 So good afternoon from my side. I'm Christoph comes from the ECB's Monetary Policy Department. So welcome to this afternoon session of our conference, which is also the last session before we're going to wrap up I'm very happy to chair the session in a conference that so far I've enjoyed very much And so the last session of today is on banking and safe assets. We're gonna have two papers The first one will be presented by Shohini Kundu, who is an assistant professor at UCLA He will speak about the aggregate effects of deposit shocks and then we are happy to have Salim Bahay From the University College of London as a discussant and as in previous sessions we're gonna have 25 minutes for the presentation then 10 to 15 minutes for the discussant and then Some minutes also for questions from the floor here in Frankfurt, but also from the online audience So if you are connected online please feel free to Raise your questions via the chat function. And so without further ado, I would give the floor to Shohini for the presentation Okay, great So first I'd like to start by thanking the organizers for giving me the opportunity to present this work today The title of this paper is the Geography of Bank Deposits in the Origins of Aggregate Fluctuations This is joint work with Sangin Park and Ashant Vats who are two excellent PhD candidates from the University of Chicago And both of them are on the market this year. So please keep an eye out for them So in the motivation the motivation for this paper comes from a simplified diagram of a bank's balance sheet In this simplified diagram a bank's assets consist of various types of loans securities and other assets and These long-term illiquid assets are financed primarily by liquid liabilities in the form of demand deposits Now this liquidity transformation is critical for financing long-term illiquid assets But it's also a key source of vulnerability for banks in the economy In this paper we introduce a new source of financial fragility, which is the geography of bank deposits We posit that the geography of banking assets and liabilities can make the economy on aggregate more susceptible to idiosyncratic shocks So what do I mean by this? Well banks collect deposits across geographies and then allocate it towards lending activities So concretely loans in Los Angeles might be financed using deposits from Washington DC And so local shocks that affect Washington DC can transmit to distant areas This problem is aggravated if bank deposits exhibit geographic Concentration and that's what we empirically test in this paper We show that local exogenous shocks to areas where bank deposits are geographically concentrated Concentrated can account for aggregate fluctuations, and that's what we refer to as the deposits channel of aggregate fluctuations So the research objective is can local deposit shocks account for aggregate fluctuations? The classical economics answer is no idiosyncratic shocks cannot account for aggregate fluctuations as implied by the central limit theorem But two recently developed lines of inquiry in macroeconomics challenged this view The first is the granular hypothesis of Gebex in which he argues that when Economic activity is fat-tailed in their finite samples the central limit theorem breaks down And the second hypothesis comes from Asimoglu, Carvalho, Ozdoglar, and Tabasco-Lehi In which they argue that idiosyncratic shocks are amplified in the presence of a network structure This paper combines both of these hypotheses First we show that bank deposits are geographically concentrated exhibiting fat-tailed and Second bank internal capital markets provide a mechanism for the amplification of idiosyncratic shocks through their network structure So the overarching objective of this paper is to study the mechanism through which the geography of bank deposits and idiosyncratic shocks can account for aggregate fluctuations There are four key findings in this paper First we introduce a new fact, which is that bank deposits are geographically concentrated Specifically we find that 30% of bank deposits come from a single county Second we construct novel bank-specific deposit shocks by combining the geographic concentration of deposits with local natural disaster-induced property damages Third we show that these idiosyncratic deposit shocks can account for aggregate fluctuations And the transmission of local deposit shocks into aggregate fluctuations occurs through the deposits channel So specifically local disaster shocks negatively affect bank deposits, which then negatively affects bank lending in other areas And lastly we argue that financial frictions are critical for the aggregation of idiosyncratic shocks These frictions include bank capital and regulatory constraints, bank's informational advantages, and the stickiness in bank borrower relationships So first is the fact, which is that bank deposits are geographically concentrated Note that this is the within bank concentration of deposits and not the within county concentration of deposits So using summary of deposits data from the FDIC We identify the share of deposits coming from each county for each bank Let me be specific on how we construct this figure We take a bank like PNC Bank and we look at where PNC raises the greatest amount of deposits from and we rank those Counties