 Good afternoon everybody. It's 3.15, so I suggest we start with a final session in this conference. So there's two papers lined up. This session is dealing with the macroeconomic effects of fiscal policy and advanced economies. And the first paper will be presented by Gernu Müller, and he has quite a number of co-authors there actually. Lydia Cox, Ernesto Passen, Raphael Schönder, and Michel Weber. And the topic is about Big G. I understand he's going to get quite granular in terms of its Big G, but going quite deeply into the numbers. You will have 30 minutes for the presentation, and Enning Weber will have 15 minutes for acting as discussant, just as a housekeeping before we kick off the presentation itself. And as a reminder, anybody who has questions or comments to raise, which we will allow to take in the Q&A round, can put up in the chat. And those who are panelists can also raise their hands and then actually ask the question. So with this, I pass the floor to Gernu from University of Tobin. This is for putting the paper on the program. It was very inspiring conference thus far. So as Isabel said, this is trying to work with Lydia Cox, Ernesto Passen, Raphael Schönder, and Michael Weber, and it's about Big G. So what is Big G? Well, if you take a traditional perspective in economics, say, on government's suspending, we start from national accounting in macro 101, and we write Y equals C plus I plus G. And if you take this perspective, then we think of G essentially as a homogeneous good, essentially a fraction of GDP, which is, we think, determined by a policymaker, say, in response to the business cycle, he or she changes G. And then we summarize the effect of these changes in G by a single physical multiplier. Okay, now in this paper, as Isabel already alluded to, we try to challenge this view a little bit and provide a new perspective, which is based both on data and some theory. In the first part of the paper, we analyze micro data and perform something which we could call somewhat pompously perhaps an autonomy of Big G. And we establish here based on this micro data that government spending is heterogeneous. Big surprise. You may say this you suspected all along. So to put some more structure onto this, we try to establish five concrete facts. And then we take these facts and take the simplest possible model and map these facts into this model in order to understand how accounting for those facts changes our view on the physical transmission mechanism. That's the paper. Okay, so the five facts are as follows. First, government spending is granular and here we simply mean the fact that government spending is concentrated among a few firms and sectors. This is not surprising, but the extent to which we document this in there we believe is quite stunning. Second, there are sector bias in the sense that the composition of government spending differs from the composition of spending which we observe in the private sector that is in the rest of the economy. Third, government spending is moderately persistent in the sense that the contracts on which government spending is based is relatively short. Also, if we estimate processes for government spending, we find relatively moderate degrees of persistence. Fourth, if we look at the aggregate variation in government spending, we can relate it to ideas in credit variation both at the firm and the sector level. We are working away from this saying stressing that it's really the ideas in credit shocks, shocks at the firm or sector level which drive aggregate variation over time. So this invites a bottom up perspective rather than a top down perspective when it comes to understanding the determinants of government spending. Last, across sectors there are heterogeneous pricing frictions and we observe that quite pronounced government spending is concentrated in those sectors where price thickness is pervasive and displays our big time in terms of our theoretical account. Our theoretical account is based on the simplest possible model. So the only innovation we do to this model is that we allow for two sectors. Still, within this simple model, we are able to account for all the five facts which we establish in the data. And the first step, we take the model to do some analytical results, so theory if you want, and here we show that multipliers can be infinitely negative if you increase government spending in the sector where prices are relatively flexible. And the next, a new result to the extent that in the one sector, new Keynesian benchmark model, the multiplier is never negative. Okay, it can be negative in our setup. Also somewhat counter-intuitively if we raise the overall stickiness we find that the multiplier depending on how you measure it can decline even though price thickness increase. This is theory results. Also at the zero lower bound, we see that the ranking of multipliers across sectors may flip. So in the last step of the analysis, we take the model and calibrate it to those five facts. And we find that the impact multiplier of a shock that originates in the sticky sector is about four times larger as a shock that originates in the relatively flexible price sector. Also, the multiplier tends to be in the calibrated model quite a bit larger than in a corresponding one sector model. And overall, the upshot is that the model, the calibrated model behaves somewhat more Keynesian than the one sector benchmark. In the sense that there is less intertemporal substitution in response to government spending and less growing out and therefore a larger multiplier. And this is interesting in light of the theory because as you are probably well aware is quite a bit of literature these days accounting for heterogeneity in the fiscal transmission mechanism. But this literature has focused on the household side and tank and whatnot. And this important literature has emphasized that heterogeneity on the household side is important for fiscal policy transmission. Now, relative to this literature, we take a step back because we say, okay, wait a minute. This is all important household heterogeneity but government spending itself is heterogeneous and that too matters for the fiscal transmission mechanism. Okay, so let me skip the other literature and just go right to the empirical analysis and just wonder about my connection. But it's okay. Yeah, you hear