Zachary Mazlish on the Political Implications of Inflation and the Impact of Transformative AI

Zachary Mazlish is an economist at the University of Oxford, and he joins David on Macro Musings to explain some recent and important macroeconomic developments, specifically the inflation linkages to the 2024 presidential election and the macroeconomic implications of transformative AI. David and Zach also discuss transformative AI’s impact on asset pricing, optimal monetary policy in world of high growth, the causes of the slowdown in trend productivity, and more.

Check out our new AI chatbot: the Macro Musebot!

Read the full episode transcript:

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: Zach, welcome to the show.

Zachary Mazlish: Thanks so much for having me on, David. It's an honor to be here.

Beckworth: Well, it's great to have you on, and you've written several interesting articles, and our common friend, Basil Halperin, is what connected us. He directed me to the Substack piece that you've written. It's really fascinating. In fact, it cleared up my thinking [and] some of the questions I had on the linkages between inflation and the election, and we're going to talk about this article that you've written on your Substack. We'll provide links to this, listeners, in the show notes, as well as Zach's website, and all of these articles that he has written. They're really interesting.

Beckworth: The article is titled *Yes, Inflation Made the Median Voter Poorer.* This is a very insightful piece. I was, again, thrilled to read it, because it cleared up my thinking, and hopefully, it will clear up the thinking of some of our listeners too, because there's been so much said about real wages, real earnings during the pandemic period. Many people have argued that this is great evidence for running the economy hot. Yes, we had some inflation. In fact, Paul Krugman, [as] you mentioned in your note, says, "Hey, but they've caught back up," but there are a number of people who have said, "Well, it's more complicated than that.” Let me read, actually, an excerpt from your paper, before we get going here, that highlights that this is a little trickier than FRED or some pundits might say on Twitter.

Beckworth: This is a quote from your paper. "There has been a fair amount of discourse on this topic throughout the last few years, and the last couple of days since the election have seen econ Twitter flooded with takes. Arin Dube, alongside Jeremy Horpedahl, continues to defend the thesis that real wages 'really have risen for the working class in America'. Jason Furman and Ernie Tedeschi have both said it’s complicated. Andres Drenik and a team of co-authors have a new paper finding that inflation reduced welfare for the median worker by $5,000, and Noah Smith goes so far as to say that in '21 and '22, Americans’ real wages suffered their biggest drop in postwar history.” You say, "What is going on here?"

Beckworth: So, all of these different conclusions, and you have worked hard to put something together. And so, I want to turn this over to you, but before I do, let me just give the punchline to our listeners, the three points of your paper. Number one, first point, "In no part of the income distribution did wages grow faster while Biden was president than they did in 2012 through 2020." We'll work through all of the measures, and you provide a robustness check on this claim by looking at multiple measures. Number two, big point, "The change in median wages is not what matters, it's the median change in wages that does." That forced me to think carefully about what it is we actually want to measure. So, that was a great point.

Beckworth: Finally, "People do not feel wages, they feel incomes,” and “median growth in total income, post taxes and transfers, was not just historically low, it collapsed and was deeply negative from 2021 onwards." So, let's start with your first point. Again, the first point is, "In no part of the income distribution did wages grow faster while Biden was president than they did [from] 2012 through 2020."

Inflation Made the Median Voter Poorer: Comparing Periods of Wage Growth

Mazlish: Yes. So, where this is coming from is that there's this paper that I think has gotten a fair amount of attention called *The Unexpected Compression,* which is by David Autor, Arin Dube, and Annie McGrew. They make the argument that the post-COVID period was a uniquely strong period of wage growth for the lower half of the income distribution. I think this paper is behind a lot of people's optimism about real wage growth during this inflationary period. I think, relative to that paper, the point I'm making is that post-COVID is different timing from post-the inflation starting and post-Biden becoming the president. And so, if you start in 2020 versus starting in 2021, that already makes a big difference to the wage measures that they use in that paper. That's not to say that they're doing anything misleading, but it's just, for thinking about the election, I think it makes sense to start thinking about what happened when Biden got in office.

Mazlish: So, the first thing I show is that if you take their exact data construction, which is using the current population survey, the CPS, and you just plot the growth in median income by income quintile, during the Biden presidency versus during Obama's second term and during Trump's term, you see that the change in the real hourly wage was no stronger for any of those income quintiles during the Biden presidency than they were during either of the previous two presidencies. For the second through fifth income quintiles, the Biden presidency looks distinctly worse than the Trump presidency. So, that's the basic point, that no part of the income distribution outperformed their 2012 to 2020 trends. I do think another caveat there is that that's just hourly wages, and I get into later that if you do more compositional adjustments and look at weekly or annual earnings, the picture looks even worse for the Biden presidency wage growth.