based on the share of deposits coming from those particular counties So county number one refers to the county in which PNC raises the greatest amount of deposits County number two is the county in which PNC raises the second greatest amount of deposits and so on and so forth So the x-axis denotes the county number and the y-axis denotes the share of deposits And here we have plotted three different lines So the blue line shows the simple average of the share of deposits Across all banks for each county number The red line presents a weighted average which weights the share of deposits coming Across all banks for each county based on the bank's total assets and the green line Plots the regression margins after taking out a county cross year and bank cross year fixed effect And we can see that regardless of the methodology the largest deposit county accounts for almost 30 percent of bank deposits Now you might be concerned that perhaps the geographic concentration of deposits is a fairly new Phenomenon or perhaps it's driven by the smallest of banks or the largest of banks So we do a battery of robustness tests in which we verify that the geographic Concentration of deposits is not a new phenomenon. In fact, we can see it from 1994 when our sample Period begins The deposit concentration isn't either driven by the smallest of banks nor is it driven by the largest of banks The share of deposits coming from the largest deposit county doesn't vary with the percentile of bank deposits bank assets bank liabilities or bank loans And the largest deposit county is scattered across the United States So it's not the case that the largest deposit county is all coming from New York City, for example So next we switch gears and try and understand how natural disasters affect aggregate deposit growth We show that local natural disasters negatively affect bank deposits both in the short run and in the long run And we identify county level shocks using local natural disaster induced property damages So this figure presents a heat map of the property damage per capita across counties from 1994 through 2018 And you can see that some areas in the United States are very prone to natural disasters Relative to other areas such as Texas for the Gulf region and some areas experience relatively less property damage per capita over the sample period and These natural disasters represent a wide range of hazard types Everything from winter winter weather and severe storms to hurricanes and tornadoes Some of these events occur very frequently such as wind and severe storms, but they don't incur nearly as high damages in terms of the total cost relative to events that occur less frequently such as hurricanes and wildfires So next what we analyze is the relationship between deposit growth in a particular county and the disaster shock in that county in the previous year And in our most conservative Specification we include county fixed effects to control for the time invariant heterogeneity across counties and state cross year fixed effects to account for state level trends and We find that a one standard deviation disaster shock is associated with a point zero seven to point eleven percentage points decline in deposit growth This is comparable with the 25th percentile of deposit growth So we find that a one standard deviation disaster shock corresponds to a loss of $570 per capita and so this result tells us that there is an immediate effect on deposits following a disaster shock Now naturally the next question that arises is is this effect transient or is it persistent? So to answer this question we study the Georgia projection in which we analyze the long-run response of deposit growth to disaster shock We find that there's an immediate decline following a disaster shock in deposit growth and this effect persists even 10 years after the initial shock and So the takeaway from this analysis is twofold First we find that local disaster shocks negatively affect local bank deposits and second that this effect is permanent So next we look at how disasters affect aggregate deposit growth by By constructing bank-specific Deposit shocks waiting the county level shocks by a share a bank share of deposits in a particular County and so here We have constructed this bank level shock by waiting county level damages per capita based on the bank share of deposits And after constructing the shock we established two key properties of the shock First that the shocks are idiosyncratic and second that the shocks are important So for the shock to be idiosyncratic, they shouldn't be foreseen or anticipated And so first we regressed our bank level shock on a multitude of bank-specific variables This includes total assets loans equity cash deposits hedging assets dividends and income and we find that there isn't any robust Pattern between these bank characteristics in the bank level shock Second from the univariate regressions We find that the explanatory power is close to zero Suggesting that these bank characteristics can't predict bank level shocks in any robust statistical or quantitative sense So even if you don't buy that we then analyze whether these shocks are idiosyncratic in two different dimensions The figure to the left Foughts the kernel density of the air one coefficient associated with these bank level shocks and