Stephan Luck on What History can Teach Us about Financial Stability

Stephan Luck is a Financial Research Advisor at the Federal Reserve Bank of New York. In Stephan’s first appearance on the show, he discusses how the 2008 Great Financial Crisis shaped his career, how he and his coauthors leverage LLMs to comb through massive amounts of historical data, what this data can teach us about responding to bank failures, what people get wrong when they try and make historical analogies to the GENIUS era, the lessons we can learn from the German hyperinflation, and much more.

Subscribe to David's Substack: Macroeconomic Policy Nexus

Read the full episode transcript:

This episode was recorded on June 1st, 2026

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

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

Our guest today is Stephan Luck. Stephan works at the New York Federal Reserve Bank and has written widely on financial stability and recently released, along with his co-authors, a sweeping historical study of US bank runs that covers a long sweep of history in the 19th and 20th century. He joins us today to discuss this work and its implications for financial stability today, as well as some of his other work on stablecoins and their German hyperinflation. Stephan, welcome to the podcast.

Stephan Luck: Thank you so much, David. It’s really a pleasure to be here. I’m, of course, a huge fan of your show, which I consider really a massive public good, a super interesting show, and it’s really an honor to be here to talk to you.

Beckworth: Well, thank you for listening, and thank you for coming on. Now, we recently met at the Wharton Financial Regulation Conference that Peter Conti-Brown and others hosted. You had a great presentation, which is really what led to you coming on here. As it turns out, Stephan, your co-author on that paper, is a previous guest as well, Emil Verner. We discussed his paper with you, which I didn’t realize.

You have been on the show already in spirit, number one, and I’ll also mention you’ve technically been on the podcast already as well, because during that podcast recording we did at the Wharton FinReg Conference, Peter and I went back and forth and did a Q&A. You were the first or second question, I believe, there. If you recognize his voice, folks, that’s because you’ve heard him before on the podcast.

Luck: I guess I should really say it’s good to be back on the show.

Beckworth: Yes. Yes, indeed. You’ve written some really fun, fascinating, and important pieces on the history of banking in the US, particularly this question: Is it solvency? Is it liquidity? What really causes bank failures? Do bank runs really matter as much as some have led us to believe? We’re going to talk about that, and then your other historical work on the national banking system and its implications for the conversations today over stablecoins. As listeners of the show know, we’ve talked a lot about the GENIUS Act, the future of stablecoins, and so Stephan has insights there as well for us.

Then, finally, we’ll talk about, again, some German hyperinflation, time permitting, because you have a super fascinating piece on that as well. Again, folks, this is why you go to conferences. You meet interesting people like Stephan. They had a nice dinner, I think, the night before the conference. Probably everything we’re going to talk about today, we talked about over dinner. This is why I love going to conferences, to meet interesting people like Stephan. Now, Stephan, tell us a little bit about yourself. How did you get into this field?

Stephan’s Career

Luck: I work at the Federal Reserve Bank of New York. I’m an economist in the Federal Reserve System. Let me just put it out there that we have a disclaimer that everything I’ll be talking about will be my personal views and not necessarily those of the Federal Reserve Bank of New York. Let me start where I am right now in my career, and then tell you a little bit how I got there.

I’m an economist in the Federal Reserve System. It’s really, I consider to be, one of the most interesting and exciting jobs that you can have as a PhD economist. My job description has two parts to it. My daytime job, which is, I think a lot about current policy issues and how to help policymakers around the Federal Reserve System and here in the Federal Reserve Bank in New York to make key policy decisions that the Federal Reserve System takes.

Just to give you one example, one concrete example, what I do in my daytime here is I coordinate briefings together with my colleagues here at the research group, but also colleagues in the supervision and in the markets group. I guess from the markets group, you had two of my colleagues in your show before, Ellen and Roberto, some really great episodes.

We do these briefings for our president, John Williams. These briefings are around issues of financial stability, monetary policy implementation, financial conditions, and the broader economic developments that are happening. This part of my job is operating at a relatively fast pace. Deliverables are due by the end of the day, end of the week, end of the month. Certainly, our life here gravitates around the FOMC cycle.

Then, there’s the second part of my job, which is I’m just an academic researcher, just like a professor that works in an economics department or in a finance department. Of course, the time horizon that you’re working on in your research is much different. It’s much longer. Really, the privilege of this job is, besides having a set of amazing colleagues that I learn from a lot, is that, these two different parts of the job really speak to each other in a really nice way.

When you’re thinking about all these policy issues, it’s really good to have this academic background that allows you to put things in a bigger framework. Also, when you’re doing academic work, it’s really good to be in tune with what policymakers care about and what the institutional details are because you have to get them right. That puts a nice check on your academic work. This is where I am right now in my career.

Now, how did I get there? Of course, like many economists of my generation, I’m 10 years out of the PhD. I was shaped by the Global Financial Crisis and the European debt crisis. They happened when I was an undergrad and early in my PhD. For me, the most memorable way to think about it is that in the summer of 2008, I was actually doing an internship in India. I used the proceeds, the money that I earned, to travel a little bit around the country. You have to remember, this was a time we didn’t have smartphones. We weren’t connected to the internet 24/7. I had just come from this hike where I really didn’t have internet access for seven days or so. I got to New Delhi. It was mid-September. There were all these news tickers saying that something big had happened. I went to my YMCA, and I bought access to internet for an hour or so for $1. I figured out, “Oh, wow, there’s all this stuff going on.”

At the time when you were pursuing an economics degree, there weren’t a ton of classes on financial crises or credit booms gone bust. None of that really was part of the curriculum. Of course, in hindsight, we overvalue these specific moments. It’s really a long process. This was the part of me where I really became interested in macroeconomics and finance. Then, of course, the European debt crisis happened, and one thing led to another.

Beckworth: Now, you ended up also at Princeton briefly, is that right? You did some work with Markus Brunnermeier?

