Moody's Talks - Inside Economics - More Science Than Fiction

Episode Date: August 28, 2026

Stripe Chief Economist Ernie Tedeschi joins the Inside Economics team to separate the signal from the noise in a busy week for the economy and financial markets. The group discusses Kevin Warsh's recent speech and the market's reaction, unpacks the latest PCE inflation data and what it means for the Fed, and explores the remarkable surge in business formations. Are more Americans becoming entrepreneurs because of shifts in how firms are organized, or is technology making it easier than ever to launch and run a business? Ernie shares insights on the rise of solopreneurs and what it could mean for the future of work, productivity, and economic growth. Questions or Comments, please email us at [e-mail]. We would love to hear from you. To stay informed and follow the insights of Moody's Analytics economists, visit Economic View. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Welcome to Inside Economics. I'm Mark Sandy, the chief economist of Moody's Analytics, and I'm joined by one of my trusty co-host, Chris DeReedies. Hey, Chris. Good to see you, Mark. Merce's off gallivanding somewhere in the world again. Economic research, I think, is what she calls it, field research. Field research. Okay. I wish we all had that kind of job.

That's a nice job to have, but we'll miss her. But we've got Matt Collier. Hey, Matt. Good to see you. Hey, Mark. Nice to see you too. Hey, Chris. It's a busy week of economic data. a lot to cover, and we always have Matt to help us out with that. Of course, the inflation data in particular, but we'll get to that in a minute. And we have a guest, Ernie Tedeski. Hey, Ernie,

good to see you. Nice to see you, Mark. Thanks for having you. Yeah, no, thanks for joining. Where are you hailing from? Where are you? So I'm normally D.C. based, but right now I am in San Diego, my hometown. Ironically enough, it's actually more humid in Southern California this week than in Washington, D.C., which I think is the first. Oh. How do you explain that? Is that El Nino or something? Yeah.

Maybe it's Warsh's speech. You're referring to Kevin Warsh or a new Fed chair. That's right. And his Jackson Hull speech. Yeah. Well, we'll come back to that in just a second. But before we get to that, I should say, you know, obviously you're the chief economist and lead, or is it chief or lead economist?

Does it, is it one or the other? My official title is head of economic insights and research, but colloquially we call, I'm the chief economist. I'm going to say chief economist, if that's okay. Yeah, chief economist of Stripe. That's cool. And you've been there for how long, Ernie? So I've been there for nine months now. Brand new role. Strip has never had a chief economist. They've never had an economic insights team. I mean, they got me because they pitched it as 70 trillion commercial transactions that Stripe is involved with on the back end, and they have data on all of them. And they needed an economist to sort of dig in and come up with external insights on what's going on with the internet economy.

It was sold. Did you say 70 trillion transactions? Is that what you said? Is that the beginning of time? Or what is that? Per year. Per year? Yeah.

Wow. So, right, Stripe is a payments company. And you can think of the data that we have as maybe similar in vain to what a credit card company would have, except one, it's just in terms of the data itself, it's more detailed. We have sort of transaction and goods level data that we record. Stripe powers, you know, so much of the e-commerce that you see. So you go to Etsy, you know, Stripe powers that and we have data on that.

Shopify is powered by Stripe. But then unlike many credit card companies, you know, we cover both consumers and B2B as well. So there's this whole insight into what business spend is doing that we have as well. Are you global? I know you're U.S. UK, but you're global. We're global. Yeah, we're present. I think we're present in almost every kind.

I don't think we're in North Korea. But we should be president in most countries. Yeah, our three biggest markets, I believe, are the U.S., the UK, and Australia right now. Got it, got it. And Stripe's been in the news recently around PayPal, OpenRouter. We won't go down that path, but a lot of interesting stuff happening at Stripe, for sure. Yeah, I think that we're, we are trying to be sort of the back end and the payments company of choice in the AI economy in particular, which I think is. you know, a good bet to make, good strategy.

And look, selfishly as the chief economist, just means more data on the AI economy for me. So I love it. Yeah, very cool. And, you know, you weren't just born and then joined Stripe. You had a, you've had a long career. I mean, I got to know you a bit when you were at the Council of Economic Advisors. You were the chief economist of the CEA, weren't you?

Yeah, that must have been a cool job. That was an incredible job. That was, so taking a step back. So my background is as never worked in a tech company before. My background is as a macroeconomist, labor economist, public finance economist. I worked for Evercore ISI on the sell side, on their monetary and fiscal policy teams. It was there for seven years. It was great.

Really enjoyed it. But then the Council of Economic Advisors came up, went over, eventually became chief economist. That was a cool job because that was managing the macroeconomic data workflow to the president. And so, you know, it was me and a team of economists that would get to see the macro data the night before and have to interpret it. Which is, I will say that that is one of the most interesting and unique professional challenges I've ever had. when you get the jobs report the night before and you can't look to Twitter or Bloomberg or, you know, like to get like a survey of reactions, it's all you. And you could have called me, Ernie.

I'm just saying you could have given me a call. I would have been happy to help you out with that. Yeah. I will say, you know, I thought I thought we were pretty good at interpreting the like economic implications of what we were looking at. But like the market, the market implications, we were. at best 50-50 and anticipating those. So that was one of the more interesting things. That's a better ratio than I have. I never, I never, really? That's what you think that

that release met? Yeah. Really? More than a few times. It was like, okay, got it. That's not what we expected. Yeah. And was Jared Bernstein your boss, C.C. Rouse? Were you both of Yeah, so I came under Cici when she was chair and then stayed on for the transition to Jared Bernstein's chair. So got to work under two great chairs. I won't tell Jared what you say here, but was he a good boss? Was he a good boss? He's a great boss. He's like, Jared as a boss is sort of like this podcast as a boss, which is like you sit.

It's casual, but like so intellectually simulating and deep. Yeah, he's great. You just sit down and you have a conversation with them about he loves talking about data, about like what's in the, you know, it's such a great environment. Yeah, really good guy. Well, thanks again for joining us. And let's talk about Kevin Warsh, the new Fed chair's speech at Jackson Hall. That happened just a few minutes ago.