Beckworth: So, a key point with that very influential study is that it's timing that matters, right? If people are basing the election on how they feel since President Biden went into office, they're not going to go all the way back to 2020, they're going to start in 2021. That's what many of us miss. And so, we have to be very careful when we do this analysis. Is that the big takeaway?

Mazlish: Yes, I think it's that. Then, the other takeaway that that section develops is that even just the distinction between hourly, weekly, and annual earnings is quite important. So, hourly wage growth looks better than weekly wage growth, and if you take this weekly earnings series that Ernie Tedeschi has constructed that also holds fixed the 2023 demographic and industrial workforce composition, and use that as your baseline, then the comparisons look even worse because it seems like people had declining hours over this '21 to '23 period, per worker, that is. Obviously, employment was growing, but on a per worker basis, they weren't catching up in weekly earnings.

Beckworth: Okay, so, the first measure was median wages. That's the big study that you mentioned, right? You're saying that Ernie Tedeschi does his own compositionally adjusted real wage measure where he controls for sex, age, race, [and] number of factors?

Mazlish: Exactly. So, there's a couple of different points going on there. The first thing I'm talking about is across all five income quintiles, and then I hone in on the 40th to 60th percentile. It's a bit confusing, because Ernie Tedeschi's median isn't the 40th to 60th. Instead, it's the triangle-weighted average of the 45th to 55th percentile. But, on his measure, it's actually the case that real wage growth in that median has been negative over the first three years of the Biden presidency. That's another important caveat, is that we only have full data for the first three years of the Biden presidency, and his numbers will look a little bit better once we have all of 2024 included.

Beckworth: Going back to the initial paper, the Autor, Dube, McGrew paper that was so influential, and I've seen it many times on Twitter— In fact, I saw a tweet recently, after the election, bringing this up again. The point you're making, though, is that you’ve got to go be careful in the measures, and you’ve got to time it right as well. But the other thing that I think I got out of this is that comparing it across other presidents, other periods, even in that first measure, the results aren't that great. The lowest income quintile was the one that did the best, but compared to historical periods in the past, it really wasn't that remarkable. Is that a fair assessment?

Mazlish: Yes. I think it's important to have in mind that what the data shows is that this 2012 to 2020 period was fairly remarkable relative to the 20 years before that, and so it's a tough reference point for Biden to compare to Obama's second term and Trump's term. But if that's the comparison that you're making, then it is the case that none of these income quintiles clearly did better, and I think when you do things carefully, it looks like they did worse, especially as you get into higher income quintiles.

Beckworth: So, the reason that I'm pushing on this is because one of the arguments made, and I made it at some point in the past, is that if you run the economy hot, it's going to be good for the labor market, particularly those in maybe lower income groups. But what you're showing is that that isn't necessarily the case. The evidence doesn't really support that, at least compared to previous historical places. Yes, the low-income group did better than other income groups, but no, it wasn't any better than any previous run of good times for that low-income group.

Mazlish: Yes. So, they did worse than they did during the Trump administration, certainly, and I think if you look at weekly or annual earnings, they did worse than they did during either of the prior two presidencies. That's the case even amongst the low-income groups. Then, if you look, say, from the middle quintile to the fourth quintile and the fifth quintile, they did distinctly poorly, even relative to a longer historical time span.

Beckworth: Okay, so, I was being too generous there, that when we look at these other measures, collectively, the running the economy hot story really doesn't have legs.

Mazlish: Yes. I think the sense in which it might still have some legs is that I'm not comparing across countries. Obviously, this was a very negative period of shocks. And so, maybe running the economy hot actually resulted in better income growth in the US than we got in any of these other countries. So, I don't think it necessarily tells us that running the economy hot was bad relative to our other options, but it does tell us that running the economy hot wasn't some cure-all that gave us this uniquely strong period of wage growth.

Beckworth: In other words, we need to do the right counterfactual here, and it could have been far worse. But unfortunately, for the election, households aren't thinking about that counterfactual. They're thinking about the actual loss in real income that they feel they're experiencing. We should say that exit poll surveys show that this was a big issue for many voters. This was probably one of the top issues when they left the voting booth. So, you have this exit poll data, and what you've done is provide a robust measure of real earnings for these households or these individuals that supports that claim. Because I've seen many people say, "These folks are confused, they're misguided. They think that they're worse off, but they really aren't." What you're saying is, "Well, actually, if you look at broad measures of real income, real wages, real earnings, [then] they actually are worse off." So, they really are identifying something, and, again, I think a key issue here is, when do you start the comparison?

