08 May 2023 |

A Bucket of Cold First Principles 

By Alex Johnson

Editor’s note – this interview with Frank Rotman, Chief Investment Officer at QED Investors, has been edited down from a longer interview, which can be found here. The edits were made to improve clarity and conciseness and to make Alex Johnson sound like less of an idiot.

Alex Johnson: Instead of asking you to predict things that are gonna happen in 2023, which I knew you’d say no to, I’ve come up with a different way of tackling this, which is what I’m calling “a bucket of cold first principles.” 

You have a very well-deserved reputation for being a rigorous thinker in fintech. So I was hoping to play a little game where I can run some predictions, maybe somewhat cliched predictions, about what’s gonna happen this year or in the next couple of years and then have you respond with, “agree”, “disagree”, or “it depends”. And I’m guessing that every one of your answers will be “it depends”, because, again, there is a lot of nuances in these topics. But more importantly, please walk us through how you would approach thinking about each prediction from a first principles perspective.

Sound good?

Frank Rotman: Sure. I’m game.

Alex Johnson: Okay. First one, a very clickbaity headline – ChatGPT will revolutionize the way that consumers manage their finances.

Frank Rotman: So, let’s start with the technology itself. ChatGPT is amazing. It is a very interesting new technology. And you know, if you play with it for a bit, you can see the power and how if this gets just a little bit better, it could serve a lot of purposes. 

By way of background, if I go back 30 years to when I was in college, I studied artificial intelligence. My background was in systems engineering and artificial intelligence, so this is a big piece of my background and something that I think and care a lot about. 

For the longest time, computing power wasn’t where it needed to be. And the data sets weren’t there to train these AI models. But now we’ve gotten to a point where it’s actually functional, and you can project that it’s going to be valuable. 

Will it be transformative in financial services? And if so, where? 

You have to think about how important an accurate answer is, right? If you need 100% accuracy, then ChatGPT and all this new artificial intelligence technology is not ready yet. Getting that last couple of percent to 100% accuracy of an answer is really important. 

For example, if you want ChatGPT to do your taxes for you, it has to be accurate. If it’s not accurate, you go to jail or you get fined. That’s a problem, right? 

So it’s not going to be ready for primetime anytime soon. For basic taxes, maybe it will get there at some point, but for complex situations that require 100% accuracy, the technology just isn’t ready. In the world of financial services, there’s so much regulation where you have to make sure that the advice that you’re giving or models that you’re implementing are accurate and that’s just a very difficult standard to pass. 

So when I get excited about ChatGPT, I get excited about its ability to write, its ability to synthesize information, its ability to make lists, its ability to take 10, 20, 30 sources of data and condense it into something that makes sense to me instead of you having to search Google to be your source of truth and having to click through 10 different websites and learn a little bit from each one, being able to synthesize and then refine and synthesize and refine. I think it’s going to be an incredible tool for people who are trying to learn things or trying to create. when it comes to approve/decline and pricing decisions for credit or for financial recommendations for investment products, these aren’t things that ChatGPT is going to be ready to do anytime soon.

Alex Johnson: OK, next one. Elon Musk’s plan to turn Twitter into a fintech super app will succeed.

Frank Rotman: It depends, but I would probably err on the side of saying that it won’t.

The movement of money is a major component of banks and banking services. It’s one of five major pillars of things that banks do. The movement of money is where a lot of platforms are making their money. The core product is something that they can sell for (hopefully) more than they manufacture it for, but a lot of these firms are now making more money, moving money than they are in their core product.

I mean, look at Shopify as an example of a giant fintech that is disguised. They make massive amounts of money just moving money between you, the small business owner that is on the platform, and the consumers that are buying products, right? So it’s a playbook That makes a lot of sense.

I think the problem with Twitter now is that it’s neither twix nor tween. There isn’t a coherent strategy, and I think the strategy is going to get in the way of layering payments functionality on top of it. I think the platform has more of a core problem that it needs to solve. If they start to go down the path of payments before they anchor their core product, then I think it’s going to struggle.

Alex Johnson: I think that’s a really good point. I agree.

And I would be remiss in my job as podcast host if I didn’t ask you to spell out the five pillars of what banks do, just so that if someone’s listening to this, they can have Frank Rotman’s list of the five pillars of what a bank does.

Frank Rotman: There are actually a lot of things that banks do, but they tend to fall into the storage of money, which is deposits, the movement of money, which is payments, the borrowing of money, which is lending, the investing of money, which is wealth management, and the transference of risk, which is insurance. Banks can’t necessarily do the insurance piece, but it’s a pillar of what we do in financial services.

Alex Johnson: Okay, great. Next one – cash flow underwriting will displace the FICO score as the primary tool used to evaluate a consumer’s creditworthiness.

Frank Rotman: I disagree with that.

Alex Johnson: Excellent!

Frank Rotman: I was the first Chief Credit Officer at Capital One way back when, so this is one I know a fair bit about.

What you are trying to do in underwriting is understand a bunch of information and behaviors about an individual or a business entity that fall into a couple of different buckets: ability to pay, willingness to pay, and stability of income, or in the case of companies, revenue.