This, we're here coming up on 11 a.m. Eastern Time on Friday the 28th of August. Let's go to the market reaction, talking about market reaction. What was the market reaction, Chris? So I'd classify it as a bit mixed, right? If you just look at the stock market and the long-term 10-year treasury rate, not much changed, right? Stock market, and the stock market's probably reacting to a lot of other things going on, AI and whatnot.

But pretty flat this morning, not much reaction. And then the 10-year as well, up, I think, two basis points. So a little bit of reaction, but nothing major. where you do see the reaction is in the Fed funds futures, right? So that's the probability that the Fed will hike, say, at the September meeting, that went up by about five percentage points. So from about a 35% chance to closer to a 40% chance of a hike. So still not over 50, but moving in that direction.

So that's one. And then if you look at shorter maturity treasuries, say the three-month or up to the one year, even the two-year, those were up for. fairly substantially, like seven, eight basis points, right? So that's a pretty big move for that kind of shorter end of the curve. So I'm sorry, did you say the futures are still not expecting a rate hike at the September meeting? Correct. A 60% chance of no hike, 40% chance of a hike.

Okay. And what was it before the speech? So it was 35%. Oh, so very small change on September. A little bit of an increase, but nothing major. And can you tell, and you may not be able to, but looking out into next year, as I recall, it was a 75% probability of rate hike by next March. Is that changed? See, I did see.

You're taking a look? I took a look. It wasn't a major change. Let me see if that's spell the case. So all in all, a reaction, but a meaningful reaction, but kind of small on the grand scale in the grand scheme of things. Yeah, that did come, I mean, now it's a 97% or no, 93% chance of a high. Okay, so that did go up. Yeah.

Okay. Hey, Ernie, I know you were saying before we got on that you were actually listening to the speech or watching it. You know, anything, does that consist, kind of the market reaction consistent with your reaction? You know, what did Warsh say that caused participants, the market participants to think that there's a greater chance the Fed's going to start tightening policy here? I thought it was interesting. So, you know, and to be clear, reaction, you know, my personal reaction is all 20 minutes old at this point. Well, that counts for a lot. That counts for a lot.

I don't know what that. I came into this speech thinking, you know, we might, he might acknowledge some of the weaker inflation reads that we've gotten month on month recently. And he actually didn't emphasize that as much as I thought he would. I thought he actually came out pretty crisp. Don't quote me on this line. He had a line in there something like the labor market is consistent with full employment. And that this is not a quote. The problem is on the price.

And I thought he was actually more hawkish on prices than I thought he might have been coming into this. No, you know, he's obviously anti-forward guidance. You know, I didn't, he didn't have any sort of explicit sort of podcast. about his reaction function or what would happen in September. But like if you were trying to divine something coming out of the speech, I thought it was interesting that leaned a little bit more hawkish on the price side. It is consistent with the market reaction. So to your treasury, which is kind of a benchmark for how the market thinks about where the fed's headed,

that's up seven, eight basis points. That is consistent. Sounds like it's consistent with your take on things too. Yeah. Yeah, that's how, you know, based on that speech, I would have expected a small market reaction in that. Got it. Chris, you want to say something? Yeah, I saw that finger go up.

That's right. So Matt is, you know, hawkish here. He's looking at the data like a hawk, I should say. And he noted that just in the last few minutes, as we were speaking, the odds of a of a hike went up to 55% in September. So markets moving around here a bit as the news gets out. Right. Hey, did you have a chance to look at the speech or watch the speech, Chris? Did you ever change to do that? I read some summaries of it. Okay.

Any reaction? Any other than what Ernie said? I mean, I read the speech because I didn't have a chance to listen to it. And that was my exact reaction. I mean, I was, he called out the better June and July inflation numbers, which we'll talk about in the second. But he pretty much dismissed them saying that, you know, that doesn't give me a sense that things are moving in the right direction here. and it felt like he was trying to establish his credentials here that I, you know, I'm, he also, the other interesting thing he said, I expected it, but I was glad to see it, was he reaffirmed the 2%

inflation target and he reaffirmed the PCE deflator, the consumer expenditure deflator as the key measure, which, you know, is something for him because he had been talking about trim mean and other measures. He did mention one interesting measure of inflation is that the percent of, of goods and services whose price increases over the past years over 3% in the, what they look in the core basket of goods and services they look at, it was over half. And he mentioned that, which is also kind of hawkish, you know, saying, hey, look. And then his talk about the economy, as Ernie, you pointed out, it was pretty upbeat. I mean, much more upbeat than my take on things.

But, you know, he was saying the economy is not a problem. But inflation is. And so that's all consistent with, you know, the kind of. the market reaction. I did notice that he began the speech, Ernie, with a little bit of talk on AI. Did you catch that? You must have seen that.

Did he say anything there? I kind of read that quickly. I didn't really see anything of consequence in that kind of conversation. Nothing of consequence. You know, I, he's asking the same questions that we are, right? Which is, there's a lot of uncertainty, of course. But, you know, Warsh brought up questions around not just the obvious questions like productivity effects and, you know, effects on the labor market, displacement,

et cetera. But, you know, I think more conceptual things like he mentioned tokens as a factor of production, which I'd never heard it framed that way before, but I thought that was really interesting. I may have to steal that framing internally. Yeah, that's good line. And how do you think about it. The other thing was, and I've talked about this too, but I like the way that Warsh talked about it, talking about, you know, who captures the profit? in the returns to AI, which I think is it's really interesting because obviously AI, in a lot of cases, not every case, it's a little different from the way that like we might textbook think

about capital investment, capital ownership. You know, if you're Ford, you own a factory, you own a production line, Ford captures the returns to that capital. Whereas in a lot of cases with AI, right, you're outsourcing the productivity gains in the return to capital to an LLM firm, right? not in every case, you might have a proprietary AI model that you own, et cetera. But like it's a, it's a different dynamic than sort of the textbook case, not in a good or a bad way necessarily. It's just different. And so we might have to think about it.

