 Okay, welcome back to the last session of this conference here today. We have two more papers coming, very topical papers. And the first one now is going to look into the future. The first one is about predicting GDP and why we think we can predict it. And the second one, we'll look into the past actually going back to 1500. So within a little bit more than one and a half hours, we're going to spend 500 and more years of data. That's the last thing before we then close and can all go for lunch. So we start off with you, Camelia. We know you from the Federal Reserve Bank of Atlanta with the paper on why does the yield curve predict GDP growth, the role of banks. Floor is yours. Hello, everyone. Thank you so much for having us. This is a joint paper with Andres Schneider and Minwe from the Federal Reserve Board. The paper is titled, why does the yield curve predict GDP growth? We're going to examine the role of banks. And this work, with us being affiliated with the Fed, is subject to the usual disclaimer. So we're motivated by the following fact. The yield curve has been an enduring predictor of GDP growth. But where this predictive power comes from, it is not quite clear. On the one hand, or one possible explanation, is that the yield curve reflects investors' aggregations of their impressions about the future state of the economy. For instance, if investors expect a slowdown in the economy, then the central bank would respond to that slowdown by lowering interest rate, short-term rates. And we know that short-term rates affect long-term rates. So that would mean a lowering of long-term rates, which would lower the yield curve. Now, if the yield curve is an aggregation of investors' expectations of future growth, then the yield curve and expected growth would be casually related. But it could also be that long-term rates respond to changes in the term premium. In that case, it's possible that the yield curve predicts future growth with a causal element. And it is this potential causal channel shown in the second part of the first line that is our interest. In particular, we want to examine if banks play a role in this potential causal channel by which the slope through the term premium fluctuations may affect future growth. So to give you the full chain of causation, we're interested in how fluctuations in the term premium affect the slope. In turn, the slope affects expected bank profits and lending. And finally, that increased bank credit affects GDP growth. For the chains in this causation link, rather the links in this causation chain A and C, there are many papers. Our focus will be on link B. The causality running from changes in the slope via the term premium on bank profits and lending. So the intuition in a snapshot is that higher term premium implies higher expected profits on maturity transformation on the core business model of banks incentivizing lending. So our paper is naturally related at least to literatures. One is the predictive power of the yield curve for GDP growth. Unlike this literature, we emphasize the role of banks. We zoom in on the role of the term premium and document the specific expected profitability channel. Secondly, we relate to the literature on banks' exposures to interest rate risk that typically focuses on short-term interest rates. Here, we look at banks' exposures to interest rate slope risks. So let me start with some very simple motivating evidence with aggregate data. This is just a number of time series regressions. In this time series regressions that run all the way back to the mid-'70s or the mid-'80s, we relate the term spread to a four-quarter-ahead GDP growth, bank loan growth, net interest margins, and ROE, so two profitability metrics, in the columns 1357. And then in the alternate columns, we decompose the term spread into the term premium and expectations. We always control what happens at the short end of the yield curve. And we see here positive correlations between the term spread, as well as the term premium component of the term spread, and four-quarter-ahead real GDP growth credit, as well as bank profits. So this tells us that perhaps there is that causal channel in the data. But of course, to establish something a little bit more credible, we need to take identification seriously. So before we move to the data and to the main results, let me give some intuition about what's going on here. We develop a model in the paper to sort of organize our thinking around the topic. This is a partial equilibrium dynamic banking model, where banks take leverage positions in assets, in particular loans, that are exposed to term premium and interest rate shocks. And there are two insights that follow from the model. One about the level effect of the term premium on bank lending, and the other one about a differential effect. The first insight is that an increase in the term premium implies greater expected profits on maturity transformation. That is an increase in banks' expected return on wealth. When we take this to the data intuitively, the expected return on wealth is the return on equity, which is profits over equity, so profits times leverage. So expected return on wealth increases in the term premium and in bank leverage. The level prediction of the model is that an increase in the term premium by raising expected profits on maturity transformation will lower the financial constraints of the bank, because the bank, by making more profits, will build capital and therefore ease its financial constraints, boosting its incentive to lend. The differential prediction of the model is that the expected return on wealth increasing in the term premium and in bank leverage from that simple ROE, so increasing function of the term premium, because of profits times leverage. That means that a leverage bank should be in a better position to take advantage of an increase in the term premium. More leverage banks should exhibit a stronger response to an increase in the term premium relative to a less leveraged bank. So I can take two testable implications to the data. First, the banks will respond to a rise in the term premium by increasing lending, in particular the supply of new loans. And so importantly here, we're going to have to somehow empirically shut down any effects of the term premium on loan demand. And secondly, more leveraged banks should exhibit stronger response to term premium fluctuations. That's the differential effect. So we have two tasks. We have the task of establishing or documenting in the data a level and the differential effect that would be consistent with the model. So let's move on to identification. I already given a flavor of the identification challenges. The first one is a typical omitted variable bias. The endogeneity issue here is that growth expectations affect both the slope or the term premium and bank lending. So this is a typical omitted variable bias. Both the left and side and the right hand side may be driven by growth expectations of investors and of agents in the economy. So we need to find exogenous variation in the term premium. We need to orthogonalize it with respect to expected future economic outlook. And the second challenge is to separate credit supply from credit demand defects. And in particular, here, we're going to deploy some loan level data that will allow us to hold credit demand at the firm level fixed. So the identification strategies we will deploy are manyfold. There are maybe two or three. So we're going to give you a portfolio of evidence that hopefully corroborates the testable implications of the model. And in a sense, I invite you to pick your favorite and to criticize the one that you don't find convincing. The first one is that for the term premium itself, which is a non-observed variable, we're going to have estimates from a well-established model. We use high-frequency term premium shocks. So defined in a way that is reminiscent of how we define monetary policy shocks, short-rate shocks. Then we're going to have an instrumental variable for changes in the term premium. So in alternative specifications, we're going to use the 14 official holdings of US treasuries. These are 14 central bank holdings of US treasuries normalized by US GDP. They are inversely related to in the time series, with inversely related to the term premium. And the identifying assumption is that these holdings are orthogonal on the US economic outlook. And rather, they are driven by central banks in international reserve management, considerations related to exchange rate management, and considerations, therefore, unrelated to the US economic outlook. Then we use the taper tantrum of the summer of 2013 as an event study because the taper tantrum gave rise to a remarkable and fairly persistent rise in the term premium that one might argue was quite unanticipated. And finally, as I mentioned before, we're going to deploy some low-level data to try and control for low demand in a way that is quite standard in the empirical banking literature with firm time fix defects, while at the same time controlling for two additional variables. So even if you're not convinced by any of these identification strategies, we have one more strategy in our sleeve, which is to control for bank-level growth expectations directly for those banks that report in the blue chip surveys. They report their one-year ahead GDP or CPI forecasts. And finally, we sort of control for the standard banglending channel by interacting high-frequency monetary policy shocks, so shocks for the short-rate with a number of balance sheet variables in all of our specifications or just directly in the time series specifications. So without further ado, let me go straight into the data and the results. The data come from the US credit register. These are big words. The credit register is actually a slice of the true credit register in the US, but it covers three quarters of the banking sector loans. This is the Y14 regulatory data. These are loan-level or bank-firm-level loan exposures that we observe as quarterly snapshots since 2013 until today. They are reported by the large stress-tested banks, and so we only look at the 15 banks for which we have data on growth expectations from the blue chip surveys, because it's so critical to control for those expectations. An advantage of the data is that it includes a lot of privately-held firms. To altogether, the firms account for 60% of non-financial business debt. There are a lot of bank-dependent firms in there. Then we have the US call report, and then we have the IV that we draw from alternative data sources. The most important element here is the term premium, an unobserved quantity. We take estimates from the Kimbrite 2005 term premium model or term structure model that was fitted to US Treasury yields since 1990. The way we measure high-frequency shocks of the term premium is to look at changes in the Kimbrite term premium on FOMC event days, and then aggregate those changes at the quarterly level. Let me go straight into the results. I'm going to have bank-level data results and loan-level data results. In the bank level, we are going to estimate the level effect of the term premium. There's not many options here. Here, we either use shocks with OLS, that's column three and six, or we use an IV, columns two and five, or we just do a simple OLS. Do not compare the estimates across columns, so they are not comparable. But what's important here is that we see a positive association of the term premium, controlling for expectations, controlling for changes in the short-rate, a positive association with loan growth, with and without credit lines. Now, of course, it's difficult to control in these regressions for loan demand. One very simple way is to put in bank MSA fixed effects to the extent to which banks are exposed to local markets, the