the figure to the right Presents the kernel density of the pairwise R square for these bank level shocks and both of these figures Indicate that the air one coefficient is centered around zero as well as the pairwise R square This suggests that these shocks lack spatial and temporal dynamics and they exhibit low persistence in low cross bank Correlation, which are two properties that you might think are associated with idiosyncratic shocks However, despite being idiosyncratic these shocks are important in the sense that they can predict aggregate declines in deposits and liquidity creation So in these two figures we present the long-run bank response in deposits and liquidity creation to the deposit shock We find that there is an immediate impact Once a bank experiences a negative deposit shock and these effects diminish five years after the initial shock occurs So in terms of magnitude we find that a one standard deviation increase in the deposit shock is associated with 0.97 percentage points decline in deposit growth and 0.19 percentage points decline in liquidity creation So next we extract aggregate and granular shocks from these bank specific shocks using the giv methodology of gebex and coyenne So we take our bank level shocks and we weight it by a bank's share of lending activity in the economy to construct these aggregate shocks And then we construct granular shocks by subtracting equal weighted shocks from the aggregate shocks And the reason why we subtract equal weighted shocks is to remove the effect of all common factors So intuitively this granular shock captures the idiosyncratic deposit growth experienced by large banks following natural disasters Now this figure presents a time series plot of our aggregate deposit shock And we conduct a narrative analysis to ensure that these shocks don't just reflect noise But rather they reflect fundamental changes in the microeconomy of different regions And you can see that our aggregate shocks coincide with major natural disasters the largest of which is hurricane Katrina But there are also major floods and wildfires that also occur throughout this time period We then conduct a narrative analysis of the crest and you can see Based on this table that oftentimes these natural disasters affect Multiple states and oftentimes there are multiple natural disasters that occur in a given year And then lastly we show that these aggregate shocks are nicely correlated with the insurance payout Suggesting that if that the aggregate shocks reflect the magnitude of the disaster that occurs in that year So next is our headline figure in which we document that the granular deposit shocks can account for aggregate fluctuations And specifically we posit that there's a negative relationship between bank deposit shocks and GDP growth So in the next set of slides We'll codify this relationship more rigorously through regression frameworks and establish that the deposits channel can Explain a substantial amount of aggregate fluctuations of aggregate fluctuations So first we regressed GDP growth at time t on the granular shock at time t minus one And we find that a one standard deviation granular shock reduces economic growth by 0.05 to 0.07 percentage points This estimate is statistically significant at the 1% level and is robust across all columns Naturally the next question that arises is well how much variation in GDP growth can be explained by these granular shocks So to answer this question, we regressed GDP growth at time t on labs of our granular shock And we find that collectively the granular shocks can explain 3.3 of variation in economic growth as indicated by the r-square Now to understand whether 3.3 percent is a large number or whether it's a small number We study how our shock compares to other shocks that are commonly used in the macroeconomic literature And we do this by conducting a kitchen sink analysis in which we run a horse race between our shock Excuse me and other shocks that are used in the macroeconomic literature This includes oil shocks monetary policy surprises uncertainty shocks term spreads government expenditure shocks and gobex's granular residual And there are two key takeaways from this analysis First is that our granular shock can explain as much variation as other macroeconomic shocks And in some instances it can explain more variation than other macroeconomic shocks And second that the effect of the granular shock is robust to controlling for other macroeconomic Shocks as you can see by the point estimate across the first row Now a concern with our analysis is that yes, perhaps the granular shock can explain as much variation as other macroeconomic shocks But it may capture the direct effect of disasters on economic growth Rather than the effect of idiosyncratic shocks to deposit growth So to address this concern We examine the long-run response of GDP growth on aggregate disaster shocks Which is measured using total property damages per capita And we compare this to the response of GDP growth to our idiosyncratic shocks Now we find that there's no statistically or economically relevant direct effect of the disaster shock on economic growth Which lends credence to our main finding that The results are driven by idiosyncratic shocks to deposit growth And now using this setup, we're then able to estimate