Interesting. Yeah, that is a good line. Token is a factor of production. Yeah. Interesting way of thinking about it. It actually makes it easier to think about it. So it's a good line. One of the funniest things about tokens is that like people,

keep trying to cram in different concepts of different things into token. So like I've heard token as a currency, token as like token as intelligence. Now token as a factor. It's like, oh, token is like this blank slate that can do everything in the economy. Yeah. Like a forest gump, you know, the economy. Yeah. Something along that. That is probably stretching it, but you know, something like that. Okay, but there's a lot to talk about. You've guys done a lot of great work at Stripe. We want to talk about AI productivity, the work you've been doing on solo partnership

and business formation. Very cool. I want to play the game. You said you're up for the game, the stats game. So we're going to play the game at some point. But before we get to all of that good stuff, let's do some of the blocking and tackling around the data. So, okay, we got a raft of economic data this week from GDP,

durable goods, got trade. got inflation, got income, got saving, some housing, a whole bunch of stuff. Where do you want to begin? What top of mind for you in terms of the economic data releases? I think if we're talking about Warsh Tre Rorsch's speech, I think inflation is probably a good entryway in now and then likely the data he's referring to. So PC deflator came out for July. So what happened there, 0.2% income. increase in both the headline and the core PCE deflator. Keeps the year-over-year rates for the headline PC at 3.7%. Core PCE 3.3%.

Doesn't sound all that interesting, at least superficially, but underneath the hood, core PCE. 0.25 or 0.249 to the third decimal point, so very close to being a 0.3% increase, which may not sound like much, but it is, at least psychologically to see that as above expectations of 0.2% growth. Headline PCE was also a little bit on the strong side. Market expectation, our expectation was for a 0.1% increase that rose 0.2%. Further, I'd add that the 3.3 year-over-year increase for the core PCE, very close to being 3.4%.

So we're getting moderation in energy markets is one story, this other sense that we're starting to see, with this hope that we're seeing some kind of moderation broadly in inflation, and we would have seen a year-over-year increase in Core PC, which would be contrary to that story, of course. The big driver for Core PC inflation is... Can I push back, though, just a little bit? I mean, because you're painting a picture that inflation is still a little hot here.

Certainly relative to expectation. But correct me if I'm wrong, a big chunk of that increase was in the cost of financial services. portfolio management, I think, is the term. And that's imputed, and that's imputed based on, or not imputed, it's based on basically movements in stock prices. So if stock prices are up, which they were in the month, then you get a bigger increase, which is kind of bogus, you know, in terms of the measurement.

And they're going to change that, right? The B.A is going to change that next month, I believe. So if you account for that, do you agree with what I just said? And if you count for that, what was inflation in the month in year over year? Yes, I agree. So 40% of the increase, core PC, 11 basis points is of the 24 basis point increase is coming from that wacky financial services portfolio management measure. You take that out. You're down to 0.1% increase. Okay. So, okay. So isn't that, doesn't that feel better? I mean, doesn't it feel like it's moving in the right direction? Ernie, what do you think about that? I mean, am I... I kind of agree with you that we should look at market-based where you, where you filter out all of those imputations. That was something that we started doing. at CEA when we were trying to get a gauge on underlying inflation because like the market

gyrations are like energy, right? Like they're essentially a volatility adder. So I think it makes a lot of sense to take them out. It's still, if you look at core market based PCE, year over year, so you're smoothing through the last couple of months, it's still 3%. Now, you could argue, okay, but like you change the metrics. So maybe it's not quite a 2% consistent target if you're using market-based. I think when I've looked at this before, I think you have to get like 2-5% or so on market-based to get something consistent with 2% PCE. You know, there are a lot of different ways to look at that.

But even by that measure, we're still hot on market-based core PCE. I think that there's, it's more than just financial services, I think, going on. So are you of the same mind as Chair Warsh that inflation is kind of more of the issue here than growth that you are? Yeah, and I've changed my mind on this. Like I in 2025, I thought, well, you know, tariffs are a one-off, not literally a one-off. Like they were, you know, obviously the policy was changing. Pass-through was variable and, you know, there are lags. And so, like, we were going to see the effects of tariffs gradually over time.

I still go back and forth on how I think about the AI effect on PCE and just how we've seen these massive increases in memory prices. Like, part of me thinks, well, that's also kind of one off. Part of me also thinks, like, no, that's like a legit structural issue with inflation, you know, with inflation, with high demand. But even if you filter through all of those things, like, it just still seems. like when you look at market-based core services, so not affected by tariffs, not affected by the memory side of things, still hot, still much hotter than it was just before the pandemic. So, you know, I think I agree with Chair Warsh that the main problem here is hot inflation, and it's still above time.

So if you were on the Fed, would you be one of the folks arguing for a rate hike in September later in the year or early next? Yeah, I think so I think just based on the only, so the only thing from a dual mandate perspective that gives me pause, so I'm a little less sanguine than Chair Warsh is on full employment right now. Given some of the weakness in, especially new graduates, younger people, I'm not so convinced that that's an AI story because we also see weakness

among non-college young people and their employment rate. So at the very least, it seems like it's broader than just AI affecting the labor market, if at all, right? Could be high interest rates, right? Could be, you know, could be sort of a bullwhip effect from the earlier, you know, pandemic turn, you know, that we saw and now we have a low-turn labor market. You know, I'm not quite sure where I land on that. But, like, inflation is just above target.

It feels like to me. And, you know, I would probably lean on the hawkish side. Yeah, the thing that bothers me about this argument where full employment is wage growth is definitively decelerating and now is falling on a real basis after inflation. I mean, that's not consistent with a full employment economy. You know, it's just not. I mean, I know the unemployment rate is low, but that's only because labor force participation is collapsed. I know there's measurement issues, but, you know, if you just kept participation constant over the past year, we'd be over 5% on the employment rate. And the whole narrative around full employment would be, you know, very, very different.

So, but anyway, hey, Chris, what about you? I know you've been kind of more doveish, right, on all this. I mean, what's your take on the inflation numbers? Do you have a different perspective? Still hot, but I am more worried about, or I am quite worried. about the labor market. Labor market. Softness that we're silly there.

