 Let's move them to the next paper. The next paper is about how do banks manage liquidity, evidence from the ECB's tiering experiment. And I see the presenter also already online. So Jean-David Seigu from the ECB. So the floor is yours. Thank you, Tobias. And I'd like to start by thanking the organizers for letting us present our paper today. So the paper I'm presenting today is called How do banks manage liquidity, evidence from the ECB's tiering experiment. It's with Luca Baldo at the Bank of Italy, and my ECB colleagues, Florian Haider, Peter Hoffman, and Olivier Verco. Let me first talk about the research question and the motivation of this paper. So the research question is actually in the title of the paper. How do banks manage their liquidity? And the answer to that question is not obvious, because banks have a lot of leeway in the management of their liquidity. Leeway in the composition of their portfolio. Do they hold reserves, sovereign bonds, or even assets that entail some haircuts? Leeway in the use of liabilities to raise liquidity, like pouring reserves from other banks. And leeway in how they dynamically rebalance their liquidity portfolio. Stating the obvious, liquidity is important for banks. It reduces the risk of runs and spillovers to the real sector. It also acts as a buffer to absorb payment shocks. Now, understanding how banks manage their liquidity as major implications. One set of implications is for spillovers. A shock on one liquid asset may or may not spillover to another market, depending on how banks manage their liquidity. And in particular, this affects the extent in which monetary policy as a shock to reserves transmits to a securities market. It has also implications for bank borrowing and lending behavior. It enhances our understanding of banks' reliance on short-term money markets and on their use of intra-group capital markets. So it's an important question to tackle. With the research question in mind, how do banks manage their liquidity? Let me talk about the methodology. So in this paper, we explored a shock that led banks to rebalance their liquid portfolio. And the shock that we explored is the introduction of the two-tier system for remunerating excess liquidity holdings that we will call tiering for short. That's been introduced by the ECB in October 2019. And according to the system, up to a limit, excess reserve holdings is remunerated at zero basis fund instead of minus 50 basis fund. And you say that again. Part of the reserve holdings up to a limit is remunerated at a rather attractive rate, zero basis fund. And if the bank has more excess reserves than that, it's remunerated at minus 50 basis fund in excess of the limit. So one can think of tiering as a decrease in the opportunity cost of holding reserves up to a limit. Hence, one can expect a higher demand for reserves for going up to the limit and a lower demand for close substitutes of reserve. So in this paper, we have two steps. In the first step, we spend a long time verifying that the shock indeed led to a rebalancing. So the prediction here is that banks rebalance their holdings of close substitutes of reserves, meaning they decrease their holdings of such assets. And our identification strategy is a difference in difference. So as to be able to solely capture the rebalancing that are induced by this shock and nothing else. In the second step, since we have established that there is a rebalancing, we study the rebalancing. And in particular, we test two alternative hypotheses. The first hypothesis is that the rebalancing preserves the portfolio composition, meaning if before the shock, a bank had one-third of sovereign bonds and two-thirds of intra-group loans, for example, it must be, that hypothesis is true, that after the shock, it still has one-third, two-third. The only difference is that after the shock, the size in euro term of this portfolio would have trunk. In order to be able to finance the increase in excess reserve. And we call this hypothesis trade-off. This term of trade-off has a reference to the corporate financing treasure. The exact reference is not important here for this talk, but you just have to understand that trade-off means that each substitute of reserves has some pros and cons. And it leads to an optimal liquidity structure that is somewhat time-invited in the absence of shocks. The second hypothesis is simply that the rebalancing is such that banks used a single substitute of reserves. For example, banks lack very much to sell sovereign bonds in order to get liquidity. Then according to that hypothesis, to accommodate their liquidity needs after the shock, that's what they did. They only decreased their sovereign bonds. And we call this hypothesis pecking order. And this term just means that they have a strong preference for a given substitute of reserves. And once, of course, they sold all of it, they go to the second best. Our identification strategy is we do counterfactual simulations aligned with theoretical benchmark. And as for the difference in different strategy and that strategy, I'll talk about the details more during this presentation. What are the results? The first set of results is that the shock led banks to rebalance holdings of three close substitutes of reserves. First, money market loans. Second, intra-group loans. And third, marketable securities holding. For a total of 3.40% of assets, which is quite large. The second set of results is that bank behavior is consistent with the existence of the trade-off, i.e. an optimal liquidity structure, and not so much with a pecking order. That is, banks rebalance proportionally to their pre-shock holdings. And it suggests that banks have a target allocation indicative of a trade-off. On the contrary, they do not seem to have an absolute preference for a given substitute of reserves. Now, what are the implications of our finding? First implication is for speedovers. Because banks are using several markets for their liquidity needs, it means that shocks on reserves are speedovers into several markets and including securities markets. The second set of implication