 Welcome everybody to the second day of our money market conference. And let's jump right away into that third session, which is on the demand for central bank reserves. And just maybe to remind you of a few logistics. So we have one and a half hours for this session, 45 minutes for each paper. So the presenter has 30 minutes and the discussion 10 minutes and then we have a we have a quick round for also questions from from the whole audience. The questions that you that you want to ask, please, please put them in the in the chat and direct them to all panelists. I will put that information in the chat as well. And then, just to say we have to be a little bit disciplined with the time because the market panel after us has to start on time at 335. Yeah, we have to we have to finish on time. So I don't know if you're okay I would maybe give you a quick signal five minutes before you before the end of your time of your presentation. And then, then we should be fine. So, so yeah let's let's jump right away into the first paper. The, the first paper is is on scars abundant or ample a time varying model of the reserve demand curve. And yeah, Gabela you're already visible so this is presented by Gabela last father from the from the New York Fed so welcome to you and over to you for the for the presentation. Yes, so let me first make sure that I can share my paper. By thanking of course the conference organizers for including our paper in the program. It is a great pleasure to be here to present it here in front of this audience. So President John Williams and our former colleague Domenico Giannone was now at us. The usual this camera applies is our own personal views that do not necessarily reflect the views of the New York Fed or the federal system at large, or the FOMC. The title suggests the paper is about banks and specifically what we are after is the reserve demand curve, which is the price of which banks are willing to borrow and lend reserves with each other as a function of aggregate reserves. So, as all of you know, this is a key monetary policy object for two reasons. The first reason is that when the FOMC communicates his stance on monetary policy, it does so by specifying a range for the rates of which banks. And the second reason is that the Fed can indeed change the aggregate level of reserves in the system, but other factors outside the Fed's control can do that too. So, from the perspective of the monetary authority, it becomes of paramount importance to get a sense of fully in real time of where you are in this curve and even more importantly of its flow. Because that tells you by how much Fed funds rates so that rates of which banks lend and borrow reserves are going to respond to reserve shocks. So the natural question is, can we have a well-identified high frequency, and here I mean daily frequency, estimate of this demand curve and its flow? And that's exactly what we do in this paper. So we propose a time-varying structural estimation of the daily frequency of the reserve demand curve over a period of 10 years going from 2010 until March 2021. So encompassing basically the old post-crisis period. Obviously, there are two challenges with such exercise. The first one is quite obvious. There are the typical endogenated issues of any demand. And here I'm thinking about disentangling supply versus demand shocks, omitted variable bias, any underlying compound factor. But there is a challenge that is more peculiar to what we're doing here, which has to do with the fact that there have been many structural changes in the market for reserves after the great financial crisis. And as a result, the curve may have moved over time significantly. So how do we address these challenges? Well, we use a time-varying structural model on the editor, and you can think of our model as having two components. So first, we propose a flexible forecasting model of the joint dynamics of prices and quantities, so the rates and the aggregate reserves. Then the second component of this strategy, we use the forecast errors for the path of reserves over time coming from the forecasting model as an instrument, for our instrumental variable estimation of the structural reserve of the market. And we use the path forecast error. Now let me anticipate right away what would be upfront daily data for our identification strategy. It's very important. Let me show you, let me give you a preview of our results right away. So here at the top, in the top figure, I'm just showing you the data. That's the time series of our main variables of increases of prices. So rates of which reserves are length and point aggregate reserves. So on the x-axis, you have time from 2010 until March 2021. The solid line is aggregate reserves in the US banking system, normalised by banks that allow to account for the growth of the banking industry over time, which is quite sizable since we're looking at more than 10 years of time. Now the dashed line instead is the weighted average rates at which reserves are borrowed in the market for reserves, minus the interest on reserve balance to control for changes in the monetary policy system. And I would be clear in a few slides about why we need to do that. The graph is color coded. So we start