me well. I can hear you but I cannot see you. Your video is just. I don't know. It's not so important, I guess. What's going on here. Okay, so, yeah, you tell me once you can't hear me anymore. So our database is this website USA gas spending of which has been set up in response to accountability and transparency law legislated in 2006, but then the data was backdated. So we have now time series from 2001 to 2018 18 years of data. And this basically spans the entire universe of government contracts. A large chunk of this but by far not everything is contract that's awarded by the Department of Defense about half of them and two thirds by value but we also have lots of contracts, which are not the fence, which is not the fence spending in total we have 57 million observations. We have 160,000 recipient companies and so forth for each contract we see the government agency which is awarding this contract and the recipient firm. We compliment these states. So this is federal spending. We compliment this data by with data from the BA in order to also see whether to what extent our facts hold up at the state and local level and for some of the facts we established we can we can show this. And the last we compliment our data with the BLS data on price adjustments in order to establish how much price utilities in those sectors is where government spending is primarily active. So here's a bird's eye view on on what we are talking about again big chief in the standard macro mall would be government consumption expenditure and cross investment to give you some number. Of course, last year in the US, that was some 4000 billions in and when GDP was somewhat more than 20,000 billion so roughly speaking, we are looking at close to 20% of GDP. Yeah, this is partly federal party state and local. Our contract data covers what is indicated with these red boxes. This is roughly speaking 40% of the federal spending. And so 16% of overall spending. We compliment this with data from the BA for state and local spending we have some additional data on intermediate inputs there so all together we end up talking about 40% of of being cheap. What do we miss in terms of federal expenditure we miss the R&D spending which comes mostly in grants. So this is not contracts data included in the database. And then also in terms of government accounting, we think of government consumption expenditure in terms of a production function, which uses intermediate goods for which we have data, but we don't have data for the value added which is pro large extent compensation of employees. In other words, we do not have data on government wages, which in the standard macro model would be part of the key. Okay, so with this let me go to the first fact and that's the granularity of government spending. And here we simply mean that government spending is concentrated among a few firms and sectors. This holds for defense and non spending as well as for state and local spending. There are different ways of establishing granularity but I mean the simplest possible way is maybe just to show this chart here. So here you see for all the contracts in our database, the concentration among firms or sectors. So in the left panel, you see over time that's a very robust pattern over time. Let's focus on the darkest blue line which hoovers around at 80%. And that's the top 1% of firms they receive about 80% of the contract. So that's a very high degree of concentration. The top 1% of all firms receive 80% of contracts. And if you turn to sector breakdown, the six digit breakdown would amount to about 1000 sectors. And if you look at the top 10, so 100, the top 100 sectors in the middle panel, they would also account for roughly 80% of government spending over time. And if you turn to the very right panel, you look at the breakdown in the two digit sector classification and here the most solid line gives you the top five out of some 25 sectors. So here the top five sectors receive 80% of government spending. In that sense, very straightforward across firms and sectors, we observe a high concentration of government spending in those sectors. Another way to establish this granularity following our backs would be to look at the distribution of contracts in the data and benchmark this against theoretical distribution. And here we take the log normal distribution which has fat tails. Thus, it is indicative of a certain concentration of contracts. And we see that here in these QQ plots, we plot the actual quantiles along the vertical axis and we plot this against the theoretical quantiles on the horizontal axis if the actual distribution aligns well with the log normal, which is simulated assuming the same mean and variance as we have in the data. You see that that would tell you that the log normal distribution approximate the actual distribution quite well. And you see that this is the case at the transaction level at the formula at the sector level. So we walk away from this first fact just stressing that there's a lot of granularity. Either in the sense that it's concentrated among firms and very few firms and sectors, or if you prefer this measure that there is a fat tail distribution. Second fact, sector bias. Here we simply mean that the composition of government spending across sectors differs from the allocation of private spending across sectors. And one way to put this down formally is that simple equation here where GK would be the share of spending going to sector K relative to overall spending. And that to the extent that their sector bias differs from the share of that sector in GDP. Graphically, we have two way two breakdowns here to make this point in the left panel you see a breakdown according to federal spending that would be the red dots. And here we measure against the vertical axis, the share of government spending in a particular sector against the horizontal axis, which gives you the share of that sector in GDP. And to the extent that these dots do not sit on the 45 degree line that is indicative of the sector bias and that happens basically for almost all sectors. Similarly, in the right panel we break up the data into defense and non defense spending and very similar picture emerges in the sense that there's big bias. To make things a little bit more concrete here, I give you a table with the two sector two digit sector classification. And