Basil Halperin on Macroeconomic Policy in an Age of Transformative AI

Basil Halperin is an assistant professor of economics at the University of Virginia. In Basil’s first appearance on the show he discusses the famous but flawed Citrini essay, why Silicon Valley’s growth expectations aren’t showing up yet in interest rates, the impact of Less Than Zero by George Selgin, what the true frictions in the economy are, the differences between Calvo and menu-cost pricing, the impact of transformational AI on emerging economies and the housing market, and much more. 

Subscribe to David's new Substack: Macroeconomic Policy Nexus

Read the full episode transcript:

This episode was recorded on March 31st, 2026

Note: While transcripts are lightly edited, they are not rigorously proofed for accuracy. If you notice an error, please reach out to [email protected]. 

David Beckworth: Welcome to Macro Musings, where each week we pull back the curtain and take a closer look at the most important macroeconomic issues of the past, present, and future. I am your host, David Beckworth, a senior research fellow with the Mercatus Center at George Mason University, and I’m glad you decided to join us.

Our guest today is Basil Halperin. Basil is an assistant professor of economics at the University of Virginia, where he works on the important and super interesting intersection of macroeconomics and AI. He joins us today to help us make sense of where the economy is going and what AI might mean for macroeconomics. Basil, welcome to the program.

Basil Halperin: Thanks, David. I’ve been listening since day one, so excited to be here.

Beckworth: Yes, we go way back, and I was trying to think where we first connected. I know it was somewhere in the blogosphere. Scott Sumner’s in the picture there somewhere, too. What is your recollection of where our paths first crossed?

Halperin: I think I do have a precise memory of that. I was a longtime reader of your blog and the econ blogosphere more generally. That’s how I got sucked into econ. At some point, I left a comment on one of your posts, and somehow we started talking from that or something, and on and off over the years since then.

Beckworth: It’s great to have you on. I’ve had other guests who’ve come on the podcast who also were readers of the blog way back when. Sometimes I’m embarrassed to go back and look at some of the things I wrote way back when. It was a fun time. Back during those days, we were trying to figure out what was going on. Why was the economy having such a slow recovery? There are all these young, budding economists like yourself. Now, you went to the University of Chicago as an undergrad. You’re an MIT macroeconomist, so a great tradition there. In fact, who was your chair? Was it Iván Werning?

Halperin: Marios Angeletos and Iván Werning were my chairs at MIT, yes.

Beckworth: Some great names there. You’ve done some interesting work, we’ll talk about today, on menu costs and thinking about what’s the best way for monetary policy to respond to shocks. I think this is very important, given that, today, we are in a world based by supply shocks, once again, big ones from the war in the Middle East. We want to get to your paper, but you have also written extensively about AI.

Global Intelligence Crisis

In fact, I had one of your co-authors on the show previously, Zachary Mazlish. We talked about a paper we’ll revisit today—very, I think, timely and an important one. I want to begin our conversation by retelling a story that we all witnessed, I believe, in February. It was an essay that was written that went really viral. In fact, I think all of us would love to have an essay that goes this viral. It was an essay by Citrini, and it was called Macro Memo from June 2028. If they were in the future, they were looking back. They spelled out in great detail this dystopian future. Very much one where there’s anemic demand, no demand at all. What was amazing, Basil, was that it had a huge impact. It was widely read. 

Not only was it widely read, but markets themselves allegedly responded to it. The stock market took a dive. This is what some say, at least. The timing lines up. This very dystopian AI essay. Just to recap—and I know many listeners remember this essay well because it made such news, such headlines—the general argument that it made was that companies would replace white-collar workers, not just blue-collar, but white-collar workers, software engineers, analysts, middle managers, with AI for massive productivity gains and cost savings.

This would lead to layoffs and, in turn, reduce consumer spending, weakening demand. Then firms would double down with further AI investment, accelerating the cycle. The result would be a ghost GDP is what the term they use, where you’d have output going up, GDP going up, but labor income almost completely vanishing. A massive shift from labor income into capital income. Very dystopian world.

I think it resonated because they provided a detailed story, even if the economics wasn’t all there. I will contend, and we have some people we know who wrote some follow-up pieces. In terms of a model, the story doesn’t hold together all the way through. We’ll talk about that. It really resonated, and that was so fascinating. How does an article go viral? They were successful. Again, allegedly, the market responded to it. What was your initial response to that article when it came out?

Halperin: Maybe two takeaways from that episode. One is it just is a remarkable showing of how forward-looking financial markets are. Financial markets are not just looking at what’s happening right now and predicting the future based on what we see right now. They’re trying to predict two, five, 10, 20 years out. I think the duration of the stock market is 10 or 20 years. The time-weighted average payment of stocks is really long-term thinking. That’s one takeaway.

The other takeaway, I think, as you hinted at, is that doing macroeconomics in words is hard. I’m not sure that anyone, maybe Milton Friedman, can do macroeconomic intuition in their head in general equilibrium. You need math to really discipline you to ensure that your stories add up, that things come together. This story that we would run out of spending in the economy because no one would be earning labor income. I wasn’t convinced that held together. The income has to go somewhere. Someone has to be spending what’s produced. Income equals output.

Beckworth: Yes, absolutely. Yes, and there was an individual named Alex Imas, who had a great piece that responded to it. He had a Substack. We’ll provide a link to it. He actually took the argument through a model. He took it through several models. If I recall correctly, he took it through three different models. One model is where you have a big income shift from labor to capital, and then the follow-up is, well, labor has a higher marginal propensity to consume than the owners of capital. That was one way you could create a world where demand collapses or it’s eviscerated. 

