Ben Moll on the Basics of HANK Models and How They Can Be Applied to Policymaking
Heterogeneous Agent New Keynesian models are becoming increasingly influential in the macro sphere, and this development could have major implications for future policy.
Ben Moll is a professor of economics at the London School of Economics, and is well known for his work on income and wealth distribution in macroeconomics and its implications for policy. Ben joins the show today to talk about this work and provide a look into the growing field of heterogeneous agent models. David and Ben also discuss the history of macro thought, the implications of different transmission mechanisms of monetary policy, and what HANK models mean for forward guidance and other more general makeup policies.
David Beckworth: Our guest today is Ben Moll. Ben is a professor of economics at the London School of Economics and is well known for his work on income and wealth distribution in macroeconomics, and its implications for policy. Ben joins us today to discuss this work and provides a look into the growing field of heterogeneous agent models. Ben, welcome to the show.
Ben Moll: Thanks, David. It's a great pleasure to be here.
Beckworth: It's great to have you on. I'm looking forward to this conversation because number one, we're going to talk about the cutting edge of modern macroeconomics. So, if you're doing cutting edge work, it's what you're doing, Ben. So, it's great to have you on. And number two, this is a great conversation because it's a continuation of what we were discussing with Eric Sims. In fact, our Eric Sims show kind of prompted us into getting together and organizing this conversation, so I'm glad to have you on. There's lots of listeners who enjoy this. In fact, I'll just mention this, Ben. The Eric Sims show was one of our best listened to shows in the past few months. So, this will probably be as well. People really do have an appetite for this technical discussion, technical material. And more importantly, it's policy relevant as well. It's not just about the models. It's about what it means for the real world and for policy.
Beckworth: Now, before we get into all the technical material and policy discussion, I'd like to hear about how you got into economics, and then, into macro specifically.
Moll: Yeah. For sure. So, I'd say overall it was a combination of some broad overall direction, mostly due to the fact that I was fascinated by the questions economics is concerned with. But also, together with a lot of randomness, I guess in retrospect, luck. So, I'm from Germany. I had one economics course in high school covering a bit of everything at a pretty basic level. And I was fascinated by a lot of the topics we covered like globalization. Why some countries are so much poorer than others and so forth. And also liked that it involved these graphs and you had to shift around these curves. In contrast, the other courses with interesting topics, like history and politics, you had to memorize all these dates instead of the curve shifting. I was terrible at that.
Moll: At the end of high school, I wanted to learn English and studied abroad, so I applied for universities in the UK. US seemed a bit far from home at the time. Then the question was what to study. Economics had been interesting. One thing I should say actually is the way it works in the UK in terms of universities is a bit different from how it works in the US. You have to decide at age 17 when you apply to university what you're going to study. Not like in the US where you decide sometime in your junior year what your degree is going to be.
Moll: So, anyway, I ended up at University College London in the degree in economics. I remember going to my first lecture and being totally shocked at economics [involving] any math and statistics. Why was this suddenly all this calculus and so on? But I got lucky in the sense that math actually suited me quite well, so I actually liked this approach. There were a lot of other people in my course who I think had equally little clue what was awaiting them. And a lot of them dropped out, but I got lucky.
Moll: And then, more generally, I think how I got into macro was also a little bit random. When I finished UCL, I applied to graduate school. So, I was pretty sure that I wanted to do that, but I was also 100% sure that I wanted to do structural micro econometrics because that was what UCL was very good at at the time. But then I ended up at Chicago for grad school and very much like the macro classes my first year ... Even then in my second year, I specialized labor, trade and development. And I really got into macro sometime later from the development angle and trying to understand across country income differences. Why some countries are so much more poorer than others and so on. So, it was a bit of a haphazard way with some detours, I would say.
Beckworth:Yeah, you've had a very interesting journey. You've been at Princeton. You're at London School of Economics. I know you've worked closely with Greg Kaplan at the University of Chicago, this heterogeneous agent model work. And I'm wondering how did you get into that? How did you get into these models? We'll get into what the models are in more detail in a few minutes, but how did you kind of become one of the leading experts in this field?
Moll: Right. So, I mean, actually as I said, I got started with this coming from the development angle. So, in particular, I was trying to understand why some countries are so much poorer than others. I wrote my thesis at Chicago about how much of differences in GDP per capita across countries can be explained by financial development. So, the idea is that maybe one of the reasons why underdeveloped countries are poor is because they have underdeveloped financial markets. And then, if you want to think about financial markets, you're going to have to think about heterogeneity. And particularly there's some good entrepreneurs, some bad entrepreneurs. What you want to achieve is that capital is allocated from the bad entrepreneurs to the good entrepreneurs. That would then lead to higher GDP per capita in the economy.
Moll: So, I'd written as part of my thesis the theoretical theory of this, the theoretical model of this. In that theory, heterogeneity was all front and center. And then, I only moved over to the business cycle macro part that we're going to talk about later, I guess, mostly later on. In particular when I was in Princeton, I was colleagues with Greg. And we had been chatting about macrostabilization policy and how that would interact with heterogeneity and that's how we got started.
Beckworth: Maybe we should begin by asking what is heterogeneous agent model macroeconomics?