MSAs where they operate and they do local lending, then those would absorb some demand shocks. But we'll do a little bit more in the loan-level data. Okay, so let's move to the loan-level data and let me show you two sets of results. First, we're going to do the event study of the taper tantrum and then zoom out from the taper tantrum period to the entire credit registry to see if we can establish more general patterns. So let's look at the taper tantrum. What happened then is that in May of 2013, former Federal Reserve Chair Ben Bernanke gave a speech in the Q&A session following his congressional testimony. He mentioned or he noted that the Fed would start tapering, it would slow down the pace of asset purchases at some future date without being precise about the date. And what happened subsequently is that market participants interpreted this speech as injecting quite a bit of uncertainty about monetary policy. And the term premium rose sharply and based on the Kim Wright model, that rise was pretty sustained over a few quarters. So we're going to zoom in on the quarters before and after this significant rise in the term premium. We're going to think of it as unanticipated and we're going to compare lending outcomes at banks before and after this significant rise in the term premium in a very standard different deep specification. Here it is. So we're going to regress loan outcomes at the bank firm level over a period of three to five quarters around the taper tantrum, not one and two because it takes time for banks lending decisions to respond to any financial or real shock and not more than five quarters because then we allow, by extending the window too much, we allow maybe contamination effects from some other events in the economy. So we look at loan outcomes in this window as a function of the bank's ex ante leverage, capital ratio, interacted with a post dummy for the post taper tantrum period. Very standard different deep. Two or three critical ingredients in this specification are that first we control for bank level one year ahead GDP growth expectations or forecasts from the blue chip surveys both before and after the taper tantrum. So clearly with the taper tantrum, with the new information they received from this speech, banks anticipations of where the economic outlook was going may have changed. So we have those, that variable in level and interacted with post. And then we also control for firm quarter fixed effects which allows us to draw identification from comparing banks with different exposure, if you will, to the term premium, the term premium rise from the taper tantrum episode. So to exploit heterogeneous exposure to the term premium in terms of leverage and compare the lending of banks with different degrees of ex ante leverage to the same firm and in the same quarter. We see how those decisions differ. That's where the identification comes from. So the results speak pretty much consistently or they corroborate the differential implications, the implications of the model in regards to the differential effects of bank leverage. The negative coefficients in the first three columns tell us that more leveraged banks reacted more. They increased lending even more than less leveraged banks after the taper tantrum compared to before and compared to less leveraged banks. Similarly, they also decreased loan spreads more than other banks. So we have strong evidence of differential effects by initial leverage and like I mentioned before, we control for a lot of different potential confounders here. So we're fairly confident that the evidence suggests that banks adjusted their lending decisions on both the extensive and the intensive margin following the taper tantrum and as a function, the response was heterogeneous according to their leverage level. Okay, so now let's zoom out to the entire credit registry and see if what we've identified by way of patterns in the taper tantrum episode generalizes to the entire period 2013 to 2019. So we run, now this is not a different if anymore, it's a very simple sort of panel regression if you will, but again at the loan level where we relay the lagged capital ratio in interaction with term premium shocks to the extensive and intensive margin lending decisions of the banks. And again, we find that more leveraged banks increase lending more. They are more likely to give new loans or for accepted loans, they are more likely to give larger loans than less leveraged banks and similarly to decreased spreads. So here we control for demand with firm quarter fixed effects where I'm hoping that these are somewhat convincing results that they're picking up the credit supply effect. We have quantity and prices also going in opposite directions. And importantly, we take every single bank control and interact it with term premium shocks with expectations and with a short rate. So it's a pretty demanding specification. Now something I have not mentioned so far is that in this somewhat black box of bank controls, a critical bank control is the share of securities to assets. And the reason why we think it's critical is because if you're familiar with some of the recent quantitative easing literature, an important channel of transmission of quantitative easing activities and in particular the flattening of the yield curve that they achieve, the lowering of term premium that they achieve is through valuation effects. They're also called windfall gains or windfall effects. Basically with interest rates going down, the value of the securities held by banks increases and that is a phenomenon that Brunermeyer refers to as stealth recapitalization. It's been documented both for Europe and for the US. And so controlling for that valuation channel is very important. It goes in the opposite direction of our channel but it is very important to control for it. Okay, so how can banks take more interest rate risk, exposure in face of a positive term premium shock? Two ways. They could increase the maturities of the loans that they make while holding leverage constant or they could increase leverage while holding maturities constant or they could do a little bit of everything. And we have evidence for both types of reactions. I wanna show you here the one regarding extension of maturities. More leveraged banks increase the maturity of their new loans even more than less leveraged banks after a positive term premium shocks. We see this for a number of alternative ways of looking at loan maturity. So to wrap up the empirics, I'm gonna speak about the impacts of the term premium on bank profitability. So this is the last table. And so here for this exercise, we go back to the long-term bank panel and we have our three identification strategies, OLS, OLS with term premium shocks and IV. And we relate the term premium to net interest margins and return on equity. We see a strong positive association. Again, this is suggestive that the channel by which banks increase lending and in turn this has positive real effects. The channel is worked in the face of a positive term with premium shock works through bank profitability, increased bank profitability. Of course, it's expected profitability. So we're looking at four quarter ahead outcomes, okay? So the economic magnitudes are reasonable. I'm gonna just talk about the level effects of the term premium. A one standard deviation increase in the term premiums which roughly 50 bips raises loan growth by 1.1 percentage points, bank names by four basis points, bank ROE by 32 basis points. The numbers are meaningful. They are not negligible. They're not implausibly large either. Okay, so I do have a little bit of time and I want to run through a number of additional results and robustness checks. I think a very natural question will be about the identifying assumption on our instrument. Our instrument recall is the holdings of US Treasuries by 14 central banks and the identifying assumption there is that those holdings, the demand for the foreign demand for US Treasuries from foreign central banks is orthogonal on the US economic outlook. You may not believe that much. You may say, well, for China especially, the Chinese business cycle is correlated with the US cycle and so that identifying assumption may be tenuous. We can drop the China's holdings from US Treasuries from the construction of the instrument and the results go through. We have placebo tests for the taper tantrum but perhaps more importantly, the taper tantrum came on the back of several waves of quantitative easing by the Fed which had increased the level of bank reserves and we may all have different opinions about how bank reserves affect the supply of loans. The empirical literature is somewhat undecided on that but controlling for bank reserves for that huge accumulation of bank reserves before the taper tantrum does not seem to affect the results either. I motivated the work by saying that we wanna look at the channel by which the slope affects real GDP growth and so in all of these credit regressions you may ask, are there real effects? And indeed, although I'm not showing it, there are real effects after the taper tantrum. As I mentioned in the credit registry we have a lot of bank dependent firms and those bank dependent firms that are more reliant on leveraged lenders tend to have better real effects, in particular higher investment rates after the taper tantrum than other firms. So there are real effects after the taper tantrum that are bank dependent firms in the US, especially among the privately held ones and they do see that there is an impact on growth from better availability of credit. And finally, I have some additional mechanism evidence. As I mentioned, there are different ways by which banks can take more interest rate risk. One is to leverage up, the other one and hold maturity is constant or the other one is to take more duration risk while holding leverage constant and we have evidence for both of these channels. So I will wrap up by discussing sort of kind of circling back a little bit to the quantitative easing literature because I wanna give some flavor of some policy implications for this work. And so I think it's important to note that we don't have different interest rate regimes in the model, the model is estimated, is calibrated for normal interest rates about not at the zero or lower bound. And so in order to very carefully speak about quantitative easing implications, what happens at the zero lower bound, we would have to do the impulse response functions for interest rates at zero, conditional on interest rates at zero versus otherwise. So and one difficulty with that as we think through the policy implications is that in a zero interest rate environment, bank behaviors may change. One reason is that profitability is just generally depressed and so banks may try to take more risk, they may develop stronger appetite for this, they may try to go up the duration ladder, the credit ladder and there's quite a bit of evidence to that effect. In fact, this is what in the literature is referred to as portfolio rebalancing effects. In some work with Margherita Botero and co-authors, we show that when negative interest rates were implemented in that environment of depressed net interest margins, banks took more risk. They lent more, but especially to riskier firms and they swapped out of low yielding assets for loans and other higher yielding assets. And the other, so this is one consideration that kind of gives us pause and a little bit of caution in thinking about the policy implications of the work, but I would like to say that our work draws our attention to a channel that suggests countervailing or counteracting events effects relative to the valuation channel of quantitative easing. As documented empirically in Chakraborty et al, Acharya et al, Rodniansky and