Then there’s a secular stagnation model. He told that story. Then he just told another story, overlapping generations model story, I believe. Because of this loss of income, savings goes down, and then the capital stock itself shrinks. He said, these are all really extreme stories, some big assumptions that don’t hold up. If anything, history tells us otherwise. Institutions adapt. Policy adapts. It’s just so unrealistic.

Now, I bring all this up because it is timely. It’s topical. We’re in a world of AI. You’ve also written yourself extensively about this. We’re going to talk about a paper in a minute where we get into what will happen to interest rates, and what would be signs that this is actually upon us? In fact, let me just stop there for a minute. You would argue, based on this paper that you wrote with Zachary and Trevor, what was the third co-author?

Halperin: Trevor Chow.

Transformative AI and Interest Rates

Beckworth: Trevor Chow. The paper is “Transformative AI, Existential Risk, and Real Interest Rates.” You would argue we probably aren’t there yet based on what’s happening to interest rates. Is that fair?

Halperin: Yes, it’s worth saying a few things about that. One is the particular scenario under consideration is this idea of transformative AI that is a very radical scenario, really, on the league of the move from pre-Industrial Revolution to post-Industrial Revolution. A 10 times speed-up in GDP growth is the benchmark that we have in mind is very rough, something like 30% GDP growth. Crazy, crazy numbers. The whole economy moving as fast as Moore’s law, more or less.

What would we see in financial markets if financial markets expected to have 30% GDP growth anytime in the next 20 or 30 years? Well, asset prices would do a lot. One very unambiguous, to my mind, asset market reaction would be that real interest rates would go up a lot. Interest rates would go up a lot. To frame that, I think we can connect this to the monetary policy world that both you and I grew up in, so to speak.

We are central bankers, of course, very interested in tracking r-star, the natural rate of interest. What determines r-star? Many things, but one very prominent determinant, in much thinking, the baseline first factor that we think of, the factor that gets us on Wikipedia page and the Ramsey rule, is that higher growth causes higher interest rates. Interest rates today are somewhat elevated compared to where they were certainly during the depths of the COVID recession, when they were deeply negative.

The 30-year real interest rate I was just looking at this morning is 2.75%, which is admittedly the highest it’s been in 20 or 25 years, but not out of the range of historical norms. Not anything like what you would see if markets were really believing what the people in San Francisco believe in terms of how soon we’re going to have really transformative changes. 

It is worth noting, folks in San Francisco and so forth are really moving tens of billions of dollars, moving their own human capital, building organizations on these beliefs that we’re going to have something like another Industrial Revolution, and it’s going to happen soon. Financial markets don’t seem to be predicting that with any strong degree of confidence, at least. I can go into more intuition on why higher growth pushes up interest rates, but that’s the story. We want to look at interest rates and see, are markets predicting higher growth or not?

Beckworth: Your paper and you yourself and your co-authors right now would say, “We’re not there yet. We haven’t hit transformative AI. There’s rapid growth.” At least the financial markets are saying that. Then, on the other hand, you have all these investors in Silicon Valley who see things differently. At some point, the realities have to clash. At some point, they’re going to go bankrupt, or we do take off.

At some point, there’s going to be a point in the road where reality does constrain what people can do with these investments. What do you see happening, I guess? We’ll come back to your paper, but what’s going to be the choke point ahead? Is it going to be investors waking up and saying, “Oh, wow, I just wasted a bunch of money,” or is it going to be AI does become transformative?

Halperin: I’m not sure that, actually, there needs to be a choke point, per se. I think it’s totally plausible that we just get something like the late ’90s, the dot-com boom, where we have a large surge in growth. AI companies make a lot of money and continue to have really fast revenue growth, like the 10 times a year that Anthropic has had last year. Not reaching the 30% GDP growth, certainly in the next five years, again, that some people take very seriously.

Maybe 20 years down the line, things change. I do think that rapid acceleration really is possible in the long run if we have something closer to full automation of human labor. Those sci-fi scenarios, I actually do think our models say that’s quite possible. Economic history says rapid economic speed-ups are quite possible. In the next five years, the next 10 years even, maybe even the next 20 years, that’s where the markets are not seeing this. Markets are pretty good at predicting the future, its own form of artificial intelligence.

Beckworth: You’re hopeful that, within my lifetime, I will see transformative AI?

Halperin: “Hopeful” is a strong word because big changes are both good and scary. Yes, I do want to say hopeful, but optimistic that we get big biomedical progress, all sorts of new consumer goods we can enjoy, et cetera.

Beckworth: I say “hopeful” because if this outcome that was outlined in this earlier paper, very dystopian, well, maybe I wouldn’t want that. We’re going to go back and revisit that because I want to flash out why we think that’s wrong and why it would not happen. I guess I’m thinking more of like, yes, I live to 120. I have a high quality of life. I see my great-grandchildren live, and I’m not living hand-to-mouth. This awesome world may happen in my lifetime, maybe it doesn’t, but we can at least think about it.

That’s what we’re going to do today. We’re going to think about what that would mean. Before we do that, let’s, again, go back to this earlier claim that, should AI really take off, transformative AI take off, we would see these scenarios, which, again, briefly, Alex Imas, we’ll link to his paper as well. Your sense is that these claims are missing something. Maybe walk us through why there would be sufficient nominal income growth or demand growth if we did have this rapid growth. What are people missing? What are these people in Silicon Valley who write these dystopian essays, what do they get wrong?

Halperin: Okay, there’s a lot of meat in there. To start with, why would interest rates go up in this world? Again, if I talk to my friends in San Francisco, and that’s really how all of us started getting in debates with these people, they say one of two things is going to happen due to AI. It’s all going to happen very soon. Number one, we’re going to have really fast economic growth. We’re going to live in untold prosperity, or AI is going to quite literally kill us all and the human species.