is for policy. Our findings on bank preferences, i.e. here trade-off, can be used to predict monetary policy impact on market trading. And finally, we find that the banks that were most in need of excess reserves were able to get those excess reserves from the banks on the other side of the spectrum that were holding a lot of these reserves. So it means that the supply of reserves in the euro area has somewhat some elastic features. Or at least to be conservative, it's not perfectly enough. Let me dive more into the paper now. And we're presenting the data. In this paper, we use two proprietary databases maintained at the ECB. The first database is on bank-level balance sheet data. We do so for 241 banks in the sample with a representative coverage across the reductions in business models. And we take this data on a window before and after the shop. So from May 2019 to February 2020. So that's five months before five months after a window. And the second database that we use gives us bank-level reserves holding data for the sample. Let me talk now about the first step. The goal of this first step is to verify that the shop indeed led to a rebalance. And for that, we consider the sources of liquidity that are close substitutive reserves. Money market loans, i.e. short-term loans among banks of the financial system in general. Intra-group loans, i.e. loans among banks of the same group. And marketable securities. For example, euro area government bonds. So what we imply is that lending to the ECB or lending to a bank or to the treasury, these are somewhat substitute investments. And our identification is a difference in difference. Now it's a good time to talk about this identification. So and I'm going to do so with this figure. So this gray bar symbolizes the holdings of excess reserves of the bank or the heterogeneity if you want in our sample. Some banks will be at the bottom of this bar. They don't have much excess reserves. And some banks will be at the top. They have a lot of excess reserves. And here I've placed the tiering limit, just abitragil located. And our goal is to define what is the control group and what is the treated group. Let me start with the control group in red here. As a control group, we took the banks that are as close to the tiering limit as possible. So you want banks that after the shock have no incentive to increase their holdings of excess reserves and no incentive to decrease their holding of excess reserves. And those are the banks that are as close as possible to the tiering limit. For the treated group, you want banks that are far below the tiering limit. So they have incentives to increase their holdings so as to get the attractive rate of zero basis point introduced by the 2-tier system. And just let me tell you that one might have thought that the right control group adjusts the bank below the tiering limit. But it's actually not true. We find out that the banks with large holdings, after the shock, are banks that are going to decrease their holdings of an excess reserve. They are actually going to transmit this excess reserve to the treated group. And that means that if you use these banks as control group, you would mis-measure your treatment effect. For example, if the treated group increases by 2%, they are holding of excess reserve. It means that the large holding group are decreasing by 2% and you will have a treated effect of 4% instead of the actual treated effect of 2%. So that's why we went for this control group. All right. What's the empirical setup? On the left-hand side, you have bank eyes holding of a certain substitute of reserves at time chain. On the right-hand side, you have two dummies. The treated dummy, which is equal to one if the bank is part of the treated group, i.e. far below the tiering limit, and zero if it's part of the control group, i.e. near the tiering. You have the tight dummy that takes a value one if T is after tiering, zero all the way. The innovation here are the fixed effects. We put country time fixed effect. And why? Because the distribution of countries in the treated group somewhat differ from that of the control group. So if there is a shock at any time in one country and you don't put this fixed effect, the shock may affect more the treated than the control group. And you will capture that in your beta, in the treatment effect. And you don't want this. So that's why we put country time fixed effect. Now, here I'm going to present you a set of graphs that represent the main results in this first step. And the three graphs are all similar. So I'm going to spend more time in the first one and less in the second after. So what do we have here? On the x-axis, you have the time. And with the red bar placed here at the time of the shock. On the y-axis, in this graph, you have net money market lending scaled by total asset. What are net money market lending? This is loans in the money market minus deposits in the money market. And in particular, if you want to raise excess reserves, you would want to decrease your net money market lending. Either lend less or borrow more. Two groups in this graph are represented, the treated group in plane and the control group in dash. Ideally, what you want is the two groups to behave similarly before the shock. And that's what the graph tells you. On the left of the red bar, the two groups are somewhat similar. But as soon as the shock hits, you see the gap between the two groups widen. And in particular, we observe a decrease in net money market lending for the treated group, which is consistent with the desire to raise excess reserve holdings. We find the same thing on the intra-group lending market, i.e. the loans that a bank grants to the affiliate of the same group minus what it borrows from the group. And here, again, we see we observe the decrease in net intra-group lending for the treated. Finally, for securities holding, we find that the treated group decrease its holding of securities after the shock. All three results are consistent with a willingness to raise excess reserve. Now, these results are summarized in the table here