Jonathon Hazell on Phillips Curves, Wage Rigidity, and How to Measure R-Star

Jonathon Hazell is an assistant professor of economics at the London School of Economics. Jonathon joins Macro Musings to talk about Phillips curves, R-stars, and nominal wage rigidity. Specifically, Jonathon and David also discuss the how to view the recent inflation experience, how to measure the natural rate using natural experiments, the downward nature of wage rigidity, and a lot more.

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: Jonathon, welcome to the show.

Jonathon Hazell: Thank you so much for having me, David. I've listened to the show many times. I grew up reading your blog. It's a real privilege to be here.

Beckworth: Thank you for reading the blog back in its glory days. I don't do that anymore. And thank you for following the podcast. I've had several of your colleagues on the show, Ricardo Reis, Ethan Ilzetzki, and Ben Moll. You're at a great place, great history of macroeconomics, London School of Economics. We'll talk more about that as we get into your career, but a real delight to have you on. You've done some amazing work on Phillips curves, on wage rigidity, a very fascinating new paper on R-star, an improvement upon what we probably have already. I'm excited to chat with you today about all of that. Before we do that, though, let's talk about you. How did you get into macroeconomics? What's your career path?

Hazell: Great. I feel really privileged because it's what I wanted from a very young age. My parents, my family always teased me that they remember me reading The Economist magazine as a child. When I was 16 through 18, the financial crisis was unfolding. I remember watching the events unfolding and thinking, "That's it. I want to be an economist." I was reading the blogs, Paul Krugman, yourself, Scott Sumner, many others.

Hazell: I just thought, this is such a vital, such an intellectual, such an important series of debates. People were trying to figure out what was going on, looking at data, using simple theory. I thought that this was a conversation that I wanted to be a part of. That's basically it. I was so influenced by these events and by the discussion about these events at the right time, being a teenager, that I went on this path ever since; economics generally, and also macroeconomics specifically, business cycles, finance, monetary economics, that kind of thing.

Beckworth: Now, did you come to the US to do your graduate work, and then go back to London afterwards?

Hazell: That's right. I started off as an undergrad at Cambridge in the UK. Then, I was lucky enough to do my PhD at MIT, which is a great program. Then coming, I guess, back to my origin story, I vividly remember reading a couple of Paul Krugman blog posts, and also an essay by Ken Rogoff, where they spoke about their experiences being PhD students at MIT back in the 1970s. I remember thinking, this is such a gilded experience, and thinking that one day, with my fingers crossed, that it could be me, and indeed it was. It was a real dream come true to come to MIT, and then also, afterwards, to manage to land this job, which has the colleagues, the environment that I could dream from.

Beckworth: Yes, I can imagine the hallway conversations at the London School of Economics must be something else. Just be able to walk down the hall and say, "Hey, Ricardo, what do you think about this, or this?"

Hazell: It's an absolute gift. One thing that is remarkable— well, maybe two things that are remarkable about LSE— the first is that we have an amazing age distribution. I have some of the best young colleagues of my age working on all of these diverse topics, on markups, on international finance, and so on. Then I have colleagues going all the way up to through Ricardo Reis, Ben Moll, Ethan Ilzetzki, Silvana Tenreyro, some of whom you've had on your podcast, all the way up through to a Nobel Prize winner like Chris Pissarides. At any point in time, I can hail any of these people down to have lunch. For me, it just feels like a real privilege to be able to have these conversations.

Beckworth: You have a very big department too. I was looking online, a lot of economists, right?

Hazell: Yes. I suppose the other thing that I think is really remarkable about the department is, I think, throughout my colleagues there, I think that there's a real commitment to try to ask really big picture, interesting questions. You'll see that from your podcast guests too. When you interview Ricardo, when you interview Ethan, or when you interview Ben, I think you're never short of a sense, I guess, that these are the people asking the really important questions that… why is inflation going up, what's going to happen to interest rates? The ones that are the core of macro, and I really appreciate the sense that we're always trying to be doing rigorous research, but also [being] part of that vital, important conversation.

Beckworth: Absolutely. Again, just to reiterate the rich history of London School of Economics, just a few names from the past; John Hicks, famous person of the IS-LM, Bill Phillips, the Phillips curve, James Meade, I'm a fan of him because he was an early nominal GDP targeter. Lionel Robbins, I guess, back in the '30s helped really establish it. William Beveridge of the Beveridge curve, very amazing of course, and F.A. Hayek in the '70s, and other names as well. A rich history, you walk down those hallways, and their presence is there with you. You're just thinking, "Man, I’ve got these great minds now, in the past." So, fun place to be, and you get to teach macroeconomics there. Tell us about that. Are you teaching graduate, or undergraduate, or both?

The Current State of Macro Education at LSE

Hazell: I teach both. I teach first year undergrads, third year undergrads, and second year PhDs. One thing that's nice about that is that I teach, I guess, the very first and the very last course that you could take as a macroeconomist. The first course that undergrads take, it's called Econ V1, Introduction to Macroeconomics, [and] the last course, which is called something like Research Methods. I really like that because I suppose it keeps you honest. On the one hand, to the PhD students, you're teaching things at the frontier. You don't need to persuade them that it's important.

Hazell: They're interested in the Phillips curve. They're interested in inflation. They're interested in unemployment. They're interested in the cutting edge. That's relatively easy to explain. But, of course, as academics, we often get trapped in our own little bubbles. It's very easy to persuade other academics inside this bubble that that's important. For me, it's a tremendous source of discipline that, then, I go back, and I teach first year undergrads who, frankly, I have to work really hard to persuade that anything I understand is valuable.

Hazell: That's so important, because it would be easy, I think, for any of us, for any academic, to get trapped [on the] sidelines, in research that doesn't really matter. But when a student asks you, “Why should I care about this?” You need to have a ready answer, and you need to talk about things like the Great Depression, the current burst of inflation, the financial crisis, and you really need to be anchored to that kind of thing. So, I love teaching first year undergrads, completely different challenge.

Beckworth: That's great. Now, at the graduate level, what are grad students learning? What is the cutting edge? Are they learning like HANK models now? What would be in their curriculum?

Hazell: I take a particular view on what I want them to learn from me, because there are many branches in modern macroeconomics. HANK models, Heterogeneous Agent New Keynesian models, [are] obviously very big. My colleagues, Ben and Ricardo, have been seminal in founding that class of models. [It’s] not my comparative advantage. I'm not a genius coder or a mathematician. The stuff that gets me really excited is empirical work, and I like to think about maybe three classes of empirical work.

Hazell: The first is, maybe, what I call old-fashioned time series econometrics, looking at the aggregate plots of unemployment, inflation, interest rates, and so on. I think that's never the last word, but I think it's very important to have that often be the first word, because you want to be familiar with what's going on with inflation today or unemployment today, the aggregate patterns that we really need to understand well. I like to do some of that. The second thing, I like to go to the microdata and talk about causal identification.

Hazell: I was very influenced by this famous paper by one of my PhD advisors, this guy called Jonathan Parker, who was able to estimate the marginal propensity to consume. He was able to estimate how much of a windfall of cash consumers are likely to spend. He was able to estimate that using this amazing natural experiment, when the federal government, at different random times, gave money to different consumers in 2009. So, I'm very influenced by going to the microdata and using these clever natural experiments to figure out what's happening to consumer spending.

Hazell: Then, the third plank of what I try to teach the students is something in-between, so regional empirical macroeconomic approaches. There's a lot you can learn from the region that, I think, has advantages versus the aggregate or versus the microdata. Because, in regional information, you can hold fixed certain things, like the stance of aggregate monetary policy, but you can also maybe get a better understanding of mechanisms than you would in the microdata alone.