Interest rates are interesting in that light because both of those scenarios push interest rates up. That’s in contrast to things like stocks, where, if we’re all dead, stocks are not very valuable if we’re dead. Extinction risk pushes stocks down. Higher growth might push stocks up. It also might not. We can come back to that. 

Sticking with interest rates, why does both higher growth and higher death risk push up interest rates? Well, it’s about this consumption smoothing idea, this classic idea, the thing that won Milton Friedman his Nobel Prize. If you expect to be earning a lot more money next year than this year, perhaps because you’re going to switch jobs, perhaps because we’re going to be in an AI-driven utopia, there’s a lot less reason to save this year. You’re going to be rich next year. No reason to save anyway.

Similarly, if you expect to be dead next year, and you don’t have a bequest motive, also no reason to save this year. You’re not going to be around next year to put those resources to use. That lower supply of savings you can think of as pushing up interest rates, just as in the same way that a lower supply of apples pushes up the price of apples. That’s really the core logic, this simple consumption smoothing channel. There’s lots of complications around that, but that’s the core idea.

Beckworth: That is the key mechanism through which this rapid growth is tied to interest rates today. It’s, again, what we think is going to happen in the future. Consumption smoothing. You can also think of this in terms of the return on capital, right?

Halperin: Yes.

Beckworth: That would be another way. It’s a complementary, similar story, right?

Halperin: Yes. Let me say a few things about that. The supply-side story, as in supply of savings, that’s like the macro 101 version, because in macro 101, it’s the supply that determines r-star. In macro 102, you might start thinking that demand also determines r-star because supply will not be perfectly elastic, to get into the technical weeds that I know your listeners love.

Higher demand will also push up interest rates. Meta, Mark Zuckerberg just issued the, to my understanding, largest investment-grade bond in history, like $27 billion or something. These firms are borrowing a lot of money to build giant data centers, giant computer clusters to train AI, to run AI. Yes, the higher demand for capital from OpenAI, from Microsoft, so on and so forth, that also probably is pushing up interest rates already today.

Beckworth: Okay, so the firms themselves, they see high profits, high returns on their investment. You know what? Let’s borrow. Let’s tap out as much debt as we can that makes sense and invest in it. In doing so, they’re pulling out those savings, which makes what remains more expensive, so rates go up. Or, alternatively, from the consumption side, we’re smoothing. They’re both operating in the same direction, and they both reinforce the same movement. Rapid growth from rapid productivity gains is going to make rates go higher.

Halperin: Exactly.

Beckworth: Sadly, we don’t have those high real rates yet that would suggest robust, rapid growth. We’re hopeful. As you said, we’re still waiting.

Halperin: Yes, I do want to emphasize that this is, to my mind, a simple gut check. 30% GDP growth, it’s just so high. 10%-plus GDP growth, it’s just so high. That would certainly push up interest rates a lot. If we had 5% GDP growth instead of 2% or 3% GDP growth that we have right now, 5% would be higher than the 4-ish we got up to in the late ’90s. 5% would be astounding.

Reading that out of interest rates, that would be hard. Again, long-term interest rates have gone up 3 percentage points or so, 4 percentage points, 3.5 maybe, since the depths of COVID. Interest rates have risen. It’s probably because of monetary policy, inflation-fighting things that many of your other guests have talked about, that you’ve spoken about. This is really ruling out that right-tail scenario on a short-time horizon.

Beckworth: All right. This is not where I want to go with the show, but I got to throw this out there since I have someone as smart as you on the program right now. In the limit, in the universe, we will all die a cold death, right? At some point, the universe, poof, it’s gone. All the usable energy is used up because of entropy. In that world, we have really high interest rates. In the limit, we’re all going to die. Is that what the model would predict?

Halperin: At the appropriate time horizon. The interest rates over the 10 to the 10 to the 10 to the 10 maturity, that would be infinity, yes.

Beckworth: All right. We got to be careful with our infinite horizon models. Don’t take them too literally, right? Because if we take them literally, then we should be backing into the present, the fact that we’re all going to die a cold death at some point in the future.

Halperin: Only for that infinitely far horizon.

Beckworth: Far horizon.

Halperin: The 30-year interest rate, that would be affected by that.

Beckworth: Okay, all right. That’s another interesting application of this theory. Again, it’s not very practical. What’s the good theory of the universe? We could look at interest rates to help us make sense of that as well. That’s not the point of where we want to go with this. We’re talking about AI. Again, you explained what interest rates tell us or don’t tell us at this point about whether we have transformative AI.

Let’s talk about, though, how does transformative AI—should it come? We’re in a place, let’s say, where we do see these higher real rates going up. It’s evident. We see rapid, say, dislocation. I think you would agree that if we do have transformative AI, there’ll be some transitions, some transition costs, some sectors would be very disrupted. There’ll be new growth here. There’ll be some old industries dying, creative destruction on steroids, right?

In that world, how do benefits get widely dispersed? How do we avoid the dystopian outcome? What are the different possible paths that we could take? I think part of this is, what is the policy response too? I believe in markets. You believe in markets. Markets respond. If we do lean towards this dystopian outcome, there’s going to be institutions that respond, policies that respond. Paint for us a picture, a tapestry here of what would be this better-case scenario.

Halperin: Basically, this seems like a really hard problem. It’s surprising that even with the rise in AI interests, no pun intended, that there’s still not more work on the development of what would the optimal redistribution policy look like. The economist’s theoretical ideal is you have some not-too-inefficient taxes from people who are gaining from the AI transition, and then you redistribute across everyone in the world to equalize marginal utilities. What is the practical way of implementing that?