Darmoni, we control for this channel, we don't do a horse race, we don't try to quantify one channel over the other, we would need to do more work to say something more confidently about that, but at least intuitively our bank profitability channel suggests counteracting effects and therefore it has implications for how we think about balance sheet management, about calibrating asset purchases and the pace of quantitative tightening. So to conclude, I hope I have convinced you that we have dug out an expected bank profitability channel by which the slope of the yield curve affects economic activity through bank lending. Our strategy was multifold in terms of identification, we threw a lot of data at this question and the headline result is that there is indeed a strong positive link between term premium shocks and bank lending with possible effects on GDP growth and therefore this causal channel offers some insight as to why the yield curve is such a strong predictor of GDP growth. All right, Camelia, thank you very much. Perfect timing. And the discussion is Clancy Happens, here from the ECB, the floor is yours. Thank you, so good morning, everyone. So this is really a very interesting paper, it was a very nice read. So I think Camelia did a great job in making abundantly clear that what this paper is really about is a question, why does the slope of the yield curve is typically associated with future growth or put differently, why is a steeper yield curve typically associated with higher future growth and vice versa, why is an inverted yield curve often seen as a good predictor of upcoming recessions. Now, the traditional answer to that would, I think very often be something along the lines of, well, this is basically spurious correlation because what the yield curve captures in the end is changes in monetary policy and the views of markets and market participants regarding future economic conditions, regarding future short-term interest rates. So obviously it's gonna tell you something about future economic growth. Now this paper takes a bit of a different stance, it's not going to say at all, it's not going to say that this spurious correlation is not there, no, it's just going to argue that there's something more to it and that's their expected bank profitability channel. So their point is going to be that a higher term premium leads to higher growth via this expected bank profitability channel and they'll proceed in two steps in their analysis. So first, they're gonna decompose long-term rates in an expected short-term rates part and a term premium. So remuneration that investors get for holding on long-term investments, long-term bond in this case. And this is nothing novel, they just take an on-the-shelf novel to do this decomposition and there's nothing wrong with that. And the real novelty about the paper is really in this expected bank profitability channel which is going to tell you that an increase in the term premium will lead to an increase in the expected interest margin for banks which will reduce financial constraints for banks, increase their bank credit supply and in this way lead to more economic growth. So if you think about it, this is kind of a big thing because it implies that the slope of the yield curve doesn't simply reflect but actually causes changes in future economic activities. So this is very different from how one traditionally tends to think about this relationship and that in itself makes this very interesting and very thought-provoking paper. So it's definitely worth your time reading it. Now, as with every empirical analysis, of course it's gonna be very challenging to establish this type of causality. And I think this is even more true for this paper than for others, exactly because of the fact that your yield curve is so correlated to so many economic factors out there. So a lot of my present discussion today will focus on that. But overall, I think this is really an excellent paper. So a very nice read, very much enjoyed reading it and what I'll try to do during the discussion is play a bit of devil's advocate regarding the mechanism you have in mind and regarding parts of the empirical analysis. So let's look a bit closer at this mechanism. So what our mechanism is going to say is that an increase in the term premium will lead to an increase in the expected and that's interest margin for banks and that's simply because there's the assumption that banks are funding themselves short-term and they're lending long-term. So this will increase their expected profits, reduce their financial constraints, increase credit supply and in this way have a positive impact on economic growth. So the last part being a sort of standard bank credit type of channel. And this is somewhat related to a recent literature on the yield curve and bank profits but the first question when I saw the channel that really came to mind is okay but why the strong focus here on the term premium and not simply the slope of the yield curve for this mechanism because in the end if you're talking about net interest margins for banks what they're interested in is the spread between long-term and short-term rates. And this is something that isn't really clarified or that isn't really clear in the current version of the paper. Now I think one way to think about it is that what they mainly have in mind is interest margin and profits on new loans. So of course if the slope in the yield curve increases due to an increase in expected short-term rates then this implies that given that these loans are funded by rolling over a short-term debt that your funding cost also increases. So in the end your net interest margin doesn't change in this case and that's why it's important to focus on the term premium here. I'm not sure whether that's the exact mechanism that you have in mind but if it is this then it's also I think very important to realize that via the same type of channel. Of course a bank also has a loan's outstanding, a large loan book outstanding and for these outstanding loans if the expected short-term rates are going up this means the expected margin on your loan book is going down which reduces profits and could have the exact opposite effect of or you could get to the exact opposite type of channel here. So the interesting thing is that so this implies a negative coefficient on your expected rates premium in your regressions and that's exactly what you get in a lot of the regressions that you are showing but you're not giving it a lot of attention in the paper at the moment while I think this is really important because if you take your mechanism seriously it should also be at work for these expectations components and secondly if the big question that you're after is really about how the slope of the yield curve predicts economic growth it's very important to say something about the relative importance of the impact of this term premium and the impact of these expectations components due to your channel. So that's I think really something important to think about for the paper. Why is this expected rates components? Why does it have a negative impact and why is it less important for bank loan granting because that's the impression that one gets currently when reading the paper. Now a last point on the relation between the term premium and credit growth so I think your regressions are very convincing in showing this positive relation between both. At the same time if you look at say aggregates more long term time series data so this shows the five year term premium that you're using between early 70s and today and the credit growth in the 12 months after for the US then indeed you see this very strong positive relation in the 70s, 80s, early 90s but later on in the sample this breaks down during some periods. I think this is a very interesting observation and could give you a bit of interest could give you interesting information about what you're really capturing with your mechanism. So one explanation there is that of course there might be different drivers over time of the term premium and some of these drivers might be more directly correlated with credit growth then than others. You gave the example of QE already during your talk QE will reduce the term premium and might activate your channel but at the same time there might also be the stealth utilization that has an impact on credit supply. Now what I'm trying to get at here is that the importance of your channel in the end will depend on what drives the term premium and how correlated the driver is with future growth and I fully understand that with your IV setup you try or the other identification setups you try to tease out say the pure term premium effect and this is very interesting because it can tell you something about the causality of your term premium but at the same time it doesn't necessarily it's not necessarily informative about the relative importance of this channel. So put differently what fraction of the correlation between the slope and growth in a couple of months is really driven by your mechanism and what's driven by the type of spurious correlation that people typically think about. I think that's an important thing to think about for the paper and to say something about how important this mechanism that you're capturing can potentially be. Okay then you already alluded to this in your talk this IV setup. There's a number of things that I'd like to say about that as well. So the problem at hand is that okay obviously if you're interested in the causal impact of the term premium, I mean you're very much aware that this could be driven by factors that simultaneously correlate with loan and GDP growth and you have multiple exercises that try to fix this. One of them being this IV setup and I think this IV setup is in a way the least convincing one here. So you're gonna use foreign official holdings as an instrument variable and as with every IV setup there's basically two assumptions that need to be satisfied. You need a strong correlation between the term premium and foreign official holdings which you show kind of convincing me in the paper although I do think that also here this might be this time variation in this relationship which might make this IV a better one during certain periods compared to others but okay that's just a minor concern. The larger concern is about the exclusion restriction here. Like as you said in principle if it's a good IV it should be uncorrelated with US economic conditions and that I think is a lot more questionable because you could come up with a number of examples that might violate this assumption. For example, there's evidence out there that there's a strong correlation between official holdings of US debt and trade between the third country and US. So any economic shock or large economic shock that's happening in the US will most likely have a strong impact on the trade partner as well will affect its dollar reserve, its dollar holdings and hence its foreign US debt holdings because that's one of the places where it's going to be investing these dollars. So that already might kind of be problematic for the instrument and in a very similar way you could come up with a number of additional examples. For example, a weakening US growth outlook leading to a dollar depreciation which then could lead to foreign exchange interventions by central banks of a third country to prevent their currency appreciating against the dollar. I think a well-known example is Japan in the early 2000s and this again would automatically lead to a link between economic conditions in the US and this instrument. So I mean, long story short, I very much prefer the term premium shock approach that you use in the other part of the paper. Compared to this IV setup, I think that's a lot more convincing and that's really a lot more convincing