If you think about this from first principles, these are the drivers of risk that need to be understood, and in order to understand, you need truth files. So a source of truth about the inputs and a source of truth about the outputs. And you need to build statistical models that basically take all of these inputs and try to predict the outputs.

If you look at cash flow underwriting, it’s very good at getting at a single thing, which is ability to pay now. So are you technically solvent or are you technically insolvent? If we add an additional burden on top of this entity or person, would they be able to make their payments? And that is a yes/no answer. You can calculate it using cash flow data. It’s a very good truth file for that.

However, when you start to get into some of the other key components of risk, cash flow data is only as good as the length. The data set that you have. So if I’m looking at an individual, if I’m ingesting 12 months worth of bank statements, I have 12 months of the stability of the income.

Alex Johnson: And just to be clear on that point, that’s roughly what most aggregators are providing, right? When you’re looking at the way that open banking, at least in the U.S., works right now, you’re not getting 10 years of cash flow history. You’re getting …

Frank Rotman: Three months to a year. Yeah, three months to a year. And that might change at some point, but you know, the reality is that, as a truth file, it is a much better truth file for now than what happened in the past.

But when you look at something like stability of income, there are other items that provide a lot of predictive value. The occupation a consumer is in is a great example of something that has a lot of long-term predictive power. There are certain job families that, if you actually look at the Bureau of Labor Statistics over a 30-year period, the unemployment rate for a specific profession is extraordinarily low. Teachers and nurses are great examples. They have incredibly low unemployment rates. Then there are other professions that have fairly high unemployment rates, right? If you look at manual labor, a lot of construction goes through boom and bust cycles. If you look at plumbers, they go through boom and bust cycles. There are professions that have instability almost built into the ecosystem that play in. And you can imagine, from a risk perspective, one is a safer risk than the other.

If you’re making a one-year loan, cash flow underwriting might be great. If you’re making a 30-year mortgage, understanding the long-term stability of income matters a lot more.

But even using stuff like employment history data has its own flaws because of something called Reg B. Reg B is all about fair lending, and there are a lot of good reasons for fair lending requirements, stemming from the United States’ long history of discriminatory lending. And because of disparate treatment of different minority groups in the past, laws were put in place so that you could not have disparate treatment going forward. 

The problem with Reg B is that the way that it is implemented is not just about disparate treatment but about the actual outcomes. So ultimately, I cannot use a lot of information in statistical models because they would result in protected classes being adversely affected by the models’ decisions. And again, there are a lot of good reasons for these regulations because of what happened in our history, but if you were to use all the data you could get to perfectly predict risk, you would actually be breaking the law.

It’s a very, very difficult topic. Almost a third-rail topic, where you have statisticians constantly trying to improve the models and looking for incremental signal, but Reg B is saying you have to pull some of that signal out because it will result in disparate treatment of protected classes. As a result, innovation has stalled in the underwriting space.

Alex Johnson: I very much agree with everything you just said, and I’ll leave that one there because you covered it much better than I ever could. 

One more – Buy Now Pay Later (BNPL) will prove to be a sustainable competitor to credit cards.

Frank Rotman: This one makes me laugh. 

When Buy Now Pay Later took off, I started to scratch my head. I was just like, “It’s just credit”. Everyone else is like, “No, it’s this magical new thing.” But I’m like, “It’s credit. You’re trading cash today for a flow of payments in the future that have variability and volatility associated with it, and you make that trade because it’s yield generating, right?”

I mean, we have invested in companies in the space. GreenSky was in our portfolio. Back in the Capital One days I helped manage an elective medical finance company that was a Buy Now Pay Later company.

Alex Johnson: Right. We’ve been doing this type of point-of-sale financing for a while.

Frank Rotman: It has been a long time. And if you actually go back to retail lending, there’s a whole bunch of codes where retailers do not have to actually have a lending license if they’re going to lend in a small number of installments. So this has been around forever, and in fact, if you actually look in countries outside of the U.S., especially where the middle class is still emerging, the number one way of getting credit would be to go to a retailer and for the retailer to break things down into installment payments for them.

Alex Johnson: Right. In Brazil, Buy Now Pay Later has been around since the 1950s.

Frank Rotman: It is the number one source of credit for individuals in the country. So it’s an interesting concept to say it’s going to compete with credit cards. I think what’s happened is point-of-sale finance has really seen a resurgence because so much more commerce is being done online and being able to have real-time decisioning and embedding these decisions at the point of sale; that’s the real innovation, right? 

So will that compete with credit cards where new credit could be issued at the point of sale? 

A lot of loans are sold in terms of monthly payments. You buy your car based on monthly payments, right? Buy Now Pay Later is really a function of saying to the consumer, “You know your budget; you know how much free cash flow you have. If you buy this thing, it will cost you an extra hundred dollars a month. Are you interested?” 

I think that’s very powerful, from a psychological standpoint and from the standpoint of reducing friction, so I do think Buy Now Pay Later will survive from this point forward.