Everyone likes to talk about universal basic income. Maybe, it’s not exactly clear what that should look like, is my read of the discussion. If the economy is growing as fast as people, models suggest that it could under advanced AI, we’ll have so much of a bonanza on our hands that one hopes that we can have the institutions to just distribute some chunk of all of this to everyone so that everyone can benefit. I don’t have a clearer picture than that, though. Maybe UBI wouldn’t be a crazy idea in the communist utopia, post-transformative AI. I don’t know.

Beckworth: It seems like, to me, fiscal policy would respond, if nothing else, because there’d be public unrest. There would be support for something to change. Again, to me, these far-fetched scenarios, if there’s massive structural unemployment, that’s the key argument being made here. Massive structural unemployment in a nutshell. If you get to that point, there’s going to be political movements. There’s going to be people in Congress calling for change, whatever that may be. Checks being sent in the mail, like you suggested.

Optimal Monetary Policy Under Menu Costs

I want to go in a different direction. I think there’s plenty of ways to get past this. Again, I believe in markets. I believe policymakers will respond. Humans will be humans. Institutions will adapt. Let’s talk about a paper of yours. I want to use this. I want to take your paper. It’s called “Optimal Monetary Policy Under Menu Costs,” you co-authored. I want to tie it into another little, short book that we’ve both read and are fans of. That’s Less Than Zero by George Selgin.

If I read him correctly, his response to my question would be, “David, the way that real benefits get dispersed across the economy is through lower prices.” He would assume, maybe he would hope for, to use a better word, stable nominal wages or stable nominal income. Maybe that’s where the policy maintains some growth of aggregate demand. Steady growth so that nominal wages do not fall, but real price level is going down. Real income, real wages, is how the benefits are widely dispersed. Maybe respond to that first before we jump into your paper. What do you think about George Selgin’s potential answer?

Halperin: Yes, I think there’s a few things worth disentangling there. One is on this book, George, of course, being your PhD adviser, if I’m not mistaken. This book has influenced me a lot. I think, to my very idiosyncratic taste, it’s one of the most conceptually important works on monetary economics in the last 30 or 40 years because it really drills in on what should central banks actually be doing from a number of directions that the formal literature eventually developed on, even if George’s work wasn’t mathematically formalized so much and wasn’t so much directly cited. He previewed works in top journals that were published 20 or 30 years later, including inspiring my paper that you mentioned with Daniele Caratelli. That’s one point, just to say that.

Beckworth: Read the book, listeners and watchers of the video. Get the book, Less Than Zero, by George Selgin.

Halperin: It’s short. It’s very readable. The second thing to say is that “Optimal Monetary Policy for the Singularity,” that’s a paper that I hope to have out this summer.

Beckworth: Oh, you do. Okay. Fantastic.

Halperin: I think this discussion is exactly relevant in that, exactly as the other paper with Zach and Trevor framed. If the real interest rate is going to rise, if r-star is going to rise, that means monetary policy needs to pay attention to that, to either not keep interest rates too low so that you have some inflationary outcome, or not have interest rates too high to have excess unemployment, like the Citrini scenario. That’s the second point.

A third point is that I think monetary policy can be helpful to not screw up how the economy goes during an AI-driven transition, but cannot solve the problem. This really is a question for fiscal policymakers. It is a question of redistribution. What monetary policy should be sure to do is just not screw things up. I keep saying Friedman in this episode, maybe appropriately because he’s the GOAT, as your boss, Tyler, would say.

What monetary policy can do versus what it can’t do: What it can do is ensure that it doesn’t screw up the economy in response to shocks. That’s the fundamental lesson of his 1968 presidential address, famous paper. Implementing the less-than-zero policy that Selgin recommends or this countercyclical inflation that comes out of my work would be one plausible way of doing that.

Beckworth: Do no harm is your advice to the Feds. Of course, I think we both agree one practical version of this would be something like a nominal GDP target or nominal income target in broad terms, but you have more specific nuance that you outline in this paper as well as, it sounds like, a nice paper coming out this summer. We’ll have to have you back on again when the other paper comes out. Let’s talk about this “Optimal Monetary Policy Under Menu Costs” you co-authored. Maybe give us the summary version of it, and then we can drill down into certain parts.

Halperin: Yes, so the question this paper tries to answer is, we have a lot of empirical evidence and a priori reason to think that prices in the world are sticky. In response to shocks, not all prices instantly update every second of the day. How should monetary policy work? What should monetary policy do under sticky prices? We have a textbook benchmark answer to that question in the form of the canonical new Keynesian model that we teach in our PhD programs, in our master’s courses, often using this Woodford textbook or this Gali textbook.

What pops out of that framework is that optimal monetary policy in response to efficient shocks is to stabilize inflation. Strict inflation targeting. This is an intellectual foundation for the real-world adoption of inflation-targeting policies that we’ve seen all over the world over the last 35 or so years. This baseline new Keynesian model is based on this canonical Calvo friction.

This Calvo friction, the way it models price stickiness, is it thinks of firm owners wake up every day, and they have some random chance, some exogenous chance that they’re not allowed to change their prices. This is mathematically convenient, analytically beautiful, but I think almost is prima facie unsatisfying. It’s not explaining why prices are sticky. It’s assuming that prices are sticky. This is the first theoretical critique. 

There’s a related empirical critique that this Calvo friction doesn’t match key features of the data. In particular, the state dependence that we see in pricing. This is things like during COVID, we had this huge shock. Prices became more likely to adjust. Prices were less sticky during COVID than they were in response to other shocks because the COVID shock was so big, the state dependence.