in bringing the message that you wanna bring. Apart from that, the only other companies I have are really more minor compared to these. I mean, maybe I only pick out the very last ones. Like in the end, if you wanna say something about the slope of the yield curve and aggregate growth, what's also, it's interesting that you're doing this exercise for the US in a way because I always think of the US as mainly being a market-based finance type of economy. So you can then question like how important would this type of channel be in this economy? On the upside, if it's already quite important here, then I mean, you're capturing a lower bound in the way potentially. So that's kind of, would be good news for the paper from that perspective. So overall, let me conclude. So I think this is a very interesting read. So definitely worth reading. It's a very tall, provoking paper. I very much enjoyed reading it and yeah, I wish you best of luck with it. Thanks a lot, Glenn. Do you wanna respond before we open the floor? Absolutely. Thank you so much, Glenn, for the fantastic comments and this is exactly what I was hoping to get. A little bit of pushback on at least one of the identification strategies. So I appreciate the comment about maybe giving more attention to the expectations component of the yield curve and trying to be clear about the science we're getting. As well as the insight you showed with the chart on the correlation between the term premium and credit growth over a long period of time, a much stronger lagged correlation before the zero lower bound essentially, then afterwards. And it's consistent with what we find in the empirical analysis that the relation we have in mind is a lot stronger historically. A little bit weaker at the zero lower bound, at least in the bank level panel. Interestingly, as we went to the credit registry which is entirely post-GFC, we were able to find effects there as well, but the valuation channel is working against us finding anything. So the fact that we find anything is sort of reassuring. Like I mentioned, we did not take, we have not undertaken an exercise where we take the zero lower bound constraints seriously. Like we don't bring the interest rates to zero as we calibrate the model and look at impulse response functions. So I will leave my comments here. That's something that we should do to be a little bit sharper on the interpretation of the results. You're pushing us to speak about the importance of the channel relative to the valuation channel and I fully agree that's definitely something on our radar. The calibration of the model we currently have right now shows that our channel dominates the valuation channel. So in the particular calibration we have, it's possible that again at the zero lower bound this may be reversed. The IV you found least convincing, the exclusion restriction is very difficult to test. I think what I'm gonna propose is maybe short of abandoning it. It is to maybe come up with a size of the violation of the exclusion restriction that I would tolerate before calling the results completely spurious. And so we can quantify that and see if that's something we can live with. And finally, yeah, it is very... So I think it's generally a little bit misunderstood that the US being a bond, a markets-based economy, it doesn't mean that the bank lending channels or bank credit channels are not very powerful and potent. There are a lot of privately held firms in the US that are bank dependent that don't have access to the one market. Just in the credit registry alone, which omits a lot of small firms and a lot of small banks, we have 100,000 firms of which only 2,000 are publicly held. So the remaining firms rely on bank debt quite a bit. And so these financial and real shocks that work their way through the banking system tend to hurt them or help them significantly. Okay, thanks a lot. There's a hand up here and then on there. As usual, please identify yourself. Andre Kerman, Drexel University. So first, it's really interesting the different empirical results that you show. I just wanna follow up a little bit on the discussant's point that about the quantitative importance. So first an observation, which is that going back to Campbell and Chiller in the late 80s, it seems, or there's a lot of evidence that the expectations part explains, I think the lion's share of slope movements now. Statistically, of course, the expectations hypothesis may be rejected, but it's the main driver. And so I think in your graph, I would appreciate to see that too. You put the term premium movements and the slope after the temper tantrum at two different scales, right? If you put them on the same scale, you would see that the term premium increases by very little relative to the slope. The other point I wanted to make is that you could also think that the term premium is driven in part by growth, changes in growth. So if you take, say, Piazzese Schneider's NBR macro annual paper, right? They show that in a consumption based asset pricing model, inflation and consumption growth, the covariance is a potentially important driver of term structure. If that growth change has an effect on the term premium, that could be even affecting in reverse causality, so to speak. But that doesn't take away from your results, but I think it's really important to document the quantitative importance. Thank you. Great paper. That's a great comment. So it's true. So in our model, we are somewhat modest. We don't try to pin down the term premium endogenously along with the bank's lending decision or asset allocation decision. We take it as, it's a partial equilibrium. So we don't pin down endogenously the term premium and interest rates. We take them as exogenous and simply examine bank's responses to fluctuations, exogenous fluctuations in the term premium and interest rates. As for the expectations, explaining the lion's share, absolutely right. I think there are some estimates from the board recently suggesting that about two thirds of the variations in the yield curve are explained by expectations. So yes, I think it's important for us to think more about the quantitative importance of the channel and sort of relate it to this other channel, the expectation channel that could be quantitatively more important, absolutely. Thank you. Very interesting. Alpsimsek, Yale, very interesting and thought provoking paper. I find it goes quite a bit against conventional wisdom because we usually think high term premium as monetary tightening, right? High term premium, higher rates for borrowers, less borrowing, less investment. In fact, the basic idea of QE is compressed term premium and stimulate the economy. And conversely, the main concern around taper tantrum was that, at the Fed, was that it might lead to an overtightening of policy. And that's why Ben Bernanke went around and tried to convince the markets that they misunderstood him, et cetera. So I think you're identifying a mechanism that might dampen this conventional wisdom. But are you saying that actually conventional wisdom is wrong and this is the main mechanism we should think about even leaving aside valuation effects? We should think of term premium as like stimulating or is it just a dampener, like to clarify that? Thank you. So for it to be a dampener, it means that we have another channel in mind that is the main channel. And then this is one on top of it that dampens the effect of that main channel. I think that we should give this expected profitability channel equal power as we do to the valuation channel through which QE, lowering of the term premium, works. Like both of them, I think, are valued channels by which banks will react to changes in the term premium. I don't see one as a dampener of the other. We can definitely put them in the horse race and see at least empirically which one dominates. We can write up a model where one dominates. We calibrate the model and depending on the calibration, one may dominate the other. So I would see them as competing channel, if you will, or coexisting rather, coexisting channels. But I think that the channel that we identify is definitely one in its own right. So there's one last question from Christian. Thanks. I'm Christian Kubica, UCV High Committee, a great presentation. I have a question about identifying assumption in your not IV setup, but the loan registry setup. So you do a convincing job using the standard methodology of firm by time fixed effects to shut down loan demand channel. Then I guess the main worry that remains is that you have some omitted variable, let's say great growth expectations. They drive up term premium. They drive up loan supply. So then you argue, okay, you use heterogeneity across banks in their capital constraint arguing that more constrained banks should react more. But at the same time, I would argue that any omitted variable, for example, great growth expectations should also affect more financially constrained banks more. So for all the omitted variables that I'd see as concerning, they should also go in the same direction, namely implying a stronger correlation between more constrained banks and the term premium relative to less constrained banks. So I was wondering how you think about that. Yeah, no, absolutely. So I think the regressions will be believable. The regression results will be believable as long as you believe that we measure as well as possible the bank level growth expectations. They enter not only in levels and interacted with the post. So we allow them to change after the taper tantrum, but we also interact them with every single right-hand side variable in terms of bank characteristics. So we interact them with leverage as well. So there's a powerful horse race there for our diff-in-diff term. We allow for confounders for leverage as well as for confounders for the post dummy capturing the rise in the term premium. And in particular, one of the terms that we put in this horse race is leverage times expectations, bank level expectations. So there's not more that we can do other than saturate this specification and try to strengthen it as much as we can this way as well as providing additional evidence from other time periods, from other data sets, from other identification strategies. I would see overall the evidence as supporting our story, but I will also very much agree that there's no perfect identification strategy. Okay. So thank you very much to both presenter and discussant for very nice. Thank you. And then I would ask Moritz and George to come on the podium. Okay. So I promise you we're going to do a time travel now going back to the 1500s. Moritz will do this for us. Moritz is from the Kiel Institute for the World Economy. And he has a paper on Central Bank balance sheets and financial crisis. So please, the floor is yours. Okay. That's working great. Well, thanks to the organizers for having me. It's a great pleasure to be back and see so many, I want to say friends and family almost in the audience. I'm actually, I am going to be that guy who jointly with Neil, Martin and Paul is going to take you on the journey into the history of Central Bank balance sheets in the next few minutes. And we are not shy to answer one very big question on which there is exceptionally little research at least on the empirical side. Namely, we're interested in the effect of Central Bank crisis interventions, lender of last resort operations that we have seen quite frequently, unfortunately I guess we want to say in recent years by all major central banks and the debate on the positive effects of such interventions, you know, cutting short run equilibria, stabilizing the financial sector and sort of intervening in potential fire sale situations are often contrasted with the potentially negative side effects, namely the more hazard, the fact that