Hazell: Then, I'm really influenced by, for instance, the landmark work that my co-author, Atif Mian, did now about 10 years ago, trying to understand the sources of the Great Recession. He showed there that regional variation in house price booms and busts seem to be really important. That's the kind of thing that I think students need to know, and where the regional approach is really helpful. So, those are the three things: empirical work, but with aggregate time series, regional stuff, and then also microdata.

Beckworth: We'll come to a paper of yours shortly where you use those very skills, the one on Phillips curves. But, before we leave graduate macro and undergraduate macro education, what would be your advice to young students who are aspiring to become the next Jonathon Hazell?

Hazell: David, as you know, you shared this question with me before, and I thought very hard about it. I think it's difficult. In the end, I came up with three observations. The first is that, I think what we do, what you and I do, what academics do, it's an extremely difficult path. It's very lonely. It's very difficult. One needs a lot of faith in one's own research abilities, even if that faith might be misplaced. On the other hand, it has tremendous rewards.

Hazell: My first piece of advice is that I think one really has to make sure that you absolutely love the research for its own sake, because it's a very hard path with the tremendous rewards of doing what you love. But, if you don't love it, then it won't be worth it. And so, coming to one of your earlier questions, from a young age, I really was quite besotted with this subject. In the end, it was an easy choice for me. But, I think understanding that you really love it is a prerequisite to it being the right choice.

Hazell: The second is that if you do really feel like you have a passion for it, I think it's important to pursue it relentlessly. For me, in my 20s, and even now, trying to do a PhD entailed lots of sacrifices. I remember, when I was an undergrad, I was known amongst my friends and my broader peer group as someone who was not partying very hard, not going out very much, because I studied extremely hard. That was what it took and what it still takes, even more now than before, to go to a top PhD program, which is an important part of trying to become an economist.

Hazell: My third point, and this is, I think, the most important, is that if you really enjoy macroeconomics, I think it's important to recognize that there are many different ways to exercise that passion, because people are very different. To give you one example, one personal hero of mine has been for all of my adult life, Tyler Cowen, who blogs on Marginal Revolution as a colleague of yours. I'm not sure he would call himself a macroeconomist, but perhaps he is to some extent. One thing that's remarkable about him is, he's obviously forged a completely unique path to figuring out what it is that's going to make him a great intellectual.

Hazell: I think it's important for everyone to figure out what that is. For instance, for me, I'm pretty well suited to doing a PhD, but I think many people aren't, and here are many other ways to exercise being a great macroeconomist; finance, central banking, public intellectuals. And so, I think that my third piece of advice I'll give to a young, aspiring macroeconomist is to figure out what is exactly the way that you think you're best fit into that, because it might not be a PhD, it might be something else.

Beckworth: That's great advice. And I should say that Tyler Cowen is my colleague and my boss. He leads the Mercatus Center. We had him on recently. He had fun talking macro with me. We went back and evaluated what macro theories did well over the past few years. So he enjoys it, but he's definitely widely read, a unique path. I'm thankful for the fact that he brought me on at Mercatus too, so great points. Your last point, I think, is especially important. It's comparative advantage. It's what we teach in Econ 101, that you may not be Paul Krugman, but you can be somebody else, and you have some value add to offer. There's always opportunity for passionate people, like you've outlined.

Hazell: Absolutely. And one of the ironies is that even Paul Krugman ended up not being Paul Krugman. We know him, as academics, as this brilliant, sort of savant trade theorist. But in the end, he, at least, decided that an even better fit for him was not being a brilliant but nerdy trade theorist, and, instead, expounding macroeconomics, and then politics, to a wide audience. So, I think that figuring out what it is for you is absolutely essential. I guess that's part of the fun.

Beckworth: Absolutely. Alright, let's move on to your work. You alluded to this paper implicitly a few minutes ago, because it requires a lot of microdata, regional data. It's an influential paper that you helped co-author in The Quarterly Journal of Economics, and it's titled, *Slope of the Phillips Curve: Evidence from US States.* I think it's also fitting that you were a part of this, because you're at the London School of Economics. As your colleague, Ricardo Reis, reminded me, Bill Phillips was there. The Phillips curve has a rich history, the very building where you work, so very fascinating paper. It really shook up many priors of people, how steep is the Phillips curve? How do we interpret Paul Volcker's disinflation? You kind of shook things up there, Jonathon. And I just want to start, before we get into what you found and what you guys did, and most of our listeners will know this, but let's just start very basic. What is a Phillips curve?

The Basics of the Phillips Curve

Hazell: You say that most of our listeners will know this, but of course, there's a lot of confusion, even with me and my co-authors, as we think about this, about what exactly the Phillips curve is, or how we should define it. And as you said… so Ricardo was probably too humble to mention that he is, in fact, the A.W. Phillips Professor of Economics. So, indeed, Phillips is central to what we do here at the LSE. The Phillips curve, I think, properly stated, the “New Keynesian Phillips curve,” says, mathematically, that current inflation can be caused by three factors.

Hazell: The first is changes in slack, changes in what's sometimes called a forcing variable. Typically, this will be measured by something like unemployment, something like the output gap. The idea here is that when the economy is booming, when unemployment is very low, when output is very high relative to trend, when labor markets are very tight, then inflation is going to rise. So, that'll be the first thing, the first way in which inflation can be caused. The second is supply shocks. If there's a supply shock, like a surge in oil prices, maybe a bottleneck, maybe some kind of issue in reallocating labor to where it needs to go, that's the kind of thing that could also, perhaps, bid up wages, and therefore, bid up prices. The third factor could be inflation expectations. Inflation might be high today, because people expect that inflation is going to be very high in the future.

Hazell: This is this landmark idea, going back to Milton Friedman, that says, well, if we knew that inflation was going to be very high in the future, perhaps because the central bank had set the inflation target very high, perhaps because there was lots of money printing, that kind of thing, then people today would bake that into their expectations, and therefore demand higher prices today, meaning inflation is high today. So, that's it, in its basic form. Inflation, the Philips curve, or the New Keynesian Philips curve, is a structural relationship between, on the one hand, inflation as an outcome variable, and, on the other hand, is inflation expectations, slack, as often proxied by unemployment, and supply shocks. I think of that as the basics. Probably, you agree, maybe not.

Beckworth: Oh, yes, completely.

Hazell: A closely related thing that people often talk about loosely [is that] the Philips curve is something like the correlation between inflation and unemployment. You can see how that could be similar to what I'm calling the “New Keynesian Philips curve,” but it might not be the same, or it might be a little bit different. And, indeed, a lot of what people try and figure out is how this correlation between inflation and unemployment is related to the structural New Keynesian Philips curve.

Beckworth: I think people often will criticize that latter definition, which is just simply reduced form correlation between slack and inflation, which is a good rule of thumb to maybe start with. It's a good way to think things through. But what you outlined, the original definition, is a structural model, I like to think of it as a form of short-run aggregate supply, going back to Econ 101. It's kind of an inverted short-run aggregate supply curve. I think it's very intuitive. If there's slack, or if the economy's overheating, you think that would have some relationship to inflation.

Beckworth: If people think that the central bank is credible, and they've grounded inflation expectations, and finally, unexpected supply shocks. All of those things make a lot of sense. If you can imagine, listeners, you have this equation. You've got inflation on the left-hand side equal to these three terms. The thing that's really fascinating in this paper that you guys look at is, you really take to task the view of how important that output gap or slack term is. There's a little parameter in front of it, which measures the slope of the Phillips curve.