Where did we come in, in our paper? We instead ground the optimal policy analysis under a different kind of price stickiness, the natural alternative, which is menu costs, where firms can change their prices at any time they want if they’re willing to pay a fixed cost of adjusting their prices, which are known as menu costs. Where this term comes from, if restaurants want to update their prices, at least historically, they needed to print new menus to update their prices. That has some costs to restaurants, so they might not instantly update their prices anytime chicken becomes more scarce, because doing so had some costs. 

You can really think of this as just a device for capturing any fixed cost of updating prices. That can be Walmart pays workers hundreds of thousands of dollars each year to walk around its many, many stores to update price tags on store shelves, or you’re a small business owner, and you need to sit down and think like, “What price should I set this year?” That’s opportunity cost of time.

That’s where we come in. We find that rather than inflation targeting being optimal, in fact, stabilizing inflation in response to these kinds of shocks is not optimal. Instead, you get something more like NGDP targeting. That’s the big picture. I can then explain why NGDP targeting is optimal.

Beckworth: Let’s go back to these two root assumptions. How do prices adjust? What are the costs to adjusting them? This Calvo story, there’s been some derision they call it the Calvo fairy because it doesn’t really exist. It was a useful tool. It was an analytical tool that helped. Then menu cost. Let me ask you this since you’re fresh out of grad school, you’re in the trenches doing research. What is the consensus today? What is a better measure of what actually happens out there? Is it menu costs or is it more of a Calvo-type price change?

Halperin: Oh. This is still a controversial question, I think. My read is that the most important stylized facts really do fit the menu cost model remarkably well. Things like Alberto Cavallo has this amazing work where he goes and scrapes prices from Amazon, Instacart, or whatever, all over the web. He has this amazing The Billion Prices Project, pricing data.

If you look at those price changes, the menu cost model just fits that data shockingly well. Things like the menu cost model predicts that in response to small shocks, firms shouldn’t be willing to change their prices. If the gain from changing your price is $0.01, but the cost of updating the price is $1, you’re not going to do it. Certainly, if it’s a big shock, that’s going to outweigh that $1 cost of changing prices. He finds these beautiful figures. He generates these beautiful figures where you see exactly that. As mentioned, the state dependence around COVID, that bigger shocks, you see the frequency of price changes happening more often.

That said, the literature has converged in the sense of finding that the time-dependent pricing of Calvo and the state-dependent pricing of menu costs can often be very similar, especially to a first-order approximation. That’s relevant for positive implications for studying what the economy does do, but for thinking about what should policymakers do, you need to start thinking about second-order approximations to the world. This might ring some bells for folks who are familiar with welfare loss functions and things like that. Once you start thinking about second-order approximations, then there are these key differences between time-dependent pricing and state-dependent pricing. That’s what motivated me to write this paper.

Beckworth: Just to be clear, the Calvo pricing approach, it’s time-dependent. Prices only change based on time. I guess my question is this, is there a good economic story for that, other than convenience analytically? Is that something that actually people argue that firms wait? They’re just, “Hey, I don’t want to do it today. I’ll do it tomorrow.” Is that an actual argument?

Halperin: I can give you the strongest argument for that, which is that if there’s observation costs, and those observation costs are some fixed cost, then maybe you only want to Google to see what your competitor’s prices are every six months. That can result in this time dependence.

Beckworth: That’s a reasonable argument. To go from there, I guess, to it’s so important that it’s going to be the one friction that drives the business cycle, that to me is, I don’t know, a little bit of a stretch in my mind. It’s an important story, but is it the story would be my question.

Halperin: Even zooming out, is sticky prices obviously the most important nominal friction in the world? I’ve written a paper on this. That’s still not clear to me. Sticky prices as opposed to sticky wages or sticky nominal debt contracts or sticky information. No one’s done a head-to-head comparison of all these different frictions. I think that’s desperately needed. Though, it’s not clear how you do that. Otherwise, I would have written the paper.

Beckworth: Just, again, to flesh this out. You have the menu cost story. You’ve got the Calvo story. In the menu cost story, is it more focused on just a few firms? The firm that actually receives a shock, it has to be big enough and costly. The benefit has to exceed the cost for them. You would imagine the firms themselves that are affected would do it. As opposed to Calvo, it’s like any firm at some point in time is going to have a price change.

Halperin: I think maybe that’s a good chance for me to give you the logic behind the core mechanism in our paper, which also really is the core logic in the two pages in George’s book where he talks about this, that we expand on and elaborate on. The core logic is, say, you have Davidville and there’s 100 firms in town, and firm number 1 becomes more productive. We know that, for microeconomic reasons, if the firm’s more productive, it can produce things more cheaply. It should be lowering its relative price. 

David, you’re the chair of the Central Bank of Davidville, if you’re aiming for a stable price level, stable inflation, then to simultaneously have that one firm lower its relative price and the average level of prices, the price level, in the economy to be stable, you need simultaneously that one shocked firm to lower its nominal price and the other 99 firms to raise their nominal price. That gets you that relative price drop and that the average level of prices is constant.

That means that every one of those 100 firms is forced to pay this wasteful menu cost of adjustment. Every firm has to pay workers to walk around the store changing price tags, or every firm has to pay attention to pricing as opposed to the natural alternative of just that one firm that was affected by the shock cutting its price, everyone else leaving their prices unchanged, then you get that correct fall in relative prices, but only that one firm has to pay that wasteful menu cost.