with dark blue from the left when it's around 2010. Then it gets lighter, lighter blue, 2011, 2012. It becomes gray for the period that goes from 2014 to 2017. And then we have pink for 2018 and 2019 and dark red for the last part of ourselves, 2020 and 2021. But what I want to focus on, of course, is the chart at the bottom. Because in that chart, we are comparing our model for the reserve market with the data. So the dots are our model-implied estimates of the reserve market functions. Those are five days ahead in sample joint workers of prices and quantities. So on the x-axis we have normalised reserves and on the y-axis we have prices. So that's rates, minus IRP. And the squares here represent the realisation of the data. So this figure is telling us the thing. The first thing is that the model is the data very well. As you can see from the part that all the dots sits on nicely on the square. The second thing is that the model is able to predict, to forecast the non-linear demand function predicted by the theory. And that would be more clear later about what the theory tells us about the reserve market curve. Here you can see if we start in 2010 with the dark blue area, there is a clearly negative slope that it decreases. Then it decreases in absolute value as reserves expense and the curve becomes flat as normalised reserves process 12-13% of banks not elastic. Now the third and most important thing that this chart is telling us is that there is not just one curve over time. There are three curves. So the reserve demand curve seems to have shifted over 2010-2019 outward and then further upward after the onset of the COVID pandemic in March 2020 as represented by the dark dots. So what are the results? The curve is shifted upward over time. It has shifted further upward at the onset of the COVID-19 outbreak. The curve was flat during the period of 2012-2017 and after March 2020 periods that people referred to as abundant, or super-abundant reserves. However, although reserves in the system in our period were almost always above a trillion and most of them above two trillions, we still observe a negative slope at the beginning of our sample in 2010-2011 and in 2018-2019. Now this slope that we observe is more gentle than the steep slope that people estimated in the 90s they referred to as the periods of scarce reserves but it's still there. Finally, the transition, the level of reserves or normalized reserves here at which the curve stops being flat and starts displaying a negative, a significantly negative slope, is around 12% of band-stalked acid, both at the beginning of our sample and in 2018-2019. So here's how the talk will go. I will spend a reasonable amount of time on the institutional background because it turns out to be key for the identification of the reserve demand curve. Then I will talk about model identification strategy. I will skip the part in which we convince you that our model has very good real-time performance in the sense of our sample forecast evaluation because they want it done. And then I will focus on empirical extinguishing on our maps. So let me start with aggregate reserves and their evolution over time. So quantity. So reserves are deposits held by banks at the head. And I think most of you have this figure in mind already but I think it's worth it to spend some time on it because it's a very important figure. This is the time series of aggregate reserves in the US banking system from 2005 until March 2021. And this figure is telling us one important message. There are two worlds, two regimes when we think about aggregate reserves. There is a pre-2008 financial crisis and there is a post-2008 financial crisis. So before 2008, reserves were in the tens of billions and they were quite stable. After 2008, as the Fed responded to the great financial crisis, reserves jumped to the trillions and they exhibited a much richer dynamics over time. So that's the time series of reserves. But how do reserves change actually in accounting? And here I need a bit of accounting and this accounting will turn out to be important. So reserves are assets from the perspective of banks or the positive institutions but they are liabilities for the Fed. So there are two ways reserves can change. The first way is quite intuitive. It's through expansions and contractions of the Fed's balance sheet. So for example, when the Fed purchases securities in the markets, it usually purchases the security from banks and when it does so, it credits their reserve balance of the Fed. There is so there is a one-to-one mapping between the expansion of the balance sheet and the expansion of aggregate reserves of the system. But there is another way the reserves can change. It has to do with the fact that reserves are not the only liability from the Fed's balance sheet. There are other liabilities, for example, the account of the U.S. Treasury, the Treasury General Account. And when this normal reserve liability is decreased, holding the size of the balance sheet constant reserves might decrease. So what are these normal reserve liability telling us? The reserves are not at those systems and that's because on a daily basis banks transact with holders of no reserve at liabilities. And I want