Peter Williams on Interest Rates, Term Premium, and the Importance of Inflation Expectations

Adding additional macroeconomic data to factor models may help provide a more stable estimate of the neutral rate of interest, or R-star.

Peter Williams is a managing director of macroeconomic research at 22V Research and was formerly at the IMF and the World Bank. Peter joins David on Macro Musings to provide a market perspective on interest rates, Treasury markets, and monetary policy. Specifically, David and Peter discuss the dos and don’ts of estimating term premiums, the importance and future of R-star, the usefulness of inflation expectations, 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: Peter, welcome to the show.

Peter Williams: Thanks, David. Great to finally be on. Really nice to make an appearance.

Beckworth: Yes, it's great to have you on. I have interacted with you for several years now on X, formerly known as Twitter, and I gained a lot of wisdom from you, because you're really good at modeling the macroeconomic issues. I know that when I had questions about term premiums or measuring r-star, you were someone I could go to, because you knew the models in depth. You knew the gory details of those models, and so you would help me better understand them myself, so I'm delighted to have you on the program today. We'll discuss that, some other things, and this conversation is long overdue, because we've been talking for a long time online. Finally, I get you on the podcast. So, Peter, tell us about your work, what you do, and how you got there.

Williams: Yes, so, right now, I'm a managing director, mostly focused on U.S. macroeconomic and Fed policy research here at 22V Research. We're an independent research shop. I'm also doing it along with… there's many other firms like us, we think they have a pretty unique and very market-facing sort of spin, so it's been quite a different seat than being in the public sector. My career path is, I think, a little different from most. I thought, for a long time, that I wanted to get a PhD when I was an undergrad, and then I realized, when contemplating the PhD process, that that probably wasn't the thing that I actually wanted to do, nor was academia probably the place for me. I ended up pivoting late in undergrad and going to get a master's, and I have a master's in financial economics, which was great quantitative training and from a theoretical and empirical perspective, but more focused on practical Wall Street applications, by and large.

Williams: After that, I spent a little bit of time in the private sector for a couple years doing some Fed watching, also U.S. macro research, before I spent a five-year stint at the World Bank and at the IMF, which is a great place to be in your early career phase. You get to learn a lot, you interact with policymakers that are pretty interesting in a fun way. And, while I was there, I had the great luck of being on the U.S. team at the IMF. And there's a lot of forecasting, there's the policy analysis work that comes with any institution like that, but especially when you have a boss who's quite understanding and supportive, as I did, it's a fantastic place to do your own research.

Williams: And so, for me at the time, in a world… the sort of 2015 through to 2020 environment, it was all about low inflation, low neutral rates, trying to think about a world which is structurally changed after the financial crisis, and I got to publish the papers there. Then I took that technical and policy experience of the private sector and, for the last couple of years, I was at Evercore ISI on this global central banking team, and now I've been at 22V Research since this past summer. We're doing very similar policy-related work, but from a Wall Street perspective focused on a relatively shorter time horizon than this, maybe, five to 10-year stuff in the public sector. This is, three, six, maybe 12, occasionally 24 months, but it's definitely a shorter time horizon and a more reactive world just because markets move every day. You don't quite have that same feel when you're in policy land.

Beckworth: Yes, or academia, which are the two places I'm most familiar with. You think about these long-term, maybe, trends, and we like to think about the short-term as well, but you're definitely someone who's got a better sense of that. Let me ask this question, since you're in the market, and, in fact, you provide research to markets as well. How does someone like you get your news? You're the one providing news to the rest of us, so, how do you get your news? What is your daily routine like? How do you digest it, put it together and think about what you're going to write about for your clients?

Williams: Yes, so, I think that in most Wall Street seats, which are seats like mine, there's sort of a rhythm around the data release in the Fed calendar that drives most of your routine. So, the first Friday of every month is non-farm payrolls. That's really important. Then you plug in CPI, PCE, those sorts of major releases, and then along with the Fed meetings, that fills in the structure of your calendar. On top of that general structure, you try and figure out, okay, what are some times that I'm going to either be interacting with clients a lot or what are the times when it's going to be a bit quieter, and I can dive in and do some more research work?

Williams: And so, from my daily routine, I think that one of the better things to do as a market participant is that you wake up every morning and you just check prices, and I think it becomes very ingrained, but to a certain extent, you want to wake up and, “What happened overnight?” Especially living pretty far West in the U.S. now, it's sort of a, "Wake up, what happened in Europe? What happened in Asia overnight? What's the sort of early beginnings of the U.S. been like?” And then, across the day, things are always happening, a lot of news flow comes in. And that's really the biggest trick for me, going from the public [sector] to the private [sector], was that the information deluge is very different when you're in the private sector relative to the public.

Williams: You're in the public sector, you're sitting around consuming working papers, you're thinking about these very long-term issues, but in the private sector, you're reacting to events much more. And you're hoping that you can anticipate some stuff that's happening. That's where the-- taking the moments when things are a bit slower to really dive into an issue, and I knew I spent, I don't know, a third of my day just playing around with data, looking at interesting new data releases, reading through news or other interesting economists, and just trying to get a sense of how things are evolving. That's all very short-term. It's like, "What's going to be the next interesting ripple inside of the bigger cyclical story?" Because a lot of times, those ripples matter tremendously from a market participants’ perspective, even if, oftentimes, you're smoothing through them in policy land.

Beckworth: So, you came to a conference that we put on late last year. It was a Bennett McCallum Memorial Conference, [we] had some Fed people there. But what does someone like you get out of a conference like that? Is it just to hear the governors or does it give you a chance to sit back and reflect and think about these issues more deeply?

Williams: Yes, so, one of the perks of conferences, especially one like that, is that it's more academic and policy-focused. It gives you a little bit of time, A, to just block off your calendar, which can be helpful just in its own right, but, also, it lets you sit and think about some of the bigger picture issues that are sort of... both what's been happening, but a lot of times it just takes one little interesting nugget from somebody speaking and you can latch onto that. Then, it's a research theme that you can maybe run with for a couple of good pieces or at least something that you can think about more going forward. And a lot of times it's-- finding those little nuggets is what we're all about. The trick is just bringing in as many perspectives as you can, and then you, as the research person, filtering through that process. And so, a slower, more thematic, and a bigger-picture event can be very helpful for that, and Governor Waller was there, which didn’t hurt.