Beckworth: Again, the conventional view of the early 1980s, what Volcker did, was he had to enact really high unemployment to get inflation down. Now, you guys really give that a beating. You really push back on that. I want to mention something in your article that should have given us pause already. You mentioned some really fascinating work by Thomas Sargent in 1982, *The End of Four Big Inflations.* He looked at hyperinflation in Austria, Hungary, Germany, and Poland in the early mid-1920s. And he shows how those things suddenly dropped, and they were too big, too quick to be explained by slack or unemployment. That there, it was a clue, I think, all along, that should have stressed the importance of this. Walk us through that journey and what you guys uncovered, and how did you uncover it?

Pushing Back on the Conventional View of the Phillips Curve

Hazell: Absolutely. I guess I should say that the thing that's always top of my mind, of course, as should be top of yours and many of your listeners, is that inflation is very high today. I guess the way I'd like to tell the story of this paper is, first, to not mention high inflation of today, and think about this as a story of maybe 1978 to 2020. That's a story of the Volcker recession and what happened afterwards in the Great Recession and so on. Tell that story, which I think our paper did a pretty good job of explaining, and then say, okay, well, what about now?

Hazell: Can we take what we learned in pre-2020 data, and then talk about the high inflation since? Because, obviously, that's the key question at the moment. So, let me go back and tell the story of this paper first; one moment on how it came about. I have three fantastic co-authors for this paper: Emi Nakamura, Jón Steinsson, and Juan Herreño. I was absolutely blessed that when I was in my first year of my PhD, so now a long time ago, Emi Nakamura and Jón Steinsson, these two sort of young guns who were reforming macroeconomics and making it much more empirical, happened to be visiting MIT, and were guest lecturing my introductory macroeconomics course as a PhD student.

Hazell: I was very taken by this very empirical approach that they were working on, and so I offered to become their research assistant, and it ended up becoming this project, which is, I think, recognizably very similar to a lot of the empirical work that Emi and Jón do, but then is also what has become, I think, the kind of work that I like to do as well. That was very fortunate, as was finding our fourth co-author, Juan, who was another student of Emi and Jón's. That's the discovery, but then let me tell you about the paper.

Hazell: We're interested in this object, the Phillips curve. Then, exactly as you said, David, the episode in US economic history, and really around the world, that really influenced how we think about the slope of the Phillips curve, is the Volcker disinflation. In the early 1980s, inflation was very high. Paul Volcker sharply tightened monetary policy, unemployment rose sharply, there was this big recession, inflation fell sharply. One conventional interpretation of this is that there's a relatively steep Phillips curve. That is to say, when unemployment rises by a lot, then inflation falls by a lot.

Hazell: That's a decent way to understand the Volcker disinflation, potentially. However, if true, then the behavior of inflation after 1985 in the United States, and again, around the world, seemed rather puzzling, which was that during the 1990s, and in the sort of 2010s, unemployment became very low, but inflation did not take off. Likewise, during the Great Recession, unemployment became very high. It was almost 10 percentage points in the United States, and inflation did not fall by very much.

Hazell: From that standpoint, the Phillips curve's slope is steep. From the standpoint of this Volcker experience, the subsequent behavior, this "missing disinflation" or missing reinflation, seemed quite puzzling. And this was… In the 2010s, this is, I think, one of the major puzzles that we had, and it was a major debate. Our first observation was to try and revive an alternative interpretation of the facts that I told you, the facts of Volcker and then the missing movements of inflation that came afterwards, with a classic idea that actually went back to Ben Bernanke, but of course, much earlier, which was what he called the anchored inflation expectations hypothesis.

Hazell: His idea was to switch gears and focus on that other force that we said was really important, which was how inflation expectations affected inflation. His observation as applied to modern events would say, look, during the Volcker period, long-run inflation expectations fell very rapidly, perhaps because Volcker was saying, "Look, I'm a really serious, tough inflation fighter. I'm really going to lower inflation, come what may." Subsequently, after 1985 or so, the Fed gets credibility, inflation expectations are very stable, and so inflation doesn't move very much.

Hazell: So in the first half before '85, you have big movements of inflation, because there's lots of movements in inflation expectations due to the fact that the Fed is still finding its feet and bringing things down. Afterwards, inflation is stable because expectations are stable. Now, if that were the story, then things would make sense again. You could have a flat Phillips curve throughout, and inflation expectations are really mattering. You mentioned this amazing paper by Sargent in 1982.

Hazell: He was arguing, basically, that this should be the case on the basis of these very extreme hyperinflationary episodes, where countries like Austria and Germany, in the midst of the interwar period, were able to lower inflation very rapidly, without changes in unemployment, by really changing how central banks operated to make inflation expectations fall really rapidly. We were interested in pursuing this idea that the Phillips curve was flat and that inflation expectations were very anchored. The way that we chose to try and argue this is to use regional data.

Hazell: Why might regional data be very useful? The main reason why we thought regional data could be really helpful is that the inflation regime that I talked about, the long run inflation expectations that could be moving around a lot, this is something that's very hard to deal with. You go to aggregate data, you look at time series of inflation and inflation expectations, and it's going to be very hard to disentangle what's causing what, what's driving what. During the Volcker disinflation, you've got all of these oil price shocks, you've got rising unemployment, you've got falling long-run inflation expectations in the inflation regime.

Hazell: [It is] very difficult to separate these things. In the regional data, things are actually quite a lot more simple, we thought. The reason for that is, in regional data, every single state has the same central bank, has the same "inflation regime." Paul Volcker was setting the aggregate inflation target for all 50 states. That meant, we thought, that if you compare two different states— they've got the same long run inflation target— in some sense, you can control for the effects of this long run inflation target.

Hazell: If I take New York, and I compare it to Massachusetts, or Texas, and I compare it to Florida, Texas and Florida both have the same Paul Volcker, they both got the same fact, that long run inflation expectations are probably changing in the 1980s, because the inflation regime is changing. I can cancel that out and focus on other sources of variation. In particular, maybe now I can revisit this slope question. I can ask, look, if unemployment is very high in Florida relative to Texas, or very low in Florida relative to Texas, by how much does inflation change?

Hazell: That would answer this slack question. If slack is much higher in Texas than in Florida, by how much does inflation change in Texas versus Florida? That would answer this question of whether or not the Phillips curve is flat or not. That was really what we decided to do. Again, to gather where we are, we were saying, look, the sources of inflation, on the one hand, could be inflation expectations. On the other hand, it could be this slack term. If we look at regional data, we can just eliminate or cancel out the inflation expectations term and focus only on the slack term. That was the insight that we thought was useful.

Beckworth: Very clever identification issue, because that's always the issue in macro, how do you identify exogenous variation? So, a couple of questions before we get to the rest of the paper. One, did you have this intuition beforehand that inflation expectations were important? That's my first question. Then secondly, how did you construct the state data? That's another interesting question.

Constructing State Data and the Importance of Inflation Expectations

Hazell: Great. Absolutely. I'll take these questions one by one. How do we realize inflation expectations were important? Here, bearing in mind, as this project started, I was a graduate student in my early 20s. I really learned from my co-authors, and one of the things that they were amazing at was really assembling the consensus wisdom from the profession and having a sense of what people thought and, therefore, what we needed to understand to really change minds.