Mazlish: Yes, absolutely, and I think, to your point, my motivation for looking into all of this is that I was a bit puzzled by the discrepancy between people saying that this was such a strong period of real wage growth, and it then seeming like people were really upset about this period of wage growth, and I do think that there are lots of other reasons why households might not like inflation, and there are lots of more complicated explanations people have come up with, but I felt like I wanted to give the simple explanation, that when people say that they feel like they've gotten poorer, they actually know what they're talking about, more of a shot. And so, that was what I found.

Beckworth: I guess that when you're going up for an election, people are going to use, as the base year, the start of the previous administration. They're not going to go back to 2020. You also note that some of this was just pure luck, because the distribution of the stimulus checks and things like that may have made them look more fondly on the Trump administration than on the Biden administration.

Mazlish: Yes. So, everything I've been talking about so far are pre-tax and those COVID stimulus earnings, but once you include the COVID stimulus checks in your measure, you see a huge boost in real take-home annual incomes in 2020 and a significant fall-off afterwards. I think, really, in particular, [in] 2022— So, the second year of the Biden administration looks like possibly the worst year of after-tax real income growth that we've ever seen in the data. That is, in fact, affected, in large part, by the timing of those pandemic stimulus packages. But I think that more research needs to be done on how households form their reference points, but it seems fairly intuitive to me that households would anchor to how they felt just before Biden took office, or even in Biden's first year, to how it developed as the inflation grew and their real incomes collapsed.

Beckworth: Okay, so, we have a good sense of real earnings, real income. Maybe one last point, and you've touched on this, but just to flesh this out before we move on to point number two, and that is that it's important to look at total income, not just wages, when we're assessing this question.

Mazlish: Yes, and I think that this is a pretty intuitive point to anyone that, even if your hourly wage goes up but your employer doesn't want you to work as much, you might not be any better off, because I think people want to have a higher annual income at the end of the day. And so, it's important to look at weekly earnings, and it's even better, in my opinion, to look at annual earnings.

Beckworth: Alright, so, [for] point number two in your essay, you stress that the change in the median wage is not what matters, but that's often what we look at in these measures. What really matters is the median change in the wage. Maybe explain that distinction, and why is it so important?

Inflation Made the Median Voter Poorer: The Median Change in the Wage

Mazlish: Yes, so, if you don't mind, I think I'll give this toy example that I gave in the piece, which was pretty helpful for my own— just getting clear on this. I'm going to walk through this example where we imagine that there's three people, person A, person B, and person C. In year one, we have person A making $4, person B making $5, and person C making $10. If, four years later— so, after a presidency— person A is now making $1, person B is making $6, and person C is also making $6, the median income increased because the median income went from the $5 of person B to the $6 of person B. But, for both person A and person C, they had a significant fall in their incomes. And so, two of the three people in this economy did not have a very good four years, and you can imagine that if an election were held, they wouldn't want to vote for the incumbent. The median change, on the other hand, as opposed to the change in the median income, would pick up on that, because the median change in this world is negative.

Mazlish: And so, the reason to look at the median change is that it tells you people's change in income between four years ago and now. But the idea is that, for understanding the election, there's no particular privilege that median income workers have. Rather, we want to know, are the majority of people better off or worse off relative to the trend and what they were expecting? The median change doesn't identify that exactly. It still requires the assumption that the distribution of changes around that median change are relatively symmetric. And I actually have looked into that a bit since writing the post and validated that assumption, and I think that's something that I want to do some more work on. But, basically, the idea is that the median change is a sufficient statistic for understanding whether the median household, like the median worker or the median voter, has gotten better off during a given stretch of years.

Beckworth: Okay, so, we want to look at the median change in wages or income. What did you find when you looked at that?

Mazlish: The Atlanta Fed, thankfully, already tracks this for us. They use the CPS data, but then they make sure that they're looking at within-person changes across 12 months and making some adjustments to the data to make sure it's compositionally correct. If you look at that, and you just look at median growth in cumulated year-over-year weekly real earnings, by president, you see that Biden's first term was the second lowest period of growth since Bill Clinton's second term and was distinctly far below either Obama's second term or Trump's term. The numbers, in case people are curious, are that during Obama's second term that median growth was 2%, during Trump's term it was 1.9% annualized, and during Biden's term it fell to 0.6%, which is a pretty steep drop off.

Beckworth: Okay, so, the picture we're painting here is that no matter how you slice it, there was really poor real income growth for households during these past four years. People understood it, and they went to the polls with it, and they voted accordingly. A question that flows out of that— why has it been so hard for us as economists, as journalists to see this? Because we've settled for the easy measure, I guess, if we want to call it that, or ones that are less clear about what you've uncovered here. So, why has it been so hard for the profession to see this as clearly as people have felt it?