the deposit elasticity of economic growth And we do this using a 2SOS specification Now this is the main contribution of the paper This paper addresses an unanswered question in the macro finance literature, which is What are the aggregate effects of deposit shocks? Identifying the effects of disruptions in bank deposits on economic growth is a major empirical challenge And that's because so far the extant literature has relied on cross-sectional estimates The issue with using cross-sectional estimates is that aggregate variables that don't exhibit cross-sectional variation Can still affect aggregate elasticity So by estimating the beta coefficient and cross-sectional regressions A partial equilibrium effect is being picked up on but not a general equilibrium effect because the intercept has not been identified This identification strategy allows us to identify the missing intercept and directly estimate the aggregate effects of deposit shocks So in column two We first regress deposit growth on the granular residual And then use this instrument this estimated value to then estimate the deposit elasticity of economic growth We find that the deposit elasticity of economic growth is 0.87 Suggesting that a 1 percentage 1 decline in deposit growth is associated with decline in economic growth by 0.87 percentage points Now this estimate may seem substantial So we then estimate the loan supply elasticity of economic growth for which there are estimates in the extant literature And we find that the loan supply elasticity of economic growth is 0.14 Which falls within the range of past work that I have with nishant as well as herenio 2020 suggesting that this Estimate of 0.87 is well calibrated Now further we can use this setup to estimate the money multiplier And what we find is that a one dollar increase in deposits increases lending by one dollar 18 cents So far we have shown that deposit shocks can affect aggregate economic growth But the question is how do local deposit shocks affect economic growth? We argue that the key mechanism through which shocks to banks affect economic growth is lending activity And we establish this by using micro data on small business lending and mortgage lending So here in our estimation We study what the relationship is between lending growth Conducted by a particular bank in a particular county in a particular year On the deposit shock that's experienced by that bank Our identifying assumption here is that banks face identical investment opportunities within a county A weaker version of this identifying assumption Is that any friction which creates a wedge between available investment opportunities across banks within a county Is unrelated to an idiosyncratic shock that happens elsewhere And in addition we control for the time and variant importance of a bank within a county by including county cross bank fixed effects And so in studying the relationship between lending growth and the bank level deposit shocks Even after including high dimensional fixed effects We find that a one standard deviation Deposit shock is associated with a decline of 1.09 to 1.85 percentage points in lending growth Now you might be concerned that perhaps this is driven entirely by the affected counties that experience natural disasters So to address this concern, we disaggregate the lending conducted in unaffected counties and affected counties And we find that the estimate is significantly higher in terms of magnitude for unaffected counties Relative to affected counties, which is consistent with the literature that's documented That affected counties experience an increase in credit demand following natural disasters We further verify these results using data on mortgages So here, uh, we study the relation between deposit shock And mortgage lending and we're able to disaggregate based on the type of mortgage So in column one, we have new home purchases in column two. We have refinancing loans in column three. We have home improvement loans and we find that the pecking order of the effects on different mortgage types is consistent with the argument that, uh Contracting frictions are more pronounced for new home purchases because borrowers don't have an established payment history unlike with home refinancing and home improvement loans And so the contraction in lending is dominant in loan types where banks face more contracting frictions We also find that the effect is dominant for loans that are more likely to be funded by deposits Which further lends credence to the deposits channel of aggregate fluctuations We establish this by comparing jumbo to non jumbo mortgages So a unique feature of the mortgage market is securitization in which Oftentimes deposits are replaced with bonds as a source of funding And this is a result of the secondary market activities of the government sponsored enterprises Fannie Mae and Freddie Mac cannot purchase jumbo mortgages. So these are more likely to be driven by deposit funding So in this regression specification We compare jumbo originations to non jumbo originations conducted by the same bank in the same county in the same year And we also include county cross bank cross jumbo fixed effects, which also allows us to relax our weak identifying assumptions and consistent with this we find that indeed Uh, the contraction in lending is pronounced