Hazell: The concrete example here was, as we were starting to figure out, “well, can a flat Phillips curve explain the data?” We thought, “okay, this seems pretty plausible.” Then, my co-authors— I remember Emi Nakamura— went around and asked a few people who she thought were very savvy and really understood the data and really understood the time series data well. In particular, I remember this clearly, she came back to all of us and said, “well, I had this great chat with this eminent economist who's called Martin Eichenbaum, who's at Northwestern.”

Hazell: And he said, "Well, I don't really believe your flat Phillips curve story because of Volcker." And we realized that Volcker was the key episode that we really need to understand if we thought that the Philips curve was flat. At some point, a while later— academia moves slowly— we were turning this around, and I was actually the one who remembered this Bernanke speech, this so-called anchored expectation hypothesis. And so everything is there somewhere, and then the magic that was absolutely my co-authors and not me was realizing that you could weave together the exciting rhetoric of these different economists with this cutting-edge econometric technique.

Hazell: But I can't take credit for that at all. Then the second thing, the data, this was also fascinating. It would be really surprising to many people to realize, to learn, that the United States does not have regional inflation measures. The United States, that it has-- you can see aggregate inflation, you can go onto FRED very easily and find it, but you can't really see regional inflation. There are some, I think, slightly problematic measures, at the city level, for a handful of large cities and some lower frequency annual or decadal measures at the state level.

Hazell: But what we have in mind when we think about inflation, which is a quarterly measure of how much prices are changing, does not exist separately for every state. And so one of the major contributions of this project was simply to construct that. Again, I can't take credit. My magnificent co-author, Juan Herreño, who's now at UC San Diego, spent literally years in the basement of the Bureau of Labor Statistics, painstakingly constructing these state-level price indices. Of course, there are so many steps there because, if you imagine, how do we construct aggregate inflation?

Hazell: We have millions or perhaps tens, even hundreds of millions of individual price quotes, which are collectively aggregated into aggregate inflation. Now imagine doing that, but now imagine doing it for every state. Now imagine doing that, but doing it in a windowless basement in Washington, D.C., using cobbled-together programs that were written by a BLS researcher 20 years prior. So, Juan is an amazing programmer and an amazing data scientist, I guess. And so he was able to do that, and in the end, the inflation indices work pretty well.

Hazell: And we're trying to work with the Bureau of Labor Statistics to make them continuously updated, so researchers can use them through to now. I suppose, then, another thing that I was really impressed to see to come together is, on the one hand, with this project, there are all of these high conceptual level ideas about the time series, about expectations, about slack. On the other hand, there's this nitty-gritty data assembly. Being able to unite both of those was a really remarkable achievement; again, not really mine. And I think it's something that I admire about my co-authors, is that they're able to, I guess, unite the high and the low, the conceptual big picture stuff in economics that we often discuss, but then also the details of assembling all of these price quotes.

Beckworth: So, you had the intuition based on previous research, Bernanke's inflation expectations story. You also did the data, hard work. Kudos to your co-author. Sounds like a miserable experience, but he did it. And so you put it all together. What did you find?

Summarizing the Empirical Results

Hazell: What did we find? Maybe, let me say it as a top-line finding, and then unfold how we found it. We discovered that the Philips curve— accounting for inflation expectations— the Philips curve is flat and has been flat going back to the 1980s, going back to this Volcker period. So, how did we find that? And this is a surprising finding. How did we find that? In the end, it was quite simple. First, we looked in the aggregate data about the co-movement between unemployment and inflation, very strong co-movement. Like I said, that's because of the Volcker period.

Hazell: But that co-movement was declining a lot over time. This is what we knew from the aggregate data. In the cross-sectional data, after you control for inflation expectations, which, like I said, is much more tractable to do in the cross-sectional regional data, the relationship between inflation and unemployment is quite weak and quite stable over time. That stability over time is very supportive of what I said about anchored inflation expectations and so on.

Hazell: Because in the cross-sectional data, once we've held fixed all of these issues about inflation expectations moving around, we find a very stable relationship. That suggests that all of this aggregate volatility and inflation expectations are driven by this big aggregate factor, things like the inflation regime, which is held fixed in the cross-section. So, that's the first thing, and then, how do we arrive at the flatness? Well, if after holding fixed inflation expectations, unemployment co-moves relatively weakly with inflation, then that means that, in some sense, the Philips curve slope is flat.

Hazell: So, in the sense that when unemployment falls by a lot, inflation doesn't rise by very much. So, that was the key finding that, from the regional data, the Phillips curve slope was flat, because after you accounted for inflation expectations, unemployment and inflation co-moved relatively weakly. Now, one thing I do want to say, and this is anticipating where I imagine our conversation will go, [is that] what we estimated was that the slope of the Philips curve was flat, but definitely positive.

Hazell: What I mean by that is that when unemployment falls, inflation does rise in the regional data. It doesn't rise by very much, but it does rise. So, supply curves, as we estimated them, still slope upwards, to give you an intuition from the start, David. They're just relatively flat supply curves. I suppose the final part of the paper is that we took this number that was estimated from the regional data, and we found that it did a pretty good job of explaining the aggregate data too, so explaining all of this relatively tranquil, moderate movement, the co-movement between inflation and unemployment, after 1985.

Beckworth: So, if you ask many macroeconomists, “How do you know if monetary policy matters?” They would use to say, "Well, Friedman and Schwartz and their book, and then Volcker. Those two episodes there show us that monetary policy matters." It sounds like what you guys have found is, you still validate that point, but for different reasons when it comes to Volcker. It was still the Fed, but it was the Fed creating credible inflation expectations, as opposed to using the unemployment to create the low inflation.

Hazell: That's exactly right. And I should say that the question that always keeps me up at night, as a good empirical monetary economist, is, how do we know that money affects inflation in the economy? And It's still so difficult to get high-quality evidence about this, but you're exactly spot on, David. What we find is that monetary policy has tremendous power to affect inflation, but to a large extent, it really seems to be through this expectations channel, as opposed to the direct effect of low unemployment causing high inflation.

Hazell: That said, if it takes very big changes in unemployment to increase inflation, that's still very much consistent with a world in which monetary policies can have very large effects. Because if you're a central bank, potentially, you can sustain quite large falls in unemployment before inflation starts to rise by too much. And so I think that a flat Phillips curve is something that would still be consistent with a pretty sort of "Keynesian view of the world," or even a monetarist view of the world. But it does suggest that the forward-looking inflation expectations [inaudible].

Beckworth: Is key, yes.

Hazell: Yes, exactly.

Beckworth: So, the story I believe you're telling in the paper is that Paul Volcker, fighting seriously against inflation, and then the Alan Greenspan Fed that followed him, that's really the important story here. If you were to construct this data set that you have, and let's say go back to the 1960s, and given that we don't think that monetary policy was as credible back then, do you think, or suspect, that the slope of the curve might be larger, or would it still be inflation expectations that are the key?

Hazell: That's a great question. We'd love to know the answers to that. We looked at this a bit, and we felt that in the aggregate time series data— which is all we have access to, because state-level data doesn't exist for this period— the aggregate time series data is quite consistent with inflation expectations being very important. For our podcast, I think it's probably easier to tell a narrative story of inflation expectations on the way up.

Hazell: The classic story in which inflation expectations are a very important determinant of rising inflation during the '60s and '70s goes something like the following: During the '60s, there's a big expansion in government spending because of the Great Society [and] because of the Vietnam War. This is Johnson and JFK, big expansion in government debt, strong pressure to keep interest rates low, nevertheless, so a large expansion in aggregate demand. This is something that continues to the 1970s, including extra political pressure on the Federal Reserve to keep interest rates low despite rising inflation.

