 Let us now turn to the second paper for this session which is on strategic complementarities in the payment system in the area of ample reserve and is presented by Gara Afonso. Gara has a long career at the Federal Reserve of New York and is an assistant vice president in the financials intermediation function of research and statistics group. The discussant of that paper will be Christine Pader, professor at University of California in Berkeley. So I'm happy to hand over the floor now to Gara for her 30-minute presentation. Please the floor is yours. Thank you. Good morning and good afternoon to some of you. We would like to thank the organizers for including our paper in the program. It has been a great conference. We would also like to thank our discussant Christine for the comments to come but also for being the first to read a draft of this paper and to everyone in the audience for waiting till the last paper in the last session and holding on with us today. I'm presenting Interbank Payment Timing still closely coupled. This is joint work with Dal Duffy and Lorenzo Perrigan at Stanford University and Hume Sheen at the BIS and they are all joining us today in the audience. Before I get to the paper, let me just say that these are my views and they don't necessarily reflect those of the New Year Fed, the Federal Reserve System or the Bank for International settlements. Prior to the global financial crisis, the balances that banks held in account at a central bank in many jurisdictions were low relative to the amount of payments that they made. They had to rely on incoming funds to make their payments. This reliance on incoming funds is going to create a high velocity of money and a strategic complementarity by which a bank's willingness to send out payments is going to depend and is going to be higher on other banks' willingness to send payments to. This complementarity is also going to raise some vulnerabilities and opens the possibility of gridlock scenarios and disruptions. And the financial crisis reserves and central banks have expanded their balance sheets to provide support to the economy and to provide accommodation to financial markets. As a result, we see that reserves that these banks hold in the account at a central bank have increased sharply. In recent years, many jurisdictions have re-evaluated the frameworks that they use to implement their policies, and they have concluded that they want to continue to operate in environments with elevator reserves. So the question we want to ask in this paper, if this is strategically complementarity in payments that we've been documented in the past in environments with lower reserves, if they still exist in an environment of elevator reserves? So to answer the question, we're going to have a closer look at the US case. We're going to use data from a large value payment system for the wire. And we find that it's a strong and positive relationship between the payments that a bank receives and the payments that it makes. This relationship persists in environments of elevator reserves. So why would we care? Well, the existence of complementarity opens the door to these grid logs in payments system and potential disruptions. And it also poses a vulnerability to financial systems. This paper is also helped us to think more broadly about what elevator reserves really mean. So in today's presentation, I'm going to first show you the data that we use. I'll introduce our wear and purify framework, show you some results, discuss robustness, and hopefully wrap up in time. Let me get to the data. In the paper, we use two data sets with confidential information, one on payments and one on reserve balances. Let's start with payments. We have information, the transaction level of all the payments sent through Fedwire Fund Services. Fedwire is a real-time gross settlement system in the US where banks with the system process every transaction that the banks send. So payments cannot be net out. It was created in 1918 and is owned and operated by the Federal Reserve. So many participants more than 5,000 have accounts and that includes banks, some businesses, and government agencies. The payment system in the time of our analysis is going to be open for 21.5 hours. So we're going to open at 9 p.m. on a preceding calendar there. It's going to be opened overnight through the day until 6.30 p.m. On an average day in 2020, this payment system process around 700,000 transactions and a value of 3.3 trillion. In the paper, we're first going to focus in 2020. In particular, we're going to look at the first 100 days on 2020. And then we're going to take a broader view and look at a 10-year period from 2010 to 2020. Our data set includes every transaction. For an analysis, we're going to aggregate transactions up to the minute level. So we're going to add up all the payments that the bank makes to all its contemporaries in a single minute. We keep talking about banks. We're going to focus only on banks, which means we're going to exclude some special accounts, including, for example, the account that Treasury has at the Federal Reserve, the Treasury General Account, or the payment systems like CHEAPs, accounts like ACH and ZLS. I still have a very large number of participants. So for computational simplicity, we're going to focus on the top 100 of those. We're going to add a top institution as measured by the daily average dollar value of payments that they process. These 100 top entities manage around 90% of the total body. For the analysis, we're going to look at the top 15 entities. And those are responsible for roughly three quarters of the body. The second data set that we're going to use is reserve balances on the positive institutions. So we use an internal Federal Reserve accounting records that includes the balance at the end of the day for every