Mazlish: It's a good question, and I think, sometimes, it's the case that with this almost simple wage data, people just assume that whatever they first see is the correct narrative about it. If you look at average hourly earnings, that's a much more confusing picture, because it's totally compositionally distorted by COVID, and I think that that is the measure that, most frequently, if you look up earnings on FRED, people first gravitate towards. So, I think that there's just these simple anchors that people got tied to. But then it's not like the Atlanta Fed data series is some obscure data series. I think that it's pretty commonly used, and if you look at it and adjust for inflation, you pretty instantly see that this was not a very strong period of earnings growth.

Beckworth: So, we touched on the third point already where we mentioned the after-tax, after-transfer position of households, so, total income. And as you noted before, that really paints a negative picture for Biden, especially in '22 and on. But have you done some updates on this that paints a better picture of distributional outcomes?

Mazlish: Yes. So, kind of to the point we were just talking about, I think that as I was thinking more about the median change, I realized that the reason the median change is better than the change in the median is that you really want to be able to compare the entire distribution of changes across periods and know, are most households in different places on the wage change distribution doing better than they did before or doing worse? I've since made some plots where you look at these after-tax annual income changes and you just plot the distributions under the Biden presidency versus under the Trump presidency.

Mazlish: You get this very clean overlay of the Biden distribution— if you can imagine two normal-looking distributions— just lying to the left of the Trump distribution. So, there are more changes that are worse at every point on the change distribution during the Biden presidency years than during the Trump presidency years. I think that tells you that there are a lot of people who, for any given worker in the distribution of how much better off they got, they tended to be a bit worse off than they were relative to their income growth during the Trump presidency.

Beckworth: Wow. That's pretty remarkable. I look forward to seeing that chart. So, again, just repeating this point, households were saying that they don't feel great. Economists were saying, "Oh, you're just confused." Lo and behold, you dig deeper into the data, they were right. They knew that they were onto something. They knew what they were experiencing. So, let me wrap up this discussion of this article before we move on to your next article and just ask, what's been the feedback to this? Because you really have pushed back against this popular narrative that economists [and] thoughtful people have been promoting. Have you gotten any negative feedback, positive feedback? What have you seen?

Assessing the Feedback to Zachary’s Article

Mazlish: I think that I've mostly been really grateful for the feedback and felt like people have had positive things to say. I've also had some people follow up with quite interesting points. David Splinter, who people might know from the whole inequality debate between Piketty and Auten and Splinter, emailed me a paper of his where they have actual tax return data and are able to track workers over time, across their tax returns. Their data only goes through 2022, so it doesn't give you a full picture, but he has a paper out. In his data, you see the same image, where 2022 is really this unique drop off in after-tax annual income. That was an interesting point.

Mazlish: I think that there definitely are some people who still seem a bit confused [about] how to reconcile all of the data, which makes sense. I do want to emphasize that this is ongoing work. I spun it up in the wake of the election, and I think that there's more ways of cutting the data, more ways of getting more granular in terms of looking at, what about the changes in incomes in the swing states, for example? That could be a very important question that my piece doesn't touch on and some people have raised with me. So, I think that people have given me lots of good leads and extra things to think about, and I've been really grateful for that.

Beckworth: Okay, and if anyone listening out there wants to commission Zach to write a full-length paper on this, such as the Brookings Papers on Economic Activity, you know where to find them at Oxford, or we can give you his contact information. So, Zach, this has been really great, really useful. Again, it really helped me clarify some of my thinking on this issue, and definitely raises questions about how we understand and assess households' standard of living and how they feel versus what some indicators say.

Beckworth: Alright, Zach, let's move on to another paper that you have written, and you have actually co-authored this, again, with our friend in common, Basil Halperin, as well as another colleague, Trevor Chow. This paper is titled, *Transformative AI, Existential Risk, and Real Interest Rates,* and this is really fascinating, because this is a topic that's been very germane, very front and central. Maybe other than the election, it's something that's getting a lot of discussion in the news. We're all using AI quite extensively now. There's talk of this big transformation, at the limit of some singularity taking over. So, you guys jump in the deep end of the pool with this. Maybe you can begin by talking about, what is transformative AI? You talk about this double-edged sword to it. What is the double-edged sword to transformative AI, and why should we think about it?