for loans that are more likely to be funded by deposits, which are jumbo mortgages Let me move this okay And then lastly, uh, we highlight the importance of financial frictions in the aggregation of idiosyncratic shocks I won't go through these results in the interest of time But we find that the lending cut is driven by constrained banks which cut lending more sharply in other areas when they experience local deposit shocks We also find that banks cut lending in areas where they lack informational advantages and they're less likely to extract rents in those areas And we highlight that firms which are more dependent on banks as a source of external financing drive the response and lending growth And borrower financial constraints and relationship frictions can exacerbate both the cut in lending as well as the real effects that are realized We conduct a number of robustness checks Uh, demonstrating that large banks are responsible for the transmission of idiosyncratic shocks that the geographic Concentration of deposits matters. It's not driven by shocks to collateral value or granularity and gdp employment And population so before I conclude and while I conclude let me just highlight the three key ingredients for this story For our results to go through there are three important factors that can explain how idiosyncratic shocks can account for aggregate fluctuations The first is that bank deposits Should exhibit geographic concentration The second is that the banks have to be large to transmit these shocks through internal capital markets And third is that the natural disaster has to be sizable If deposits were perfectly diversified Then even the largest natural disaster would end up being a drop in the bucket when it comes to effecting Funding for the bank If the bank is a small bank and it experiences a large natural disaster It won't be able to transmit these shocks through internal capital markets And if there's a little bit of rain even for the largest bank in the most concentrated market It's unlikely to wipe out deposit funding. So to conclude This paper documents a new source of bank fragility, which is the geography of bank deposits We find that bank deposits are geographically concentrated and we highlight the role of internal capital markets in propagating idiosyncratic shocks Specifically, we find that idiosyncratic shocks can account for 3.3 percent of variation in economic growth and the mechanism through which This occurs is that natural disasters negatively affect deposits, which then negatively affects bank lending And lastly, we highlight the role of financial frictions in magnifying the deposit channel These frictions include bank capital constraints, informational advantages, and the stickiness in bank borrower relationships Thank you, and I'm looking forward to the atticilline's discussion Thank you very much, Shohini, for this very clear and very interesting presentation And let me then hand over to Salim for his discussion Okay, can you hear me okay? Yes, yes, perfect. So thank you very much to the organizers for inviting me to discuss this paper. I really enjoyed reading it and it's a real pleasure to get to discuss So, you know I don't need to motivate this question. I don't think at all to this to this audience that they're looking at the aggregate effects of shocks to to bank credit supplier and the paper is is very well done There's a neat identification strategy There's incredibly thorough empirics as you saw from from the presentation that the question is important And you know, I've seen your family put this before I knew she would do a fantastic job in summarizing the paper So I'm not going to try and compete with her In doing so, instead I want to basically just take it as as read and focus discussion on some comments and reactions that I had So I want to do basically three things in this discussion First is just press a little bit harder on identification And you know, this fundamentally is a paper identifying an aggregate shock, which is very difficult It's very challenging And so as sort of like an intellectual exercise I wrote down another model to see whether I could actually get a situation where the granular IV they used is correlated with output absent any banking frictions It's possible and we can cripple about how level those mechanisms are But the main the main point of that is to I wanted to press the authors in a bit harder about what variation they're using and what's generally going to result essentially I'll explain that in the second discussion Um, then I want to talk about the mechanism and in particular the role of the wholesale market and smoothing these shocks And the allocation of deposits, which is not something that's really discussed In the paper. I think it's an important feature of of of how these shocks could propagate And lastly, I basically have a question about about non-legal deposits And it wasn't clear to me from from reading the paper how exactly You think you've you've dealt with that and so I wanted to raise that and put out as a final point Okay, so I'm going to jump into this little more about I wrote it down As a starting point it's super simple to explain it in about about five minutes Um, and then it will allow me to think about identification a bit more So I'm going to have a situation where there's