 Hello everyone, good afternoon. Welcome back to our session. So our session is entitled the central bank operational frameworks. And of course this is a central theme. In particular for us here at the ECB these days with a review of our operational framework in full speed. So the papers that will be presented in this session but I would say more generally the papers presented in the conference, keynote speeches and also the market panel will provide key inputs that we would be gratefully taking into account for the completion of the operational framework review. My name is Tomas Lasopoulos. I'm deputy director general for market operations. I'll be chairing this session and our session has two papers that have a very clear link between them. They both tackle the question of the demand for reserves and this is in fact a key question for central banks. It's a key question both in terms of managing the transition away from central banks QE portfolios but also for helping and informing the choice of the steady state operational framework and the resulting balance sheet size and composition. And in a way the two papers are complementary because the first paper focuses very much on bank demand for reserves whereas the second paper also addresses the demand for liquidity by known bank financial intermediaries and in fact shows that this can be in some environments and conditions a more decisive factor for determining when the balance sheet shrinkage can stop. Now let me turn to our four speakers for this afternoon and let me start with Annette. Annette Wissing Jorgensen will be our first presenter this afternoon. She is a senior advisor at the board of governors of the Federal Reserve but of course she has a long and distinguished career also in academia. And to discuss Annette's paper we have joining us virtually online Stefano Coradine from the ECB's research department. And our second presenter this afternoon is also from the board of governors of the Federal Reserve, Sria Anbiel and who is a group manager in the division of monetary affairs. And finally to discuss Sria's paper we are pleased to have Professor Rafael Robulo from SEMFI and director of SEMFI and of course a leading authority in banking and bank regulation but also monetary policy and monetary policy implementation in particular. So let's move immediately without further delay to the first paper. Annette the floor is yours, you have 25 minutes. Thank you very much. Thank you. So today I'd like to talk about banks demand for central bank reserves and I should say this is joint work with David Lopez, the leader whom you hopefully did not see present the same paper here at a different ECB conference a few weeks ago. Importantly since there are some numbers at the end these are just my views and David's views that usually disclaimer applies. Alright so thinking about the role of reserves in U.S. monetary policy if you think back to before the financial crisis the Fed was in a system where reserves didn't earn interest there was no IOR therefore reserve demand was low. Reserve supply was low even relative to demand in the sense that there was a very sharp reserve scarcity and this allowed the Fed to change the equilibrium short market rate by very small amounts plus minus a couple of billions was enough to have a large impact on the equilibrium market rate. Fast forward to the financial crisis the CO lower bound became binding the Fed and other central banks went into unconventional policies with four guidance and QE. With QE reserve supply expanded massively and also the Fed started paying interest on reserves of course something that the ECB had already been doing for a long time before that. So our focus here is how exactly does the Fed which is our focus control the equilibrium interest rate in this new ample reserve setting and secondly and very related how can we use an understanding of reserve demand to think about quantitative tightening. And here we're going to take an interest rate volatility perspective and try to estimate that point of curvature of where you start seeing increases in equilibrium rate being pretty large you know for a given size of supply reduction. I want to say I've taken a different angle in my paper for this in ECB center conference this year for thinking about this QT issue which was saying OK let's start by observing that the central bank doesn't really need a particular balance sheet size in order to hit a given policy stance when it can pay interest rate on reserves. So effectively with the balance sheet you can do one more thing which could be for example supplying a lot of safe and liquid reserves sort of following the Friedman rule logic. In the paper a key point in that paper key point is that one should also think about the assets that the central bank holds. So not only do you add safety and liquidity when you supply reserves but you may also take away safety and convenience if you buy German bonds or US treasury as a central bank. All right so back to the current paper I'm going to do this in two steps. First I'll lay out a framework for thinking about the demand also the supply of reserves about the equilibrium and also the question of interest rate control. And then I'm going to estimate this out empirically and then get back to thinking about what it what it says in terms of what can we do with this from the perspective of actual policy. And there's going to be two useful outputs hopefully one will be we're going to be able to construct a schedule for how to set the interest rate and reserve as a function of the balance sheet size. If you want to hit a given target that's called the ISOFED funds rate curve. The second one will be we can estimate how much QTS feasible from an interest rate volatility perspective. All right so just since we are you know not at the Fed let's just quick recap on the moving parts of the Fed's balance sheet on the asset size side treasuries MBS are the main ones. Loans are relatively small so this is unlike the ECB which historically supplied more of its reserves to lending to banks and of course later has also added a lot of securities portfolio on the liability side. The Fed and other central banks supplies their assets by funding it with the autonomous factors that's currency and the government deposits called the Treasury General Account in the US. And then with in blue the interest paying deposit the interest paying liabilities reserves which you can think of as just banks checking account with the Fed. And then the overnight reverse repo facility which you can think of investments by non banks with the Federal Reserve typically a money market funds in the form of repo. All right so to put the paper in perspective the black line here graph side the total size of the Fed's balance sheet you can see it goes up a lot more than it goes down. There was a round of quantitative tightening in 2018 2019 it's already come up you know it didn't ended maybe as well as as one could hope with a big spike in short market in short market interest rate suggesting perhaps too much reserve scarcity. And where the paper fits in of course isn't this latest round of QT which you can see is well underway. All right I don't this is a you guys already know the US. Okay so reserve demand will do with demand supply and equilibrium so reserve demand. Going back to the basics of banking comes from banks liquidity management problems so if you think about narrow banking the banks will back their deposits one for one with reserves. If you think about fraction reserve banking the banks will back their deposits with some fraction that being put into reserves and where we are now is in this ample reserves banking where we not totally sure how the reserve demand function looks. But nonetheless you know from this it should be clear that a key input of course should be how big is a banking sector which we are going to process here by deposits and then of course something about the how expensive is for the banks to hold reserves. All right so our framework is going to be so the pretty reduced form but hopefully enough that we can take it to the data. There's going to be four main ingredients. The first one will be that the fat pays into some reserves. The second one will be the reserves have liquidity benefits. You could think of these as coming from the fact that if a bank has reserves and faces a deposit outflow then it can satisfy that outflow without incurring any cost of selling assets or you know delaying payments in the extreme. I'm going to use V to denote the convenience value so that's like the total dollar value of having reserves. I think this is an expected savings from having reserves and thus not having to incur other costs. And then the key element here will be the derivative of V with respect to the reserves. That's what we call the convenience yield and that's the marginal value of an additional reserves. It's going to be decreasing in reserves presumably as more and more reserves are less and less useful. It's going to be increasing in deposits as an additional dollar reserves is more useful for managing a larger amount of deposits. The third feature will be bank balance sheet cost which for simplicity will have just as a constant fraction of assets. Those are in there to capture the empirical fact that oftentimes we see short market rates going below the IOR which is something one can get out with the balance sheet cost. Finally, and this is quite related to three years paper, we're going to have a cost of posting collateral in VPO transaction so that I can speak not just to the Fed fund trade but also to VPO rates. I'll capture that with a W function. You could think of that cost as capturing foregone securities lending revenues if you have to post a collateral in VPO borrowing rather than being able to lend it out. With that, you can think of the reserve demand function just as coming from the bank's first order condition for borrowing to hold reserves. In that little bank profit expression, there's three rows. The first one is interest income. The second one is interest expense. The third one is from these various features that I already talked through. Okay, so suppose we define reserve demand and that'll be our focus as coming from the first order condition for borrowing in the Fed funds market and holding reserves. Then the reserve demand curve simply traces out the highest interest rate the bank is willing to pay to borrow. That's on the left-hand side as a function of how many reserves are held. The second thing that's to pay is going to be the net benefit of reserves which comes from the fact that reserves pay interest. You can think of that as a store of value motivation for holding reserves. And then the convenience yield from the liquidity benefits and then thirdly the balance sheet cost. There's going to be many different first order conditions one could look at. So linking into 3S paper would link mostly to the last one. So the second one of how much will the bank pay in the VPO market to borrow to hold reserves. That's going to be that additional term that has to do with the cost of posting collateral. So when you think about reserve demand, you can think of it being relative to whatever source of funding and thus the y-axis would