The Significance of Transformative AI and its Double-Edged Sword

Mazlish: I think, to start off, we're going in for the wackiness here, where transformative AI is the extreme scenarios that very much the forefront of Silicon Valley is considering and maybe not what the typical economist is expecting from AI. But based off some work from Tom Davidson, transformative AI has been defined, and we define it, as a situation where either real GDP growth over 10Xs and rises to something like 30% per year, or where this AI kills us all and destroys humanity as we know it. Again, I think we understand that lots of economists think that these are totally science fiction scenarios, but there's been some real work out there by people where they are considering these possibilities from AI, and we want to speak to those possibilities.

Beckworth: So, you mentioned a greater than 30% increase in global GDP, so, that would be a benchmark or a threshold that you'd have to cross to be in this space.

Mazlish: Yes. Obviously, it's a bit semantic and hazy. I think that if GDP growth started growing at 20% per year, we'd still want to say that that's transformative AI, but 30% is a benchmark that I believe Tom Davidson has used in some of his work, and we adopted a similar one.

Beckworth: Yes, so, the double-edged sword is that we have really robust, amazing real GDP growth, or we all die. Is that the double-edged sword here?

Mazlish: Yes. Sorry, I got away from that. That's the idea. 

Beckworth: Yes, so, we end up in a world of the Terminator or we end up in utopia. So, we're going to focus on utopia today, or maybe it's not utopia. There'll be some challenges. Your paper's point is that there'll be some transformational issues that we've got to go deal with [in] getting to this utopia. But let's focus on the better case scenario. What would be the implications for asset pricing? Because this is where you go in the paper.

The Impact of Transformative AI on Asset Pricing and its Policy Challenges

Mazlish: The basic logic is something that grad students would learn in their first week, really, of a macro PhD course, and, presumably, undergrads might even come across it as well, which is just that when growth expectations are higher, that's going to push up real interest rates. The most classic logic for that is a consumption smoothing logic, where if you know that you're going to be richer in the future, you want to borrow resources from that future to consume now, and this greater demand for borrowing is going to push up real rates.

Beckworth: You could also say that there's less need to save for the future if you're going to be incredibly wealthy and taken care of in the future. The reason that we save for the future is because we may not have that income that we have today, so that would increase rates. Could you also tell an investment story, too? As firms see higher return to capital, productivity is soaring, [that] for a given stock of savings, they're going to tap into that, use it up so that there's nothing left, so rates have to go up. Is that also an interpretation of that?

Mazlish: Absolutely, and I do think it's worth emphasizing that we're really seeing that, where all of these big tech firms are having massive amounts of capex that maybe we've never seen before, and it's all to build these data centers to train the next AI models and to work with AI. And so, the demand for borrowing— I think that Matt Levine had a piece about the credit markets for AI, where it's specifically about this huge boom in demand for borrowing to finance AI investment, which, like you said, David, would also push up real rates under the very similar logic.

Beckworth: So, what are the challenges that this creates for policymakers, for the real world?

Mazlish: Obviously, there's so many things that we don't know about exactly how it's going to develop. I think the big questions that lots of people are focused on are concerns about what it's going to do to people's jobs and whether or not it's going to automate people's jobs. I think that there's a lot of great work being done on that. It's very interesting to me that, when you're looking at this, a lot of times what happens to wages or what happens to jobs depends on something like whether the elasticity of substitution between capital and labor is greater or less than one, and it's funny how this little parameter in our model being greater or less than one tells you whether this is a great world for workers or only a great world for capitalists. And I think there are richer models that complicate that story in lots of important ways, but I do think that the labor issue is probably the one that's most forefront for policymakers.

Beckworth: So, how high could rates go?

Mazlish: So, in the paper what we do is that you can take a basic Euler equation that comes from the Ramsey model and just plug in, what if expected growth is 30%, and what if you add no complications to the model? In that extremely simple model, you're going to get something like, real interest rates go to roughly 30% per year. That's not necessarily the prediction we're making, but what we then show— and, I think, more of the empirical contribution of the paper— is that if you look in the last 30 years at the relationship between real interest rates and growth expectations, there is a strong positive relationship.

Mazlish: And the coefficients that we get, whether you're regressing it in changes or in levels, are around one. So, that suggests that for every percentage point increase in expected growth, there's about a one percentage point increase in real interest rates, holding all else equal. Obviously, there are lots of things that aren't being held equal in the real world, but I think that that's a useful benchmark for thinking about, if growth rates were to go to 10%, [then] that could really push up real interest rates on the order of 10 percentage points.

Beckworth: That's a nice rule of thumb that we often see used today. People look at, well, how fast is real GDP growing? We would expect real rates to track that, and what you're doing is providing a precise estimate of that given some rapid growth in the future. What are the implications for financial markets?

