Reserve expects to launch a new system this month aiming to make payments in the U.S. banking system available immediately, around the clock. Although it is a boon to consumers and many businesses, some analysts warn that FedNow could destabilize banks’ reliance on customer cash, fanning the flames of deposit flight that became the bane of several regional banks this spring.
instantaneous transactions allow customers to pull cash with ease, and without notice. That threatens smaller banks and likely requires stiffer cushions to mitigate adverse scenarios, according to [Noor] Menai, an advisory council member of the Federal Reserve Bank of San Francisco. Regulators are expected to propose tougher capital rules for the banking sector in the coming months.
FedNow’s launch comes at a precarious time, during the Fed’s year-plus effort to raise interest rates and quell inflation. Banks are competing with money-market funds and other higher-yielding products for deposits, and paying up to borrow elsewhere. U.S. commercial-bank deposits were down $705 billion in June from a year ago.
FedNow has become a magnet for lots of anxiety in and around the financial industry; everything from concerns about CBDCs to conspiracy theories about how it is part of a plot to destroy the crypto industry to these more mundane concerns about it accelerating the flight of deposits.
None of it makes any sense to me.
I guess I can see why it’s tempting to try (however illogically) to connect the dots, given the multifaceted role of the Federal Reserve (setting monetary policy, supervising banks, and providing payments infrastructure). But it still doesn’t make sense.
In this specific case, why would FedNow make the failures of banks like SVB and First Republic more likely? The introduction of mobile banking undoubtedly hurt banks’ deposit franchise values (read this paper on “Bank Walks” for a lot more detail), but that’s a function of convenience (the ability to chase rates or participate in a bank run from your couch) not speed (how fast the transaction clearance and settlement). And if you want to address the disparity between the speed at which bank customers can take money out and the speed at which banks can bring money in from external funding sources, the answer isn’t to hobble bank customers (though banks participating in FedNow will have flexibility in how they implement it). The answer is to modernize the Fed Discount Window and Federal Home Loan Bank system.
A flaw in Revolut’s payment system in the US allowed criminals to steal more than $20mn of its funds over several months last year before the company could close the loophole, according to multiple people with knowledge of the episode.
The problem stemmed from differences between European and US payment systems, which meant that when certain transactions were declined Revolut would erroneously refund accounts, handing them its own money, according to three people with knowledge of the situation.
The problem first appeared episodically in late 2021. Organised criminal groups then took advantage of the fault early in 2022, according to three people with knowledge of the situation, encouraging individuals to try to make expensive purchases that would go on to be declined. This would then be cashed out via ATMs.
Although Revolut recouped part of the roughly $23mn stolen by pursuing some of those who had taken funds, the net loss was about $20mn — equal to almost two-thirds of its annual net profit in 2021, those people added.
Whoops! Whoops! Whoops!
Revolut’s systems failed to pick up the mass fraud and the problem came to light when a partner bank in the US notified the fintech that it was holding less cash than expected, the people told the Financial Times.
The loss relating to the theft was not specifically disclosed in the delayed 2021 results.
The fintech is still awaiting its banking licence in the UK, more than two years after first announcing its application, far longer than the typical turnround time of less than a year.
The UK’s Financial Conduct Authority ordered an independent review of Revolut’s policies to prevent and detect financial crime in 2020.
Auditor BDO separately warned that Revolut’s revenues could have been “materially misstated” as it was unable to satisfy itself of the “completeness and occurrence” of about two-thirds of its revenues reported for 2021.
A few stray thoughts:
- “The problem stemmed from differences between European and US payment systems”. That’s just brutal. It’s not even like Revolut can say, “Well, sure, we took some losses, but we managed to gain a huge foothold in the massive U.S. market.” As far as I know, they’ve seen little-to-no traction in the U.S. as of yet. Just lots of fraud!
- It’s pretty interesting to me that this started happening episodically with customers in 2021 and then caught the attention of criminal groups in 2022, who started exploiting it on a much larger scale. This wasn’t a “hacker working by himself to systematically test Revolut’s defenses and finding a weak point.” This was “a few customers notice something weird happening, and then organized criminal groups hear about it and attack.” The first thing is just something you need to be ready to react to when it inevitably happens. The second thing is something that, ideally, Revolut would hear about and get ahead of before it became a big problem.
- Speaking of which, WHERE THE HELL WAS THE MONITORING?!? If you are operating a banking business, with lots of individual customer accounts, your core job is making sure the numbers add up. Apparently, Revolut didn’t realize that it had a problem until its partner bank in the U.S. notified it that it didn’t have as much money in its FBO account as it was expecting. What was Revolut doing?!?
- I guess we now know why Revolut was having all that trouble with the simple task of submitting financial statements to regulators. Two-thirds of its 2021 net profit was stolen! Good job by BDO adding in that qualified opinion!
- I can’t possibly see how Revolut gets that UK banking license.
2 Fintech Content Recommendations
#1: The Checking Account War Is Over (And The Fintechs Have Won) (by Ron Shevlin, Forbes) 📚
New Shevlin Data Drop! Update the neobank adoption slide(s) in your presentation decks!
(Also, stick around for Ron’s analysis. It’s excellent, as usual).
#2: How to Protect Customers from ChatGPT-Driven Scams (by Simon Taylor, Sardine) 📚
I sometimes get asked, “How will generative AI impact fraud management?”
My answer is always that it’ll make it substantially more difficult because it’ll lower the cost for fraudsters to produce significantly better social engineering scams at scale.
The follow-up question is obvious – “OK, what do we do about that?”
Simon has helpfully outlined his and Sardine’s thoughts on that question. Very much worth a read!