The Basics of Heterogeneous Agent Model Macroeconomics
Moll: Absolutely. So, I'd say that heterogeneous agent macro is simply a way of thinking about classic macroeconomic questions. What's the effect of monitoring fiscal policy, what explains long run growth and so on in terms of distributions of micro variables, as opposed to how people traditionally think about these questions, namely in terms of aggregate only. To me that's attractive for two reasons. The first one is conceptually it provides an integrated approach to macro and distribution or inequality. And maybe even more importantly, empirically, it provides an integrated approach to micro and macro data. And there's simply a lot more micro data than macro data, I guess, and the hope is that maybe you can use that to improve our understanding of the macroeconomy.
I'd say that heterogeneous agent macro is simply a way of thinking about classic macroeconomic questions. What's the effect of monitoring fiscal policy, what explains long run growth and so on in terms of distributions of micro variables, as opposed to how people traditionally think about these questions, namely in terms of aggregate only.
Beckworth: Yeah. It also provides a lot of answers to questions or debates out there. I'll just give an example. Yesterday I was watching the press conference after the FOMC meeting, you know, Jay Powell gets up and he answers questions from the media. And the last question that was asked was about the poor old savers getting harmed by low interest rates. One of the things that you get from this literature that you worked on is there are these multiple channels, and just to focus on the savers losing interest income is a very narrow incomplete view, which is understandable maybe given what we typically teach or have taught in the past. But kind of a general equilibrium effects of this. There's a whole lot more going on. We'll get to these later. And that question is just not well founded. It assumes a very narrow dimension. So, there's lots of interesting insights that fall out of this. And so, I'm excited to have you on the show today.
Beckworth: Where does heterogeneous agent macro fall in the division between long run and short run macro? So, long run, we typically think of longer economy growth, as you mentioned earlier, why are some countries richer than others. And then, short run is business cycles. Is it a part of one of those or is it its own unique, distinct field? How would you classify it?
How to Classify Heterogeneous Agent Macro
Moll: Yeah. Absolutely. So, I think it's more, just like another dimension. So, you could think of a 2x2 table with short run and long run in one dimension. The other dimension is representative agent macro. So, macro without heterogeneity versus macro with heterogeneity and distribution. Then there's interesting stuff in the four entries of that table. So, you can apply heterogeneous agent macro tools or I like to sometimes call it distributional macro tools to both the study of short run and long run issues.
Beckworth: Okay. I like it. That's a neat little matrix there. That's nice way of framing that.
Beckworth: So, let's talk about the history of this because this is not new. In preparing for the show, I read an interesting article you sent me by Beatrice Cherrier about the history of heterogeneous agent macro. And apparently, it goes back quite a bit. In fact, lot of work at the University of Minnesota, Neil Wallace, Edward Prescott, some of their students. I would also mention David Andolfatto of the St. Louis Fed. On Twitter we're getting these exchanges a lot. I mentioned HANK. That's HANK with H-A-N-K, new Keynesian and he'll mention H-A-N-C. He's thinking the overlapping generation model.
Beckworth: So, there's been work done before on this, but you've taken in a new direction. Maybe a very prominent direction. And you have your own history of it. So, Beatrice Cherrier has her history of this, but tell us your history of it. You have a three generation history of macro thought and you're able to place your work within that division. So, tell us how you see the history of macro thought.
The History of Macro Thought
Moll: Yeah. Absolutely. But first let me say that I very much recommend that article by Beatrice who's a historian of economic thought and she really did a very nice job of looking into the history of this. And it's a good counterpoint to an argument to an argument you sometimes encounter, in particular in people making arguments against macroeconomics in the media and so on, which is that all of macroeconomics is this sort of wildly unrealistic representative agent macro. And what you can see very nicely if you read Beatrice's article, and I mean that's something that I think every macroeconomist has been saying for a long time is that's just wildly inaccurate. That's just not true. Heterogeneous agent macro has been around for a long time. So, I very much recommend that article.
Moll: So, yeah, okay. Let me tell you my history of thought of the role of distribution in macroeconomics. As you said, David, I find it useful to categorize macroeconomic theories into of four time periods. So, there's before modern macro time period, I would say, which is from 1930 to the 1970s. The 1930s are the birth of macro I would say with Keynes and so on. Then there's what sometimes I call the first generation of modern macro from the '70s to the '90s. Then there's a second generation of modern macro from the '90s to the financial crisis. And then, there's the third generation of modern macro, which is after the financial crisis.
Moll: I should say upfront that with all the times to fit some really rich literature into this very simplistic narrative is not going to work perfectly, but I'll point out a little bit where it fails. One thing that's interesting to think about is what are the drivers of this evolution. To why have heterogeneous agent models become more prominent, I would argue that it's three things. One is better data. The second one is better computers and algorithms. And the third one is current events, so in particular rising inequality and financial crises. Things that are hard to think about without thinking about heterogeneity.
To why have heterogeneous agent models become more prominent, I would argue that it's three things. One is better data. The second one is better computers and algorithms. And the third one is current events, so in particular rising inequality and financial crises.