to focus on two important examples here. One is the Treasury General Account. The account of U.S. Treasury has a defect. So when then, why is newly issued Treasuries from the Treasury? They do so. They pay for the security by using their own reserves. So there is an increase in the Treasury General Account and a decrease of aggregate reserves of the system when banks submit the payment. And the similar dynamics occurs for tax payments. When banks submit tax payments to the Treasury, it won't be up to the clients. The other account, the other non-reserved liabilities that we like to focus on is the overnight reserve of the city, simply overnight reserve, which is used by money market funds to place cash at the Fed to reverse repos color-utilized by Treasuries. But when money funds place cash at the Fed, they are instructing their custodian banks to make the transfer. And the custodian banks are using their own reserve balance. So for an increase in the overnight hierarchy, there must be a decrease in aggregate reserves in the system. So if no reserve liability is very small, a negligible part of the tax balance sheet or stable over time, intuitively, it wouldn't matter for our access. But it turns out that they are neither small nor stable over time. So in the chart at the top, I'm showing you the ratio between the dollar value of all reserve liabilities excluding currency in circulation over aggregate reserves. And as you can see, it's a sizeable number and it changes a lot over time. So for example, it started around 40% in 2010. Then it dropped to less than 20% in 2011. It stayed there until 2014. And then it started to steadily increase over time, exceeding 80% of aggregate reserves in the system in much 20 countries. But not only sizeable, it also displays a very rich dynamics. So at the bottom here, I'm focusing on the two liabilities that I mentioned earlier, the TGA on the left and the overnight RRP on the right. As you can see, the TGA was almost always below 100 billion from 2010 until 2015. And then it started to increase steadily and it exceeded 1.5 trillion in mushroom defense. The overnight RRP on the other hand did not exist before 2013. But then it's usually exploded until before 2013. Then it's usually exploded with big exceeding 400 billion. And now you can see in this chart but the overnight RRP in September has exceeded 1.5 trillion, just like the TGA. Now let me talk a bit about the market in which reserves are traded. So that's called the Fed Funds Market in the US. It consists of unsecured lending, mainly overnight. And the rates at which reserves are traded, called Fed Funds Rates, are the rates targeted by the Fed in its more variable sentimentation, as I said before. Now, what does the theory tell us about these rates? So absent frictions, they should always be above the interest paid on reserves balances, which was zero, by the way. And that's because no bank would ever incentive to lend its reserves at a rate which is below the rate that it earns. Or it's a count of just letting cash stay. At the same time, it should always be below that this can win the rate, which is the rate at which banks can borrow from the bank, because no bank is an incentive to borrow from another bank at a higher rate than what they get from the Fed. I'm abstracting from market segmentation and stigma, but that's not important for what we are trying to do. So what does the theory tell us about these reserves? So all the models of the reserve demand curve identify two regions. There is a region in which the curve displays a very steep slope that's region of scarce reserves, and it's usually around aggregate reserves required. Then there is a region far away from aggregate reserves requirement, in which the curve is perfectly flat. And that's called the region of abundant reserves. Now, what's in between scarce and abundant? Well, that depends on the model you're looking at. All models without trading frictions, they predict a piecewise linear relationship with a king between flat and steam. Now, more modern models that allow for trading frictions actually predict an intermediate region in which the curve displays a gentle slope, so a smooth region of transition. Now, let me talk briefly about monetary policy implementation to motivate why what we're doing is important. And let me start with pre-2008, okay? So before 2008, reserves at the site were in the terms of billions, they were now remunerated, thanks to strong incentives to actively trade with each other on a daily basis, and the demand curve had a very steep slope. So that meant, from the Fed's perspective, that the Fed could keep the target rate by just tuning with small adjustment the reserve supply through the open market operations, because even a small change in the reserve supply would imply a sizable change in the price. But how can you do monetary policy with abundant reserves? Well, when reserves are in the trillions and they are remunerated, banks have lower incentives to trade, so supply tuning is ineffective. Even some large, relatively large changes in the supply will not materially affect the food. So the Fed started implementing monetary policy for administrative rates, so basically it changed the interest on reserve