Luck: Right. At the end of my PhD, if you fast forward, the last two years of my PhD, I spent at Princeton University as a visiting student. Really, for me, what happened during the PhD, I spent, naturally, a lot of time thinking about theoretical models and bank runs and sovereign defaults. How did I get to working on economic history? That actually came later. That came when I took my first job at the Federal Reserve Board, where I started in 2016.

Around that time, two things happened to me. The first thing that happened was that I realized that I wasn’t really a very good theorist. A lot of my models were quite simple. The assumptions and the results were quite close to each other. The second part was that I got to meet one of my future co-authors, Sergio Correia, who started at the same day as I did and who then had the office right next door to my office. We figured we were both really interested in economic history. We also figured it was very hard to make really big impact papers with economic history, just for the lack of data. It was just really hard to do something that’s really representative and not just more of a case study-like paper.

What happened then, Sergio and I, we chatted a lot. One day, we came across this really nice experiment in the 19th-century banking markets in the United States during the National Banking Era, where there was this very nice discontinuity across, otherwise, very similar markets of how hard it would be to contest that market, to enter that market as a bank. We had this really nice experiment. We knew if we wrote a paper, no matter what we’d find, this would be a good paper.

We ended up publishing this paper in the JPE. It’s a paper on the effects of banking competition on financial stability and growth. At the time, when we had just the experiment, we didn’t have the data. I think this is really where Sergio shaped my view and what then led to my broader research agenda. He had this vision that, instead of hiring a bunch of RAs and have them type data, he actually had this vision that we could use the computer to do this.

We could use optical character recognition. We could use advances in layout recognition to build a Python pipeline to really systematically extract data from historical documents. That was really his vision. Sergio’s quite famous because he developed a code to really efficiently estimate models with multidimensional fixed effects. He’s quite famous for that. The fixed effect revolution in applied microeconometrics has a lot to do with his work.

What people maybe don’t know, he’s more generally this wizard-like genius who can have the computer do anything you want. I was just along for the ride, sitting next to him, watching him code as we together developed this Python code that allowed us to extract data systematically from historical documents. The first document we went after is the Office of the Comptroller of the Currency’s annual report, which, in its appendix, published every year the balance sheet of each and every national bank from the Civil War, essentially throughout the Great Depression until 1941.

That’s where we started. Once we had some success with that and realized that this is possible, we developed this broader research agenda. This research agenda is really all about using historical microdata to go after fundamental questions in macroeconomics and finance. There’s a ton of questions in macroeconomics and finance that you just cannot answer easily without making strong assumptions, just using contemporary data.

Just take banking as an example. In banking, if you want to think about bank runs and bank failures, there is a ton of government interventions in the banking sector that essentially are targeted to prevent this kind of thing from happening, right? If you’re looking at the contemporary banking sector and you’re concluding that bank runs never happened, then, of course, you’d immediately interject, “Well, that’s just because the government is providing deposit insurance, is doing lending-of-last-resort activities, and all kinds of other interventions to prevent those things from happening.”

Now, you have two options what to do. You can write a model, or you could actually go back in history and study those laboratories in which the government was not doing these interventions. This kind of logic of how you can use historical microdata applies to a lot of different questions. Just think of hyperinflations. There is just not that many hyperinflations in modern history.

We naturally sometimes have to go back in history to study them. Sometimes it’s the absence of a shock. Sometimes it’s the availability of a natural experiment. Sometimes it’s just these identification issues that come from the fact that the government is intervening broadly. It’s this mix of technology making it feasible for us to use history to speak to these fundamental questions. That’s what led to my research agenda and where I am now with my research.

Bank Failures

Beckworth: That’s a great segue into the first paper we want to talk about. It’s titled “Bank Failures: The Roles of Solvency and Liquidity.” This is what you presented at the FinReg conference, where we met. It’s also a culmination of what you just talked about, the use of technology, LLMs. This is also where you mentioned Sergio, but this is also where Emil Verner comes into the story as well. Maybe give us a quick overview of this paper. Also, I want you to spend some time again on how you used LLMs because there’s a lot of young grad students, maybe young economists, listening. This is the future. Help them see what they need to be doing.

Luck: Let me first say that this paper that you’ve just mentioned is based on a research agenda where I’m incredibly privileged not to be at the Federal Reserve only, but also to work with Sergio and Emil on this agenda. The general goal that we have is we want to understand the nature of banking crises. Why do they happen? What are their consequences? Think about policy implications. A big part of that agenda is using the historical data.

Everything I was just saying about digitizing data the way Sergio and I did it a few years back, everything we did is superfluous now because you can just use AI to do these things. We spend a lot of time on doing things that if we had just waited, the tech companies were actually giving us that technology. It doesn’t mean that we regret doing it. Sergio, Emil, and I have developed this agenda.

Underlying this paper that you’ve mentioned, we have a set of papers. There’s a paper on failing banks, on supervising failing banks, and a paper on bank runs with and without bank failures. This latest paper is, in some sense, an AI paper in that we really leverage large language models to help us construct a structured dataset out of originally unstructured data. What we did in this paper is we wanted to understand bank runs better.

The problem is that bank runs empirically are somewhat understudied. As I indicated earlier, they don’t tend to happen that often in the contemporary banking system exactly because the government is doing all kinds of things to prevent them. It’s very hard to evaluate the policies. It’s very hard to understand the dynamics around them just with contemporary data. It’s very well understood that the historical US banking system had a lot of bank runs.

The problem with bank runs in general, though, is that they don’t show up necessarily in a regulatory report. When we digitize the balance sheets or income statements of a bank, they will not contain information necessarily of whether the bank was subject to a bank run previous to filing a financial statement. Even if a bank fails, we only know whether it failed or not. We don’t necessarily know whether it had a bank run before it failed.