Hazell: So, the classic example that people cite is when Nixon pressured Burns in the mid-1970s to keep interest rates low despite rising inflation. And all of this was the kind of thing that led actual inflation to rise, in part because people expected that the Federal Reserve would tolerate very high inflation. So, I think this is the classic narrative. I don't have much to add other than, say, that that seems to have a great degree of truth to me and seems to explain inflation pretty well and that sort of story goes all the way back to Milton Friedman.

Hazell: The one challenge that we encountered is, in a way that seems ridiculous, almost, to me as a modern observer, is that large periods of the 1970s had price controls. And so a lot of the behavior of inflation was pretty strange in the 1970s, because price controls interfere with inflation a lot, because price controls do seem to really work in the sense that when it's illegal to raise prices, prices don't rise by very much. Of course, they have all of these bad side effects, and when they're removed, inflation tends to rebound. So, inflation behaves quite strangely over the early part of the 1970s, probably because of price controls, but it's a little hard to say.

Beckworth: Interesting. So, both the lack of data, and even if you had the data, it'd be clouded by these price controls, very interesting. One last question about the Phillips curve, and then we'll move on to your other work. As you know, in this recent conversation about the inflation surge, and now we have this disinflation happening, there's been several points made, and maybe you can respond to them. One is that the Fed's credibility was important to the disinflation coming down. That's one point.

Beckworth: The other point is that some are invoking a nonlinear Phillips curve, so they would say, "Yes, we agree with Jonathon about the flat curve, but that's because it's nonlinear. It's flat to a point then, pop, it goes up.” And so, Gauti Eggertsson has a paper that he just completed, and I would mention one other paper because I had him on the show, [but] Joe Gagnon, Kristin Forbes and Chris Collins had a paper similarly arguing that there's nonlinear Phillips curves around the world. So, maybe speak to the recent experience and the role that you think inflation expectations played, and to what extent can we infer a nonlinear Phillips curve?

The Recent Inflation Experience and Nonlinear Phillips Curves

Hazell: Great. I've been thinking a lot about this, but I should say that my thoughts are very much in progress, [and] any of it could be revised. I think that the nonlinear Phillips curve is plausible, but I think that a flat Phillips curve story is still plausible. Let me try and unfold and explain why. The first point to make is that, during the most recent period, long-term inflation expectations did not rise by very much. So, during this earlier period, like the Volcker period and so on, long-term inflation expectations are changing a lot.

Hazell: That seems consistent with the Federal Reserve allowing the monetary regime to change, some "unanchoring of inflation expectations.” That absolutely did not happen this time around. It doesn't seem like inflation expectations were very unanchored. Of course, from the standpoint of our paper, at least the headline of our paper, the behavior of inflation during the current period seems to be a little bit puzzling, because inflation expectations didn't move very much, at least long-term inflation expectations, yet inflation rose very quickly.

Hazell: One natural response to that, so this is my second point, is to say that maybe the Phillips curve is nonlinear, which is to say that, in the range of data that one observes between 1978 and 2020, inflation tends to budge relatively little when unemployment falls. But, perhaps when unemployment gets extremely low, like we saw in the recent period, or other measures of labor markets slack get extremely tight, like we saw in the recent period, perhaps then, suddenly, inflation takes off. This is the idea of the nonlinear Phillips curve.

Hazell: It clearly has some intuitive truth to it, which is that if every single worker is employed, and you, David, are the final unemployed worker left, and every firm wants to hire you, one thinks, probably, that you could have a pretty good shot to bid up your wage as much as you want and then maybe that wage would pass through into rising prices. So, the nonlinear Phillips curve story is that that's the argument that we have in mind… so Gauti and Pierpaolo and others have made the case that it's there in the data, which I think is a very reasonable case.

Hazell: One difficult thing about the nonlinear Phillips curve, and any story that confronts the time series data, is that so many things are going on in the time series. At the same time, as you have the non-linearity that they wish to detect of a tight labor market and rising inflation, many other things are happening too, for instance, many supply shocks to oil, to labor market frictions, to bottlenecks, and so on. I think there could well be some truth to the nonlinear Phillips curve story. I'm not willing to give up on the flat Phillips curve story just yet.

Hazell: I'm still curious about investigating it, and that's for the following reason. Between the end of 2020 and roughly now, the United States underwent a gigantic and very persistent demand shock. Even with a flat Phillips curve, one might expect that very big demand shock to have large effects on inflation. And I think— just to not lose sight of this big demand shock— I think it's helpful to put some numbers on it. If you take the fiscal stimulus of the last quarter of December 2020 and the first quarter in March 2021, there was fiscal stimulus around 13% of US GDP.

Hazell: The CARES Act, which was in March 2020, added in fiscal stimulus of another 7% of GDP, I think. I might be getting the exact numbers wrong. But, if you take the fiscal stimulus of 2020 and 2021, you are looking at a fiscal stimulus of something in excess of 20% of GDP. This is extraordinary. This is unprecedented in peacetime. Start with that and then apply a multiplier to it. Imagine that the Federal Reserve, at least initially, doesn't respond by raising interest rates, which seems to be what happened in the data.

Hazell: Maybe you start to apply a multiplier like 1.5 or two to that. Now, you are ending up with, really, truly staggering numbers, like a total fiscal demand impulse of something like 30% to 40% of GDP, just bonkers numbers. Even with the flat Phillips curve that we estimated, you'd expect a very large response to inflation. I suppose that's why I, for one, am still curious about the flat Phillips curve, because I think it's easy to lose sight of just how big the demand shock was. Back in 2021, people like Blanchard and Summers were saying that inflation's really going to take off.

Hazell: By and large, they weren't invoking non-linearities, they were just saying, "Look, fiscal stimulus is so big. Fiscal stimulus is so big that responsive inflation might be big, too." Again, I might be wrong on the precise numbers, but I think it's a hypothesis worth investigating, that what was really going on was just a very big demand shock. Now, one challenge with the story of a flat Phillips curve and a big demand shock is the behavior of unemployment. Unemployment in the United States is roughly 3%. It was also roughly 3% in 2019, but, of course, in 2019, inflation was not very high.

Hazell: And so I think that if we're going to go down the big demand shock story, we do need some explanation for why unemployment wasn't incredibly low, because that's what you would need for this story to work. But one can think of reasons why, perhaps, unemployment has reached its rock bottom and slack was showing up elsewhere in the labor market, for instance by workers doing lots of job-to-job switching. So, I guess, to summarize, to come back to your original question, I think it's quite possible that a nonlinear Phillips curve could be what's going on.

Hazell: I think it's equally possible that you have a big demand shock along a flat slope. I think, also, that supply shocks would've played some role. I'm not sure how much. So, I would say that the jury's still out. I think one thing that I learned from this process is just that this is how difficult macroeconomics is. Here we are, after the fact, with all of our PhDs and our learning, and we're still really struggling. It gives me a deep sense of sympathy for, for instance, the current Fed. I might say, "Look, the Fed was very behind the curve in 2021, but you and I are only just now, perhaps, catching up with the curve in 2024." I guess that's my view, but it's difficult to know for sure. And it could be nonlinearity, but not sure just yet.

Beckworth: That is so fascinating. I hadn't thought of that before, but that's definitely a possibility, right? Given the huge size of the fiscal stimulus and support for monetary policy, you could have a real small parameter, real small number on that slope parameter, and still have sizable inflation. If you look at core inflation, 6%. If you look at headline, which is oil, take that out, that totally maps on. It could be a plausible interpretation of what happened.