So for me, it's still a 50-50 mandate, if you will. You wouldn't cut rates, but you would just hold rates on change. I would hold that certainly for the moment. All right. Yeah, consistent with me. So, hey, Matt, back to the PC inflation number, the consumer expenditure deflator. Anything else you want to call it? I know there's a bunch of changes coming, methodological changes that are going to be

incorporated next month and that's going to reduce the overall rate of inflation. Is that right? Do you want to just call that out? Yeah, just scale. we're looking at 3.3, this time next month, when we get August PC deflator, this month's year-over-year growth will be three flat, maybe 3.1% based off of these three methodological changes that the BEA are making. One primarily is focused on what we're talking about with portfolio management, so less directly impacted by equity market prices. second computer software although that's still a little bit unclear on how that's the the the new compositional measure is going to be used it's some cpi some ppi measures but in general pretty confidently it's going to have a negative effect because it's not just going to be the cpi component

that's been running hot being being accelerating recently over the past year or two because of the i i build out so that effect will have a little bit of a lowering effect and then third smallest effect being legal services just the way that's been changed and that's kind of a collection issue that the BLS has had with CPI. So I think the two to three tenths of a percentage point reduction in core PCE is a pretty safe way to think about it. We can get into how that might be revised in subsequent months as a kind of follow-up methodological change is going to be implemented, but for the time being September's meeting in a few weeks, 3.3% year-over-year core PC inflation. The next time the Fed meets, that will likely look like 3%. So I don't think

the Fed's going to be whipsawed by any of those technical changes. And in general, I think the conclusion is one that we're all seeing or we're all agreeing with here, which is too high, but not accelerating terribly towards 4% or anything. Yeah, my sense is that we come up to the next meeting, we're going to get another good inflation print that's flattered by these methodological changes, but even just abstracting from that, it's going to be another point two kind of number, you know, kind of in closer to target. And I don't see the job market improving. It feels to me, like if you have wage growth that's in negative territory, we didn't even talk about real incomes. They're flat at best on a year-over-year basis. Saving rates are very low. Consumer spending feels a little

punk to me that, you know, we're going to come in with a kind of a softish economy with the inflation numbers moving in the right direction. That's just not the fodder, you know, were a rate increase. But, but Ernie, before we move on, did you want to, you mentioned this, you alluded to it, but maybe you can flesh it out a little bit. How big an impact do you think AI is having on the inflation numbers? Actually, real quick, Matt, I should know this. Are these revisions going to be applied retroactively? They are, at least back through the annual updates are typically five years. I think that the BA will implement that through. I don't know how close you are to thinking about the way that they're going to be measuring financial services.

So they're dependent on the quarterly services survey now for wages and that's on a different lag. So there's going to be a revision to where that data is all available, which will be Q1 at the time of the change. Then we'll get the revised Q2 estimates in December. So that will happen. So there'll be kind of two tiers of revisions that happen all at once. And the biggest effect will be where we don't have that QSS data. So June and July is year over year. I'm sorry, Q2.

and then July, year-over-year rates will be under a different way of measuring financial service wages. When you look at, to answer your question mark, like when you, under the old, under the current old, soon-to-be-old methodology, when you look at computers and peripherals on the PCE side, and there's a big difference here between the effect on PCE and CPI, you know, I think that the the acceleration was adding, you know, 20 basis points or so to year-on-year PCE. inflation. So like look, like not not the difference between, you know, like being at target and above target, but like that's that's a meaningful amount. That'll go down, you know, as they revise their methodology. The story that I tell is that so, so the computer that I'm talking to you on is a, is a brand new laptop that I bought several months ago. And I didn't, I couldn't get quite the components that I wanted because I would have to wait like a couple. It's a PC, not a Mac. It's a Dell, actually.

So I got my Dell back in April. Okay. Priced out. Couldn't get the components I wanted. But I was like, you know, I need this sooner rather than later. So just bought it. And then on a lark, like, you know, a month or so ago, I went back and I priced out the exact same specs on the Dell site. And it was twice as much as what I had paid for.

Holy cow. In those years. So it's by far the best performing asset in my portfolio, which is kind of crazy. Yeah. Yeah, probably could pull out the chips and sell them for even a bigger markup, probably. I seriously thought about that. Yeah, yeah. That's good.

Well, let's play the stats game. And then we'll move on to some of the other work that you've been doing. The stats game is we before to stat. The rest of the group tries to figure that out through clues, questions, deductive reasoning. The best one is that's not so easy. We get it right away. One that's not so hard we never get it. and if it's apropos to the topic at hand,

and I'm not sure what that is, but so we can do anything we want all the better. And Matt, would you want to go first? Do you have a good stat? Sure. I hesitate because we always go to Marissa. I feel a little lost without Marissa, so.

I feel lost too. Yeah, right? Yeah. 2.3%. 2.3% in the data that came out this week? Yes. government data yes

well geez I mean why the hesitation yeah not it's pseudo government data I think that's the right is it fed data is it there you go okay it's fed data

Fed data 2.3% what what would that be um came out this week did industrial production come out this week I don't think so I don't think it did no that didn't come out this week

week. A Fed release related to growth in some way? No, that's not how I'm characterize it first and foremost. Okay. Related to inflation. Yes. Yes. Is this something that Kevin Warsh just called out in his speech? I didn't see the speech this morning, but it's very in his lexicon. Is it a trend mean inflation? Yeah, there you go. Dallas Fed? It's the Dallas Fed's trim-meaned PC inflation. So, Zurny.

Do you see how that was done? That was quite, I'd say that was, wouldn't you say that was masterful? I'm just, not to put words in your mouth. I'm just. No, no, no, that was, that was like 20 questions with my kids, but you did it in like five. That was amazing. And I was, and with Matt's head fakes, you know, oh, it's, what, it's not really government statistic.

Yeah. Quasi government. Quasi. Well, according to the president. it's fully part of the government, you know. It is the government. Okay, that was a good one, Matt. That was really good.