Beckworth: Right, and I understand that people were pulled in to hear his talk at an important time in Fed policy, what was going to happen. Now, I don't think he moved markets any with the speech he gave, because it was a much more reflective, "Where has monetary policy been and come in the past 20-some years?” Let me now draw, again, upon your role as a market participant and observer, someone who writes these research notes, and ask a question that I've been thinking about, a lot of people have been thinking about, and that is, why didn't we get a recession last year? Because the market was predicting a recession, or most observers were. The yield curve was supposedly predicting a recession. A lot of people had it baked into the outlook, but it did not happen. I know not everyone said a recession was coming, but I would say that probably a majority did, and we were wrong. So, what is your take on what happened?

Why Didn’t We Have a Recession in 2023?

Williams: So, you have to acknowledge that we got most of the things which happen around recessions. The housing market became quite weak, classically recessionary, [and] probably, maybe, the best single recessionary indicator. The broader goods and manufacturing cycle looked relatively weak. We had some financial stress, Silicon Valley Bank, obviously, in particular, but more broadly, you can see credit conditions tightening, starting in mid-2022, even. We had some speculative tech sectors and some of the speculative parts of markets… All of the classic recessionary stuff sort of happened. We skated through it.

Williams: So, It's important to acknowledge that we did see cyclical weakness, but we sort of got lucky, I think, largely on the back of two things. One, higher nominal topline growth gave firms and households room to smooth through shocks in a world that, if your top line is growing at 5% or 7%, you have a lot more buffer than a world where your top line is growing at 1% or 2%. That typically gets lost in most macro models, because they're usually focused on the real side. And so, that income smoothing effect often, in nominal terms, gets lost. Everybody's thinking about real terms, but sometimes the nominal-- because most of your costs are nominal, though it's in the short-run, that matters a lot. And then, beyond that, we're still dealing with the legacy of fiscal policy. That's largely faded by this point, but I think that early last year, there was still some pent-up excess savings… however, exactly, you want to frame that debate. Household wage growth was extremely active, still. And we’re still seeing this pent-up services sector catch up and rebound and now, perhaps, even to some excess. And I think those three things, really, with the first two being the more interesting parts, are really what did it.

Williams: We, in a sense, got lucky, but it was a very unique cycle. I wouldn't necessarily ascribe it to luck so much as, we were coming out of a recovery in a very asynchronized way across the whole of the economy. It seems a bit natural that a lot of things would operate at very different speeds after that. I think that still, to some extent, characterizes where we are more so on the inflation side than the real side. But It's still a dominant theme in most of the macro debates we're having right now. I think a lot often gets lost in pure top-down perspectives.

Beckworth: Does that make it harder to forecast in an environment like this?

Williams: Certainly. Yes, I think one of the challenges that's really been apparent over the last couple of years is that both the cyclical relationships between a lot of the leads and lags are broken. The yield curve, obviously, was one I didn't put a whole lot of stock in to begin with. But, certainly, now, we can at least say its predictive power has been attenuated some, overall charitably, but more broadly, it's just harder, because the data is asynchronized across different sectors. I think that's where the real challenge is.

Williams: It's easy to find certain spots of weakness. The trick is knowing what parts of this weakness have another causal mechanism or are likely to resolve on their own. Where are we looking for true cyclical weakness that's going to break the labor market? Because if you're not breaking the labor market, it's not really a recession. Maybe you'll see activity, be a little bit weaker, but, fundamentally, if the labor market holds up, things are okay, or they're going to be okay.

Beckworth: Well, I like your explanation that the reason the economy was so resilient in spite of the sharp tightening cycle of the Fed and other shocks that hit the economy, was because we had a relatively robust level of nominal income growth. As someone who promotes nominal GDP targeting, I like to hear that story told, because it just reinforces, I think, the argument, at least one argument, for nominal GDP or, equivalently, nominal income targeting. I guess my question then is this, was that sustained level of nominal income growth, was it luck or was it, again, a result of policy? You touched on the policy side, the residual fiscal policy, the cash balances households held. Maybe it's a mix of both, that it was fiscal policy, but it was luck that it lasted as long as it did.

Williams: Yes, it's a challenge. I'm sure that [based on] your perspective, depending upon where exactly you sit in the economy, [you] might answer that a little bit differently. If you're a worker benefiting from tremendously tight labor markets, you call it, perhaps, good luck or even just a nice change of pace after a relatively anemic labor market for so long. If you're a small business owner, especially in the high-touch services category, you probably have a very different framing over the last couple of years. But I do think the legacy of it, the lesson, I think, especially in the short run, is that fiscal policy is tremendously powerful.

Williams: There are lessons for that, speaking to the post-GFC period, for sure, but also going forward, that's something to keep in mind. A fiscal policy, especially with very direct cash transfers, in effect, can be tremendously stimulative, especially when you're in a very… you know, you would say that you would expect the effects of fiscal policy to be attenuated as the economy heated up, and It's challenging to assess what exactly the cyclical position of the economy was for much of 2021 or 2022, given the hits to supply, tremendous demand growth. But, I think, pretty clearly, we've learned that fiscal policy can do a lot and that relatively stimulative monetary policy for a while probably mattered, but I think fiscal [policy] dominates, at least in environments like that. In more normal circumstances, that's more of an open debate, but at least in environments that, especially with a very lagging monetary policy response, fiscal wins.

Beckworth: Yes, and that environment raises a question about interest rates. I want to segue into that and spend some time there thinking about interest rates, because as we noted, the Fed raised rates rapidly, sharply, to historically high levels, especially given where we've been the previous decade or so. And there was this big question, well, how high is too high? Or has the Fed gone far enough up to be above the neutral rate so that it is truly tight? In order to answer that question, though, we've got to think about, what is your theory of interest rates? How do you understand interest rates at a very basic level? So, how do you think through interest rates, again, as someone who's active in the market, [when] you have to write these notes for your clients?