change in your graph, but the intuition is always the same. So graphing this out, you can see then that reserve demand is just money demand for banks. The reserve demand function is downward sloping because this convenience shield is declining in reserves. The vertical dimension is determined by the interest rate in reserves minus the balance sheet cost and the line is going to asymptote to the IOR minus fee if the convenience shields go to zero. So there's some saturation eventually. All right, moving on to supply. The reserve supply, you can think of in terms of just the fat balance sheet, a simplified version is put here where the starting point, especially in the fat case, is that the fat buys some securities and has to pay for them. Some of the securities are paid for by the autonomous factors and the rest with reserves. So the starting point for how large reserves will be will be securities minus autonomous factors. We're going to call that net securities. Now, if the central bank here, the fat has a lending facility, then of course that's another way to supply security, to supply reserves. And on the other hand if the central bank has an investment facility for non-banks, the ORP facility in the fat case, well that's going to reduce the amount of reserves funding that the fat needs and will thus be subtracted. So reserves is basically net securities plus loans to banks minus whatever reserves the central bank has taken in in the form of the ORP. All right, so if you graph this out, I'm going to distinguish the cases with and without the ORP facilities since both are in our sample. If there's no ORP facility, then the supply curve sort of starts out at a vertical line of equal to net securities, and then the fat stands willing to add further reserves at the lending facility rate, which at the fat, of course, is the interest rate and this can be the primary credit rate. So the supply curve is vertical and then becomes flat. If there's the ORP facility, well then the fat doesn't just stand ready to add reserves but also to subtract reserves, and so then you get this second flat part as illustrated in the right picture at the ORP rate. Okay, so then throwing these two guys into the same picture, you can see that basically there's three things that can happen with the equilibrium, depending on where the demand curve intersects the supply curve. So the simplest case is one where the demand intersects the supply on the vertical part, their reserves are just equal to the net securities. The second case, sorry, in this case neither the discount window nor the ORP facility is used and the case is the simplest one. Maybe it could be the case as it is in a lot of the sample post-GFC that the demand curve hits the supply curve on the bottom flat part. So that's going to basically happen and you can sort of play around with moving the demand curve around. If there's not enough demand relative to supply, then there's going to be take up at the ORP facility and the way that that's going to show up in the picture is that reserves will be less than net securities and the difference will be the ORP facility which I have called the non-bank facility as indicated on the axis. And this of course sort of shows you why the ORP facility was introduced. It's a way to make sure that the market equilibrium industry doesn't fall below the floor set by the ORP rate. All right, so the third possible equilibrium is one where the reserve demand crosses supply at the top flat part of supply. So in that case the banking sector is going to borrow more reserves from the Fed through these loans to banks and you can see where that shows up on the horizontal axis. So this is going to happen if there's a lot of demand relative to supply. Formally it's going to happen if demand evaluated at the lending facility rate exceeds the supply, the net security supply, then you can see take up at the discount window. All right, so in terms of what we learned about interest rate control from this simple framework, the central bank controls the short market interest rate through a bunch of different channels. The first one is the vertical part of the supply curve, the net securities. But then also through controlling importantly the demand curve, the IOR, which remember shift the demand curve just up and down. The flat parts on the supplier curves also can matter if demand is really low or really high. So basically all the three administered rates matter. And it's also important to emphasize that the private sector plays a key role here in terms of actually using these facilities and thus facilitating either a reduction or an increase in reserves. All right, so let me turn to estimating this out and I'm going to focus on the post-GFC sample. So to do this, David and I, we assume a log functional form for the convenience yield so that it's log linear in reserves and deposits. And then we throw in a reserve demand shock and then you get to the red estimating equation in the middle. If you wanted to link this back to sort of more old school money demand literature, you can reorganize it and solve for reserves. And then you'll see the equation in blue which tells you this is what Bob Lucas would have called a semi log money demand function. All right, so I wanted to just emphasize a few points about the estimation. The first one being the importance of controlling for deposits. So deposits have gone up, they have gone up even relative to GDP, but the important part here is that they have gone up just in dollar terms. That started before the financial crisis and has kept going. Driven by, in the right picture, in red, an increase in liquid deposits, which are sort of the most reserve-intensive intents to manage. Okay, second point about the estimation is that we need to instrument for reserves. You probably already caught that one because if the demand curve hits a supply curve on the bottom flat part, then if you start moving around the demand curve that there's a reserve demand shock, the split of net securities between reserves and the ONRP will change. So that's a no-no for OLS because then you have a reserve demand shock, affects reserves, so you can run OLS. So we instrument reserves with the sum of reserves plus ONRP, which is equal to net securities when the lending to banks is small. We also, in the paper, instrument for deposits. I'm going to skip that here because it doesn't change the results much. What's important is to control for deposits because they go up over time, but not to instrument for them. Okay, so the estimation result then starts with the first stage of the IV estimation to divide. Not surprisingly, reserves plus ONRP is a very strong instrument for reserves. More importantly to the left, you can see that we get the expected signs. For example, the coefficient of log of reserves is negative. That's why we got the downward sloping demand. It says the convenience yield is declining in the reserve amount. And then we get the positive coefficient on deposits saying that the value of additional reserves is higher if the banking sector is larger, essentially. Translating it into that semi-log money demand, one can calculate that a 10 basis point drop in the Fedfront's IOR spread leads banks to be willing to hold 50% more reserves. So the reserve demand is very elastic, but it's not flat despite the large reserves in much of the sample. All right, so just to show you that the fit is really nice. It's actually easier to focus on the reduced form of the IV, which directly links the left-hand side variables to the instruments. And you see they are squared here from this incredibly simple model. It's like 90%. And the fit is good throughout the sample, as shown in the figure to the right. A way to summarize the estimation is to graph out the implications of the reduced form. And here I want to contrast the raw data to the left with the estimation result to the right. So in the data, I have to the left the Fedfront's IOR spread graphed against reserves plus ORP. You can see that it's trying to look like a downward sloping demand curve, but it's like something is pulling it out over time. And that's just, we argue, the growth of the banking sector over time. Once you control for that, which I have done by reviding the estimation equation in red into the one in blue to define supply adjusted for deposit, so adjusted for the need for supply, and then change that to put the x-axis, put that variable on the x-axis. You can see now there's a pretty nice fit. Notice also from that top point that the estimation is not surprised by September 2019 in the sense that given the reserve plus ORP supply at a time and the amount of deposits at the time, the regression tells us that that's when reserves were the scarcest over the sample and thus the actual and the predicted spread to the highest. All right, so just a word about why deposits went up. So we think that a key channel here is that there is, households just have more financial assets to invest. So if you do the household portfolio theory 101, then the households will put some in cash, think of that as deposits, some in bonds, some in stocks. So if they have more money to invest, you would get higher reserve demand. In the left picture here, I'm showing you household financial assets to GDP. You can see it's going up dramatically. The right figure is the portfolio share for deposits, which in the post-GFC period is sort of amazingly stable around 16, 17%. So that immediately tells you what might be a good instrument for deposits. It's the household financial assets. We also use the level of interest rates so we can do an O by D test, but the main strength is coming from financial assets. So just to flag, if you do the estimation instrumenting both for reserves and deposits, you can see the fit again is good. All right. So then let's think about what this all implies for policy. And let's start with how to set the interest rate on reserves. So to summarize the estimation, in the left picture, I'm graphing what is the predicted spread coming out of the IB reduce form given deposits at the end of the sample. So the estimation ends in the middle of updating, but the estimation ends in 2022 months time. So deposits were about 17.7 trillion. If you plug that in, you have the estimated coefficients A, B, and C. You can trace out the predicted spread for various values of reserves plus O and RP. That's what all those little red dots are. And the gray area is the range of data used in estimation. So now that says we have a pretty good handle on how the Fedfront's IR spread moves with supply. So from that, then we can think about how to set the interest rate on reserves to hit a desired target given the balance sheet size. So for example, in the right picture, I have illustrated it for a case where the Fed, they want the effective funds rate to clear near 4%. So intuitively, the way to achieve this depends on the size of the balance sheet. So for example, look at the left side of the gray area. Suppose the Fed had a balance sheet around 700 billion. Then reserves would be quite scarce. So if the Fed had a reserves plus O and RP of 700 billion, then reserves would be quite scarce. And there would be a pretty large positive spread. Therefore, if you want the Fed funds