Mazlish: Just one thing I want to say on that that I think is maybe somewhat interesting is that there's been a lot of papers on this relationship between real interest rates and expected growth. It's at the heart of the Euler equation. It's a very central macroeconomic relationship. I think that part of our contribution is that it's hard to measure real rates and it's hard to measure expected consumption growth. Doing both of those well is really key to identifying this, and a lot of previous work has relied on some sort of statistical model of inflation that you then subtract from nominal rates to get a real rate, and then also having some statistical model of expected growth.

Mazlish: We get around that by either using inflation-linked bonds to give you direct measures of real rates, so TIPS in the US, and if we don't have access to inflation-linked bonds, we use long-term inflation and growth expectations from this consensus economic survey. When we do all of that, and additionally control for default risk, which is something that I think people also need to do if they're going to run these sorts of regressions, you get this very strong relationship. So, I think that the implication for financial markets is that real interest rates will go up. Then, that's obviously going to have somewhat ambiguous effects on the stock market because you might think that expected dividends are growing, but so is the rate at which you're discounting them. And so, depending on which of those increases more, stock prices could go up or down.

Beckworth: One of the implications that flows out of your analysis, though, is that you might be able to see where AI is going based on these indicators you just outlined, correct?

Mazlish: Yes. And so, this is where I think it's a bit of a matter of faith and priors and how much you believe in the efficient market hypothesis. I don't think that people necessarily have beliefs about that that are responsive to the evidence. It seems like everyone comes in— either they're convinced by the arguments for why markets are efficient, or they think it's totally silly that markets would be efficient. But if markets are efficient, and if you really believe that markets are efficient, then looking at this real interest rate is one of our best measures of whether this smart, efficient aggregator of information that is the market expects AI to deliver a boom in growth or not. And I think in the world where you don't buy that market efficiency story as much, if you think that the market is eventually going to catch on and adjust once it becomes obvious that we're getting this growth, and you yourself believe this growth is coming, then there's more of an opportunity to make a trade. And obviously, usual disclaimer about— this is not investment advice, but I think it's an interesting angle to think about.

Beckworth: Yes, absolutely. I'm a champion of the efficient market hypothesis, and I would say that it's the best we have. Yes, it's not perfect. Market signals can sometimes be wrong, but it's definitely, I think, far better than what we'd get elsewhere, because it accumulates up all of human understanding and wisdom. It puts skin in the game so that the information is tied to risk-reward. And I would finally note, circling back to our earlier conversation, you know what? Prediction markets did really well in the presidential election, right? It's a form of a financial market. They did far better than the polls and what many people expected. So, we've got to give it some credit.

Mazlish: Yes. We also do have some empirical evidence in the post that, historically, when 10-year real rates are high, that has corresponded to higher output growth over the next 10 years. Again, there's a bit of a limited sample where we have good measures of real rates that, also, we can look forward 10 years from and have the data on. But I'm very much on your side, David, where I think that efficient markets is a very good benchmark and we should take seriously what the market's thinking.

Beckworth: Yes. I loved your paper, because it forced me to start thinking through some of the implications of this world where we have rapid real growth. You mentioned really high real rates, which, as you said, maybe we discount asset prices in the future more, but maybe those asset prices are so high already because of higher future cash flows. They will still be high even after we discount them. And I was just thinking through all of the many problems that this could solve or maybe issues it could address. Let me just throw a few out, and if you have any response to it, let me know.

Beckworth: First off, if we had such rapid growth, it would solve many of the problems that we now have to think about, like climate change. If we have such rapid growth— this is the growth that could help us adapt to a new climate. It could provide solutions. I just think that this is part of the answer to climate change. It's also a way to think and deal with a decline in population growth rates. We talk about a decline in population growth rates as actually being an important driver of economic growth. If AI can somehow offset this— Now, I've talked to some people who are skeptical that it will, but let's go with this. Let's run with this belief that it does happen. Man, it would take care of that as well. We wouldn't have to worry as much about a declining labor force when we have robots and AI doing a lot of the work for us, so win-win.

Beckworth: Let's go to an issue that is really plaguing our country, and that I think has been affecting other advanced economies as well, and that's the housing stock. We don't have enough housing in the US. There are many reasons why we don't build, but I think at the very core of our housing shortage is this tension between your house being an asset and also being a consumption good. Housing is the primary way we accumulate wealth in America, but we need it to live. It's also a consumption good. And people become NIMBYs because it is what they're going to retire with, the wealth that they're going to retire with.

Beckworth: If you could find another way to accumulate wealth— If we're having 10% real GDP growth, the stock market is soaring, [then] I imagine that household portfolios of stocks might mitigate the focus on houses being this thing that we have to preserve at all costs, and it might open up more housing supply. Maybe I'm reaching there, I don't know. Last thing is, we would definitely eliminate the zero lower bound problem if we've got 10% real interest rates, right? This whole literature we've spent years discussing and thinking about— it's bygones. No need to worry about it anymore. Any thoughts on any of those hot takes I just shared with you?