balances and the discount win rate. And by doing so, it changes the opportunity cost of banks of all name, their reserves in their accounts. So changing overnight in the interest on reserve balances, the IRB, basically corresponds to vertical shifts in the reserve demand curve, which we want to control for, because they would contaminate our estimation of the reserve demand curve. We are not interested in these vertical shifts, we are interested in the curve itself. Okay, let me close the institutional section by talking about drivers of the demand for reserves by bank. And the literature has identified many thin drivers of an upward pressure in the precautionary demand for reserves for the crisis. The first one is quite obvious, is the new regulatory supervisory framework. So think of liquidity and poverty ratio, living bills, supervisory stress taxes. And even setting aside regulators and supervisors, banks themselves have changed their internal liquidity risk management in response to the crisis. And now they have a higher demand for safe and liquid assets such as reserves. And then there is a third driver, which I think is very important and sometimes people forget about it. And I find that over time after the crisis, the interbound market for reserves has become quite a trope in terms of, so the volume in terms of liquidity has become much lower than what it used to. So the lack of debt in the late day funding markets combined with the need to submit large intraday payments has increased the precautionary demand of banks for reserves. And my co-author and colleague, Gara Fonso, will present a paper exactly on dystopic later in this afternoon. So let me move now to model identification trash. So we postulate these structural demand curve. So that's the demand curve that if we can see that we postulate for bank reserves. So as is the weighted average bank transfer rate minus the IRB, and that's the dependent variable. The main independent variable is aggregate reserves in the banking system normalized by banks for the last few. So all parameters in the model, including the variance in front of the structure are allowed to be time-varyed. Our object of interest is of course beta. That's the time-varying elasticity of rates to reserve shocks because that's the time-varying slope of the demand curve. So what's the trick behind this model? They're basically using a high frequency time-varying linear model to cast the volatility to capture the non-linearities predicted by the team. So that's the assumption here for this to work. And the assumption is that the structural parameters that govern the structural equation have all more slowly than the liquidity shocks that hit banks every day. And that's reasonable because it took banks months to adjust to the post-crisis frame for the drives that was mentioned. So let's talk about the endogeneity issues we are facing. So the first one is obvious and has to do with Fed's interventions. And on a daily basis to open up operations, that's correct. But it still does respond to unusual dislocations in the Fed's market. And here the example I want you to have in mind is in September 2019 when Fed funds rates and overnight repo rates spiked up on September 16th and 17th when the effective Fed funds rates actually reached the target proportion. Over the following days, the Fed responded by expanding their reserve supplies to the market, and within a few days both Fed funds and repo rates went back to their private liabilities. Second type of liquidity has to do with no reserve liabilities and that's why it's time so much time. So not only they change mechanically the level of reserves in the system but they also correlate to reserve the main shocks. And the reason is that the main holders of these no reserve liabilities are key money market participants think of the Treasury, money market funds having or no reserve liabilities and they are acting both in the Fed funds market and in the repo. So they are actually the users of these no reserve liabilities both affect and depend on banks and banks overnight. And again, an important example is what happened in mid September 2019 when spillover from the repo market was made onto the Fed funds market at the same time where it correlated with changes in aggregate users. So let me be a bit more specific to fix ideas and I give you two examples and four examples of entertaining. The first one has to do with the window dressing of European banks that are on offense. So our on offense European banks reduced their wholesale wholesale short-term borrowing. They do so to improve their regulatory capital ratios that are calculated. Now that means that they demand for reserves so they demand for borrowing in the Fed funds market also decreases because that's wholesale overnight. At the same time though they reduce also their borrowing from money market funds which are their main lending wholesale. And what do money market funds do when they face this decrease in the demand for funding while they place their cash at the overnight European? Which implies that the overnight European goes up and aggregate reserves go down. So the good thing about this compounding factor is that it reverts within a few days and it's highly