rate to clear at 4%, you need to shade the IR down to allow for the spread. And that's what you can see in the right picture. It comes out to like 3.75 or something like that at the left part of the picture. Conversely, if the Fed had a really large balance sheet, well then the Fedfront's IR spread would be negative. And therefore, to hit a 4% Fed funds target, you need to set the IR higher than the 4% target. So this red line to the right has a name. We call it the ISOFED funds curve. That's a terminology borrowed from a very good theoretical metric by Bianchi and Bidio. And we think this is sort of one of the first empirically estimated, hopefully the first ISOFED funds curve for the US. The other thing we can speak to is how much QTE is feasible before we might get to the point where there's a lot of curvature in reserve demand. And the left picture here is reserves plus ORNRP in dollars, and the right one scales by GDP just because that's how a lot of people think about it. In both pictures, the rightmost line is where we are at the end of the sample. That was at a reserve plus ORNRP supply of about 5.3 trillion as of last week. We are now down to 4.4. And then we use our estimated reduced form to think about how large the predicted value would be for various counterfactual levels of reserve plus ORNRP supply. Okay, so for example, let's start with possibility two. So we estimate that if reserve plus ORNRP was reduced to about 2.84 trillion, which is around 11% of GDP, substantially below today's level, then the predicted spread would be the same as in September 2019, which was around three basis points. That's the second line from the left in both pictures. So what this tells you is that one thing that, for example, would be quite risky from an industry volatility perspective, would be for the Fed to run down reserve plus ORNRP all the way down to 7% of GDP, which it did going into September 2019. That's the leftmost line in both pictures, and that would lead to substantially more reserve scarcity than in September 2019. Intuitively, because even though in both cases reserve GDP was around 7%, the banking sector has just grown in terms of deposits since then, so 7% of GDP is much scarier now than it was back then. So if you're concerned that even option two might be too scarce, given how September 2019 turned out, you could consider larger values and we consider a third option, for example, one would be where you said the predicted spread to zero, that works out to about 3.3 trillion in reserve plus ORNRP. Importantly, these are not numbers that are constant over time. As I have said many times, the banking sector grows over time, so these numbers are going to evolve with deposits. This is easy to account for. You just update the little blue deposit numbers before you draw the pictures. This is something one can do on an ongoing basis. All right, so then just two... I can move the slide since I'm out of time, so that's fair. So let me just say, remember that since the autonomous factors are very volatile, if you want to keep the reserve plus ORNRP equal to whatever number you like best here, with the autonomous factors fluctuate, you need to change the overall asset size correspondingly in order to make sure that an increase in the autonomous factors doesn't suddenly push the answer below the number that you have here. Thank you very much. Thank you very much, Annette, right on time. Let's move to Stefano online. We'll provide a discussion of the paper. Stefano, welcome. You have 15 minutes for your discussion. Yes, we share the slides. Do you see my slides? Not yet. There we go. Thank you very much for the invitation. Let me start with the usual disclaimer that these views are expressed here. I'm not necessarily reflected on one of these CBN in your system, and I take also the opportunity to thank colleagues in the research market operation and monetary policy directorate for the discussion that I had on the topic, on the reserve demand, and on this paper. The main reason is that the first time I read this paper was more than one year ago, because essentially the paper raised so many questions that also inside the ECB we started also to look at the paper to try to see whether we could replicate some of these facts for the euro area. So my presentation will be structured in two parts. First, I will have three main comments on the paper. Then I will share some of the number, so the exercise that we did for the euro area. I think that the novelty here of this paper is that you derive the reserve demand from a bank optimization problem where the deposits are a key variable. So the idea here is that the reserves earn an interest, but on top of that they also provide a convenience that is due to transaction cost savings. Essentially when the banks face a deposit outflow, instead of selling securities or liquidating loans, they can just use the reserves. And the main advantage of the framework that is proposed here in this paper is essentially that they provide parameters for the wide range of US reserves so you can recover the entire demand function. And, you know, you can run as Aneta did in the last part of her presentation also some policy exercise. And of course the big question here is that to try to assess the quantity of tightening to reduce reserves, but at the same time you want to keep control on the short-term rates. So the key ingredient here is this convenience here, that is defined as a sort of benefit function that is going to depend on reserves and deposits. You can also interpret as a sort of minus a cost function and the idea is quite simple. So more reserves reduce the price of reserves, more deposits are going to increase the price of reserves because essentially the cost of liquidity management is going up. And so the reserve demand is derived from very simple and elegant equilibrium relation. On the left side you have the marginal cost of borrowing in the federal fund markets. This is going to be the federal fund rate. And on the right side you have the marginal benefit of holding reserves. And now here you have three components. The first one is the classic interest on reserves paid by the Fed. The second one is the contribution is this marginal liquidity benefit from additional reserves minus you have a marginal cost of regulation. Here regulation penalizes the balance sheet expansion. Now to estimate this equation one, what you need is you need to define this benefit function. And this is the key ingredient. So the choice here is just to have these log linear firms. So the reserves and deposits enter separately. If you want to take a more traditional approach, it would have been to have just one variable that is reserve scaled by the deposits. In this case it's like in the loop model, you have the classic duplicate above reserves and deposits double. Here the choice by the authors is different I think that main advantage, this is my understanding for reading the paper, respect of the choice one is in terms of model fitting. And in fact, it's quite remarkable to see, you know, on the left hand side that the reserve demand so it looks quite unstable. But then when you control for deposits, the model fit is a super tight. The square is super high. You know, on the right hand side essentially is the figure that you have in the paper. Here I'm just plotting the fitted values with a blue line. And then the red line is the effective fund rate minus the investment interest on reserves. And you can see that the fit is very nice. Now, this is my first comment. I think the point of what I find quite interesting is that, you know, the implied elasticity of deposits is 1.79. So essentially what these estimates are telling us is that the price of reserves is more sensitive to deposits than to reserves. Now, this deposit is the second additional variable. So my question is, I find a little bit, to some extent, surprising and interesting to see such large sensitivities of deposits taking into account that also deposits are larger than reserves. I wonder whether it's due also, maybe to the functional form that is chosen. But I think this elasticity is a knowledge but this to some extent is not really discussed. The second comment is more about whether this relation isn't stable or there is more than one regime. So here I'm referring to this paper by New York Fed, a force of general knowledge by the Williams. I think the paper was presented two years ago here at the same conference. And if you look at this paper, essentially they discuss free regimes over the same sample. For example, if you take the reserves over bank assets, this is the left-hand side panel, you can clearly see these three regimes. So in the first regime is a sort of expansion from 2010 to 2014. And then you go from 2015 to mid-March with a contraction and then you have an expansion again. It's quite interesting. I really like the chart on the right-hand side when now you look at this reserve demand. And you know, it doesn't look anymore to some extent very unstable, but what you see essentially is that you see three curves, not one. And what is nice and interesting is that the location of this reserve demand has shifted over time. Now, of course, reading this paper, one could say, well, it's just, you know, deposits are the only demand cross-shifter or the main one. But I guess there can be, potentially there might be also other factors that can explain this shift from the location of demand that you see on the right-hand side panel. I think in that regard, it's quite interesting also to see this recent paper by Lagos and Navarro that they use a sort of quantitative, theory-based approach to assess how variation in key parameters can rationalize this shift in demand that you see on the right-hand side. Overall, I think that the question is, you know, what drives this convenience here? You know, one potential candidate can be, to some extent, his regulation, right? So the regulation here, if you look at the model, is a reduced form with a sort of linear cost that penalizes action and essentially the result in the demand curve is just a shift down. But, you know, if you read a lot of policy documentation, you talk to colleagues in policy, you know, most of the time you hear this argument that, you know, banks might have precautionary reserve motifs to comply with the liquidity regulation. This is why they hold large amount of reserves. Secondly, if you look, for example, also the liquidity coverage duration requirement, also banks might have a preference for meeting these requirements using reserves type of other HQLA assets. So I would like to see maybe a little bit more discussion also in the paper about other potential drivers for this demand of reserves. Let me move to the second part, to the application to the euro area. And I think here I'm talking again about regimes because usually what we have in mind in the ECB is like free regimes. You have the regime before Lehman that was like a sort of neutral allotment with low stable access reserves. And then you have a second regime moving from October 2008 to February 2015 where we moved to the fixed rate full allotment with moderate access reserve. Now, what is key here is that the liquidity is entogenously determined by banks' needs, the