We can do some things around that with financial data by measuring deposit outflows, but we’re still having a substantial measurement issue. What we did is, essentially, we said, “Well, why don’t we use historical newspapers and try to figure out, ask those newspapers whether there’s a bank run or some kind of bank distress event in a given bank at a given point in time?” Essentially, what we’ve done is we’ve downloaded 372 million newspaper articles that are publicly available.

Beckworth: Wow.

Luck: We exploit the fact that in the history of the United States, there’s a lot of newspapers. A lot of these newspapers are available digitally to us. Some great work from Melissa Dell and co-authors who actually made it quite simple for you to search these newspapers. What we’ve done is, essentially, we’ve said, “Okay, let’s search the newspapers for articles that potentially talk about bank runs.”

At this point, everything I’ve described, for none of this, you need AI, because this is just, sort of, you have a document, you make it searchable. Where does the large language model come in? Well, the large language models come in when you identify a set of newspaper articles that potentially say something about a bank run, but you’re not exactly sure whether they would. Once we’ve searched this body of 372 million newspaper articles, we still have hundreds of thousands of articles that are potentially talking about bank runs.

Just to give you an example, a false positive would be a boat runs into a riverbank, or a specific person who owns a bank is running for office. Those are false positives that you need to weed out. The large language models are actually very good at that. Long story short, what we do is we use the large language model to create a set of newspaper articles that actually talk about true bank distress events. Then we use the large language models also to identify whether different articles are talking about the same actual event.

We combine them into different articles into one specific event. Then we essentially have a dataset now. In our case, it ends up being a dataset of more than 3,500 bank runs and a bunch of other distress events of bank suspensions and bank failures that just come out of the newspapers and make the historical newspapers speak to us. This would be something, I think, that would not have been feasible before large language models, because they can understand the context. Just the sheer amount of computing power needed wasn’t available before.

Beckworth: Yes, this is so amazing. This paper, there’s been a few others that have done similar things. This, at least in my view, seems like the future. Every generation, expectations go up. If you’re going to get published in a good journal, what counts as good work? The expectations are going up and up. This is an example of that, but it’s a good thing. We want to be able to use all that data that’s been sitting there latent in the archives. Now, we can do it, and we can learn insights. Let’s move on. Give us a summary of the key takeaways from your paper.

Luck: Yes, so I think the key takeaways from this paper, but maybe the broader research agenda, is that, first of all, bank failures and banking crises tend to be always and everywhere related to some kind of deterioration of bank fundamentals. They don’t happen somewhere out of the blue. They’re not some bolt out of the blue. They happen in the context of rising asset losses and a deterioration of the ultimate business model, also on the funding side. I’ll give you more detail on what I mean by all of that.

Beckworth: Sure.

Luck: The second insight is that bank runs, as a cause of bank failures or the cause of banking crises, I believe, tend to be overrated. That doesn’t mean that bank runs don’t happen. It doesn’t mean that they’re also not important. I think, historically, in the narrative of US history and in the way we think about fragility in the banking system, we’ve just put a little bit too much emphasis on them as a cause of why bad things happen.

Let me elaborate a little bit on both. First of all, with respect to fundamentals being at the heart of bank failures and the heart of banking crises, there’s one important fact that we bring forward. That’s actually coming out of a different paper, the failing banks paper that Emil talked about when he was on your show a while ago. It’s just that bank failures tend to be extremely predictable.

It’s not super surprising in the contemporary banking sector where bank failures are really a supervisory decision. It’s a supervisor that closes a bank based on observed hard information of the bank’s capitalization. Supervisors may take time. There are some legal issues involved. The fact that bank failures are predictable over the last 50 years or so, that is not super surprising. What’s really interesting, they’re also really easy to predict in the historical context before the depression, including during the depression.

That’s when there’s no deposit insurance. Depositors, when the bank failed, actually tended to realize quite substantial losses and needed to actually wait for quite a while to receive the proceeds out of the receivership. It was quite a risky investment to have a deposit in the bank, potentially. On top of that, next to the fact that depositors would realize losses, which you think would make them very attentive and wary of certain risks in their bank or the banking system, the accounting standards back then were not nearly as good as they are today.

First of all, the granularity of balance sheets is much lower than today. The accounting standards underlying them is making things such as book equity informative about the bank’s capitalization is much worse back in historical setting. We didn’t start out the agenda believing that we would find that. We were really surprised when we established that finding, that there’s this really strong predictability.

Now, predictability in and of itself doesn’t really fully tell you something about the causes of failure. You could argue that a bank failure, even if it’s predictable, can still be caused by a bank run, the fact that depositors withdraw. When I use the word “cause” in this context, what I mean is the counterfactual being, had the deposits not flown out of the bank, then the bank would have survived.

Now, the second argument we made in the failing banks paper was that, well, let’s just look at the proceeds and the receiverships of the banks once they have failed and understand how deeply insolvent was a bank. We just exploit the fact that we can see when a bank failed historically, how much funds did the receiver collect and then pay out to depositors. Now, the logic here is if a bank run is the cause of the bank failure, what the bank run does is it destroys the franchise value of the bank. It pushes the assets into the hands of the receivership.

There’s an argument to say, “Well, a banker is better at managing its assets than someone else, some outsider.” Those two things are not observable. What we did in that paper, we said, “Well, let’s take the observed recovery rate on the assets and failure and, say, make an assumption on what you think the franchise value was and make an assumption on what you think is the loss from the fact that the receiver is taking the assets and then holding them to maturity and just collecting the funds and then making payouts.”

What we just found in the history, this just happens to be a fact that we discovered, is that historical national bank failures just involve very, very low recovery rates. It’s very hard for the majority of the failures to make an argument that the franchise value was so high and the receiver was so inefficient. Even though we see a lot of evidence that are actually experts in unwinding these banks, it’s very hard to construct the case that the majority of these failures would have come from a bank run. In fact, it looks more like most bank failures were failures of already deeply insolvent institutions, even though, as I said, sometimes bank runs happened.