Do you want to explain? A measure that just technically is attempting to get at underlying inflation by ignoring the stuff that's disinflating or deflating and the things that are accelerating rapidly. So really just getting to the middle third of all the components within the PC and saying, well, what does that look like? like there's our measure of underlying inflation. It doesn't matter if you're cutting off, lapping off really important, heavily weighted components, but what's happening in the middle of the distribution there? And Chair Warsh at the beginning of his term as chair called that out explicitly as a potential better indicator of where inflation is at where it's headed. And of course, that's running much lower than the published measures of PC and CPI inflation. So that

that warrants a degree of skepticism, but in general, that still remains to be the case. It's lower than PC inflation, much closer to the Fed's target. I don't think anybody should hold that out as the sole indication of inflation, but that's where it is today. Hey, Ernie, what do you think of those trim mean or these different inflation measures that Cher Warsh has up until today, talk, been talking about? Do you put much weight on them? So from a policy perspective, no, the thing I worry about with, the thing I worry about

with explicitly trimmed mean, maybe a little less median inflation, but definitely with trim mean, is that like you're by design, by definition, filtering out the canaries in the coal mine, right? And so if you think that inflationary shocks start in a narrow part of the consumer basket and then spill over into other parts, then you're not seeing that. You're introducing a blind spot into the data. And this is exactly what we saw in 2021, remember, like the inflationary pressure began with goods, with automobiles in the early part of the year. And if you look at trimmed mean measures of inflation, right? Like you would have had a incorrect, doveish view of inflation early on in 2021 and totally missed the shocks flowing through the system. Yeah, makes a lot of sense to me.

Hey, you want to go next, Ernie? Or do you want to afford a stat? I have a very, you know, this is not a very creative one, but 4.2%. 4.2%. Is it something, government data? Yes. It is. Did it come out this week? Yes. Is it in the GDP accounts? I got it. Yes, it is.

Okay, go ahead, Chris. Go ahead. Real domestic private. Look at that. real domestic what? Yeah, real domestic private output consumption. Private domestic final purchases. Oh, oh, okay. Sorry, sorry.

It's a mouthful. Yeah, okay. Oh, is that real? I didn't know that. Is that real? Wow. That's, yeah, quarterly, quarterly real. Quarterly, the second quarter,

and that goes to the boom in investment spending, I guess. driving a lot of the train there. That's right. Actually, consumption was up too. Yeah, consumption. Got revised up there. Yeah, both of them were strong on a quarter to quarter basis.

Right. But yeah, so for those, for listeners who don't know, PDFP, private domestic final purchases, I colloquially refer to that as core GDP. So that's just business fixed investment and personal consumption expenditure. So it filters out things like, inventories, government spending, net exports. The reason why I like it is because it's actually a better predictor of where GDP growth ends up the next quarter than GDP itself is.

So it's a good way to sort of, not that the components that it's filtering out aren't important, inventories, government trade, very important. It's that they tend to be volatile. And so if you're trying to predict where GDP is going, this is a better measure. And it's coming in quite strong, which, you know, narratively, given how strong investment is, is maybe not surprising. But 4.2% real is no joke. Yeah, it's not consumption, right? Because that's, as I said, that's pretty punk. I mean, it's 2% at best. And we got a data point for July, and that was 2%-ish, you know, year-year. That's all in that. Hold on. So I'm looking at the day. So real PCE was, 3.5% quarterly annualized.

No, cute, yeah, yeah, on the quarter, right, but year or year. Yeah. I mean, it was 2% last year. It was 2% in the first half of the year annualized. It was 2% in the year over year through July. Yeah, you can find periods. But the underlying rate of, in my view, real consumer spending is 2%, which is okay. I mean, that's fine.

So the juice is all investment. And that's all AI, right? I mean, non-A.I. investment pretty soft. I mean, I know it was up in the quarter, but over the last year or so it's been, I think, even negative. I mean, so it's all AI. So it gets me into another question. Do you mentioned the effect of AI on PCE inflation. You say it was 20 basis points, 0.2 percentage points. What do you think its contribution to real GDP growth is AI all in? Yeah. So I, so, and here I think you have to make a distinction between like gross final demand

and GDP. So I think if you're talking gross final demand and you just literally add up the components of, you know, computers for consumers, CAPEX, data centers, et cetera, for investment, exports among American computer firms, you get something close to 1 to 1.5 percentage points. But the thing is, is that on the investment and the PCE side,

so much of that is imported that if you're literally talking about GDP. So activity within our domestic borders, it's much, much smaller than that. It's probably 0.3, 0.4 percentage points, I think, when I calculate it. But again, like, that's a good headwind. Yeah. Sorry, tailwind to having your economy. That's not a joke, but it's not the like, you know, that to me is not like a miraculous contribution to GDP. That's like a solid sort of, you know, tailwind that we might get every five or 10 years or so. But look, like if it sustains itself, then that's, you know, then that becomes a more

historic type of, you know, positive economic job. Well, in terms of the AI contribution, do you put much weight on the impact that the surge in equity prices has had on consumer spending, that, you know, that 2% growth of there's, some part of that is related to wealth, so-called wealth effects? Is that called your calculation? So no, that's not part of my calculation at all. So it's interesting. You don't believe in the wealth effects?

You don't think they're going to. I, I 100% believe in wealth effects. What's sort of interesting is that when you look at the change in equity prices for the very top, and then you try and then you apply assumptions about wealth effects at the very top, let's say, you know, 1% of wealth increases, 2% of wealth increases, consistent with the economic literature. You know, you get a, you get a meaningful amount of consumption. that you can sort of assume or impute to that. But it accounts for like a historically normalish share of consumption growth that we've seen over time. Now, the 1% or the 2% may be wrong, right?

Like the marginal propensity to consume out of wealth may be higher for some reason that it has at the top than in recent history. And so maybe the top are contributing more than they have previously. Maybe there's a worry that like AI, you know, driven wealth might is volatile and might decline. And so they're sort of getting their kicks in now. You know, that's, there's just a lot of uncertainty, right? Yeah, I do the arithmetic, even with one or two cents. I mean, the total market cap is up $40 trillion in four years.