euro, for example. And then you have the third period where, you know, there is this injection of large amount of excess reserve, be it QE, PSPP and PEP, and then the euro. Now, if I look now at the evolution of reserves and deposits, I think it's quite interesting to notice also for the euro area that there is a strong movement between excess reserves and deposits in particular in the last period since March 2015. Yeah, it's almost a correlation of 92%. That can be also, to some extent, can raise some challenges when you estimate the model. You know, if I take the model and try to estimate this to be a sub-sample, I think what is interesting to notice is that so I run always the regression with large reserves and then I add also deposits in the second and fourth column. What you can see is that there is this positive coefficient associated with deposits in the second period from 2008 to 2015. It's also quite interesting to see how the elasticity of reserves changes. So to some extent this is provided to some extent some supportive evidence also for the frame for the euro area, but when you move to the last period, maybe because here the world is too flat, you know, you don't see much going on. Deposits do not play much a role. Now, I have to say that, you know, just taking the model here and running the regression, of course, is one possibility, but I think and it's quite clear also from the paper that, you know, you need to take into account seriously also the institutional context when you apply this type of framework. I think we had also in the third, this March 2015, we had also the two tier system for remunerating excess reserves. This is easy to take into account in this regression to some extent is not going to change much, but this is my view also. I think there are two factors that are quite interesting in the euro area. First of all, you know, when you take the framework and, you know, instead of applying to just to the area, you apply to the countries, maybe to the four major one, you can see that this elasticity can change. And finally, that is, I think for me, is a very important channel that also was discussed during the conference that, you know, the banks receive reserves on when they borrow from the euro system to their refinancing operation via calculus. This is a very important channel. Of course, you want to include that in the framework, but I think here the main difficulties is you need to find who's a good instrument for that when you want to apply this additional channel. Let me conclude. I think definitely this is a massive paper. I have to admit that it's not the first time that also my policy work is triggered by a paper by a net. I bet it's not going to be the last one. And let me thank you for your attention. Thank you very much, Stefano. Anette, would you like to come back to some of the points made by Stefano? Yeah, he makes a very good point about that the illicit reserve demand with respect to deposits is higher than one. If you put in, we really should have done this liquid deposits rather than total deposits. Liquid deposits are the ones that really necessitate liquidity management as opposed to time deposits. Then that illicit is much closer to one, so we'll do that in the next version of the paper. That's also much more consistent with what you get in cross-sectional regressions and what you would expect, you know, in terms of the supervision and regulation, I think is sort of an important point you bring up in terms of why is there such a high demand for reserves, which from liquidity management, also in the paper you mentioned by Lagos and Navarro, they sort of argue that just like managing cash and managing deposit flows, you wouldn't expect that much reserve demand. It's possible that we just got lucky that, you know, the LCR says you have to hold liquidity as a function of deposits, you put in deposits and it's working great, so that's obviously hard to distinguish from the liquidity management motive. Maybe you don't need to as a central bank. If you're just trying to figure out when reserves are scarce, maybe this is fine, but of course from an economic perspective it's important. In terms of the comparison to the paper by Afonso and others, the reason that they find the need to have time-marrying parameters I think is because they put in reserves to assets. I suspect that if they had put in reserves relative to liquid deposits they could get by with much more stable parameters over time, then they would essentially be back to the specification that we had, aside from taking a log. In terms of the comparison with ECB, in the central paper I do try to estimate this out for the ECB and you're correct that this is synchronometrically harder because reserves and deposits are much higher, much more correlated in the euro area than in the US so it's just synchronometrically difficult. The Bank of England has a great blog post where they try out this framework on their data and they find actually a good fit and also a higher coefficient when I ran it on the Bank of England data a higher coefficient on deposits and reserves again and so hopefully there are some robustness to be had but it's going to be key to look at places where reserves and deposits are not super-highly correlated. I do want to emphasize, though, even if they're super-highly correlated it is important to try to get this to work out because if deposits are growing over time it's not like the ECB can just say, look, our fit is great even if we kick out the deposits so just run the spread on reserves then you would get that reserve scarcity would kick in at a way lower level that is actually the case and you might end up with sort of a September 19 event but it's difficult but you know, I mean maybe if there's some generality to the coefficient across jurisdictions it would be helpful to start looking at what kind of illicities do they get at the Bank of England, what kind of illicities do we get at the FAD and other places where you can estimate it a little bit better given the data. But thanks a lot for a very productive discussion. Thank you, Anita and Stefano. Let's open the floor for some questions also for online participants. I see we have a question in the back. Hi, I'm Ansgar from Imperial College. I was wondering the result at the end, the sort of September 19 result if you change the size of your balance sheet too much you struggle to hit your target rates. Should I think of that as a long run steady state thing so that trade off is always there or could there be sort of some ratchet effect? You know that all the markets, all the general equilibrium, all the asset markets you didn't model, etc. have gotten used to a big FAD and so in the short run if you try to push back against that there's just a big adjustment cost and you start hitting that curvature. So both of them could kind of be consistent with your data because you're only estimating one demand curve. I'm wondering more broadly how you think about that. Thank you. Yes. The key ingredient of your model is this convenience yield of reserves. Now the events of last spring suggest that the policies are more volatile. You can have these sort of digital runs which presumably increases the convenience yield of reserves but on the other hand if you have the Fed stepping in as a lender of last resort you don't need to have reserves because you can always borrow. So I mean there are these two forces that are going to make it difficult to use your estimated equations for the current environment. Right? Maybe I abused my role here and I have a question about the distribution of reserves within the system. So you use aggregate data and there is a concern and I think part of that may have played out at least part of the literature for September 19 alludes to that that it is in fact different parts of the banking system having a different level of reserves compared to their assets or compared to their deposits that might lead to situations where you start having the inflection in the demand curve at a higher level than what macro estimates would suggest so what would you respond to that? Maybe we should just take that first round. So I think I'll ask a question in the back about the ratchet effect. So this is I think sort of a general terminology for the idea that if you have more reserve supply than reserve demand it sort of causally affects reserve demand as opposed to just moving the equilibrium along a given demand curve. In this context it sort of comes down to where the deposits went up because of QE and there are sort of different views on this like Chara, Raja and I think Viraal is coming tomorrow to present an argument that there is this ratchet effect. I had argued that there's also a large role for just deposits going up because household financial assets goes up and Will Diamond has a nice paper trying to quantify this out. People are kind of all over the place on this. I think it's still an open question. Whatever drives the increase in deposits I think everyone agrees that if there's a lot of deposits then there's going to be a lot of reserve demand and so from a central banking perspective in some sense again do you need to understand why the deposits came to exist? Possibly not. Then with respect to Raphael's question about can the March 2023 work fit into the graphs? I think it fits quite well. Remember I had a graph where you have the supply goes up and then becomes flat and then you have reserve demand and I said if demand shifts out then there's going to be borrowing in the discount window. So that fits right into the framework and then as reserve demand shifts down as this sort of precaution and demand for holding more reserve due to high deposit volatility fades into the whole thing works in reverse. So I think that works fine. In terms of the distribution of reserves I'm not sure if I could get a better fit. If I had perfect data and I could model everyone's reserve demands I'm not sure that would help me. I think what you're giving me is a sort of a reason for where the curvature coming from is that the marginal bank is now really reluctant to change their reserve holdings but if anyone can think of ways to sort of incorporate the 180 to actually get a better fit that's a good research area. Okay I think we need to move to the second paper now so Sria the floor is yours. You have also 25 minutes. Okay thank you. First please thank you so much to the conference organizers for accepting the paper into the program. We were very pleased to present this paper here today especially with this terrific program and audience. This work is joint with Romina Reprecht and Alyssa Anderson who are both at the board with me. Romina is actually in the audience and Ethan Cohen who is a first year PhD student at the University of Minnesota. And before I begin the views expressed in this paper today are ours and should not be reflected as of the Federal Reserve Board or the Federal Reserve System. So the motivation for this paper actually came from our policy work and that policy work which Annetta talked about at length in her talk was how has the Federal Reserve's new monetary policy framework which is now called ample reserves which I'll define a little bit later how has that changed the demand