The Broader Macroeconomic Effects of Rapid Growth

Mazlish: I think that they're all good takes. I think that the important takeaway— and I'm sure other people know more about this, but my understanding from some of Daron Acemoglu's work is that the scenario we really want to worry about is where AI is a bit better than existing alternatives, because then it displaces labor but doesn't create this cornucopia of wealth that we can redistribute and make everyone better off with. But I think in the world where AI is generating upwards of 5% GDP growth per year, there's going to be a lot of new wealth to go around.

Mazlish: And to your point about people not having to rely on their house because they can be in stocks, even more simply, if Treasuries are returning 8% per year or whatever, you can just buy Treasuries and make lots of money. It doesn't become that hard to accumulate wealth, and I think that there certainly are lots of important issues to think about, how we make sure people who historically don't have much liquid wealth in financial market assets get to benefit from this. But I'm completely with you, that it seems like there's many more good possibilities in that world than bad ones.

Beckworth: Let me throw out one more hot take, Zach, and I apologize for throwing these curve balls at you. What would be an optimal monetary policy in this world? Here's my hot take on this, okay? I know our friend in common, Basil, would agree with me, because he comes from the same perspective. But George Selgin has this book called Less Than Zero: The Case for a Falling Price Level, and he argues that when you get rapid productivity growth, if you can stabilize nominal income— so, here we are back at a nominal GDP target— if you can stabilize aggregate nominal income or stabilize total dollar incomes, allow this rapid growth to be reflected in a gently falling price level so that everyone benefits from it—

Beckworth: In fact, going back to our earlier discussion, everyone was very cognizant of the fact that inflation was eroding their real gains, but you could flip that to the other side. If we had mild deflation, it could be the opposite story. Now, for most macroeconomists, this is a tough pill to swallow, because when you say deflation, the first thing that comes to mind is the 1930s, the Great Depression, which is an awful case of deflation. But you could have a mild productivity-driven version. And I think, in my mind, this would be the way it would work.

Beckworth: If we start with these rapid productivity gains— so, let's think of factories or firms across the world. Per unit cost of production is falling as productivity goes up. There's some competition out there. Lower per unit cost of production is translating to lower output prices. Therefore, we see lower price levels, again, given stable aggregate demand, and people benefit from it. So, that would be my view of how this could work with monetary policy. Any thoughts on my optimal monetary policy in such a world?

Optimal Monetary Policy in a World of High Growth

Mazlish: Yes. I'm completely with you. I think that it would be a wacky world for monetary policy, and I think, certainly, even if productivity growth was just 5% per year for a few years, that would put [on] deflationary pressures that we've never really seen before, and I think that it would require the Fed really keeping rates low to avoid deflation in that world. So, I'm very sympathetic to the arguments in Selgin's great book, and I think it would be a funny time. On some level, I think we sometimes worry about the business cycle, sometimes we worry about growth. If we're getting 10% growth per year, the setting of monetary policy might not be the first thing on people's minds, but I'm sure that—

Beckworth: Fair point.

Mazlish: -you and I would still be paying attention to it.

Beckworth: Yes, that is a great point. This podcast may cease to exist, because most of it does focus on the business cycle dimension, including monetary policy. If we're growing at 10% a year— again, it's going to cover all kinds of problems. Even if we have a financial crisis, guess what? 10% real growth a year is going to quickly clean that mess up, too. How we do financial stability, how we do monetary policy becomes second order importance. So, I hope that we have this amazing world, Zach. It sounds fantastic. I hope that the people in Silicon Valley are correct and we end up there. In the time we have left, I want to go to another article briefly, and this article is going to be a bit of a downer. We've come from this wonderful world of 10% or more real GDP growth. I want to be there, but I want to come back to reality, Zach. 

Beckworth: So, you have another article that deals with the actual observed real productivity slowdown. The title of this paper is called, *Decomposing the Great Stagnation: Baumol's Cost Disease vs. “Ideas Are Getting Harder to Find.”* Again, Basil's name comes up. You two are the co-authors of this. But again, compared to this great future AI world, we're here in reality, and we've had a decline in productivity growth, really since the '70s, and you attempt to explain it through Baumol's cost disease plus some other factors. Maybe walk us through those potential stories that can explain it.