Moll: But anyway, the story I'd say is something like the following. So, from the '30s to the '70s, in the era before modern macro, there's [inaudible] for example, which is if you'd ever looked at this, it's about aggregates. There's no role for inequality of distribution by design. Distribution does play some role in growth theory. In particular, there's this work by Kaldor and Pasinetti, the Cambridge UK theorists, that's concerned with factor income distribution. So, capitalist versus workers. But not personal income distribution. There's very little work on personal income distribution. And then there's some disconnected empirical work, so Kuznets Curve and so on.
Moll: Then at some point, for some reason that I don't want to get into too much here, modern macro comes along in the '70s. And people start driving these micro founded dynamic theories of the macro economy. The first generation of this series, they're basically mostly all representative agent models. The typical one I'm thinking about here is the representative agent RBC model. Again, obviously in the representative agent model there's no role for inequality of distribution by design.
Moll: What that also means is that you cannot talk to a number of important empirical facts. For example, over the last 20 or 30 years we've seen the economy growing as a whole, but this growth has arguably been unequally distributed and maybe also poor people tend to be hit harder in recessions and so on. So you can't talk about that with representative agent obviously. And also, it's kind of hard to think about welfare, so things like who gains and who loses in response to a monetary policy or other policy.
Moll: So, then people pointed this out pretty quickly. Then sometime in the '90s, and this is also what's nicely described in Beatrice's article, people started building these heterogeneous agent models. In particular, they started to incorporate macro heterogeneity, particularly in income and wealth into these models. That's the early heterogeneous agent models. There's people, as you said, particularly in Minnesota, like Rao Aiyagari, Truman Bewley at Yale, Mark Huggett, Ayse Imrohoroglu, Tony Smith, Per Krusell, my colleague here at LSE, Wouter den Haan and so on. What they do is they built these heterogeneous agent models where, as I said, before you represent the economy with a distribution of these micro variables.
Moll: That's kind of a huge step forward, but one thing that's interesting is that in these early heterogeneous agent models with this second generation, if you want, of theories of distribution in macro, the typical finding is actually that heterogeneity doesn't matter all that much for macro aggregates. And the well-known articulation of this is a result of Krusell and Smith. They call this approximate aggregation. The reason for this is that in these baseline models, there's a lot of linearity. Meaning, in particular that sort of rich people are just scaled version of poor people in these models. And so, in particular as a result inequality doesn't matter much for macro. But also, you can see that this linearity is a knife edge thing.
Moll: There's an interesting thing in terms of the history of this, which is actually at the same time in the literature there's some more nuanced cautionary results. In particular, even in the Krusell Smith paper itself, they have a section for, somewhere a little bit further back in the paper, with an extension where heterogeneity does matter for the behavior of the macro economy. But I don't know for some reason, this gets lost and it's fair to say that around the time the economist's perception is that inequality and heterogeneity doesn't matter all that much for macro.
Moll: The problem with this is of course that in the data, rich people are not just scaled versions of poor people. For example, they have lower marginal propensity to consume out of transitory income changes and so on.
Moll: Then more recently, in particular after the financial crisis, in the late 2000s I would say, a new generation of theories came along. And I would say that defining future of these newer heterogeneous agent theories is that they take micro data much more seriously than the early theories. This leads them to emphasize things like household balance sheets, credit constraints, marginal propensities to consume that are high on average but heterogeneous, non-homostaticities and non-convexities. So, in short, things that make you move away from this knife edge case where heterogeneity doesn't matter. So, the typical finding then is that distribution does matter for macro in these more recent theories. This is I think also where a lot of my own work falls.
Moll: So, again, it's like these three generations in modern macro. And the first one is just representative agent. So, there there is no heterogeneity. So, obviously you can't even think about it. Then in the second generation, so from the '90s to the financial crisis, it's at least in most cases or at least that's the perception, it's just a one way relationship that the macro economy affects distribution but not the other way round. And then, more recently people started thinking about this very rich interaction between heterogeneity or inequality on the one hand, and macro on the other hand. So, that's how I would summarize it.
Beckworth: Yeah. It's very interesting. As you mentioned earlier, this affects both long run macro issues as well as short run. We tend to focus on the short run issues on the show, but obviously both are important. But I did want to get later into the implications for monetary policy and some of the discussions they're having today about what they should be doing. But I want to come back to this question of why this surging interest now. Now, you mentioned, there's better data, micro data. Computationally is a whole lot easier to do these models. In fact these models, as I read are pretty challenging to implement. There's different techniques. Also, the interest from the great financial crisis has stoked the desire to learn more about this issue. And you have been a part of that conversation and work that's been done. I was reading that you now have a class where you go around and you educate central bankers on how to use these models. So, why don't you tell us about that?
Educating Central Bankers on Heterogeneous Modelling
Moll: Right. So, yeah. That's exactly right. Yeah, thanks for pointing it out. So, yes. In the summer, Greg Kaplan at the University of Chicago and I teach a master class for central bankers and also economists at other policy institutions, say the Treasury, about monetary and fiscal policy with heterogeneity. That's the name. We're doing it again this year and it'll be in August in Chicago at the Becker Friedman Institute at the University of Chicago. So, the background on this is that Greg and I together with my former Princeton colleague, Gianluca Violante, we'd written this paper that we are going to talk about later, I guess. *Monetary Policy According to HANK*. And HANK, just to be clear again, as David I think already said, is an acronym that stands for heterogeneous agent new Keynesian models. So, H-A-N-K.