Hazell: Sure. A different way to put it is, if I say to you that there's a fiscal stimulus of 20% of GDP, and I give you a multiplier that's greater than one, [is it] arguably a surprise that inflation only rose to 6%? Because, at the time, I had some sympathy for the Biden administration because, at the time, they thought, "Look, the worst mistake is to let a lingering recession carry on." They absolutely didn't let that happen, but perhaps they made a mistake in the other direction.

Beckworth: Yes. And again, I think you can weave a story where both inflation expectations maybe muted some of the otherwise big impact on inflation from the stimulus. Then, also, that Phillips curve slope played a role as well. It's so fascinating. I like to always bring on as a data point… Jonathon, as you know, I'm a big fan of nominal GDP targeting, even though we don't do it, but implicitly you can use it as a rule of thumb. If you look at the pre-pandemic trend path of nominal GDP compared to where it is today, it's about $2 trillion larger or above that path. That had to come from somewhere. That didn't miraculously appear, that's a policy choice. And I think we could argue, maybe that was necessary, maybe that was worth the trade-off, but the point is, where did it come from? How did it get there?

Hazell: The thing that I was surprised [about] from, now, the 15 years that I've spent being a macroeconomist is that I thought in 2020, as perhaps you did, that the real danger was a 2010 to 2015 style.

Beckworth: Oh, yes.

Hazell: It's just grindingly high unemployment. That, of course, didn't happen. The opposite happened, which was an absolutely booming economy at the cost of high inflation. What I've been very surprised by is how much less popular the current economy is than the economy of 2010 to 2015 in the United States. I would've thought that the current economy would be fantastic and everyone would be relatively happy despite the high inflation. I would've thought that all of the problems that people were complaining about for the last 12 years, low interest rates, high unemployment, had been completely solved, and that wasn't the case. So, one place where I really changed my priors was on how much people dislike inflation versus unemployment and, therefore, on the value of absolutely giant fiscal stimulus. I found that very surprising.

Beckworth: Likewise. I found it very fascinating that, in polls, you suddenly see the number one concern being this high inflation. There's an Ipsos poll that tracks it around the world, and, for a long time, you see inflation being more important than COVID, and racial issues, and all of these other things that you think are important. There's a study also done, I forget who did it, but they looked at Google searches for inflation, and they found this threshold effect, that people don't really look for inflation until it got around 3% to 4%, and suddenly, boom, there's this increase.

Beckworth: So, there's some point at which people really become cognizant and worry about it. So, that's a great observation, that politicians, policymakers need to be mindful that people care, and it could have a bearing on what they're able to do moving forward. Let's move forward to your other work. I want to spend some time on this notion of R-star, which is this idea of a real interest rate that brings— it's where desired investment, desired savings comes together. That's one way to look at it, or it could be viewed as the rate where you have price stability.

Beckworth: We have a number of measures of it already. We have the famous Laubach-Williams, or the Holston-Laubach-Williams measure, from the New York Fed. Theirs is constructed using structural time series data. There's some other ones that have come out, the Richmond Fed has a few. You could look at market estimates, look at the five-year, five-year forward, but you have come out with a novel measure, you and your coauthors, and it's from a paper titled, *Measuring the Natural Rate Using Natural Experiments.* Walk us through this paper and what you found.

*Measuring the Natural Rate Using Natural Experiments.*

Hazell: Great, thanks. Thanks a lot. This is a paper with Atif Mian, who's a professor at Princeton, and then Verónica Bäcker-Peral, who's a young researcher at Princeton, pre-PhD, but has been just absolutely magnificent to work with on this project. Okay, so, just to set the scene, so why do we care about this thing called the R-star? Most of your readers are probably familiar with it, but just to be clear, we think there's this idea, which I think goes back to Wicksell, which is what he calls the natural rate of interest. This is what clears the market for saving and investment while ensuring stable inflation and full employment. It's the interest rate at which the economy is at a steady equilibrium. It's very interesting, because movements in this natural rate are key for detecting structural shifts in the economy, such as whether or not the secular stagnation era has ended. One example that's always at the top of my mind is to think, "Okay, right now, interest rates are relatively high, asset prices are relatively low."

Hazell: But five years ago, during secular stagnation, the opposite was true. Asset prices were high, interest rates were low. Now, do we think that we're going to return to this world of low interest rates and high asset prices in the next few years, or are we now in a permanent world of high interest rates and low asset prices? Very difficult to know, and so that, I think, the key macro question for market participants: How would we know the answer? Well, suppose that we could observe this kind of phantasm object R-star, then we'd know what interest rate was naturally being gravitated towards. And then, we could figure out whether or not the economy is heading back to a low-interest rate world or not; crucial for central bankers, crucial for market participants, and so on. Of course, it's very difficult to measure. It's a long-run equilibrium of the economy once all of the transitory factors have subsided. And so, these influential, impressive papers by, for instance, Laubach-Williams, what they're trying to do is, they're trying to use clever econometric techniques to infer R-star, the long-run interest rate.

Hazell: Of course, it's very difficult, because to use these structural techniques, one has to be sure that the structure of the economy is correctly specified. And we, macroeconomists, know that we rarely understand the structure of the economy correctly, and so it's quite difficult to know what R-star is. In practice, this is going to lead to real issues. And so— I actually saw this tweet from you, David— different measures of R-star disagree now, or at least when you tweeted, by something like 200 or 300 basis points, like really big amounts. Some measures of R-star say we're in the low-interest rate world, some say we're in the high-interest rate world. And so there's a theoretical lure to R-star, but in practice, when we try to measure it, [it’s] really difficult to do so. And so, that one challenge is just that the point estimates disagree a lot between these different structural models. A second one would be nerdier, but I think equally important, which is that the standard errors associated with these estimates are giant.

Hazell: So, the last time I checked the Laubach and Williams measure, it spans something like the 95% confidence interval span, something like five or six percentage points of interest rates, really big amounts. Now, again, I don't mean to be mean-spirited. I think it's a crucial object to measure and these people like Laubach and Williams, and successors like Lubik and Matthes, but, you know, really breaking the frontier. What we wanted to do is see if we can come up with different measures to come out of that. So, that's the preamble, why we should care about R-star.

Hazell: We're going to take a different approach. We're going to try and measure the natural rate of return using natural experiments and microdata. We like this, because it seems to be less sensitive to specification issues. It seems to be more precise. But, on the flip side, we're going to be measuring a different object, an object that's specific to UK housing. In particular, we're going to be measuring the rate of return on UK housing in the long run. The object that we were talking about before, that you mentioned, R-star is something like the rate of return on government bonds in the United States in the long run.

Hazell: We're measuring the rate of return on housing in the UK in the long run. These two things might not be exactly the same. Why might they not be the same? Because, of course, the government bond is like a risk-free asset, whereas UK housing might have risks associated with it. We're measuring a different object, but we're going to have a really neat way to measure it. We're going to learn a lot of stuff. And, hopefully, that's going to shed some light on these questions. So, in particular, we're going to make use of this very strange feature of UK property, which is kind of intriguing.

Hazell: It will be surprising to US listeners, so I'll explain it briefly. Most apartments in the UK are sold as what's known as leaseholds. What are leaseholds? Leaseholds are long leases. When I was a graduate student, I rented once per year. What is a leasehold? Instead of renting once per year, you rent for something like 100 years or 200 years, and these leaseholds are freely traded. Moreover, leaseholders have the right to extend their lease, conditional on paying freehold as the value of a lease extension. And, we can observe the value of these lease extensions.