there's N counties in the US and time index by T as a single trade of all consumption good And it's a representative household in each county and a bank in each county too Okay, so it's about our bank area Asian The households are going to maximize lifetime utility from consumption And they're going to consume an endowment And income from deposits of an indulgent interest rate here RT And their savings will be indulgent as to ET And then they'll have some period budget constraint And then there'll be some disasters and I'm going to assume that these disasters Are going to damage The endowment so one thing I'm posing here is that disaster does affect How put there was a supply surprising result to me I led the paper that actually if you just took an aggregate disaster shocks in the US they don't I'm going to make this assumption assumption that they do You know, we can cripple about that as well. So in period one This this county could receive a disaster poverty pie Which lowers its endowment to a level which is lower than y bar the average level of income That can they would receive In period from period two onwards It will always receive Y bar as its endowment and then we said heterogeneity and the initial level of income So it means some more heterogeneity going on as well which determine the initial size of savings and then turn the banks out So banks in this model, which is this is what it's really trying to include Are basically pure Essentially frictionless price taking intermediaries. The only thing I'm going to do is just force them to raise deposits From households in the same county when they're located. Okay, so they raise deposits always from one place so we just we just Inbuilt the the geographical Concentration that shahoney mentioned but they're going to Raise as much as they as they want at the very interest rate And they're going to lend to a national that tends to firm Okay, so there's no lending friction here at all and there will be some aggregate and evenology which just takes the sum of all the deposits Raised by all banks and and generates some aggregate Return from doing that in t equals one f and b concave and then from t goes into onwards We're going to have a little technology. Let's just jump directly to a steady state Everyone's savings in in period one will just persist into the infinite future. Okay And then GDP is just simply the aggregate output From from all lending plus the endowment And then equilibrium will just be a set of household order equations which pins down the amount of of Savings that households have a very interest rates and the marginal return on on lending Okay, which is a national variable and then rt Will will clear the market And then the key point here is that banks are available In this model. I could just have easily had every single household invest in physical capital and their investment would fluctuate as As they get hit by these disasters. Okay And I know similar to this model. So n will be equal to 100. I have 100 counties and 500 possible It's actually I'm not going to be concentrated for sorts these numbers are relevant But the point I wanted to make here is that The uncertainty is basically just coming from from sampling right so in some periods Five counties make it by an example disaster in other periods four megahit by disaster or whatever So all the variation out there in this model is just driven by Uh, the failure of a large number which was a key part of shahen's talk In the frequency of of national disasters. Okay, so in some sample some simulations There will be a lot of national disasters and others that will be a few. Okay So then we can just do exactly the same exercise as As shahen did. I can construct my granular instrument. That's simply the disaster Indicated whether a region hit a disaster or not I can weight it by the size of the bank's balance sheet in that region Which is just its share of deposits over the national share Mindless this this equal weighting thing, which is what the granular IV relies upon I can compute the correlation between this IV and output. Okay, and and actually the way this Watch the honey is really pushing for is that, you know, if everyone's the same And we're all exentered integral. There's no granularity. This this this instrument goes to zero Even though they may have some natural disasters There's no There's usually quite a short system that out and the correlation between between the instrument and output to zero And this fits the paper and you also get the result the disasters have a permanent impact on deposits and bank lending But that's simply through Households being permanent income consumers. Okay um Now let me just tweak this in a couple of very small ways. So the first thing I want to do is add worth heterogeneity So I'm going to make households have Different ex ante levels of wealth. So period zero endowments will be different. So some will have a high endowment Others have a low endowment. I don't make preferences on the bottom So I say I did it When you're ready the wealthy you have a lower margin to consume. Okay, that's why I'm sort of just forcing into the model here, and then when you do that and you simulate this model the correlation between The granular IV and an output and period two becomes positive and that would sort of push up the