Moll: This work on heterogeneity and that paper about how heterogeneity affects monetary policy by us and many others also, has received a fair amount of interest in the central banking community. And so, Greg, Gianluca and I have been getting a lot of invitations from central banks around the world that basically said, "We have a lot of people in our research departments that would love to know more about how can we start incorporating heterogeneity into the type of models we work with. Can you come over and come by and tell us about it, both in terms of the tools, and the policy implication?" So, we've been doing this for a while and traveled around the world a lot to different central banks.
Moll: We've developed this new mini course, but then, last year Greg and I figured how about instead we bring all these interested people in one place. Where in particular the idea was that maybe these people, who also were interested in learning about heterogeneity in macro, they would then be able to get to know each other and interact with each other and exchange what they know and so on. So, for the first time in August last year, we had this master class in Chicago with about 20 participants and at least Greg and I thought it was really great. As far as we could tell, I guess, the participants seemed to agree. And we had classes during the day, both theory and very hands on stuff like computations and discussions about policy. And then, during the night we usually had some sort of social activity. And so, it was four days full on all about macro with heterogeneity.
Moll: And so, given the positive feedback we got last year, we're definitely going to do it again this coming August. We hope we can again attract a fun and interesting and broad crowd like last year. And we have a website where you can find more details if you're interested. Hosted by the Becker Friedman Institute, there's all the information there. There's a little video. That will give you a flavor of what we do there. I think one can find it by just googling, for example, BFI for Becker Friedman Institute, heterogeneity, and then there's also the information there.
Beckworth: Okay. So, Ben, you have a chorus. This is a growing field. Let's actually talk about the HANK model. Let's jump into the weeds here. And I want to use as a launching point your article that you mentioned, *Monetary Policy According to HANK*. As you mentioned, HANK is this short hand. This acronym for heterogeneous agent new Keynesian model. As opposed to RANK. So, there's HANK, there's RANK. RANK is the representative agent new Keynesian model, which is kind of the poster child for all the critics to pick on, but has kind of been the workhorse model because there's a three equation result that falls out of this. It's easy to think through and it's used in policy circles a lot. But you have HANK and maybe eventually we can talk about TANK, which is what I talked with Eric Sims about, which is the two agent new Keynesian.
Beckworth: But let's get into HANK because there's a lot of interesting details and implications that flow out of it. So, let's work our way through. And maybe as a starting point, why don't we start by talking about how HANK is different in terms of the households that you bring into the model.
HANK and RANK Differences in Terms of Households
Moll: Right. So, I mean, at a very basic level relative to RANK obviously the big difference in HANK is that the households are heterogeneous and we try to bring to the table all that we've learned in the last 20 or 30 years on heterogeneous agent modeling, and also what we know empirically in terms of household consumption and saving behavior and household balance sheets and so on. So, the goal in that article was to basically reassess the monetary transmission mechanism to household consumption once we take on board all the richness in micro consumption behavior and household balance sheets that we see in the micro data.
The big difference in HANK is that the households are heterogeneous and we try to bring to the table all that we've learned in the last 20 or 30 years on heterogeneous agent modeling, and also what we know empirically in terms of household consumption and saving behavior and household balance sheets and so on.
Moll: So, one way of thinking about it, I guess, is you can think about it as of the standard representative agent new Keynesian three equation model. We take one of these equations, so in particular this IS curve or early equation. We throw that out and we replace that with this very rich consumption side of the economy. The other two equations are still the same, but how households respond to interest rate and income changes in particular, that we base on this much better empirically founded theory of consumption behavior.
Moll: I should say just to be clear that we're definitely not the first to develop a framework like this where you combine heterogeneous agent on the new Keynesian side, but I would say we did it really in a state of the art way. And then, it gives you the starkest results.
Beckworth: So, Ben, you have these different households and you differentiate them across different dimensions, I think liquidity. But aren't there three households and how are they different?
Moll: Yeah. I mean, so that's right. We do have these different types of households at the end, but one thing that's very important is that in contrast to ... in particular, I say the TANK models that you've talked about with Eric, where these different types are in the TANK model, these different types are just exogenously imposed. So, some people in the population are spenders and some people are savers. In our, I would argue, more realistic HANK models, these different types arise endogenously. So, in particular the way it works is just ex ante in our theory, people are actually mostly the same. They differ maybe in terms of the income type whether they're more educated or not, I guess, is one way of thinking about it.
Moll: But then what arises is that endogenously you get these different types. So, in particular, some people because they get lucky over time and they're well educated, they've high permanent income types, and they keep their jobs for a long time, they accumulate a lot of liquid wealth. Then at some point they buy a house and so on. And so, then they're wealthy. In contrast, other people, they get unlucky. They get a sequence of bad income realizations. For example, they get laid off and have to live off unemployment benefits. And then, those guys are the ones who are going to end up with low liquid wealth.