Exploring the Causes of the Productivity Slowdown

Mazlish: I just want to clarify that this is a short research note where we're taking some work that other people have done and making a few modifications to just try and robustly estimate how much Baumol's cost disease has contributed to this productivity growth slowdown. I do just want to shout out this paper by Duernecker, Herrendorf, and Valentinyi, which does this estimate, and we just build on it in a few ways that we think are interesting, but I want to make sure that they get the credit for that.

Mazlish: To your question, the basic idea is that there is productivity growth slowdown within an industry. So, within agriculture, or even more narrowly, like farms, or within information technology, you can imagine that productivity slows down within that industry. What we want to say is that that within-industry productivity growth slowdown is the channel of ideas getting harder to find, which Nick Bloom and co-authors have drawn a lot of attention to, and we think is, in many ways, the leading explanation for why productivity growth has slowed down.

Mazlish: Then, there's another channel, which is this Baumol's cost disease channel, and that comes from the compositional effects where if some sectors are fast-growing and some sectors are slow-growing, in terms of productivity, if people continue to want to consume the slow-growing sectors— and you might think of services versus manufacturing here— those service sectors are going to become a larger and larger share of the economy. That means that aggregate productivity growth is going to just be mechanically driven down by this compositional effect of the slower-growing sectors taking up a larger part. And so, we separate out productivity growth into this within-industry component, this reallocation component, and then a few additional terms that are about whether the inputs into production, labor, and capital, are being allocated towards sectors where they're more or less productive, or the output elasticity— how much additional output you get from an additional unit of capital or labor— is higher or lower.

Beckworth: Okay, and what do you find?

Mazlish: What we find is that, just to fix numbers here, if you look at post-World War II to 1973 as one period, and then post-1973, the oil shock, as another period, annual TFP growth has slowed down about one percentage point, give or take depending on exactly how you construct things. Of that one percentage point annual slowdown, 25%, so 0.25 percentage points, are due to this Baumol's cost disease channel. Then, basically, the remaining 75% are due to the within-industry ideas getting harder to find channel.

Beckworth: The big challenge then is that ideas are getting harder to find?

Mazlish: That does seem to be the dominant explanation. I think that we wrote the paper because we're, in part, very interested in this Baumol story, and we wanted to see how big of an effect it was, and I think that 0.25 percentage points of productivity growth a year is a very significant amount. Compounded over 50 years, that's going to give you something like roughly a 30% difference in aggregate TFP, which means that incomes would be 30% higher if we didn't have this Baumol channel happening. So, that is a big effect, but the majority of the slowdown seems to be within-industry forces, which maybe could be other things than just “ideas are getting harder to find,” but in our simple way of thinking about it, we wanted to call it “ideas are getting harder to find.”

Beckworth: Is it fair to say that Baumol's cost disease is inevitable as the economy grows, [and there is a] bigger service sector? So, it's going to be there no matter what, [but] you just have to find a way to deal with it. Maybe you have to have more rapid productivity gains in the other part of the economy, more idea generation to keep up with Baumol’s cost disease.

Mazlish: Yes. I think that that's definitely an important point, is that Baumol's cost disease isn't a sign of any sort of inefficiency. It just means that people have gotten richer and now want to consume more of these slow-growing sectors. And I think this is a point that Dietrich Vollrath elaborates on in his book, Fully Grown, where he's saying that it's okay that we've grown and we're now growing slower. The fact that we've grown is what's gotten us here.

Mazlish: I do think, back to the earlier discussion about AI, that it's not necessarily inevitable that Baumol's cost disease is going to drag us down, because you might imagine that these sectors we think of as intrinsically low productivity growth are not perpetually low productivity growth. But if there is an asymmetry in a sector’s productivity growth, and crucially, if people continue to want to consume the low productivity growth sectors, then this Baumol channel is going to kick in.

Beckworth: Okay, so, circling back and closing to AI and these findings that you just shared with us, are you hopeful that AI can create more ideas? It can not only affect the products, the outcomes, but the science itself, help generate ideas, help make research more effective, make a big difference?

Mazlish: I'm certainly hopeful. I think, going back to our belief in market efficiency, real rates are a bit higher. They are pretty high relative to what they were in 2018, but they're still not signaling that we're about to hit some sort of singularity and take off. I take that seriously, but I am pretty optimistic. The things it does in my own research process, all of the time, I'm pretty wowed by, and maybe the research that I'm doing isn't the sort of science that accelerates productivity growth, but I can imagine that people who are doing productivity growth-enhancing scientific research are also benefiting from these tools, so, I am pretty optimistic.

Beckworth: Well, with that, our time is up. Our guest today has been Zach Mazlish. Zach, thank you so much for coming on the program.

Mazlish: Thank you so much, David. Really great to be on. Huge fan of the show.