Two things about this.
First, the fact that these tokens are not the same as equity isn’t a meaningless technicality. It’s a really important distinction!
As OpenAI made clear in its statement, it didn’t approve or endorse this offering. Robinhood just acquired $1 million worth of OpenAI shares from employees or investors through secondary transactions without OpenAI’s permission.
One of the problems with this model, when it inevitably gets rolled out on a larger scale, is that retail investors won’t understand what they’re investing in because private companies are, well, private. They don’t have to disclose any information, unless they want to. Their investors (like Softbank, which recently pumped in $40 billion) get some information. Retail investors who OpenAI doesn’t know or care about won’t get anything. Shit, Robinhood won’t get anything! All that Robinhood and its customers know is that OpenAI was last valued at $300 billion, and AI is a hot market right now. That’s not the basis for a good investment!
OpenAI clearly understands and is worried about this. It doesn’t need the liquidity that Robinhood can bring to the table (and it probably isn’t thrilled about the impact of an even more irrational secondary market for its shares on its employees and investors). And it doesn’t want consumers making stupid, speculative bets on tokens that bear its name. That’s bad for the brand!
The same thing is true for SpaceX, Stripe, and all the other name-brand private companies that are overburdened with hype. They don’t want this market to exist, and they will likely take actions to make it more difficult for tokens that are loosely affiliated with them to be sold.
Which brings us to the last sentence in Tenev’s tweet:
Our giveaway plants a seed for something much bigger, and since our announcement we’ve been hearing from many private companies that are eager to join us in the tokenization revolution.
Just because OpenAI, SpaceX, and Stripe don’t like this model doesn’t mean that other, less famous private companies won’t.
If your private company isn’t popular enough to catch the attention of Masayoshi Son, it may still be mysterious and intriguing enough to catch the attention of Robinhood’s retail investors. And if it is, and you could use more liquidity from the private markets or media attention on your brand, you may welcome the opportunity to join the “tokenization revolution”.
In economics, this is known as adverse selection. Private companies that know their business isn’t good enough to attract significant investment from venture capital firms work with Robinhood to sell tokens that give retail investors exposure to their last valuation, without disclosing any information about their business to those investors.
This isn’t going to end well, but we shouldn’t be surprised Robinhood is pushing in this direction. That entire company is essentially just an engine that generates a steady stream of adverse selection for its customers. This is just the latest flavor.
(Movie Fan’s Note — I love Will Smith in this movie, with one small exception. After he crashes his plane and the alien ship chasing him, he climbs onto the spaceship, opens the hatch, punches the alien, and says, “Welcome to Earth.” It’s such an insane and discordant part of the movie that it really feels like Smith told Emmerich, “You gotta let me punch an alien and say a cool catchphrase.”)
I gotta call my brother, my housekeeper, my lawyer … Ah, forget my lawyer.
What Happened?
Federal banking regulators are being instructed to forget about the theory of disparate impact. Evan Weinberger at Bloomberg reports:
The Office of the Comptroller of the Currency informed examiners and at least one fellow banking regulator last week that it stopped using disparate impact theory as part of fair lending reviews, according to multiple people familiar with the matter who requested anonymity to avoid retaliation.
The change follows an April executive order from President Donald Trump directing agencies to stop relying on the theory, which says facially neutral practices can have systemic discriminatory effects.
The OCC will instead rely solely on disparate treatment as the standard for bringing cases, according to a June 25 internal OCC email obtained by Bloomberg Law. That means examiners will have to find overt acts of discrimination rather than conducting statistical analyses of banks’ loan books.
So what?
Quick review — in the U.S., you are not allowed to discriminate against a protected class (race, national origin, sex, age, etc.), across a whole range of specific activities, including lending (which is governed, principally, by the Equal Credit Opportunity Act or ECOA).
Apart from overt evidence of discrimination (which we thankfully don’t see much of in lending anymore), there are two primary ways that regulators look discrimination in lending. You can look for policies, processes, or analytic models that are structurally biased against a protected class, such as the inclusion of age or race in a credit risk model. This is called disparate treatment. Or you can look at the outcomes of policies, processes, and models that are, on their face, neutral, to see if any protected class is disproportionately harmed. This is called disparate impact.
Since the implementation of ECOA via Reg B, lenders have been scrutinized using both theories, with disparate impact becoming a particularly prominent tool in regulators’ toolboxes over the last 12 years or so.
When a lender’s policy, process, or model is challenged under the disparate impact theory, they generally have the option to push back by arguing that it is necessary to meet a substantial, legitimate, non-discriminatory need (e.g., managing default risk) and that there is no less-discriminatory alternative (LDA) that can also meet that need.
Now, the Trump Administration is instructing regulators to eliminate disparate-impact liability as much as possible, which raises two interesting points.
First, like many actions undertaken in this administration, no laws are being revised here. Legally, nothing has changed with the theory of disparate impact. ECOA isn’t being updated, and all the existing case law (including the Supreme Court case Texas Dept. of Housing & Community Affairs v. Inclusive Communities Project) still applies. Federal regulators may choose not to pursue enforcement actions based on disparate impact for the next four years, but state attorneys general can still do so. As can private litigants, using public information to demonstrate disparate impact and bring their own cases.
Second, while I understand (and, in some cases, agree with) the philosophical concerns regarding disparate impact, I think it (and the requirement to seek out less-discriminatory alternatives where available) is a really useful mechanism for detecting and reducing subtle forms of bias. This helps address structural and historical forms of discrimination, but just as importantly (from my perspective) it gives regulators a tool to assess the harm caused by credit decisioning systems that doesn’t rely on an explicit understanding of how those systems were designed, which will become increasingly useful in the age of large language models.