Moll: So, then what you end up having and what we emphasize is these three types of households, as you said, which are what we call the poor hand-to-mouth, the wealthy hand-to-mouth, and the non-hand-to-mouth. And hand-to-mouth here just means having low liquid wealth. Poor hand-to-mouth means having both low liquid wealth and low illiquid wealth. So, liquid wealth would be things like just the balance of your checking account. Illiquid wealth would be whether you have a house or in particular a 401k retirement account. And so, you have these poor hand-to-mouth people who have low liquid wealth and low illiquid wealth. You have these rich hand-to-mouth people who have low liquid wealth and at the same time though high potentially illiquid wealth. So, they may have a house even though they have a low balance in their checking account.
Moll: But again, just to be clear, the key difference to the TANK view that, say, Eric and these other people have is that these different types arise endogenously. So, it's not like you have it written on your forehead when you're born which type you are. Sometimes you're one type and sometimes you're another type.
Beckworth: Yeah. And maybe to summarize this, you have the three households. You have the poor hand-to-mouth, the wealthy hand-to-mouth, and then, kind of ... I'll call this the wealthy, but you call it the non-hand-to-mouth. But the RANK model, kind of the workhorse new Keynesian model only has that last group, the one that's not hand-to-mouth or the wealthy one. And my understanding is that group, that one group is the household that optimizes intertemporally. So, they literally think about the future, the present, they're responding to changes in interest rates. Whereas those other two households are much more ... they're living paycheck to paycheck. They're not thinking about optimizing their portfolios based on some small interest rate change. So, they're much more, what you would call I guess, the marginal propensity to consume households.
Beckworth: And all of this has very big implications for the transmission mechanisms of monetary policy. And so, why don't we talk about those? You have a chart with a bunch of channels that illustrates how these different households respond differently to shocks to monetary policy. So, maybe we can walk through some of these channels and discuss why it's important to know that these different transmission mechanism channels exist.
Transmission Mechanisms Channels of Monetary Policy and Their Implications
Moll: Right. Exactly. So, yeah. Let's take as a starting point maybe exactly the representative agent new Keynesian model and how monetary policy works there, which was also really the starting point of our monetary policy according to HANK paper. The way it works there is that ... so when the central bank cuts interest rates consumption goes up, and you can decompose that into what we call direct effects and what we call indirect effects. So, the direct effects would be interest rates are lower, therefore it's a bad time to save. Good time to borrow. And therefore you save less and borrow more and consume more.
Moll: The indirect effects are what happens in general equilibrium in particular because people now increase their consumption. They demand more. Therefore, firms have to produce more to meet that demand. They have to increase their production, hire more labor, and therefore they're going to drive up labor income. And then, that labor income ends up back in people's pockets and increases consumption.
Moll: The key thing however is in the representative agent new Keynesian model it's all about the direct effects and very little about the indirect effects. So, in any quantitatively realistic decomposition you could do, you would get that more than 95% is about the direct effects. So, which is basically just pure intertemporal substitution. And less than 5% is about indirect effects. Another way of saying this is that just because in a representative agent new Keynesian model people are these permanent income type consumers, they have low marginal propensities to consume. And so, therefore they react a lot to interest rate changes, and react very little to income changes.
Moll: The first approximation therefore in a RANK model it's all about intertemporal substitution. Instead when you introduce heterogeneity and more realistic household balance sheet, you end up with a much, much more realistic description of all the different channels through which monetary policy could affect household consumption. And there's both additional direct effects and additional indirect effects. In particular, in terms of the direct effects, these additional different types of income effects. These income effects are for example what you had mentioned earlier when you talked about the pensioners complaining to Jay Powell about not liking interest rate cuts.
Moll: So, the income effect are some people are borrowers. Borrowers like interest rate cuts because it puts more money in their pocket. But some people are savers and savers don't like interest rate cuts because that gives them less income from their savings. Similarly, you will have valuation effects from inflation. So, if you cut interest rates typically that will generate inflation. If you have a lot of nominal assets, that will wipe out some of the value of these assets and you're also worse off.
Moll: Finally, there's these potentially income effects working through mortgage rates. So, if I have a big mortgage, then in general, it's also good if interest rates fall because, for example, I can maybe refinance my mortgage later.
Moll: So, that's the direct effects. So, already there you paint a much richer picture. Depending on the structure of people's household balance sheets, what assets and liabilities they hold, and these direct effects maybe either positive or negative. And it's unclear how they play out in the aggregate. But on top of that then you have these indirect effects, which one of them I've already mentioned is that as consumption increases by some people, and also maybe investment increases, you're going to get increases in aggregate demand. And therefore, labor income which then puts additional money into people's pockets and they spend more out of that. But on top of that, you also have that asset prices change when the central bank adjusts interest rates. And therefore, maybe house prices go up or stock prices go up. And people react to those.
Moll: And then, the final thing as sort of yet another wrinkle you can have is that also fiscal policy in general will have to respond in some way to monetary policy. In particular, just because the government budget constraint will typically be affected one way or another if the central bank cuts interest rates just because the interest that the government pays on government bonds will typically be affected by the monetary policy rate in some way or another. It also matters what fiscal policy does. In particular, the technical reason is that in these models Ricardian equivalence breaks and therefore because people are constrained and therefore it matters what fiscal policy does in response to monetary policy. And so, all of these things are things that don't matter in representative agent model.