Hazell: Essentially, we can observe, what is the price of extending the lease for a property from something like a 90-year lease to an 180-year lease. The price of that extension basically says, how much do markets value property in the very, very far future. In other words, what is the long-run rate of return on property? That's the basic idea. We're going to use the fact that we can observe extensions for UK property to, say, what's the very long-run value of UK property? Related to this R-star thing, with R-star, we're interested in the very long-run value of government bonds.

Hazell: Here, we're going to measure the very long-run value of UK property instead. So, because we have all of this microdata, it turns out to be relatively tractable. We have something like 130,000 different lease extensions. It turns out that we can measure the long-run value of UK property very precisely. We can measure it at a point in time. We can measure it every quarter. And we can measure it without making many structural assumptions about the economy. Previously, I talked about R-star when I was trying to make lots of assumptions about the economy in order to measure R-star. Here, we basically observe, without making many assumptions, what the value of property is in the long-run. So, that's the key idea, using this peculiar feature of UK assets, UK housing, to measure the very long-run value of property, which is sort of a cousin… It's something somewhat adjacent to R-star, but it's a long-run discount rate of housing instead of government bonds.

Beckworth: What did you find? What's the trajectory of this real return?

Hazell: It's sort of remarkable. We find that the long-run rate of return on housing is about five percentage points in 2003, and it falls to about two percentage points by the end of 2023, so from five to two. These might not seem like big numbers, but these are actually very large numbers. It's the equivalent to a doubling of the price-rent ratio. It's quite similar to the amount by which R-star seems to have fallen, but it's also, we think, kind of remarkable that the long-run value of properties has basically doubled in the UK from 2003 to 2023. That's the first finding, a big fall in the long run rate of return of property, so quite similar to the magnitudes by which R-star has fallen. The second finding, which we think is particularly relevant to the current debates, is that the long-run rate of return on housing hasn't really risen since the start of 2023. It's risen, but by a very small amount. Now, what does that tell us?

Hazell: It comes back to the motivation that I told you at the start, which is that, exactly as you tweeted, there's a lot of discussion, a lot of debate right now about whether or not interest rates are going to return to that previous low level or if we're going to be in a higher one. Now, we have very long-run valuations without having to make many assumptions. Our very long-run valuations say, tentatively, that valuations are going to remain high, that interest rates and rates of return are going to remain low in the future. That seems like a useful thing for policymakers. We hope, we expect, that policymakers will carry on measuring R-star the ways they do, but, if everything works out and goes to plan with this paper, our measurement might also enter the pantheon of indicators that policymakers will measure, because it has these advantages of microdata and the precision of making relatively few assumptions.

Beckworth: You have a website online that updates and provides real-time estimates of it, correct?

Hazell: That's very kind… and with replication packet. Our hope is that, for people who are interested, for people who want to use this in their own work, they can download the measure, they can replicate it themselves, they can extend it. All of the data is publicly available, and the website is on my webpage, on Atif's webpage, and on Veronica's webpage.

Beckworth: We'll provide a link in the show notes as well. The only unfortunate thing about it is we don't have one for the US, right? We need--

Hazell: No. Not exactly, because the US does not have this peculiar [inaudible].

Beckworth: Yes, so it'd be great to have this real return on capital and another way to gauge where the trend is going. But, I guess, in general, then, this suggests that rates will stay low for some time going forward. That's consistent with my priors. Again, I'm speculating here, but I suspect that the same things that kept rates low before the pandemic, or with this, are more pronounced demographics, risk aversion, regulations, a number of things that would suggest continued low rates once we get to the final end of the pandemic bump.

Hazell: Yes, I think that's right, exactly. So, my priors are the same as yours, that we have all of these slow-moving structural factors that made rates low before and that haven't changed. The ones that I would list are the global savings glut, so another famous paper by Ben Bernanke that, around the world, many countries, and in particular China, were saving a lot and still are. A second factor would be demographics. That's only likely to get more extreme. A third factor would be the growth slowdown. That doesn't seem to have fully reversed around the world.

Hazell: To take the other side of it, to hedge, to be a two-handed economist, if I thought, okay, why would it be the case if that interest rates would return to permanently higher levels? I think the main reason for that would be a big acceleration in growth rates. Where could that come from? In the short run, the US does seem to be growing quite rapidly, much more rapidly than pre-COVID. You already know my views on that. I think that's probably something to do with fiscal stimulus. In the long run, fiscal stimulus will very likely peter out, of course. What could cause much higher growth rates in the long run?

Hazell: I'm quite bullish about AI. So, here, I've been reading a lot of blog posts by your colleague, Tyler, amongst others. I'm pretty bullish about AI. Did AI increase GDP growth rates by a couple of percentage points per year? Maybe. It seems conceivable, hard to know for sure. That's the kind of thing that I think would lead to a durable rise in interest rates. If I were a financial market participant, and I was allocating my bets, I would be allocating towards falling interest rates, but I'd have some hedge against a big spike, and I'd be monitoring the deployment of AI closely to see how quickly that's changing.

Beckworth: I hope that does happen. I hope we do have higher rates as a result of increased productivity growth from AI and other innovations moving forward. That'd be awesome, fantastic. Alright, one other topic that I want to touch on with you in the time we have left, and this relates to work you've done on downward wage rigidity, which is kind of a key tenet of modern macroeconomics. There's also some people [that] focus on output price rigidity, some focus on financial contracts as a sticky price as well. But downward wage rigidity is kind of key, it goes back to Keynes. And you have a really fascinating paper titled, *[Downward Rigidity in the Wage for New Hires].* Walk us through this briefly. I'll just tell you up front, it was fascinating to see that you can still have wage rigidity for employment relationships that haven't even formed yet, or are going to form. It's a very powerful insight, and it suggests that there really is something there to wage rigidity, downwardly.

The Downward Nature of Wage Rigidity

Hazell: I think so. So, let me back up. The bit that I find exciting is, almost, just for a second, recapping the classic point. The classic point, I believe, belongs to Keynes, but it may be earlier. You might know more history of economic thought than I do. Why does unemployment rise during recessions? Because of downward wage rigidity, just like you were saying. Wages don't fall during recessions, so unemployment rises because the cost of labor remains high, even as labor demand falls. Firms stop hiring, they start firing workers. So, that makes a lot of sense. You go to the data, wages rarely fall.

Hazell: I'll also weave in some economic history. Up until the 2000s, I think that this was basically viewed as a settled question, most of the ways you looked at it, nominal wages didn't fall very much. Downward wage rigidity seemed to be a very satisfactory explanation of why unemployment rose during recessions. At some point in the 2000s, I think that things were turned on their heads for very smart reasons by my colleague, Chris Pissarides. He pointed out that, actually, maybe a particularly important kind of wage is the wage for a newly hired worker.

Hazell: Why is that the case? Because when I'm a firm, if I'm deciding whether or not to hire a worker, what matters is not my incumbent workers who I've already hired, but the wage for a newly hired worker. Now, if that wage were flexible, if that wage fell a lot during recessions, then I might carry on hiring lots of workers even during recessions. If I'm hiring lots of workers even during recessions, that's really going to stabilize the rise in unemployment. Unemployment threatens to rise, newly hired workers say, "Look, I'm really happy to take a very low wage."

Hazell: Then, unemployment doesn't rise, because all of these workers are hired out of unemployment. So, Pissarides said, "Look, maybe what really matters is the wage for new hires. Moreover…" and this is the second point, "…by introspection," Pissarides said, "it seems really quite likely that wages for newly hired workers are, in fact, quite flexible. Most of the intuitions that we have for why wages should be downwardly weighted apply to continuing workers. As a continuing worker, I'd be angry as hell if LSE cut my wage in nominal terms.