it sometimes feels that central banks have to run faster just to stand still in every crisis. And so we, as I said, we're not shy. We're going to lever the history of Central Bank balance sheets in the modern period going back indeed to the 1600s to study the causal effects. And I'm sure we're going to discuss that causal statement here during financial crisis on macroeconomic outcomes. More precisely, what we're going to do is we have this new data set for Central Bank balance sheets and their composition for 17 economies since the early years. We're going to use that data to study long run evolution of balance sheets. When we clean up, I give you some stylized facts that you might think are interesting. And then we'll zoom in on balance sheet expansions. And importantly, we're going to show and argue that throughout the history of central banks, there have always been sort of distinct schools when thinking about lending or lender of last resort operations. So in the 19th century, there was the famous hanky badger debate for those of you who study sort of history of economic thought, opposing those who said like central banks should not get their hands dirty and let purge the financial system that's in trouble. And there were the liquidationists in the early 20th century into the interwar period. You know the famous Mellon statement that let the banks go and let's purge the rottenness out of the system. Public sectors shouldn't get involved in saving these speculators. And until recently there is a tradition of not far from here maybe of sort of monetarist people who are very skeptical of generally speaking doing things that could be interpreted as being discretionary rule deviation in crisis. So we're going to use this history of economic thought. We're going to use the intellectual predispositions of governors in charge and correlate them with the probability of providing support in crisis times. I mean to argue that these sort of the governor in charge and typically his intellectual or ideological predisposition to giving support to using the central bank balance sheet in crisis times is exogenous and we're going to use that to estimate the causal effects of these interventions. Now I'm going to convince you I hope that this does make sense. So we have the near universe of modern era financial crises and I'm going to code for each governor in charge, the school, the ex-ante sort of belief system that he had and we'll ask in the first stage does this actually correlate with the probability of the central bank using or taking on that land of last resort role, expanding the balance sheet, we say yes, that does and then we're going to causally estimate that. Okay, so there's a big literature on central bank balance sheet and the effects of LLR, but I think I mean actually maybe I need to take that back. It's not that big. There is not that much that we have on the empirical side, maybe some of you know the Honore and Klingerbeel paper, Mike Bordew and co-authors have done something, but it's not overly big and I don't think there is a paper, I mean maybe a discussion, maybe a fond something, but I'm not aware of that. There's definitely a large literature on financial crisis and their effect and recently Ulrich Mahmendi and others have used that idea of central bank government believes as an instrument to test economic outcomes, making the argument that these are predisposed, these are predetermined and they're actually exogenous to the situation that the governments then find themselves in. Okay, so the agenda, the evolution of central bank balance sheet, then we'll zoom in on balance sheet expansions and then the main part is really going to be that study of the causal effects of LLR interventions. Okay, so this is the stylist fact part, we have annual central bank balance sheet data for 17 countries, covering the size and detail composition starting from the 1600s. We're going to have macroeconomic data and crisis chronologies from BVX, which is Baron Werner Schoen and JST's Jordan Schillig-Taylor data sets. We're going to make the data available as usual and there's historical sources that we use to classify the ex-anti-ideological predisposition of in total 112 central bank governments prior to financial crisis. This is the data set, you just get the countries, it's the usual kind of western focused data set for which we have long run data. This is the evolution of central bank total assets over GDP in these 17 countries. This is a chart that you probably have seen in one or the other version, not going back quite as far, but you've maybe seen it for the 20th century showing this massive spike obviously in the size of balance sheets relative to GDP across the OECD world, I should say, or the industrial world plus Japan and a few others. One of the interesting facts that comes out of our work is if you scale this not by GDP but by the size of the financial sector, if you think about the financial stability role of central banks, maybe that is for some questions the appropriate scaling, things look very different. By this scale, the central bank balance sheet are actually quite small relative to what's out there now in terms of financial assets, of borrowing and lending. It's gone up a little bit, but we are in that perspective. Excuse me, we are relatively safe. It looks like not safe is a too strong statement, but we are in a less extraordinary time. Same actually happens if you scale central bank holding of government debt relative to GDP. Again, it looks quite unprecedented that we are almost at times that typically were, or at levels that were typically reserved for war times. We are at least in the major economies in peace right now. Again, if you scale this by the share of government debt outstanding, things look a little bit more benign. Of course, it has happened that overall financial debt relative to GDP has increased substantially and that government debt relative to GDP has gone up quite a lot. The scaling really matters for how you want to think about the size of central bank balance sheet in this moment and some might say, if you focus on the financial stability role versus the stabilization role, you come to quite different conclusions. Okay. Major balance sheet expansions and their drivers, sorry. What you see here are expansions by event type and expansions are defined as 15% nominal year-on-year asset growth. We could do this with 20% or 30%. It wouldn't change the basic message of this chart which is balance sheet expansions over time have shifted from, I don't know why we code wars in green, it looks far too friendly, from war times, from government finance in terms of emergency to what you see out here is the red and then there's some natural disaster as well, is mainly a financial crisis intervention. When we observe balance sheet expansions in the sample in the 20th century, you see this shift from interventions in war times, typically government finance operations, you see this here in the two world wars as well, two financial crisis interventions. And you can do this a little bit more formally and just estimate the central bank's sensitivity to crisis events over time. So this is pre-industrialization, industrialization period, the first globalization, the worker period and post-worker too. You see that in financial crisis, typically in the early times there were some interventions, but it's not clear that central banks actually intervened almost, intervened sort of as a rule, but now you could always say that we are in a world where central bank interventions in financial crisis have almost become, central bank balance sheet expansions, I should say, during crisis times have almost become systematic. They are expected certainly also, I mean, at least the probability is very high, 60%, so there's some chance that market participants will price that in as well. Okay, so these are interventions. Now let's talk about in the last 15 minutes about the main idea and the main, I think if you will, the main new contribution here of this paper, namely, what do we do to test and estimate the causal effects of central bank liquidity support in crisis times. So to start with, we're gonna define the land of last resort role as central bank lending to banks that are unable to borrow at viable rates from distressed markets. We're gonna operationalize that in a very crude way and maybe we can do better here if we had more detailed balance sheet data, I should say, as an annual central bank balance sheet growth of 15% around the crisis time. We're gonna have central bank land of last resort operations that increase bank reserves by a balance sheet expansion so that we use that definition and we bring that together with a full set of crisis states from BVX, Baron Berners-Jong and JST. If you just track as an event study, and this is similar to what Mike Bordeaux and others have done before, the evolution of real GDP and the trajectory of GLGDP in crisis with and without liquidity interventions. So zero here is the year of the crisis and you see how real GDP evolves into the crisis and then after the crisis, you would come to the somewhat paradoxical conclusion and some of these papers make that point. If you will, that's kind of where the literature stands that it's not clear that these central bank interventions from this very high flying altitude macro perspective have a positive effect. Oh, the average trajectory without liquidity injections is actually better than the one with liquidity injections. But obviously, there is a large endogeneity or causality problem where crisis might simply warrant more support. So we need to do better than this and this is what we propose to do. Obviously, what we need is a quasi-random variation in liquidity support during crisis and that's not easy to come by but we think we can exploit these longstanding differences and opposing schools of thought on the pros and cons of discretionary central bank interventions and crisis times. So in the 19th century, I mentioned the hanky-badger debate. In the late 19th and early 20th century, there's the whole debate about the real-build doctrine and liquidation is the idea that central banks would only take real sector collateral and not financial sector collateral and the real-build skies were consequently opposed to lender of last resort operations just vis-a-vis banks and then there's the rules-focused monetarists. This is sort of the Brunner-Meltzer tradition that kind of lives on in various forms over time. So we argued that governors that were ex-anti-influenced or sympathetic to these schools could be less likely to intervene. So liquidation is school, the real-build, school, the hanky, school as opposed to budget and we're going to use that as an instrumental variable, namely the beliefs of central bank governors regarding the LLR benefits. Here's an example. There's Richard Koch. He was governor of the Reichsbank in the early 20th century and he was an intellectual tradition of the liquidationist. He was a, as the Bursons had called him, a fierce defender of non-intervention. So it comes to 1907 crisis. Koch refused to intervene despite a lot of pressure to do so. And then there's a quote. He was intent on cleansing the Reichsbank balance sheet of all non-trade bills. We refuse to let the Reichsbank be a cheap source of liquidity for commerce, et cetera, et cetera. So this is the arc of intellectual history that we're going to use. The hanky is the real-build tradition, the liquidationist, the Mellon school. Some of these, they overlap, but there is at each point, I find this quite interesting actually, at each point in time there is this debate right today where you would open the Financial Times at any given moment and that would be, you know, with some likelihood in any given week, someone warning of the more hazard effects