Moll: The summary, the way I always say it is I guess is that things are just much, much more subtle in these heterogeneous agent models. There's all these different complicated channels that affect household consumption that you wouldn't even think about if your view of the world is a representative agent model.
The summary, the way I always say it is I guess is that things are just much, much more subtle in these heterogeneous agent models. There's all these different complicated channels that affect household consumption that you wouldn't even think about if your view of the world is a representative agent model.
Beckworth: Yeah. And so, the critique about savers being harmed by low interest rates misses the rich story. And also, I think more importantly misses the fact that these people might be better off even if they do get a little less interest income. They might be better off in other dimensions. Maybe employment income goes up. Maybe there's some other channel. Maybe their asset prices go up. So, there's a much richer story to tell.
Moll: Yeah exactly. Yeah, 100%. Yeah, yeah. It's very hard to figure out how these different channels will tend to balance in the end.
Beckworth: I want to give a concrete example of this because this is very abstract. I want to reference a paper that you also sent me. I won't attempt to mention the authors’ names. Maybe you can because probably I'll mispronounce it, but the title of the paper is *Household Balance Sheet Channels of Monetary Policy: A Back of the Envelope Calculation for the Euro Area.* And they also have a little column they've written up for Vox EU. But what they do is they look at the effect of QE programs or the asset purchase programs in Europe. This critique could be applied there as well. People who depend on fixed incomes or interest income might be harmed by this.
Beckworth: Some really interesting results. I encourage our listeners to take a look at this paper. But what they find is that the households with the lowest incomes benefit the most. They show that on average their incomes rise about 3.5% compared to only 0.5% for the higher income households. So, low income households actually benefited a lot from QE in Europe. More importantly, they show that it worked through employment channels. It didn't work through asset price channels, so much as it did employment income and just increased demand for labor, which is an important story to tell because you often do hear, for example, a critique of QE is it's only for the rich. It helps the rich because it elevates asset prices, but this paper, for example, show that it actually helped the low income as much or more.
Moll: Yeah. No, exactly. Their names are Slacalek, Tristani, and Violante.
Beckworth: Thank you.
Moll: And where Gianluca is my former colleague at Princeton. So, yeah, exactly. I sent you this paper and I very much like this paper not so much because of the final message they find, which is that QE was actually beneficial for the lower income households. It's more about what I like about it is the approach. I think if you redid this in other countries the way they did it, you may well find that it goes the other way because the balance of these direct and indirect effects may well go the other way depending on the characteristics of the countries.
Moll: But the main thing I really like about this paper and this also ties back to another question I think we alluded to already, which is the difference between RANK and TANK and HANK in terms of explaining things to policymakers. What I really like about this paper is the way they built up things from the ground up using microdata. So, they just do this very simple exercise. They just say, okay, let's think about all the different transmission mechanisms in this HANK models. The nice observation they make is we can get quite far by just using some microdata on three key things, one is household portfolios. So, what assets and liabilities do they hold. The second one is their exposure, in particular of their income and their asset prices to aggregate fluctuations. And then, the final thing is household's marginal propensity to consume.
Moll: What they show is you can do these very simple back of the envelope calculations where ... So, in their case they go to the Euro area, household finance and consumption survey, which is HFCS, and they just split people into these three groups that we already talked about. They say, "Overall there's 10% of people who have neither liquid wealth nor illiquid wealth. That's poor hand-to-mouth. There's 12% of people who have low liquid wealth." And then, there's the rest, and then they just do simple things. For example, they say, "Of these 12% of wealthy hand-to-mouth people they have on average negative 17,000 Euros of assets in that data set that are directly exposed to an interest rate change. In particular these households have a lot of mortgages, which in Europe are often flexible negative rate mortgages depending on the country. So, these guys would benefit lot from interest rate cut." They then do a similar exercise regarding exposure to inflation. For example, then they find that this group has negative 29,000 Euros of assets that are exposed to inflation. Then you do another exercise like this for how asset prices would respond and how people are exposed to asset prices. And then, you do another one. How their labor income would move and who holds how much labor income.
Moll: And then, in the end you take evidence on MPCs as I've already said and you glue it all together, and just by these simple back of the envelope calculations, you get an estimate of how much a 1 percentage point cut in interest rates should increase aggregate consumptions through all these various channels, which I think in their case is something like 0.7%.
Moll: And so, the point is I just very much like this approach and that's where a HANK model leads you. Where in contrast in a representative agent model your story would have been instead the representative agent substitutes intertemporally and has an intertemporal elasticity of substitution of, I don't know, one half. And therefore, a 1% interest rate cut causes consumption to rise by half a percentage point, and that's the end of the story. So, I just like this building it from the ground up and much more nuanced way of approaching this. I also actually think it's much easier to communicate to policymakers.
Beckworth: No, absolutely. Yeah, I think this is an important point because a lot of times you hear the critique of monetary easing during the crisis. “Oh you’re just tapping into future spending. You're just bringing forward future spending" or "You're not really making that much difference. You've just shifted stuff across time" which is kind of the intertemporal idea, right? But if there's a much richer story, if there's other channels with which this works, then you can't apply that critique as much. I think it's useful because it really does broaden the scope and the channels to which monetary policy works. I think one of the drawbacks of the standard new Keynesian model is that it narrowly focuses on that intertemporal channel and it lends itself to criticism, I guess, is a big thing.