depositor institution in the US. In our analysis, we're interested in opening balances. So we're going to approximate the opening balance by the day before closing balance. Going back to the top 15 large institutions by payments, those institutions hold roughly 40% of all reserves in the system. So in this paper, we're interested in understanding if there exists a strategic complementarities in payments. And if this relationship persists in environments of elevator reserves. So to do that, we're going to look very closely at the relationship between the cumulative receipts of a bank and the payments that it makes. Before I show you the empirical analysis, I would like to share with you some features of the data that are going to influence our model of decisions. In this slide, we show the distribution of payments on the left and that of receipts on the right. We aggregate payments to the minute level. So we add up all the payments that a bank made on a minute and we add them up to take a measure of the payments per minute. And that's what we show on the left panel. So what we see here is that the distribution of payments is very skewed. There are minutes when banks made as much as 40 billion in payments, but they're important to you. There are many minutes when banks do not send any payments. And these zero payments are going to play an important role in our analysis. So we have to take them into consideration as we design our strategy. Let me show you how the relationship between payments and their cumulative receipts looked. In our 2020 sample of 100 days, we have over 2 million observations. So scatterplots becomes very difficult to interpret. So we're going to show you a hit map to give you a sense of what the different mass is. Where a light color indicates a high density. So in here, we can see when we plot the relationship between the payments that a bank makes in a minute and the receipts over the past 15 minutes, we see that there's a lot of minutes when banks spend no payment. And this is a feature that we just described. We also saw that in the histograms before. We would love to also understand how this relationship is when banks do pay and do make payments. And that's what we have on the right panel. So again, exclude for a moment the minute where banks don't make any payments and show you the relationship between the payments a bank makes in a minute and its cumulative receipts. We now see that a linear relationship is emerging. So when we designed our empirical approach, we will not be able or we will not use a linear model. We would like to have a flexibility of a model that allows us to capture both the minutes when banks do not send payments, but also the linear relationship that emerges when banks do send payments. So instead of an OLS, in our analysis, we're going to estimate the topic regression. So where are these zeros coming from? So early on, I mentioned that the payment system in the U.S., but why it opens at 9 p.m. of the preceding calendar day. So this is what we see here. I'm going to show you the intraday timing of payment on an average day in 2020. We're going to aggregate all the payments for all the banks. I'm plotting in this case by every 10 minutes just to make a bit more smooth the shape of payment by the days to see how they evolve throughout the day. When fire opens, there are some payments that get sent. There's very low activity overnight. Payments increase in the morning. And then there's the relevant characteristics of payment systems like the one in the U.S., which is a high concentration of payment just before the market closes. So during the night, there's a lot of minutes we find do not send payments. Let me show you our model. We're interested in capturing the simplest relationship that can give us the relationship between how a bank pays based on how much it receives. But we're going to look at the payments that I make on minute m of the E.T. As a function of the cumulative receipts over the previous 15-minute window on the E.T. All our specifications are going to include bank fix effects, date fix effects, and we also include a series of period-of-the-day fix effects. The idea of this period-of-the-day fix effect is to capture some of the dynamics of intraday payments that we discussed earlier. So in particular, we're going to have an opening indicator that is going to capture activity in the payment system for the first half hour to take into account that payments can be queued ahead of the market opening and they get processed when the market opens at 9 p.m. We also want to add an indicator variable that captures the activity that happens overnight because very few banks send payments overnight. The next set of period-of-the-day fix effects are intended to capture the concentration of payments that we see in the afternoon. So we're going to include a series of demivirals every half hour from 2 p.m. all the way to 6 p.m. And finally, the last one is intended to capture the fact that banks cannot process payments on behalf of the clients in the last 30 minutes of the day. We're going to cluster our standard errors at the bank level. Let's see what we find. So Column 1 shows our main results. It shows the relationship between how much a bank paid in a given minute and how much it has received in the previous 15 minutes. Our point estimate is 0.61. It's possible it's a different from zero. Since we estimated the Tobin model, we cannot directly interpret this coefficient as the marginal effect on receipts. We need to adjust it by a factor. So after adjustment, the value is 0.421. What does it mean? It means that a 1% increase in the cumulative receipts that the bank gets over the past 15 minutes translates into a 0.4% increase in the payments that that bank sends out over the next minute. This documents the system of this complementarity