of too much central bank support and someone saying, like, well, no, no, this is necessary. We can't hold Main Street hostage for what's going on on Wall Street. So this is really like a long-standing, long-standing sort of intellectual fight out there that we think we can lever and you see here some of these examples, which we, the monetarists here, the Bruno Meltzer Taylor School, as we call them. So at each given moment, you find people who are like, no, this is not, we should be more cautious here. So how do we do this? We code central bank government beliefs before the crisis. That's important. So we use material, we use sources, we use newspapers, dictionary, I'm sorry, sort of national biographies. Here the example is the Gallica from the BNF. Before the crisis date and look through them and look at statements that would identify sort of the intellectual home of these governments. Here's the Koch example again. We pay particular attention to statements revealing moral hazard concerns. So President Koch is the fierce defender of the gold standard loathed by the bi-metalist, cleanses the rice, cleanses the rice bank balance sheet. And these are statements from 1903. So well before the 1907 crisis and we say we can sort of classify him here as a hawk. Then we come up with this whole panorama of hawks and doves or pragmatists. So doves and pragmatists, we'll be not sure. What we really can do is identify the hawks and then we just put the doves and pragmatists in one camp. So Ben Bernanke, we call him like a pragmatist. He's not a hawk, but then there's some Mervin King. Many of us in the room know him. We call him a hawk because he's been, until at some point in the crisis, quite outspoken about the moral hazard concerns and wasn't super happy to intervene. You might disagree with some of those Jean-Claude Trichet as a pragmatist. We can disagree with some of these classifications, but we do them for all crisis that we have in the dataset. So first, a very simple look at the data. Hawks and doves, what's the probability of hawkish or dovish governors expanding the balance sheet between 15% or more in each of the years following a crisis? You see there's a difference. The blue guys are the doves. They're much more likely. And that probability rises as we go along. And first look maybe at the crisis outcomes as well. Here now we classify, maybe look in the middle at the log wheel GDP so that's similar to what I showed you before. The blue one is the average trajectory under dovish governors and then the average trajectory under hawkish governors. There's a slight difference both in GDP and CPI, et cetera. You don't see it so much on the monetary side and I come to that in a second. Okay, so this is not very formal yet. Let's get a little bit more formal. So the first stage, the probability is we're going to estimate the probability of seeing a balance sheet expansion in the crisis defined as an annual balance sheet growth of 15% in the year of the crisis or in the year between crisis and T plus 1. And we sort of classifies this by doves and hawks. For the doves we have about 22 crisis without intervention and 21 crisis with intervention. For the hawks you see there's a big difference in the mean. The hawks have, there's 26 crisis without intervention and then in nine of those crisis the hawks also intervene. So we have some nice variation here that we can play with. You also get a sense for one of the big problems I guess that macro has, which is like even if you look at the universe of modern era crisis you end up with 90 or so and over a few hundred years. So if that's your level of observation, the individual crisis or even the business cycle we still don't have that much data to work with. Okay, so we're going to estimate the first stage which is the balance sheet expansion. There's going to be this variable here for your hawkish government or not and then there's going to hold a whole set of macro financial control variables. Then on the second stage we're going to use that for the effect of the balance sheet intervention and this YIT here that it's going to be a local projection of the effects on GDP and then other outcome variables after the intervention. We have country fixed effects. We have the usual control variables that I'm going to show you here. There's going to be inflation. There's going to be money. There's going to be GDP per capita. The things you want to see, we're going to control for the three-year growth of real private sector bank lending. Control for the size of the credit boom that might correlate with the severity of the crisis, etc. In the end, we would still argue the big assumption you need to make is that the sort of the hawk-duff allocation to the crisis, if you will, is quasi-random and has nothing to do with that severity ex-ante. So here's the first stage. Hawkish governors are about 40% less likely to intervene or to expand the balance sheet in the crisis to provide liquidity to the financial sector to take on that land of last resort role. All else equal, no? All the country fixed effects with macro controls. There is a clear difference. You have a hawk in charge. And there's an assumption, obviously, that the governor kind of is very important for the decision-making of the central bank overall, but we see that in the data. And then in the second stage, we're going to track the path of real GDP of CPI and of money of horizons going forward and we're going to estimate the impact of central bank liquidity interjection with confidence bands comparing the path with intervention to it than what we think is a plausible counterfactual without liquidity interjection. And what we end up with is good news for central banks and it's also good news for people who believe this is a financial stability, this budget, land of last resort role is important and beneficial overall because we find, if you buy into our identification, we find large effects. If you look at the real economy here, the largest differences in the trajectories with and without liquidity interjection. Essentially, being a dove supporting, sort of in this setup, being more or less randomly allocated to the crisis, being a dove expanding, supporting the financial sector early, it gives you a better GDP outcome, it gives you better price outcome and it gives you more money growth and it's part of implicit in the balance sheet expansion. So all in all, you end up with a stabilization package that looks much better than what you get in the hawkish case. Look at investment in stock prices, also something that I think some of us will have in mind from previous crises that in the average trajectory without liquidity provision is much worse, it takes much longer for investment to recover, the stock prices are much more depressed for a longer period where the interventions push them up and I think a lot of that sort of rhymes with what we've seen in crises in the past few years. So is this about really land of last resort operation or is this about something else? Is this about the central bank monetizing the deficit and this is about fiscal stabilization through the back door? See here, what happens to central bank government assets over the crisis between hawks and doves, there's barely a difference, so this is not about an intervention to support buying up government debt, supporting fiscal operations, this is really driven by difference in the increase in central bank reserves. This is about the operations with the financial sector. We also find, if anything, that under hawks, government expenditure after crisis increases more potentially because there's more need for that to stabilize the economy and also government debt increases more under hawkish governance in crisis than under doves again, those controls for all kinds of macro financial variables. I'm not going to talk about those, they're all in the paper. I want to use my last minute or two to bring across one additional element that gives a little bit of hope for the hawks. Namely, we are looking for evidence that more hazard could, after all, sort of come back to haunt you. What we look at here is the share of country years in which countries, after a crisis with a liquidity injection compared to a crisis without a liquidity injection, experience another bad credit book. Arguably, it's sort of a little bit of a rough criterion, but you can see here a difference that if you, as a country, as an economy come out of a crisis where the hawk in charge and the financial sector had worse economic outcomes, the financial sector potentially learned the lesson, I don't know, but you get the share of bad booms happening in the next 15 to 20 years is much lower, whereas there's some evidence, and I'll give you some regressions in a second, that seems to suggest that intervening in the short term comes potentially at the cost of the next bad boom, the next crisis, the next bad boom bust episode being more likely. Again, I'm intentionally sort of cautious on making strong statements because obviously we have all kinds of identification issues that should not distract from what I think is quite nice identification in the previous part. But here are some of these regressions, so the liquidity injection, the last crisis, dicts with some statistical significance the likelihood of getting into next, in the boom bust episode within the next 20 years. Okay, that's it, conclusion. Central bank balance sheets have grown a lot relative to real economy size but not relative to financial sector size, so I think that's something we want to keep in mind when we think about the future of central bank balance sheet. Large liquidity injection stabilizes the economy for natural markets in crisis times and boosts the recovery, so there is evidence that assuming that land of last resort a function is valuable, and maybe there's some more hazard concern. Thank you very much. Moritz, the discussion is Georgia Premier from Northwestern University, 15 minutes. All right, so thank you very much for the invitation. This is a nice and important paper. It involves an impressive amount of work, as you probably might have noticed from Moritz's presentation, and it's very dense when it comes to results and facts. And so the first thing that I'm going to do I'm going to summarize the paper in a single slide, so that's a little bit of a heroic task, or at least my reading of the important results in the paper. So the paper is a very well-defined goal, which is to study the effect of central bank liquidity support during financial crisis. The question obviously is an old one. It's not a new one, but the approach is rather new, because instead of focusing on a single episode, for example, the great financial crisis in the US, or on the history of a single country, the paper constructs a novel data set that covers 17 advanced economies since 1580-something. So very long data set. And the first contribution of the paper is the data set itself, but then only based on this data set, the authors can construct a host of interesting stylized facts. Here are a couple of them that we saw in Moritz's presentation, the fact that the size of the central bank balance sheet relative to GDP skyrocketed over the last 20 years, but the situation is very different if we focus instead on the size of the central bank balance sheet relative to lending to private firms. And so this, of course, a corollary of this means that lending to private firms has exploded in the last few years, something to keep in mind. But the others are very ambitious, so they don't want to stop at collecting or documenting stylized facts. They really want to get at the causal effect of intervening for a central bank during a crisis on macro outcomes. And this is a difficult task, because central bank interventions are endogenous to political and economic circumstances, whether