A Market Practitioner’s Theory of Interest Rates

Williams: I think the trick is that there's never just one interest rate, right? There's the fed funds rate at the very-- it sort of anchors the yield curve. Then, as you get out, there are a whole myriad of interest rates, whether you're talking about private sector borrowing, longer-term Treasuries. But, I think from a private sector perspective, the fed funds rate matters for certain parts of the economy, but it's really more medium and longer-term yields that matter more.

Williams: That's where, especially over the last couple of months, since late October especially, we've seen a somewhat different story between the fed funds rate obviously staying plateaued for a while, and market rates, easing off a fair bit. To me, whenever I'm thinking about the impact on the economy, usually in my head, I just have a 30-year mortgage rate, right? If I'm picking one, that's the easiest. It's both because it's the single most impactful for a lot of households. It's very clearly tied to a lot of cyclical indicators. You can sort of tie that back into Treasury rates and then from there into the Fed.

Williams: The trick with building a sort of single theory of interest rates is that you have to pick a place where you want to start. If it's the fed funds rate, you're talking… there's some degree of inertia. Where's the fed funds rate now? You're thinking through forward-looking inflation developments in the labor market. Then more broadly, there's these medium and longer-term anchors that filter into the rest of the yield curve, i.e. what is trend inflation? Where's R-star?

Williams: Both of those building blocks are relatively more challenging to assess. I think for a lot of market participants it's kind of like, "Well, we'll sort of revise it a bit as we go, but it's often not the key part." Whereas from a Fed policymaker's perspective, those structural debates often eat up quite a lot of time and thought, just because that's sort of the anchor on policy, you're heading out to the medium term, where I think for a lot of market participants is the long run, is the thing where you-- that's where you get to eventually, but you make profits day to day.

Williams: That's often a dichotomy that you're working with as the private sector anyway. The Fed is thinking on this more longer-term or structural basis and you're confronted with day-to-day realities, with a whole lot of money and transactions moving around. It becomes pretty challenging. So, I think that it all starts with the Fed, and then it gets dramatically more complicated from there. I think that that's really the trick. There's not really a singular meta-theory, which is often the challenging part of explaining it, because, realistically, so much of it anchors off of where we start, which is often a bit uncomfortable for people too.

Beckworth: So, market participants do think about R-star, at least over the long run, and then trend inflation rates. So, when they're thinking about 10, 20-year Treasuries, the long end of the Treasury yield curve, they truly have to wrestle with and come up with estimates of the underlying structural or R-star rate. Is that fair?

Williams: I think to a certain extent. I think for a lot of market participants and investors, there's a general view on where a long-term interest rate is likey to prevail. What's that trend rate? How much you break it down into, is this productivity? Is this some other distortions? Is this long-run inflation? Those dynamics are relatively harder to pin down. And so I think, as you also see in policy land for that matter, too, it's a relatively slow adaptation process that often has very punctuated, distinct shifts, big business cycle shifts, as we get new inflation regimes.

Williams: You tend to see in punctuated equilibrium that things are relatively constant for a long time and then they shift pretty dramatically. And you can see that, after COVID, certainly, and after the tightening cycle we've had, but also you see it after the financial crisis and in the opposite direction. It takes a while for people to internalize just how sluggish growth is going to be for how long, how potent this ELB is. Then, eventually, long-term rates drift out, but it takes a while.

Beckworth: Speaking of this distinction between [how] the shorter end of the yield curve is being shaped by Fed policy, near-term movements, business cycle frequency developments, and then longer term, again, the structural forces, the expected path of R-star, the expected trend inflation rate, you plug those things in. So, one of the interesting developments that feeds off of that is that the 10-year Treasury [rate] late last year was getting close to 5% and then it dropped quite dramatically, down to 4%, I believe. We're just a little bit above 4% now. It dropped all the way down, actually, below 4% briefly. What do you attest that to? What explains that, because Fed policy didn't change, as you noted, right? The Fed kept its rate parked there. Were markets just expecting rate cuts too soon or was it some other development in the economy or just some shocked expectations?

Williams: I think when— especially that summer normalization and then drop-down in rates afterwards— I think that there are three forces that happened all at once. It sort of overlapped, and it at least gets you to 5% and then we get off of 5% again. I think one, rates moved down very dramatically and acutely after SVB, and so that financial crisis tail got priced in, and it slowly drifted back out. Second, momentum is very powerful, and I think, especially over the summer when we were seeing very rapid GDP growth, inflation is starting to look a little bit better, but especially on a 12-month basis, it is still quite high.

Williams: Markets are coming to grips with, "Well, the tightening cycle might not be done yet. There's a lot of inflation uncertainty out there pulling rates up." And I think there's another-- and this is less satisfactory academically, I think, but it is a bit of an irregularity. The typically long-term Treasury rates end the cycle within, call it 50 basis points of where the fed funds rate ends the cycle. Sometimes it's a little bit below that, but typically the curve likes to be flat. If you are not taking any risk and you're getting paid 5.5%, you're taking a lot of duration risk and you're only getting paid 4%, it may not be particularly attractive. And so, I think that risk balancing channel often plays a role towards in-cycle dynamics.

Williams: Then we get to October, the Fed starts to pivot a little bit, the inflation data gets a bit better, there's a bit more of a recessionary concern, especially over the medium term, and that pulls rates down. And then, since then, we've seen this tug of war. You know, with the soft landing as the base case, it's a recessionary tale on one side and then this sort of no landing or, at least, this persistently “a little too hot” world on the other side and then figuring out where things are going, because the different data flows that have come in has often told quite different stories about that over the last couple of months. So, markets are [inaudible] between them, but accepting that the most likely move, the next move for the Fed, is going to be cuts. The question is just, how many, how quickly, and for what reason?

Beckworth: Another story that was told during this time was the large budget deficit. Last year we had a large budget deficit despite it being several years out of the pandemic, the recovery was doing well, and so there was a lot of talk about a term premium really taking off because of this fear. And I do think it's a legitimate fear. I do think that if things continue as they are presently in law, we're going to be in some trouble at some point in the future. And so, that story made sense for a little while, but then suddenly again, 10-year Treasury yields dropped dramatically and there was no change in fiscal policy. There was no change in the long-term path of debt-to-GDP. So, what do you think about term premium as a story for movements in long-term yields?