That’s telling you, well, a lot of historical bank failures in a setting in which bank failures should often be caused by bank runs, according to some of the models that we teach, tended to not really have some of the features that we would expect. Now, what we find then in this new paper, which I’m really excited about, is we merge the financial data, the call report data, with the data that tells us whether a bank run happened or not.

We just try to understand what are the causes of bank runs, when do they happen, and do banks fail when they’re subject to a bank run, because a bank run doesn’t need to be the same thing as a bank failure. I think here, the key takeaway is then we find that it is indeed the case that once in a while, there is a run on the bank that looks, in its call report, in its financial statement, perfectly healthy, although it’s much more likely to have a bank run in a bank that looks fundamentally weak according to its financial statement.

What’s really interesting, and that was really mind-blowing to us, is that if you’re a bank that’s healthy in your financial statement, it looks fundamentally sound, you can be subject to a bank run, but you will just not fail. That’s absent maybe the Federal Reserve, absent deposits, all of these government interventions. It’s a historical fact that when a bank was strong in its sense that it had good fundamentals, it actually could be subject to a bank run, but it would not fail. When the bank run happens, it kills some banks, but it kills the banks that have already very weak fundamentals. 

Beckworth: Yes. So much there, but at the end of the day, no matter how you slice or dice the data, what you’re telling us is that it’s ultimately about sufficient capital, fundamentals, solvency.

Luck: Yes. One of the striking findings in that paper, too, is that when you study the dynamics, let’s take a bank that’s subject to a bank run and ends up surviving the bank run. One of the key findings is that most bank runs actually don’t result in failure. They do quite a bit, but most of them don’t. When you survive, we can split the sample into banks that, before the bank run happens, have strong fundamentals and weak fundamentals.

What you observe is that a bank with good fundamentals is essentially going through the bank run in an unscarred way. Whereas if the bank has weak fundamentals, well, then the bank will actually lose deposits and loans permanently. It’s telling you something. Really, the state variable that you’re interested in as a policymaker is the fundamental solvency of the system because the better capitalized banks are, the more profitable they are, the healthier their business model is, a liquidity event on its own has just very little scope to do big damage.

In fact, in this paper, we could trace it all to real economic activity at the city level. We can trace manufacturing activity. Again, you’ll find this, if there’s a run that involves a failure or a run on a bank that’s weak, you’re going to see a decline in real economic activity. It tells you the financial shock is potentially feeding through to the real economy. If the banking system is in good shape, the liquidity event on its own is not going to leave a big dent. Real economic activity just continues as it was, even absent the bank run.

Beckworth: You’ve made a very convincing case that it’s weak fundamentals, it’s issues of solvency that nine times out of 10 or more is going to be the primary cause. Bank runs, the typical popular portrayals simply are misleading. Everything from movies, like It’s a Wonderful Life, to just, I guess, maybe our attraction to fear and psychology.

More fundamentally, in the profession, a very famous model, Diamond–Dybvig. I have to bring this up, Stephan, because there are so many people who will invoke that in a heartbeat, right? “Oh, that’s clearly a case.” Not only that, they’ll go to the Great Depression and invoke it there. I think what you’re telling us is that, no, not so fast, right? That model may be useful as an intellectual exercise, and maybe in a few cases. More often than not, that’s not really what happened historically.

Luck: I think that, to me, given what the data are telling us, it seems like a fair statement. It’s funny you bring up George Bailey from It’s a Wonderful Life because what’s really interesting is that we always cite George Bailey almost always when we teach bank runs. We tell the story of this movie. I just happened to watch this movie again this Christmas with my wife. It’s a really funny movie, except for the fact that George Bailey is just, as an actor, way too old for the character he’s playing. That’s just a side note.

What’s really interesting, what everybody just seems to have forgotten, and which our paper actually resurfaces because it’s in the newspaper articles, George Bailey is able to fend off the run. His institution, which is not a bank, it’s a savings and loan, but his institution is fundamentally solvent. People are panicking. That looks like a Diamond-Dybvig run. What people then forget is, in the Diamond-Dybvig model now, George Bailey would be fire-selling his assets. Illiquidity first would drive the insolvency, and then would lead to closure of his institution.

In the movie, what he does is he stands in front of the depositors that are panicking, and he calms them down. We find in our paper, actually, we asked the large language model to give us context as what did the managers do to prevent the bank run. How did the bank survive the bank run if it ended up surviving? What did it try to do even if it failed? You find a lot of exactly what George Bailey did. They tried to actually calm down the panic, if it’s a panic. They tried to get funding somewhere else. If they’re solvent, they’re able to do that.

I think Diamond-Dybvig is a fantastic model. I’ve loved that model all my life. It’s beautiful to teach. It’s a bit of the question, what do we do with it? It’s an intellectual exercise. Where do people take it? I would say yes. Maybe Diamond-Dybvig, we’ve taken a little bit too far in terms of bringing it into the policy work. For someone like me, who talks to policymakers on a regular basis, this is a bit frustrating if someone takes a very theoretical argument and tries to apply it into an actual real-life policy problem.

That’s, I think, where the data helps. I think the bank-run literature has had the issue that we just did not have enough data. We have some really great papers on bank runs using contemporary data. Manju Puri and Rajkamal Iyer have been really, over the last 15 years, the people who pushed this and had very successful papers. We also had some really great historical papers. Eugene White’s work comes to mind, and Calomiris, Mason, really good work.

I think they had actually more evidence on microdata. They had more bank runs and more bank failures than we have in the contemporary setting, but still, they were constrained a little bit by the data collection. You couldn’t just study the universe of all bank runs. The person who came closest, I think, is Elmus Wicker, who actually read the newspaper articles, as many as he could. He was just one person. He was a brilliant person, but he could just not do what the computing power can do in terms of reading every newspaper article.

Policy Implications of Bank Failures

Beckworth: Yes, he did extensive labor if you put into that project of his. Now, let’s talk about the policy implications. The clear one is there should be sufficient capital buffer. Banks should be solvent. They should be taking care of themselves, but it also raises the possibility that liquidity regulations really aren’t as consequential as we often think they are. Let me provide just a little pushback to that conclusion.