estimates that Shelly presented that would mean that essentially these deposit shocks are positive declarative output, right? So it's the inverse of the results we found in the data So it's just it's just making making your results seem weaker and the interesting business is really straightforward banks and wealthy regions More wealthy households have lower margins to consume. So when they get hit by a disaster their deposits fluctuates by less and so Disasters disproportionately hit these wealthy regions also hit larger banks And therefore the impact on aggregate savings is is smaller. Okay And that basically stabilizes output. Okay compared to a sort of just a shock that hits Poor regions and you actually can reverse that by changing preferences around. It's just it's purely a function of these As I assumed, okay The second thing you could try to try doing was looking at heterogeneity and how how hard disasters hit Okay, and and what I had in mind when I did this was the idea that you know a dollar of property damage Um does different things depending on where the disaster happens. So imagine a dollar of property damage in new york made you more Um That or a property damage it's more to gvp or more to the endowment of the people affected Uh, uh property damage in a rural area for instance, okay Maybe that assumption is is incredible. That's what I was going for If you if you have this uh, this setup, okay Uh, then you actually just get the exact correlation that uh, shahon which is that there's a As a correlation between the ground IV and alpha which is negative Okay, and the idea here is simply precautionary saving if you live in one of these regions where A disaster has a bigger impact because also different the given size has a bigger impact on your Income, uh, you want to save more and that means you have bigger banks and more disasters. Okay So I mean these are just different things that fall out of the model. Um, there's a bigger picture Uh, uh, I wanted to get from this intellectual exercise really which is that you know, the key empirical result in this paper is that disasters Hit harder when they occur in regions where big banks raise other positives Right, that's how the weighting is uh, is uh, is constructed. So they hit harder as the output impact is bigger Uh, there's a bigger impact on on lending as well. Okay And you know, I fully take the point that the regions where banks as a whole Raise their deposits from these sort of concentrated reasons for all banks seems to be virtually were distributed across across the US But there was less evidence in the paper. I could see about how Where the big banks are getting their deposits from I was trying to find that And I think you need to do more to convince to read or you could do more to convince to read that There is nothing special about the regions where the larger banks are getting their deposits from Uh, that would explain your your results. So just that's just give me some some basic summary statistics What are their characteristics? Do they differ on their income levels? Which is sort of pointing to This sort of well-heard certainty point. Are they more dense? Are they local distribution hubs? Are they regional financial centers? Etc. All these things would generate a situation where uh, you know I'm thinking more disaster in new york where large u.s. Banks deposits Does have a bigger impact than disaster in a county where maybe smaller banks are are raising their deposits Okay, so I think I think it's more than just where the deposit's coming from. It's where the big banks are deposits from Uh, maybe I missed it, but I really cannot find that in the paper the second uh, and you know and just just for full disclosure the paper does control for for um For population weighting and GDP weighting and point weighting, but my point is more that there could be other mechanisms that work So that's sort of my my pressing on on notification um, then in terms of of the mechanism so You know taking taking the results hold its face value. What they're saying is that there is a permanent reallocation amongst banks in deposits where more of these eucnecratic disasters hit and when it hits a region where You know big banks are concentrated that will lead to even bigger reallocation deposits, presumably that was on my my reading of the paper Precisely because you know you can imagine a situation where there's a disaster People start making payments. They import resources from outside their county that reshuffles deposits around the US and One implication of this since loans are a liquid Is that this reallocation will require clearing in the wholesale market And this market isn't perfect because I've noticed for a long long time in this whole conference about imperfections in the money market And so I was wondering, you know, if you have this persistent and efficient allocation Of deposits, figure out the aggregate supply with just the fact that suddenly because we have these disasters The deposits are now in the wrong place compared to where loans are amongst banks rather than amongst geographies Doesn't that raise potentially the cost of intermediation? for the bank system as a whole because suddenly they have to reshuffle money around themselves to support the distribution of lending and you know higher costs of intermediation