central bank intervenes, how much it intervenes depends on the state of the economy. So you cannot simply run a regression of macro outcomes during a crisis on whether or not a central bank has intervened or the extent of the intervention. So they need an instrument, and their clever idea is to use as an instrument in their IV strategy the beliefs of the appointed central banker, the belief before the crisis. And this is an analysis that they conduct on a shorter sample. Now, based on this instrumental variable strategy, they document two important results. The first one is that central bank balance sheet expansion of at least 15% during or right after a financial crisis bolsters real GDP by about 21% over the subsequent three years. This is a very large amount. And the second rate, this is about short-term gains of policy intervention, but they also have a result about possible long-term risks of policy intervention. So this is the graph that, you know, I said one slide and two figures. I'll show you the first figure. This is the second figure, which I will use later on for a little exercise. This shows the difference between log real GDP in countries in which the central bank intervenes during a financial crisis and countries where the central bank does not intervene. But the second result is about long-term negative consequences of central bank liquidity support. And they show that central bank liquidity support in the current crisis makes the next crisis more likely. This is exactly what we probably would expect. This moral hazard concern is it implies this. So if I were a monetary historian, I could spend all of my remaining 10 minutes and probably more on the construction of the data set, arguing about specific choices. There are many, many choices that the authors had to make in order to construct the data set, to process the data. There is an 87-page appendix that goes into these painful details, but I'm not a monetary historian, so I decided to trust the authors entirely about their choices, or some of their choices, and instead I will organize my comments around these two points. I want to discuss possible threats to identification. I will try to argue that the beliefs of the appointed central bank even before the crisis are to some extent endogenous to the economic circumstances. And so this is going to serve as a note of caution about the causal interpretation of some of these results. And the second part of the comments is that I will do a heroic exercise, a very back-of-the-envelope calculation, because I want to try to put the short-term gains of the central bank intervention on the same scale as these long-term risks to see whether they are of comparable magnitude or whether maybe the short-term gains are just clearly superior to any long-term risk that it's obvious that the central bank should intervene or maybe things are less clear-cut. And I will conclude that they are less clear-cut. Okay, so let's start with the possible threats to identification. I want to discuss basically whether the instrument that they use is really exogenous. Remember that the instrument used are the pre-crisis beliefs of the appointed central bank, of the central banker in charge at that point. And for the instrument to be exogenous, we need to believe that these views, these beliefs of central bank should correlate with a macro outcome during a crisis only through the effect of beliefs on whether a central bank decides to provide liquidity support or not. And there are reasons to be skeptical. This is a little bit of an extreme assumption. The paper is very open about one possible limitation, one possible threat to identification. In fact, I'm quoting the paper and the argument here is very simple. If there is a dovish central banker in charge or if the market anticipates that there will be in the future a dovish central banker in charge, then this anticipation of a dovish crisis management might encourage financial restaking and make the next crisis more likely. Of course, if this is true, if this is an important consideration, this will lead to an understatement of the effect of central bank liquidity support in the sense that the effect of central bank liquidity support in reality will be even bigger, larger than what Moritz and co-authors estimate. So let me try to argue that there are other reasons to be a little bit concerned and these other reasons might actually lead to overestimating the effect of central bank liquidity intervention. And the idea that I would like to discuss is that excessive restaking today might actually lead to the appointment of a hawkish central banker. Maybe the people in charge or the markets realize that there is excessive restaking and there is the need of a central banker that tightens financial regulation. Now, if excessive restaking today lead both to a more likely appointment of a hawkish central bank and a more likely crisis tomorrow, you might actually overstate the effect of central bank liquidity support. Now, is this just a theoretical curiosity or is it what happens in reality? The truth is that I don't know especially I have no idea how to think about the 1600 and the 1700, but I do know that for type of historical happy source that I've looked at, like, for example, the great inflation in the United States, the fact that inflation had been high at the end of the 1970s for at least ten years at that point led to the appointment of Paul Volcker as a central banker. Paul Volcker had a reputation of being a hawkish central banker, a reputation of being a central banker that would have likely fought inflation and in fact it was appointed central banker. So this endogeneity of the appointments of central bankers I think could be a concern. So again, I'm not entirely sure how important concern number one or concern number two are in practice. This is just a note of caution about interpreting, you know, literally causally some of those estimates. Let me go to the second set of comments which is about comparing long-term gains, sorry, short-term gains with long-term risks. Remember that the paper estimates that central bank liquidity support today in the current crisis reduces the severity, reduces the recession in the current crisis but makes the next crisis more likely. And so can we compare these two things? So when we look at a time span of about, say, 20 or 25 years, who's better off? The country in which the central bank intervenes today or the country in which the central bank decides not to intervene? Of course, this depends on parameter values. For example, one key parameter is how severe the next crisis is going to be. Okay? And let me show you some possible... So my little simulation exercise which is loosely based on some of the results of the paper involves the simulation of many possible GDP paths. Okay? Let me show you four possible cases. So this is a simulation that starts with a crisis. Okay? All of the simulations start with a crisis today the last five years. That's a typical scenario. And the red country is a country with a central bank that intervenes with liquidity support today while the blue country is a country with a central bank that decides not to intervene. Based on the results in the paper there is a discrepancy in real GDP today depending on whether the bank intervenes or not. The country in which the bank intervenes is better off. In this particular simulation I constructed it by assuming that there are no more crisis in the future. And I also assumed that whatever distance there is in GDP red and GDP blue this persists forever. Now, another possible... So this is the initial crisis and the central bank liquidity support reduces the recession. Another possible simulation is a simulation in which there is a crisis today and then the country in which the central bank decides not to intervene is subject to another crisis after 15 years while the country where the central bank intervenes has no more crisis within 20 years. That's another possible... These are going to be stochastic simulations so these are examples of possible paths. Case number three is the opposite. There is the crisis today and the country in which the central bank decided to intervene is actually subject to a crisis after 10 years again. Now, the paper's estimate suggests that this is more likely than the upper row. And this is one last case in which there are subsequent crisis in both countries but the crisis in the country with intervention happens earlier. Okay? So again, the results in the paper suggest that the bottom row is more likely than the upper row. All right, so I'm going to do this many, many times. Having this crisis happening stochastically and calibrating the exercise on some of the results of the paper. For example, I'm going to calibrate the GDP difference in the current crisis between countries with an intervention in central bank and a non-intervention in central bank based on this. This is what the paper estimates. So I'm going to use this number. I'm going to assume that this is a distance of 5, this is a distance of 8%, this is a distance of 7%. Then I'm going to calibrate the probability that the subsequent crisis within 20 years happens in the country that has decided not to intervene today. I'm going to calibrate to 4% per year based on this graph where the average probability is 4%. Then I'm going to calibrate the probability of a subsequent crisis happening in a country where the central bank instead decided to intervene today to 14%. Don't take these numbers too seriously in the sense that take his numbers seriously but my interpretation and my use of those numbers into the context of this simple exercise not too seriously. So I took this 14% from this table that basically estimates that they increase the probability of the next crisis for a country whose central bank decides to intervene today is between 4% and 16% more than in the country that does not intervene. So I collapsed this to 10% and I used 14%. And finally, this is the key parameter here that determines everything. How costly is going to be the next crisis? I'm almost done, Michael, and I'm going to analyze two cases. One with a milder crisis. I don't know how mild maybe this is already fairly severe, which is a crisis of 5% losses in GDP per year for five years. And the next one is a severe crisis, 10%. And I'm going to assume that when the crisis happens again, both countries react the same. I don't want to introduce that additional dimension, that additional difference. And what I'm going to do is I'm going to simulate many possible GDP path and then I'm going to average GDP over 25 years in these two type of countries, the countries that initially reacted and the countries that initially didn't. And this is the blue here is the distribution of average GDP in countries that decided to react in the early crisis. And it's normal, the mean is normalized to be one. Here you can see that countries that decided not to react initially they're most likely worse off. So there's only a limited probability that these are better off. And this is in the case of a fairly mild crisis. If instead the subsequent crisis is more severe given that the country that intervened today is more likely to get that subsequent crisis then the two distribution are much more comparable. In the sense that this is the range of possible the blue are the range of possible outcomes in countries whose central bank intervened at time zero. And the yellow are the ranges of possible outcomes in the countries where the central bank did not intervene. And the two distributions here are fairly similar. Actually the model outcome is better without intervention relative to the intervention. So again don't take the numbers of this exercise too seriously my purpose here was simply check whether the gains are in order of magnitude