it's all in such a way that the top three sectors in terms of government spending come up on top. And you see that together they amount to some 70% of spending manufacturing being the largest sector the second largest is professional scientific and technical services. And the third one is administrative and waste management. So these three sectors while accounting for 70% of government spending account only for some 18% of GDP. You know, if you want a sector which is big in terms of GDP, but small in terms of government spending there will be healthcare social assistance. Okay. Fact three is moderate persistence. Here we look at the duration of contracts. And we also look at the duration of firms in our sample where we have 18 years of data and we can look how long certain firms are in the data set. And here I should stress that this data is very comprehensive. All contracts above $25,000 have to be recorded in the database. So if a firm drops out of our sample, we know it's basically not catering to the government. The contract duration median is only 31 days here. I have a nice example. Let me tell you my example. Let's just an arbitrary contract from September 2008 when Sykes property and appraisal services were awarded a contract for single family housing appraisals. So this is a job which can be done in a couple of days. And then at the opposite end of the spectrum, we have a contract the longest lasting contract in our sample is a 40 year and 10 month contract awarded to by the Department of Energy to Stanford University for the operation and management of the National Accelerator Lab. Okay, so that's 43. That's the exception, of course. You see that there are a few longer lasting contracts. The mean is 123 days here in the left panel. I give you the distribution, the cumulative distribution function of the contract. And on the horizontal axis, you see the duration in days and the dashed line marks 365 days. So there will be contracts to the left of this, which last less than a year and that's killed the majority of contracts. That's the dark line in the lighter line is the multi transaction contracts which naturally last longer, but even also here the large majority of contracts. It has fairly short durations. Now, in the right panel, I was talking about this already you see the share of firms which stay in the data set for a number of years. So the large majority of firms stay only one year, meaning they're no longer catering to the government, and then only amongst the most important firms, or let's let's look at the top 10% of firms that will be the blue line with circles. You see only a small fraction say 5% of those firms stay in the in the sample for the entire period for 18 years. Okay. So another way to to establish persistence as we do it in standard macro analysis would be to estimate a one process we do so at the sector level using the two digit classification. And we end up with parameters around point four or the entire ranges between point two and point 67, which is somewhat smaller than what we have in the data. This is not entirely irrelevant to the extent that this persistence. If you take a neoclassical perspective governs the wealth effect of government spending shock so to speak. Fourth fact, ideas in credit shocks dominate when it comes to accounting for the variation in the accurate spending in the in the short run. Yeah. How do we see this? Well, we do estimate as I said, a one processes on on at the sector level in a panel and we include time fixed effects to see whether there's a common component. Once we do this, the R square increases from 97.9 to 98.3%. So there is no common component in other words, there's no aggregate component. Also, if we look at the ideas in credit innovation, so these shock grosses, we don't see a systematic pattern. So some shocks are positive some innovations are positively correlated others are negatively correlated. So it is really this bottom up thing, which which we observe here in the data. Another way to establish this following up X would be to compute the granular residual. This is what we do here with the gamma term. Here we look at the deviation of spending growth in one firm. I or one sector I relative to the firm or sectoral average weighted with that with that sectors of firm share in the previous period to see whether in some sectors of firms there's important variation relative to what's going on in the in the in the in the average. And whether that granular residual can then account for overall changes in government spending, we do this following up X for firms and sector separately in those columns, including lags, not including lags. Either way, we get a fairly high R squared. That's in the very bottom line. You see our squares between 0.3 and 0.39 suggesting that this granular residual is indeed accounting for aggregate fluctuations. And so ideas in credit fluctuations move this around rather than, you know, having an accurate movement in shifting these things at the micro level. Last fact, last fact. When we look at the price thickness across sectors and here again we use the two sector classification. We observe that government spending is concentrated in sectors where price thickness is pervasive in the left panel. You see the two blue balls and against the horizontal axis. You see the share of those sectors in spending. This is close to 30%. And on the vertical axis we measure the frequency of price changes, which is about 10%. This is a monthly frequency. So in those sectors only some 10% of firms to change their price in a given month. And that's different from the other sectors where about 22% of firms change their prices. We robustify this finding by looking at the inverse of the contract duration as a proxy for the stickiness. And here we obtain a very similar picture. Okay, that fact will turn out to be important when we turn to theory, which is what I do now. Okay, so the question is, to what extent does this heterogeneity matter? Yeah. And to address this question, we start from the basic textbook version of the New Keynesian model, monopolistically competitive environment, carbon pricing as a representative household, which allocates consumption over time. Monetary policy targets inflation. That's kind of important to keep things transparent. We don't assume a tailor or we say really monetary policy goes for