predicted. Now the second type of indigeneity has to do with the Treasury. And for example with Treasury office. So on settlement dates banks demand for overnight funding especially Ripple Fund. And the reason is that they finance their purchase of Treasury's and new Treasury's through overnight Ripple Fund. Now this puts up for pressure on the reserve demand because borrowing in the Fed funds market so borrowing reserves overnight and borrowing Ripple's collateralized by Treasury's so there is an upper pressure in the reserve demand. But at the same time when banks submit their payments to the Treasury for the new securities, they use the reserves. So there is an increase in the TGA and a corresponding decrease in reserves. So these are shortly a very frequent compounding factor and corporate tax payments display a similar demand. They have a very similar demand. So how do we deal with indigeneity? So first of all between the data we drop one-day wind those around one times and these take care of the wind addressing of European banks. And that's the first example that telling you that using daily data is key for this type of exercise. But then we have a more general which is an instrumental variable approach. As I said, it's two components. So in the first part we build a forecasted model of the joint dynamics of quantities so reserves Q and rates prices X. The model is basically an adaptation of the daily time-varying VR with stochastic volatility proposed by Prime Minister in 2005. And again the key assumption is that the forecasted model the parameters of the model move more slowly than the daily forecasted model. Then we take the past forecast errors coming from this forecasted model for the path of reserves over time as instruments in an IV estimation of the structure of the manipulation. To fix ideas we estimate our forecasting VR we obtain forecast errors here. We use five days ago forecast errors as instruments for Q in our structural application. So the IV estimate that we written is usual as a ratio of two covariances yet the cool thing is that we allow these covariances to be ten-vary but instead of doing instead of using the usual two-stage square we pull back our covariances from the estimation of the reduced form VR. And that allows us to write the theta as a ratio of two impulse response functions. A denominator of the impulse response function of prices to quantity and a denominator of the impulse response function of the impulse response function. So in terms of additional IV I just want you to think of this IRS denominator as the reduced form regression and one other denominator as a first-stage. So since we are doing IV let me convince that our instrument is exogenous and again the data will not be key. So the exclusion restriction is that the forecast forecast errors are uncorrelated with demand structure errors. So let me tackle Fed's intervention first. So since 2008 the policy has been implemented through administer rates. So I know that you have a market approach and Fed's research supply only responded to unusual dislocations in the Fed's market. And that happened typically with a delay of at least a day. Think of it as a time-consuming thing. So this is telling you that it is key to use daily data to build the model and construct the forecast error. Now let me talk about no reserve liabilities and very much made. So the key thing here is that these confounding factors that typically are transitory and they last for less than five days. And good examples, again, are treasury options to explain this. So let me briefly talk about the relevance of our instrument very briefly. So it turns out that our sorry just to say is relevant and this comes from the fact that we are looking at the post-prize period where the path of reserves displays great persistence and the second reason is that our model is good in the path of reserves. Now let me move to the results and let me close with the results. So as I said at the beginning the main result, the first result of the paper is that the reserve demand curve is shifted over time. So here in this slide I'm showing you at the top the same chart I showed you at the beginning. So those are in-sample joint forecast of prices and quantities from 2010 to March 2021 and the bottom chart I'm showing you the out-sample joint forecast. So that's our model implied demand curve in-sample and the one at the bottom in real-time. So the first thing to notice is that there is markedly seen and that speaks to the real-time performance of our model. Then the second thing to notice in terms of economics is that there are these clear outburst shifts over 2010 to 2019 so the blue and then March 2020 as represented by the red the dark red dots Now let's move to the main object of interest in our data which is the LSTC of fat transfer rates to stocks and reserves So you are learning the LSTC of fat transfer rates and basis points to stocks and normalized reserves percentage points the dark blue over time so 2010 to March 2021 the dark blue line is our posterior median or S2 and the shaded areas represent credible sets around. So what this picture is telling us is that there was a significantly negative slope at the beginning, 2010 to 2011 then it