Beckworth: Also, this paper and I guess HANKs in general, I mean, again going back to QE, I have my own criticisms of them. But they do lend themselves to thinking more carefully about the counterfactual. So, you could say, "Oh, the poor savers" but what's the counterfactual? Counterfactual might have been the massive unemployment, all these other channels working, a huge decline in income for everybody. And it could have been much worse situation than we already have.
Beckworth: But I want to come back to the policy. This point you made about the policy implications and thinking through this in a tractable way. And this was the motivation that brought the show together between you and me. This was the question I asked Eric Sims and we talked about this on Twitter a little bit. But if you're a policymaker, how do you use HANK? So, the big model is RANK being used by central bankers, if they use them. I mean, I think some central bankers have even simpler models in their mind, simple Phillips Curve in their head. But if you're using a RANK, there's three equations. You kind of can think through the scenarios, do a back of the envelope quick calculation in your head. How would you sell this to policymakers as a way to think through their policy actions?
HANK Applied to Policymaking
Moll: Right. So, yeah. We've actually over the past few years done a fair amount of this. We've spent a fair amount of time within central banks. I mean, I have to say that reception by central bankers has actually been really fantastic and they've been really interested in our work. And so, that's why we also have this master class for example. And so, I mean, yeah, one way to sell this, so if you really want to talk to the policymakers at the very top I would say is exactly these simple back of the envelope calculations that we just talked about. So, we just had John Williams here at LSE a few weeks ago who's the president of the Federal Reserve Bank of New York. So, I'd asked him "What do you find the most useful in terms of approaches to modeling?" And he actually said, "Yeah, these little back of the envelope calculations can be very useful." I think just because they're already thinking about this a lot anyway.
Moll: So, I'd say one thing that actually goes in our favor and in the favor of HANK models is that I think quite often people actually have quite good and maybe even much better intuition about how HANK models work than about how RANK models work. In particular because the real world has a lot of heterogeneity in it. We just happen to, I think in general, have good intuition about the world we live in rather than some parallel universe where there's a representative agent. And so, I don't think in terms of the intuition it's so hard to sell it.
I'd say one thing that actually goes in our favor and in the favor of HANK models is that I think quite often people actually have quite good and maybe even much better intuition about how HANK models work than about how RANK models work.
Moll: And then, the thing that comes out of it is that sometimes, and maybe that's one of the things you're getting at, the policy implications are actually quite different in terms of what comes out of these HANK models.
Beckworth: I want to touch on that last policy implication point that the intuition may be even stronger for a HANK model than a RANK model. I heard your colleague talk about this, Greg Kaplan. He did an interview on HANK models and he made this point that some of these insights take us full circle back to introductory macro classes where you do really simple models. Like an aggregate demand and supply model. Where sometimes you don't always think through expectations, forward looking. You just tell a story. And I think many people at some level probably think more like a HANK model than a RANK model. Is that a fair interpretation?
Moll: Yeah. I think that's right. Yeah. In a sense these models are very in quotation marks "old Keynesian". And the story we tell about these direct and indirect effects, indirect effects being important in the sense is very similar, at least qualitatively to the standard Keynesian cross you learn in your undergraduate. So in particular that's sort of impulse and a multiplier. What we call direct effects is really the impulse and what we call indirect effects is the multiplier. And the key point is that the RANK model, the standard representative agent new Keynesian model is not very Keynesian at all in the sense that it has a small multiplier. Whereas our HANK type model is much more Keynesian in the sense that it's much closer to the way people think in their undergrads where the impulse is not that big, but the multiplier is relatively bigger.
Beckworth: Okay. This is all very interesting, but we're running low on time and I do want to touch on some of the current debates we're having right now. So, the ECB just announced they're going to do a review of their policy the way they do monetary policy. The Fed’s been doing one for the past year. One of the things they've talked about doing, at least here in the US, is adopting some kind of makeup policy, which means a price level target or nominal GDP level target where you make up for past mistakes. One of the reasons you would do that is because it provides powerful forward guidance. You're committing yourself to acting a certain way in the future. And so, my question is what does HANK mean for forward guidance? Number one. And then, number two, for these makeup policies more generally.
What HANK Means for Forward Guidance and Makeup Policies
Moll: Yeah. So, for forward guidance it's actually an interesting question. So, in particular one intuition you may have is that because these direct effects are less strong, so people respond less to interest rate changes, you could think that also monetary policy in general, I guess, and forward guidance in particular could be weaker. But it turns out that’s actually, at least from a theoretical point-of-view, not necessarily true. So, in particular the key point is that while the direct effects are smaller, the indirect effects are larger, and these things can potentially then in the aggregate go one way or another. So, in particular they could potentially exactly offset.