stabilizing prices completely. Yeah, there's zero inflation. And this makes things very straightforward. We say that government spending is financed through lump sum taxes as usual. And the only innovation relative to the workhorse model is now that we say we have two sectors rather than one. Yeah. And those sectors differ in their steady state shares in government spending. So gamma would be the weight of spending that goes to sector one. Omega would be the weight of private spending that goes to sector one. And then you have the share of private spending and output. That's the latter Cedar so that indulgingly we determine the size of sector one and as a weighted average of those shares going of the private and public shares going to that sector. Importantly, the sectors also differ in terms of price rigidity. So alpha would be the color parameter and that's different in the two sectors. Yeah. Now the model is super simple. Let me skip the algebraic exposition and tell you what we do with the model. So we approximate the model as usual around a deterministic steady state linearize. Now the steady state is asymmetric in the sense that we have the sector bias. Yeah. I'd be more specific about how we map the five facts into the model when we turn to the calibration. For now, the important assumption is that prices are completely flexible in sector one while they are sticky in sector two. And this is very stylist and makes it easy to get closed form results. And so what's also important, we have spending at the sector level. So we have shocks taking place in sector one and sector two. This is very stylized. I was pressing fact four. Fact four is that it's really the idiosyncratic shocks which dominate the overall fluctuations here in the simplest possible model. We could think of this boils down to assuming, okay, there's two sectors in each sector you have a shock rather than having an aggregate shock. Okay, now in this world, you have the terms of trade, the relative price of the two sectors as an endogenous variable. And that's the only endogenous variable. And so you can still solve this paper, this model paper pencil. And then we having so for the terms of trade, we can then solve for consumption. And then we arrive at equation seven, which I find quite exciting. It gives you the solution of consumption as a function of the terms of trade tower from the previous period as an endogenous state. We can ignore this for now and focus on G1 and G2, which are spending shocks taking place in sector one and sector two respectively. Now these guys here are just scaling factors so that we are looking at shocks, which are normalized to be equal to 1% of GDP. What's exciting is the TDA guys, the TDA guys are measuring the impact of the spending shocks on consumption and they are positive. They are positive and here's a minus in front, meaning you always get drowning out of private expenditures in response to the spending shocks. However, however, the TDA one differs from the TDA two. G1, I should remind you is sector one spending and sector one differs from sector two in that price are completely flexible by prices are sticky in sector two. Now, as a result, TDA one can become arbitrarily large, it can be arbitrarily large, meaning you get lots of drowning out. And this is really the heart of the model how changing how accounting for the sector composition of spending is really changing the fiscal transmission mechanism. So what's going on? Think of an increase of government spending in sector one. This is inflationary. Monetary policy is sitting there preventing inflation to go up. So given the spending impulse, this inflationary spending impulse, it has to put hard on the brakes because since government spending is raised in sector one where prices are very flexible, inflation is very responsive. So this means a lot of counteracting action from the central bank interest rates go up in order to reduce consumption because only by lowering aggregate demand, which is consumption in this model. The central bank can prevent inflation from rising. So maybe the easiest ways to if you contrast this with TDA two if spending goes up in the sticky sector inflation is not very responsive and therefore monetary policy has not to react very much. Okay, so it's really in order to keep inflation in check that monetary policy brings about a contraction of consumption. But the important thing here for monetary policy is that it only has a very blunt instrument. It can only steer consumption by adjusting the policy rate and consumption falls on all sectors. So it may in response to higher spending in sector one, which is very inflationary also has to bring down consumption in sector two, which buys very little in terms of reduced inflation because prices are very sticky in sector two. Now this response for consumption map directly into output multipliers. Remember TDA one can be very large and as a result gamma, which would be a measure of the output multiplier can be negative. And that's a new result to the extent that the one sector model only knows non zero multipliers. Okay, let me spend the last bit of my presentation on how we calibrate this model to capture the five facts. There's a bunch of parameters which are the same in as in a standard model. And then we have the five facts. And here we say, okay, sector two represents the top three sectors in terms of government spending for in which government spending, which account for 70% of government spending and their sector bias. So we say that that sector is relatively small, meaning that private spending falls largely on sector one where prices are more flexible. We capture the persistence simply by estimating a one processes on data for the top three sectors and the remaining sectors. We end up with moderate values here. We account for fact four, namely that there's ideas in credit trucks driving the aggregate variation over time by looking at sector specific trucks rather than at aggregate trucks. And lastly, we calibrate this model to capture the degree of price thickness, which we do find in the BLS data for those sectors. So sector two is now a sticky sector, sector one is much less sticky, but it's not flexible prices as in the previous analysis. And here you see what's going on in the model. This is the impulse response of output to a spending shock normalized to 1% of