That’s the core logic of what we show to be optimal in the model. The way that gets you NGDP targeting or nominal wage targeting or this countercyclical inflation is that if this firm was having a positive productivity shock, it’s producing more stuff. Output’s going up. Y is going up. It’s cutting its price. Every other firm is keeping prices constant, but that one firm is lowering its prices. The average level of prices in the economy is then falling. Y up, P down. In a baseline setup, those are one for one. P times Y, nominal GDP, is kept constant. You can also see this as nominal wage targeting. Stabilizing nominal wages ensures that the nominal costs of all those unaffected firms is stabilized so that they don’t want to change their prices. Nominal wage targeting is another way we frame optimal policy in the paper. All of these terms—NGDP targeting, nominal income targeting, nominal wage targeting—all of these things are pointing to this countercyclical inflation, looking through shocks rather than aiming for strict inflation stabilization, which is the baseline new Keynesian logic.

Beckworth: The argument you’re making, if I understand correctly, is that it’s more efficient for just the one or two or three firms that have been hit by the shock to make the price adjustments than to force price changes on everybody. Maybe each firm has a smaller price change than the one or two or three firms that have been hit, but it’s not efficient in the aggregate because everyone is changing for no good reason outside of those firms that were hit. Let the firms that are hit do the price adjustments. Keep it to them. Yes, maybe the aggregate price level is affected, but the relative prices are where they’re supposed to be because of the productivity changes.

Halperin: That’s exactly it.

Beckworth: Let me go back to your point about what are the most important rigidities or frictions in the economy. We raised the question if output prices truly are—because, like you mentioned, the state contingents during COVID prices rapidly changed. There’s sticky wages, and that seems to make a lot of sense to me and maybe people who suffer from money illusion. I remember working in academia. I’m no longer there, but I remember there were many years they wouldn’t give a pay raise. People, they lived with it. They weren’t happy, but effectively, you’re getting a real wage cut because inflation was going up. If you actually did a nominal wage cut, man, the faculty would be up in arms.

There really is no real difference between those two approaches. One, you do it through inflation, through money illusion. The other, you don’t. I think nominal sticky wages is an important one. There’s also this literature on sticky information. People don’t update. Again, this is maybe state contingent. When inflation is taking off, maybe you pay more attention.

Also, a sticky debt contract. Jim Bullard is another advocate of nominal GDP targeting for a different reason. He thinks debt contracts in nominal terms don’t adjust, and that creates problems for market clearing as well. Are there any other ones I’m leaving out, any other ones you would add to that list?

Halperin: Those are the most important ones. I’ll make two comments. One is, as you may even be aware, all of those frictions actually, which to my mind are the most important classes of nominal frictions, all of those point to something like NGDP targeting being optimal rather than inflation targeting besides price stickiness, which historically we would think of Calvo as saying inflation targeting is optimal. That’s why I was excited to work on my paper because it shows that once you think of price stickiness as being due to menu costs rather than that Calvo friction, you get that NGDP targeting-esque result, that countercyclical inflation result.

A lot of work across sticky wages, information frictions, nominal debt contracting, and now sticky prices showing this countercyclical inflation result as being optimal. I can also preview yet another work in progress—too much to do—on bounded rationality, also finding a similar flavor of results. Again, I’ll say to George Selgin’s credit, a lot of this, all of these results besides the sticky wage and the bounded rationality results were previewed in Less Than Zero. Again, I really think that that was a very important conceptual work.

Beckworth: Classic. We have all these frictions. They all point to something like nominal GDP targeting. To be clear, I advocate that, and I think you would, too, simply on practical grounds. It may not be the perfect version. Your model might have something a little bit different. I’d love to go back to the example of Michael Woodford and Gauti Eggertsson. They would call for an output gap adjusted price level target. That’s a mouthful. That’s effectively a nominal GDP level target. They said, yes, we will settle.

In fact, Michael Woodford had an essay at Jackson Hole, famous one, 2011, where he advocated something like a nominal GDP level target, even though on paper, he wanted an output gap adjusted price level target. He said nominal GDP level targeting is pretty darn close, and it’s a whole lot easier to think about than having some abstract modeled version.

Halperin: The way I think about this is that NGDP targeting or something like it is eclectically optimal. It’s from an eclectic—

Beckworth: Eclectically optimal.

Halperin: —set of perspectives. It’s optimal, which, to my theorist brain who spends all day pushing Greek letters around a whiteboard, that’s not totally satisfying because all of these frictions point to slightly different optimal policies. If Jay Powell or Christine Lagarde wants to come to research economists during a framework review or strategic review and ask, what is the academic baseline optimal policy, I do think that the literature in the last 20 years or so has come to this perspective that across this eclectic range of perspectives, nominal income targeting is what’s pointed to being optimal.

Beckworth: Let’s take these insights and apply them to the current context of the oil shocks that we’re seeing right now. We have a Middle East war, and who knows how bad it’s going to get. I was just reading about how Iran bombed Qatar’s liquid natural gas facility, which apparently it’s very complicated, hard to build. It’s not only oil price shocks, but natural gas shocks, and these things probably won’t come down for a while. How does this play into your research in this broader discussion about these rigidities and nominal GDP targeting?

Halperin: Yes, interesting question. If you wanted to take the model in my work seriously, you would say, “Oh, this oil shock is one shock to a particular relative price, and then this exact logic from Davidville of one firm out of 100 goes through, except in this case, it’s a case where prices should be allowed to rise, and GDP should be allowed to fall due to the negative oil shock.”

I actually think that oil shocks are unique and should be studied separately in that energy prices are very flexible, so there’s not so much menu costs on changing a lot of energy-related prices because these things fluctuate so much. Instead, there’s alternative perspectives like Aoki 2003 is this great paper that’s known but still underrated, I think, where he points out that if you have one good in the economy, like energy, that has flexible prices while everything else in the world is sticky, then you should just, as a monetary policymaker, forget about that flexible good. It’s not going to screw anything up because it’s flexible. This is a justification for paying attention to core inflation rather than overall inflation level. This is another argument for looking through shocks, looking through energy shocks in particular, but from a different perspective. That’s how I think about energy shock issues.