I’m very sympathetic to this paper. I know this is something you probably heard before. There was this paper that looked at the Great Depression and looked at banks in Mississippi, those that were under the Atlanta Federal Reserve, the sixth district. They were supervised by the Atlanta Fed. Then, I believe the St. Louis Fed supervised the northern part of the state.

It was a nice natural experiment. It’s the same state, same culture. Everything else should be the same, except for one part of the state’s regulated by the St. Louis Fed, and one’s regulated by the Atlanta Fed. It seems to suggest that there was a difference, right? The supervisory role played a difference to, I believe, access to the discount window. Now, am I interpreting those results as saying that liquidity does matter? If so, how would you respond to it?

Luck: Certainly, liquidity matters, and liquidity interventions matter. In this paper from Richardson and Charles that you mentioned, of course, if you have ample liquidity provision, you can reduce the number of failures. The way I think of it is it’s not a question of whether liquidity interventions—I’m not claiming they don’t matter. The question is whether they’re the optimal policy.

If we think of classic lending-of-last-resort doctrine, which says, “Oh, we should be lending to solvent institutions,” if you take some of my research in that context, the paper we’ve just talked about on bank runs with and without bank failure, there are banks that are solvent and subject to runs. What we find is that absent a lender of last resort, the solvent institutions are able to borrow from other institutions.

I actually have another paper with Markus Brunnermeier, who you mentioned earlier, and Kristian Blickle on the German crisis of 1931 on the papers titled, “Who Can Tell Which Banks Will Fail?” The German banking crisis of 1931 is one of the most massive financial meltdowns in history. It’s not just the banking crisis. It’s a currency crisis as well. It’s a political crisis. It’s a huge bank run on the banking system. It’s a collapse of the stock market.

What we find using microdata in that instance is that the banks that are actually solvent and subject to deposit outflows at the height of the crisis are still able to borrow from other banks that are having excess liquidity. That’s just telling you, if you came in with what Bagehot means to most people in the modern context—just a side note, Bagehot was really writing about a very different world than we have today. He was thinking about the gold standard and the convertibility of gold. Maybe to begin with, it’s hard to apply his insights to the modern data. 

Well, if we lend to solvent institutions, the counterfactual seems to be, at least that’s history is telling us, they would have been able to borrow anyway. It’s the lending to the insolvent institutions that, actually, the Bagehot would tell us not to do that actually can have big impact. Of course, you can prop up banks that are insolvent and keep them alive much longer with liquidity interventions.

The question is, is that the right policy? My research would say, “Well, maybe sometimes it’s the only feasible policy.” Fine, but the better policy would be to actually find out which institutions are insolvent and then recapitalize them, or if that’s not feasible, resolve them. That’s going to be a much more effective policy tool than making loans. By the way, this is the story of, in some sense, the Great Depression’s Reconstruction Finance Corporation.

It’s also the story of TARP a little bit. The key moments are when policy shifts away from providing liquidity alone, but to actually recapitalizing the banking system, is that’s when you start to resolve banking crises in a much more effective way. I’m not saying liquidity interventions don’t matter. The question is, more, are they actually giving you the bang for the buck that you’re seeking as a policy seeker?

Beckworth: Right. There’s scarce resources for the use of public funds, so how do we best use them? Let’s apply those insights to a more recent example. We had Silicon Valley Bank a few years back. How would you view that experience through the findings of your paper?

Luck: When we were presenting our failing banks paper, I would, of course, it was the timing, just required that you’d always talk about it in that context. The way I would look at Silicon Valley Bank, with hindsight, of course, is that this seemed like a failure very much the way banks used to fail in the 19th century, is that, here’s a bank. It has a ton of uninsured deposits. In Silicon Valley Bank’s case, it was almost all of them. It was deeply insolvent. I can argue a little bit why I believe that’s the case. 

Then the run should have happened for a long time, and it actually happened much later than you would have thought, looking at the data, in line with depositors being somewhat sleepy. I believe Silicon Valley Bank was deeply insolvent. The FDIC realized the loss of around $20 billion on it.

If you were to argue that Silicon Valley Bank was not fundamentally insolvent, you would have to argue that there’s some franchise value associated with their lending activity, of which they didn’t have a ton, or you would have to argue at the same time that their deposits, which were uninsured corporate deposits, had actually a very low deposit beta. Plausibly, it was very high. That just means the depositors were very sensitive to movements in macroeconomic conditions.

When the Fed funds rate goes up, they require about the same return. It’s plausible that those investors, which are highly sophisticated, would do that. The only way to argue they were not fundamentally insolvent then is to say, “Well, the FDIC’s auctioning process was really flawed.” It may not be perfect, but I doubt that it would generate, out of nowhere, a $20 billion loss.

That being said, I think if you, like me, believe that Silicon Valley Bank was deeply insolvent, in fact, was somewhat more insolvent about six months before it failed than when it actually failed, because if you look at the evolution of the 10-year Treasury throughout that time and the pricing of their Treasury securities, they were actually more insolvent in October of 2022. Then they look very much like a 19th-century bank failure. It’s like, this business model just did not pan out. For some reason, the depositors are taking their time to actually close the bank.

Beckworth: Okay, so your data is drawn from the 1800s up through the mid-1900s. It’s, again, worth noting that you draw these great insights pre-FDIC, so no one can say, “Oh, it was FDIC.” That’s the whole point of your exercise here. You can say what actually happened in practice when banks weren’t protected. There weren’t government interventions. Let’s come to the present, because this is probably a question you get.

Again, just to play devil’s advocate here, I know you’ve got a good answer for this. Today, we have these massive GSIBs, banks that are huge. We also have a global dollar system, right? Could it be that today with the global dollar system, we have a bank like Lehman who fails, it causes runs in money markets. The eurodollar market is hit. Commercial paper market’s hit, money market’s hit. Could dynamics be a little bit different today, whereas maybe we should think more closely about liquidity regulations and issues?