GDP in the left panel. That shock takes place in sector one. In the right panel, it's a shock taking place in sector two. Okay, now the blue line is the benchmark, the symmetric baseline case where you have a multiplier of point two that's rather low. And that is understood because monetary policy targets inflation. It basically gives you here the flex price allocation, the REC multiplier, which is moderate. Okay, so that's the benchmark. And then we can introduce our five facts progressively to see how they change the picture. You see, if you account for different pricing frictions in the two sectors, these matters speak time. Now the multiplier is negative in response to sector one shock. It's positive in response to sector two shock. And it's much larger than in the baseline. And if we introduce all our features in the end, I don't go through everything here. I just tell you, in the end, the red line gives you the responses for the five facts model. It doesn't make a difference much for sector one shock effects, but it matters speak time for sector two shocks. Here now the effect is much bigger. Why is that? Well, intuitively, it's simply because in sector two, there's more price stickiness and therefore raising spending there requires less of a counteracting response of monetary policy. And so there's less crowding out. Bottom line, there's no such thing as a fiscal shock. It really depends on where it originates in this economy. In terms of empirical performance, let me stress one more thing here. Let's also look at the interest rate response to a sector two shock because sector two after all is the sector where most of the fiscal action is taking place. If we look at this sector, and we look how at how the interest rate response to a shock here, we can compare our five facts model to the symmetric baseline case. And we see now a much more moderate response of the interest rate in response to the shock. This is actually what we have in aggregate time series models. There's work by more for Oolik, my own work with Prasetti and Porto. And by the way, there's constantly this issue that interest rates are not very responsive to fiscal shocks. And this is precisely what our model predicts if you consider a sector two shock. Why is that sector two is relatively sticky. You don't need much of a monetary policy response. A little bit of monetary tightening is sufficient to stabilize inflation because most of the private spending happens in the relatively flex price sector. So a little bit of reducing consumption is enough to keep inflation in check. Okay, let me skip this in the interest of time and show you a last graph here. A last graph here where we look at the effective lower bound. That's the pink lines. And it's again the sector one shock and the sector two shock in the left and the right panel. And now you see that the ranking of multipliers flips. In the left panel, you see the output response to the sector one shock and it's now much larger. Also in the right panel, it's a bit larger. In both cases, the multiplier is now larger than one, but in particular the sector one shock effect changes quite a bit. Because now at the zero lower bound, the impact on inflation is not met by a contractionary monetary response because arguably the central bank is finding the interest rate too high to begin with. So it's not raising interest rates in response to the higher stimulus. So here we see then a much, much amplification in response to the shock. However, we don't see much amplification when it comes to sector two shock. It's just a tiny increase. And that again is consistent what we have in the data. At least if we trust Rami and Subaru's recent JPEPs. Where they document not much of an increase in the multiplier at the ERP. So that's consistent once we look at sector two shocks. Finally, let me conclude our anatomy of BigG delivers five facts about government spending. There's granularity. There's sector bias. There's only moderate persistence of government spending. It is the sector shocks, the isocratic shocks, which drive the aggregate variation. And we find that government spending is concentrated in relatively sticky sectors accounting for those two facts in a two sector new Keynesian model. We find that in general interest rate responses are more moderate. There's less crowding out and the multiplier is larger. And so in a nutshell, if you want capturing for the micro facts helps the model to account for what we have in terms of macro evidence times years evidence. Thank you. Thank you again. I would pass immediately the floor on to Henrik for his discussion. Thanks a lot, Isabel. And thanks for inviting me to discuss this paper. I hope you hear me well and see the slides by now. Now, that's a very nice paper. And before I jump into it, let me make two disclaimers. First of all, I'm not relative to one of the authors of the paper, Michael, even though we share the same surname. So no hidden incentives here. And the second disclaimer, I'm working at a policy institution, the Bundesbank. And this is why I'm expressing my own views here. Now, this paper makes quite a few contributions and contributions. And it starts from the observation, which is quite a plausible one that if you look at government spending G, this of course is not one large transaction. But instead, G is composed of many smaller transactions. And the authors make this visible using this new database for the U.S., which has all the large share of the government procurement contracts. And from this database, they go and extract their five facts. I just go over them briefly. First, government spending is concentrated in just a few firms or sectors. The sector composition differs depending on whether you look at the government or the private sector. So that's what they call a bias. There is moderate persistence of government spending at all levels of disaggregations, so the contract level, firm level and the sector level. The spending to just a few influential firms or sectors is going to dominate aggregate spending growth. And then there's a statement about price stickiness. Namely, you get a concentration of government spending in sectors with relatively sticky nominal prices. Now, these are the facts. And then, as Gernaut explained quite nicely, they set up a theory, multi-sector, Ukraine's in model accounting for these facts. And