and reliance of banks in incoming funds to make out the payments. Another way to capture this complementarity is by taking into account that on days when banks have opening balances are very high when the payment system opens and banks have very high balances, we expect this relationship to be less strong. To look at that, we're going to add to our baseline model in Column 3. We're going to look at days when banks have very high balances in aggregate. This is the day when the payment system opens and these banks have really high balances. By very high, we mean the top-decile of the days. We see that on those days, the strength of this relationship between how much bank pay receive and how much they pay is weaker. So on days when balances are unusually high, banks are synchronizing the payments less. In the paper, we consider this use of robustness to analyze some possible dimensions that can influence our results too. First, we look at the role of bank balances. We have just discussed the role of aggregate reserves in the system. But we could think that as the bank balance increases, the bank will rely more on those balances and less on incoming funds to make those payments. We wanted to include balances in our analysis to see if our relationship vanishes as we include the bank's opening balance on that day. So this is what we have in specification 3. We see that there's no effect of including the balances. And the idea behind this is that bank balances don't change significantly from day to day. And our baseless specification already includes bank fix effects. So we already take into account the balance that the bank have during this period. Next, we would like to ask ourselves if our results are driven by a spurious relationship that could be driven, for example, for days with very high payment volume or very low payment volume. So maybe it's not a strategic complementarity, it's just driving by days with high activity and low activity in the payment system. So again, our results in our specification includes date fix effects. So this date fix effect should capture an observable factors that are constant across banks and minutes, but that change from day to day. So that will take into account days where payments are very high or very low. We also want to control explicitly for the past payments, and that's what we do in specifications 4 and 5. So specification 4, we're going to include all the payments that the bank has made up to 15 minutes ago before it makes a decision to send payments out. And in specification 5, we look at the payments in the previous minute. We see that the magnitude of our coefficient of interest has declined, but there's still a positive and significant relationship between how much a bank pays out and how mighty it receives, even after controlling explicitly for the payments that the bank has made that day. Another series of analysis that we look is at the payment window. So in our framework, we're looking at how much a bank pays out in the next minute and how might it receive in the previous minute. We put these windows to be consistent with the seminal work by McAndrews and Porter that document the existence of this complementarity in environments with scarce reserves. In the paper, we concede the other windows. Our results do not change qualitatively when we alter the window. So for example, in the column 6, instead of looking at how much you receive over the past 15 minutes, we consider a 30-minute window. And in column 7, instead of looking at the payments you do in the next minute, we look at five minutes ahead. So just to recap, we have found a strategic complementarities in payments in 2020 in the United States where reserve balances were over 2.5 trillion. We see that on days when banks opened the day with very high balances, the strength of this relationship is weaker. So on days with very high balances, banks synchronize their payments less. Next, we would like to understand if the timing of payments intraday is also different on those days. To do that, I'm going to look at the share of payment that banks process at different times of the day. So we start by looking at the business day and we're going to divide it in small pages of time. In the graph C in the presentation, we use 20-minute periods. But we also try 15 and 30, so it doesn't really affect results. In each period, I want to see if the share of receipts is different on days with high balances or days with low balances. So we estimate this relationship and instead of showing you 65 tables, we're going to try to summarize all that information in the chart. The chart on the left shows the results of estimating the offset of the expression. The one on the right is showing us the slope of this relationship between the share of receipts and the balances. A blue symbol indicates that the coefficient is positive and is statistically different from zero. Red corresponds to negative and is statistically significant from zero. And our gray symbols are values where the coefficient was not statistically different from zero. So what we see in the right panel is the share of payments that get sent and received during a business day. This is similar to the pattern that we have explored already for intraday payments. The more interesting is the one on the right. This is going to tell us, this is this coefficient here, is going to tell us if the share of payments is different when balances are high. And what we see is that on days with high balances we find a positive coefficient. So the share of payments sent in the morning is higher. In the afternoon we find a negative coefficient. So the share of payments is lower. These are compliments. So the idea is that on days when banks have very high balances the share of payments that is received early in the day is higher. So if your question is, is this