with somewhat more details. What you can see is the treatment effect in the third column is negative, as we expected. i.e. the bank sold some substitutes of reserves in order to be able to increase its holding of excess reserve. There are two back-of-the-envelope calculations that we can do here. The first one is the total treatment effect in percentage of assets. If you do the sum of the free coefficient, you get 3.40% of assets. This is close to 2.80%. What is this 2.80%? This is the average increase in excess reserves in the treated group. And it turns out that those two figures, numbers, are statistically the same. What does it mean? It means by focusing on solely those three substitutes of reserve, we capture most or even all of the action. The second back-of-the-envelope calculation we can make is to do that in euro terms. So if you take this 3.40% and you multiply by the average bank size, you get 2.30 billion per bank. Multiply again by the number of treated banks and you get roughly 200 billion in our sample. This is the same magnitude as 227 billion, which is the total amount of unused allowances. What does it mean? It's a term that tells you that prior to the shock, if the banks that are below the tiering limit were to go exactly at the limit, it would need to increase excess reserve by 227 billion. Once again, it's close to roughly 200 billion we capturing here, most of the effect of the action. Now in the paper, we have more results about this rebalancing and maybe I'll have time to talk about them after the discussion. But let me now focus on the second step. What is this goal? The goal of a step is to study the rebalancing among the substitutes of reserves. And here again, we test two alternative hypotheses. The first one, the trade-off hypothesis, is a world where the rebalancing preserves the portfolio composition. Again, if prior to the shock, I had one-third of sovereign bonds and two-third of intragroup loans, I still have one-third, two-third after the shock. The second, the alternative hypothesis, the picking order, is simply that banks lack very much to use a given substitute of reserve and they use solely this one. What is our strategy here? Methodology is to compare actual rebalancing to counterfactual rebounds. Now it's a good place to explain it. Sorry, Jean-David, you have five more minutes, so... That's good. Thank you. This symbolizes the asset side of the bank. Bank up some reserves, some substitute X and some substitute Y of reserve. And observe that here in this example, they hold 50% of X and 50% of Y. We start with that prior to the shock. Then we observe by how much reserves the bank increase its... By how much the bank increase its holding of excess reserve. In this example, it's everything that is above the black line. Then we ask ourselves, in theory, how can the bank finance perform this increase in excess reserve? Well, one benchmark you may think of is actually to sell some of X and some of Y in the same proportion. And that's represented in this figure. And we call it the trader counterfactual. Why same proportion? It's just because the bank started with 50% of X and 50% of Y. If it had started with 1, 3, 2, 3, it would be 1, 3, 2, 3, 4. Then we go on and design another benchmark. Theoretically, the bank, what it could do is only use X and not touch Y. And this is the pecking order counterfactual. As a fourth step, we look at what actually happened to this bank. And in this example, the bank sold both X and Y a bit more X than Y. And we ask, in statistically speaking, what actually happened is it's equal to the trade-off benchmark where X and Y have been sold in the same proportion or the pecking order benchmark where X only has been used. Said differently, we test two sets of new hypothesis. The first set gives you the distance between the actual rebalancing and the trade-off rebalancing. And the second set gives you the distance between the actual and the pecking order rebalancing. And the new hypothesis tells you that the distance is new. And we do that via OTD, G-square test. Because we are, because this is a generalization of the T-test because we're working in a vector environment. Let me spend the remaining time on this table and then I conclude. This table gives you the results. The first column for the trade-off and second, third and fourth column for the pecking order. It gives you the distance between the actual rebalancing and the trade-off rebalancing. If the distance is close to zero, you will fail to reject the new hypothesis and the P value will be above the conventional level. If it's far from zero, you will reject the new hypothesis. What are the results here? If you look at the first column, the distance is close to zero and the P value is above the conventional level. This means that the rebalancing preserved the portfolio composition. If you, on the contrary, if you look at the second, third and fourth column, that simulate the pecking, the distance between the actual and the pecking order counterfactual, the distance is sometimes very far from zero and the P values are way below the conventional level so we can reject the pecking order. It's time for me to conclude. So in this paper, we ask how do banks manage their liquidity? And we do so by using a short balance, banks to rebalance free substitutes of reserves, money market loans, intra-group loans, marketable securities for that. And then we study this rebalancing. We find that banks rebalance proportionately to their pre-shock holdings which is consistent with a target liquidity structure. Conversely, they did not seem to exclusively rely on one source of liquidity. And these are important implications. First, shocks on reserves are speedovers into several markets, including securities market. Second, or findings on bank preferences can be used to predict the impact of monetary policy on market trading. And finally, the third implication is that the supply of reserves in the euro area has some elastic features. With a small pivot, that hearing was a cautiously calibrated policy. And so this third implication is valid only in this