Term Premium as a Story for Long-Term Yield Movements

Williams: The issue I have with term premiums— and as somebody who's written research papers on the subject, I often feel a little bit remiss in some ways— is that, in real time, it's hard to disentangle the two things. A lot of stuff that acts as the term premium also acts on expected rates and vice versa. Fiscal policy, it's boosting near-term growth, it's presumably affecting Fed policy in an endogenous sense as well, and it's also raising term premium over the long run. So, that story was there, and there's certainly some longer-run concerns that the US needs to have about fiscal policy in an infinite horizon sense. We need to fix some things, but in the short run, a lot of stories like that are just a function of momentum, and for a long time, rates were really moving up, that creates a bit of its own dynamic as well. Momentum is a very powerful force in markets, and then you start looking for stories which coincide with that.

Williams: So, I think that the fiscal policy story over the summer was potent while it happened, but as soon as price changed, that story seemed to have faded away as you pointed out. That's one issue where I start with reasoning from the term premium in that sense, because it's very convenient to tell stories with, but it can be relatively hard to separate the two in real-time in a way that's particularly satisfying. It’s like, why is this happening now? And if it's happening now, why hasn't it been sustained?" That can also be a trick with reasoning from these decompositions rather than just the accepting that the 10-year yield is doing anything. It's a bit nihilistic, but--

Beckworth: Right, Right. But one of the theories for interest rates, or at least long-term rates, is this expectations theory, that you look at the expected path of the Fed's overnight target interest rate, and then you add in some term premium, and then, of course, embedded in those would be real versus inflation terms. But looking just at the expectation of the overnight rate and then looking at the term premium, that's a nice little decomposition, and I've played around with that as many others have as well. 

Beckworth: And one source that we often draw from is the New York Fed’s nice decomposition of the term premium. I've got some cool results playing around with that data, but you have cautioned me in the past to be careful, especially with the most recent estimates of term premiums. What is your caution in using models that estimate term premiums? And then, what would you recommend someone myself do if I am wanting to get a sense of what the term premium is?

Estimating Term Premiums: Cautions and Recommendations

Williams: The first one's easier to answer than the second, I hate to say. The cautions basically say, so when you're thinking of a term premium model, it’s interest rates, as you just said, interest rates are expectations plus a term premium, which also means that any term premium model is also, implicitly or explicitly, depending on how it's set up, a model of expected interest rates. And so, I think the trick you have to keep in mind is that a lot of these models impart a very strong degree of inertia to rates, because, historically, if you run statistical tests, interest rates are often pretty close to a random log. Not exactly one, but they're pretty close.

Williams: That means that if interest rates are run very low right now, they'll be expected to persist very low in the future, or very high, and you sink at the same degree of persistence. That can that can be a reasonable assumption under some criteria, but I think it often loses a lot of the nuance, especially at very extreme levels of rates. If rates are very high right now and the model's saying, "Oh, rates are going to be pretty close to high for a long time," well, that might be a fair assumption under certain circumstances, but we know we're mid to late cycle, the Fed has hiked very aggressively, so you have to sort of think about how the distribution of risks can really be skewed. I think that's where a lot of the term premium models— I think, really, almost every one that I’ve ever seen— often gets stuck, because they just impart a little bit too much inertia to rates.

Williams: And so, the good check on term premium models— and I've written this to clients plenty of times, and I’ve said it a zillion times otherwise— is thinking about backing it out. Does this really make a lot of sense if we take it and invert it to say, “what's the expected rate over the next 10 years?” Particularly, you can model that and look at the implied forward rates over time. Sometimes they're very reasonable, and you have to gut check it, though, because every model— and this is true with neutral rate models, too, with assessing inflation expectations as well. Whenever you're dealing with these unobservable concepts, you have to gut check them against reality. And some people are more inclined to just accept the model, because it's a good model, it fits statistically well.

Williams: It has nice forecasting properties over the short term, but if you're talking about something over a 10-year horizon, you have to assess that over 10 years. And there, it’s much more about balance of risks, trends in long-term rates, neutral rates. That's where lot of those models, I think, become a little bit weak, because it's harder to integrate in the long-term macro structural and short-term realistic financial market dynamics. Doing the two of those together is, computationally, quite challenging, but you also have to think about the theory on both sides in quite a rigorous way. It becomes quite difficult as well.

Beckworth: So, you have attempted to provide a way out for us. You have a 2020 IMF paper titled, *A Macroeconomic Approach to the Term Premium.* Walk us through how you attempt to resolve this quandary.

Williams: That paper, in effect, builds off of some related work from the San Francisco Fed and some similar associated authors there. Also, it has some ties in this Board of Governors Kim-Wright model, but it's pretty distinct from the New York Fed's approach. Basically, what we tried to do is we allowed for long-term interest rates to be time-bearing, which most of these models don't do, which is criticism number one if your long-term rate is assumed to be constant over time. That's probably an issue, especially if you're estimating a model since the '60s or '70s.

Williams: So, we went for time-bearing rates, and then we linked some basic yield curve factors— slope and curvature, in effect, which are very common ways of decomposing the yield curve— to underlying macroeconomic dynamics. And the model does a pretty good job at tying together these markets and macro dynamics. There's a whole range of very granular debates inside the literature about how information is incorporated into markets, and if adding macro variables tends to help.

Williams: And I think it serves as a good disciplining device precisely because you're providing a bit more medium-term and longer-term structure under the data in a way that markets sometimes lose if you just make some more standard assumptions about the way these models should set up and what they imply about reality. So, I think that the macro data provides a nice disciplining device even if there's some debates about it. Then the trick is, "Okay, how does a model like this work after COVID?” And [inaudible], when you have unemployment rates in the high teens, inflation going crazy, I think that incorporating that information from a macro model back into rates, that's where it sort of looks a little bit worse ex-post and has been one of the challenges in trying to assess these things after an entirely new environment rather than in the post-COVID world or the post-GFC world where the challenge was how do you think about markets and pricing over time in a world of very low rates that are persistently de-anchored to the downside, and in an environment where term premium seemingly had moved in a, persistently close to zero in either direction, sort of environment. 