just a 2020 effect which was already a very unusual year or this is something that we see in environments of elevator reserves more broadly. So to answer this question we're going to take a step back and take a period of 10 years from 2010 to 2020. I'm going to look at the two sets of analysis that we do in this work. First we look at the existence of this complementarity. So we're going to look at the relationship between the payments the bank receives and how much it pays out. And then second we're going to move into the timing of payments on the day and see if it's different on days when balances are usually high. So let me start with the complementarity. So we're going to take 10 years of data and we're going to run our baseline analysis the one that looks at the relationship between incoming payments and outgoing payments at the quarterly level. And we're going to show the estimated coefficient by a blue line the solid blue line in this graph. The red one indicates reserve balances that these top large banks hold at the time. What we see here is that as the Fed expanded the balance sheet and reserve balances increased the strength of the synchronization of this complementarity is weaker. As the Fed normalizes the balance sheet the strength regained and it became stronger. In terms of the timing of payments as before we see the one of interest is the panel on the right we see that on days when banks have a high balance at the beginning of the business day the share of payments that they process early in the day is higher. So let me drop up. So in this paper we study the relationship between the payments that a bank receives and the payment it makes. We find that the 1% increase in the cumulative receipts over the past 15 minutes translate into a 0.4% increase in the payment that that bank makes over the next minute. We find these results in environment of elevator reserves. We see that on days when balances are unusually high this relationship is weaker. Suggesting that banks synchronize their payments less. Before they look at the timing of payments intraday and see that on those days with very high balances banks receive a higher share of the payments early in the day. Why is this existence of a strategic complementarity so important? Where complementarity is open the door to potential gridlocks it also amplifies the effect of operational events and cyber events. Linking to the previous paper complementarity are amplified by market fragmentation. So we need to take into account this complementarity in payments when we think about the design and implementation of digital monies such as central bank digital currency of stable coins. And now to link back to the beginning of the conference today. Since the global financial crisis we've seen that reserve balances have increased startling. But also the uses and roles that these reserves do play. In the US reserves are used as part of the liquidity regulation and use them for their own risk management practices. They are important for supervisory internal stress testing that are part of living wills. As we show today they also play a role in the intraday management of payments. So these works help us think more broadly about what elevated reserves really mean. Thank you. Thank you very much Gara for this interesting presentation and let me hand over immediately to Christine for her comments as a discussant. So thank you very much for asking me to discuss this paper. I know everyone always says that but I really mean it in this case. I have a theory paper that as an assumption takes is given the fact that banks payment flows affects their real activities. So somebody who does mostly applied theory just to find out that it's actually true. The assumption works. It was really exciting for me. So what is the experiment that they do? They take just what seems to me to be just a tremendous amount of data and they break the day into one minute intervals and they relate the dollar value in a particular minute to cumulative receipts over the previous 15 minutes. And so this just captures to a first order how sensitive banks actions are and banks collateral management is to inflows. And the numbers I think are actually pretty big. So a 1% increase in receipts goes to a 0.4% increase, right? So I mean something is there. And I guess always the big question is why do we care? Why is this interesting? Well, first of all, fundamental to capitalism is the smooth functioning of the payment system. And so anything that affects the payment system is obviously something that we want to take very seriously. And really, I mean, one of the reasons why we regulate banks is because of their role in the payment system. If they weren't in the payment system, we would, we might not regulate them in the same way. And so things like strategic hoarding of liquidity, which is another way of thinking about this complementarity, could have systemic implications. And so this is something that we really should be worried about. And after the global financial crisis, there were these sort of loose arguments based on large balance sheets. Oh, nothing really matters anymore. They're just like, oh, they have all this cash sitting around. And so the argument was that there was no more strategic complementarity. But there is, right? And so the question is, how do we want to think about it? And what sort of value what I would suggest or what I would encourage the authors to do just to sort of push the paper and push the agenda forward. So the complementarity is there, but I guess we kind of want to know why. We sort of want to know what the banks are doing. And one of the sort of pivotal graphs that Gareth showed towards the end was the complementarity is between reserve balances. So the idea is that if banks have a lot of money overnight, then essentially they're going to