bigger than the potential risks. And I don't think that's the case they're more comparable they're at least of the same order of magnitude. So this was my second comment the first comment was instead arguing that there are some concerns about the exogeneity of the instruments and so that's a little bit of a note of caution. Thank you very much. Thank you. Thank you, George. I can I'm very happy to see these simulations I think your assumptions about the cost of the second crisis are actually quite severe even in the mild case so I guess that is good news for the intervention regarding threats to identification I think yes you were very gracious not even to talk about all the measurement issues and classification issues that obviously arise I think we can test that second channel which indeed we did not discuss in the paper so the idea that in a financial boom the appointment of a harkish governor becomes more likely I think there's ways to at least get a sense for that in the data which I guess my hunch would be that these appointments follow political schedules they are like sort of probably predetermined with I would be surprised if we find much but it's something we can test I think and we'll do that I mean maybe leave it for discussion thank you. Great paper one benefit of a lender of last resort facilities preventive right once you have the LRR you know since badger hot that actually crisis become less likely because you're eliminating the bad equilibrium and in fact Michael Bordeaux has a very nice paper comparing the UK and the US experiences in late 19th or early 20th century they had very similar business cycles but the UK had fewer crisis and he argues because UK had a lender of last resort so I'm wondering you're showing very nice that conditional on having a crisis lender of last resort is very useful but maybe actually there's additional benefits that you have fewer crisis to begin with when you have a dovish central bank in place and I would also note that this actually goes against the moral hazard argument that it's not necessarily the case that you have more crisis with dovish if you think of this preventive mechanism strong enough it might cut against the moral hazard argument are we collecting are we collecting? I totally agree this is conditional on having one I guess you want to balance this against having a dove in charge might increase in centres for risk taking especially if the regulation interaction is uncertain but in theory maybe in practice not so much but it's a good point Oresti already has a mic so let's give Oresti Oresti Tristani I was on a very interesting paper I was wondering about you said you have a lot of information about these balance sheets as well the central bank balance sheet and I was wondering whether there's a role played by the duration of the intervention so how fast the balance sheet goes back and also what is on the asset side you showed reserves we know that some expansions of balance sheets involve purchases of government bonds but I guess some other ones were more on the bank side and also a final issue I was wondering one other measure whether ok not really the effects on moral asset but maybe the potential dissatisfaction with what happened with the intervention could be gauged by looking at whether a dove-ish central bank is followed typically by a hawkish one I was wondering whether you can see this from the data Four great questions in one speed of intervention unfortunately we only have annual data so that really you want to say as quickly as you see trouble emerging can you see difference in the timing having an effect we don't have that granularity in time we told you we have about 90 interventions or 90 crises 26 with the doves and I think 9 with the hawks and with that have seen these interventions so there's a little bit of a data issue how finally we can cut that but it's a very interesting point we'll look at what we can do on the asset side I think we believe that this is mostly private sector assets that are bought but the data on the liability side of central bank balance is much better than on the asset side so again if we had better we've looked at it we don't have the systematic overview there is another I think interesting point that I might want to bring in I think that relates to what you said is we also obviously look at the path of interest rates so to make sure that this is really the effect of the balance sheet expansion and not also of like you could think that the doves cut interest rates quicker and there's a difference in interest rates so that's taking care of. I think it's a very great attempt on an excruciatingly difficult identification problem and one can quibble of course about classifications like I think the difficulty is people change their ideology I think Mervin King is a great recent example Hawke turned dove presiding over the biggest balance sheet expansion in the history of the country my question goes as follows do you think it's really the ideology of the governors or of the country the reason why I'm asking that is that most central banks in history actually are either given an explicit last resort mandate or not so if you go let's say to 1900 France Germany were very different France and the UK had given their central banks explicit mandates Germany and the US did not and of course there was very much influenced by the ideologies of these countries at the time they later on changed with the establishment of the Fed etc etc I don't know if you can distinguish but it's I think something in terms of really what is causing this maybe something you would want to look into great question to some extent there is a country fixed effects on these instrument in the first stage that might take care of it a little bit change in countries over time it's a fair point I think from my reading of history I would certainly and I referred to colleagues over there over there I don't know over there there might be some ideological consensus in countries at some point that is even more important than the governor's upbringing we're making an assumption here of course that the governor himself plays a very important role in guiding this is more than just an extension of the consensus it's an interesting it's a good point I agree what was the other question there was a good sorry let's collect them now maybe it has a third one so Carlo you can answer so I had a related question I had a related question to what Luc was asking so it seems that people have difficulties with the instruments so I said why not another instrument and I thought do you have information on whether or not independent or not throughout history because I thought take the case of the ban de France part of its history it was receiving instructions to limit its ability to act on its own so if you have variation across central banks or across time within central banks you could get an exhaustion variation which cannot be criticized for being endogenous to restaking because the reason why some central big independent was completely remote to any economic considerations in many cases that's why I think of trying that Carlo great paper just on the moral hazard concern one well known precaution is also the principle to lend in a crisis at a penalty rate and against good collateral so that's well known and I was just wondering whether you can also observe this data in your historical data set so that you can distinguish also between central banks which have in a way respected this principle and those that may have not thanks and then Carlo last question Villa from ACB now great paper thanks to quick question the first on the identification I was just thinking that probably the important assumption the identification is the persistence of beliefs or the governor before and after the appointment and I was trying to understand what is that would make the enforcement of this persistence realized today of course is the market testing the whatever policymaker says that would move assets in seconds while I'm not sure that hundreds of years ago whatever government would say would be tested by any market discipline so this would probably affect your results the second is when you balance short-term gain and long-term gain whether you also account for you have a graph that says that shows that there is a huge increase in investment so now decent paper shows that for example a titan can last a market policy can last about 10 years with affecting also TFP or the accumulation of knowledge whether this is something that you also account for thanks I mean maybe maybe you wanted I mean there's this you're referring to Oscars and Allen's paper on these long-term effects for monetary policy feel free if you wanted to come in I will maybe start with Arnaud and Roland in central bank independence we cut and check if that makes a difference for the intervention probability didn't but that doesn't mean that we shouldn't look closer into this see Roland know we'd love to know enormous enormous work and then I do I mean I think this we thought about this sort of the governor persistence and believes we thought about maybe test looking at the market reaction at the appointment of that governor and see if mostly he is seen as a as a hog or duff but it's very difficult with these historical data but I think it's a good idea we indeed we do assume that this kind of hard coded you know your hog you or you I don't know you come from the monstrous tradition and you don't change your views even in the evidence of other facts I don't you know maybe sometimes that's that's corresponds to reality I should say before I hand over to for last comment I should say that this actually this paper grew out of a central paper that I did with Neil I think in 2014 so thanks to the ECB for really kickstarting that research no I don't think I would okay and I think that was a very nice concluding reminder and that paper generated a big discussion we can continue that over lunch but there's two more things before I mean one is a round of applause for the two of you and the second I mean look love and open the conference and I would want to give him the floor to also formally close it all right so that's it's not closed yet we still go for lunch all together but now it's the most important part of the conference which is to thank those who made it happen and all of you have ever organized conference like this one in your life knows how much is happening behind the scene and behind the curtain and often it's forgotten and I don't want to make that mistake today so we're going to have a few of the colleagues here on stage right now and it's a little surprise for them because they don't know so that's the fun part and I want to start with Britta Bertram actually was not even supposed to be in charge of this event but took over from a sick colleague which is of course the most exemplary way of team work you can imagine so Britta please don't be shy on stage is on stage be careful with the steps by the way this stage is a little bit holy in the building because normally it's the real important guns that are standing here and in particular our president Christine Lagarde gives a press conference here so I'm not sure you realize that alright but anyway let's just continue all the trainees as well of course they've been going around with the microphone Catalina everyone come on stage the the communications team whoever has a hand free I mean NASA I know you always do others feel free to come on stage if the camera keeps rolling we have a whole assistant team so please come on stage and then the last but not least the organizing committee Bartosz first and foremost but this time we had a really large one and that I think explains why we had such a wonderful collection of papers so please Bartosz lower everyone of you please come on stage