Williams: That’s one way you can approach it, but realistically, with the term premium modeling, I think one of the lessons of thinking about the post COVID period is that a lot of times when you're sort of correcting to the structural problems of the previous cycle, especially when it's been a very pronounced cycle, you end up getting to a world where something changes and you now have to reassess how that approach might be best. And there, the non-linearity of the ZLB immediately after COVID, after the GFC, is the thing that's hardest to model in these models realistically, because it's sticky, it's non-linear, it's hard to sort of fold in. Then, from there— there's some models that sort of do this— but then, from there, you're like, okay, we have this non-linearity that's very challenging to incorporate, but also very extreme macroeconomic readings. But, in effect, it's been quite inertial. And so, folding all of those things in together and assessing the model ex post, you might suggest that there's a lot of challenges in sort of doing really any approach to the term premium realistically, I think.

Williams: That's where, I think— coming back to the disciplining device of what is Fed policy and then thinking through from Fed policy to very longer-term structural assumptions— I think if I was to write another term premium paper, I would lean more on the approach of Kim-Wright from the Board of Governors. That’s a very well-known paper as well, because they discipline the models, short-run dynamics and longer run dynamics, by using surveys of economists. And while economists' surveys are not perfect, they make mistakes, they probably, on that, aim a little bit to market pricing, because it's good, from a career perspective, not to look too particularly different from what markets are saying.

Williams: But I think that survey-based approach is very robust to extreme macroeconomic dynamics, because you're already incorporating those strange dynamics into the surveys, and it also helps around the zero lower bound. And there’s always the learning process with every research agenda and paper. And, I think, if I was to do it again, I'd probably steer more in that direction than I did before, just because the nature of the shock has been so different after the last couple of years. You're trying to think through, "Okay, having learned what we’ve learned, what's more robust?"

Beckworth: So, you're saying that if you are interested in the term premium decomposition, other than re-estimating the model in your paper, which would be your first choice for us to do, your second option would be to go look at the Kim-Wright data that the Board of Governors puts up and updates on its website regularly— I believe that FRED also hosts it as well— as opposed to going to the New York Fed decomposition. But one last question on the New York Fed's term premium series. Is it fair to say, though, that over time, it gets better? That at a particular point, as time goes on, you get far enough past that point that the challenges are less pronounced farther back? Or are they still a problem?

Williams: Let's sort of reframe that a little bit. Most models are going to perform a bit better the farther back in the sample you look, I think, especially with a lot of these models [where] you've had so much time variation. Certainly, when you're talking about thinking through the 90s and the aughts, you may get relatively similar results from everything, because they're all anchored off of a pretty similar, close to averages sort of data sample. Back then, it's just a bit easier than the post-GFC and post-COVID world, realistically. So, the farther back you go, the less the model choice matters. I think that as you get closer in time, you have to be a bit more cautious in using it [inaudible], frankly.

Beckworth: So, you want to, again, discipline your estimate of the term premium with actual and forecasted macroeconomic data. Okay, let's use that as a motivation to move into this idea of R-star, and we've touched on it already. We've had several shows on this as well, but it's an important concept that guides Fed behavior. Yes, there are plenty of critics of R-star out there. Every time I tweet something [about] R-star, I get someone who retweets a meme, "I don't believe in the R-star. It's something made up in your mind," which is funny. Nonetheless, Fed officials do look for it or think about it at least.

Beckworth: I think, from a modeling perspective, it's a nice way to think about or to have a benchmark to estimate where we are. And I go back to Fed Chair Jay Powell's 2018 speech, where he talked about navigating by the stars. And we’re talking about R-star here, he brought up U-star, the natural rate of unemployment, Y-star, which is potential real GDP. I want to read a quote from that 2018 speech at the Jackson Hole conference, and he says, "Navigating by the stars can sound straightforward. Guiding policy by the stars in practice, however, has been quite challenging of late because of our best estimates of the location of the stars have been changing significantly." Again, this is in 2018. For example, their estimate of U-star, the natural rate of unemployment was always too high, and they had to keep adjusting it down.

Beckworth: If you look at the Summary of Economic Projections, the long-run value keeps getting dropped. Then, just recently, he had a speech, Jay Powell, at Jackson Hole once again— so this was last year's meeting— and he had this clever wording where he talks about navigating by the stars under cloudy skies. So, it seems a very tough task to be using this thing called R-star. And you actually wrote a paper on it called, *Reading the Stars.* So, you're going to help us cut through the cloud, the uncertainty, and make sense of it. But before we do that, before we talk about your paper and your attempt to better get at these measures, what is your sense of how important one should take the notion of R-star? From a market perspective, is this an idea that’s just in some academic’s head, some papers and journals, or is it something we should take seriously, both because of the Federal Reserve and because there really is something like that out there in the real world?

*Reading the Stars* and the Importance of R-Star

Williams: You have to take it seriously, for both reasons that you just said. One, the Fed pays attention to it. So, to some extent, you are generally-- it's probably best that you do as well as a market participant. But, also, you can think about the general specifics of any given model of R-star, any quarterly estimate, and that's probably not particularly important. I think that the importance of R-star— and particularly nominal things like i-star, let's call it, if it's the nominal neutral rate— is that so much of the risks to the cycle to policy are shaped by how close to the zero lower bound i-star has been.

Williams: After the GFC, i-star is very close to, let’s call it 2.5%, maybe 2%, or even a bit below. And the zero lower bound is the defining feature of that entire cycle from a rates perspective, in effect. Whereas, post-COVID, seemingly, maybe i-star has changed a little bit, but with a sort of higher inflation world, that sort of asymmetry fades away, and you're no longer in this environment where you're really skewed to one side, because you don't have policy space. Now we have policy space, and so, specific estimates of R-star perhaps matter a little bit less, but they matter a little bit less because, realistically, policy has sort of moved away from the absorbing state that is the zero lower bound, into a more normal stance of policy, and whether or not we get back to 2.5% this cycle or until the next recession, whatever remains to be seen, as does the stability of that 2.5%, I think. But I think that the models and the estimates matter, but they matter less for their precise point values, and more so for the broader stories they're telling.

Beckworth: Okay. So, tell us about your paper, *Reading the Stars.* It’s another IMF paper from 2020.