We’re gonna have to work on our communication.
What happened?
Block has put out a statement justifying why it does not furnish repayment data to the credit bureaus for its Afterpay pay-in-4 BNPL product:
Recent industry announcements in the U.S. about Buy Now, Pay Later (BNPL) credit reporting have sparked important discussions about the future of consumer credit access and consumer data. Afterpay does not currently report to credit bureaus in the United States, and we won't until we see concrete evidence that BNPL data reflecting responsible payment behavior will help, not hurt, the credit scores of our customers.
So what?
This statement is in reaction, I think, to FICO’s recent BNPL announcement (which I covered extensively in the newsletter and on the podcast), and I find it fascinating.
The company states, “We won’t furnish data until we see concrete evidence that BNPL data reflecting responsible payment behavior will help, not hurt, the credit scores of our customers,” and yet … that’s exactly what FICO and Affirm demonstrated with their recent joint study and it appears to be one of the primary motivations behind the design of the new, BNPL-focused versions of the FICO score.
Now, to be clear, I have some concerns with the approach that FICO and Affirm have taken on furnishing and scoring BNPL data, but the one thing we can say with certainty is that it absolutely is designed to help, not hurt, BNPL users’ credit scores.
Does Afterpay not trust the work that Affirm and FICO have done on this? Or are Afterpay’s customers different enough from Affirm’s that this methodology might impact the two customer bases differently?
Or, perhaps, Afterpay just doesn’t want to share its data!
I think this might actually be the answer (even though Afterpay would never admit it) because the arguments they make in the statement doesn’t make much sense.
For example, the company argues that its loss rates are low and therefore its underwriting models are superior to credit scores, and therefore the negative impact of reporting BNPL data to the credit bureaus proves that the credit scoring system itself is not (yet) compatible with these modern innovations:
Afterpay demonstrates that modern underwriting can expand access while managing risk effectively – internal models consistently outperformed traditional credit scores in predicting repayment, allowing approval of 13% more customers while maintaining loss rates below 1% (Represents Afterpay Pay-in-Four loss rates through Q1 2025).
In fact, 98% of Afterpay purchases incur no late fees and 95% of installments are paid on time. This demonstrates our commitment to responsible lending and strong customer outcomes. The data also aligns with recent findings from CFPB economists, whose research shows that consumers repaid their BNPL loans 98% of the time. This independent validation of what we've observed in our own data reinforces our belief that responsible BNPL products can serve customers effectively without immediate credit reporting.
Maybe!
Or maybe your loss rates are low not because of your underwriting prowess, but because small-dollar pay-in-4 loans are structurally safer than other consumer credit products? And maybe because of that structural difference, it might make sense, in some cases, for the credit scores of habitual BNPL users to go down a bit, even though they are paying on time?
Maybe?
I don’t know. I just wish BNPL providers would be more honest in their communication on this issue. It’s perfectly fine to say, “It’s our data and we don’t feel like giving it away, especially if it also hurts our customers’ credit scores.”
(Movie Fan’s Note — The best parts of the movie are when Will Smith and Jeff Goldblum are together. It’s kinda crazy that they don’t meet until the end of the film, but the last act wisely leans into their odd couple dynamic, and the result is wonderful.)
Oops
What happened?
The CFPB published a technical correction to some old data that it had published:
In 2015, the CFPB released a widely cited report that provided a benchmark for estimates of consumers with limited credit histories. More specifically, the report included estimates of the adult population in the United States who, in December 2010, did not have a credit record (“credit invisible”) or who had insufficient credit history to have a credit score (“stale unscored” and “insufficient unscored”).
Subsequent analysis with updated data and a methodological correction reveals that the original estimate of credit invisibles should be roughly cut in half, with an almost commensurate increase in credit records that were unscored.
So what?
This one is dedicated to Ron Shevlin, who asked how many old newsletters I would need to rewrite based on these updated figures.
To answer Ron’s question, I don’t know, and I’m too busy to go back into the archive and check.
However, I do find this interesting, for a few reasons.
First, I’m surprised the current CFPB, which isn’t exactly firing on all cylinders right now, had the bandwidth to go back and correct these numbers. Truthfully, I’d prefer that they focus their limited resources on ensuring that, I don’t know, military servicemembers get reimbursed by the credit union that had been ripping them off. But sure, yeah, this is good too!
Second, I think it’s essential to emphasize that the estimate for the overall number of U.S. consumers without a credit score remained unchanged, despite a decrease in the estimate for the number of credit-invisible consumers (those without a credit file).
The original data source that the CFPB used in 2015 to estimate the number of credit invisible consumers omitted records that contained only deferred student loans, collections, or closed accounts. When you account for that, we see that many consumers previously thought to have no bureau file actually had one. This is what drove the 2010 credit invisible figure down from 25 million to 13 million.
However, while those consumers did have credit files, most of them did not have files that were sufficiently thick or recently updated to be scored by traditional credit scoring models, such as the FICO Score. Thus, the CFPB’s revised estimate for unscorable consumers in 2010 went up from 17 million to 29 million.
So, all in all, the estimate for the total number of folks in the U.S. who couldn’t access affordable credit in 2010 remains the same (approximately 42 million).
Third, the good news is that by 2020, that number had dropped from 42 million to about 32 million. And interestingly, the big contributor to that was a significant drop, between 2010 and 2020, in the number of true credit invisibles (7 million, down from 13 million), rather than in unscorable consumers (25 million, down from 29 million).
This suggests that we’ve been doing a much better putting credit invisible consumers on the credit bureaus’ radar (perhaps through mechanisms like rent reporting), but we still have a long way to go when it comes to helping consumers build and maintain scorable credit histories.