Moll: There's a nice paper by Ivan Werning at MIT who makes this point where he has a clean theoretical benchmark where the size of the direct effects and the indirect effects exactly offset always. And then, the consumption response to monetary policy is always exactly the same regardless of how much heterogeneity there is, how many people are born constrained, et cetera. It's sort of a Keynesian cross logic again. While the impulse is less in HANK the multiplier is larger. And then, the two things offset. And that's true for monetary policy in general, but for forward guidance in particular.
Moll: This is a benchmark result and typically in practice when you bring these models to the data in a realistic way, it's not going to hold exactly. But it's just to make to the point that forward guidance or monetary policy doesn't necessarily have to be less powerful. I think what's fair to say is that both monetary policy and forward guidance, that their results on aggregate consumption will depend on factors, I would say, that are more out of the control of the Fed than what you think if you had in mind a RANK model to think about the world.
Moll: So, in particular, what matters for how powerful forward guidance is for monetary policy is all the things that determine how much household disposable income responds. So, it matters how fluid are labor markets, how much does firm investment respond. And even what is fiscal policy doing. So, all of these are things that the Federal Reserve Bank can't really affect that much. So, there I think for forward guidance is unfortunately ... or that's just the reality of things. I guess, again, a little bit subtle and nuanced and a little bit murky in terms of the takeaway.
So, in particular, what matters for how powerful forward guidance is for monetary policy is all the things that determine how much household disposable income responds.
Beckworth: Very interesting. Let me follow up with just the level target discussions. If the Fed were to stick with its inflation target or go to, say, a price level target, do the HANK models shed any light on that distinction?
Moll: Yeah. I mean, not really unfortunately I would say. Sorry to disappoint. I know you're a big proponent of nominal GDP targeting.
Beckworth: That's okay.
Moll: I think it comes back to this point that I've made before, which is that the HANK model is really a three equation new Keynesian model, and then, you just throw out one of the equations to ... I guess or early equation. And you keep doing the other two. So, in particular, you keep the new Keynesian Phillips Curve. So, the paradigm for price setting is a really the same as in the new Keynesian model. I mean, I haven't fully thought this through, but I think all the things about price level targeting or inflation targeting would be the same in HANK as in RANK. I mean, I should say the reason why we adopted these assumptions is not because we necessarily believe them. But we wanted to change one thing at a time. We wanted to start from a well understood benchmark, and then, bring our competitive advantage to the table, which was the well modeled household heterogeneity and the distributional results.
Beckworth: So, you guys use survey data to calibrate the models and to simulate them. And you feel like the data is good enough where you feel confident about what you're doing, right?
Moll: Yeah. I mean I think household data has been getting better and better over time. Both for the surveys, and then, more recently we also use a lot of administrative data. And so, for example, I have bunch of work with ... so, not in the HANK paper that we published in the AER. There we just use US data, in particular the Survey of Consumer Finances. But I have some other work, for example, where we use the Norwegian tax records on the universe of Norwegian households. And there you just have...
Beckworth: Amazing data.
Moll: … in observation and you just have everything about them. It's the data they use to assess the income taxes. In Norway the key thing is they have a wealth tax, so therefore you know all the assets, the liabilities that these households have. You have the information that the tax authorities collect. So, in that sense, yeah. I think the data is pretty good already. And it's going to keep getting better and better. So, in that sense I'm definitely not worried about that.
Beckworth: Very interesting. In closing what do you see HANKs going in the future?
Where HANKs Are Going in the Future
Moll: I think what's already been happening is that, as I've said there seems to be a fair amount of interest from central banks around the world. And so, I think what will happen or at least I hope what will happen is that basically we will get better and better on the modeling side in terms of bringing microdata to the table to really take seriously the micro consumption behavior. And then, I think what will hopefully happen is that these HANK models will become another model in the menu of models that central banks have access to or they choose from when they try to answer a question.
Moll: So, I would say we're definitely not saying HANK should be the only model out there and should replace all other models because I think the right choice of models depends on the question you're asking. And so, I think what would be a good outcome would be if for some questions, say, you're interested in what's the effect of house price changes on the macro economy, or how does monetary policy affect household balance sheets by affecting, say, house prices and so on. If those type of questions the central banks would answer with HANK models, then I think that would be already a great outcome. And I should say that the central banks are actually kind of a very natural place to start this or adopt these frameworks because they're quite often also the ones that have the very high quality microdata that you can actually naturally use to discipline these models. It's just that at the moment, usually the modeling guys and the guys with the microdata, they sit in different departments and they don't talk as much as they should. In a sense what we're saying is "You guys should just talk more to each other." So, I think the more the interplay there can be between the people that understand the models and the people who are the experts for the microdata, I think there would be lot of gains from trade from that.
Beckworth: No. I think you're absolutely right. And I don't know if you saw this, but in the last FOMC minutes that came out in the last meeting, the last ones that are available, they actually talked about a HANK model in there. They talked about the implications of different channels through the different households. So, if that's any kind of sign, it looks like a very promising development and front for all the work that you're putting in the HANK models. The fact that they're discussing at these high level meetings.
Moll: Okay. That's great. No, I had not seen that.
Beckworth: Yeah, yeah. I thought of you when I read that. With that our time is up. Our guest today has been Ben Moll. Ben, thank you so much for being on the show.
Moll: Thank you so much for having me. It's been lots of fun.
Photo by Eva Dang