Williams: I think that the impetus for the paper was obviously Chair Powell's speech in 2018. I spent a long time working on it and synthesizing the ideas, along with two of my co-workers on the US team at the time. The idea there was that, in effect, we've seen this at that point, Chair Powell had mentioned R-star, it's hard to navigate by very uncertain… and that was one part of the debate after the GFC, was the lower R-star world. Now, we've seen R-star estimates falling some, but I think that just as important, and just as obvious to me as I was sitting in that seat, was the debate around U-star and the rest of the business cycle, and the rest of the other structural parts of the economy that looks like, after the GFC in such a low inflationary world, those have all changed. And I think that that was just as much of an impulse and an impetus for writing the paper as was the R-star discussion.

Williams: What we did in the paper, is that most R-star estimates typically come out of relatively simple structured macro models. Typically, they have three or four data inputs, they're finicky to estimate for a variety of reasons, which we can dive into a bit more if you want to, but they're going to be tricky to estimate, so it's hard to just throw a bunch of data and then get out an R-star estimate, especially if you're not sitting there working on a single model for years and years and years on a huge staff and build it up. So, our approach is, "Okay, let's try and find a way that we can structure a bunch of macro data that will help us better triangulate both R-star, but also the neutral rate of unemployment, the NAIRU, however we want to frame it.”

Williams: We can think about the level of potential GDP growth, and we can stitch all of this together. So, instead of just looking at inflation, GDP growth, and real interest rates, we added in wage growth, which tells us both about trend inflation, but it also tells us a little bit about trend productivity growth. It also tells us about the state of the business cycle itself. We added in some other labor market measures to help triangulate the business cycle a bit better. And the idea there is that all the data we're adding in to hopefully have a bit more precision and a bit more real-time stability in estimating the results— You have a lot of unknown things you're trying to figure out, both specific parameters but also the states, like R-star, U-star, the level of trend productivity growth, all if that. So, there's just a lot of challenges, so the more data you can bring, the better, assuming the data is actually useful information, and there is sort of a trick. And so, we added in all of this data and the result was, basically, you get substantially more stable estimates in real time as you estimate the model, because you're just bringing more data to bear on the same fundamental questions, like where's R-star, where's the business cycle? And those two things feed back. 

Williams: I think that my quip on the term premium was thinking about term premium models as expected rate models. My quip on R-star, similarly, would be that models of R-star can sound very complicated, lots of things going on, but fundamentally, R-star is telling us, how does GDP growth evolve relative to our expectations and relative to what we think it should be doing, given the output gap? So, given if you're on the output gap, if it's really negative, you expect growth to be a bit above trend, but if it's really positive, the economy is super hot, you'd expect growth to be a bit below. That helps you infer where the stance of policy is at any given point in time. And so, the better you can pinpoint the state of the business cycle, that's better on its own merits, but it's also going to help you pinpoint, R-star because the input that drives all of the R-star estimates is, fundamentally, the business cycle and a little bit of the real rate. That's the trick.

Williams: And so, we did some other disciplining devices. We structured the model in such a way that it's very tightly anchored to [where] everything in equilibrium is at 2%. If you're at 2% inflation, that's equilibrium, and then everything else centers around that. I think that was an important insight as well, because it often gets a little bit lost in some of these other models. They don't necessarily discipline to the Fed’s target. They often prefer relatively different concepts that gets lost along the path. And I think that is also where I'm a little innately suspicious of some of these other models, not that the authors have done anything wrong, but it's just that you have to make some underlying assumptions.

Williams: From our seat then, and particularly from my seat now, the underlying assumptions I wanted to make are those that center around the Fed policy decisions and Fed optimality. And I think that was the most singularly important decision, and everything else ran from there. The results—and I think, perhaps, [it was] not too surprising— but basically, it showed that R-star had fallen substantially during the GFC. That's what, basically, every model shows, but actually, the R-star had fallen more substantially than most other models, and it's, basically, zero for most of the post financial crisis period, and then picking up relatively gradually, it's cyclical but persistent forces after the financial crisis started to fade away. And then thinking there as the legacies of tight fiscal policy right after the financial crisis in 2011/2012, fiscal policy probably became too tight too soon.

Williams: We had this banking stress, we had balance sheet issues, and all those leave-- They're not exactly secular, but they're also not normally cyclical in the way we typically frame relatively short-term shocks. They're all relatively long-lived effects and, eventually, those started to resolve away, and I think that was a really important sight. We see this in a lot of the other papers too, that around 2015 or so, directionally, basically, every estimate of R-star starts to pick up, the level being somewhat indifferent there, but that story is pretty similar there, that some of these semi-structural forces start to fade away. Then, you get back to a relatively normal economy by fall of 2016, normal with a lower level of rates, but at least things were becoming more healthy. Then, COVID hits, and everything gets snarled yet again.

Beckworth: Yes, so this brings to mind the possibility of different versions of R-star. There's different estimates of it, but I'm thinking in particular of different horizons, so business cycle R-star and then, maybe, a long-term, medium-run R-star. And, of course, what comes to mind is the famous Laubach-Williams or the Holston-Laubach-Williams version of that.  So, I'm wondering what your thoughts are on where R-star is headed, long-term. Are we going to be back in a world like we had pre-2020, like the naughts, low R-star levels, or do you think we are going to be in a different world fundamentally and at a higher R-star level, which will then, ultimately, leave Treasury yields, policy rates, all of those things, at a higher trajectory going forward?

The Future of R-Star

Williams: The last two business cycles both present very potent examples that, often, cyclical shocks have permanent effects. I remember a paper from years ago, *[Is the Cycle the Trend?]* I think some of my colleagues at the IMF wrote that title, and I think that's very true. After the GFC, we had anemic growth, we had these tremendous shocks, everything re-enters downwards. I think that after COVID, we've seen the reawakening of the Phillips curve, however we want to think about that. We've seen the reintroduction of supply shocks in an inflationary as opposed to deflationary way for the first time in, basically, 20 years, maybe even 30 or 40 or 30-something years; my lifetime, basically. 