disappeared from 2012 to 2017 and then it will emerge again in 2018 and 2019 and the curve became flat again as the pair responded to the COVID-19 outbreak by expanding. I also want to emphasize that this affects actually this negative LSTC at the beginning our sample in 2018 to 2017 are actually economically important if you think that a one standard deviation shall in aggregate reserves would explain 50% of the in-sample standard deviation of fat transfer rates in 2010 and 30% of the standard deviation However the numbers are much more than the numbers researchers used to obtain for the scarce period or at least for the pre-2008 period in the 90s So there are several robust instructions in the paper of course the idea of our main robust instructions is to explicitly control for family transfers that people have put forward like other money market that can also not only can be lower on to the fat transfer rate but also correlate with aggregate reserves and clear examples that are repo rates, fee bill yields or money market funds and the idea behind what we do is to just augment our forecasting model to explicitly control for this compounded fact This for example is the LSTC that we obtain when we control explicitly for repo rates in our forecasting model and as you can see the results are remarkable this year So let me conclude with the question or maybe the question I think is key for the fat and I believe some of you may be wondering about it quite some time and the question is so for now I've just shown you our estimates of the LSTC, the slope of the curve we're trying but the question could be for what level of results actually the curve stops being flat because the theory tells it that the demand curve is no limit so in terms of the old model the question is where is the king in the demand curve or if you want to think in terms of the more modern models where does a gentle slope start for more here I'm showing you the same chart I showed you before at the bottom so let's aware that I'm a structural estimation of the sample structural estimation of the time varying LSTC of the reserve demand curve the only difference from our baseline the only difference is that at the top now I'm showing you the path of aggregate normalised reserves in the system so aggregate reserves divided by banks of points and the dash red vertical lines correspond to the points in time at which the 95% credible set crosses zero so in frequentist terms if you want those are the points in which the LSTC stops being significant from the 95% points and as you can see the point in 2011 was slightly below 12% and it's slightly above 12% of banks of the LSTC at the beginning of 2018 so there seems to have been a shift to the right in this point however if you think in terms of money those two points roughly 12% for banks of the LSTC correspond to completely different levels in dollar values but they corresponded to less than 1.6 freedom at the end of 2011 and slightly more than 2 treatment at the beginning of 2018 so to wrap up in this paper we propose a structural time varying estimation of the reserve demand curve over 2010 to 2021 and the whole post crisis period we do so by using a combination of a stochastic volatility time varying for casting model casting era and we combine it with an instrumental variable approach applied at the date I didn't show you about our forecasting model as excellent out sample real-time performance but the most important thing is that what we show is that the reserve demand curve has moved upward over time consistent with drivers the presence of upward pressure in banks demand precautionary demand for reserves and its slope has changed significantly also in particular in 2018-2019 we estimate a significantly negative slope even though reserves were around 2 treatment thank you Galileo thanks a lot for the presentation and yeah let's move right away into the discussion and we have Hubert O. Ennis from the Richmond Fed about the floor zeroes thank you let me start by saying this is a really good paper and my plan is sort of to share some thoughts with you about the paper you know I have this disclaimer down there that I don't think of myself as having views but in case some of what I said comes across as views those are not of the views of the reserve but what I'm going to try to do is just share some thoughts to start let me just say you know briefly the sort of motivation behind this kind of discussions is that the Fed intends to implement monetary policy using a floor system and this is sort of the language of the Federal Open Market Committee and on the 30 2019 they said they intend to implement monetary policy with a regime of ample supply of reserves that ensures control over short term interest rates but setting administer rates and no active management of the supply of reserves and there's a bunch of keywords here you know ample control active management but we'll maybe talk about some of those key question how much is ample and this is kind of what they are addressing and the Fed tries some other stuff they try to ask the banks with some mixed results and what these guys are going to do is they're going to try to estimate the sensitivity of rates to the level of outstanding