That's $10 trillion a year, multiply by 0.01 or 0.0. That's a lot of growth. I mean, no? I mean, a lot of growth, but I guess I would say the rich always contribute a lot to PCE to,

you know, but it's a change in that matters, right? No, agree. Yeah. But yeah, so,

yeah, so the exercise that I ran, so I was looking at the, the, the Z1 data from the Fed, the financial accounts, yep. The financial accounts

and the distribution accounts, which I know well. Yeah, exactly. So combining those, And, and yeah, so when you do, when you make, when you do that exercise, not on a cumulative basis, but like on a year on year basis. Um, you know, like you get, you know, I think you get the top 10% contributing between 20 to 30% of nominal PCE growth every year. Um, you know, pretty consistently. And obviously it's very volatile. Um, and that's about what we're getting now. I think we're a little bit higher than that now when you do that exercise, you know, maybe. 35% right now, but not remarkably, you know, extraordinarily higher than normal.

Interesting. Interesting. Well, I can't believe we're already at 45 minutes into this conversation. What the heck? I actually, I want to do my statistic, though, before we go on because I think it's a good, sorry, Chris. I don't have you had a good thing. And I'm not sure you guys will get this, but Ernie, this is in your ballpark. And the stat is 579,000.

579,000. This is for the month of July. Is this your labor market metric? No, it's related to your work on entrepreneurship. Is this business formation? Indeed, business formation in the month of July. Record high business formations. It looks a little weird.

I mean, a lot of the increase is, in retail, in restaurants, hospitality. It's not in AI-related stuff. I saw a little bit in professional business services, so maybe some of that's going on. But information services, I didn't see it. The one thing, and this is a good segue into your work. Here's another statistic.

Tell me, tell me what this is related for the month of July, $152,000. $152,000. So $570,000. There was 570,000, 79,000 applications to form a business in July. This is IRS data. What was the, what was the, what's the 152, 152K? Is that high propensity businesses?

It is indeed high propensity, high propensity. Yeah. And I want you to explain that. You did great, masterful. The word is masterful. That's like, that's like what, that's like getting wordal on the first strike. Yeah, there you go. Yeah.

And that's a big difference because that's not at a record high. And that's actually you started, it's a little on the soft side compared to where we've been since the pandemic. So, but, but anyway, this, this leads into work that you are doing at Stripe because of your, the ability to get insight on this based on your data about business formation, solo entrepreneurship. Do you want to, you want to start, you want to give us a sense of that, that work? Yeah. Yeah, sure. So maybe I'll start at the 30,000 foot level and then we can talk about the Stripe data. So we're seeing, as you alluded to, Mark, a second surge in business formation.

So the first surge was in 2020 with the pandemic, made a lot of sense. People were at home, didn't have a lot to do. Also, we had some policies that incentivized business formation like the PPP program at the time. And so you see almost this step function increase in business formation in 2020. But what's interesting is that you see it both in overall formations and do see a step-up increase in high propensity application. So businesses that are likely to hire workers. Then in 2024, so we have no more PPP at this point, not a pandemic. The paycheck protection program, right? The paycheck protection program, right.

Yeah, go ahead. That was going to say you, and so you applied for a loan that then if you kept workers was fully forgiven. And to, obviously, to qualify for that money, you had to be a business. So it did, it did incentivize business formation, some of which were fraudulent, right, as a way to access those funds, some of which were not fraudulent. It was just, it was an incentive for business formation. But that, you know, that paid out its last, I think, I think it was basically done by 2022. So that's not a factor now. We don't have a pandemic as far as I know, happening right now. There is more work from home. But we saw this surge again in 2024, but what makes this recent surge different is that we did not see an increase or an acceleration in those high

propensity businesses. This is all non-likely employer firms. So putting my CEA hat on, I would say, that's kind of sketchy. I wonder if this is real stuff going on. Well, the nice thing about Stripe is that Stripe has two ways of measuring this or looking into this. One is just overall business signups with Stripe directly on our platform. The other thing is that we have a special product called Stripe Atlas, which helps you incorporate in Delaware specifically, which is a cool signal because, like, if you incorporate in Delaware, you're probably a serious business, right? And you're, that's where you would incorporate if you plan on getting.

V.C. money, for example, in the future. Both of those metrics have accelerated at the same time that we see this acceleration in the overall data. We can also look at things like the amount of time it takes these companies to reach meaningful revenue thresholds, like $1 million or $5 million. And there, like the amount of time that the latest cohort of businesses, the 2025 cohort of businesses have taken to reach a million dollars in revenue is significantly faster than earlier pre-AI cohorts like 2023 or 2022. So that's not the story that you would expect if this were fraud or like just sort of potential businesses.

These are meaningful businesses. Just to reiterate what you're saying to make sure I have this right, you're saying, okay, we saw a surge in business formation during the pendent. A lot of that was related to the circumstances created by the pandemic, the gig economy, and then you ascribe a lot to the PPP because I put cash into people's pockets and they use that as an opportunity to go off and start a company. And we've been kind of elevated, if I look at the data, business application data, formation data.

It's been elevated all along, but more recently in the last year or two, it started to pick up again. But this time, it's not these high propensity formations. These are companies that don't plan to hire. These are what you're calling solopreneurship. These are single-person companies that are enabled or empowered by artificial intelligence. That they don't need to go out and hire people because they got AI, basically. That's right. Well, yeah.

So definitely on the solopreneurship side. So these are single founders seem to be driving the increase and seem to be driving the increase, not just in the United States, but France has very good business formation data as well that can hone in on solopreneurs. And in the public data in France, we've seen an acceleration in solopreneurship there as well. So it's interesting for a couple of reasons. One is the like AI connection, which you mentioned. Second is what it means for employment, since these are obviously like just a numbers game, right?