Williams: Then, we see tremendous fiscal policy in a way that is, basically, World War II-esque in its expanse, and also duration, actually, if you think about multiple years of very [inaudible] fiscal policy. Then, all of that seems to, I think-- Structurally shifted up R-start. The underlying concept, however we want to measure it, was likely shifting up pre-COVID. And then, I think that all of that stuff and the escape from the ZLB as an anchoring phenomenon, and escape from the low nominal growth world changes states on R-star, changes where we are on R-star. I think, for me, my best guess is that i-star… you think a little bit of trend inflation shifting a little bit as well, and a little bit of real rate is at least 3% going forward. I’d lean a bit more, probably, towards 3.5% at this point. A lot of that, it goes to the directional stuff around COVID. I think that's all directionally clear, like what's the magnitude? I think if you look at the strength of the economy over the last couple of years, yes, the cyclical parts got hit, as you might expect, from Fed tightness, but overall things held up pretty well, and it continues to hold on relatively well through 2023, after the similar fiscal policy and everything else start to wear off.

Williams: I think that both shows the power of positive hysteresis, not usually a concept we think of with R-star, but one where there’s an actual tie up. Like, the economy’s been running really hot and that begets further investment, that begets labor supply. All of that boosts R-star or at least makes the sustainable level of real rates— whatever we want to think about whether they should drive R-star higher theoretically— clearly the economy has tolerated very high levels of real rates for longer than most people have expected.

Beckworth: Let's move from real rates and trends in real rates to inflation expectations in the time we have left, because you've also worked in this area. You continue to work on it in your current job. You had a 2020 paper titled, *Inflation Expectations in the U.S: Linking Markets, Households, and Businesses.* So, the question is, how useful are inflation expectations? Is it really a big deal? Again, the critics will say, “Oh, Beckworth, you and your inflation genie. You think it's magical, it can do all of these wonderful things.” And then I've talked to others who I'm sympathetic to, this other camp, that says, “Well, if the Fed has built up credibility, that inflation genie actually is sort of real. It does actually matter. Credibility does matter.” And I would dare say that it's probably played some role in preventing inflation from taking off even more than it did. But what are your thoughts on it? And maybe you can use that as a segue into your paper.

How Useful Are Inflation Expectations?

Williams: Yes, so I think inflation expectations matter, but in a structural underlying sense of like, how do I bargain for work? How do I, as a firm, think about top line growth expectations on a forward looking basis, and so how did all of those things feed back together? How much surveys of households, businesses, and even TIPS pricing really tell us about that underlying concept of importance is more challenging to estimate, realistically. Part of that is the de-anchoring of what we're always concerned about, right? It's like, okay, inflation spikes, and then it's going to shift up again, because expectations will be pulled forward. And then the question is like, okay, what's the underlying part of expectations? Do we see them in surveys and not like, “number go up, bad.” That happens a lot. But, especially in the short run, a tremendous amount of these moves… from a firm's perspective, it's input costs, typically like a commodity [inaudible]. For households, it's food and energy prices. And there's asymmetric responses. You get pissed off more when food prices go up than when they go down a little bit, and so you pay asymmetric attention here. I think that there's a little strand of literature on that, but in effect, you have to sort of think about these like asymmetries in the short run and then how does that get you into the long run? And you can try and build models which are structurally analogous to term premium models, which is like tying together different horizons of surveys.

Williams: You're going to get a five-year measure [which] is almost surely going to be less volatile and less variable and tell you more about underlying expectations than a one-year measure, which is a lot of these short-term effects. That's true whether or not you're looking at the TIPS market, household surveys, business surveys, all of the different measures we have out there. So, if you structure those dynamics together, you can get a pretty good sense. It tells you a good sense of, okay, what are people expecting right now? And the thing that it doesn't capture particularly well and, I think, theoretically, is very challenging to do-- so especially in a sort of anchored expectations context in the U.S., where we don't really have a whole lot of data that extends back into the period when inflation was super high and volatile. That's kind of a trick.

Williams: Maybe it's better to think about these studies in an EM context where inflation expectations are more volatile, but it's like, okay, we have this like long-run measure we're thinking about, like long-run, infinite horizon inflation expectations, something like that, and the question is, how does that really move? Nobody has particularly good theories for that. Yes, there are lots of nice econ papers looking back at history, but as far as, how do we operationalize that in the passthrough from short-term expectations into long-run expectations, from policy shocks into long-run expectations, that's much more challenging.

Williams: I think that building out a robust theory there is quite difficult. And I'll say that over the post-COVID experience as opposed to when I was writing that paper and the pretty robust result was that inflation expectations hit the anchor to the downside. After COVID hit, we see something opposite. We saw pretty large shocks, especially in short-term inflation expectations measures. And I think the Fed was, perhaps, excessive in its immediate response to some of these measures. The lurch to 75 basis points after these hot inflation expectations readings from consumers, yes, it seems, perhaps, a bit overdone, but the thing you see in that result is that, yes, it was revised down. The Fed famously— well, not famously, but niche famously— lurches into hiking 75 basis points [after] this preliminary reading from the University of Michigan's Consumer Inflation Expectations Survey comes in very hot. The final reading, however, looks much less punchy.

Williams: Basically, what happens is that the most educated sample of college-educated workers in that survey revised down their inflation expectations because they see the Fed pushing very aggressively. It's only a few hundred people who are effectively driving this shift in the survey, but it's an interesting result, and I think it does go to show that inflation expectations matter, and they can be influenced and shaped by policy, but actually trying to pin them down into a macro-econometric model that's useful for forecasting is much more challenging than just trying to say, “Estimate them right now on the spot since here's where they are.” That's relatively easy to do, but trying to build out a grand model that can tell you about the risks over five or ten years, given a whole bunch of different policy outcomes and inflation shocks, that's a much more challenging ask.

Beckworth: Do you find it challenging to use these surveys? I think one of the things that's been observed is that people are less responsive in the surveys. You just mentioned that there are only a few hundred people in the Michigan survey. I know that there's more in other surveys, but the concern is that maybe we're not really getting a true sense of the underlying inflation expectation by households. I know there was a paper done where people tried to get around this. They actually looked at TIPS, purchases of TIPS, by households versus, say, institution[s], trying to get a sense-- or there were the I-bonds, the famous I-bonds. A lot of people bought them up, retail folks bought up those, because they were worried about inflation expectations. So, there's things like that that you could turn to, but are we actually, maybe, less informed than we were in the past because people aren't responding to surveys about inflation expectations as thoroughly as they did in the past?