be less sensitive to payments and payment flows. I would encourage the authors to distinguish carefully between daylight balances and central bank reserves or overnight balances. So I'd be very interesting to see some sort of document like or some sort of documentation like this that distinguishes between daylight balances and reserves. Okay. So how would they do that? What would one be one way of sort of seeing the sort of essentially collateral management issue that happens over the daylight hours or during the payment system opening hours? So suggestion number one, the Fedwire is certainly not the only interbank payment system in the U.S. And the other systems, so for example the ACH, which distinguishes between push and pull and they're batched, they, those transactions happen at very, very specific times of the day. So I would encourage thinking a little bit more about the dummy variables for time of day effects and trigger them exactly to what's going to the opening and closing of say the ACH, right? And if collateral is costly and banks are doing some, have sort of divided their balance sheet into a little bit that they're doing their trading stuff and a little bit that they're doing into their payment system, then the ACH settlement and above average ACH settlement, so for example, should tighten constraints. Even if you don't have data on what those flows are and the push and the pull, you would still know that those times of day should be slightly different. The other thing that I would suggest the authors could do, I mean I certainly couldn't do it but I'm sure the authors collectively can do it, is we know that Fedwire is not free. They're sort of marginal transfer costs but there's also overdraw fees and collateral limits. So a couple of things. They're sort of a boundary, a collateral boundary that the banks have to respect before they start getting into trouble. They either have to pay overdraw fees or they get their fingers slapped. So it would be interesting to using the data that you have if you could relate banks' payment flows also to a variable that essentially is their distance from this boundary. Right, you have all the master accounts and I think you should be able to do it. It's horribly complex but it should be possible. The other thing that's sort of interesting about these collateral requirements and when you hit sort of overdraw problems is these costs have changed over time and so that might give you some sort of interesting heterogeneity in your very long time series. I don't know if you can see this. It might be too small. What this is is a snapshot from the Fedwire account management system. So all the people who hold master accounts basically have this updated on their computers and what does it show? It basically shows opening balances. It shows essentially pre-funded ACH and basically also if you can see this there's a section that looks at daylight overdraft balance. So it gives you the collateralized capacity, debit caps and so forth. So these, as I understand it, this information is updated in 15-minute increments and given that this is part of the suite of services that Fedwire gives to its large clients who have master's accounts, this, you might be able to get this data in which case this would add a huge amount of richness and would really, really speak to how banks are using their collateral. So in summary, I think understanding the payment system is extremely important. It's very, very obviously very, very difficult to get data on the sort of retail use of payments unless it's some sort of unusual situation. But certainly how these wholesale payment systems work is very, very important to understand. And I think what we want to know is whether or not banks in their role in this RTGS systems is it inefficient? Does their dual role as banks who provide loans and do other things and trading, is it sort of inefficient because it forces them to conserve collateral in a specific way? I would love to know the answer to this question. So these are very interesting data and thank you for writing the paper. Thank you very much, Christine, for your very valuable comments. When handing over now back to Gaara for possible response, maybe you could also kindly address a question which Jean-Charles Rocher has asked in the chat, i.e., can you distinguish the payments initiated by banks from the ones initiated by their clients? So for example, if they instruct selling securities, I mean, whether such transactions can be differentiated. Thank you. And the floor is yours for the final word. Thank you. I'll answer Jean-Charles' question easier. Yes, it can be. There's a field in the payments that indicates if the bank is a transaction is on behalf of a customer or is a transaction that is done on behalf of the bank. So this is something that we could use. I agree. And then back to Christine, thank you so much for the discussion. I agree with your comments. The idea of using... So in the paper at this time, we only focus on the payments by banks and we are dropping out all that richness of all, you know. Sources of other shocks that we could explore much better in our analysis is in the to-do list. I thought about starting with chips as another large payment system by private only probably owned and see how that affects that relationship. But ACH is another one on the to-do output. It's a great, I mean, it's a great... It's our next avenue to try to respond to richness. And then in terms of the second comment, I have not thought about that. It is really interesting. In some way, overdrafts are really not happening. So the dollar value of overdrafts have decreased significantly. It became cheaper in that is collateralized, although there's the continuity cost of how you do that. But this takes us back to the question of what elevated reserves really mean. I