would in turn reduce Clear supply. So it's not about the aggregate supply but because it's per se it's about how they allocate it as a result Of the of the shop and so, you know, one simple test. This is simply is your it's a giv Collective measures of stress in the internet market for instance, or perhaps in money markets and the last thing this is part part the question is The key facts in the paper and this is this is one of the sources of variation that Showing to the results rely upon is that 20 to 30 percent of banks Banks in the u.s. raised 20 to 30 percent of the deposits from a single county Okay, and this is this is really the source of the variation for the The liv and sort of implicit in this and it's an assumption that the deposits that are supplied in this county are Related or reacts to disasters in the county, right? So they are somehow From from local agents, right? That's kind of the idea behind this um, and so, you know, the question I have was well, where does this concentration Come from And you're seeing the big national banks do this. So it can't just be regional specialization There's some discussion in the paper about online deposits and shuffling of headquarters around which I think is very welcome But the concern I have is that basically banks Book all their non-retail deposits through one or two Plants, okay So and that may not be the hq. It could be just, you know, a branch. They have a new york one of us sort of financial sector um So all the money to have the deposits covered by the same on the positive data I I presumed yes, I couldn't get the finance on this moment. Yeah for fdic If I look at the flow of funds around 25 percent of us deposits Excuse me. Just one second. I have time within the next minute. Oh, yeah, I'm going to use both points. It's fine uh So about 25 percent of from the non-bank financial sector mfs corporates rest of the world the federal government We'd expect these deposits sort of sit in one particular branch And there's also, you know, interbank lending and corresponding bank activity that will also take place In in so that's one or two branches And so, you know, I was wondering basically are these concentrated deposits really local? And you know as a test at the bank level Is there a difference in sensitivity between deposits and the disasters? Depending on concentration So that's what I want to say. It was a super interesting paper And I want to sort of push you to think a little bit more about your results by having this very impressive piece of work I thank you very much for the opportunity to discuss it Thank you very much. Salim. We now have uh, some time for a couple of questions either from Here frankfurt or from the online audience and and of course also I would like to give shohini Some time to react to your great discussion Maybe we're waiting for questions if shohini would already like Sure, so thank you salim for the very insightful discussion I think it gives us a lot to think about in the next iteration of this paper Um, let me just touch on two quick points So I think so far the paper is kind of agnostic to what actually happens to deposits following a disaster shock We haven't explored whether the deposits exit the banking system or whether there's reallocation Towards other banks in other regions, which from a policy perspective Has different implications in the sense that if you think that the Uh deposits are leaving the banking system entirely You might think that perhaps there's a failure of the discount window or the central bank and stepping in providing funds in the short run If you think that perhaps there's reallocation in interbank markets Perhaps it's the second channel you alluded to on interbank stress And I think the interbank stress channel is is very interesting and something that we should look More into but it's just something that um so far. We've been agnostic to as to what happens after Disaster hits The last point about the geographic concentration of deposits This point is well taken and I think it's something we've struggled with in this project Which is that we're in part limited by the data that's available to us the summary of deposits data Uh has some guidelines on how they book deposits in different regions, but we can't Entirely rule out whether deposits that are booked in one region may actually be attributed to another region In the worst case it's possible that it could understate our results if we end up attributing natural disasters in areas where deposits Don't necessarily experience an outflow or where the disaster didn't occur, but that's that's an accounting Concern that we've we've tried to do a number of tests to get at measurement concerns You know the headquarters switching how that leads to the allocation of deposits But we're largely limited by the granularity of the summary of deposits data And I think it's it's a it's an open question. Why is there geographic concentration of deposits? How does this vary? um Depending on the type of concentration and it's it's something we're looking into Thank you very much shahini. So i'm i'm looking around here. Um In the physical audience, is there any Any question you would like to raise at this moment? If that's not the case, then let me thank shahini and also salim for for two excellent presentations and discussions