I guess the main result there is that once you count for these facts, you tend to get larger aggregate fiscal multiplier. So, I mean, as I said, this is a great paper. I found it quite exciting to read and it's full of a lot of interesting new facts and insights. And if you're not happy after reading the paper with what you got, then you still have an appendix of about 100 pages. And after reading this, you certainly got to be happy. Okay, so this is a very rich paper. And I'm quite sure it's certainly not the last paper written with this newly available data because this data is extremely rich and very, very interesting. Now, I wanted to make three comments. First of all, how about looking at even more dimensions of heterogeneity in the data? And I understand this is kind of a tough call. I mean, I told you about the appendix and it's probably a bit too much, but nevertheless, I think these are important dimensions and I at least wanted to point them out. My second comment is about what we learn about the sectoral government spending process. And when I say sectoral, I also mean the cross-section of firms, right? So, I refer to this interchangeably. Now, the third comment is about the intensive versus extensive margin of sectoral government spending. So, let me jump into the first one, more dimensions of heterogeneity. Now, that data, Gernot and his co-authors have, they comprise both government consumption and investment, right? So, consumption would be purchases of intermediate goods and services and investment is investment in structures, equipment, software, what have you. Now, we know that government consumption, that consumption and investment in the private sector behave quite differently also at the disaggregate level, right? So, it would be really nice if you could show that your five facts also hold for consumption separately and investment separately. Now, that's the first dimension. The second dimension, implied unit prices. You show in your theory in a quite transparent way as I think that the fiscal multiplier depends, so the aggregate fiscal multiplier depends on relative prices between the two sectors in your theory. Now, this raises the question whether you can say something about prices and relative prices using your data. And I was wondering whether you have information or could compute a contract implied unit prices because those unit prices, in case they exist, they would be very interesting to compare across different sectors. And you could also compare them with the respective prices paid by the private sector, right? In the respective NAIC sector. Now, that was the second dimension, the third dimension. Now, government spending is often pre-announced. We know this, there's a big literature on this. And if you look at your big and sizeable cross-section, it's quite likely I find that the pre-announcement horizon is going to vary across these different sectors. Now, for example, think of an infrastructure project, building a bridge, say you need quite a bit of time in order to prepare such a project. While if you acquire services, that is probably something that is much, much faster. Now, as another example, think of a sector in which the average or median value, the value of the average or median contract is either very small or very large, right? And for the large value of the median contract, you would also expect that you need quite a bit of time in order to pre-announce and implement such a project. Now, the question of course is, is there such variation, cross-section variation in the pre-announcement horizon? And if so, can that variation be exploited to learn more about physical shocks and their pre-announcement effects? Now, my second comment is about what we learned about the sectoral government spending process. Now, before the big G-paper life was simple, at least when you thought about aggregate government spending shocks, it usually was an A1 process. And once you specified it, you just needed to calibrate the mean, the G, the persistence, the row and the variation in the shock. And that was about it. Now, with the big G-paper things become more comprehensive and general, and you probably have to think of a multi-variate spending process, right? Now, to illustrate this here, I put down a VAR, which now makes G a vector which contains the cross-section of government spending across all those different sectors of firms, what have you. And then the VAR tells you how the cross-section's deviation from its mean is going to evolve over time, and all this is going to be driven by the reduced form epsilon with the variance covariance matrix sigma. Now, it's pretty clear that this VAR is just illustrative, and it's clearly not parsimonious enough for the paper's large cross-sections. Remember, if they look at sectors, it's about 1,000. If they look at firms, it's about 5,000, right? So you're not going to capture this VAR, that's pretty clear. But in the theory, that may be a bit different, right? Now, I think that the paper is extremely explicit and detailed about two objects of this process, and this is the average value of the cross-section. That's what fact one and two are about, right? So the big concentration of government spending in just one sector, in just a few sectors, and the bias relative to the private economy. And the other dimension on which the paper is very explicit is the diagonal elements of the A1 matrix, right? Moderate persistence in government spending, irrespectively of the level that you look at. That's fact three. But there is more to this process, and I think there the paper is a little bit less detailed. Now, one thing is the entire leg structure, here I captured this illustratively by A1 and A2, which of course is informative about the dynamics below the effects between the sectors, right? And then there's the variance-covariance matrix, which of course is informative about contemporaneous below versus between the sectors. And of course, fact four of the paper is exactly about these things. But I think maybe the authors can do a bit more along these lines. And of course, the reason is that these objects A and sigma are going to shape the dynamics of the cross-section in deviation from its mean. And again, the paper's theory suggests that these dynamics actually matter for the magnitude, but also for the time variation of the aggregate physical