Williams: It's kind of what you say, just inflation expectations, because you could say, you know, change the sentence, people respond to surveys generally much less often than they used to, and this has been one of the great challenges of post-COVID macro analysis in real-time anyways, especially for a lot of things like the initial surveys as opposed to the final versions of all of the government surveys. Those initial response rates have just collapsed post-COVID. And look, we've seen some minor recoveries in certain areas, but not dramatically. And I think it makes any sort of analytical job more challenging. But on the inflation expectations point, I think the trick— and this is true before COVID, and it's true now— is that you have to think about the models for all of the different data sources in the inflation expectations space, whether it's like ISM Prices Paid, the University of Michigan survey, [or] the New York Fed survey. They all tell a little different and separate stories, and they're also idiosyncratic.

Williams: You have to think about the broader swath telling the story, thinking through all of them, and being— as much as one possibly can be and not getting trapped in your own nerves— but being as much of a Bayesian updater as you can be. I think that's the trick. There's a lot of short-term noise in these things. There's sampling bias, they over-indexed the commodity prices, all of this. But, in stitching them together, it's trying to be as signal-extractive as you can, not underreacting to noise, also not overreacting to noise, but also not underreacting to the much more harder-to-parse signal.

Williams: And I think that's going to-- especially in the inflation expectations space, post-the last year and a half, has been very challenging because, also, the Fed has differentially cared about inflation expectations at various points in time, which makes watching inflation expectations and mapping that into broader interest rates and market reactions, quite challenging. We've seen some, sometimes, where they seem to matter quite a lot to the Fed, and then other times, they seem close enough, and we'll let it go.

Williams: And I think, right now, the Fed, with the data itself, drives how concerned the Fed has to be, right? If inflation expectations just jumped a lot tomorrow, but inflation itself was well-behaved, who cares? In effect, you might be a little bit confused or a little bit skeptical, but in the end, the inflation data is what matters, that's where the mandate is. Everything else is feeding in to risks around that, frameworks around that. I think that's the issue with these measures, is that they help guide the risk, but they also can inject some noise into that risk management process.

Beckworth: Okay, one last question here as we close out the show— related to inflation since we're on that topic— and that is, will the "last mile" be as easy, in terms of getting 2%, as it has been so far?

The Difficulty of Reaching the 2% Finish Line

Williams: No, I don't think so. If we break down inflation along Chair Powell's broad lines, then maybe. Even with a few extra wrinkles in there for the benefit of complication, we've seen used car prices— which are a huge source of inflation over the past couple of years— coming down relatively nicely. How much farther used car prices can fall remains to be seen, and that is, perhaps, the biggest short-term uncertainty in inflation forecasting. It's like, what are used cars going to do? I don't know that anybody has any particular edge there, even people inside the used car market, but, certainly, that remains an issue. Eventually, that source of deflation is going to taper out. My guess is that it's at some point over this coming summer, similar with the rest of the core goods deflation. There's a risk, from an optimistic sense, that deflation could keep running for a very long time, even a couple of extra years. I think, realistically, given what we've seen so far-- also in core goods, the rest of core goods deflation probably runs out. And so that leaves you with housing inflation which, as formally measured— and this has been gone over a lot— is quite inertial, given the way that sampling inside of the housing surveys works for CPI, PCE, rents inflation, but nevertheless, it’s pretty inertial. 

Williams: That's come down slowly to where it is headed, but I guess it’s a bit of an open question as well given that market rent estimates are also quite all over the place. Certain measures, like single family rents in particular, look like they might be stuck around 5%, which is not an inflation target consistent level. Then, we think about core services, ex-housing, which is where Chair Powell tends to anchor on the most, where he's focused the most. Over the last couple of months, especially, that's looked a little bit bouncier, especially in the CPI world for some relatively technical reasons, but more broadly, it's not been coming down particularly fast. I think we've seen wage growth potentially stabilizing at levels that also seem a little bit too high. Maybe high productivity buys us faster wage growth, but it seems to me like the concern-- I think framing it like “the last mile is the hardest” is a reasonable enough way to approach it. But, if we're thinking about the shocks that happen and how do they dissipate is, perhaps, a different way of framing it.

Williams: The core goods, the supply chain shock, the food and energy shocks, those have happened, and those are largely fading out of the data at this point, and both in an inflationary sense, but now also fading out of the data over the next, let's call it, 6 to 12 months, who knows, in a deflationary sense. You're left with the legacies of tighter labor market bargaining, relatively snarled labor markets, and the possibility that wage growth assumptions, on the part of firms and employees, have reset a little bit higher, and there's still a lot of relative wage level re-equilibration yet to come. Marginal workers, like people who are getting rehired, saw relatively punchy wage growth, and people who are not getting hired, not switching jobs, saw relatively slower wage growth, and so some of that re-equilibration is going to just take a lot of time. And I think the question is, do we think about those as new shocks, or is it just the same old shock from a couple of years ago working its way through the system?

Williams: And so the "last mile" question is tricky, because, to me, that might just be the same shock that's taking a long time to feed through and the Fed could end up pushing again a bit harder against it anyway, just because they don't really trust the inflation process at this point. I think that it seems like a pretty reasonable base case for the Fed. And this isn't the S&P, a Fed that's very cautious about cutting rates, even when inflation is, basically, at— from a statistical sense when we think about the noise around inflation— the inflation is, basically, at target over the Fed's forecasting horizon.

Williams: But rates come down very slowly because there's a lot of caution there, and they don't really trust that the inflation process is going to be well-managed going forward. So, I think that the good parts of deflation are largely behind us, and the question is, we're just going to, very slowly, I think, from here, see a more challenging deflationary process. In a real-world sense, I don't know how much 2.5 versus 2.4 PCE matters for most people, but from the Fed's perspective, and those of us in Fed-watching seats, that can be quite impactful for sure.

Beckworth: Okay, and with that, our time is up. Our guest today has been Peter Williams. Peter, thank you so much for coming on the program.

Williams: Thanks so much, David. Great to finally be on.

About Macro Musings

Hosted by Senior Research Fellow David Beckworth, the Macro Musings podcast pulls back the curtain on the important macroeconomic issues of the past, present, and future.