context. All right. Thank you very much. And I look forward to the discussion by Vaso. Thank you, Jean-David. Thanks a lot. So then, yeah, let's move right away to the discussion. And we have Vaso Janu from Bayes Business School at the University of London. Vaso, the floor is yours. Thank you for the opportunity to discuss this paper. So let me begin by giving a short overview of what the paper is about. So the paper is essentially asking the question, how do banks manage their liquidity? They started the different sources and uses of liquidity. So different markets on central bank reserves, interbank market, lending within affiliate groups, security holdings, and so on. So the key aspect of the paper is that they study how a shock to one of these markets spills over possibly to other markets. So the shock they are starting is a shock to the cost of holding central bank reserves. When the ECP has changed, it's the cost for banks of holding excess reserve. They exploit the two tier, what they call the tiering system, and that lends itself to a different analysis since the effect was heterogeneous across banks based on the holdings of central bank reserves they were holding prior to this change in October 2019. Now, why do we care about starting this? I mean, one, the obvious reason is the analysis is going to be informative about how monetary policy about monetary policy transmission. But it also is informative about, so the first one is important for central banks how they conduct their monetary policy. The second is understanding possibly the preferences banks may have about their liquidity. The author study effectively two types of possibilities, example possibilities, one which they call the trade-off theory is the idea that banks have a preference for a stable portfolio. They have various trade-offs, cost and benefits associated with a particular source of liquidity and they choose a certain ratio is optimal for them and therefore we will expect as the shock changes they are going to keep these weights constant. The other possibility is that they actually have a pecking order. So they prefer one source of versus another and they are going to exhaust when something becomes more available they exhaust any additional liquidity from that particular source. So the expect we will expect the shock in that case that will lead to changes in their relative weights. Here is what they find. So the first, I mean the key results are just to summarize is that the shock here is a decrease in the cost of central bank reserves. I won't go into the specifics exactly already John David explained that quite well but the adjustment we observe here is indeed that the banks did adjust their holdings of various other sources to raise their reserves. Now, one result is that the banks acted swiftly that any price effects that they saw they were only short-lived then they revert very quickly back to the steady state and sort of the supply of reserves looks fairly elastic in this context. In terms of where they, how did they raise the results? I mean, there are three things that they are checking them and what comes out is that the increase came through an increase in net borrowing mostly from borrowing from banks who were on the other side of the spectrum and they were not so much affected by the decrease in the cost of holding results. They also decreased lending to the net lending to affiliates and they also decreased their security holdings of government bonds. In particular, the effect seems to be stronger. I'm not quite sure why but on domestic government bonds. So when it comes to the question between taking off versus trade off the result seems to be supportive of the trade off rather than the packing or the hypothesis. So in summary, these are the results. So now let me give you very briefly my comments and thoughts which one, I mean, when thinking about the paper and thinking about the literature many of the papers in this space focus on one market of liquidity and this paper essentially it's actually studying several markets and speed lovers from one to another. The key result that comes from the paper about the fact that banks don't seem to have a preference in a banking order has implications in monetary policy and transmission in the sense that any particular shock in this context will lead to smaller price pressures than otherwise. They are in that market or others seems there is no strong preference for one or the other. So the facts on pricing are going to be smaller. It's also implies that the monetary policy is more predictable if banks have a, yeah, they have preferences more consistent with the trade off theory rather than the packing order. Many papers in this space focus on crisis period. So they look at the moment where there is a lot of risk and they study the facts and the contagion at that time. Now this paper informs about stock transmission in normal times. I mean, the offers don't use that. It's my choice of words. They don't say exactly normal times but what I would like to draw attention here and ask for more information on it. I mean, the context here is not a shock. It's not a crisis period. That is a period of negative interest rates. I mean, they define the initial state estate prior to the shock. There's also ample aggregate liquidity. So my question is how much of the results may be specific to the initial state estate and it will be helpful if the authors could expand a little the discussion on this to help the reader understand whether these are specific to this period or we are expecting them to give an existing theories and other evidence from other papers if they can guide the reader in that to think about this. Now in terms of the treated and control groups, I mean, as I mentioned, the treated groups and control are basically distinguished and Jean David already explained that quite well based on their reserve holdings prior to the shock relative to whatever the tiering allowance limit was. Now, if you want to look at it, I have a little graph here which sort of distinguish this. So you could think of the treated group as those who are way