Hazell: But, if I were a newly hired worker, LSE just offers me a contract. I'm just happy to get off the job market and become employed. I have no reference point. What does it even mean for the wage to be downwardly weighted? And so by introspection, Pissarides said, "Look, probably wages for newly hired workers are, in fact, quite flexible. This is what matters.” And so, in fact, the downward wages, to the hypothesis, isn't as compelling as before. And so I think this was one of the truly interesting insights that came out of the 2000s about theory and empirics.

Hazell: The final thing is, Pissarides went one step further and he surveyed work on the flexibility of the wage for new hires. He found that, when one looked at the data, average wages for new hires seem to be quite flexible. What do I mean by an average wage? Take the wage of all newly hired workers, hired at a point in time, compare it in booms versus busts. The average wage for new hires seems to be a lot higher during booms than busts. So, I think that this really overturned the apple cart. It's an amazing paper. It's in Econometrica in 2009.

Hazell: So, what to make of it? How do we think about downward wage rigidity afterwards? Is there downward wage rigidity? It turns out that there's a sort of empirical challenge to what Pissarides was doing, and that's to do with job composition. So, let me tell you, and slightly more concretely, what Pissarides finds and then why it's a bit more complicated. Imagine that there's an economy of high-wage bankers and low-wage baristas. Imagine, during booms, that, mostly, the economy's hiring bankers. During busts, mostly, the economy's hiring baristas.

Hazell: During booms, the average wage for new hires is going to be pretty high, it's mostly bankers. During busts, the average wage for new hires is going to be pretty low, it's going to be mostly baristas. The average wage for new hires looks pretty flexible, even if wages vary very little for bankers and baristas. Even if wages are rigid for bankers and rigid for baristas, so the wage for new hires is in fact rigid, this old Keynesian thing, the data that Pissarides was looking at, that's averaging across different kinds of jobs, averaging across bankers and baristas could, in fact, look quite flexible.

Hazell: The real difficulty then, I think, coming out of this debate, was that we didn't really have ways that we could control for the composition of who was being hired. The ideal here, in this example I've given you, is, you want to see, looking only at banker jobs, how does the banker wage vary for new hires? Looking only at barista jobs, how does the barista wage vary for new hires? One doesn't want to pool between them, because then the shift in composition between bankers and baristas can mess things up.

Hazell: So, that was the starting observation from my paper, which is that what one really wants is, one wants a measure of wage rigidity for new hires that can correct for job composition. So, what we did, me and my co-author, Bledi Taska… so Bledi Taska worked, then, at a company called Burning Glass Technologies, which scrapes online vacancies on the internet. So, most hiring now, in the United States, is done online, or has been since about 2012 or so; a subset, not a large subset, so, you know it's an important caveat to the paper, a subset of like 5% of vacancies post-wages.

Hazell: But, what is very useful is that these vacancies, not only post-wages, have very detailed descriptions of the kinds of jobs that these vacancies are doing. In particular, you can see the firm, you can see the job, you can see the occupation, and so on. So, imagine a physical location of Starbucks in Cambridge, Massachusetts, that regularly posts vacancies for baristas and pays them an hourly wage. We can track all of that information. We can see the hourly wage for baristas across multiple vacancies posted by the Starbucks. We can see, sort of, how the wage changes for this specific job.

Hazell: That means that we can purge for job composition. We can just look for every job in the economy, every job in our data, the wages rise, the wages fall for new hires across successive new vacancies. That was why we liked the data, because one can look at job level wage changes for new hires, for new vacancy postings, without this challenge of composition. In the end, the finding is very simple, which is that, surprisingly, wages for new hires are, in fact, quite downwardly rigid, though flexible upwards.

Hazell: So, across successive barista vacancies in that Starbucks in Cambridge, the wage will almost never fall. It'll rise sometimes, but it'll almost never fall. During expansions, when unemployment falls by a lot, the wage will rise by a lot. But during contractions, when unemployment rises, the wage won't fall by very much. Just as Keynes originally conceived, but now we have it for a measure of the new hire wage for the wage posting vacancies. 

Beckworth: That is a really fascinating finding, counterintuitive. My colleague and boss, Tyler Cowen, thinks that this is probably one of the most important findings in recent years, in terms of this debate about sticky downward wages. What's the story behind that? Why would it be the case that new hire wages are downwardly rigid? Because, again, you might think intuitively, like you just said, like you being hired at LSE. You'd have no reference point, right? So, what do you think is happening in the background that explains this?

Hazell: Absolutely. And I should say that Tyler, when I was a job student, Tyler was very kind. He blogged about my paper. I've always been a huge fan of Mercatus and Marginal Revolution, but it was— so for that reason, it was a very proud moment for me. Here's what, I think, is likely the story, and then I will, maybe, say a bit of evidence about it from another paper I have. I think, probably, it's to do with internal equity, as originally conceived of in this very famous book by this Yale professor, Truman Bewley, which is called, Why Don't Wages Fall During Recessions? It's an amazing book. Bewley imagines the following, that, I think, probably applies to my analysis. He says, "Look, the first logic, the simple logic, that you might have, is that wages for new hires would fall during recessions,” exactly because of the LSE professor example. I become a new professor, I'm willing to accept a lower wage, because I have no reference point.

Hazell: Then Bewley says, "Not so fast." What about this idea that he calls internal equity? Internal equity works like the following. I arrive at LSE, and I go around the hallways, and I say, "I just got hired, great job, I'm on $10 an hour.” Then, my colleague, who's the same rank as me, he's not a tenured, chaired professor, he's just another assistant professor who got hired just the year before, he says, “Well, you're only being paid $10 an hour, I'm being paid $20 an hour, they've screwed you.”

Hazell: I'd be furious, immediately. I'd be marching into the chair's office, I'd say, “What do you mean? We're doing the same job. He's being paid so much more than I am.” That's the internal equity idea of Bewley, which says that there might be various reasons why firms are constrained to pay the same wage for new hires and continuing workers. One reason might be the fairness story that I just told, which is that it's very difficult to pay otherwise similar workers different wages, because the worker who's paid worse might revolt, essentially.

Hazell: One can imagine other stories, too. So, for instance, there might be managerial frictions. Imagine, now, that we're not at LSE, but we're part of some big bureaucracy, a very large multinational company, or whatever. They might just have a pay scale. They might just pay all janitors, or managers of a certain rank, the same wage. It might just not be worth their while to differentiate between different managers. So, I think something like that is probably going on. I have a paper called, *National Wage Setting,* that's with Heather Sarsons, also with Bledi Taska, again, and with Christina Patterson.

Hazell: Heather and Christina are fantastic assistant professors at Chicago Booth. And both of them, with me, what we found was that, using the same data, firms tend to pay very similar wages for the same job, even in different geographic regions. So, for instance, Starbucks, for baristas, tends to pay the same wage in Cambridge as it does in New York, as it does in San Francisco. And so that's quite consistent with firms having a relatively limited ability to differentiate wages between observably different workers of the same position, and that would be consistent with this internal equity story that Bewley was telling.

Beckworth: So interesting. I could continue on with this conversation for some time, but we are now at the end of the program. I want to thank you, Jonathon, for coming on the show. Our guest today has been Jonathon Hazell. Jonathon, thank you again for coming on the program.

Hazell: David, thank you. It's been great to listen to the podcast episodes, to read your blog over the years. So, i'm looking forward to learning even more from you in the future. Thanks a lot.