Beckworth: I’ve heard some folks talk about core nominal GDP targeting, so maybe narrowing because sometimes exports, imports in a small open economy can affect that measure as well, so maybe consumption spending, something like that. That is very interesting point, if the prices are flexible.

Which leads me to another question. Basil, you could imagine a world where prices become more flexible. In fact, we’re seeing this. We’re seeing more and more. You work with Uber. In fact, you probably have seen real-time price adjustments and stuff. Sticky output prices may become less and less of a phenomenon. I don’t know what you think about that. Do you see a world where we could get past many of these rigidities, or will they always be here?

You mentioned the behavioral ones, the cognitive ones. Maybe the information would always be with us. I guess my question is, given technology, given AI, heck, let’s go back to AI, do we get past some of these rigidities going forward, or will some of them always be with us, and therefore, we need to have a view of monetary policy like you outlined in the paper?

Halperin: I love this question. My short answer is I would love to read someone digging into this, but my extended speculation is that, number one, you could imagine, is there a market failure here where should the government be trying to encourage firms to adopt technologies that make it easier to change prices, things like those electronic shelf labels that Walmart will now have on its store shelves to change prices more flexibly, and things like that? If this nominal rigidity is a market failure, what’s the optimal policy there? Would be super interested in that. What are the macroeconomic implications of adopting these electronic shelf labels? I thought about partnering with some firms to study that.

In terms of the sci-fi scenarios for in the long-run future, how could things go? That’s somewhere where I don’t feel like I’ve seen someone do a clean analysis, but I’m sure I could be missing relevant points. What comes to mind is an old Fischer Black paper where he talks about an economy without nominal prices at all, and that instead of paying in dollars, per se, you would just go to the grocery store and pay in a whole bundle of assets, whatever you have in your checking account. It could be just an S&P 500 index fund. That’s how you pay at the grocery store. If we have such frictionless payment technology, we don’t need to hold a medium of exchange. We don’t need nominal prices. We can just pay in real units. It would be as if you went to the movies and just brought along your grocery basket and just gave them some of your groceries. It didn’t matter to them, those groceries, because they could just frictionlessly exchange those groceries for whatever the movie theater wanted to have on hand.

Beckworth: I guess if I had to speculate, and I think what you said makes sense, but if I had to speculate, I think we will always be human. We’ll always have cognitive biases, and these give rise to money illusion. They give rise to information sensitivity, not wanting to go look at stuff until you need to look at stuff. Sticky information gives rise to even sticky nominal debt contracts. Why is the market for nominal Treasury bonds so much larger than for TIPS? People like fixed nominal prices on their debt contracts.

These are all, to me, cognitive biases or behavioral reasons that I don’t know if we can get past them. We’d have to rewire our humanity maybe to do that. Even if we went to a world—I guess going back to what you were saying a minute ago—a world of completely indexed transactions, everything’s indexed, so there is no nominal distortion anywhere, we would still have people who would have these tendencies. I guess that’d be my thought on that.

Halperin: I think that makes a lot of sense. Though in the sci-fi limit, I think things can get really crazy. I think this is what you were probably referring to by modifying our humanity. More prosaically, that could be like we all have AI agents that do our shopping for us, and that technological advancement allows us to solve our behavioral biases. Then certainly in the transhumanist utopia, we have whatever other brain modifications that are not desirable, I don’t know.

Beckworth: That’s it—transhumanism. We won’t be human anymore. Noah Smith had a great piece a month or two ago about how AI is going to take away our agency and limit some of the more important creative ideas, decisions—we could turn it over, like you said, AI will do my shopping, my contracting, my investing for me. What is David Beckworth doing? Am I just at home painting art? I hope there’s more to life than that. Now, my wife’s an artist. I should be careful as I say that. These are all super interesting questions.

Look, that’s way off. In the near term, we deal with the real constraints, these real rigidities that we’ve outlined. Your paper does a good job showing why menu costs and why something like nominal GDP targeting—what’s the term you use?

Halperin: Eclectically optimal is the term I totally made up.

Beckworth: Eclectically optimal, I like that. I like that. Very nice. Let’s just start using that term. I think going forward, that would be it. I think on a practical level, nominal GDP targeting is useful when you have something like an oil shock, even though it is flexible. You should definitely look through those. I know the tension for central bankers right now is, yes, we want to look through them but we’ve also gone through four or five years of above target inflation with a spate of mega supply shocks, war in Ukraine, COVID, now this. 

If you do it long enough, then inflation expectations are not anchored and so we become what? An emerging market can’t look through supply shocks as easily as a well-established one with inflation-fighting credibility. That is the tension. I think that’s why something like a nominal GDP target could be a good, useful cross-check. I had a Substack out recently where I said, “Look, you don’t have to be explicitly targeting nominal GDP, but you could cross-check to get a baseline measure of what you think potential GDP is, add 2% to that. If you’re persistently outside of that, then there’s excess aggregate demand that you should respond to. Otherwise, look through it.”

Halperin: It’s an excellent post. I highly recommend listeners and watchers read that.

Transformative AI and its Macro Implications

Beckworth: Let’s go back to your paper on transformative AI and interest rates because we haven’t completely exhausted that awesomeness yet. I want to go to the implications again of a world with rapid economic growth due to AI. As you outlined, we would have higher real rates. I’m trying to think through what this would mean for a number of things. Let’s start with this, national debt. All right? On one hand, the interest costs and debt would go through the roof. On the other hand, the taxable base in terms of real income would be much larger. I don’t know. Have you thought through those two possibilities and where it lands in the middle?