Luck: I think one real strong, important limitation to point out from using the historical evidence is that the historical evidence only speaks about commercial banking and not about all these other types of financial activities that we do see in the United States today, where we have these large financial conglomerates who do so much more than commercial banking; who provide market banking activities, investment banking, insurance, so all kinds of other activities. It’s very hard to just apply those insights that are really about deposit-taking institutions to these big, global financial conglomerates.

One thing I will say as a general thing that I’ve learned from doing the historical research is that the more sophisticated the investors are, the more they are able to understand between what is a pure liquidity event and what is an actual solvency event. That makes me think that in a world with informed capital, that’s the world where the private sector can figure out these things the best. It’s actually in the less informed space where I think the government has more scope to providing insurance and making sure we don’t have panic.

That will be the only thing that I would say is a general lesson, what I’ve learned from the historical evidence, I would apply here. That being said, I really want to be clear that you cannot use evidence from 19th-century national banks to think about the current financial conglomerates.

Beckworth: That’s a great point, though. Investors today are more informed than ever. If you were in a bank back in the 1870s, you may not have access to all the information, like what’s happening to banks in the northeast, or the south. This is a very different beast, both in a good way, and maybe in a bad way. You could argue there’s greater risk, but also there’s greater knowledge and information, more discipline from the market too.

National Banking System

All right. Let’s move on, but in the same vein. Let’s draw upon some history. Let’s talk about something else you’ve written on. It’s the national banking system. You can outline this more if you want, but at the time of the Civil War, the federal government ushers in a national banking charter for national banks. Maybe you can talk about its motivation for doing that as opposed to the state banking system that existed up until that point. You have a piece where you say there were implications from that development to stablecoins today, so maybe make that connection for us.

Luck: Yes, I’d love to. Let me just point out, the piece is on Liberty Street Economics, which is a blog post we have here on the research group at the New York Fed. Any of your listeners who don’t read it should check it out once in a while. It has a lot of cool stuff. This is an outlet where people like me sometimes just write up things that they’ve learned by thinking about ongoing policy issues that are not necessarily part of a bigger research paper.

What I’ve done last summer, of course, the GENIUS Act was passed, and established a framework for how we can have stablecoins in the United States. What was really interesting then when I was talking to my colleagues here in the group that are real experts on this, and I was reading up on the topic is, a lot of people would always make this comparison of stablecoins to the US monetary system we had in the United States before the Civil War, which was essentially a system, it was called the free banking era.

It doesn’t mean you could do anything you want, it just meant there was a very clear set of rules under which you could establish a bank. A tradition that was actually continued after the Civil War in the National Banking Era. Essentially, what happened back then is that each state had its own state-chartered banks, and the notes that were circulating in the economy were bank notes.

Banks would have to buy government bonds, before the Civil War, it was of the states. Then they would issue notes that were backed by these government bonds. But because some states were more dubious actors than they are today, they would invest in risky things such as canal building, et cetera, it wasn’t necessarily a risk-free asset to have a state bond. What would happen is that if there were concerns about the value of the state bond, then the bank notes essentially lost value because you were thinking about the bank going bust.

That’s why it led to these multiple different bank notes circulating in the economy, privately issued, having different values in different parts of the country. A lot of people were saying the GENIUS Act is going to lead to a system where we have all these different privately issued currencies circulating. It’s going to be this whole mess. I said to myself, “I’m not sure that is actually the right historical comparison. It strikes more that the better comparison is the National Banking Era.”

What the National Banking Acts, during and after the Civil War, did is they essentially took that state system but said, “If you charter a bank under national law, and you buy a certain type of government bond issued by the Treasury, then you can issue a national bank note.” Maybe for your listeners to really envision this, back in the day, the most common note that would circulate in the economy were notes printed by banks. In the sense that, the bank note did not have a president of the country on it, but the president of the bank that had issued those specific note types.

What’s really the cool thing about these national bank notes, they remind me a lot of stablecoins under the GENIUS Act. Essentially, what the bank would do, it would buy Treasuries, and then issue the notes. In default, the notes were fully backed by the Treasuries. Now, the bank would have a whole commercial banking business on top of that where it takes deposits, it issues equity, it makes loans, invest in securities, but there is this note-taking business that is part of the bank. These bank notes, of course, circulated widely. They were hugely successful because of the problems we had before the Civil War of not having this uniform currency.

Now, what are stablecoins under the GENIUS Act? Under the GENIUS Act, not all of it’s exactly clarified yet, but essentially, what we’re going to have is we’re going to have issuers issuing stablecoins that are going to then be backed by Treasury bills and notes, but also uninsured deposits, and potentially some other types of investments.

Here’s this comparison that you can make, which is, you’re now going to have an institution. It’s a private institution. It’s issuing a form of money that can be used. In the case of stablecoins, it’s going to be used in digital finance. In the case of the national banking notes, it was just used as the standard currency circulating. What’s in the interesting comparison historically, what does it tell us about the potential success of stablecoins?

If you look at national banking notes, they were a huge success, in that, you had a uniform currency. They grew quite a bit, but then they also stopped growing. Why was that? It was because the deposit that the bank would issue would actually, to some extent, be superior to the note, in that, it would actually be able to pay interest, and offer some certain types of payment services that the national bank note did not.

As over time, after the Civil War, the banking system grows, the corresponding network deepens, interbank transfers become possible, wire transfers become possible, deposits actually become the more attractive investment. Notes continue to grow, but somewhat slowly, and really deposits start to take off.

We have to be careful when applying very simple and historical comparisons, but there’s something here that resonates with me, in that, the payment system in the contemporary setting can actually compete with a lot of the services that stablecoins can provide, at least at the domestic retail level. Now, banks right now could offer you an insured deposit, so it’s safe, that can be wired almost instantaneously around the system and pay you interest.