multiplier. So what could be done here? Of course, it would be nice to get a fairly parsimonious process in order to describe the big cross-section G here. And maybe one way is to think about the influential sectors versus the fringe sectors, so the essential ones versus the inessential ones. Then of course, explain a little bit or capture a little bit spillovers across these sectors. And what I would actually find quite interesting is to think about whether there are further variables such as relative prices between the sectors that are important to describe the evolution of the cross-section G. Now, my last comment is about intensive versus extensive margin of sectorial governance spending. Now, by intensive margin, I mean changes in the value of ongoing contracts, right? Sometimes these contracts are negotiated or extended, and then the value is going to change. Now, the extensive margin would be new contracts in a given period, or of course expiring contracts, right? Both of them would contribute to the extensive margin. And if I understand correctly what the paper at the moment does, it's going to pull both margins, right, and considers them jointly. Now, to think more about this, let's think about the evolution of the stock of government spending in a given sector K. And for the moment, let's ignore the extensive margin, right? So changes in ongoing contracts, let's ignore those. Now, then the stock of government spending is today is going to be equal to the stock yesterday, plus what we get in terms of the value of new contracts today minus the value of expiring contracts in the previous period. Now, this is just accounting, but now let's add an assumption, namely that contracts expire at a given rate, one minus roll. And then you get that the stock today is going to be a fraction of the stock yesterday, plus the value of new contracts today. Now, that's a fairly simple equation, and you could demean it. So consider the deviation of the stock from its unconditional mean. And then you see that what drives innovations in this deviation is going to be unexpected variation in the value of new contracts. And to me, this suggests that if you want to learn about these innovations to the stock, then what you could do is look at the value of new contracts instead of looking at the value of all contracts. Now, I was wondering whether a specificity of the data, namely what the authors call the September seasonality. So that's a seasonality that arises because in September the fiscal year ends. And as a result, you get a big increase in the number and I guess also in the value of new contracts. Now, I was wondering whether you could exploit the September seasonality in order to disentangle the extensive versus the intensive margin. Of course, there's a caveat and this is that the seasonality is going to be anticipated and it's probably not totally trivial to deal with this anticipation effect. Here's a graph that Gernot also showed the contract duration in the data set and you see that the dashed line again, that's contracts with the duration of a year. You see that a fairly large share of contracts is shorter than a year. So I think this intensive versus extensive margin issue is really not such a big issue if you think about annual frequency. But if you go to shorter frequency, for shorter frequencies, say a quarter or maybe even a month, then this issue I think is actually quite important. Now, I have just a few things to say and these are only minor things to say about the theory. So I talked again about this directly and instead got to conclude. I think this is a very complete paper on an extremely fundamental issue. And for me, this paper works very well. So I'm by now happy to throw overboard the big G fiction. I think the paper raises quite important follow up questions, which are quite multifaceted, I have to say. And my big picture conclusion from this paper is it seems fair to replace the big G fiction by big sector G's. And once you do this, then this is going to alter the transmission of fiscal shocks. So that's it from my side. Thanks a lot. Thanks a lot Henning for this very nice discussion and actually it's a great paper. I should have said that immediately after the presentation of Gernot. I mean, also getting such a granularity into the data and it helps us much better to understand also how fiscal policy transmits to the economy. Gernot, I don't know how you want to do it, but you want to respond directly to Henning. I saw in the chat there's already three questions that came in, mainly on the data actually. So would you prefer to respond first to this question? Let me just thank you and Henning. I think that is really brief because he raised a lot of points and he was very generous in his discussion. I mean, he casted this as follow up issues and I'm totally on board. I mean, we haven't looked at this really because we are struggling hard to sort of organize our thinking along those five facts. But there's clearly more to be done. I agree with this. So there is the implied unit price. It's a great idea. I think this will be very hard to tease this out and analyze it systematically, but we could do more and the same with the pre-announcement thing. Let me just say here, I mean, these contract durations, we know we observe the starting date of a contract and then the end date. And then, of course, you could also take up the issue to what extent the delivery differs from the signing of the contract. So there's another anticipation period, but this is hard to tease out, but I understand from a theory point of view. This would be very interesting to learn more here. On the multivariate VRs, I like these ideas. We haven't done this and you're right. I think we could look at various aggregates and then run some parsimonious VRs and see whether there are some spillovers. Even though on impact, there seems to be very little spillovers, but it could take time for them to materialize. Totally agree. And the implicit and the intensive extensive margin of distinction is also a very useful idea. So I will talk to my co-authors about this, but this could put more structure on our analysis. So thank you a lot. Very useful comments. You're welcome. So as I already indicated, there were three questions in the chat, mainly on the data. I'm just, I'm sure you were busy with the presentation, so you