reserves and kind of get the sense of ample from that so the paper estimates an aggregate daily excess demand for bank reserves in the US and I say excess because in principle some banks demand reserves and others supply reserves the Fed provides the net aggregate supply but there are endogenous like Gabriele explained so well there's a bunch of endogenous and autonomous factors that change the supply so what they do they consider this equation S is the spread between the effective funds rate and the interest on reserves and Q is the aggregate quantity of reserves in the system normalized by total bank assets and so it's just basically price on quantity regression the slope cost and you know it's a sophisticated you know procedure and so they have the slope coefficient this beta and all the other coefficients are allowed including the variance of the shop is allowed to change over time and Gabriele explained very well why this is important in this exercise there's a classic endogeneity problem this sort of you learning is the first exercise in econometrics like the supply maybe moving in response to market conditions and so you get the supply and the demand moving simultaneously and what they do is they use a idea approach to estimate the coefficient beta based on sort of exogenous changes in Q and it's kind of a nice procedure but I'm not going to talk much about it Gabriele on this topic I think it's important is that Gaara told us that you know most of the lending in the Fed funds market is being done by FHLBs fair home loan banks their government sponsor enterprises that cannot earn interest on reserves and this creates a composition effect that combined with some internal management practices of these GSEs tends to push the effective rates below the interest on reserves and Gabriele touched on this topic also Gaara told us that most of the borrowing is done by foreign banks to arbitrage interest on reserves so this makes the pricing in the Fed funds market the ideas in practice in my opinion and so if you are accustomed to think about the pool model of the money I feel like that's not necessarily a good framework to understand the dynamics of rates and how rates move as total standard reserves change during this period and the regulatory changes that Gabriele mentioned and other things tend to influence the demand and also this demand and this supply are very kind of the demand is coming from FBOs and the supply is coming from HLB that are not earning interest on reserves so it's tricky the standard logic of that one tend to use for downward sloping demand curves is really I think not a great abstraction to use Gabriele emphasized the endogeneity I'm going to square it if you know in their paper they discuss several channels that make the endogeneity the Fed moves supply and reaction to market conditions and Gabriele explained that really well non-reserve Fed account balance is moving reaction to market conditions repo markets and Fed fund markets are tightly connected affecting reserve supply and rate but they are running examples of these events in September 2019 the Fed intervene and successfully calm this what appear to be this functional markets I get the sense that expectations of Fed intervention partly explain the spike in the repo rates in 2019 so it's sort of like even the spikes were endogenous in my mind think of think of the case where there were funds in the sideline so there were there was short term funding available after a long period of very high reserves but those arbitrage rates that were needed to counter the spike did not happen had not happened for a while and then there were set up costs and other administrative procedures that were needed to arbitrage to turn on the arbitrage machine and so on top of that market participants probably anticipated that the Fed would intervene and squash those arbitrage opportunities in a few hours or days just that it happened and some paying those costs was un-economical because you wouldn't have a lot of time to take advantage of the arbitrage so I think it's important to keep in mind that the way markets function depends on the anticipation of central bank intervention okay so there are results the authors find that the rates start to become sensitive to the quantity of total standing reserves the level of reserves is approximately 12% of total assets in the system and that's approximately $2.6 trillion of reserves right now now I want to say that the Fed recently placed a standing repo facility that in principle might have changed that because early on in the discussion of this facility one of the motivations was to reduce the demand for reserves by balance now the slope coefficient is roughly equal to minus 1 when reserves are 8% of assets so that's about 1.8 trillion of reserves now the way I look at things is the Fed targets a range of 25 basis points wide for the effective Fed funds rate and the range of short-term fluctuations in total reserves is around $2 billion now there are these slow moving things but the blips are around $2 billion maximum so that's like 1% of banking assets that translate into a one basis point fluctuation in rates so it suggests to me that even with reserves at 8% these things will work fine