for money in the financial system. So central banks two tools usually right to do monetary policy one is interest rates and the other is the size of their balance sheet. And I would argue that the monetary policy transmission channels for interest rates are probably well known in the existing literature. In the paper we talk about two channels we call one the bank deposit channel and the other we call the non-bank deposit channel. And in the bank deposit channel that's in the very familiar paper now by Dreschler, Schnabel, and Savov and the QGE 2017 and in that paper they show that when the Federal Reserve raises interest rates that leads to a decline in bank deposits and lending. Now then Kairang Xiao came along the RFS in 2020 and he asked okay well if banks are losing deposits where does that money go? And so he shows that when the Fed raises interest rates that leads to a decline to an increase excuse me in non-bank deposits and this is what we call the non-bank deposit channel and when I talk about non-banks I'm specifically referring to money market mutual funds. And so in this paper we're going to ask okay so we know these two channels of monetary policy transmission for interest rates well what about when the Federal Reserve uses its balance sheet so specifically how do banks and non-banks demand money when the Fed uses its balance sheet as a monetary policy tool specifically we're thinking about quantitative tightening so when the Federal Reserve is reducing the size of its balance sheet as it's doing now and so we're going to write down and solve a structural model and then use that structural model we're going to calibrate the model to the existing to the current monetary policy tightening cycle and then we're going to try to see if we can figure out what the equilibrium size of the Federal Reserve's balance sheet should be and indeed how much reserves there should be on the liability size of the Fed's balance sheet and I'm going to put a disclaimer in now reassuringly we come to very similar estimates as Annetta's paper okay so what we find are two main results so we're going to show that the demand for money by non-banks that these more money market mutual funds and the capacity of the repo market to absorb this demand is actually going to be the first binding constraint on the size of the Fed's balance sheet and not the demand for money or reserves by banks which I think is a little bit different to what the existing literature has shown and so once we after we calibrate the model and then we run the model at IORB which is the interest on reserves that the Fed pays banks at 4.65% which it was in February of this year we estimate that the Fed could reduce its balance sheet by about 2.3 trillion which is consistent with Annetta's work and this is important because it's consistent with the Fed maintaining what we like to call as an ample reserve framework so that's the first takeaway the second takeaway is that I think we show this kind of novel complementarity between the two monetary policy tools of interest rates and balance sheets and we show that the higher the Fed sets its policy rate we show that the smaller its balance sheet can be and I will explain the mechanism driving that when I talk about the results of the model so before I get into the model I just want to briefly explain some how monetary policy is done at the Fed since we're at the ECB today and Annetta did already much of the work for me so here is that red line that she I think was also read in her talk as well which is the demand for reserves by banks so on the x-axis is I have reserves and on the y-axis is I have price of reserves and the Fed talks a lot now about what is called an ample reserve framework this is the monetary policy framework that the FOMC is committed towards and that's denoted by the blue vertical line which I've noted as supply and this is the idea that we're on the flat portion of the demand curve so small changes in the quantity of reserves is not going to affect the price of reserves but how can the Fed make sure that the price of reserves doesn't go to zero or below and so now we've moved to an administered rate policy so now we have two policy rates which one is called the interest on reserve balances which is essentially that's the interest that banks get when they put money at the Fed and the second rate is called the ONRP rate which is the overnight reverse repo facility rate and that's the bottom dotted horizontal line which is always going to be lower than the IORB rate but economically I like to think of the ONRP rate as the non-bank reserve interest rate so it's the same thing as banks Donbanks can put money at the Fed and they get an interest rate so it's the same thing economically so right now the reserves are around 3.3 trillion and then what we're going to use the model for is okay what is X and that's the vertical line so essentially how small can the Federal Reserve's balance sheet be that's still consistent with being in this ample reserve framework meaning on the flat portion of the demand curve and this is essentially what we're going to use the model for to ask what the equilibrium size of the Fed's balance sheet should be taking into account the demand for reserves by banks and the demand for non-bank reserves by these money market mutual funds so before I also get into the model I just wanted to also show what's been going on over the last year and a half so the Federal Reserve started hiking interest rates in March 2022 and it started declining the size of its balance sheet in June 2022 so on the X axis here I've plotted from January 2022 till I think maybe last month and I have three lines the two takeaways are as follows so the yellow line is bank reserves but the blue line is what I like with the non-bank reserves this is what is the take up at the overnight reverse repo facility and the red line is repo volumes I am specifically plotting overnight treasury repo which Nanowenshin talked about at length the take away from this chart here is the red line has been going up the blue line has been going down but I argue that the yellow line has stayed pretty flat and this is over a period of monetary policy tightening with the Federal Reserve increasing interest rates and declining the size of the balance sheet and these empirical observations I think are unique because well we think a lot about bank reserve demand but we've been doing monetary policy tightening for about almost two years now and I think the yellow line has stayed flat so the model is trying to capture what essentially the red line is doing moving up repo volumes and the blue line going down which is essentially a non-bank demand for reserves at the Fed so let's get into the model so the model is built upon two papers one by Rocker Mentor and Ben Lester in 2017 it's about the federal funds market and then Kyron Shows 2020 RFS paper that illustrates the non-bank deposit channel we have a two period model and five types of agents so we have banks or non-banks which are money market mutual funds broker-dealers households and firms are going to have unit mass for each type of agent and if you will a sixth agent the central bank, the Federal Reserve which will implement monetary policy so it's going to set the interest on reserves which denoted as capital R the overnight reverse repo rate which is lower case R and it's going to arrive in the model exogenously with a big balance sheet so it's going to buy government bonds from households which is BCB and this model is really trying to understand quantitative tightening and so that's why we exogenously make the central bank come in with a big balance sheet so I'm not really going to show you much math here today I find it very helpful to just show the T accounts of all the agents so the mechanisms of the model hopefully are easily understood so let me just first start with the central bank which is in the top left so on the asset side of the balance sheet the central bank has bonds treasury securities specifically in our model a general good PX which we have in the model to create a demand for money in the model the liability side is two-fold reserves which is the cash held at the Fed by banks and then the ONRP which is the cash held at the Fed by the non-banks moving to banks banks receive deposits DB and they can invest in either loans or reserves money funds which is the middle left those are non-banks they receive deposits they also have two investments they can either lend in repo markets or put their money at the Fed with the ONRP now households are the agent that's going to be deciding who gets the deposits so they have equity in the model and then they invest that equity in the money fund deposits and obviously how much they decide to allocate between banks and money funds is going to be a function of the interest rate that banks give them versus the money funds give them and then finally we have dealers now dealers they're going to be getting bonds and they have to finance all of them in the repo market and this is what's going to be creating the demand for financing in the repo market because as the central bank is going to reduce the size of its balance sheet using BCB then that means that the dealers have to step in and buy those treasury securities which means they have to finance them in the repo market but before I get there let me just quickly talk about the deposit market as I mentioned households will allocate their equity between banks and money funds and that's going to be a function of the interest rate that either agent provides so the bank deposit rate is going to be a function of capital R capital R which is the interest on reserves that's how much they get for holding at the Fed plus IL which is the second term that's how much they get on loans the interest rate they get on loans minus the monitoring costs IDM which is the interest rate that money funds provide is going to be a function of row which is going to be the repo rate minus some cost for them such as the fees they might impose and so the important thing in the model is that a pass through of an increase in IORB which is capital R is going to be larger for the money fund deposit rate than for the bank deposit rate and this is essentially the bank deposit channel and the non-bank deposit channel I talked about earlier so this is very much so in the data so as a result in the model because IDB is going to be less than IDM as R goes up capital R goes up that means that money is going to be flowing from bank deposits to money funds when the Federal Reserve raises interest rates now banks their maximization functions as follows they can either take that deposits that they get from the households and they can either invest in loans which is the first term minus those monitoring costs or invest at the Fed and put them in reserves so we're going to impose a constraint on banks which is the delta in the second line and that delta is trying to capture the delta is a minimum reserves deposit ratio that we impose in the model that banks hold reserves because Federal Reserve no longer has reserve requirements but the delta is also there to kind of capture things that we I think was just talked about in the previous paper the fact that some banks just have