haven't had a chance to look into the chat. So one comes from Mako Bassetto. I mean, he's asking whether there's an issue about the labeling of spending versus transfers. So I think he picked up on your presentation when you mentioned that healthcare spending is so small and he's asking, is it maybe because it's considered transfer rather than direct spending? And then there's two questions from Jacopo Cimaldomo. One is on the, he's asking the amounts that are reported in the contracts don't necessarily want to one correspond to what you see in the public accounts. So he thinks some contracts may be interrupted or others are only executed partially. So he's asking, can you control for these effects? I was wondering if he actually have information on these effects more generally. And then a second question that Jacopo is asking is that you mentioned that 80% of spending is directed to the top 1% of firms. So the biggest and so what is the share of firms operating in the defense sector among these firms? And where is the defense spending allocated when you're looking at it in your slide 12 of your presentation? All right. Thanks a lot for this question. So I mean, first the transfer thing. So it's true. I mean, we don't include transfers. So I mean, yes or no labeling. I mean, we follow I think standard practice in the sense that our G does not include transfers and our contract data does not include transfers. So we cannot speak to the issue. Now you could say that by changing transfers, the government is equally impacting the economy or in similar ways. But we just don't have in debt data. We don't have data on transfers. So that's a downside. On the other hand, it corresponds closely to what in the baseline model we look at when we do the big cheap thing. But I agree this would be important also to investigate on Chaco post thing. I mean, with the point to be controlled for for these disruptions. I mean, in the in the database, there's this category modifications and these modifications can be sizable and we do see them. But it's at a conceptual level, to be honest, it's not entirely clear, you know, across the entire universe of these contracts. What's the best way to do with these modifications. So in the end, we do not regard them for the for the simple fact that, you know, this could mean different things. Some of the things you you mentioned, but yeah. So I guess in the way we compute the facts, I would say they are robust with respect to these modifications. And we would get similar numbers, irrespectively of whether we throw out this modification or we take them in. But yeah, so this is maybe there's there's also more to be done here and try to understand better what the nature of these modifications is. On the defense spending we have in the appendix of the paper. I don't know whether it's post summaries at least posted on our websites. I do the entire statistics break up for separately for defense and non defense spending. And this type of granularity is not driven by defense spending in the sense that once we do this only for non defense spending. The share of Well, okay, maybe so that I have it here in front of me it's it's really between 75 and 80% also the share of the top 1% firms in overall spending if you only consider those contracts which are not department of defense contracts. So it's not that a few military firms account for this this kind of concentration you also have very very similar patterns in the non defense spending part of the contract which is about half the data set. Somehow my videos, I don't know what's going on. Yeah, we still hear you perfectly at least I do but yeah now you're back again so you're coming and going but it's good to see you once in a while at least in there. I'm checking if there's more questions from either the panelists who can also raise their hands or from the audience. I mean, maybe I asked a question myself I mean you know you you presented in your model results I mean like how you know, you know, depending on to which sector kind of fiscal spending is erected whether it's one where is the more sticky prices or not. That of course you would see different macro effects and then you attribute a lot of it to to monetary policy. But of course we have a lot of models these days where you would say, you know, monetary policy is constrained by the effective lower bound. How would that change our results and of course in reality we have no standard measures so you may still say there are other tools to which monetary policy reaches the economy but you know in these models very strictly then say monetary policy does not react. We analyze a little bit the ERB and so since I mean this is a new Keynesian model and in the new Keynesian model it's really the interaction of monetary and fiscal policy which determines the overall effect of fiscal policy and that that's really at the heart of our model and it's really this need for monetary policy to counteract the stimulus which differs depending on where the shock originates and here the sector bias and the granularity are key. And so you absolutely right the effective lower bound things turn upside down because as in the one sector model. We get larger multipliers in the flex price sector because here you want a lot of inflation coming out of the fiscal stimulus and that lowers the real interest rate and that stimulates aggregate demand as in the standard model. However in the other sector where garbage spending is is more pronounced. You don't have that kind of an effect because their inflation is not very responsive in the first place and so whether monetary policy response to this or not does not matter so much in the end. We like that result to the extent that it conforms well with that. In the rainy Subaru paper, which documents that that at the effective lower bound multipliers do not increase that much at least not as much as the one sector New Keynesian model. So we don't see this in the data and we can explain this to the extent that we say, yes, there's not much of a difference if we talk about the sticky sector anyway as the sector where spending is predominantly taking place. Thanks. I see no hands raise or additional questions but I see seven them has switched on her camera because she rightly noticed that it's quarter past past four so I would suggest we move to the second paper in this session.