in terms of targeting a range okay so why am I thinking that well there are benefits of having a small there are important benefits of having a small balance sheet for the central bank political economy you know reasons and also you know price risk and this perception that so things that impact the central bank independence which I think are very important to sort of protect so my I come out with some broader policy questions should choose small and short-lived fluctuations in overnight interest rates measuring basis points be avoided at all costs and I wonder is this about bond trading and should it be about industrial production and more broadly about GDP and social welfare now this picture is these are 5-year periods and then I plotted the FedFans rate volatility so it's the fluctuations in the FedFans rate in these windows and the windows are 92 to 96 2002 to 2006 so that's black and green and then you get 2012-2016 and then the red is three years for 2019 2017 to 2019 and you can see that the September 2019 blip is kind of small relative to what we were asking to see before and I'm not thinking about how well we do with controlling rates I'm talking about how the US economy worked during these periods so just to conclude this I think is a good and important paper it's very well executed has ingenious and sophisticated estimation procedures I think the finding seems reasonable to me though the interpretation seems less obvious to me and that's kind of what I try to discuss with you and then the policy implications is again even I don't have it clear I think they seem to suggest that maybe 12% of reserves over total assets is a reasonable it's a good place to be in terms of total reserves but I'm not sure that it's the right way to think about it and thank you very much and great paper thank you Hubert well just to check maybe whether there's a question from the audience just seeing whether there's anything in the chat I don't see anything so given the advanced time unfortunately maybe just back to you Gabriela whether you have a one minute response or two minute response to Hubert has just laid out sorry for the pressing on the time I understand and do my best so first of all thank you so much Hubert for the wonderful discussion I will pass upon a few points so the theory and the fact that holds model or other models may not be the right models to look about I think you're absolutely right we want to be agnostic about the theory we need them to understand we just look what theoretical models tell us about the curve and then we try to bring those theories to the data in a natural I agree with you in particular that I think there's need for most theory that looks at market segmentation in a role of different players in this market because they're not all the same it's not just about liquidity shock needing identical banks so there's something more and at the same time I think there should be more models about that and more empirical analysis about the reserve demand curve at the individual bank level or bank time so that's key now I want to briefly talk about interpretation and what we mentioned near the end about the timing of these attacks and again I want to reiterate these are my own personal views not the views of the Federal Reserve System so I agree with you I think in the sense of what our paper is telling us is that there was a negative slope let's take September 10th there was a significantly negative slope back then how big was the slope it wasn't big enough probably to explain the huge jump with the Federal Reserve System so this is telling us that it was not just about movements along the curve there was something else so this is pointing to the importance of the slow moving factors that push the curve up and down not just the slope but also the position of the curve changes and may get closer to the upper limit of the target range and which we have in the model but and in particular is pointing out to the role of stocks to high frequency demand stocks that is exactly what we want to filter out when we estimate structurally the reserve demand factor so I agree with you and I think that one of the messages from the paper is that researchers should spend a long time thinking about modeling these high frequency demand stocks in the reserve demand factor that correlate also with how we are preserving this system and there is a growing literature there and our colleague Ivan Kopan is a paper with Darryl Duffy Anton Matan is a paper with Simon Porter in some sense speaks to the same talk so I agree with you and I think that we should not only think about ample mass scarcity and abundance in terms of the reserve demand factor itself in the movement along the curve but we should also think about it in terms of how to correlate about the shocks the high frequency demand shocks that affect that constraint and that is the basis Thank you so much Thank you Gamela Thank you very much for presenting a very interesting and very relevant paper I am sure there will be ample opportunity to put this paper to test in the future so we will look forward to that or look forward to your results in the future and also thank you So thank you for all that, of course, for the discussion. Thanks a lot.