Then you ask yourself, how’s a domestic retail investor going to benefit from a stablecoin? The historical comparison makes you think, well, domestically, it may not be as much. Now, that’s maybe also just because the stablecoins domestically are not solving a problem that the national banking notes were solving back then.

It’s a different question abroad. It could be that there are people abroad right now that are holding dollar notes because they are worried about the inflation risk, expropriation risk of their own currency, their own domestic currency, and that’s why they prefer to hold dollars. For them, it may be interesting to hold stablecoin. There’s some potential for there to be demand abroad.

Beckworth: Yes, that’s very interesting. I love how you made that comparison historically to that period. I agree with you. I think domestically maybe tokenized deposits will be the hot thing if that. Clearly, I think that stablecoins’ future is in cross-border payments in places with less stable monetary regimes.

You mentioned how deposits eventually became more and more highly demanded relative to the notes. It’s simply competition of the two types of assets that led to this outcome. Let me throw in another possibility. I believe I read somewhere in the past that also during this time, the US government was beginning to pay off its debt, so the stock of debt went down. Therefore, banks couldn’t get the collateral to issue it.

If I use that, if that’s the case, I may be wrong, that clearly won’t be an issue today, right? We’re going to have plenty of Treasury bills that will be issued. If anything, it’ll be the other way around. We will be supporting the US government in its issuance of T-bills.

Luck: Right. Back then, it’s quite interesting the way that under the National Banking Era worked was that the Treasury issued bonds. Then a subset of those bonds were eligible to be used as collateral to secure note circulating. There’s this interesting fact that the amount of notes issued never reached the limit of the amount of available collateral. There are some technical reasons why this may have not been the case, but really this is telling me that this wasn’t a constraint.

Another part of the historical parallel that’s really interesting is that during the National Banking Era, there’s, of course, the Federal Reserve doesn’t exist. There’s no central bank, so seigniorage is really being earned by these private issuers. In that case, the national banks. In some sense, you could think of national banks before the Federal Reserve as each bank being a private bank, but also a tiny central bank that gets some of the seigniorage.

This is another parallel to the stablecoins under the GENIUS Act. In some sense, we’re allowing now private sector investors to earn some of the seigniorage that comes from issuing, in this case, digital money. That’s going to be important to think about what does that mean for, even if we can crowd in demand for Treasuries by, say, foreign retail investors holding US stablecoins, it may have implications for seigniorage, which, of course, also matters for the Treasury. There’s going to be some interesting things for us to observe once stablecoins really take off, how that’s going to affect the deficit.

Beckworth: I’m glad you brought that point up because that was also the key insight as well from your paper that this is just replicating that experiment in terms of it created demand for Treasury. Back then, as you note in your piece, one of the motivations for the National Banking Era, one was we want a uniform currency, sure, but another one was that we got to find someone to buy our debt. We’re funding a civil war. We need to get a captive audience. Let’s force banks into buying these securities. The analogy today is we’re running large, large deficits. Anything else that can help? This won’t solve our problem, but definitely extend the runway. There is that analogy. 

German Hyperinflation

We’ve been talking about balance sheets to really underlining all these conversations so far, Stephan. I want to go in the last few minutes we have to a paper that you have written with some co-authors, including Markus Brunnermeier. You alluded to this earlier, I believe, on the German hyperinflation and this debt inflation channel. Walk us through the bird’s-eye view summary of this paper.

Luck: Yes. Let me tell you about that paper. This paper is a nice example of how you can use historical data to learn something about a fundamental question in macroeconomics. In this case, it’s just about how do large inflationary shocks transmit to the real economy. It’s just something we can’t observe. Large inflationary shocks is just something we don’t observe in developed economies over the last 40, 50 years. This was a really good laboratory for us. Besides that, it’s really one of the most iconic events in macroeconomic and financial history.

What we uncovered when we were studying the episode is that there is something like what we call the debt inflation channel active in the transmission of monetary policy to the real economy. What is the debt inflation channel? It’s just essentially the reverse of Irving Fisher’s debt deflation channel. It’s just capturing the idea that when you have a large inflationary shock unexpected, you’re going to transfer, you’re going to redistribute away from debtors toward creditors, especially if financial contracts are long term, fixed rate, and not indexed to inflation.

One of the things we here studied is digitized microdata at the firm level. We understood what did their liability structure look like, to which extent had they issued bonds, to which extent were they long term. What we found essentially as the inflation was picking off after World War I, it’s those firms that had a lot more debt going into the inflation that actually contributed more to the economic activity during the inflation, in the sense that they invested more and they hired more people.

It solved a little bit of a puzzle that you get at the time series that if you study the German hyperinflation, it’s actually this very long period that runs from the end of World War I to essentially the fall of 1923. For the first three years, the inflation is considerable. It’s very, very high. The price level goes up by a factor of 30 or something until summer 1922, but the economy is actually doing really, really well in the sense that unemployment is low, investment is high.

It doesn’t mean that people weren’t miserable; they certainly were because inflation at that level is highly corrosive and people were spending a lot of their resources in just figuring out what the prices are. But at some aggregate level, the economy seemed to be doing good.

What we found is that, that cannot be explained by classic new Keynesian channels. We found that actually, at that time, wages were already flexible, prices were already quasi-flexible. What we figured then is that the financial friction may actually be what’s driving this ability for the inflation to still be having a positive effect on economic activity just for the mere fact that if I was working for Macro Musings, you’d be paying me a salary. If we woke up tomorrow and there’s a big inflationary shock, I would just come and say, “David, I would like to have a higher salary. Otherwise, it’s just not worth my time to work for you.”

If I was a bank and I lent to you and there was a big inflationary shock, I couldn’t come to you and say, “Well, you know what, David, inflation actually went up. I think we should change the terms of our loan.” Through that simple logic, large inflationary shocks can just have a much larger effect through these financial channels. That’s really, I think, what we found in that paper.