Halperin: Yes. Another excellent question and one where you see people making important mistakes all the time. There’s two or three forces that work here. One is right now the US has, I don’t even know, 90% debt to GDP ratio. That’s 90% debt to the current level of GDP. If we start having really fast technological growth, that makes it much easier to pay off that existing stock of debt, like what happened in the ’90s when the debt was drawn down and there were fears that we would run out of debt, out of Treasuries to issue. That’s one force, the depletion, so to speak, of the existing stock of debt.

Then on top of that, there’s the two competing forces that you mentioned. Faster growth means higher tax revenue, but faster growth in general equilibrium means higher interest rates, that is, higher rates on new debt. The average maturity of Treasuries is like six years or something so the US has to roll over its debt stock every six years or something like that. After six years, we’re going to be stuck with all those higher new interest rates. Hence then, the question is this R versus G, or really R minus G, is the effect of AI on interest rates or on growth larger? What determines that?

In principle, many things, but again, we can think of a baseline. The baseline benchmark that economists think of is that whether R or G goes up by more depends on—this is going to be super nerdy—this super important parameter, the elasticity of intertemporal substitution. How willing are people to trade off consumption today versus consumption tomorrow? In particular, is this greater than or less than one? Economists, macroeconomists and financial economists, spend a lot of time getting in fights, trying to debate, is this thing greater than or less than one? People have different views. I think as a benchmark, it’s like approximately one.

And this is what the empirical estimates in my paper with Zach and Trevor would find where—by the way, we do see in that data, in that paper, we have some nice data showing that R and G are indeed lined up. Higher growth leads to higher interest rates or correlationally leads to higher interest rates. We estimate something around one for the slope of that relationship, this elasticity of intertemporal substitution. I would say that higher growth leads to one-for-one higher interest rates so that new debt is not any easier or any harder to pay off as a response to AI. That would say that AI is not a magic solution to our debt problem. We have to figure out how to bring down debt.

I also do want to flag though that the best microeconomic estimates to my read find that this critical elasticity of intertemporal substitution parameter is less than one. In fact, our best estimates are 0.7 in my paper. That would mean that interest rates rise by more than growth goes up. That would mean that it’s harder to finance government debt in response to acceleration and growth. That is interesting to think about.

Let me make even one more point if I may or maybe even two more. One is the effect on, you might think emerging markets versus developed countries. If we had developed advanced robotics today, I would certainly be interested to get a robot in my home to do the dishes, fold the laundry. If you live in a country where labor is very cheap, you might be less excited, less willing to pay for an advanced robot. 

For that and other reasons, you could imagine higher growth in the United States, but not such a large acceleration in growth in, say, emerging market economies. Global capital markets are integrated, so if interest rates in the US go up, interest rates in other countries are also going to go up. You could end up with something like the other Volcker shock, where Volcker raising interest rates pushed down inflation in the US, but also pushed up real interest rates in other countries like in Latin America and contributed to the debt crises there in the ’80s. You could imagine AI causing sovereign debt crises for countries that don’t benefit as much from AI. That is, I think, something to keep in mind.

Beckworth: That is interesting. There are these distributional issues globally, not just nationally. It’s one thing to solve and think about the problems in the US or in Europe. If it spills overseas, it could be real problems there in terms of fiscal solvency concerns. It’s complicated. The answer is it’s complicated. Potentially bad if the elasticity in intertemporal substitution is less than one, we might be in trouble. Bottom line is we need to get our fiscal house in order no matter what. Even if we do have this awesome transformative change, let’s rein in those deficits. There’s the fiscal question.

The last question I’ll ask tied to this world of transformative AI is what does it mean for individual or household wealth accumulation? I think of two things in particular, participation in the stock market, but also the other big thing is home ownership. One of the main ways Americans get wealth is through their homes. Of course, 30-year mortgages are a unique phenomenon in America. It lowers the financing cost of homes for many people. Any thoughts on how a world of high interest rates would play out in terms of household wealth accumulation?

Halperin: Interesting. A couple of considerations come to mind. One is, again, this consumption smoothing idea that if we know that we’re going to be a lot richer in the future, then we don’t need to save as much today. My friends working in AI labs are not putting money in their 401k or whatever, things like that. Indeed, capital markets are integrated, so you might expect mortgage rates to go up, but also your income will be going up a lot faster. Hopefully, you should be able to keep paying that same mortgage rate if those things are rising proportionally.

Then a third factor is that another asset price that AI could affect is the price of land. Not housing per se, but land, where you could imagine ending up back in a kind of Malthusian world where natural resources are the only thing scarce in the world of abundance because we can’t produce more land, at least until we think about colonizing the stars and so forth. That could push up the price of land, making housing more expensive through that channel in particular. How all those things net out, not totally clear to me.

Beckworth: Interesting. It might be useful if you have the ability to actually buy a home or land, but physical land, even if it’s just a lot out in the country or something, just to hang on to it. Another implication is we really do need Elon Musk to colonize Mars to solve that constraint as well. 

Well, with that, our time is up. Our guest today has been Basil Halperin. Thank you so much for coming on the program.

Halperin: Thanks, David. This was a lot of fun.

Outro: Macro Musings is produced by the Mercatus Center at George Mason University. Dive deeper into our research at mercatus.org/monetarypolicy. You can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. If you like this podcast, please consider giving us a rating and leaving a review. This helps other thoughtful people like you find the show. Find me on Twitter @David Beckworth, and follow the show @Macro_Musings.