some preferences for holding lots of reserves that we just don't understand and they might be related to some regulatory ratio such as LCR and SLR there's preferences to hold safe assets now let's move on to non-banks in the repo markets the money funds so now the money funds they also have two investments they receive deposits from the households but they can either lend in the repo market and receive row or lend their deposits at the ONRP at the Fed and that interest rate is lowercase r so how is demand for liquidity how is that driven in the model so the demand for liquidity comes from the broker dealers in our model and that demand is created as the Federal Reserve reduces the size of its balance sheet by reducing the amount of money that's going to come from the money funds and so the money funds are going to decide how they're going to allocate those deposits and that's going to be a function of okay what's the repo rate which is going to be a function of the demand for financing by those dealers that's the first term and then the second term is essentially how much they would get in interest rates and then the second term is essentially how much they would get in interest rates essentially how much they would get in interest at the ONRP so we write down the model and we solve the model and we get two equilibrium so we're just going to go over the equilibrium now so the first equilibrium is what we call the access liquidity equilibrium which is also you can think of as the ample reserves equilibrium for the ample reserves framework that the Federal Reserve is currently in that equilibrium is characterized as follows so the demand for liquidity is low the Federal Reserve is still reducing the size of its balance sheet but the demand that dealers need to finance those bonds is low relative to the liquidity that money funds have in deposits and so that's going to mean that RO which is the repo rate that money funds can lend to is going to be equal to the overnight reverse repo rate and because they're equal that means that the money funds are just going to put all of their money at the Fed instead the second equilibrium I like to call the scarce liquidity equilibrium and so that means now that the repo rate is greater than R there's enough demand in the repo markets by dealers to finance those treasury bond holdings that leads to an increase in the repo rates so money funds adjust by taking money from the ONRP and lending in the repo market instead leading to RO which is greater than R and that means that there is no more take up or investment or non-bank reserves at the Fed because of these two equilibriums then we back up what we like to call a critical threshold of central bank holdings which we call BCB tilde and this is the level right at the tipping point where you go from excess liquidity to scarce liquidity such that the repo rate is going to be equal to the ONRP rate that's the rate that non-banks get at the Fed but ONRP take up the amount of non-bank reserves at the Fed is going to be zero and so this is the threshold this is the minimum size of the Federal Reserve's balance sheet that's consistent with an excess liquidity regime it's that tipping point going from excess liquidity to scarce liquidity and you can think of I'm trying to back up that X which is that vertical dash line I showed you in the background slide that's the smallest balance sheet that Fed can have that's consistent with an ample reserve framework so before I go into some of the results I'm just going to go through the mechanisms one more time in the T accounts so these T accounts hopefully you're a bit familiar but now this is what's going on so in the excess liquidity regime as the Federal Reserve is reducing the size of its balance sheet you can see BCB going down but if I quickly go into the money funds now they're allocating less towards the ONRP which is those non-bank reserves and more into repo now from the dealers you can see here that the BD the amount of bonds that they need to finance is going up and they're financing that in the repo market this is where the fund stuff occurs and the scarce liquidity regime again the central banks on the asset side the bonds that the Federal Reserve holds it's going down but now this is where the households come in so now the households are going to be adjusting the deposits they have from banks towards money funds why so as the central bank is reducing the size of its balance sheet by reducing the amount of Treasury securities it holds that's creating demand for financing by the dealers you can see their T account the asset side and the liability side is going up this is putting pressure upwards on the repo rate RO which is now going to be above the ONRP rate which is the non-bank reserve rate and so money funds are therefore going to start shifting money from the ONRP and into repo and since now RO is now going up and is greater than R they can then give a better interest rate to the households so IDM the interest rate that they can offer to the households is now going up a lot more than IDB which is the interest rate that banks can offer and this is causing households to adjust their deposits by moving from banks to households okay so we write down the model we solve it and then we calibrate it to the current monetary policy tightening cycle which we is from September 15, 2021 which is the peak level of reserve balances on the Fed's balance sheet to the end of last year and so we're going to calibrate the model to match several moments the first is RO the repo rate we're going to be looking at the overnight tri-party general collateral rates this is the GC market overnight against Treasury repo aggregate bank deposits the interest rate on bank deposits and money fund deposits which is and then aggregate non-bank reserves at the Fed and then aggregate reserves, just amount of bank reserves at the Fed and we also have some independent parameters in the model so we're going to be putting in the current policy rates so the average over this period is 1.38 and 1.28 we're going to be putting in the average interest rate on banks' outside investments this IL is like the loan interest rate on what banks get if they were not to invest in reserves we take this from the call reports and so we're going to be putting in the average interest rate of all the way to the average of loans and securities that banks can hold we put in how much Treasury securities are held at the Fed how much Treasury securities exist in the economy and then this is that delta is that minimum reserve to deposit ratio which is one incentivizes banks to hold reserves and is trying to proxy for these preferences for banks to hold reserves because of regulatory ratios so how do we do so here is the model after we calibrate it and then we run it for the February 2023 FOMC where the Federal Reserve did hike interest rates and I think we do pretty well so the repo rate we match pretty well the interest rate on money fund deposits bank deposits, aggregate O&RP gauges that non-bank reserves aggregate reserves I think we do really pretty well what we don't do well is the interest rate on bank deposits we predict 18 basis points we predict 3.28% when in reality it's 18 basis points so if you have any take away in this entire presentation if you have any money invested in the US please make sure it's not in a bank and you put it in a money market fund and this is a bit of a puzzle I think when we looked at a lot of other structural model papers this seems to be something very hard for a lot of papers to model pretty well it's a bit of a puzzle why households keep all their money in banks but we do a lot of robustness in the paper to make sure it's not driving our results and speaking of results let me talk about them so the first result is that we claim that the capacity of the repo market and the demand for money by non-banks is going to bind before reserve demand so this graph here on the x-axis we have capital R which is the interest on reserves and on the y-axis we back out the capital of reserves and so we're doing it here for 4.65% which is at that time what we calculated the model for but now we know that R is like five and a quarter so the blue dotted line you can think of anything above that line that were in excess liquidity equilibrium anything below the blue dotted line were in the scarce liquidity equilibrium the yellow line is what we backed out as actual bank reserve demand so we in the paper say ok if you're the Federal Reserve if you want to reduce the size of your balance sheet to BCB Tilda which is that critical level of the central banks balance sheet where we go from that excess liquidity to scarce liquidity that's the blue dashed line which would equate to about $3.2 trillion in reserves but if you were to ignore that and just think about bank reserve demand that would go down to about $2.1 trillion of reserves which is you know a big difference and that's fine right because the Federal Reserve ok you're only monitoring banks you're only supervising banks that you typically think of that's fine but what the red dotted line shows that if you were to reduce the amount of reserves past the red dotted line that indicates then that the repo rate would start to become much much higher than the Federal Reserve's policy rates and since the Federal funds rate which is the rate that the Federal Reserve typically controls follows the repo rate not the other way around we argue that if the Fed were to reduce the amount of reserves on its balance sheet past the red dotted line that's when the Fed would start to lose interest rate control so this is why we argue in the paper that it's really this capacity of the repo market and the non-bank reserve demand that it's going to be the first binding constraint on the size of the Fed's balance sheet on the interest rate control rather than just thinking about bank reserve demand alone and then the next natural question might be well why does the capacity of the repo market matter is it an actual friction so Adam Darrell and David Yang have a paper that showed that reserve balances recent all-time low on September 16th 17th 2019 it was also mentioned I think several talks here today we argue that reserves also reached an all-time low on that day but we also argued that this friction this capacity of the repo market of dealers being able to finance those treasury securities also started to bind on that day and we run the model using our calibrated model calibrated values to predict what the model would have said happened on September 16th 2019 and I think we do pretty well so we do get that spike in row which is the repo rate and we are able to match bank deposits aggregate ONRP take-up and aggregate reserves again we are not able to match the interest rate on bank deposits well again but I think this is always a little bit of a puzzle in the literature so another take away perhaps from this paper is that yes reserves reached an all-time low on September 16th 17th 2019 but there were also frictions in the repo market that played an important role contributing to that repo spike and that was due to the non-bank demand for money. The second result is this complementarity between this fact that as the Federal Reserve raises interest