Beckworth: One of the big takeaways from that paper, at least for me, and correct me if I’m wrong here, is that sticky nominal debt contracts really matter. It’s an important rigidity in the economy. Is that fair?

Luck: It certainly was an important rigidity in the context of the German hyperinflation. I just want to be giving some context on why they probably were more important then than they are today, in that, the German hyperinflation happened after a period of extreme price stability. Before World War I, of course, the major economies are on the classic gold standard, and the aggregate price level hardly moves.

As firms are issuing bonds before World War I, they’re issuing bonds at times with a maturity of 100 years and interest rates that are fixed. We actually found, when we looked at bond data, there’s actually only three prices for bonds, three interest rates quoted. It’s 4%, 4.5%, and 5%. That’s all there is. There isn’t any further variation. Those kind of financial contracts, they really created the potential for the inflation to clean the slate and really relax the financial constraints on firms.

It’s a bit different today, where in the US especially, a lot of debt is floating rate, and it’s much more short term. For the German hyperinflation, we sure believe that the financial frictions mattered quite a bit.

Beckworth: There’s shorter debt, and we definitely don’t have that kind of price stability, like prices are flat. We do have well-anchored inflation expectations, for the most part, right? We still see 30-year mortgages and long-term debt. There’s still enough long-term debt, though, where this should be something of a meaningful friction, right?

Luck: Yes. I certainly believe that it’s a possibility in the sense that it’s something we should study. Some people have studied. It’s not that it’s fully understudied. I think we’ve focused a lot on the new Keynesian frictions. I think our paper illustrates that there is a potential for these other frictions to matter. Now, I haven’t done research myself that suggests that in the current US context, it is a key driver, but it certainly potentially is.

Beckworth: Okay, so I’m pushing you on this because I have a point I’m trying to make here. You’ll see what it is. If you’ve listened to the show for a long time, you’ll know where I’m going at some point.

Let’s just imagine a world, Stephan, where aggregate demand is relatively stable, so the central bank is doing a good job offsetting demand shocks. As a result, the only inflation—and these won’t be big shocks, they won’t be like hyperinflation—but the only inflation shocks are coming from supply shocks. Again, assuming the Fed, or whatever the central bank is, is offsetting demand shocks.

Aggregate demand is kind of neutralized, it’s stable, but supply shocks are still present. As a result, if inflation is just coming from supply shocks, inflation becomes countercyclical. To be concrete, if there’s a negative supply shock, a recession happens, inflation temporarily goes up. Conversely, if it’s a positive supply shock, which it’s AI right now, you get some disinflation, you have a boom. So inflation becomes countercyclical case. We get that.

If that’s the case, and if there are meaningful sticky nominal debt contracts, well, then the real debt burdens become procyclical, so countercyclical inflation leads to procyclical real debt burden. During a recession, inflation goes up, and the real debt burden goes down, but during a boom, the real debt burden rises.

What am I going through all this? Because this is a story that Jim Bullard and a few others have told for why nominal GDP targeting could actually enhance financial stability. I had to bring this plug in here at the end. The issue is how important is this fixed nominal debt rigidity. If it’s not that important, this is not a great story, but if it is important, then what happens is, by stabilizing aggregate demand, you will effectively be turning debt contracts into something that acts more like equity. It adjusts because you get this procyclical real debt burden.

I guess the question then ultimately is, how meaningful is this rigidity? There are other rigidities—sticky wages, sticky prices, sticky information. I’ve kind of drunk the Kool-Aid element that there’s sticky nominal debt contracts, but I want to be fair. What is the empirical evidence? You’re saying it’s not clear yet how important that is. It’s there, but it’s not going to show how important it is.

Luck: I think, for me, what’s clear is that, in a pure qualitative sense, these frictions matter in the sense that they exist and there is redistribution. For this friction to actually matter for the macroeconomy, you need to have two things. You need to have agents, if it’s a consumer or a firm, a household or a firm, they need to be substantially constrained to begin with, for there to be an effect, once you relax the financial constraint with inflation, or you tighten it with deflation. That is a bit of a question that’s not always fully clearly answered to me. To what extent who is constrained and when, and what are the effects of the actions?

The second effect we haven’t really talked about is that this is just sort of all partial equilibrium that we’ve talked about. There is always the flip side. In the debt inflation, we’re going to have the creditor winning, but you have the debtor losing, and that, of course, matters in GE. That’s the part that I just have to be fully honest, I just don’t have a full understanding of. By the way, this also applies to my understanding of the German hyperinflation. We really focus on the benefits for the creditors, but we really have a hard time quantifying the losses for the debtors

I think the sum of all of that is going to tell us whether this channel matters. I think more work is needed. I just got sidetracked with bank runs after the hyperinflation paper, so maybe I’ll get back to this some other day.

Beckworth: No, but this is all good. Again, all your research really goes back to this core issue of financial stability and how do we maintain it and improve it going forward, the importance of balance sheets. As you said, your origin story was the Great Financial Crisis. This is what really spurred you into this research agenda, into your career. Here we are, it’s been a while since then, do you think, in closing, that we’re doing enough research and thinking about these financial stability issues?

Luck: I’m very optimistic about the future in the sense that we have all the evidence, or more than ever, we have evidence that we need, in part by unlocking these historical settings that can guide us. I believe we’re doing a lot, and we have everything we need at hand to do it good and do it right.

Beckworth: Okay, our guest today has been Stephan Luck. Stephan, where can people find you online?

Luck: Well, you can find me on my New York Fed, or you just Google my name, you’ll find my Google homepage. Let me put in a plug. I have a homepage together with my colleagues Sergio and Emil, which is finhist.com, where we provide vast material on bank runs historically, but also bank balance sheets, so anyone interested in that topic should check that out.

Beckworth: We’ll provide links to that in the transcript as well. Stephan, thank you so much for coming on the podcast.

Luck: Thank you, David. It was a pleasure to be here today.

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