below the limit and those around the control are those around the limit above and below. The idea is that because those are very close to the allowance limit, they have little incentives to adjust while these ones are far away, incentives to adjust are much larger. Now, of course, one question I had when looking at this, I mean, one of the results of the paper and I think Jean David concluded with a reflection on that is that one of the results of the paper is that that in fact, it's swiftly and then the supply of reserves is elastic and my question is, is this a result or is it a design feature? So the limit was a choice. So monetary authorities have probably picked that limit very carefully to ensure that they're gonna be no undue shocks to the system. So I don't know if I can generalize that. Maybe the conclusion here isn't about elasticity per se, but maybe that that particular experiment or that particular intervention was effective and without effects, without long-run effects in the market. Now, the other comment I have, I mean, we can see from the analysis in the paper, the control banks tend to be different than the treatment banks, that's all right. They tend to be, the control group tend to be larger banks with smaller deposit base, more central bank reserves by definition and more security is issued and so on. I think it would be interesting to sort of see how much, and that goes back to my earlier question. It would be interesting to see how much of this differences on where they are on the space of reserves is stable over time. And so of course, benchmarking it to whatever is the available level of reserves at a given point in time. So you may be doing an analysis and that could be potentially useful to see how much of this analysis here is specific to this period with a negative interest rates or an ample liquid in the market or there are some stable patterns in terms of who is present in the available space of reserves at each point in time. Now, a comment. Just one minute left, sorry. Yeah, I'm closing, no problem. So one of the key questions, one of the key results of the paper, so that was about the elasticity comment and the negative interest rates. But now one of the comments is, one of the key results is that banks appear to have a preference for a stable structure of liquid assets and in turn to that structure after the shock, I mean, there is a trade off, cost and benefits and that makes optimally certain ratio for them. It will be interesting, I think, if the authors could expand some on the paper to provide some information about which structure is optimal for whom and how that varies with different characteristics. Now that can be in the form of discussion and guidance on corporate finance theories where similar ideas are pressed and it can also be incorporated with an analysis in the paper. I wasn't quite sure how exactly to bring it in the current system. One possibility which I was looking at is to try to have heterogeneous treatment effects and exploit the cross-section of the time series in this case, given that the experiment per se doesn't let itself naturally for different periods. Overall, I enjoyed reading this paper. This is very nice and interesting paper well clearly written, rich with many interesting results and thank you for giving me the opportunity. Thanks, bye-bye. Thank you, Vasa, for the discussion. Thanks a lot. So I think we have, well, actually we have exhausted our time but maybe let's have five more minutes on general discussion. So I'm just looking at whether there's anything otherwise I would hand over to Jean-David to maybe respond. I don't see anything in the chat. Maybe just one question from my side. I mean, maybe related also a bit of what Vasa was saying about the heterogeneity. I was just wondering whether you looked into the question of heterogeneity across banks and to what extent, for example, the stable portfolio result varies, for example, across different banking types. Perfect. Anyway, but maybe you can just take on the question. Thank you. Thank you, Vasa and Tobias. So let me start with the heterogeneity. Let me, actually, thank you, Vasa, very much for this. You touched on the very, very good points and I'm not going to be able to address all of this but let me start with the heterogeneity. It's a great, it's a great suggestion. We haven't done so for the Peking order trade-off results and that's something that we could do with a small limit that, you know, you don't want to, you cannot refine so much this analysis because you need a critical sample size but otherwise that's something that we can do. For the rebalancing, we did explore the heterogeneity and we didn't see anything that was major. So the heterogeneity and the rebalancing didn't seem to be so large and did not warrant us to mention that. Let me talk about the steady state, the fact that maybe some of the results are dependent on the steady state and I would say that most of them are. I mean, it's like any empirical paper you really bounded by your sample time. So what we could do is we would take your suggestion to see if what we find can be also found in the literature. I would like to add that the literature on bank liquidity management is not so large. So let's see what we find there. For the fact that T-Ring was carefully designed, I think that let me push back a little bit here and I think it's actually a positive feature in the sense that if BCB didn't do a good job there, the rebalancing we would have observed would may have been very different from the rebalancing that banks wanted to implement. I.e. we would probably not have been able to observe the true bank preferences. But because T-Ring was well calibrated, there was not too much stress on the banking system and therefore we were able to observe the true preferences of banks. And finally, on your comment about the characteristics of the stable overtime, that's definitely something we can look into. All right, thank you very much. Thanks a lot, Jean-David, for a very interesting presentation and of course also thanks a lot to you, Vaso, for your discussion.