rates more than that means that the equilibrium size of the balance sheet can decline more as well and the mechanism behind that is because as the Federal Reserve increases interest rates, money funds can offer a better rate to households they are able to invest more money into the repo market which allows to cater to that extra dealer demand as those treasury securities the Fed no longer holds. So in conclusion, we show in this paper there is also a deposit channel of monetary policy as the Fed does quantitative tightening and that non-banks have an important role in determining how the Federal Reserve does monetary policy implementation in terms of quantitative tightening. Thank you. Thank you very much. Rafael you have 15 minutes for your discussion. Well thanks very much for the discussion. This is not an easy paper to discuss but I must say that I learned a lot reading this paper. Now everybody knows that sorry, what happened? Can I have the slides back? Okay, right. So since the global financial crisis central banks have combined these conventional interest rate tools with these unconventional quantitative tools and QT so they've gone from scarce to an ample reserve regime so the policy rate becomes the interest rate on bank reserves. Now, the paper addresses a key issue in monetary policy implementation. I think that this is something that here in Frankfurt they are sort of studying carefully what are the effects and the limits of QT how do they compare with changes, increases in the policy rate. The paper incorporates institutional features of the U.S. financial system, in particular these behavior of banks and non-banks money market mutual funds and also these institutional features of Fed monetary policy, the distinction between the interest rate on reserve balances and the overnight reverse repo rate, the own RRP. The main results in my view are the following for a given policy rates and ample reserves QT mainly affects reserves of non-banks. The limits of QT therefore depend on the holdings of reserves by non-banks and this is the way in which you can understand the title stop believing in bank reserves because the heart of the matter is what happens to the non-banks not to the banks. The second result is that a switch to the scarce reserve regime depends on the policy rates or the policy rates and you can have more QT with higher rates so these are the main results of the paper. Now, the paper structure as follows there are some aggregate time series evidence there is a theoretical model, there is a calibration and then there is the discussion of the results. I have four main comments which are in this slide first, I think that this is an ambitious paper on a very important topic for central banks and let me add that I have found surprising that there hasn't been that much research not by academies for them this is very far away from their interest but inside central banks and we are thinking about this during the QT but what about QE have you had many years and where are the papers right? Now the paper in a way seems work in progress the results are very promising but I think that there is a fair amount of polishing that needs to be done. In my view the theoretical model has too many what I call peculiar features that I will describe in a minute and this is going to be the focus of my discussion and finally I think that I couldn't resist saying that this aggregate time series evidence doesn't add anything this is visual correlations of endogenous variables please don't do that anymore I think that this is becoming very popular but it doesn't tell you anything ok let me just briefly summarize the theoretical model these two periods, five types of private agents, the household banks non-banks and dealers plus the government a central bank, households have an initial endowment and then investing bank and non-bank deposits the firms are going to produce and sell consumption goods to the households and household can only pay firms with bank deposits so there is like a cash in advance type of demand for bank deposits. The banks are funded with household deposits there is no equity capital, they invest in research and loans to other unnamed agents they are subject to linear balance sheet costs, subject to a reserve requirement and on the other hand the non-banks are funded also by households deposits they invest in research and loans to dealers and they are subject to linear balance sheet costs the dealers are funded by the non-banks and they are going to invest in government debt, ok that's the natural model. Oh yes there is a central bank that says the total amount of research held by banks and non-banks, the interest on reserve banks let me deviate from your notation I prefer to use in the small r for interest rates, net interest rates rb is the interest rate on banks paid by bank research and rn is the interest rate on research by non-banks and of course there is a gap currently 10 basis points between the two rates, ok peculiar features of the model. In the model there are two types of goods, there is a general good which is produced by the government the central bank and this is a special good produced by firms ok well I mean central banks producing goods is not something that I find most appealing but I am a theorist, ok I am willing to do almost anything but I mean this is second the determination of the positive rates is obtained by bilateral bargaining between banks and depositors I think that this is probably an unnecessary complication and finally there is an exogenous fixed loan spread which hopefully could be endogenized properly the second comment is that I think that there are some unnecessary elements, the dealers are funded by non-banks and they invest in debt so I guess that you could assume that non-bank could directly invest in government debt and you kill the dealers they don't play any role in the setup banks reserve requirements don't play any role, they are calibrated to a very high level 13% because of this is the September 2019 so again not really essential what I think are missing elements is lending to banks by non-banks banks issuing commercial paper that is funded by non-banks and this is an important adjustment mechanism not in the model leverage constraints for banks I think that this is also some important missing element, you should have some way of limiting borrowing by banks from non-banks otherwise there is an average trash opportunity because of the gap between the two policy rates and otherwise non-banks would not keep any reserves, they would move other reserves to the banks and the banks would get the higher policy rate so what am I going to do in the remaining nine minutes I'm going to sketch it's only a sketch, a simpler theoretical model that yields similar results the ingredients is there's a conventional central bank that doesn't produce any goods there's households with bank deposits in the utility function there's a local monopoly banks setting the loan and deposit rates and there is competitive non-banks so again two periods I killed the dealers households have an initial endowment they invest in bank and non-bank deposits and firms are going to be borrowing from banks to produce output the banks are monopolists with respect to households and firms these are local monopolies there is an economy-wide market for non-bank deposits they borrow from households and possibly non-banks they invest in reserves and loans to firms and they are subject to a leverage ratio which puts us an upper bound on asset side, this is important as I argued before the non-banks are competitive they borrow from households, they invest in reserve government debt and loans to the banks so I'm going to focus on the ample reserve regime so this would be the balance sheet of the non-banks reserves I have the government bonds, remember I killed the dealers and loans to banks which is the new feature and then on the liability side they have these deposits with a sub-index end because these are the non-banks so if the amount of reserves is positive a zero profit condition, remember that they are competitive it would imply that the deposit rate is equal to the loan rate and is equal to the interest on reserves that obviously of course simplifies a lot the analysis balance sheet of non-banks we have the deposits and borrowing from the non-banks reserves and loans to firms on the asset side and of course if there is a gap between the two policy rates the upper bound on asset side is going to be binding otherwise there would be an arbitrage opportunity so the sum of reserves plus loans to the firms is going to be equal to this upper bound so of course if they borrow from the non-banks at the RRP they are going to be making a small profit and this could be considered a subsidy to the banks equilibrium loan and deposit rates the interest on reserves is the opportunity cost of loans and the equilibrium loan rate it would be something like that the loan rate maximizes the spread between the loan rate and the opportunity cost multiplied by the downward sloping demand for loans and that would be the equilibrium loan rate the equilibrium deposit rate is obtained in the analogous manner the spread between the marginal revenue of the deposits which is the RB and the deposit rate and of course in the case of the deposits the demand the household supply of the deposits depends on both the interest that they get with the banks and the interest that they get from the non-banks which is higher what is the effect of QT on banks? well, loan rates and loan quantities only depend on the interest rate on bank reserves if you keep the interest rate on bank reserves there is going to be no effect on deposit rates and deposit quantities and there is going to be no effect on loan rates and loan quantities so QT doesn't have any effect on banks in this simplified version what about non-banks? well QT only affects the size of the balance sheet of the non-banks reserves go down government bonds go up, this is the open market operation involved in QT and therefore no change in household deposits or in loans to banks QT is neutral, it doesn't have any real effects in this simplified model ok limits of QT, well given these two policy rates QT can proceed as long as the amount of reserves held by the non-banks is positive so it's basically the same result as the paper the limits of QT depend on the holdings of reserves by non-banks stop believing in bank reserves there is a message that I take it from the paper which is in the title what about, I mean I'm going to play around with changes in interest rate but I'm going to split, I'm going to say ok what happens if we increase the interest rate on reserves of the non-banks now if the amount of non-bank reserves is positive the zero profit condition stays the same as before and so an effect of an increase in Rn for a fixed RB is that there are going to be a shift from bank to bank deposits and this is going to increase in non-bank lending to the banks because nothing is going to happen on the banks side so this is the, for the non-banks the deposits go up, no change in reserves is going to these additional funds are going to be re-loaned to the banks, no change in reserves and in the case of the banks the deposits go