think banks will work thinking on buffers more than thinking of really overdrafts. So reserve balances have reached a level that they're going to breach their internal buffers before they get an overdraft. And that goes to your comment too. So it would be very interesting to try to assess how their behavior gets changed as they get closer to those buffers, which is a bit more challenging than looking at the balance and saying, you know, you're going to zero. I can see you're about to reach overdraft because that's something that we can quantify more easily. But I completely agree. It's a very valuable line of work. And that's where first we're going to try to explore the variation through the payments that we don't include, and then we'll move into the second one. Thank you again, Gara, for the very interesting presentation and comments and for also Christine for her comments. Now we have reached not only the end of this session, but we have actually reached the end of the conference. I would therefore very much like to hand over to Katrin Assenmacher, who will wrap up the conference now. Katrin, over to you. Yeah, thank you, Hamut. And it's my pleasure and a great honor for me to conclude this conference. Today and yesterday, we were able to listen to a lot of interesting contributions. And we had a selection of teams that are particularly relevant for the implementation of monetary policy, namely the structure and the functioning of money markets, management of central bank reserves, the working of the payment system, and recent trend in financial innovations, namely CBDC and decentralized finance. Overall, it was a highly topical program, and I thank the money market conference organizers very much, that's Maria and Cio, Peter Hoffmann, Daniel Kieden, and Sebastian Weber for putting this program together. I know that this has not been an easy task, as there have been many high quality paper submissions to the conference, and they had to have a tough choice. I would also like to thank the discussants, for their insightful and constructive comments that gave the audience, and surely also the authors of the paper, new perspectives on these presentations. Let me look back to the main takeaways from the conference. So first, there were two presentations by Marie Herova and Leili, who dealt with the severe financial stress in spring 2020 that affected money market and non-bank actors. This work adds to discussions about the effectiveness of the existing monetary policy framework, and how well it alleviates the liquidity crisis in the financial system, in which non-banks, such as mutual funds in the Euro area or prime MMFs in the U.S., play an increasingly important role. To other papers, study potential sources of instability and funding markets. Jay Kahn examined increased involvement of hedge funds and basis trade of treasure fruiting using evidence from March 2020, when liquidity in the treasury market led to large sales of basis trade positions, which exacerbated stress in the treasury market. On a related topic, Dimitri Chebotarov argued that conservative CCP repo hackers have been another source of instability, since they may induce negative selection and credit spreads in CCP record market. So this was kind of the private sector part, turning to the side of central banks. Another focus of the conference was on modeling and analyzing central bank reserves. So Gabriele Laspada presented the modeling framework to monitor the market of reserves in real time, and to check the very important question whether monetary policy operates in the region that's characterized by scarcity of central bank reserves or not. Jean-David Sego studied banks' access reserve management in the Euro area, relying on the introduction of the ECB Q2 system that acted as a shock to the value of liquidity holding. So this finding supports a trade-off view of liquidity management, according to which banks prefer a stable structure of liquid asset. All these topics were also taken up in the market participants panel, which offered an additional insight from a practitioner's perspective. Overall, in addition, innovations and digital payments, challenges they bring and the effectiveness of their regulation were also on the program. There was a model by Yelib analyzing the optimal stablecoin regulations, leading with the risks that arise when stablecoins issues transform risky assets into digital hookers of stable value. Such models provide a framework that can be used to evaluate regulatory proposals and shows that capital requirements are better in the welfare sense than regulation to limit price and volatility. And now Gower focusing on traditional payment systems analyzed the strategic complementarity of in-going and outgoing payments, finding that strategic complementarity still matters despite the existence of ample reserves. I also very much enjoyed the keynote address by Hao Jiangzhu on CBDC design, which showed how the options like remuneration and payment convenience interact with the structure of the banking system. Finally, such a conference would not have been possible without people acting behind the scenes and making everything work. And this work is probably even less visible in the virtual than in an onsite event, but it's no less cumbersome or demanding. And here I would like to express my thanks to Britta Bertram, who has this year managed most of the administrative work, which is indispensable to make such a conference a success. So thanks a lot, Britta, for taking care of everything that was needed to make this work and for all your preparations that were really very extensive and very profound. With that, I wish you all a nice evening and I hope to see you again at next year's event, which will hopefully take place in Frankfurt at ECB. So have a good evening and a nice afternoon. Goodbye to all.