These are literally just solo founders, right? So they punch below their weight in terms of employment, maybe in the long run. On the AI side, in the strike on that, are you kind of saying, so this would be pure productivity gain, right? I mean, you're saying they're generating a question. Yeah. Like like like so one theory is this is like you said, this is pure productivity gain. And if you want to go even bigger, maybe this is deconstruction of the firm. Right. So right. These solo preeneers are almost like, you know, we just have more and more independent contractors performing services that in the past firms would have brought in half because the costs were lower. But that, you know, this is a very cozy and framework for for for how we think about. it. Like, it could be that coordination costs have gotten so low, right, that firms can now just basically outsource a lot of different services that normally they would hire in for and that

these solopreneurs are filling that, that demand. It could be that these solopreneurs actually don't end up looking that much different in the medium to long run than they have in the past, and they hire people eventually. I'm not, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I think that this sort of batch of solopreneurs are different than what we've seen before. And so I suspect that, like, there are going to be more of them, but they will end up being small than they have in the pets. They'll stay solopreneurs for them. So you are ascribing this phenomenon to, there's potentially many reasons, but you do think AI is a big part of the story here. Yeah, and so we can look at that a couple of ways. So in the Stripe Atlas data where the incorporation in the state of Delaware, we actually get, you know, we get their business plans, we get information on their products. We can identify AI native companies. That is, if you're a company that is selling a product that is directly related to AI and LLM, a piece of software that is AI powered. And, you know, roughly half of the surge that we've seen in Stripe Atlas incorporations have been

AI native. So that's like, that's like the most reliable chunk of the AI story. I think it's actually even broader than that. I don't think that this is purely a story about AI native companies. I think this is AI assisting everyone who is thinking about starting a business with the hurdles and the fixed costs in starting a business. So AI is helping you write your business plan, come up with, you know, like it's helping you with the product itself. You know, my wife runs a a clothing firm, and she will occasionally use AI to give her ideas on dress patterns, right, for what she's selling. AI can tell you which tax authorities, right, to register with.

That's not going to show up as AI natives in our Stripe Atlas data, but it does, right, but that is still like a phenomenon that is, I think, helping drive business formation. It's almost like AI is becoming a co-founder in a lot of ways versus a, you know, I use it mainly as a research assistant, but if you're thinking about starting a new business and your risk averse, you might be more willing to go out there and start a business now that you know that you at least have a semi-reliable source of answers on some of the questions that you have. Have you, I know the data is available across industry and I think you have some cool scatter plots, you know, kind of looking adoption versus AI across the industry. I'm sorry. And also there's regional data as well. Have you looked into that? And is that providing any insight into what's going on?

So this was really interesting. Yeah, we just had a piece that came out this week on the sort of the geographical implications of the recent surge in business formation. And so here was my prior going into this. I didn't do this work, but I had a prior going into it, right? Right. Right. Like most economists, we have pirated about everything, right? Yeah, that's right. Look, it's our job.

It's our job. The, you know, the surge in business formation and the pandemic was very dispersed, and you saw it sort of leave central cities and go out into, you know, the outer rings in the suburbs. Makes perfect sense, given work from home. I would have thought that in the age of AI, human relationships would have become more valuable, agglomeration effects would have become more valuable. And you would have seen some of that post-pandem, that pandemic pattern reverse itself and you'd see more concentration of new businesses in city centers. That's not what we're seeing. In fact, we're seeing to the, and I think a lot of

this is new and uncertain and fresh, but to the extent that we've seen it in the last year, it's the opposite. Like we've actually seen more sort of disbursement out to the periphery, you know, in suburbs and away from central cities, which, yeah, which I think is fascinating. That is counterintuitive. I wonder what is that all about? Is that cost maybe? It's just, I can live in a, I can get a much nicer home, bigger home out in the excerpts and more rural areas and I still can do my work. You know, any thought as to what's going on there? possible. So take the framework that I mentioned earlier of AI as a co-founder and think about it from a risk aversion framework. If you lowered risk aversion for potential entrepreneurs, where is the

marginal business going to be? And so maybe the conclusion is, well, the marginal business, when you apply these lower, you know, risk costs, lower coordination costs, maybe they're out in the suburbs. Maybe it's, you know, second, you know, second earners in households, people that have not already sort of signaled that they or have in the past formed businesses, right, and therefore moved closer to city centers. These are the people that have always avoided that in the future. And so just from an entrepreneurship and then like a productivity standpoint, it's fascinating with the implications of that are because it's very different, I think, than prior surges. business formation. Hey, we're running out of time, Ernie, but I wanted to talk to you about productivity and AI more broadly, because I know you've been doing a lot of thinking around this. It hasn't shown up, or maybe you have a different perspective.

For my kind of perch, it really hasn't shown up in the macroeconomic data yet. Now, we've got measurement problems, no doubt, and maybe once we get all the data in 10 years from now, we'll see that there was some productivity increases in a macro sense. But if we take the data as at face value at the current point in time, so far it feels like productivity growth has improved, no doubt, in the last few years. But in my kind of perspective, it's just a normalization. It's kind of back to where it was on average since the beginning of time, you know, getting 2%-ish kind of non-farm business productivity growth. Nothing more than that.

So it doesn't give you the sense that AI's business as usual kind of productivity gains. It's not giving you any extra juice over what we've seen historically. Do you concur with that perspective and, you know, or are you seeing something different in the data that you're looking at? I don't entirely agree with that. So I think if you look at, okay. Yeah, you're right. So if you look at non-farm labor productivity, you know, I would, I think of the trend in that is 1.5% year and year. and we've moved to more like two to two and a half percent, I think, over the last few years.

So look, not a surge that is anywhere near what we saw in the dot-com boom, at least in the height of the dot-com boom. But that's a meaningful shift up, and it's been, you know, durable-ish for two to three years. We always have, look, you're right, we have to give a large grain of salt to productivity data. but we're at the point now where we've seen it sort of around that level for two or three years. I think that there's at least a short run sort of change in story happening. The mystery comes in when you look at

total factor productivity data, which you have to give even more grains of salt to. But like, look, a lot of very smart people have modeled this and think about it. And so like if the surge, if the, I won't say surge, If the step up in productivity recently were a micro productivity story from AI, you're using a model. You are like whatever, 20% more productive coding or writing, whatever.