down, they shift to the non-banks and the non-banks recycle these funds back to the banks and therefore this would be the change in the effect of a change in the deposit rate of the reverse ripoff facility what happens if you change the interest on reserve balances, well the two first order conditions that I didn't write before but you can write very easily imply that loan and deposit rates are going to go up this is going to reduce bank loans it's going to increase in bank deposits it's going to lead to an increase in bank reserves because of the you have smaller loans so you just fill the gap with additional reserves and it's going to have an ambiguous effect on bank profit so this is what happens, the deposits go up loans to firms go down given the upper bound on the banks balance sheet because of the arbitrage opportunities reserves are going to go up and the loans by the non-banks are going to go down so no change in the size of balance sheet and in the case of the non-banks the deposits go down and all these funds are going to be I'm going to be translating to reduce loans to the banks so because of the shift from non-bank to bank deposits ok finally what happens this is the exercise done in the paper if both rates increase by the same amount at the same time well the balance sheet of the banks first the deposits go down loans to firms go down but now there is a difference I mean there is no change in the size of the balance sheet by the leverage constraints so reserves must go up and therefore the banks have to borrow more from the non-banks and in the case of the balance sheet of the non-banks what happens is that the deposits go up loans to the banks go up because of the shift from bank to non-bank deposit remember that the bank deposit market is not competitive so the deposit rate stays this is one of the puzzles in your basically zero whereas the non-banks go up so therefore that moves households from banks to non-banks and so there's going to be a reduction in reserves it's total risk to be unchanged because this is just a pure changes in interest rates ok so summing up this alternative model that I have sketched I think avoid some of the shortcomings of the theoretical model in the paper it yields similar results in particular the fact that limits of QT depend on the holding of reserves by the non-banks the alternative model yields some contrasting results so if you increase both policy rates it's going to reduce non-bank reserves and therefore it leaves less room for QT with higher rates so this is the opposite result but of course the model is different so it wouldn't be surprised that you get a different result ok just to conclude last two slides first I think that as I said at the beginning the paper addresses a key issue from a novel perspective incorporating these institutional features about the US financial system and the Fed monetary policy I think that there are many interesting questions to be addressed with this type of model the effect of for example equating the two policy rates the interactions between monetary policy and bank regulation remember the key role in my model of the limit to the size of the bank's balance sheet obviously in a place like this differences with ECB's monetary policy implementation but I think that to conclude I think that much more research is needed I mean it is my bias but I think that theoretical contributions would be especially welcome and let me just conclude with something that well should be a sentence very familiar to all of you I think that we need richer models the simple models of these two periods like the one that I have sketched cannot address Bernanke's conundrum with quantitative easing or quantitative tightening is that it works in practice but it doesn't work in theory in this type of models QT is neutral doesn't have any effects on real variables so in order to get real effects you have to think about the effect of QT or QE on credit spreads on sovereign spreads whatever I mean you must have models that have much more heterogeneity across other dimensions in the economy so there's a big amount of work to be done on these matters so thank you very much thank you very much Trafal Shreya would you like to come back to some of the points made by Professor yes first of all thank you very much for that excellent discussion and the model makes us think which definitely will make sure that all our results are robust I think the one thing and I might have gotten this wrong is that I think your model doesn't have a repo market and I think that was one of the channels that we wanted to show and that the capacity of the repo market is also a friction on the size of the Fed's balance sheet but I think your point is well taken but the repo it is in the model but it operates only through the banks I mean the banks should be there but perhaps there are other agents that's not there yeah we have gotten that feedback before to combine the banks and the dealers and we'll revisit that the reason we kept it separately because we were like dealers don't hold reserves so it's nice to keep them separate from a thinking perspective and also your point is well taken that the model doesn't have any real side implications yeah thank you very much I will definitely need your slides okay so we're already past a lot of time but let's maybe take one or two questions and Eta so you would like to... I just like clarifying thing too I don't want to cut out the questions any other questions okay let's take those please Hi, Christian Kubitsa, ECB thanks a lot for the presentation very much enjoyed it I'm trying to wrap my head around how I should think about your paper in relation to the first paper that we saw today because your paper basically argues we should first increase rates and then we are able to do quantitative tightening the first paper today argued that quantitative that we should do first do quantitative tightening in order to reduce scarcity effects that prevent monetary policy transmission so what should we do first conventional or unconventional monetary policy and maybe the difference is related to the different institutional setup because we don't have the reverse repo facility in your area so do you think that's the main reason for this difference in perspective or is there something else behind it we're very curious to hear what you think about that thanks maybe let's take that final question from the gentleman there Hi, Miigle Gervari from Banc de France I'm a co-author of the first paper but I'm not going to take your question for sure I'm going to let someone essentially this very difficult point I think your paper to me is the first one which really provides an explanation of what happened in September 19 other than scarcity of reserves I think your point is more that what was going on is more on the repo market side I think it's very welcome lots of people have this intuitions, discussions but nobody could well show it so now my question is if we want to go further and to try to empirically kind of try to discriminate between scarcity of reserves or excess of bonds would you have any idea of whether by looking at the Fed funds market versus the repo market and the fact that as we saw also in the keynote speech today the repo market spiked up much more than other money market rates is there anything you can exploit there to convince us that it's not reserve scarcity but actually what's going on the bond market that's spiked, thank you thank you, Anette would you still like to make your clarifying? Yeah, so just clarifying the links, so remember the picture I had with the reserve demand and then I said you're going to get an ORP take up if there's too much supply relative to demand so if you in that context lower supply at first dollar for dollar you're just going to get a reduction in ORP take up then at some point you're going to start seeing the Fed funds rate lift off and you know so for lift off the floor I think being actually in complete agreement the point at which the ORP hits zero is going to happen before you start seeing substantial curvature and that's also what happened in the last QT that you saw the ORP goes to zero quite a lot before September 19 so in that sense there's no contrast I also just want to remind you remember and I took the first order condition I said there's many different first order conditions we could do you could borrow in many different ways to invest in repo to invest in reserves so it's not like there's different first order conditions for one and the other it's like there's a reserve demand relative to repo and in that sense I think it all is pretty consistent I would agree I think one thing I thought your paper also said that you would recommend increasing interest rates first and then doing quantitative tightening or do you make a statement on the sequence because that's in our paper if you increase interest rate the key thing is whether you want to manipulate deposits if you start moving the interest rate then you're going to get all the interesting effects of money flowing from one to the other one but that's an important result of our model in the sense that we do recommend that you have to increase interest rates first in order to have room to reduce the balance sheet and I think also the second part of that question was the difference between ECB so this is very much a learning trip for me because I think that money market mutual funds and non-banks are not as much of a player here as it is at the Fed in the US so I think when the Federal Reserve increased the size of its balance sheet significantly during the COVID pandemic I think that was really when we saw a lot of this non-bank demand for money and I would argue that bank reserve demand kind of got saturated at that point already very very big so maybe there could be a point where non-banks do start playing a lot big of a role in the ECB if the balance sheet did get very very large and the ECB started providing essentially a tool like the O&RP here so that would be my but I think that's the biggest difference is the O&RP between the Fed and the ECB and then the second question which was empirically being able to disentangle the scarcity of reserves versus the capacity of dealer balance sheets in September 2018 yeah I think that's very hard I think both are at play I don't know which bound first and I I think that's a very interesting question that I have to think about how we could ever exploit that empirically thank you for that though Well I mean this is a point about the sequencing now inflation shoots up the Fed or the European Central Bank has to tighten monetary policy there's no question about that QT is doing nothing for so therefore you have to raise interest rates all these discussions are an academic discussion interest rates must go first and then of course you want to do the other stuff for other reasons but in terms of the policy issue it is very clear the sequencing that you have to follow thank you also for this very clear advice and and with this we conclude this session the conference will reconvene at half past three with a very exciting market panel so I strongly encourage everyone to come back also virtually and in the meantime please join me in thanking and congratulating the speakers