You would expect to see commensurate surges in total factor productivity as well, because that's what economists think as sort of the pure technological. Do you want to say one sidebar just to quickly explain TFP? Go on. There's a whole other podcast. Yeah, no. So, but briefly, TFP tries to, TFP shows you the portion of productivity that remains after you filter out increases in just capital and labor. Yeah. So anything that remains in theory should be like pure technology or efficiency, right? That hasn't been strong. That's actually been weak. Yeah. The San Francisco Fed's measure of TFP, utilization adjusted TFP, is basically zero year over year. BLS has another measure that's not quite that bad, but it's still not extraordinary.

You know, it's at 0.7 percent, I think, year on year. So we haven't been seeing surges there. So it looks more like what's happening now is a utilization story that like because of, it is an AI story, but it's not a micro-productivity story. Because of all the demand for CAPX and AI, what firms are doing is they are running their existing capital hotter as one of the strategies for meeting that demand. So the more GPU cycles, you are using your existing factories more to create more chips and semiconductors.

You're pushing the bounds of your own capital further. And that's a real thing. That's an actual thing that can affect productivity. It's just different from like the LLMs, you know, making you 50% more productive and then aggregating that up to. overall product. That's interesting. And calling out TFP and TFPP utilization adjusted, something that came on our radar screen just in the last couple of podcasts. So you're saying it's still an AI story, but it's a story about AI supporting increased or more intensive utilization of the existing capital stock. Right. Can you make that intuitive? How do I get my mind around

that. Yeah, sure. So, so, so think of it this way. The surging CAPX that we've seen, that doesn't, that doesn't immediately or necessarily affect productivity. It has to, like, you have to put it in place and workers have to make use of it before it. It leads to productivity. But like a, but a correlated thing, right, is if you're a firm and you're spending a lot on a new data center because there's so much demand for AI. you're probably also like feeling the demand pinch and you're doing other things that are going to try to increase capacity as well, one of which is more capital utilization, making better or more use of your existing capital at the same time. And that does affect mechanically measured productivity.

And so what the San Francisco Fed decomposition suggests is that a lot of the surge over the last two years has been actually a rise in utilization rather than a rise in underlying TIA. And I will say as an advertisement, you've got a great substack. So I highly recommend. And I was looking at that prior to the conversation. You have a really cool chart with that decomposition that really makes it clear. It's been a few months, but you mentioned, or I've seen you write about the change in construction workers going from, say, office space to high value data centers and that showing up, you know, as a residual increase in productivity. Is that something you're still believe in? Is that still happening since you, I think it was probably Q1, maybe was that story?

Yeah. And you were definitely, yeah. Yeah. So, so yes, I think that's another channel by. So that's an indirect channel by which CAP-X could increase productivity, right? It's like the opportunity cost or the opportunity, you know, the opportunity potential of moving labor from something that's a little bit lower value to something that's higher value. And again, that's real.

Like, none of this is to say that like these are blind spots or these are fake things that we should ignore. They are real. The point I'm making is that it's not, it's not the productivity story that a lot of observers have in mind or are expected. Yeah. Hey, I'm going to pin you down just a little bit. You can dodge if you want, but I'll try.

And I did this with David. We had David Arter on a couple weeks ago. I don't know what the question is, but I'm sure he'll have answered it much better than I. No, no, no. Well, he answered it. He actually obliged and took a crack at it. And this is a straightforward question because it's all about me, Ernie,

know, what I got to do. And I got to put the paper. I got to actually have an explicit forecast for TFP, non-farm business productivity, GDP and everything else, you know, out at infinitum for 100 years. You know, so let's take the next 10 years because there's a lot of folks out there that have taken a crack at estimating or forecasting what they think the juice is going to, what I call the juice is going to be from AI. And, you know, if you look at the range of projections from economists, not the technologists who were, you know, 30 percent productivity growth over the next year kind of dystopic view.

But the economists, economists, the range from zero, you know, basically to, let's say, 1%. So you got Asimoglu, you know, down here with basically nothing, maybe a tenth of a percent per annum over the next. This is TFP, to the whole factor, pretty. Got it. To say, Goldman Sachs on the other end, where. you know, I'm doing a TFP adjustment. They come in around 1% per annum extra over the next 10 years. We're sitting at 3 tenths of a percent per annum over the next 10 years.

Where would you land in that spectrum? Or maybe you're one of those AI crazy technologists. So I will oblige with the caveat, it doesn't need to be said that this is like highly uncertain and basically science fiction in a lot of ways. Right. Right. You know, I would probably, I would, so on average over the next 10 years, I would probably come in close to you all at like a quarter percentage point. But I would have a very, I'd have a very variable sort of time pattern around that. So I, I think that the main thing holding back AI models from affecting TFP right now is they do inner ring work like task related work very well. right now. But there's this human bottleneck, this sort of management friction that I think is

preventing those models from converting into revenues, converting into output in all cases. So I think that will be the focus of the next phase of AI. And so I think that like, I think that we'll get to something more like, you know, between 0.5 and 1 percentage points to annual TFP, maybe three or four years from now as we sort of solve that. And then I think it's going to look a lot like the dot-com booth, right, where you get, you might have some years where you get even above one percentage point of GDP to TFP. And then I think it will probably have played itself out by the end of that decade. And so it'll be, you know, the effect will be much lower then. And so you average it out. And you get something between a quarter and a half percentage.

Well, I think that's more science than fiction, I'm telling you, because that's our forecast. So that's exactly our forecast, which means we're both going to be wrong. It's right. Both be wrong for sure. Because we've said it, now we have to bet against it. Yeah, we're doomed. We're doomed. Hey, Ernie, we kept you long enough.

Really do appreciate the conversation, learned a lot. I want to thank you for all the hard work you're doing with the Stripe data. That's, you know, very valuable kind of contribution to the. conversation for sure. And I'm glad you're on the case because no one better than you to kind of make sense of all that data that you're getting. I appreciate that. Thank you. And best of luck with the company. It sounds like you got a lot of cool stuff going on there. And, you know, I think you joined a great company. So congratulations with that. Guys, anything else before we call it a podcast? Matt, Chris, anything? Any part of words? Okay. All right. Well, with that, dear listener,

I hope you enjoyed the conversation. We will talk to you next week. Take care now.